1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 /* 29 * The objective of this program is to provide a DMU/ZAP/SPA stress test 30 * that runs entirely in userland, is easy to use, and easy to extend. 31 * 32 * The overall design of the ztest program is as follows: 33 * 34 * (1) For each major functional area (e.g. adding vdevs to a pool, 35 * creating and destroying datasets, reading and writing objects, etc) 36 * we have a simple routine to test that functionality. These 37 * individual routines do not have to do anything "stressful". 38 * 39 * (2) We turn these simple functionality tests into a stress test by 40 * running them all in parallel, with as many threads as desired, 41 * and spread across as many datasets, objects, and vdevs as desired. 42 * 43 * (3) While all this is happening, we inject faults into the pool to 44 * verify that self-healing data really works. 45 * 46 * (4) Every time we open a dataset, we change its checksum and compression 47 * functions. Thus even individual objects vary from block to block 48 * in which checksum they use and whether they're compressed. 49 * 50 * (5) To verify that we never lose on-disk consistency after a crash, 51 * we run the entire test in a child of the main process. 52 * At random times, the child self-immolates with a SIGKILL. 53 * This is the software equivalent of pulling the power cord. 54 * The parent then runs the test again, using the existing 55 * storage pool, as many times as desired. 56 * 57 * (6) To verify that we don't have future leaks or temporal incursions, 58 * many of the functional tests record the transaction group number 59 * as part of their data. When reading old data, they verify that 60 * the transaction group number is less than the current, open txg. 61 * If you add a new test, please do this if applicable. 62 * 63 * When run with no arguments, ztest runs for about five minutes and 64 * produces no output if successful. To get a little bit of information, 65 * specify -V. To get more information, specify -VV, and so on. 66 * 67 * To turn this into an overnight stress test, use -T to specify run time. 68 * 69 * You can ask more more vdevs [-v], datasets [-d], or threads [-t] 70 * to increase the pool capacity, fanout, and overall stress level. 71 * 72 * The -N(okill) option will suppress kills, so each child runs to completion. 73 * This can be useful when you're trying to distinguish temporal incursions 74 * from plain old race conditions. 75 */ 76 77 #include <sys/zfs_context.h> 78 #include <sys/spa.h> 79 #include <sys/dmu.h> 80 #include <sys/txg.h> 81 #include <sys/zap.h> 82 #include <sys/dmu_traverse.h> 83 #include <sys/dmu_objset.h> 84 #include <sys/poll.h> 85 #include <sys/stat.h> 86 #include <sys/time.h> 87 #include <sys/wait.h> 88 #include <sys/mman.h> 89 #include <sys/resource.h> 90 #include <sys/zio.h> 91 #include <sys/zio_checksum.h> 92 #include <sys/zio_compress.h> 93 #include <sys/zil.h> 94 #include <sys/vdev_impl.h> 95 #include <sys/spa_impl.h> 96 #include <sys/dsl_prop.h> 97 #include <sys/refcount.h> 98 #include <stdio.h> 99 #include <stdio_ext.h> 100 #include <stdlib.h> 101 #include <unistd.h> 102 #include <signal.h> 103 #include <umem.h> 104 #include <dlfcn.h> 105 #include <ctype.h> 106 #include <math.h> 107 #include <sys/fs/zfs.h> 108 109 static char cmdname[] = "ztest"; 110 static char *zopt_pool = cmdname; 111 112 static uint64_t zopt_vdevs = 5; 113 static uint64_t zopt_vdevtime; 114 static int zopt_ashift = SPA_MINBLOCKSHIFT; 115 static int zopt_mirrors = 2; 116 static int zopt_raidz = 4; 117 static int zopt_raidz_parity = 1; 118 static size_t zopt_vdev_size = SPA_MINDEVSIZE; 119 static int zopt_datasets = 7; 120 static int zopt_threads = 23; 121 static uint64_t zopt_passtime = 60; /* 60 seconds */ 122 static uint64_t zopt_killrate = 70; /* 70% kill rate */ 123 static int zopt_verbose = 0; 124 static int zopt_init = 1; 125 static char *zopt_dir = "/tmp"; 126 static uint64_t zopt_time = 300; /* 5 minutes */ 127 static int zopt_maxfaults; 128 static uint16_t zopt_write_fail_shift = 5; 129 130 typedef struct ztest_block_tag { 131 uint64_t bt_objset; 132 uint64_t bt_object; 133 uint64_t bt_offset; 134 uint64_t bt_txg; 135 uint64_t bt_thread; 136 uint64_t bt_seq; 137 } ztest_block_tag_t; 138 139 typedef struct ztest_args { 140 char za_pool[MAXNAMELEN]; 141 spa_t *za_spa; 142 objset_t *za_os; 143 zilog_t *za_zilog; 144 thread_t za_thread; 145 uint64_t za_instance; 146 uint64_t za_random; 147 uint64_t za_diroff; 148 uint64_t za_diroff_shared; 149 uint64_t za_zil_seq; 150 hrtime_t za_start; 151 hrtime_t za_stop; 152 hrtime_t za_kill; 153 traverse_handle_t *za_th; 154 /* 155 * Thread-local variables can go here to aid debugging. 156 */ 157 ztest_block_tag_t za_rbt; 158 ztest_block_tag_t za_wbt; 159 dmu_object_info_t za_doi; 160 dmu_buf_t *za_dbuf; 161 } ztest_args_t; 162 163 typedef void ztest_func_t(ztest_args_t *); 164 165 /* 166 * Note: these aren't static because we want dladdr() to work. 167 */ 168 ztest_func_t ztest_dmu_read_write; 169 ztest_func_t ztest_dmu_write_parallel; 170 ztest_func_t ztest_dmu_object_alloc_free; 171 ztest_func_t ztest_zap; 172 ztest_func_t ztest_zap_parallel; 173 ztest_func_t ztest_traverse; 174 ztest_func_t ztest_dsl_prop_get_set; 175 ztest_func_t ztest_dmu_objset_create_destroy; 176 ztest_func_t ztest_dmu_snapshot_create_destroy; 177 ztest_func_t ztest_spa_create_destroy; 178 ztest_func_t ztest_fault_inject; 179 ztest_func_t ztest_vdev_attach_detach; 180 ztest_func_t ztest_vdev_LUN_growth; 181 ztest_func_t ztest_vdev_add_remove; 182 ztest_func_t ztest_scrub; 183 ztest_func_t ztest_spa_rename; 184 185 typedef struct ztest_info { 186 ztest_func_t *zi_func; /* test function */ 187 uint64_t zi_iters; /* iterations per execution */ 188 uint64_t *zi_interval; /* execute every <interval> seconds */ 189 uint64_t zi_calls; /* per-pass count */ 190 uint64_t zi_call_time; /* per-pass time */ 191 uint64_t zi_call_total; /* cumulative total */ 192 uint64_t zi_call_target; /* target cumulative total */ 193 } ztest_info_t; 194 195 uint64_t zopt_always = 0; /* all the time */ 196 uint64_t zopt_often = 1; /* every second */ 197 uint64_t zopt_sometimes = 10; /* every 10 seconds */ 198 uint64_t zopt_rarely = 60; /* every 60 seconds */ 199 200 ztest_info_t ztest_info[] = { 201 { ztest_dmu_read_write, 1, &zopt_always }, 202 { ztest_dmu_write_parallel, 30, &zopt_always }, 203 { ztest_dmu_object_alloc_free, 1, &zopt_always }, 204 { ztest_zap, 30, &zopt_always }, 205 { ztest_zap_parallel, 100, &zopt_always }, 206 { ztest_traverse, 1, &zopt_often }, 207 { ztest_dsl_prop_get_set, 1, &zopt_sometimes }, 208 { ztest_dmu_objset_create_destroy, 1, &zopt_sometimes }, 209 { ztest_dmu_snapshot_create_destroy, 1, &zopt_rarely }, 210 { ztest_spa_create_destroy, 1, &zopt_sometimes }, 211 { ztest_fault_inject, 1, &zopt_sometimes }, 212 { ztest_spa_rename, 1, &zopt_rarely }, 213 { ztest_vdev_attach_detach, 1, &zopt_rarely }, 214 { ztest_vdev_LUN_growth, 1, &zopt_rarely }, 215 { ztest_vdev_add_remove, 1, &zopt_vdevtime }, 216 { ztest_scrub, 1, &zopt_vdevtime }, 217 }; 218 219 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) 220 221 #define ZTEST_SYNC_LOCKS 16 222 223 /* 224 * Stuff we need to share writably between parent and child. 225 */ 226 typedef struct ztest_shared { 227 mutex_t zs_vdev_lock; 228 rwlock_t zs_name_lock; 229 uint64_t zs_vdev_primaries; 230 uint64_t zs_enospc_count; 231 hrtime_t zs_start_time; 232 hrtime_t zs_stop_time; 233 uint64_t zs_alloc; 234 uint64_t zs_space; 235 ztest_info_t zs_info[ZTEST_FUNCS]; 236 mutex_t zs_sync_lock[ZTEST_SYNC_LOCKS]; 237 uint64_t zs_seq[ZTEST_SYNC_LOCKS]; 238 } ztest_shared_t; 239 240 static char ztest_dev_template[] = "%s/%s.%llua"; 241 static ztest_shared_t *ztest_shared; 242 243 static int ztest_random_fd; 244 static int ztest_dump_core = 1; 245 246 static boolean_t ztest_exiting = B_FALSE; 247 248 extern uint64_t metaslab_gang_bang; 249 extern uint16_t zio_zil_fail_shift; 250 extern uint16_t zio_io_fail_shift; 251 252 #define ZTEST_DIROBJ 1 253 #define ZTEST_MICROZAP_OBJ 2 254 #define ZTEST_FATZAP_OBJ 3 255 256 #define ZTEST_DIROBJ_BLOCKSIZE (1 << 10) 257 #define ZTEST_DIRSIZE 256 258 259 static void usage(boolean_t) __NORETURN; 260 261 /* 262 * These libumem hooks provide a reasonable set of defaults for the allocator's 263 * debugging facilities. 264 */ 265 const char * 266 _umem_debug_init() 267 { 268 return ("default,verbose"); /* $UMEM_DEBUG setting */ 269 } 270 271 const char * 272 _umem_logging_init(void) 273 { 274 return ("fail,contents"); /* $UMEM_LOGGING setting */ 275 } 276 277 #define FATAL_MSG_SZ 1024 278 279 char *fatal_msg; 280 281 static void 282 fatal(int do_perror, char *message, ...) 283 { 284 va_list args; 285 int save_errno = errno; 286 char buf[FATAL_MSG_SZ]; 287 288 (void) fflush(stdout); 289 290 va_start(args, message); 291 (void) sprintf(buf, "ztest: "); 292 /* LINTED */ 293 (void) vsprintf(buf + strlen(buf), message, args); 294 va_end(args); 295 if (do_perror) { 296 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), 297 ": %s", strerror(save_errno)); 298 } 299 (void) fprintf(stderr, "%s\n", buf); 300 fatal_msg = buf; /* to ease debugging */ 301 if (ztest_dump_core) 302 abort(); 303 exit(3); 304 } 305 306 static int 307 str2shift(const char *buf) 308 { 309 const char *ends = "BKMGTPEZ"; 310 int i; 311 312 if (buf[0] == '\0') 313 return (0); 314 for (i = 0; i < strlen(ends); i++) { 315 if (toupper(buf[0]) == ends[i]) 316 break; 317 } 318 if (i == strlen(ends)) { 319 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", 320 buf); 321 usage(B_FALSE); 322 } 323 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { 324 return (10*i); 325 } 326 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); 327 usage(B_FALSE); 328 /* NOTREACHED */ 329 } 330 331 static uint64_t 332 nicenumtoull(const char *buf) 333 { 334 char *end; 335 uint64_t val; 336 337 val = strtoull(buf, &end, 0); 338 if (end == buf) { 339 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); 340 usage(B_FALSE); 341 } else if (end[0] == '.') { 342 double fval = strtod(buf, &end); 343 fval *= pow(2, str2shift(end)); 344 if (fval > UINT64_MAX) { 345 (void) fprintf(stderr, "ztest: value too large: %s\n", 346 buf); 347 usage(B_FALSE); 348 } 349 val = (uint64_t)fval; 350 } else { 351 int shift = str2shift(end); 352 if (shift >= 64 || (val << shift) >> shift != val) { 353 (void) fprintf(stderr, "ztest: value too large: %s\n", 354 buf); 355 usage(B_FALSE); 356 } 357 val <<= shift; 358 } 359 return (val); 360 } 361 362 static void 363 usage(boolean_t requested) 364 { 365 char nice_vdev_size[10]; 366 char nice_gang_bang[10]; 367 FILE *fp = requested ? stdout : stderr; 368 369 nicenum(zopt_vdev_size, nice_vdev_size); 370 nicenum(metaslab_gang_bang, nice_gang_bang); 371 372 (void) fprintf(fp, "Usage: %s\n" 373 "\t[-v vdevs (default: %llu)]\n" 374 "\t[-s size_of_each_vdev (default: %s)]\n" 375 "\t[-a alignment_shift (default: %d) (use 0 for random)]\n" 376 "\t[-m mirror_copies (default: %d)]\n" 377 "\t[-r raidz_disks (default: %d)]\n" 378 "\t[-R raidz_parity (default: %d)]\n" 379 "\t[-d datasets (default: %d)]\n" 380 "\t[-t threads (default: %d)]\n" 381 "\t[-g gang_block_threshold (default: %s)]\n" 382 "\t[-i initialize pool i times (default: %d)]\n" 383 "\t[-k kill percentage (default: %llu%%)]\n" 384 "\t[-p pool_name (default: %s)]\n" 385 "\t[-f file directory for vdev files (default: %s)]\n" 386 "\t[-V(erbose)] (use multiple times for ever more blather)\n" 387 "\t[-E(xisting)] (use existing pool instead of creating new one)\n" 388 "\t[-T time] total run time (default: %llu sec)\n" 389 "\t[-P passtime] time per pass (default: %llu sec)\n" 390 "\t[-z zil failure rate (default: fail every 2^%llu allocs)]\n" 391 "\t[-w write failure rate (default: fail every 2^%llu allocs)]\n" 392 "\t[-h] (print help)\n" 393 "", 394 cmdname, 395 (u_longlong_t)zopt_vdevs, /* -v */ 396 nice_vdev_size, /* -s */ 397 zopt_ashift, /* -a */ 398 zopt_mirrors, /* -m */ 399 zopt_raidz, /* -r */ 400 zopt_raidz_parity, /* -R */ 401 zopt_datasets, /* -d */ 402 zopt_threads, /* -t */ 403 nice_gang_bang, /* -g */ 404 zopt_init, /* -i */ 405 (u_longlong_t)zopt_killrate, /* -k */ 406 zopt_pool, /* -p */ 407 zopt_dir, /* -f */ 408 (u_longlong_t)zopt_time, /* -T */ 409 (u_longlong_t)zopt_passtime, /* -P */ 410 (u_longlong_t)zio_zil_fail_shift, /* -z */ 411 (u_longlong_t)zopt_write_fail_shift); /* -w */ 412 exit(requested ? 