ztest.c revision 168676
1153838Sdfr/* 2153838Sdfr * CDDL HEADER START 3153838Sdfr * 4178828Sdfr * The contents of this file are subject to the terms of the 5178828Sdfr * Common Development and Distribution License (the "License"). 6178828Sdfr * You may not use this file except in compliance with the License. 7178828Sdfr * 8153838Sdfr * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9153838Sdfr * or http://www.opensolaris.org/os/licensing. 10178828Sdfr * See the License for the specific language governing permissions 11153838Sdfr * and limitations under the License. 12178828Sdfr * 13178828Sdfr * When distributing Covered Code, include this CDDL HEADER in each 14178828Sdfr * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15178828Sdfr * If applicable, add the following below this CDDL HEADER, with the 16178828Sdfr * fields enclosed by brackets "[]" replaced with your own identifying 17178828Sdfr * information: Portions Copyright [yyyy] [name of copyright owner] 18178828Sdfr * 19178828Sdfr * CDDL HEADER END 20153838Sdfr */ 21178828Sdfr/* 22178828Sdfr * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23178828Sdfr * Use is subject to license terms. 24178828Sdfr */ 25178828Sdfr 26178828Sdfr#pragma ident "%Z%%M% %I% %E% SMI" 27178828Sdfr 28153838Sdfr/* 29178828Sdfr * The objective of this program is to provide a DMU/ZAP/SPA stress test 30178828Sdfr * that runs entirely in userland, is easy to use, and easy to extend. 31153838Sdfr * 32178828Sdfr * The overall design of the ztest program is as follows: 33178828Sdfr * 34153838Sdfr * (1) For each major functional area (e.g. adding vdevs to a pool, 35153838Sdfr * creating and destroying datasets, reading and writing objects, etc) 36178828Sdfr * we have a simple routine to test that functionality. These 37178828Sdfr * individual routines do not have to do anything "stressful". 38153838Sdfr * 39178828Sdfr * (2) We turn these simple functionality tests into a stress test by 40178828Sdfr * running them all in parallel, with as many threads as desired, 41178828Sdfr * and spread across as many datasets, objects, and vdevs as desired. 42178828Sdfr * 43181344Sdfr * (3) While all this is happening, we inject faults into the pool to 44178828Sdfr * verify that self-healing data really works. 45178828Sdfr * 46178828Sdfr * (4) Every time we open a dataset, we change its checksum and compression 47178828Sdfr * functions. Thus even individual objects vary from block to block 48178828Sdfr * in which checksum they use and whether they're compressed. 49178828Sdfr * 50178828Sdfr * (5) To verify that we never lose on-disk consistency after a crash, 51178828Sdfr * we run the entire test in a child of the main process. 52178828Sdfr * At random times, the child self-immolates with a SIGKILL. 53178828Sdfr * This is the software equivalent of pulling the power cord. 54153838Sdfr * The parent then runs the test again, using the existing 55178828Sdfr * storage pool, as many times as desired. 56178828Sdfr * 57178828Sdfr * (6) To verify that we don't have future leaks or temporal incursions, 58178828Sdfr * many of the functional tests record the transaction group number 59153838Sdfr * as part of their data. When reading old data, they verify that 60178828Sdfr * the transaction group number is less than the current, open txg. 61178828Sdfr * If you add a new test, please do this if applicable. 62178828Sdfr * 63153838Sdfr * When run with no arguments, ztest runs for about five minutes and 64153838Sdfr * produces no output if successful. To get a little bit of information, 65153838Sdfr * specify -V. To get more information, specify -VV, and so on. 66153838Sdfr * 67153838Sdfr * To turn this into an overnight stress test, use -T to specify run time. 68153838Sdfr * 69153838Sdfr * You can ask more more vdevs [-v], datasets [-d], or threads [-t] 70153838Sdfr * to increase the pool capacity, fanout, and overall stress level. 71153838Sdfr * 72153838Sdfr * The -N(okill) option will suppress kills, so each child runs to completion. 73153838Sdfr * This can be useful when you're trying to distinguish temporal incursions 74153838Sdfr * from plain old race conditions. 