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