0 : 1); 413 } 414 415 static uint64_t 416 ztest_random(uint64_t range) 417 { 418 uint64_t r; 419 420 if (range == 0) 421 return (0); 422 423 if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r)) 424 fatal(1, "short read from /dev/urandom"); 425 426 return (r % range); 427 } 428 429 static void 430 ztest_record_enospc(char *s) 431 { 432 dprintf("ENOSPC doing: %s\n", s ? s : "<unknown>"); 433 ztest_shared->zs_enospc_count++; 434 } 435 436 static void 437 process_options(int argc, char **argv) 438 { 439 int opt; 440 uint64_t value; 441 442 /* By default, test gang blocks for blocks 32K and greater */ 443 metaslab_gang_bang = 32 << 10; 444 445 /* Default value, fail every 32nd allocation */ 446 zio_zil_fail_shift = 5; 447 448 while ((opt = getopt(argc, argv, 449 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:z:w:h")) != EOF) { 450 value = 0; 451 switch (opt) { 452 case 'v': 453 case 's': 454 case 'a': 455 case 'm': 456 case 'r': 457 case 'R': 458 case 'd': 459 case 't': 460 case 'g': 461 case 'i': 462 case 'k': 463 case 'T': 464 case 'P': 465 case 'z': 466 case 'w': 467 value = nicenumtoull(optarg); 468 } 469 switch (opt) { 470 case 'v': 471 zopt_vdevs = value; 472 break; 473 case 's': 474 zopt_vdev_size = MAX(SPA_MINDEVSIZE, value); 475 break; 476 case 'a': 477 zopt_ashift = value; 478 break; 479 case 'm': 480 zopt_mirrors = value; 481 break; 482 case 'r': 483 zopt_raidz = MAX(1, value); 484 break; 485 case 'R': 486 zopt_raidz_parity = MIN(MAX(value, 1), 2); 487 break; 488 case 'd': 489 zopt_datasets = MAX(1, value); 490 break; 491 case 't': 492 zopt_threads = MAX(1, value); 493 break; 494 case 'g': 495 metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value); 496 break; 497 case 'i': 498 zopt_init = value; 499 break; 500 case 'k': 501 zopt_killrate = value; 502 break; 503 case 'p': 504 zopt_pool = strdup(optarg); 505 break; 506 case 'f': 507 zopt_dir = strdup(optarg); 508 break; 509 case 'V': 510 zopt_verbose++; 511 break; 512 case 'E': 513 zopt_init = 0; 514 break; 515 case 'T': 516 zopt_time = value; 517 break; 518 case 'P': 519 zopt_passtime = MAX(1, value); 520 break; 521 case 'z': 522 zio_zil_fail_shift = MIN(value, 16); 523 break; 524 case 'w': 525 zopt_write_fail_shift = MIN(value, 16); 526 break; 527 case 'h': 528 usage(B_TRUE); 529 break; 530 case '?': 531 default: 532 usage(B_FALSE); 533 break; 534 } 535 } 536 537 zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1); 538 539 zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time / zopt_vdevs : UINT64_MAX); 540 zopt_maxfaults = MAX(zopt_mirrors, 1) * (zopt_raidz_parity + 1) - 1; 541 } 542 543 static uint64_t 544 ztest_get_ashift(void) 545 { 546 if (zopt_ashift == 0) 547 return (SPA_MINBLOCKSHIFT + ztest_random(3)); 548 return (zopt_ashift); 549 } 550 551 static nvlist_t * 552 make_vdev_file(size_t size) 553 { 554 char dev_name[MAXPATHLEN]; 555 uint64_t vdev; 556 uint64_t ashift = ztest_get_ashift(); 557 int fd; 558 nvlist_t *file; 559 560 if (size == 0) { 561 (void) snprintf(dev_name, sizeof (dev_name), "%s", 562 "/dev/bogus"); 563 } else { 564 vdev = ztest_shared->zs_vdev_primaries++; 565 (void) sprintf(dev_name, ztest_dev_template, 566 zopt_dir, zopt_pool, vdev); 567 568 fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666); 569 if (fd == -1) 570 fatal(1, "can't open %s", dev_name); 571 if (ftruncate(fd, size) != 0) 572 fatal(1, "can't ftruncate %s", dev_name); 573 (void) close(fd); 574 } 575 576 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 577 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 578 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0); 579 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 580 581 return (file); 582 } 583 584 static nvlist_t * 585 make_vdev_raidz(size_t size, int r) 586 { 587 nvlist_t *raidz, **child; 588 int c; 589 590 if (r < 2) 591 return (make_vdev_file(size)); 592 593 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); 594 595 for (c = 0; c < r; c++) 596 child[c] = make_vdev_file(size); 597 598 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); 599 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, 600 VDEV_TYPE_RAIDZ) == 0); 601 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, 602 zopt_raidz_parity) == 0); 603 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, 604 child, r) == 0); 605 606 for (c = 0; c < r; c++) 607 nvlist_free(child[c]); 608 609 umem_free(child, r * sizeof (nvlist_t *)); 610 611 return (raidz); 612 } 613 614 static nvlist_t * 615 make_vdev_mirror(size_t size, int log, int r, int m) 616 { 617 nvlist_t *mirror, **child; 618 int c; 619 620 if (m < 1) 621 return (make_vdev_raidz(size, r)); 622 623 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); 624 625 for (c = 0; c < m; c++) 626 child[c] = make_vdev_raidz(size, r); 627 628 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); 629 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, 630 VDEV_TYPE_MIRROR) == 0); 631 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, 632 child, m) == 0); 633 VERIFY(nvlist_add_uint64(mirror, ZPOOL_CONFIG_IS_LOG, log) == 0); 634 635 for (c = 0; c < m; c++) 636 nvlist_free(child[c]); 637 638 umem_free(child, m * sizeof (nvlist_t *)); 639 640 return (mirror); 641 } 642 643 static nvlist_t * 644 make_vdev_root(size_t size, int log, int r, int m, int t) 645 { 646 nvlist_t *root, **child; 647 int c; 648 649 ASSERT(t > 0); 650 651 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); 652 653 for (c = 0; c < t; c++) 654 child[c] = make_vdev_mirror(size, log, r, m); 655 656 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 657 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 658 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 659 child, t) == 0); 660 661 for (c = 0; c < t; c++) 662 nvlist_free(child[c]); 663 664 umem_free(child, t * sizeof (nvlist_t *)); 665 666 return (root); 667 } 668 669 static void 670 ztest_set_random_blocksize(objset_t *os, uint64_t object, dmu_tx_t *tx) 671 { 672 int bs = SPA_MINBLOCKSHIFT + 673 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1); 674 int ibs = DN_MIN_INDBLKSHIFT + 675 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1); 676 int error; 677 678 error = dmu_object_set_blocksize(os, object, 1ULL << bs, ibs, tx); 679 if (error) { 680 char osname[300]; 681 dmu_objset_name(os, osname); 682 fatal(0, "dmu_object_set_blocksize('%s', %llu, %d, %d) = %d", 683 osname, object, 1 << bs, ibs, error); 684 } 685 } 686 687 static uint8_t 688 ztest_random_checksum(void) 689 { 690 uint8_t checksum; 691 692 do { 693 checksum = ztest_random(ZIO_CHECKSUM_FUNCTIONS); 694 } while (zio_checksum_table[checksum].ci_zbt); 695 696 if (checksum == ZIO_CHECKSUM_OFF) 697 checksum = ZIO_CHECKSUM_ON; 698 699 return (checksum); 700 } 701 702 static uint8_t 703 ztest_random_compress(void) 704 { 705 return ((uint8_t)ztest_random(ZIO_COMPRESS_FUNCTIONS)); 706 } 707 708 typedef struct ztest_replay { 709 objset_t *zr_os; 710 uint64_t zr_assign; 711 } ztest_replay_t; 712 713 static int 714 ztest_replay_create(ztest_replay_t *zr, lr_create_t *lr, boolean_t byteswap) 715 { 716 objset_t *os = zr->zr_os; 717 dmu_tx_t *tx; 718 int error; 719 720 if (byteswap) 721 byteswap_uint64_array(lr, sizeof (*lr)); 722 723 tx = dmu_tx_create(os); 724 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 725 error = dmu_tx_assign(tx, zr->zr_assign); 726 if (error) { 727 dmu_tx_abort(tx); 728 return (error); 729 } 730 731 error = dmu_object_claim(os, lr->lr_doid, lr->lr_mode, 0, 732 DMU_OT_NONE, 0, tx); 733 ASSERT3U(error, ==, 0); 734 dmu_tx_commit(tx); 735 736 if (zopt_verbose >= 5) { 737 char osname[MAXNAMELEN]; 738 dmu_objset_name(os, osname); 739 (void) printf("replay create of %s object %llu" 740 " in txg %llu = %d\n", 741 osname, (u_longlong_t)lr->lr_doid, 742 (u_longlong_t)zr->zr_assign, error); 743 } 744 745 return (error); 746 } 747 748 static int 749 ztest_replay_remove(ztest_replay_t *zr, lr_remove_t *lr, boolean_t byteswap) 750 { 751 objset_t *os = zr->zr_os; 752 dmu_tx_t *tx; 753 int error; 754 755 if (byteswap) 756 byteswap_uint64_array(lr, sizeof (*lr)); 757 758 tx = dmu_tx_create(os); 759 dmu_tx_hold_free(tx, lr->lr_doid, 0, DMU_OBJECT_END); 760 error = dmu_tx_assign(tx, zr->zr_assign); 761 if (error) { 762 dmu_tx_abort(tx); 763 return (error); 764 } 765 766 error = dmu_object_free(os, lr->lr_doid, tx); 767 dmu_tx_commit(tx); 768 769 return (error); 770 } 771 772 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { 773 NULL, /* 0 no such transaction type */ 774 ztest_replay_create, /* TX_CREATE */ 775 NULL, /* TX_MKDIR */ 776 NULL, /* TX_MKXATTR */ 777 NULL, /* TX_SYMLINK */ 778 ztest_replay_remove, /* TX_REMOVE */ 779 NULL, /* TX_RMDIR */ 780 NULL, /* TX_LINK */ 781 NULL, /* TX_RENAME */ 782 NULL, /* TX_WRITE */ 783 NULL, /* TX_TRUNCATE */ 784 NULL, /* TX_SETATTR */ 785 NULL, /* TX_ACL */ 786 }; 787 788 /* 789 * Verify that we can't destroy an active pool, create an existing pool, 790 * or create a pool with a bad vdev spec. 791 */ 792 void 793 ztest_spa_create_destroy(ztest_args_t *za) 794 { 795 int error; 796 spa_t *spa; 797 nvlist_t *nvroot; 798 799 /* 800 * Attempt to create using a bad file. 801 */ 802 nvroot = make_vdev_root(0, 0, 0, 0, 1); 803 error = spa_create("ztest_bad_file", nvroot, NULL, NULL); 804 nvlist_free(nvroot); 805 if (error != ENOENT) 806 fatal(0, "spa_create(bad_file) = %d", error); 807 808 /* 809 * Attempt to create using a bad mirror. 810 */ 811 nvroot = make_vdev_root(0, 0, 0, 2, 1); 812 error = spa_create("ztest_bad_mirror", nvroot, NULL, NULL); 813 nvlist_free(nvroot); 814 if (error != ENOENT) 815 fatal(0, "spa_create(bad_mirror) = %d", error); 816 817 /* 818 * Attempt to create an existing pool. It shouldn't matter 819 * what's in the nvroot; we should fail with EEXIST. 820 */ 821 (void) rw_rdlock(&ztest_shared->zs_name_lock); 822 nvroot = make_vdev_root(0, 0, 0, 0, 1); 823 error = spa_create(za->za_pool, nvroot, NULL, NULL); 824 nvlist_free(nvroot); 825 if (error != EEXIST) 826 fatal(0, "spa_create(whatever) = %d", error); 827 828 error = spa_open(za->za_pool, &spa, FTAG); 829 if (error) 830 fatal(0, "spa_open() = %d", error); 831 832 error = spa_destroy(za->za_pool); 833 if (error != EBUSY) 834 fatal(0, "spa_destroy() = %d", error); 835 836 spa_close(spa, FTAG); 837 (void) rw_unlock(&ztest_shared->zs_name_lock); 838 } 839 840 /* 841 * Verify that vdev_add() works as expected. 842 */ 843 void 844 ztest_vdev_add_remove(ztest_args_t *za) 845 { 846 spa_t *spa = za->za_spa; 847 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 848 nvlist_t *nvroot; 849 int error; 850 851 if (zopt_verbose >= 6) 852 (void) printf("adding vdev\n"); 853 854 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 855 856 spa_config_enter(spa, RW_READER, FTAG); 857 858 ztest_shared->zs_vdev_primaries = 859 spa->spa_root_vdev->vdev_children * leaves; 860 861 spa_config_exit(spa, FTAG); 862 863 /* 864 * Make 1/4 of the devices be log devices. 865 */ 866 nvroot = make_vdev_root(zopt_vdev_size, 867 ztest_random(4) == 0, zopt_raidz, zopt_mirrors, 1); 868 869 error = spa_vdev_add(spa, nvroot); 870 nvlist_free(nvroot); 871 872 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 873 874 if (error == ENOSPC) 875 ztest_record_enospc("spa_vdev_add"); 876 else if (error != 0) 877 fatal(0, "spa_vdev_add() = %d", error); 878 879 if (zopt_verbose >= 6) 880 (void) printf("spa_vdev_add = %d, as expected\n", error); 881 } 882 883 static vdev_t * 884 vdev_lookup_by_path(vdev_t *vd, const char *path) 885 { 886 int c; 887 vdev_t *mvd; 888 889 if (vd->vdev_path != NULL) { 890 if (vd->vdev_wholedisk == 1) { 891 /* 892 * For whole disks, the internal path has 's0', but the 893 * path passed in by the user doesn't. 894 */ 895 if (strlen(path) == strlen(vd->vdev_path) - 2 && 896 strncmp(path, vd->vdev_path, strlen(path)) == 0) 897 return (vd); 898 } else if (strcmp(path, vd->vdev_path) == 0) { 899 return (vd); 900 } 901 } 902 903 for (c = 0; c < vd->vdev_children; c++) 904 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != 905 NULL) 906 return (mvd); 907 908 return (NULL); 909 } 910 911 /* 912 * Verify that we can attach and detach devices. 913 */ 914 void 915 ztest_vdev_attach_detach(ztest_args_t *za) 916 { 917 spa_t *spa = za->za_spa; 918 vdev_t *rvd = spa->spa_root_vdev; 919 vdev_t *oldvd, *newvd, *pvd; 920 nvlist_t *root, *file; 921 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 922 uint64_t leaf, top; 923 uint64_t ashift = ztest_get_ashift(); 924 size_t oldsize, newsize; 925 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; 926 int replacing; 927 int error, expected_error; 928 int fd; 929 930 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 931 932 spa_config_enter(spa, RW_READER, FTAG); 933 934 /* 935 * Decide whether to do an attach or a replace. 