75153838Sdfr */ 76153838Sdfr 77153838Sdfr#include <sys/zfs_context.h> 78153838Sdfr#include <sys/spa.h> 79153838Sdfr#include <sys/dmu.h> 80153838Sdfr#include <sys/txg.h> 81153838Sdfr#include <sys/zap.h> 82153838Sdfr#include <sys/dmu_traverse.h> 83153838Sdfr#include <sys/dmu_objset.h> 84153838Sdfr#include <sys/poll.h> 85153838Sdfr#include <sys/stat.h> 86153838Sdfr#include <sys/time.h> 87153838Sdfr#include <sys/wait.h> 88153838Sdfr#include <sys/mman.h> 89153838Sdfr#include <sys/resource.h> 90153838Sdfr#include <sys/zio.h> 91153838Sdfr#include <sys/zio_checksum.h> 92153838Sdfr#include <sys/zio_compress.h> 93153838Sdfr#include <sys/zil.h> 94153838Sdfr#include <sys/vdev_impl.h> 95153838Sdfr#include <sys/spa_impl.h> 96153838Sdfr#include <sys/dsl_prop.h> 97153838Sdfr#include <sys/refcount.h> 98153838Sdfr#include <stdio.h> 99153838Sdfr#include <stdio_ext.h> 100153838Sdfr#include <stdlib.h> 101153838Sdfr#include <unistd.h> 102153838Sdfr#include <signal.h> 103153838Sdfr#include <umem.h> 104153838Sdfr#include <dlfcn.h> 105153838Sdfr#include <ctype.h> 106153838Sdfr#include <math.h> 107153838Sdfr#include <errno.h> 108153838Sdfr#include <sys/fs/zfs.h> 109153838Sdfr 110static char cmdname[] = "ztest"; 111static char *zopt_pool = cmdname; 112static char *progname; 113 114static uint64_t zopt_vdevs = 5; 115static uint64_t zopt_vdevtime; 116static int zopt_ashift = SPA_MINBLOCKSHIFT; 117static int zopt_mirrors = 2; 118static int zopt_raidz = 4; 119static int zopt_raidz_parity = 1; 120static size_t zopt_vdev_size = SPA_MINDEVSIZE; 121static int zopt_datasets = 7; 122static int zopt_threads = 23; 123static uint64_t zopt_passtime = 60; /* 60 seconds */ 124static uint64_t zopt_killrate = 70; /* 70% kill rate */ 125static int zopt_verbose = 0; 126static int zopt_init = 1; 127static char *zopt_dir = "/tmp"; 128static uint64_t zopt_time = 300; /* 5 minutes */ 129static int zopt_maxfaults; 130 131typedef struct ztest_args { 132 char *za_pool; 133 objset_t *za_os; 134 zilog_t *za_zilog; 135 thread_t za_thread; 136 uint64_t za_instance; 137 uint64_t za_random; 138 uint64_t za_diroff; 139 uint64_t za_diroff_shared; 140 uint64_t za_zil_seq; 141 hrtime_t za_start; 142 hrtime_t za_stop; 143 hrtime_t za_kill; 144 traverse_handle_t *za_th; 145} ztest_args_t; 146 147typedef void ztest_func_t(ztest_args_t *); 148 149/* 150 * Note: these aren't static because we want dladdr() to work. 151 */ 152ztest_func_t ztest_dmu_read_write; 153ztest_func_t ztest_dmu_write_parallel; 154ztest_func_t ztest_dmu_object_alloc_free; 155ztest_func_t ztest_zap; 156ztest_func_t ztest_zap_parallel; 157ztest_func_t ztest_traverse; 158ztest_func_t ztest_dsl_prop_get_set; 159ztest_func_t ztest_dmu_objset_create_destroy; 160ztest_func_t ztest_dmu_snapshot_create_destroy; 161ztest_func_t ztest_spa_create_destroy; 162ztest_func_t ztest_fault_inject; 163ztest_func_t ztest_vdev_attach_detach; 164ztest_func_t ztest_vdev_LUN_growth; 165ztest_func_t ztest_vdev_add_remove; 166ztest_func_t ztest_scrub; 167ztest_func_t ztest_spa_rename; 168 169typedef struct ztest_info { 170 ztest_func_t *zi_func; /* test function */ 171 uint64_t *zi_interval; /* execute every <interval> seconds */ 172 uint64_t zi_calls; /* per-pass count */ 173 uint64_t zi_call_time; /* per-pass time */ 174 uint64_t zi_call_total; /* cumulative total */ 175 uint64_t zi_call_target; /* target cumulative total */ 176} ztest_info_t; 177 178uint64_t zopt_always = 0; /* all the time */ 179uint64_t zopt_often = 1; /* every second */ 180uint64_t zopt_sometimes = 10; /* every 10 seconds */ 181uint64_t zopt_rarely = 60; /* every 60 seconds */ 182 183ztest_info_t ztest_info[] = { 184 { ztest_dmu_read_write, &zopt_always }, 185 { ztest_dmu_write_parallel, &zopt_always }, 186 { ztest_dmu_object_alloc_free, &zopt_always }, 187 { ztest_zap, &zopt_always }, 188 { ztest_zap_parallel, &zopt_always }, 189 { ztest_traverse, &zopt_often }, 190 { ztest_dsl_prop_get_set, &zopt_sometimes }, 191 { ztest_dmu_objset_create_destroy, &zopt_sometimes }, 192 { ztest_dmu_snapshot_create_destroy, &zopt_rarely }, 193 { ztest_spa_create_destroy, &zopt_sometimes }, 194 { ztest_fault_inject, &zopt_sometimes }, 195 { ztest_spa_rename, &zopt_rarely }, 196 { ztest_vdev_attach_detach, &zopt_rarely }, 197 { ztest_vdev_LUN_growth, &zopt_rarely }, 198 { ztest_vdev_add_remove, &zopt_vdevtime }, 199 { ztest_scrub, &zopt_vdevtime }, 200}; 201 202#define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) 203 204#define ZTEST_SYNC_LOCKS 16 205 206/* 207 * Stuff we need to share writably between parent and child. 208 */ 209typedef struct ztest_shared { 210 mutex_t zs_vdev_lock; 211 rwlock_t zs_name_lock; 212 uint64_t zs_vdev_primaries; 213 uint64_t zs_enospc_count; 214 hrtime_t zs_start_time; 215 hrtime_t zs_stop_time; 216 uint64_t zs_alloc; 217 uint64_t zs_space; 218 uint64_t zs_txg; 219 ztest_info_t zs_info[ZTEST_FUNCS]; 220 mutex_t zs_sync_lock[ZTEST_SYNC_LOCKS]; 221 uint64_t zs_seq[ZTEST_SYNC_LOCKS]; 222} ztest_shared_t; 223 224typedef struct ztest_block_tag { 225 uint64_t bt_objset; 226 uint64_t bt_object; 227 uint64_t bt_offset; 228 uint64_t bt_txg; 229 uint64_t bt_thread; 230 uint64_t bt_seq; 231} ztest_block_tag_t; 232 233static char ztest_dev_template[] = "%s/%s.