936 */ 937 replacing = ztest_random(2); 938 939 /* 940 * Pick a random top-level vdev. 941 */ 942 top = ztest_random(rvd->vdev_children); 943 944 /* 945 * Pick a random leaf within it. 946 */ 947 leaf = ztest_random(leaves); 948 949 /* 950 * Generate the path to this leaf. The filename will end with 'a'. 951 * We'll alternate replacements with a filename that ends with 'b'. 952 */ 953 (void) snprintf(oldpath, sizeof (oldpath), 954 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf); 955 956 bcopy(oldpath, newpath, MAXPATHLEN); 957 958 /* 959 * If the 'a' file isn't part of the pool, the 'b' file must be. 960 */ 961 if (vdev_lookup_by_path(rvd, oldpath) == NULL) 962 oldpath[strlen(oldpath) - 1] = 'b'; 963 else 964 newpath[strlen(newpath) - 1] = 'b'; 965 966 /* 967 * Now oldpath represents something that's already in the pool, 968 * and newpath is the thing we'll try to attach. 969 */ 970 oldvd = vdev_lookup_by_path(rvd, oldpath); 971 newvd = vdev_lookup_by_path(rvd, newpath); 972 ASSERT(oldvd != NULL); 973 pvd = oldvd->vdev_parent; 974 975 /* 976 * Make newsize a little bigger or smaller than oldsize. 977 * If it's smaller, the attach should fail. 978 * If it's larger, and we're doing a replace, 979 * we should get dynamic LUN growth when we're done. 980 */ 981 oldsize = vdev_get_rsize(oldvd); 982 newsize = 10 * oldsize / (9 + ztest_random(3)); 983 984 /* 985 * If pvd is not a mirror or root, the attach should fail with ENOTSUP, 986 * unless it's a replace; in that case any non-replacing parent is OK. 987 * 988 * If newvd is already part of the pool, it should fail with EBUSY. 989 * 990 * If newvd is too small, it should fail with EOVERFLOW. 991 */ 992 if (newvd != NULL) 993 expected_error = EBUSY; 994 else if (pvd->vdev_ops != &vdev_mirror_ops && 995 pvd->vdev_ops != &vdev_root_ops && 996 (!replacing || pvd->vdev_ops == &vdev_replacing_ops)) 997 expected_error = ENOTSUP; 998 else if (newsize < oldsize) 999 expected_error = EOVERFLOW; 1000 else if (ashift > oldvd->vdev_top->vdev_ashift) 1001 expected_error = EDOM; 1002 else 1003 expected_error = 0; 1004 1005 /* 1006 * If newvd isn't already part of the pool, create it. 1007 */ 1008 if (newvd == NULL) { 1009 fd = open(newpath, O_RDWR | O_CREAT | O_TRUNC, 0666); 1010 if (fd == -1) 1011 fatal(1, "can't open %s", newpath); 1012 if (ftruncate(fd, newsize) != 0) 1013 fatal(1, "can't ftruncate %s", newpath); 1014 (void) close(fd); 1015 } 1016 1017 spa_config_exit(spa, FTAG); 1018 1019 /* 1020 * Build the nvlist describing newpath. 1021 */ 1022 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 1023 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 1024 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, newpath) == 0); 1025 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 1026 1027 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 1028 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 1029 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 1030 &file, 1) == 0); 1031 1032 error = spa_vdev_attach(spa, oldvd->vdev_guid, root, replacing); 1033 1034 nvlist_free(file); 1035 nvlist_free(root); 1036 1037 /* 1038 * If our parent was the replacing vdev, but the replace completed, 1039 * then instead of failing with ENOTSUP we may either succeed, 1040 * fail with ENODEV, or fail with EOVERFLOW. 1041 */ 1042 if (expected_error == ENOTSUP && 1043 (error == 0 || error == ENODEV || error == EOVERFLOW)) 1044 expected_error = error; 1045 1046 /* 1047 * If someone grew the LUN, the replacement may be too small. 1048 */ 1049 if (error == EOVERFLOW) 1050 expected_error = error; 1051 1052 if (error != expected_error) { 1053 fatal(0, "attach (%s, %s, %d) returned %d, expected %d", 1054 oldpath, newpath, replacing, error, expected_error); 1055 } 1056 1057 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 1058 } 1059 1060 /* 1061 * Verify that dynamic LUN growth works as expected. 1062 */ 1063 /* ARGSUSED */ 1064 void 1065 ztest_vdev_LUN_growth(ztest_args_t *za) 1066 { 1067 spa_t *spa = za->za_spa; 1068 char dev_name[MAXPATHLEN]; 1069 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 1070 uint64_t vdev; 1071 size_t fsize; 1072 int fd; 1073 1074 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 1075 1076 /* 1077 * Pick a random leaf vdev. 1078 */ 1079 spa_config_enter(spa, RW_READER, FTAG); 1080 vdev = ztest_random(spa->spa_root_vdev->vdev_children * leaves); 1081 spa_config_exit(spa, FTAG); 1082 1083 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 1084 1085 if ((fd = open(dev_name, O_RDWR)) != -1) { 1086 /* 1087 * Determine the size. 1088 */ 1089 fsize = lseek(fd, 0, SEEK_END); 1090 1091 /* 1092 * If it's less than 2x the original size, grow by around 3%. 1093 */ 1094 if (fsize < 2 * zopt_vdev_size) { 1095 size_t newsize = fsize + ztest_random(fsize / 32); 1096 (void) ftruncate(fd, newsize); 1097 if (zopt_verbose >= 6) { 1098 (void) printf("%s grew from %lu to %lu bytes\n", 1099 dev_name, (ulong_t)fsize, (ulong_t)newsize); 1100 } 1101 } 1102 (void) close(fd); 1103 } 1104 1105 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 1106 } 1107 1108 /* ARGSUSED */ 1109 static void 1110 ztest_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 1111 { 1112 /* 1113 * Create the directory object. 1114 */ 1115 VERIFY(dmu_object_claim(os, ZTEST_DIROBJ, 1116 DMU_OT_UINT64_OTHER, ZTEST_DIROBJ_BLOCKSIZE, 1117 DMU_OT_UINT64_OTHER, 5 * sizeof (ztest_block_tag_t), tx) == 0); 1118 1119 VERIFY(zap_create_claim(os, ZTEST_MICROZAP_OBJ, 1120 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 1121 1122 VERIFY(zap_create_claim(os, ZTEST_FATZAP_OBJ, 1123 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 1124 } 1125 1126 static int 1127 ztest_destroy_cb(char *name, void *arg) 1128 { 1129 ztest_args_t *za = arg; 1130 objset_t *os; 1131 dmu_object_info_t *doi = &za->za_doi; 1132 int error; 1133 1134 /* 1135 * Verify that the dataset contains a directory object. 1136 */ 1137 error = dmu_objset_open(name, DMU_OST_OTHER, 1138 DS_MODE_USER | DS_MODE_READONLY, &os); 1139 ASSERT3U(error, ==, 0); 1140 error = dmu_object_info(os, ZTEST_DIROBJ, doi); 1141 if (error != ENOENT) { 1142 /* We could have crashed in the middle of destroying it */ 1143 ASSERT3U(error, ==, 0); 1144 ASSERT3U(doi->doi_type, ==, DMU_OT_UINT64_OTHER); 1145 ASSERT3S(doi->doi_physical_blks, >=, 0); 1146 } 1147 dmu_objset_close(os); 1148 1149 /* 1150 * Destroy the dataset. 1151 */ 1152 error = dmu_objset_destroy(name); 1153 if (error) { 1154 (void) dmu_objset_open(name, DMU_OST_OTHER, 1155 DS_MODE_USER | DS_MODE_READONLY, &os); 1156 fatal(0, "dmu_objset_destroy(os=%p) = %d\n", &os, error); 1157 } 1158 return (0); 1159 } 1160 1161 /* 1162 * Verify that dmu_objset_{create,destroy,open,close} work as expected. 1163 */ 1164 static uint64_t 1165 ztest_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t object, int mode) 1166 { 1167 itx_t *itx; 1168 lr_create_t *lr; 1169 size_t namesize; 1170 char name[24]; 1171 1172 (void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object); 1173 namesize = strlen(name) + 1; 1174 1175 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize + 1176 ztest_random(ZIL_MAX_BLKSZ)); 1177 lr = (lr_create_t *)&itx->itx_lr; 1178 bzero(lr + 1, lr->lr_common.lrc_reclen - sizeof (*lr)); 1179 lr->lr_doid = object; 1180 lr->lr_foid = 0; 1181 lr->lr_mode = mode; 1182 lr->lr_uid = 0; 1183 lr->lr_gid = 0; 1184 lr->lr_gen = dmu_tx_get_txg(tx); 1185 lr->lr_crtime[0] = time(NULL); 1186 lr->lr_crtime[1] = 0; 1187 lr->lr_rdev = 0; 1188 bcopy(name, (char *)(lr + 1), namesize); 1189 1190 return (zil_itx_assign(zilog, itx, tx)); 1191 } 1192 1193 void 1194 ztest_dmu_objset_create_destroy(ztest_args_t *za) 1195 { 1196 int error; 1197 objset_t *os, *os2; 1198 char name[100]; 1199 int basemode, expected_error; 1200 zilog_t *zilog; 1201 uint64_t seq; 1202 uint64_t objects; 1203 ztest_replay_t zr; 1204 1205 (void) rw_rdlock(&ztest_shared->zs_name_lock); 1206 (void) snprintf(name, 100, "%s/%s_temp_%llu", za->za_pool, za->za_pool, 1207 (u_longlong_t)za->za_instance); 1208 1209 basemode = DS_MODE_TYPE(za->za_instance); 1210 if (basemode != DS_MODE_USER && basemode != DS_MODE_OWNER) 1211 basemode = DS_MODE_USER; 1212 1213 /* 1214 * If this dataset exists from a previous run, process its replay log 1215 * half of the time. If we don't replay it, then dmu_objset_destroy() 1216 * (invoked from ztest_destroy_cb() below) should just throw it away. 1217 */ 1218 if (ztest_random(2) == 0 && 1219 dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_OWNER, &os) == 0) { 1220 zr.zr_os = os; 1221 zil_replay(os, &zr, &zr.zr_assign, ztest_replay_vector); 1222 dmu_objset_close(os); 1223 } 1224 1225 /* 1226 * There may be an old instance of the dataset we're about to 1227 * create lying around from a previous run. If so, destroy it 1228 * and all of its snapshots. 1229 */ 1230 (void) dmu_objset_find(name, ztest_destroy_cb, za, 1231 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 1232 1233 /* 1234 * Verify that the destroyed dataset is no longer in the namespace. 1235 */ 1236 error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os); 1237 if (error != ENOENT) 1238 fatal(1, "dmu_objset_open(%s) found destroyed dataset %p", 1239 name, os); 1240 1241 /* 1242 * Verify that we can create a new dataset. 1243 */ 1244 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0, 1245 ztest_create_cb, NULL); 1246 if (error) { 1247 if (error == ENOSPC) { 1248 ztest_record_enospc("dmu_objset_create"); 1249 (void) rw_unlock(&ztest_shared->zs_name_lock); 1250 return; 1251 } 1252 fatal(0, "dmu_objset_create(%s) = %d", name, error); 1253 } 1254 1255 error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os); 1256 if (error) { 1257 fatal(0, "dmu_objset_open(%s) = %d", name, error); 1258 } 1259 1260 /* 1261 * Open the intent log for it. 1262 */ 1263 zilog = zil_open(os, NULL); 1264 1265 /* 1266 * Put a random number of objects in there. 1267 */ 1268 objects = ztest_random(20); 1269 seq = 0; 1270 while (objects-- != 0) { 1271 uint64_t object; 1272 dmu_tx_t *tx = dmu_tx_create(os); 1273 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, sizeof (name)); 1274 error = dmu_tx_assign(tx, TXG_WAIT); 1275 if (error) { 1276 dmu_tx_abort(tx); 1277 } else { 1278 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1279 DMU_OT_NONE, 0, tx); 1280 ztest_set_random_blocksize(os, object, tx); 1281 seq = ztest_log_create(zilog, tx, object, 1282 DMU_OT_UINT64_OTHER); 1283 dmu_write(os, object, 0, sizeof (name), name, tx); 1284 dmu_tx_commit(tx); 1285 } 1286 if (ztest_random(5) == 0) { 1287 zil_commit(zilog, seq, object); 1288 } 1289 if (ztest_random(100) == 0) { 1290 error = zil_suspend(zilog); 1291 if (error == 0) { 1292 zil_resume(zilog); 1293 } 1294 } 1295 } 1296 1297 /* 1298 * Verify that we cannot create an existing dataset. 1299 */ 1300 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0, NULL, NULL); 1301 if (error != EEXIST) 1302 fatal(0, "created existing dataset, error = %d", error); 1303 1304 /* 1305 * Verify that multiple dataset holds are allowed, but only when 1306 * the new access mode is compatible with the base mode. 1307 */ 1308 if (basemode == DS_MODE_OWNER) { 1309 error = dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_USER, 1310 &os2); 1311 if (error) 1312 fatal(0, "dmu_objset_open('%s') = %d", name, error); 1313 else 1314 dmu_objset_close(os2); 1315 } 1316 error = dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_OWNER, &os2); 1317 expected_error = (basemode == DS_MODE_OWNER) ? EBUSY : 0; 1318 if (error != expected_error) 1319 fatal(0, "dmu_objset_open('%s') = %d, expected %d", 1320 name, error, expected_error); 1321 if (error == 0) 1322 dmu_objset_close(os2); 1323 1324 zil_close(zilog); 1325 dmu_objset_close(os); 1326 1327 error = dmu_objset_destroy(name); 1328 if (error) 1329 fatal(0, "dmu_objset_destroy(%s) = %d", name, error); 1330 1331 (void) rw_unlock(&ztest_shared->zs_name_lock); 1332 } 1333 1334 /* 1335 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. 1336 */ 1337 void 1338 ztest_dmu_snapshot_create_destroy(ztest_args_t *za) 1339 { 1340 int error; 1341 objset_t *os = za->za_os; 1342 char snapname[100]; 1343 char osname[MAXNAMELEN]; 1344 1345 (void) rw_rdlock(&ztest_shared->zs_name_lock); 1346 dmu_objset_name(os, osname); 1347 (void) snprintf(snapname, 100, "%s@%llu", osname, 1348 (u_longlong_t)za->za_instance); 1349 1350 error = dmu_objset_destroy(snapname); 1351 if (error != 0 && error != ENOENT) 1352 fatal(0, "dmu_objset_destroy() = %d", error); 1353 error = dmu_objset_snapshot(osname, strchr(snapname, '@')+1, FALSE); 1354 if (error == ENOSPC) 1355 ztest_record_enospc("dmu_take_snapshot"); 1356 else if (error != 0 && error != EEXIST) 1357 fatal(0, "dmu_take_snapshot() = %d", error); 1358 (void) rw_unlock(&ztest_shared->zs_name_lock); 1359 } 1360 1361 #define ZTEST_TRAVERSE_BLOCKS 1000 1362 1363 static int 1364 ztest_blk_cb(traverse_blk_cache_t *bc, spa_t *spa, void *arg) 1365 { 1366 ztest_args_t *za = arg; 1367 zbookmark_t *zb = &bc->bc_bookmark; 1368 blkptr_t *bp = &bc->bc_blkptr; 1369 dnode_phys_t *dnp = bc->bc_dnode; 1370 traverse_handle_t *th = za->za_th; 1371 uint64_t size = BP_GET_LSIZE(bp); 1372 1373 /* 1374 * Level -1 indicates the objset_phys_t or something in its intent log. 1375 */ 1376 if (zb->zb_level == -1) { 1377 if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { 1378 ASSERT3U(zb->zb_object, ==, 0); 1379 ASSERT3U(zb->zb_blkid, ==, 0); 1380 ASSERT3U(size, ==, sizeof (objset_phys_t)); 1381 za->za_zil_seq = 0; 1382 } else if (BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) { 1383 ASSERT3U(zb->zb_object, ==, 0); 1384 ASSERT3U(zb->zb_blkid, >, za->za_zil_seq); 1385 za->za_zil_seq = zb->zb_blkid; 1386 } else { 1387 ASSERT3U(zb->zb_object, !=, 0); /* lr_write_t */ 1388 } 1389 1390 return (0); 1391 } 1392 1393 ASSERT(dnp != NULL); 1394 1395 if (bc->bc_errno) 1396 return (ERESTART); 1397 1398 /* 1399 * Once in a while, abort the traverse. We only do this to odd 1400 * instance numbers to ensure that even ones can run to completion. 1401 */ 1402 if ((za->za_instance & 1) && ztest_random(10000) == 0) 1403 return (EINTR); 1404 1405 if (bp->blk_birth == 0) { 1406 ASSERT(th->th_advance & ADVANCE_HOLES); 1407 return (0); 1408 } 1409 1410 if (zb->zb_level == 0 && !(th->th_advance & ADVANCE_DATA) && 1411 bc == &th->th_cache[ZB_DN_CACHE][0]) { 1412 ASSERT(bc->bc_data == NULL); 1413 return (0); 1414 } 1415 1416 ASSERT(bc->bc_data != NULL); 1417 1418 /* 1419 * This is an expensive question, so don't ask it too often. 1420 */ 1421 if (((za->za_random ^ th->th_callbacks) & 0xff) == 0) { 1422 void *xbuf = umem_alloc(size, UMEM_NOFAIL); 1423 if (arc_tryread(spa, bp, xbuf) == 0) { 1424 ASSERT(bcmp(bc->bc_data, xbuf, size) == 0); 1425 } 1426 umem_free(xbuf, size); 1427 } 1428 1429 if (zb->zb_level > 0) { 1430 ASSERT3U(size, ==, 1ULL << dnp->dn_indblkshift); 1431 return (0); 1432 } 1433 1434 ASSERT(zb->zb_level == 0); 1435 ASSERT3U(size, ==, dnp->dn_datablkszsec << DEV_BSHIFT); 1436 1437 return (0); 1438 } 1439 1440 /* 1441 * Verify that live pool traversal works. 1442 */ 1443 void 1444 ztest_traverse(ztest_args_t *za) 1445 { 1446 spa_t *spa = za->za_spa; 1447 traverse_handle_t *th = za->za_th; 1448 int rc, advance; 1449 uint64_t cbstart, cblimit; 1450 1451 if (th == NULL) { 1452 advance = 0; 1453 1454 if (ztest_random(2) == 0) 1455 advance |= ADVANCE_PRE; 1456 1457 if (ztest_random(2) == 0) 1458 advance |= ADVANCE_PRUNE; 1459 1460 if (ztest_random(2) == 0) 1461 advance |= ADVANCE_DATA; 1462 1463 if (ztest_random(2) == 0) 1464 advance |= ADVANCE_HOLES; 1465 1466 if (ztest_random(2) == 0) 1467 advance |= ADVANCE_ZIL; 1468 1469 th = za->za_th = traverse_init(spa, ztest_blk_cb, za, advance, 1470 ZIO_FLAG_CANFAIL); 1471 1472 traverse_add_pool(th, 0, -1ULL); 1473 } 1474 1475 advance = th->th_advance; 1476 cbstart = th->th_callbacks; 1477 cblimit = cbstart + ((advance & ADVANCE_DATA) ? 100 : 1000); 1478 1479 while ((rc = traverse_more(th)) == EAGAIN && th->th_callbacks < cblimit) 1480 continue; 1481 1482 if (zopt_verbose >= 5) 1483 (void) printf("traverse %s%s%s%s %llu blocks to " 1484 "<%llu, %llu, %lld, %llx>%s\n", 1485 (advance & ADVANCE_PRE) ? "pre" : "post", 1486 (advance & ADVANCE_PRUNE) ? "|prune" : "", 1487 (advance & ADVANCE_DATA) ? "|data" : "", 1488 (advance & ADVANCE_HOLES) ? "|holes" : "", 1489 (u_longlong_t)(th->th_callbacks - cbstart), 1490 (u_longlong_t)th->th_lastcb.zb_objset, 1491 (u_longlong_t)th->th_lastcb.zb_object, 1492 (u_longlong_t)th->th_lastcb.zb_level, 1493 (u_longlong_t)th->th_lastcb.zb_blkid, 1494 rc == 0 ? " [done]" : 1495 rc == EINTR ? " [aborted]" : 1496 rc == EAGAIN ? "" : 1497 strerror(rc)); 1498 1499 if (rc != EAGAIN) { 1500 if (rc != 0 && rc != EINTR) 1501 fatal(0, "traverse_more(%p) = %d", th, rc); 1502 traverse_fini(th); 1503 za->za_th = NULL; 1504 } 1505 } 1506 1507 /* 1508 * Verify that dmu_object_{alloc,free} work as expected. 1509 */ 1510 void 1511 ztest_dmu_object_alloc_free(ztest_args_t *za) 1512 { 1513 objset_t *os = za->za_os; 1514 dmu_buf_t *db; 1515 dmu_tx_t *tx; 1516 uint64_t batchobj, object, batchsize, endoff, temp; 1517 int b, c, error, bonuslen; 1518 dmu_object_info_t *doi = &za->za_doi; 1519 char osname[MAXNAMELEN]; 1520 1521 dmu_objset_name(os, osname); 1522 1523 endoff = -8ULL; 1524 batchsize = 2; 1525 1526 /* 1527 * Create a batch object if necessary, and record it in the directory. 1528 */ 1529 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 1530 sizeof (uint64_t), &batchobj)); 1531 if (batchobj == 0) { 1532 tx = dmu_tx_create(os); 1533 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, 1534 sizeof (uint64_t)); 1535 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1536 error = dmu_tx_assign(tx, TXG_WAIT); 1537 if (error) { 1538 ztest_record_enospc("create a batch object"); 1539 dmu_tx_abort(tx); 1540 return; 1541 } 1542 batchobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1543 DMU_OT_NONE, 0, tx); 1544 ztest_set_random_blocksize(os, batchobj, tx); 1545 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, 1546 sizeof (uint64_t), &batchobj, tx); 1547 dmu_tx_commit(tx); 1548 } 1549 1550 /* 1551 * Destroy the previous batch of objects. 1552 */ 1553 for (b = 0; b < batchsize; b++) { 1554 VERIFY(0 == dmu_read(os, batchobj, b * sizeof (uint64_t), 1555 sizeof (uint64_t), &object)); 1556 if (object == 0) 1557 continue; 1558 /* 1559 * Read and validate contents. 1560 * We expect the nth byte of the bonus buffer to be n. 1561 */ 1562 VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db)); 1563 za->za_dbuf = db; 1564 1565 dmu_object_info_from_db(db, doi); 1566 ASSERT(doi->doi_type == DMU_OT_UINT64_OTHER); 1567 ASSERT(doi->doi_bonus_type == DMU_OT_PLAIN_OTHER); 1568 ASSERT3S(doi->doi_physical_blks, >=, 0); 1569 1570 bonuslen = doi->doi_bonus_size; 1571 1572 for (c = 0; c < bonuslen; c++) { 1573 if (((uint8_t *)db->db_data)[c] != 1574 (uint8_t)(c + bonuslen)) { 1575 fatal(0, 1576 "bad bonus: %s, obj %llu, off %d: %u != %u", 1577 osname, object, c, 1578 ((uint8_t *)db->db_data)[c], 1579 (uint8_t)(c + bonuslen)); 1580 } 1581 } 1582 1583 dmu_buf_rele(db, FTAG); 1584 za->za_dbuf = NULL; 1585 1586 /* 1587 * We expect the word at endoff to be our object number. 1588 */ 1589 VERIFY(0 == dmu_read(os, object, endoff, 1590 sizeof (uint64_t), &temp)); 1591 1592 if (temp != object) { 1593 fatal(0, "bad data in %s, got %llu, expected %llu", 1594 osname, temp, object); 1595 } 1596 1597 /* 1598 * Destroy old object and clear batch entry. 1599 */ 1600 tx = dmu_tx_create(os); 1601 dmu_tx_hold_write(tx, batchobj, 1602 b * sizeof (uint64_t), sizeof (uint64_t)); 1603 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 1604 error = dmu_tx_assign(tx, TXG_WAIT); 1605 if (error) { 1606 ztest_record_enospc("free object"); 1607 dmu_tx_abort(tx); 1608 return; 1609 } 1610 error = dmu_object_free(os, object, tx); 1611 if (error) { 1612 fatal(0, "dmu_object_free('%s', %llu) = %d", 1613 osname, object, error); 1614 } 1615 object = 0; 1616 1617 dmu_object_set_checksum(os, batchobj, 1618 ztest_random_checksum(), tx); 1619 dmu_object_set_compress(os, batchobj, 1620 ztest_random_compress(), tx); 1621 1622 dmu_write(os, batchobj, b * sizeof (uint64_t), 1623 sizeof (uint64_t), &object, tx); 1624 1625 dmu_tx_commit(tx); 1626 } 1627 1628 /* 1629 * Before creating the new batch of objects, generate a bunch of churn. 1630 */ 1631 for (b = ztest_random(100); b > 0; b--) { 1632 tx = dmu_tx_create(os); 1633 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1634 error = dmu_tx_assign(tx, TXG_WAIT); 1635 if (error) { 1636 ztest_record_enospc("churn objects"); 1637 dmu_tx_abort(tx); 1638 return; 1639 } 1640 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1641 DMU_OT_NONE, 0, tx); 1642 ztest_set_random_blocksize(os, object, tx); 1643 error = dmu_object_free(os, object, tx); 1644 if (error) { 1645 fatal(0, "dmu_object_free('%s', %llu) = %d", 1646 osname, object, error); 1647 } 1648 dmu_tx_commit(tx); 1649 } 1650 1651 /* 1652 * Create a new batch of objects with randomly chosen 1653 * blocksizes and record them in the batch directory. 1654 */ 1655 for (b = 0; b < batchsize; b++) { 1656 uint32_t va_blksize; 1657 u_longlong_t va_nblocks; 1658 1659 tx = dmu_tx_create(os); 1660 dmu_tx_hold_write(tx, batchobj, b * sizeof (uint64_t), 1661 sizeof (uint64_t)); 1662 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1663 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, endoff, 1664 sizeof (uint64_t)); 1665 error = dmu_tx_assign(tx, TXG_WAIT); 1666 if (error) { 1667 ztest_record_enospc("create batchobj"); 1668 dmu_tx_abort(tx); 1669 return; 1670 } 1671 bonuslen = (int)ztest_random(dmu_bonus_max()) + 1; 1672 1673 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1674 DMU_OT_PLAIN_OTHER, bonuslen, tx); 1675 1676 ztest_set_random_blocksize(os, object, tx); 1677 1678 dmu_object_set_checksum(os, object, 1679 ztest_random_checksum(), tx); 1680 dmu_object_set_compress(os, object, 1681 ztest_random_compress(), tx); 1682 1683 dmu_write(os, batchobj, b * sizeof (uint64_t), 1684 sizeof (uint64_t), &object, tx); 1685 1686 /* 1687 * Write to both the bonus buffer and the regular data. 1688 */ 1689 VERIFY(dmu_bonus_hold(os, object, FTAG, &db) == 0); 1690 za->za_dbuf = db; 1691 ASSERT3U(bonuslen, <=, db->db_size); 1692 1693 dmu_object_size_from_db(db, &va_blksize, &va_nblocks); 1694 ASSERT3S(va_nblocks, >=, 0); 1695 1696 dmu_buf_will_dirty(db, tx); 1697 1698 /* 1699 * See comments above regarding the contents of 1700 * the bonus buffer and the word at endoff. 1701 */ 1702 for (c = 0; c < bonuslen; c++) 1703 ((uint8_t *)db->db_data)[c] = (uint8_t)(c + bonuslen); 1704 1705 dmu_buf_rele(db, FTAG); 1706 za->za_dbuf = NULL; 1707 1708 /* 1709 * Write to a large offset to increase indirection. 1710 */ 1711 dmu_write(os, object, endoff, sizeof (uint64_t), &object, tx); 1712 1713 dmu_tx_commit(tx); 1714 } 1715 } 1716 1717 /* 1718 * Verify that dmu_{read,write} work as expected. 1719 */ 1720 typedef struct bufwad { 1721 uint64_t bw_index; 1722 uint64_t bw_txg; 1723 uint64_t bw_data; 1724 } bufwad_t; 1725 1726 typedef struct dmu_read_write_dir { 1727 uint64_t dd_packobj; 1728 uint64_t dd_bigobj; 1729 uint64_t dd_chunk; 1730 } dmu_read_write_dir_t; 1731 1732 void 1733 ztest_dmu_read_write(ztest_args_t *za) 1734 { 1735 objset_t *os = za->za_os; 1736 dmu_read_write_dir_t dd; 1737 dmu_tx_t *tx; 1738 int i, freeit, error; 1739 uint64_t n, s, txg; 1740 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; 1741 uint64_t packoff, packsize, bigoff, bigsize; 1742 uint64_t regions = 997; 1743 uint64_t stride = 123456789ULL; 1744 uint64_t width = 40; 1745 int free_percent = 5; 1746 1747 /* 1748 * This test uses two objects, packobj and bigobj, that are always 1749 * updated together (i.e. in the same tx) so that their contents are 1750 * in sync and can be compared. Their contents relate to each other 1751 * in a simple way: packobj is a dense array of 'bufwad' structures, 1752 * while bigobj is a sparse array of the same bufwads. Specifically, 1753 * for any index n, there are three bufwads that should be identical: 1754 * 1755 * packobj, at offset n * sizeof (bufwad_t) 1756 * bigobj, at the head of the nth chunk 1757 * bigobj, at the tail of the nth chunk 1758 * 1759 * The chunk size is arbitrary. It doesn't have to be a power of two, 1760 * and it doesn't have any relation to the object blocksize. 1761 * The only requirement is that it can hold at least two bufwads. 1762 * 1763 * Normally, we write the bufwad to each of these locations. 1764 * However, free_percent of the time we instead write zeroes to 1765 * packobj and perform a dmu_free_range() on bigobj. By comparing 1766 * bigobj to packobj, we can verify that the DMU is correctly 1767 * tracking which parts of an object are allocated and free, 1768 * and that the contents of the allocated blocks are correct. 