%llua"; 234static ztest_shared_t *ztest_shared; 235 236static int ztest_random_fd; 237static int ztest_dump_core = 1; 238 239extern uint64_t zio_gang_bang; 240extern uint16_t zio_zil_fail_shift; 241 242#define ZTEST_DIROBJ 1 243#define ZTEST_MICROZAP_OBJ 2 244#define ZTEST_FATZAP_OBJ 3 245 246#define ZTEST_DIROBJ_BLOCKSIZE (1 << 10) 247#define ZTEST_DIRSIZE 256 248 249static void usage(boolean_t) __NORETURN; 250 251/* 252 * These libumem hooks provide a reasonable set of defaults for the allocator's 253 * debugging facilities. 254 */ 255const char * 256_umem_debug_init() 257{ 258 return ("default,verbose"); /* $UMEM_DEBUG setting */ 259} 260 261const char * 262_umem_logging_init(void) 263{ 264 return ("fail,contents"); /* $UMEM_LOGGING setting */ 265} 266 267#define FATAL_MSG_SZ 1024 268 269char *fatal_msg; 270 271static void 272fatal(int do_perror, char *message, ...) 273{ 274 va_list args; 275 int save_errno = errno; 276 char buf[FATAL_MSG_SZ]; 277 278 (void) fflush(stdout); 279 280 va_start(args, message); 281 (void) sprintf(buf, "ztest: "); 282 /* LINTED */ 283 (void) vsprintf(buf + strlen(buf), message, args); 284 va_end(args); 285 if (do_perror) { 286 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), 287 ": %s", strerror(save_errno)); 288 } 289 (void) fprintf(stderr, "%s\n", buf); 290 fatal_msg = buf; /* to ease debugging */ 291 if (ztest_dump_core) 292 abort(); 293 exit(3); 294} 295 296static int 297str2shift(const char *buf) 298{ 299 const char *ends = "BKMGTPEZ"; 300 int i; 301 302 if (buf[0] == '\0') 303 return (0); 304 for (i = 0; i < strlen(ends); i++) { 305 if (toupper(buf[0]) == ends[i]) 306 break; 307 } 308 if (i == strlen(ends)) { 309 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", 310 buf); 311 usage(B_FALSE); 312 } 313 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { 314 return (10*i); 315 } 316 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); 317 usage(B_FALSE); 318 /* NOTREACHED */ 319} 320 321static uint64_t 322nicenumtoull(const char *buf) 323{ 324 char *end; 325 uint64_t val; 326 327 val = strtoull(buf, &end, 0); 328 if (end == buf) { 329 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); 330 usage(B_FALSE); 331 } else if (end[0] == '.') { 332 double fval = strtod(buf, &end); 333 fval *= pow(2, str2shift(end)); 334 if (fval > UINT64_MAX) { 335 (void) fprintf(stderr, "ztest: value too large: %s\n", 336 buf); 337 usage(B_FALSE); 338 } 339 val = (uint64_t)fval; 340 } else { 341 int shift = str2shift(end); 342 if (shift >= 64 || (val << shift) >> shift != val) { 343 (void) fprintf(stderr, "ztest: value too large: %s\n", 344 buf); 345 usage(B_FALSE); 346 } 347 val <<= shift; 348 } 349 return (val); 350} 351 352static void 353usage(boolean_t requested) 354{ 355 char nice_vdev_size[10]; 356 char nice_gang_bang[10]; 357 FILE *fp = requested ? stdout : stderr; 358 359 nicenum(zopt_vdev_size, nice_vdev_size); 360 nicenum(zio_gang_bang, nice_gang_bang); 361 362 (void) fprintf(fp, "Usage: %s\n" 363 "\t[-v vdevs (default: %llu)]\n" 364 "\t[-s size_of_each_vdev (default: %s)]\n" 365 "\t[-a alignment_shift (default: %d) (use 0 for random)]\n" 366 "\t[-m mirror_copies (default: %d)]\n" 367 "\t[-r raidz_disks (default: %d)]\n" 368 "\t[-R raidz_parity (default: %d)]\n" 369 "\t[-d datasets (default: %d)]\n" 370 "\t[-t threads (default: %d)]\n" 371 "\t[-g gang_block_threshold (default: %s)]\n" 372 "\t[-i initialize pool i times (default: %d)]\n" 373 "\t[-k kill percentage (default: %llu%%)]\n" 374 "\t[-p pool_name (default: %s)]\n" 375 "\t[-f file directory for vdev files (default: %s)]\n" 376 "\t[-V(erbose)] (use multiple times for ever more blather)\n" 377 "\t[-E(xisting)] (use existing pool instead of creating new one)\n" 378 "\t[-T time] total run time (default: %llu sec)\n" 379 "\t[-P passtime] time per pass (default: %llu sec)\n" 380 "\t[-z zil failure rate (default: fail every 2^%llu allocs)]\n" 381 "\t[-h] (print help)\n" 382 "", 383 cmdname, 384 (u_longlong_t)zopt_vdevs, /* -v */ 385 nice_vdev_size, /* -s */ 386 zopt_ashift, /* -a */ 387 zopt_mirrors, /* -m */ 388 zopt_raidz, /* -r */ 389 zopt_raidz_parity, /* -R */ 390 zopt_datasets, /* -d */ 391 zopt_threads, /* -t */ 392 nice_gang_bang, /* -g */ 393 zopt_init, /* -i */ 394 (u_longlong_t)zopt_killrate, /* -k */ 395 zopt_pool, /* -p */ 396 zopt_dir, /* -f */ 397 (u_longlong_t)zopt_time, /* -T */ 398 (u_longlong_t)zopt_passtime, /* -P */ 399 (u_longlong_t)zio_zil_fail_shift); /* -z */ 400 exit(requested ? 