1769 */ 1770 1771 /* 1772 * Read the directory info. If it's the first time, set things up. 1773 */ 1774 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 1775 sizeof (dd), &dd)); 1776 if (dd.dd_chunk == 0) { 1777 ASSERT(dd.dd_packobj == 0); 1778 ASSERT(dd.dd_bigobj == 0); 1779 tx = dmu_tx_create(os); 1780 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (dd)); 1781 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1782 error = dmu_tx_assign(tx, TXG_WAIT); 1783 if (error) { 1784 ztest_record_enospc("create r/w directory"); 1785 dmu_tx_abort(tx); 1786 return; 1787 } 1788 1789 dd.dd_packobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1790 DMU_OT_NONE, 0, tx); 1791 dd.dd_bigobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1792 DMU_OT_NONE, 0, tx); 1793 dd.dd_chunk = (1000 + ztest_random(1000)) * sizeof (uint64_t); 1794 1795 ztest_set_random_blocksize(os, dd.dd_packobj, tx); 1796 ztest_set_random_blocksize(os, dd.dd_bigobj, tx); 1797 1798 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (dd), &dd, 1799 tx); 1800 dmu_tx_commit(tx); 1801 } 1802 1803 /* 1804 * Prefetch a random chunk of the big object. 1805 * Our aim here is to get some async reads in flight 1806 * for blocks that we may free below; the DMU should 1807 * handle this race correctly. 1808 */ 1809 n = ztest_random(regions) * stride + ztest_random(width); 1810 s = 1 + ztest_random(2 * width - 1); 1811 dmu_prefetch(os, dd.dd_bigobj, n * dd.dd_chunk, s * dd.dd_chunk); 1812 1813 /* 1814 * Pick a random index and compute the offsets into packobj and bigobj. 1815 */ 1816 n = ztest_random(regions) * stride + ztest_random(width); 1817 s = 1 + ztest_random(width - 1); 1818 1819 packoff = n * sizeof (bufwad_t); 1820 packsize = s * sizeof (bufwad_t); 1821 1822 bigoff = n * dd.dd_chunk; 1823 bigsize = s * dd.dd_chunk; 1824 1825 packbuf = umem_alloc(packsize, UMEM_NOFAIL); 1826 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); 1827 1828 /* 1829 * free_percent of the time, free a range of bigobj rather than 1830 * overwriting it. 1831 */ 1832 freeit = (ztest_random(100) < free_percent); 1833 1834 /* 1835 * Read the current contents of our objects. 1836 */ 1837 error = dmu_read(os, dd.dd_packobj, packoff, packsize, packbuf); 1838 ASSERT3U(error, ==, 0); 1839 error = dmu_read(os, dd.dd_bigobj, bigoff, bigsize, bigbuf); 1840 ASSERT3U(error, ==, 0); 1841 1842 /* 1843 * Get a tx for the mods to both packobj and bigobj. 1844 */ 1845 tx = dmu_tx_create(os); 1846 1847 dmu_tx_hold_write(tx, dd.dd_packobj, packoff, packsize); 1848 1849 if (freeit) 1850 dmu_tx_hold_free(tx, dd.dd_bigobj, bigoff, bigsize); 1851 else 1852 dmu_tx_hold_write(tx, dd.dd_bigobj, bigoff, bigsize); 1853 1854 error = dmu_tx_assign(tx, TXG_WAIT); 1855 1856 if (error) { 1857 ztest_record_enospc("dmu r/w range"); 1858 dmu_tx_abort(tx); 1859 umem_free(packbuf, packsize); 1860 umem_free(bigbuf, bigsize); 1861 return; 1862 } 1863 1864 txg = dmu_tx_get_txg(tx); 1865 1866 /* 1867 * For each index from n to n + s, verify that the existing bufwad 1868 * in packobj matches the bufwads at the head and tail of the 1869 * corresponding chunk in bigobj. Then update all three bufwads 1870 * with the new values we want to write out. 1871 */ 1872 for (i = 0; i < s; i++) { 1873 /* LINTED */ 1874 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 1875 /* LINTED */ 1876 bigH = (bufwad_t *)((char *)bigbuf + i * dd.dd_chunk); 1877 /* LINTED */ 1878 bigT = (bufwad_t *)((char *)bigH + dd.dd_chunk) - 1; 1879 1880 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 1881 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 1882 1883 if (pack->bw_txg > txg) 1884 fatal(0, "future leak: got %llx, open txg is %llx", 1885 pack->bw_txg, txg); 1886 1887 if (pack->bw_data != 0 && pack->bw_index != n + i) 1888 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 1889 pack->bw_index, n, i); 1890 1891 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 1892 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 1893 1894 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 1895 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 1896 1897 if (freeit) { 1898 bzero(pack, sizeof (bufwad_t)); 1899 } else { 1900 pack->bw_index = n + i; 1901 pack->bw_txg = txg; 1902 pack->bw_data = 1 + ztest_random(-2ULL); 1903 } 1904 *bigH = *pack; 1905 *bigT = *pack; 1906 } 1907 1908 /* 1909 * We've verified all the old bufwads, and made new ones. 1910 * Now write them out. 1911 */ 1912 dmu_write(os, dd.dd_packobj, packoff, packsize, packbuf, tx); 1913 1914 if (freeit) { 1915 if (zopt_verbose >= 6) { 1916 (void) printf("freeing offset %llx size %llx" 1917 " txg %llx\n", 1918 (u_longlong_t)bigoff, 1919 (u_longlong_t)bigsize, 1920 (u_longlong_t)txg); 1921 } 1922 VERIFY(0 == dmu_free_range(os, dd.dd_bigobj, bigoff, 1923 bigsize, tx)); 1924 } else { 1925 if (zopt_verbose >= 6) { 1926 (void) printf("writing offset %llx size %llx" 1927 " txg %llx\n", 1928 (u_longlong_t)bigoff, 1929 (u_longlong_t)bigsize, 1930 (u_longlong_t)txg); 1931 } 1932 dmu_write(os, dd.dd_bigobj, bigoff, bigsize, bigbuf, tx); 1933 } 1934 1935 dmu_tx_commit(tx); 1936 1937 /* 1938 * Sanity check the stuff we just wrote. 1939 */ 1940 { 1941 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 1942 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 1943 1944 VERIFY(0 == dmu_read(os, dd.dd_packobj, packoff, 1945 packsize, packcheck)); 1946 VERIFY(0 == dmu_read(os, dd.dd_bigobj, bigoff, 1947 bigsize, bigcheck)); 1948 1949 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 1950 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 1951 1952 umem_free(packcheck, packsize); 1953 umem_free(bigcheck, bigsize); 1954 } 1955 1956 umem_free(packbuf, packsize); 1957 umem_free(bigbuf, bigsize); 1958 } 1959 1960 void 1961 ztest_dmu_check_future_leak(ztest_args_t *za) 1962 { 1963 objset_t *os = za->za_os; 1964 dmu_buf_t *db; 1965 ztest_block_tag_t *bt; 1966 dmu_object_info_t *doi = &za->za_doi; 1967 1968 /* 1969 * Make sure that, if there is a write record in the bonus buffer 1970 * of the ZTEST_DIROBJ, that the txg for this record is <= the 1971 * last synced txg of the pool. 1972 */ 1973 VERIFY(dmu_bonus_hold(os, ZTEST_DIROBJ, FTAG, &db) == 0); 1974 za->za_dbuf = db; 1975 VERIFY(dmu_object_info(os, ZTEST_DIROBJ, doi) == 0); 1976 ASSERT3U(doi->doi_bonus_size, >=, sizeof (*bt)); 1977 ASSERT3U(doi->doi_bonus_size, <=, db->db_size); 1978 ASSERT3U(doi->doi_bonus_size % sizeof (*bt), ==, 0); 1979 bt = (void *)((char *)db->db_data + doi->doi_bonus_size - sizeof (*bt)); 1980 if (bt->bt_objset != 0) { 1981 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os)); 1982 ASSERT3U(bt->bt_object, ==, ZTEST_DIROBJ); 1983 ASSERT3U(bt->bt_offset, ==, -1ULL); 1984 ASSERT3U(bt->bt_txg, <, spa_first_txg(za->za_spa)); 1985 } 1986 dmu_buf_rele(db, FTAG); 1987 za->za_dbuf = NULL; 1988 } 1989 1990 void 1991 ztest_dmu_write_parallel(ztest_args_t *za) 1992 { 1993 objset_t *os = za->za_os; 1994 ztest_block_tag_t *rbt = &za->za_rbt; 1995 ztest_block_tag_t *wbt = &za->za_wbt; 1996 const size_t btsize = sizeof (ztest_block_tag_t); 1997 dmu_buf_t *db; 1998 int b, error; 1999 int bs = ZTEST_DIROBJ_BLOCKSIZE; 2000 int do_free = 0; 2001 uint64_t off, txg_how; 2002 mutex_t *lp; 2003 char osname[MAXNAMELEN]; 2004 char iobuf[SPA_MAXBLOCKSIZE]; 2005 blkptr_t blk = { 0 }; 2006 uint64_t blkoff; 2007 zbookmark_t zb; 2008 dmu_tx_t *tx = dmu_tx_create(os); 2009 2010 dmu_objset_name(os, osname); 2011 2012 /* 2013 * Have multiple threads write to large offsets in ZTEST_DIROBJ 2014 * to verify that having multiple threads writing to the same object 2015 * in parallel doesn't cause any trouble. 2016 */ 2017 if (ztest_random(4) == 0) { 2018 /* 2019 * Do the bonus buffer instead of a regular block. 2020 * We need a lock to serialize resize vs. others, 2021 * so we hash on the objset ID. 2022 */ 2023 b = dmu_objset_id(os) % ZTEST_SYNC_LOCKS; 2024 off = -1ULL; 2025 dmu_tx_hold_bonus(tx, ZTEST_DIROBJ); 2026 } else { 2027 b = ztest_random(ZTEST_SYNC_LOCKS); 2028 off = za->za_diroff_shared + (b << SPA_MAXBLOCKSHIFT); 2029 if (ztest_random(4) == 0) { 2030 do_free = 1; 2031 dmu_tx_hold_free(tx, ZTEST_DIROBJ, off, bs); 2032 } else { 2033 dmu_tx_hold_write(tx, ZTEST_DIROBJ, off, bs); 2034 } 2035 } 2036 2037 txg_how = ztest_random(2) == 0 ? TXG_WAIT : TXG_NOWAIT; 2038 error = dmu_tx_assign(tx, txg_how); 2039 if (error) { 2040 if (error == ERESTART) { 2041 ASSERT(txg_how == TXG_NOWAIT); 2042 dmu_tx_wait(tx); 2043 } else { 2044 ztest_record_enospc("dmu write parallel"); 2045 } 2046 dmu_tx_abort(tx); 2047 return; 2048 } 2049 2050 lp = &ztest_shared->zs_sync_lock[b]; 2051 (void) mutex_lock(lp); 2052 2053 wbt->bt_objset = dmu_objset_id(os); 2054 wbt->bt_object = ZTEST_DIROBJ; 2055 wbt->bt_offset = off; 2056 wbt->bt_txg = dmu_tx_get_txg(tx); 2057 wbt->bt_thread = za->za_instance; 2058 wbt->bt_seq = ztest_shared->zs_seq[b]++; /* protected by lp */ 2059 2060 if (off == -1ULL) { 2061 dmu_object_info_t *doi = &za->za_doi; 2062 char *dboff; 2063 2064 VERIFY(dmu_bonus_hold(os, ZTEST_DIROBJ, FTAG, &db) == 0); 2065 za->za_dbuf = db; 2066 dmu_object_info_from_db(db, doi); 2067 ASSERT3U(doi->doi_bonus_size, <=, db->db_size); 2068 ASSERT3U(doi->doi_bonus_size, >=, btsize); 2069 ASSERT3U(doi->doi_bonus_size % btsize, ==, 0); 2070 dboff = (char *)db->db_data + doi->doi_bonus_size - btsize; 2071 bcopy(dboff, rbt, btsize); 2072 if (rbt->bt_objset != 0) { 2073 ASSERT3U(rbt->bt_objset, ==, wbt->bt_objset); 2074 ASSERT3U(rbt->bt_object, ==, wbt->bt_object); 2075 ASSERT3U(rbt->bt_offset, ==, wbt->bt_offset); 2076 ASSERT3U(rbt->bt_txg, <=, wbt->bt_txg); 2077 } 2078 if (ztest_random(10) == 0) { 2079 int newsize = (ztest_random(db->db_size / 2080 btsize) + 1) * btsize; 2081 2082 ASSERT3U(newsize, >=, btsize); 2083 ASSERT3U(newsize, <=, db->db_size); 2084 VERIFY3U(dmu_set_bonus(db, newsize, tx), ==, 0); 2085 dboff = (char *)db->db_data + newsize - btsize; 2086 } 2087 dmu_buf_will_dirty(db, tx); 2088 bcopy(wbt, dboff, btsize); 2089 dmu_buf_rele(db, FTAG); 2090 za->za_dbuf = NULL; 2091 } else if (do_free) { 2092 VERIFY(dmu_free_range(os, ZTEST_DIROBJ, off, bs, tx) == 0); 2093 } else { 2094 dmu_write(os, ZTEST_DIROBJ, off, btsize, wbt, tx); 2095 } 2096 2097 (void) mutex_unlock(lp); 2098 2099 if (ztest_random(1000) == 0) 2100 (void) poll(NULL, 0, 1); /* open dn_notxholds window */ 2101 2102 dmu_tx_commit(tx); 2103 2104 if (ztest_random(10000) == 0) 2105 txg_wait_synced(dmu_objset_pool(os), wbt->bt_txg); 2106 2107 if (off == -1 || do_free) 2108 return; 2109 2110 if (ztest_random(2) != 0) 2111 return; 2112 2113 /* 2114 * dmu_sync() the block we just wrote. 2115 */ 2116 (void) mutex_lock(lp); 2117 2118 blkoff = P2ALIGN_TYPED(off, bs, uint64_t); 2119 error = dmu_buf_hold(os, ZTEST_DIROBJ, blkoff, FTAG, &db); 2120 za->za_dbuf = db; 2121 if (error) { 2122 dprintf("dmu_buf_hold(%s, %d, %llx) = %d\n", 2123 osname, ZTEST_DIROBJ, blkoff, error); 2124 (void) mutex_unlock(lp); 2125 return; 2126 } 2127 blkoff = off - blkoff; 2128 error = dmu_sync(NULL, db, &blk, wbt->bt_txg, NULL, NULL); 2129 dmu_buf_rele(db, FTAG); 2130 za->za_dbuf = NULL; 2131 2132 (void) mutex_unlock(lp); 2133 2134 if (error) { 2135 dprintf("dmu_sync(%s, %d, %llx) = %d\n", 2136 osname, ZTEST_DIROBJ, off, error); 2137 return; 2138 } 2139 2140 if (blk.blk_birth == 0) /* concurrent free */ 2141 return; 2142 2143 txg_suspend(dmu_objset_pool(os)); 2144 2145 ASSERT(blk.blk_fill == 1); 2146 ASSERT3U(BP_GET_TYPE(&blk), ==, DMU_OT_UINT64_OTHER); 2147 ASSERT3U(BP_GET_LEVEL(&blk), ==, 0); 2148 ASSERT3U(BP_GET_LSIZE(&blk), ==, bs); 2149 2150 /* 2151 * Read the block that dmu_sync() returned to make sure its contents 2152 * match what we wrote. We do this while still txg_suspend()ed 2153 * to ensure that the block can't be reused before we read it. 2154 */ 2155 zb.zb_objset = dmu_objset_id(os); 2156 zb.zb_object = ZTEST_DIROBJ; 2157 zb.zb_level = 0; 2158 zb.zb_blkid = off / bs; 2159 error = zio_wait(zio_read(NULL, za->za_spa, &blk, iobuf, bs, 2160 NULL, NULL, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_MUSTSUCCEED, &zb)); 2161 ASSERT3U(error, ==, 0); 2162 2163 txg_resume(dmu_objset_pool(os)); 2164 2165 bcopy(&iobuf[blkoff], rbt, btsize); 2166 2167 if (rbt->bt_objset == 0) /* concurrent free */ 2168 return; 2169 2170 ASSERT3U(rbt->bt_objset, ==, wbt->bt_objset); 2171 ASSERT3U(rbt->bt_object, ==, wbt->bt_object); 2172 ASSERT3U(rbt->bt_offset, ==, wbt->bt_offset); 2173 2174 /* 2175 * The semantic of dmu_sync() is that we always push the most recent 2176 * version of the data, so in the face of concurrent updates we may 2177 * see a newer version of the block. That's OK. 2178 */ 2179 ASSERT3U(rbt->bt_txg, >=, wbt->bt_txg); 2180 if (rbt->bt_thread == wbt->bt_thread) 2181 ASSERT3U(rbt->bt_seq, ==, wbt->bt_seq); 2182 else 2183 ASSERT3U(rbt->bt_seq, >, wbt->bt_seq); 2184 } 2185 2186 /* 2187 * Verify that zap_{create,destroy,add,remove,update} work as expected. 