0 : 1); 401} 402 403static uint64_t 404ztest_random(uint64_t range) 405{ 406 uint64_t r; 407 408 if (range == 0) 409 return (0); 410 411 if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r)) 412 fatal(1, "short read from /dev/urandom"); 413 414 return (r % range); 415} 416 417static void 418ztest_record_enospc(char *s) 419{ 420 dprintf("ENOSPC doing: %s\n", s ? s : "<unknown>"); 421 ztest_shared->zs_enospc_count++; 422} 423 424static void 425process_options(int argc, char **argv) 426{ 427 int opt; 428 uint64_t value; 429 430 /* Remember program name. */ 431 progname = argv[0]; 432 433 /* By default, test gang blocks for blocks 32K and greater */ 434 zio_gang_bang = 32 << 10; 435 436 /* Default value, fail every 32nd allocation */ 437 zio_zil_fail_shift = 5; 438 439 while ((opt = getopt(argc, argv, 440 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:z:h")) != EOF) { 441 value = 0; 442 switch (opt) { 443 case 'v': 444 case 's': 445 case 'a': 446 case 'm': 447 case 'r': 448 case 'R': 449 case 'd': 450 case 't': 451 case 'g': 452 case 'i': 453 case 'k': 454 case 'T': 455 case 'P': 456 case 'z': 457 value = nicenumtoull(optarg); 458 } 459 switch (opt) { 460 case 'v': 461 zopt_vdevs = value; 462 break; 463 case 's': 464 zopt_vdev_size = MAX(SPA_MINDEVSIZE, value); 465 break; 466 case 'a': 467 zopt_ashift = value; 468 break; 469 case 'm': 470 zopt_mirrors = value; 471 break; 472 case 'r': 473 zopt_raidz = MAX(1, value); 474 break; 475 case 'R': 476 zopt_raidz_parity = MIN(MAX(value, 1), 2); 477 break; 478 case 'd': 479 zopt_datasets = MAX(1, value); 480 break; 481 case 't': 482 zopt_threads = MAX(1, value); 483 break; 484 case 'g': 485 zio_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value); 486 break; 487 case 'i': 488 zopt_init = value; 489 break; 490 case 'k': 491 zopt_killrate = value; 492 break; 493 case 'p': 494 zopt_pool = strdup(optarg); 495 break; 496 case 'f': 497 zopt_dir = strdup(optarg); 498 break; 499 case 'V': 500 zopt_verbose++; 501 break; 502 case 'E': 503 zopt_init = 0; 504 break; 505 case 'T': 506 zopt_time = value; 507 break; 508 case 'P': 509 zopt_passtime = MAX(1, value); 510 break; 511 case 'z': 512 zio_zil_fail_shift = MIN(value, 16); 513 break; 514 case 'h': 515 usage(B_TRUE); 516 break; 517 case '?': 518 default: 519 usage(B_FALSE); 520 break; 521 } 522 } 523 524 zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1); 525 526 zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time / zopt_vdevs : UINT64_MAX); 527 zopt_maxfaults = MAX(zopt_mirrors, 1) * (zopt_raidz_parity + 1) - 1; 528} 529 530static uint64_t 531ztest_get_ashift(void) 532{ 533 if (zopt_ashift == 0) 534 return (SPA_MINBLOCKSHIFT + ztest_random(3)); 535 return (zopt_ashift); 536} 537 538static nvlist_t * 539make_vdev_file(size_t size) 540{ 541 char dev_name[MAXPATHLEN]; 542 uint64_t vdev; 543 uint64_t ashift = ztest_get_ashift(); 544 int fd; 545 nvlist_t *file; 546 547 if (size == 0) { 548 (void) snprintf(dev_name, sizeof (dev_name), "%s", 549 "/dev/bogus"); 550 } else { 551 vdev = ztest_shared->zs_vdev_primaries++; 552 (void) sprintf(dev_name, ztest_dev_template, 553 zopt_dir, zopt_pool, vdev); 554 555 fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666); 556 if (fd == -1) 557 fatal(1, "can't open %s", dev_name); 558 if (ftruncate(fd, size) != 0) 559 fatal(1, "can't ftruncate %s", dev_name); 560 (void) close(fd); 561 } 562 563 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 564 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 565 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0); 566 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 567 568 return (file); 569} 570 571static nvlist_t * 572make_vdev_raidz(size_t size, int r) 573{ 574 nvlist_t *raidz, **child; 575 int c; 576 577 if (r < 2) 578 return (make_vdev_file(size)); 579 580 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); 581 582 for (c = 0; c < r; c++) 583 child[c] = make_vdev_file(size); 584 585 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); 586 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, 587 VDEV_TYPE_RAIDZ) == 0); 588 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, 589 zopt_raidz_parity) == 0); 590 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, 591 child, r) == 0); 592 593 for (c = 0; c < r; c++) 594 nvlist_free(child[c]); 595 596 umem_free(child, r * sizeof (nvlist_t *)); 597 598 return (raidz); 599} 600 601static nvlist_t * 602make_vdev_mirror(size_t size, int r, int m) 603{ 604 nvlist_t *mirror, **child; 605 int c; 606 607 if (m < 1) 608 return (make_vdev_raidz(size, r)); 609 610 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); 611 612 for (c = 0; c < m; c++) 613 child[c] = make_vdev_raidz(size, r); 614 615 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); 616 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, 617 VDEV_TYPE_MIRROR) == 0); 618 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, 619 child, m) == 0); 620 621 for (c = 0; c < m; c++) 622 nvlist_free(child[c]); 623 624 umem_free(child, m * sizeof (nvlist_t *)); 625 626 return (mirror); 627} 628 629static nvlist_t * 630make_vdev_root(size_t size, int r, int m, int t) 631{ 632 nvlist_t *root, **child; 633 int c; 634 635 ASSERT(t > 0); 636 637 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); 638 639 for (c = 0; c < t; c++) 640 child[c] = make_vdev_mirror(size, r, m); 641 642 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 643 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 644 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 645 child, t) == 0); 646 647 for (c = 0; c < t; c++) 648 nvlist_free(child[c]); 649 650 umem_free(child, t * sizeof (nvlist_t *)); 651 652 return (root); 653} 654 655static void 656ztest_set_random_blocksize(objset_t *os, uint64_t object, dmu_tx_t *tx) 657{ 658 int bs = SPA_MINBLOCKSHIFT + 659 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1); 660 int ibs = DN_MIN_INDBLKSHIFT + 661 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1); 662 int error; 663 664 error = dmu_object_set_blocksize(os, object, 1ULL << bs, ibs, tx); 665 if (error) { 666 char osname[300]; 667 dmu_objset_name(os, osname); 668 fatal(0, "dmu_object_set_blocksize('%s', %llu, %d, %d) = %d", 669 osname, object, 1 << bs, ibs, error); 670 } 671} 672 673static uint8_t 674ztest_random_checksum(void) 675{ 676 uint8_t checksum; 677 678 do { 679 checksum = ztest_random(ZIO_CHECKSUM_FUNCTIONS); 680 } while (zio_checksum_table[checksum].ci_zbt); 681 682 if (checksum == ZIO_CHECKSUM_OFF) 683 checksum = ZIO_CHECKSUM_ON; 684 685 return (checksum); 686} 687 688static uint8_t 689ztest_random_compress(void) 690{ 691 return ((uint8_t)ztest_random(ZIO_COMPRESS_FUNCTIONS)); 692} 693 694typedef struct ztest_replay { 695 objset_t *zr_os; 696 uint64_t zr_assign; 697} ztest_replay_t; 698 699static int 700ztest_replay_create(ztest_replay_t *zr, lr_create_t *lr, boolean_t byteswap) 701{ 702 objset_t *os = zr->zr_os; 703 dmu_tx_t *tx; 704 int error; 705 706 if (byteswap) 707 byteswap_uint64_array(lr, sizeof (*lr)); 708 709 tx = dmu_tx_create(os); 710 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 711 error = dmu_tx_assign(tx, zr->zr_assign); 712 if (error) { 713 dmu_tx_abort(tx); 714 return (error); 715 } 716 717 error = dmu_object_claim(os, lr->lr_doid, lr->lr_mode, 0, 718 DMU_OT_NONE, 0, tx); 719 ASSERT3U(error, ==, 0); 720 dmu_tx_commit(tx); 721 722 if (zopt_verbose >= 5) { 723 char osname[MAXNAMELEN]; 724 dmu_objset_name(os, osname); 725 (void) printf("replay create of %s object %llu" 726 " in txg %llu = %d\n", 727 osname, (u_longlong_t)lr->lr_doid, 728 (u_longlong_t)zr->zr_assign, error); 729 } 730 731 return (error); 732} 733 734static int 735ztest_replay_remove(ztest_replay_t *zr, lr_remove_t *lr, boolean_t byteswap) 736{ 737 objset_t *os = zr->zr_os; 738 dmu_tx_t *tx; 739 int error; 740 741 if (byteswap) 742 byteswap_uint64_array(lr, sizeof (*lr)); 743 744 tx = dmu_tx_create(os); 745 dmu_tx_hold_free(tx, lr->lr_doid, 0, DMU_OBJECT_END); 746 error = dmu_tx_assign(tx, zr->zr_assign); 747 if (error) { 748 dmu_tx_abort(tx); 749 return (error); 750 } 751 752 error = dmu_object_free(os, lr->lr_doid, tx); 753 dmu_tx_commit(tx); 754 755 return (error); 756} 757 758zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { 759 NULL, /* 0 no such transaction type */ 760 ztest_replay_create, /* TX_CREATE */ 761 NULL, /* TX_MKDIR */ 762 NULL, /* TX_MKXATTR */ 763 NULL, /* TX_SYMLINK */ 764 ztest_replay_remove, /* TX_REMOVE */ 765 NULL, /* TX_RMDIR */ 766 NULL, /* TX_LINK */ 767 NULL, /* TX_RENAME */ 768 NULL, /* TX_WRITE */ 769 NULL, /* TX_TRUNCATE */ 770 NULL, /* TX_SETATTR */ 771 NULL, /* TX_ACL */ 772}; 773 774/* 775 * Verify that we can't destroy an active pool, create an existing pool, 776 * or create a pool with a bad vdev spec. 777 */ 778void 779ztest_spa_create_destroy(ztest_args_t *za) 780{ 781 int error; 782 spa_t *spa; 783 nvlist_t *nvroot; 784 785 /* 786 * Attempt to create using a bad file. 787 */ 788 nvroot = make_vdev_root(0, 0, 0, 1); 789 error = spa_create("ztest_bad_file", nvroot, NULL); 790 nvlist_free(nvroot); 791 if (error != ENOENT) 792 fatal(0, "spa_create(bad_file) = %d", error); 793 794 /* 795 * Attempt to create using a bad mirror. 