2188 */ 2189 #define ZTEST_ZAP_MIN_INTS 1 2190 #define ZTEST_ZAP_MAX_INTS 4 2191 #define ZTEST_ZAP_MAX_PROPS 1000 2192 2193 void 2194 ztest_zap(ztest_args_t *za) 2195 { 2196 objset_t *os = za->za_os; 2197 uint64_t object; 2198 uint64_t txg, last_txg; 2199 uint64_t value[ZTEST_ZAP_MAX_INTS]; 2200 uint64_t zl_ints, zl_intsize, prop; 2201 int i, ints; 2202 dmu_tx_t *tx; 2203 char propname[100], txgname[100]; 2204 int error; 2205 char osname[MAXNAMELEN]; 2206 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; 2207 2208 dmu_objset_name(os, osname); 2209 2210 /* 2211 * Create a new object if necessary, and record it in the directory. 2212 */ 2213 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 2214 sizeof (uint64_t), &object)); 2215 2216 if (object == 0) { 2217 tx = dmu_tx_create(os); 2218 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, 2219 sizeof (uint64_t)); 2220 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL); 2221 error = dmu_tx_assign(tx, TXG_WAIT); 2222 if (error) { 2223 ztest_record_enospc("create zap test obj"); 2224 dmu_tx_abort(tx); 2225 return; 2226 } 2227 object = zap_create(os, DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx); 2228 if (error) { 2229 fatal(0, "zap_create('%s', %llu) = %d", 2230 osname, object, error); 2231 } 2232 ASSERT(object != 0); 2233 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, 2234 sizeof (uint64_t), &object, tx); 2235 /* 2236 * Generate a known hash collision, and verify that 2237 * we can lookup and remove both entries. 2238 */ 2239 for (i = 0; i < 2; i++) { 2240 value[i] = i; 2241 error = zap_add(os, object, hc[i], sizeof (uint64_t), 2242 1, &value[i], tx); 2243 ASSERT3U(error, ==, 0); 2244 } 2245 for (i = 0; i < 2; i++) { 2246 error = zap_add(os, object, hc[i], sizeof (uint64_t), 2247 1, &value[i], tx); 2248 ASSERT3U(error, ==, EEXIST); 2249 error = zap_length(os, object, hc[i], 2250 &zl_intsize, &zl_ints); 2251 ASSERT3U(error, ==, 0); 2252 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2253 ASSERT3U(zl_ints, ==, 1); 2254 } 2255 for (i = 0; i < 2; i++) { 2256 error = zap_remove(os, object, hc[i], tx); 2257 ASSERT3U(error, ==, 0); 2258 } 2259 2260 dmu_tx_commit(tx); 2261 } 2262 2263 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); 2264 2265 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 2266 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 2267 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 2268 bzero(value, sizeof (value)); 2269 last_txg = 0; 2270 2271 /* 2272 * If these zap entries already exist, validate their contents. 2273 */ 2274 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 2275 if (error == 0) { 2276 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2277 ASSERT3U(zl_ints, ==, 1); 2278 2279 VERIFY(zap_lookup(os, object, txgname, zl_intsize, 2280 zl_ints, &last_txg) == 0); 2281 2282 VERIFY(zap_length(os, object, propname, &zl_intsize, 2283 &zl_ints) == 0); 2284 2285 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2286 ASSERT3U(zl_ints, ==, ints); 2287 2288 VERIFY(zap_lookup(os, object, propname, zl_intsize, 2289 zl_ints, value) == 0); 2290 2291 for (i = 0; i < ints; i++) { 2292 ASSERT3U(value[i], ==, last_txg + object + i); 2293 } 2294 } else { 2295 ASSERT3U(error, ==, ENOENT); 2296 } 2297 2298 /* 2299 * Atomically update two entries in our zap object. 2300 * The first is named txg_%llu, and contains the txg 2301 * in which the property was last updated. The second 2302 * is named prop_%llu, and the nth element of its value 2303 * should be txg + object + n. 2304 */ 2305 tx = dmu_tx_create(os); 2306 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2307 error = dmu_tx_assign(tx, TXG_WAIT); 2308 if (error) { 2309 ztest_record_enospc("create zap entry"); 2310 dmu_tx_abort(tx); 2311 return; 2312 } 2313 txg = dmu_tx_get_txg(tx); 2314 2315 if (last_txg > txg) 2316 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); 2317 2318 for (i = 0; i < ints; i++) 2319 value[i] = txg + object + i; 2320 2321 error = zap_update(os, object, txgname, sizeof (uint64_t), 1, &txg, tx); 2322 if (error) 2323 fatal(0, "zap_update('%s', %llu, '%s') = %d", 2324 osname, object, txgname, error); 2325 2326 error = zap_update(os, object, propname, sizeof (uint64_t), 2327 ints, value, tx); 2328 if (error) 2329 fatal(0, "zap_update('%s', %llu, '%s') = %d", 2330 osname, object, propname, error); 2331 2332 dmu_tx_commit(tx); 2333 2334 /* 2335 * Remove a random pair of entries. 2336 */ 2337 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 2338 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 2339 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 2340 2341 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 2342 2343 if (error == ENOENT) 2344 return; 2345 2346 ASSERT3U(error, ==, 0); 2347 2348 tx = dmu_tx_create(os); 2349 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2350 error = dmu_tx_assign(tx, TXG_WAIT); 2351 if (error) { 2352 ztest_record_enospc("remove zap entry"); 2353 dmu_tx_abort(tx); 2354 return; 2355 } 2356 error = zap_remove(os, object, txgname, tx); 2357 if (error) 2358 fatal(0, "zap_remove('%s', %llu, '%s') = %d", 2359 osname, object, txgname, error); 2360 2361 error = zap_remove(os, object, propname, tx); 2362 if (error) 2363 fatal(0, "zap_remove('%s', %llu, '%s') = %d", 2364 osname, object, propname, error); 2365 2366 dmu_tx_commit(tx); 2367 2368 /* 2369 * Once in a while, destroy the object. 2370 */ 2371 if (ztest_random(1000) != 0) 2372 return; 2373 2374 tx = dmu_tx_create(os); 2375 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t)); 2376 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 2377 error = dmu_tx_assign(tx, TXG_WAIT); 2378 if (error) { 2379 ztest_record_enospc("destroy zap object"); 2380 dmu_tx_abort(tx); 2381 return; 2382 } 2383 error = zap_destroy(os, object, tx); 2384 if (error) 2385 fatal(0, "zap_destroy('%s', %llu) = %d", 2386 osname, object, error); 2387 object = 0; 2388 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t), 2389 &object, tx); 2390 dmu_tx_commit(tx); 2391 } 2392 2393 void 2394 ztest_zap_parallel(ztest_args_t *za) 2395 { 2396 objset_t *os = za->za_os; 2397 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; 2398 dmu_tx_t *tx; 2399 int i, namelen, error; 2400 char name[20], string_value[20]; 2401 void *data; 2402 2403 /* 2404 * Generate a random name of the form 'xxx.....' where each 2405 * x is a random printable character and the dots are dots. 2406 * There are 94 such characters, and the name length goes from 2407 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. 2408 */ 2409 namelen = ztest_random(sizeof (name) - 5) + 5 + 1; 2410 2411 for (i = 0; i < 3; i++) 2412 name[i] = '!' + ztest_random('~' - '!' + 1); 2413 for (; i < namelen - 1; i++) 2414 name[i] = '.'; 2415 name[i] = '\0'; 2416 2417 if (ztest_random(2) == 0) 2418 object = ZTEST_MICROZAP_OBJ; 2419 else 2420 object = ZTEST_FATZAP_OBJ; 2421 2422 if ((namelen & 1) || object == ZTEST_MICROZAP_OBJ) { 2423 wsize = sizeof (txg); 2424 wc = 1; 2425 data = &txg; 2426 } else { 2427 wsize = 1; 2428 wc = namelen; 2429 data = string_value; 2430 } 2431 2432 count = -1ULL; 2433 VERIFY(zap_count(os, object, &count) == 0); 2434 ASSERT(count != -1ULL); 2435 2436 /* 2437 * Select an operation: length, lookup, add, update, remove. 2438 */ 2439 i = ztest_random(5); 2440 2441 if (i >= 2) { 2442 tx = dmu_tx_create(os); 2443 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2444 error = dmu_tx_assign(tx, TXG_WAIT); 2445 if (error) { 2446 ztest_record_enospc("zap parallel"); 2447 dmu_tx_abort(tx); 2448 return; 2449 } 2450 txg = dmu_tx_get_txg(tx); 2451 bcopy(name, string_value, namelen); 2452 } else { 2453 tx = NULL; 2454 txg = 0; 2455 bzero(string_value, namelen); 2456 } 2457 2458 switch (i) { 2459 2460 case 0: 2461 error = zap_length(os, object, name, &zl_wsize, &zl_wc); 2462 if (error == 0) { 2463 ASSERT3U(wsize, ==, zl_wsize); 2464 ASSERT3U(wc, ==, zl_wc); 2465 } else { 2466 ASSERT3U(error, ==, ENOENT); 2467 } 2468 break; 2469 2470 case 1: 2471 error = zap_lookup(os, object, name, wsize, wc, data); 2472 if (error == 0) { 2473 if (data == string_value && 2474 bcmp(name, data, namelen) != 0) 2475 fatal(0, "name '%s' != val '%s' len %d", 2476 name, data, namelen); 2477 } else { 2478 ASSERT3U(error, ==, ENOENT); 2479 } 2480 break; 2481 2482 case 2: 2483 error = zap_add(os, object, name, wsize, wc, data, tx); 2484 ASSERT(error == 0 || error == EEXIST); 2485 break; 2486 2487 case 3: 2488 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); 2489 break; 2490 2491 case 4: 2492 error = zap_remove(os, object, name, tx); 2493 ASSERT(error == 0 || error == ENOENT); 2494 break; 2495 } 2496 2497 if (tx != NULL) 2498 dmu_tx_commit(tx); 2499 } 2500 2501 void 2502 ztest_dsl_prop_get_set(ztest_args_t *za) 2503 { 2504 objset_t *os = za->za_os; 2505 int i, inherit; 2506 uint64_t value; 2507 const char *prop, *valname; 2508 char setpoint[MAXPATHLEN]; 2509 char osname[MAXNAMELEN]; 2510 int error; 2511 2512 (void) rw_rdlock(&ztest_shared->zs_name_lock); 2513 2514 dmu_objset_name(os, osname); 2515 2516 for (i = 0; i < 2; i++) { 2517 if (i == 0) { 2518 prop = "checksum"; 2519 value = ztest_random_checksum(); 2520 inherit = (value == ZIO_CHECKSUM_INHERIT); 2521 } else { 2522 prop = "compression"; 2523 value = ztest_random_compress(); 2524 inherit = (value == ZIO_COMPRESS_INHERIT); 2525 } 2526 2527 error = dsl_prop_set(osname, prop, sizeof (value), 2528 !inherit, &value); 2529 2530 if (error == ENOSPC) { 2531 ztest_record_enospc("dsl_prop_set"); 2532 break; 2533 } 2534 2535 ASSERT3U(error, ==, 0); 2536 2537 VERIFY3U(dsl_prop_get(osname, prop, sizeof (value), 2538 1, &value, setpoint), ==, 0); 2539 2540 if (i == 0) 2541 valname = zio_checksum_table[value].ci_name; 2542 else 2543 valname = zio_compress_table[value].ci_name; 2544 2545 if (zopt_verbose >= 6) { 2546 (void) printf("%s %s = %s for '%s'\n", 2547 osname, prop, valname, setpoint); 2548 } 2549 } 2550 2551 (void) rw_unlock(&ztest_shared->zs_name_lock); 2552 } 2553 2554 static void 2555 ztest_error_setup(vdev_t *vd, int mode, int mask, uint64_t arg) 2556 { 2557 int c; 2558 2559 for (c = 0; c < vd->vdev_children; c++) 2560 ztest_error_setup(vd->vdev_child[c], mode, mask, arg); 2561 2562 if (vd->vdev_path != NULL) { 2563 vd->vdev_fault_mode = mode; 2564 vd->vdev_fault_mask = mask; 2565 vd->vdev_fault_arg = arg; 2566 } 2567 } 2568 2569 /* 2570 * Inject random faults into the on-disk data. 2571 */ 2572 void 2573 ztest_fault_inject(ztest_args_t *za) 2574 { 2575 int fd; 2576 uint64_t offset; 2577 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 2578 uint64_t bad = 0x1990c0ffeedecade; 2579 uint64_t top, leaf; 2580 char path0[MAXPATHLEN]; 2581 char pathrand[MAXPATHLEN]; 2582 size_t fsize; 2583 spa_t *spa = za->za_spa; 2584 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ 2585 int iters = 1000; 2586 vdev_t *vd0; 2587 uint64_t guid0 = 0; 2588 2589 /* 2590 * We can't inject faults when we have no fault tolerance. 2591 */ 2592 if (zopt_maxfaults == 0) 2593 return; 2594 2595 ASSERT(leaves >= 2); 2596 2597 /* 2598 * Pick a random top-level vdev. 2599 */ 2600 spa_config_enter(spa, RW_READER, FTAG); 2601 top = ztest_random(spa->spa_root_vdev->vdev_children); 2602 spa_config_exit(spa, FTAG); 2603 2604 /* 2605 * Pick a random leaf. 2606 */ 2607 leaf = ztest_random(leaves); 2608 2609 /* 2610 * Generate paths to the first two leaves in this top-level vdev, 2611 * and to the random leaf we selected. We'll induce transient 2612 * I/O errors and random online/offline activity on leaf 0, 2613 * and we'll write random garbage to the randomly chosen leaf. 2614 */ 2615 (void) snprintf(path0, sizeof (path0), 2616 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + 0); 2617 (void) snprintf(pathrand, sizeof (pathrand), 2618 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf); 2619 2620 dprintf("damaging %s and %s\n", path0, pathrand); 2621 2622 spa_config_enter(spa, RW_READER, FTAG); 2623 2624 /* 2625 * If we can tolerate two or more faults, make vd0 fail randomly. 2626 */ 2627 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); 2628 if (vd0 != NULL && zopt_maxfaults >= 2) { 2629 guid0 = vd0->vdev_guid; 2630 ztest_error_setup(vd0, VDEV_FAULT_COUNT, 2631 (1U << ZIO_TYPE_READ) | (1U << ZIO_TYPE_WRITE), 100); 2632 } 2633 2634 spa_config_exit(spa, FTAG); 2635 2636 /* 2637 * If we can tolerate two or more faults, randomly online/offline vd0. 2638 */ 2639 if (zopt_maxfaults >= 2 && guid0 != 0) { 2640 if (ztest_random(10) < 6) 2641 (void) vdev_offline(spa, guid0, B_TRUE); 2642 else 2643 (void) vdev_online(spa, guid0, B_FALSE, NULL); 2644 } 2645 2646 /* 2647 * We have at least single-fault tolerance, so inject data corruption. 2648 */ 2649 fd = open(pathrand, O_RDWR); 2650 2651 if (fd == -1) /* we hit a gap in the device namespace */ 2652 return; 2653 2654 fsize = lseek(fd, 0, SEEK_END); 2655 2656 while (--iters != 0) { 2657 offset = ztest_random(fsize / (leaves << bshift)) * 2658 (leaves << bshift) + (leaf << bshift) + 2659 (ztest_random(1ULL << (bshift - 1)) & -8ULL); 2660 2661 if (offset >= fsize) 2662 continue; 2663 2664 if (zopt_verbose >= 6) 2665 (void) printf("injecting bad word into %s," 2666 " offset 0x%llx\n", pathrand, (u_longlong_t)offset); 2667 2668 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) 2669 fatal(1, "can't inject bad word at 0x%llx in %s", 2670 offset, pathrand); 2671 } 2672 2673 (void) close(fd); 2674 } 2675 2676 /* 2677 * Scrub the pool. 2678 */ 2679 void 2680 ztest_scrub(ztest_args_t *za) 2681 { 2682 spa_t *spa = za->za_spa; 2683 2684 mutex_enter(&spa_namespace_lock); 2685 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE); 2686 mutex_exit(&spa_namespace_lock); 2687 (void) poll(NULL, 0, 1000); /* wait a second, then force a restart */ 2688 mutex_enter(&spa_namespace_lock); 2689 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE); 2690 mutex_exit(&spa_namespace_lock); 2691 } 2692 2693 /* 2694 * Rename the pool to a different name and then rename it back. 2695 */ 2696 void 2697 ztest_spa_rename(ztest_args_t *za) 2698 { 2699 char *oldname, *newname; 2700 int error; 2701 spa_t *spa; 2702 2703 (void) rw_wrlock(&ztest_shared->zs_name_lock); 2704 2705 oldname = za->za_pool; 2706 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); 2707 (void) strcpy(newname, oldname); 2708 (void) strcat(newname, "_tmp"); 2709 2710 /* 2711 * Do the rename 2712 */ 2713 error = spa_rename(oldname, newname); 2714 if (error) 2715 fatal(0, "spa_rename('%s', '%s') = %d", oldname, 2716 newname, error); 2717 2718 /* 2719 * Try to open it under the old name, which shouldn't exist 2720 */ 2721 error = spa_open(oldname, &spa, FTAG); 2722 if (error != ENOENT) 2723 fatal(0, "spa_open('%s') = %d", oldname, error); 2724 2725 /* 2726 * Open it under the new name and make sure it's still the same spa_t. 2727 */ 2728 error = spa_open(newname, &spa, FTAG); 2729 if (error != 0) 2730 fatal(0, "spa_open('%s') = %d", newname, error); 2731 2732 ASSERT(spa == za->za_spa); 2733 spa_close(spa, FTAG); 2734 2735 /* 2736 * Rename it back to the original 2737 */ 2738 error = spa_rename(newname, oldname); 2739 if (error) 2740 fatal(0, "spa_rename('%s', '%s') = %d", newname, 2741 oldname, error); 2742 2743 /* 2744 * Make sure it can still be opened 2745 */ 2746 error = spa_open(oldname, &spa, FTAG); 2747 if (error != 0) 2748 fatal(0, "spa_open('%s') = %d", oldname, error); 2749 2750 ASSERT(spa == za->za_spa); 2751 spa_close(spa, FTAG); 2752 2753 umem_free(newname, strlen(newname) + 1); 2754 2755 (void) rw_unlock(&ztest_shared->zs_name_lock); 2756 } 2757 2758 2759 /* 2760 * Completely obliterate one disk. 2761 */ 2762 static void 2763 ztest_obliterate_one_disk(uint64_t vdev) 2764 { 2765 int fd; 2766 char dev_name[MAXPATHLEN], copy_name[MAXPATHLEN]; 2767 size_t fsize; 2768 2769 if (zopt_maxfaults < 2) 2770 return; 2771 2772 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 2773 (void) snprintf(copy_name, MAXPATHLEN, "%s.old", dev_name); 2774 2775 fd = open(dev_name, O_RDWR); 2776 2777 if (fd == -1) 2778 fatal(1, "can't open %s", dev_name); 2779 2780 /* 2781 * Determine the size. 2782 */ 2783 fsize = lseek(fd, 0, SEEK_END); 2784 2785 (void) close(fd); 2786 2787 /* 2788 * Rename the old device to dev_name.old (useful for debugging). 2789 */ 2790 VERIFY(rename(dev_name, copy_name) == 0); 2791 2792 /* 2793 * Create a new one. 2794 */ 2795 VERIFY((fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666)) >= 0); 2796 VERIFY(ftruncate(fd, fsize) == 0); 2797 (void) close(fd); 2798 } 2799 2800 static void 2801 ztest_replace_one_disk(spa_t *spa, uint64_t vdev) 2802 { 2803 char dev_name[MAXPATHLEN]; 2804 nvlist_t *file, *root; 2805 int error; 2806 uint64_t guid; 2807 uint64_t ashift = ztest_get_ashift(); 2808 vdev_t *vd; 2809 2810 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 2811 2812 /* 2813 * Build the nvlist describing dev_name. 2814 */ 2815 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 2816 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 2817 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0); 2818 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 2819 2820 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 2821 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 2822 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 2823 &file, 1) == 0); 2824 2825 spa_config_enter(spa, RW_READER, FTAG); 2826 if ((vd = vdev_lookup_by_path(spa->spa_root_vdev, dev_name)) == NULL) 2827 guid = 0; 2828 else 2829 guid = vd->vdev_guid; 2830 spa_config_exit(spa, FTAG); 2831 error = spa_vdev_attach(spa, guid, root, B_TRUE); 2832 if (error != 0 && 2833 error != EBUSY && 2834 error != ENOTSUP && 2835 error != ENODEV && 2836 error != EDOM) 2837 fatal(0, "spa_vdev_attach(in-place) = %d", error); 2838 2839 nvlist_free(file); 2840 nvlist_free(root); 2841 } 2842 2843 static void 2844 ztest_verify_blocks(char *pool) 2845 { 2846 int status; 2847 char zdb[MAXPATHLEN + MAXNAMELEN + 20]; 2848 char zbuf[1024]; 2849 char *bin; 2850 char *ztest; 2851 char *isa; 2852 int isalen; 2853 FILE *fp; 2854 2855 (void) realpath(getexecname(), zdb); 2856 2857 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ 2858 bin = strstr(zdb, "/usr/bin/"); 2859 ztest = strstr(bin, "/ztest"); 2860 isa = bin + 8; 2861 isalen = ztest - isa; 2862 isa = strdup(isa); 2863 /* LINTED */ 2864 (void) sprintf(bin, 2865 "/usr/sbin%.*s/zdb -bc%s%s -U /tmp/zpool.cache -O %s %s", 2866 isalen, 2867 isa, 2868 zopt_verbose >= 3 ? "s" : "", 2869 zopt_verbose >= 4 ? "v" : "", 2870 ztest_random(2) == 0 ? "pre" : "post", pool); 2871 free(isa); 2872 2873 if (zopt_verbose >= 5) 2874 (void) printf("Executing %s\n", strstr(zdb, "zdb ")); 2875 2876 fp = popen(zdb, "r"); 2877 2878 while (fgets(zbuf, sizeof (zbuf), fp) != NULL) 2879 if (zopt_verbose >= 3) 2880 (void) printf("%s", zbuf); 2881 2882 status = pclose(fp); 2883 2884 if (status == 0) 2885 return; 2886 2887 ztest_dump_core = 0; 2888 if (WIFEXITED(status)) 2889 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); 2890 else 2891 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); 2892 } 2893 2894 static void 2895 ztest_walk_pool_directory(char *header) 2896 { 2897 spa_t *spa = NULL; 2898 2899 if (zopt_verbose >= 6) 2900 (void) printf("%s\n", header); 2901 2902 mutex_enter(&spa_namespace_lock); 2903 while ((spa = spa_next(spa)) != NULL) 2904 if (zopt_verbose >= 6) 2905 (void) printf("\t%s\n", spa_name(spa)); 2906 mutex_exit(&spa_namespace_lock); 2907 } 2908 2909 static void 2910 ztest_spa_import_export(char *oldname, char *newname) 2911 { 2912 nvlist_t *config; 2913 uint64_t pool_guid; 2914 spa_t *spa; 2915 int error; 2916 2917 if (zopt_verbose >= 4) { 2918 (void) printf("import/export: old = %s, new = %s\n", 2919 oldname, newname); 2920 } 2921 2922 /* 2923 * Clean up from previous runs. 2924 */ 2925 (void) spa_destroy(newname); 2926 2927 /* 2928 * Get the pool's configuration and guid. 2929 */ 2930 error = spa_open(oldname, &spa, FTAG); 2931 if (error) 2932 fatal(0, "spa_open('%s') = %d", oldname, error); 2933 2934 pool_guid = spa_guid(spa); 2935 spa_close(spa, FTAG); 2936 2937 ztest_walk_pool_directory("pools before export"); 2938 2939 /* 2940 * Export it. 2941 */ 2942 error = spa_export(oldname, &config); 2943 if (error) 2944 fatal(0, "spa_export('%s') = %d", oldname, error); 2945 2946 ztest_walk_pool_directory("pools after export"); 2947 2948 /* 2949 * Import it under the new name. 2950 */ 2951 error = spa_import(newname, config, NULL); 2952 if (error) 2953 fatal(0, "spa_import('%s') = %d", newname, error); 2954 2955 ztest_walk_pool_directory("pools after import"); 2956 2957 /* 2958 * Try to import it again -- should fail with EEXIST. 2959 */ 2960 error = spa_import(newname, config, NULL); 2961 if (error != EEXIST) 2962 fatal(0, "spa_import('%s') twice", newname); 2963 2964 /* 2965 * Try to import it under a different name -- should fail with EEXIST. 2966 */ 2967 error = spa_import(oldname, config, NULL); 2968 if (error != EEXIST) 2969 fatal(0, "spa_import('%s') under multiple names", newname); 2970 2971 /* 2972 * Verify that the pool is no longer visible under the old name. 2973 */ 2974 error = spa_open(oldname, &spa, FTAG); 2975 if (error != ENOENT) 2976 fatal(0, "spa_open('%s') = %d", newname, error); 2977 2978 /* 2979 * Verify that we can open and close the pool using the new name. 2980 */ 2981 error = spa_open(newname, &spa, FTAG); 2982 if (error) 2983 fatal(0, "spa_open('%s') = %d", newname, error); 2984 ASSERT(pool_guid == spa_guid(spa)); 2985 spa_close(spa, FTAG); 2986 2987 nvlist_free(config); 2988 } 2989 2990 /* ARGSUSED */ 2991 static void * 2992 ztest_suspend_monitor(void *arg) 2993 { 2994 spa_t *spa; 2995 int error; 2996 2997 error = spa_open(zopt_pool, &spa, FTAG); 2998 if (error) { 2999 (void) printf("Unable to monitor pool '%s'\n", zopt_pool); 3000 return (NULL); 3001 } 3002 3003 while (!ztest_exiting) { 3004 mutex_enter(&spa->spa_zio_lock); 3005 while (!ztest_exiting && list_is_empty(&spa->spa_zio_list)) 3006 cv_wait(&spa->spa_zio_cv, &spa->spa_zio_lock); 3007 mutex_exit(&spa->spa_zio_lock); 3008 3009 (void) sleep(3); 3010 /* 3011 * We don't hold the spa_config_lock since the pool is in 3012 * complete failure mode and there is no way for us to 3013 * change the vdev config when we're in this state. 3014 */ 3015 while ((error = zio_vdev_resume_io(spa)) != 0) { 3016 (void) printf("I/O could not be resumed, %d\n", error); 3017 (void) sleep(1); 3018 } 3019 vdev_clear(spa, NULL, B_TRUE); 3020 } 3021 spa_close(spa, FTAG); 3022 return (NULL); 3023 } 3024 3025 static void * 3026 ztest_thread(void *arg) 3027 { 3028 ztest_args_t *za = arg; 3029 ztest_shared_t *zs = ztest_shared; 3030 hrtime_t now, functime; 3031 ztest_info_t *zi; 3032 int f, i; 3033 3034 while ((now = gethrtime()) < za->za_stop) { 3035 /* 3036 * See if it's time to force a crash. 3037 */ 3038 if (now > za->za_kill) { 3039 zs->zs_alloc = spa_get_alloc(za->za_spa); 3040 zs->zs_space = spa_get_space(za->za_spa); 3041 (void) kill(getpid(), SIGKILL); 3042 } 3043 3044 /* 3045 * Pick a random function. 3046 */ 3047 f = ztest_random(ZTEST_FUNCS); 3048 zi = &zs->zs_info[f]; 3049 3050 /* 3051 * Decide whether to call it, based on the requested frequency. 3052 */ 3053 if (zi->zi_call_target == 0 || 3054 (double)zi->zi_call_total / zi->zi_call_target > 3055 (double)(now - zs->zs_start_time) / (zopt_time * NANOSEC)) 3056 continue; 3057 3058 atomic_add_64(&zi->zi_calls, 1); 3059 atomic_add_64(&zi->zi_call_total, 1); 3060 3061 za->za_diroff = (za->za_instance * ZTEST_FUNCS + f) * 3062 ZTEST_DIRSIZE; 3063 za->za_diroff_shared = (1ULL << 63); 3064 3065 for (i = 0; i < zi->zi_iters; i++) 3066 zi->zi_func(za); 3067 3068 functime = gethrtime() - now; 3069 3070 atomic_add_64(&zi->zi_call_time, functime); 3071 3072 if (zopt_verbose >= 4) { 3073 Dl_info dli; 3074 (void) dladdr((void *)zi->zi_func, &dli); 3075 (void) printf("%6.2f sec in %s\n", 3076 (double)functime / NANOSEC, dli.dli_sname); 3077 } 3078 3079 /* 3080 * If we're getting ENOSPC with some regularity, stop. 3081 */ 3082 if (zs->zs_enospc_count > 10) 3083 break; 3084 } 3085 3086 return (NULL); 3087 } 3088 3089 /* 3090 * Kick off threads to run tests on all datasets in parallel. 3091 */ 3092 static void 3093 ztest_run(char *pool) 3094 { 3095 int t, d, error; 3096 ztest_shared_t *zs = ztest_shared; 3097 ztest_args_t *za; 3098 spa_t *spa; 3099 char name[100]; 3100 thread_t tid; 3101 3102 (void) _mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL); 3103 (void) rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL); 3104 3105 for (t = 0; t < ZTEST_SYNC_LOCKS; t++) 3106 (void) _mutex_init(&zs->zs_sync_lock[t], USYNC_THREAD, NULL); 3107 3108 /* 3109 * Destroy one disk before we even start. 3110 * It's mirrored, so everything should work just fine. 3111 * This makes us exercise fault handling very early in spa_load(). 3112 */ 3113 ztest_obliterate_one_disk(0); 3114 3115 /* 3116 * Verify that the sum of the sizes of all blocks in the pool 3117 * equals the SPA's allocated space total. 