796 */ 797 nvroot = make_vdev_root(0, 0, 2, 1); 798 error = spa_create("ztest_bad_mirror", nvroot, NULL); 799 nvlist_free(nvroot); 800 if (error != ENOENT) 801 fatal(0, "spa_create(bad_mirror) = %d", error); 802 803 /* 804 * Attempt to create an existing pool. It shouldn't matter 805 * what's in the nvroot; we should fail with EEXIST. 806 */ 807 (void) rw_rdlock(&ztest_shared->zs_name_lock); 808 nvroot = make_vdev_root(0, 0, 0, 1); 809 error = spa_create(za->za_pool, nvroot, NULL); 810 nvlist_free(nvroot); 811 if (error != EEXIST) 812 fatal(0, "spa_create(whatever) = %d", error); 813 814 error = spa_open(za->za_pool, &spa, FTAG); 815 if (error) 816 fatal(0, "spa_open() = %d", error); 817 818 error = spa_destroy(za->za_pool); 819 if (error != EBUSY) 820 fatal(0, "spa_destroy() = %d", error); 821 822 spa_close(spa, FTAG); 823 (void) rw_unlock(&ztest_shared->zs_name_lock); 824} 825 826/* 827 * Verify that vdev_add() works as expected. 828 */ 829void 830ztest_vdev_add_remove(ztest_args_t *za) 831{ 832 spa_t *spa = dmu_objset_spa(za->za_os); 833 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 834 nvlist_t *nvroot; 835 int error; 836 837 if (zopt_verbose >= 6) 838 (void) printf("adding vdev\n"); 839 840 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 841 842 spa_config_enter(spa, RW_READER, FTAG); 843 844 ztest_shared->zs_vdev_primaries = 845 spa->spa_root_vdev->vdev_children * leaves; 846 847 spa_config_exit(spa, FTAG); 848 849 nvroot = make_vdev_root(zopt_vdev_size, zopt_raidz, zopt_mirrors, 1); 850 error = spa_vdev_add(spa, nvroot); 851 nvlist_free(nvroot); 852 853 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 854 855 if (error == ENOSPC) 856 ztest_record_enospc("spa_vdev_add"); 857 else if (error != 0) 858 fatal(0, "spa_vdev_add() = %d", error); 859 860 if (zopt_verbose >= 6) 861 (void) printf("spa_vdev_add = %d, as expected\n", error); 862} 863 864static vdev_t * 865vdev_lookup_by_path(vdev_t *vd, const char *path) 866{ 867 int c; 868 vdev_t *mvd; 869 870 if (vd->vdev_path != NULL) { 871 if (vd->vdev_wholedisk == 1) { 872 /* 873 * For whole disks, the internal path has 's0', but the 874 * path passed in by the user doesn't. 875 */ 876 if (strlen(path) == strlen(vd->vdev_path) - 2 && 877 strncmp(path, vd->vdev_path, strlen(path)) == 0) 878 return (vd); 879 } else if (strcmp(path, vd->vdev_path) == 0) { 880 return (vd); 881 } 882 } 883 884 for (c = 0; c < vd->vdev_children; c++) 885 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != 886 NULL) 887 return (mvd); 888 889 return (NULL); 890} 891 892/* 893 * Verify that we can attach and detach devices. 894 */ 895void 896ztest_vdev_attach_detach(ztest_args_t *za) 897{ 898 spa_t *spa = dmu_objset_spa(za->za_os); 899 vdev_t *rvd = spa->spa_root_vdev; 900 vdev_t *oldvd, *newvd, *pvd; 901 nvlist_t *root, *file; 902 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 903 uint64_t leaf, top; 904 uint64_t ashift = ztest_get_ashift(); 905 size_t oldsize, newsize; 906 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; 907 int replacing; 908 int error, expected_error; 909 int fd; 910 911 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 912 913 spa_config_enter(spa, RW_READER, FTAG); 914 915 /* 916 * Decide whether to do an attach or a replace. 917 */ 918 replacing = ztest_random(2); 919 920 /* 921 * Pick a random top-level vdev. 922 */ 923 top = ztest_random(rvd->vdev_children); 924 925 /* 926 * Pick a random leaf within it. 927 */ 928 leaf = ztest_random(leaves); 929 930 /* 931 * Generate the path to this leaf. The filename will end with 'a'. 932 * We'll alternate replacements with a filename that ends with 'b'. 933 */ 934 (void) snprintf(oldpath, sizeof (oldpath), 935 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf); 936 937 bcopy(oldpath, newpath, MAXPATHLEN); 938 939 /* 940 * If the 'a' file isn't part of the pool, the 'b' file must be. 941 */ 942 if (vdev_lookup_by_path(rvd, oldpath) == NULL) 943 oldpath[strlen(oldpath) - 1] = 'b'; 944 else 945 newpath[strlen(newpath) - 1] = 'b'; 946 947 /* 948 * Now oldpath represents something that's already in the pool, 949 * and newpath is the thing we'll try to attach. 950 */ 951 oldvd = vdev_lookup_by_path(rvd, oldpath); 952 newvd = vdev_lookup_by_path(rvd, newpath); 953 ASSERT(oldvd != NULL); 954 pvd = oldvd->vdev_parent; 955 956 /* 957 * Make newsize a little bigger or smaller than oldsize. 958 * If it's smaller, the attach should fail. 959 * If it's larger, and we're doing a replace, 960 * we should get dynamic LUN growth when we're done. 961 */ 962 oldsize = vdev_get_rsize(oldvd); 963 newsize = 10 * oldsize / (9 + ztest_random(3)); 964 965 /* 966 * If pvd is not a mirror or root, the attach should fail with ENOTSUP, 967 * unless it's a replace; in that case any non-replacing parent is OK. 968 * 969 * If newvd is already part of the pool, it should fail with EBUSY. 970 * 971 * If newvd is too small, it should fail with EOVERFLOW. 972 */ 973 if (newvd != NULL) 974 expected_error = EBUSY; 975 else if (pvd->vdev_ops != &vdev_mirror_ops && 976 pvd->vdev_ops != &vdev_root_ops && 977 (!replacing || pvd->vdev_ops == &vdev_replacing_ops)) 978 expected_error = ENOTSUP; 979 else if (newsize < oldsize) 980 expected_error = EOVERFLOW; 981 else if (ashift > oldvd->vdev_top->vdev_ashift) 982 expected_error = EDOM; 983 else 984 expected_error = 0; 985 986 /* 987 * If newvd isn't already part of the pool, create it. 988 */ 989 if (newvd == NULL) { 990 fd = open(newpath, O_RDWR | O_CREAT | O_TRUNC, 0666); 991 if (fd == -1) 992 fatal(1, "can't open %s", newpath); 993 if (ftruncate(fd, newsize) != 0) 994 fatal(1, "can't ftruncate %s", newpath); 995 (void) close(fd); 996 } 997 998 spa_config_exit(spa, FTAG); 999 1000 /* 1001 * Build the nvlist describing newpath. 1002 */ 1003 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 1004 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 1005 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, newpath) == 0); 1006 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 1007 1008 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 1009 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 1010 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 1011 &file, 1) == 0); 1012 1013 error = spa_vdev_attach(spa, oldvd->vdev_guid, root, replacing); 1014 1015 nvlist_free(file); 1016 nvlist_free(root); 1017 1018 /* 1019 * If our parent was the replacing vdev, but the replace completed, 1020 * then instead of failing with ENOTSUP we may either succeed, 1021 * fail with ENODEV, or fail with EOVERFLOW. 1022 */ 1023 if (expected_error == ENOTSUP && 1024 (error == 0 || error == ENODEV || error == EOVERFLOW)) 1025 expected_error = error; 1026 1027 /* 1028 * If someone grew the LUN, the replacement may be too small. 1029 */ 1030 if (error == EOVERFLOW) 1031 expected_error = error; 1032 1033 if (error != expected_error) { 1034 fatal(0, "attach (%s, %s, %d) returned %d, expected %d", 1035 oldpath, newpath, replacing, error, expected_error); 1036 } 1037 1038 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 1039} 1040 1041/* 1042 * Verify that dynamic LUN growth works as expected. 1043 */ 1044/* ARGSUSED */ 1045void 1046ztest_vdev_LUN_growth(ztest_args_t *za) 1047{ 1048 spa_t *spa = dmu_objset_spa(za->za_os); 1049 char dev_name[MAXPATHLEN]; 1050 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 1051 uint64_t vdev; 1052 size_t fsize; 1053 int fd; 1054 1055 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 1056 1057 /* 1058 * Pick a random leaf vdev. 1059 */ 1060 spa_config_enter(spa, RW_READER, FTAG); 1061 vdev = ztest_random(spa->spa_root_vdev->vdev_children * leaves); 1062 spa_config_exit(spa, FTAG); 1063 1064 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 1065 1066 if ((fd = open(dev_name, O_RDWR)) != -1) { 1067 /* 1068 * Determine the size. 1069 */ 1070 fsize = lseek(fd, 0, SEEK_END); 1071 1072 /* 1073 * If it's less than 2x the original size, grow by around 3%. 1074 */ 1075 if (fsize < 2 * zopt_vdev_size) { 1076 size_t newsize = fsize + ztest_random(fsize / 32); 1077 (void) ftruncate(fd, newsize); 1078 if (zopt_verbose >= 6) { 1079 (void) printf("%s grew from %lu to %lu bytes\n", 1080 dev_name, (ulong_t)fsize, (ulong_t)newsize); 1081 } 1082 } 1083 (void) close(fd); 1084 } 1085 1086 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 1087} 1088 1089/* ARGSUSED */ 1090static void 1091ztest_create_cb(objset_t *os, void *arg, dmu_tx_t *tx) 1092{ 1093 /* 1094 * Create the directory object. 