3118 */ 3119 ztest_verify_blocks(pool); 3120 3121 /* 3122 * Kick off a replacement of the disk we just obliterated. 3123 */ 3124 kernel_init(FREAD | FWRITE); 3125 error = spa_open(pool, &spa, FTAG); 3126 if (error) 3127 fatal(0, "spa_open(%s) = %d", pool, error); 3128 ztest_replace_one_disk(spa, 0); 3129 if (zopt_verbose >= 5) 3130 show_pool_stats(spa); 3131 spa_close(spa, FTAG); 3132 kernel_fini(); 3133 3134 kernel_init(FREAD | FWRITE); 3135 3136 /* 3137 * Verify that we can export the pool and reimport it under a 3138 * different name. 3139 */ 3140 if (ztest_random(2) == 0) { 3141 (void) snprintf(name, 100, "%s_import", pool); 3142 ztest_spa_import_export(pool, name); 3143 ztest_spa_import_export(name, pool); 3144 } 3145 3146 /* 3147 * Verify that we can loop over all pools. 3148 */ 3149 mutex_enter(&spa_namespace_lock); 3150 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) { 3151 if (zopt_verbose > 3) { 3152 (void) printf("spa_next: found %s\n", spa_name(spa)); 3153 } 3154 } 3155 mutex_exit(&spa_namespace_lock); 3156 3157 /* 3158 * Create a thread to handling complete pool failures. This 3159 * thread will kickstart the I/Os when they suspend. We must 3160 * start the thread before setting the zio_io_fail_shift, which 3161 * will indicate our failure rate. 3162 */ 3163 error = thr_create(0, 0, ztest_suspend_monitor, NULL, THR_BOUND, &tid); 3164 if (error) { 3165 fatal(0, "can't create suspend monitor thread: error %d", 3166 t, error); 3167 } 3168 3169 /* 3170 * Open our pool. 3171 */ 3172 error = spa_open(pool, &spa, FTAG); 3173 if (error) 3174 fatal(0, "spa_open() = %d", error); 3175 3176 /* 3177 * Verify that we can safely inquire about about any object, 3178 * whether it's allocated or not. To make it interesting, 3179 * we probe a 5-wide window around each power of two. 3180 * This hits all edge cases, including zero and the max. 3181 */ 3182 for (t = 0; t < 64; t++) { 3183 for (d = -5; d <= 5; d++) { 3184 error = dmu_object_info(spa->spa_meta_objset, 3185 (1ULL << t) + d, NULL); 3186 ASSERT(error == 0 || error == ENOENT || 3187 error == EINVAL); 3188 } 3189 } 3190 3191 /* 3192 * Now kick off all the tests that run in parallel. 3193 */ 3194 zs->zs_enospc_count = 0; 3195 3196 za = umem_zalloc(zopt_threads * sizeof (ztest_args_t), UMEM_NOFAIL); 3197 3198 if (zopt_verbose >= 4) 3199 (void) printf("starting main threads...\n"); 3200 3201 /* Let failures begin */ 3202 zio_io_fail_shift = zopt_write_fail_shift; 3203 3204 za[0].za_start = gethrtime(); 3205 za[0].za_stop = za[0].za_start + zopt_passtime * NANOSEC; 3206 za[0].za_stop = MIN(za[0].za_stop, zs->zs_stop_time); 3207 za[0].za_kill = za[0].za_stop; 3208 if (ztest_random(100) < zopt_killrate) 3209 za[0].za_kill -= ztest_random(zopt_passtime * NANOSEC); 3210 3211 for (t = 0; t < zopt_threads; t++) { 3212 d = t % zopt_datasets; 3213 3214 (void) strcpy(za[t].za_pool, pool); 3215 za[t].za_os = za[d].za_os; 3216 za[t].za_spa = spa; 3217 za[t].za_zilog = za[d].za_zilog; 3218 za[t].za_instance = t; 3219 za[t].za_random = ztest_random(-1ULL); 3220 za[t].za_start = za[0].za_start; 3221 za[t].za_stop = za[0].za_stop; 3222 za[t].za_kill = za[0].za_kill; 3223 3224 if (t < zopt_datasets) { 3225 ztest_replay_t zr; 3226 int test_future = FALSE; 3227 (void) rw_rdlock(&ztest_shared->zs_name_lock); 3228 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d); 3229 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0, 3230 ztest_create_cb, NULL); 3231 if (error == EEXIST) { 3232 test_future = TRUE; 3233 } else if (error == ENOSPC) { 3234 zs->zs_enospc_count++; 3235 (void) rw_unlock(&ztest_shared->zs_name_lock); 3236 break; 3237 } else if (error != 0) { 3238 fatal(0, "dmu_objset_create(%s) = %d", 3239 name, error); 3240 } 3241 error = dmu_objset_open(name, DMU_OST_OTHER, 3242 DS_MODE_USER, &za[d].za_os); 3243 if (error) 3244 fatal(0, "dmu_objset_open('%s') = %d", 3245 name, error); 3246 (void) rw_unlock(&ztest_shared->zs_name_lock); 3247 if (test_future) 3248 ztest_dmu_check_future_leak(&za[t]); 3249 zr.zr_os = za[d].za_os; 3250 zil_replay(zr.zr_os, &zr, &zr.zr_assign, 3251 ztest_replay_vector); 3252 za[d].za_zilog = zil_open(za[d].za_os, NULL); 3253 } 3254 3255 error = thr_create(0, 0, ztest_thread, &za[t], THR_BOUND, 3256 &za[t].za_thread); 3257 if (error) 3258 fatal(0, "can't create thread %d: error %d", 3259 t, error); 3260 } 3261 3262 while (--t >= 0) { 3263 error = thr_join(za[t].za_thread, NULL, NULL); 3264 if (error) 3265 fatal(0, "thr_join(%d) = %d", t, error); 3266 if (za[t].za_th) 3267 traverse_fini(za[t].za_th); 3268 if (t < zopt_datasets) { 3269 zil_close(za[t].za_zilog); 3270 dmu_objset_close(za[t].za_os); 3271 } 3272 } 3273 3274 if (zopt_verbose >= 3) 3275 show_pool_stats(spa); 3276 3277 txg_wait_synced(spa_get_dsl(spa), 0); 3278 3279 zs->zs_alloc = spa_get_alloc(spa); 3280 zs->zs_space = spa_get_space(spa); 3281 3282 /* 3283 * If we had out-of-space errors, destroy a random objset. 3284 */ 3285 if (zs->zs_enospc_count != 0) { 3286 (void) rw_rdlock(&ztest_shared->zs_name_lock); 3287 d = (int)ztest_random(zopt_datasets); 3288 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d); 3289 if (zopt_verbose >= 3) 3290 (void) printf("Destroying %s to free up space\n", name); 3291 (void) dmu_objset_find(name, ztest_destroy_cb, &za[d], 3292 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); 3293 (void) rw_unlock(&ztest_shared->zs_name_lock); 3294 } 3295 3296 txg_wait_synced(spa_get_dsl(spa), 0); 3297 3298 /* 3299 * Right before closing the pool, kick off a bunch of async I/O; 3300 * spa_close() should wait for it to complete. 3301 */ 3302 for (t = 1; t < 50; t++) 3303 dmu_prefetch(spa->spa_meta_objset, t, 0, 1 << 15); 3304 3305 /* Shutdown the suspend monitor thread */ 3306 zio_io_fail_shift = 0; 3307 ztest_exiting = B_TRUE; 3308 mutex_enter(&spa->spa_zio_lock); 3309 cv_broadcast(&spa->spa_zio_cv); 3310 mutex_exit(&spa->spa_zio_lock); 3311 error = thr_join(tid, NULL, NULL); 3312 if (error) 3313 fatal(0, "thr_join(%d) = %d", tid, error); 3314 3315 umem_free(za, zopt_threads * sizeof (ztest_args_t)); 3316 3317 spa_close(spa, FTAG); 3318 3319 kernel_fini(); 3320 } 3321 3322 void 3323 print_time(hrtime_t t, char *timebuf) 3324 { 3325 hrtime_t s = t / NANOSEC; 3326 hrtime_t m = s / 60; 3327 hrtime_t h = m / 60; 3328 hrtime_t d = h / 24; 3329 3330 s -= m * 60; 3331 m -= h * 60; 3332 h -= d * 24; 3333 3334 timebuf[0] = '\0'; 3335 3336 if (d) 3337 (void) sprintf(timebuf, 3338 "%llud%02lluh%02llum%02llus", d, h, m, s); 3339 else if (h) 3340 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); 3341 else if (m) 3342 (void) sprintf(timebuf, "%llum%02llus", m, s); 3343 else 3344 (void) sprintf(timebuf, "%llus", s); 3345 } 3346 3347 /* 3348 * Create a storage pool with the given name and initial vdev size. 3349 * Then create the specified number of datasets in the pool. 3350 */ 3351 static void 3352 ztest_init(char *pool) 3353 { 3354 spa_t *spa; 3355 int error; 3356 nvlist_t *nvroot; 3357 3358 kernel_init(FREAD | FWRITE); 3359 3360 /* 3361 * Create the storage pool. 3362 */ 3363 (void) spa_destroy(pool); 3364 ztest_shared->zs_vdev_primaries = 0; 3365 nvroot = make_vdev_root(zopt_vdev_size, 0, zopt_raidz, zopt_mirrors, 1); 3366 error = spa_create(pool, nvroot, NULL, NULL); 3367 nvlist_free(nvroot); 3368 3369 if (error) 3370 fatal(0, "spa_create() = %d", error); 3371 error = spa_open(pool, &spa, FTAG); 3372 if (error) 3373 fatal(0, "spa_open() = %d", error); 3374 3375 if (zopt_verbose >= 3) 3376 show_pool_stats(spa); 3377 3378 spa_close(spa, FTAG); 3379 3380 kernel_fini(); 3381 } 3382 3383 int 3384 main(int argc, char **argv) 3385 { 3386 int kills = 0; 3387 int iters = 0; 3388 int i, f; 3389 ztest_shared_t *zs; 3390 ztest_info_t *zi; 3391 char timebuf[100]; 3392 char numbuf[6]; 3393 3394 (void) setvbuf(stdout, NULL, _IOLBF, 0); 3395 3396 /* Override location of zpool.cache */ 3397 spa_config_path = "/tmp/zpool.cache"; 3398 3399 ztest_random_fd = open("/dev/urandom", O_RDONLY); 3400 3401 process_options(argc, argv); 3402 3403 argc -= optind; 3404 argv += optind; 3405 3406 dprintf_setup(&argc, argv); 3407 3408 /* 3409 * Blow away any existing copy of zpool.cache 3410 */ 3411 if (zopt_init != 0) 3412 (void) remove("/tmp/zpool.cache"); 3413 3414 zs = ztest_shared = (void *)mmap(0, 3415 P2ROUNDUP(sizeof (ztest_shared_t), getpagesize()), 3416 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); 3417 3418 if (zopt_verbose >= 1) { 3419 (void) printf("%llu vdevs, %d datasets, %d threads," 3420 " %llu seconds...\n", 3421 (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads, 3422 (u_longlong_t)zopt_time); 3423 } 3424 3425 /* 3426 * Create and initialize our storage pool. 3427 */ 3428 for (i = 1; i <= zopt_init; i++) { 3429 bzero(zs, sizeof (ztest_shared_t)); 3430 if (zopt_verbose >= 3 && zopt_init != 1) 3431 (void) printf("ztest_init(), pass %d\n", i); 3432 ztest_init(zopt_pool); 3433 } 3434 3435 /* 3436 * Initialize the call targets for each function. 3437 */ 3438 for (f = 0; f < ZTEST_FUNCS; f++) { 3439 zi = &zs->zs_info[f]; 3440 3441 *zi = ztest_info[f]; 3442 3443 if (*zi->zi_interval == 0) 3444 zi->zi_call_target = UINT64_MAX; 3445 else 3446 zi->zi_call_target = zopt_time / *zi->zi_interval; 3447 } 3448 3449 zs->zs_start_time = gethrtime(); 3450 zs->zs_stop_time = zs->zs_start_time + zopt_time * NANOSEC; 3451 3452 /* 3453 * Run the tests in a loop. These tests include fault injection 3454 * to verify that self-healing data works, and forced crashes 3455 * to verify that we never lose on-disk consistency. 3456 */ 3457 while (gethrtime() < zs->zs_stop_time) { 3458 int status; 3459 pid_t pid; 3460 char *tmp; 3461 3462 /* 3463 * Initialize the workload counters for each function. 3464 */ 3465 for (f = 0; f < ZTEST_FUNCS; f++) { 3466 zi = &zs->zs_info[f]; 3467 zi->zi_calls = 0; 3468 zi->zi_call_time = 0; 3469 } 3470 3471 pid = fork(); 3472 3473 if (pid == -1) 3474 fatal(1, "fork failed"); 3475 3476 if (pid == 0) { /* child */ 3477 struct rlimit rl = { 1024, 1024 }; 3478 (void) setrlimit(RLIMIT_NOFILE, &rl); 3479 (void) enable_extended_FILE_stdio(-1, -1); 3480 ztest_run(zopt_pool); 3481 exit(0); 3482 } 3483 3484 while (waitpid(pid, &status, 0) != pid) 3485 continue; 3486 3487 if (WIFEXITED(status)) { 3488 if (WEXITSTATUS(status) != 0) { 3489 (void) fprintf(stderr, 3490 "child exited with code %d\n", 3491 WEXITSTATUS(status)); 3492 exit(2); 3493 } 3494 } else if (WIFSIGNALED(status)) { 3495 if (WTERMSIG(status) != SIGKILL) { 3496 (void) fprintf(stderr, 3497 "child died with signal %d\n", 3498 WTERMSIG(status)); 3499 exit(3); 3500 } 3501 kills++; 3502 } else { 3503 (void) fprintf(stderr, "something strange happened " 3504 "to child\n"); 3505 exit(4); 3506 } 3507 3508 iters++; 3509 3510 if (zopt_verbose >= 1) { 3511 hrtime_t now = gethrtime(); 3512 3513 now = MIN(now, zs->zs_stop_time); 3514 print_time(zs->zs_stop_time - now, timebuf); 3515 nicenum(zs->zs_space, numbuf); 3516 3517 (void) printf("Pass %3d, %8s, %3llu ENOSPC, " 3518 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", 3519 iters, 3520 WIFEXITED(status) ? "Complete" : "SIGKILL", 3521 (u_longlong_t)zs->zs_enospc_count, 3522 100.0 * zs->zs_alloc / zs->zs_space, 3523 numbuf, 3524 100.0 * (now - zs->zs_start_time) / 3525 (zopt_time * NANOSEC), timebuf); 3526 } 3527 3528 if (zopt_verbose >= 2) { 3529 (void) printf("\nWorkload summary:\n\n"); 3530 (void) printf("%7s %9s %s\n", 3531 "Calls", "Time", "Function"); 3532 (void) printf("%7s %9s %s\n", 3533 "-----", "----", "--------"); 3534 for (f = 0; f < ZTEST_FUNCS; f++) { 3535 Dl_info dli; 3536 3537 zi = &zs->zs_info[f]; 3538 print_time(zi->zi_call_time, timebuf); 3539 (void) dladdr((void *)zi->zi_func, &dli); 3540 (void) printf("%7llu %9s %s\n", 3541 (u_longlong_t)zi->zi_calls, timebuf, 3542 dli.dli_sname); 3543 } 3544 (void) printf("\n"); 3545 } 3546 3547 /* 3548 * It's possible that we killed a child during a rename test, in 3549 * which case we'll have a 'ztest_tmp' pool lying around instead 3550 * of 'ztest'. Do a blind rename in case this happened. 3551 */ 3552 tmp = umem_alloc(strlen(zopt_pool) + 5, UMEM_NOFAIL); 3553 (void) strcpy(tmp, zopt_pool); 3554 (void) strcat(tmp, "_tmp"); 3555 kernel_init(FREAD | FWRITE); 3556 (void) spa_rename(tmp, zopt_pool); 3557 kernel_fini(); 3558 umem_free(tmp, strlen(tmp) + 1); 3559 } 3560 3561 ztest_verify_blocks(zopt_pool); 3562 3563 if (zopt_verbose >= 1) { 3564 (void) printf("%d killed, %d completed, %.0f%% kill rate\n", 3565 kills, iters - kills, (100.0 * kills) / MAX(1, iters)); 3566 } 3567 3568 return (0); 3569 } 3570