1095 */ 1096 VERIFY(dmu_object_claim(os, ZTEST_DIROBJ, 1097 DMU_OT_UINT64_OTHER, ZTEST_DIROBJ_BLOCKSIZE, 1098 DMU_OT_UINT64_OTHER, sizeof (ztest_block_tag_t), tx) == 0); 1099 1100 VERIFY(zap_create_claim(os, ZTEST_MICROZAP_OBJ, 1101 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 1102 1103 VERIFY(zap_create_claim(os, ZTEST_FATZAP_OBJ, 1104 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 1105} 1106 1107/* ARGSUSED */ 1108static int 1109ztest_destroy_cb(char *name, void *arg) 1110{ 1111 objset_t *os; 1112 dmu_object_info_t doi; 1113 int error; 1114 1115 /* 1116 * Verify that the dataset contains a directory object. 1117 */ 1118 error = dmu_objset_open(name, DMU_OST_OTHER, 1119 DS_MODE_STANDARD | DS_MODE_READONLY, &os); 1120 ASSERT3U(error, ==, 0); 1121 error = dmu_object_info(os, ZTEST_DIROBJ, &doi); 1122 if (error != ENOENT) { 1123 /* We could have crashed in the middle of destroying it */ 1124 ASSERT3U(error, ==, 0); 1125 ASSERT3U(doi.doi_type, ==, DMU_OT_UINT64_OTHER); 1126 ASSERT3S(doi.doi_physical_blks, >=, 0); 1127 } 1128 dmu_objset_close(os); 1129 1130 /* 1131 * Destroy the dataset. 1132 */ 1133 error = dmu_objset_destroy(name); 1134 ASSERT3U(error, ==, 0); 1135 return (0); 1136} 1137 1138/* 1139 * Verify that dmu_objset_{create,destroy,open,close} work as expected. 1140 */ 1141static uint64_t 1142ztest_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t object, int mode) 1143{ 1144 itx_t *itx; 1145 lr_create_t *lr; 1146 size_t namesize; 1147 char name[24]; 1148 1149 (void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object); 1150 namesize = strlen(name) + 1; 1151 1152 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize + 1153 ztest_random(ZIL_MAX_BLKSZ)); 1154 lr = (lr_create_t *)&itx->itx_lr; 1155 bzero(lr + 1, lr->lr_common.lrc_reclen - sizeof (*lr)); 1156 lr->lr_doid = object; 1157 lr->lr_foid = 0; 1158 lr->lr_mode = mode; 1159 lr->lr_uid = 0; 1160 lr->lr_gid = 0; 1161 lr->lr_gen = dmu_tx_get_txg(tx); 1162 lr->lr_crtime[0] = time(NULL); 1163 lr->lr_crtime[1] = 0; 1164 lr->lr_rdev = 0; 1165 bcopy(name, (char *)(lr + 1), namesize); 1166 1167 return (zil_itx_assign(zilog, itx, tx)); 1168} 1169 1170void 1171ztest_dmu_objset_create_destroy(ztest_args_t *za) 1172{ 1173 int error; 1174 objset_t *os; 1175 char name[100]; 1176 int mode, basemode, expected_error; 1177 zilog_t *zilog; 1178 uint64_t seq; 1179 uint64_t objects; 1180 ztest_replay_t zr; 1181 1182 (void) rw_rdlock(&ztest_shared->zs_name_lock); 1183 (void) snprintf(name, 100, "%s/%s_temp_%llu", za->za_pool, za->za_pool, 1184 (u_longlong_t)za->za_instance); 1185 1186 basemode = DS_MODE_LEVEL(za->za_instance); 1187 if (basemode == DS_MODE_NONE) 1188 basemode++; 1189 1190 /* 1191 * If this dataset exists from a previous run, process its replay log 1192 * half of the time. If we don't replay it, then dmu_objset_destroy() 1193 * (invoked from ztest_destroy_cb() below) should just throw it away. 1194 */ 1195 if (ztest_random(2) == 0 && 1196 dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_PRIMARY, &os) == 0) { 1197 zr.zr_os = os; 1198 zil_replay(os, &zr, &zr.zr_assign, ztest_replay_vector); 1199 dmu_objset_close(os); 1200 } 1201 1202 /* 1203 * There may be an old instance of the dataset we're about to 1204 * create lying around from a previous run. If so, destroy it 1205 * and all of its snapshots. 1206 */ 1207 (void) dmu_objset_find(name, ztest_destroy_cb, NULL, 1208 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 1209 1210 /* 1211 * Verify that the destroyed dataset is no longer in the namespace. 1212 */ 1213 error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os); 1214 if (error != ENOENT) 1215 fatal(1, "dmu_objset_open(%s) found destroyed dataset %p", 1216 name, os); 1217 1218 /* 1219 * Verify that we can create a new dataset. 1220 */ 1221 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, ztest_create_cb, 1222 NULL); 1223 if (error) { 1224 if (error == ENOSPC) { 1225 ztest_record_enospc("dmu_objset_create"); 1226 (void) rw_unlock(&ztest_shared->zs_name_lock); 1227 return; 1228 } 1229 fatal(0, "dmu_objset_create(%s) = %d", name, error); 1230 } 1231 1232 error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os); 1233 if (error) { 1234 fatal(0, "dmu_objset_open(%s) = %d", name, error); 1235 } 1236 1237 /* 1238 * Open the intent log for it. 1239 */ 1240 zilog = zil_open(os, NULL); 1241 1242 /* 1243 * Put a random number of objects in there. 1244 */ 1245 objects = ztest_random(20); 1246 seq = 0; 1247 while (objects-- != 0) { 1248 uint64_t object; 1249 dmu_tx_t *tx = dmu_tx_create(os); 1250 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, sizeof (name)); 1251 error = dmu_tx_assign(tx, TXG_WAIT); 1252 if (error) { 1253 dmu_tx_abort(tx); 1254 } else { 1255 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1256 DMU_OT_NONE, 0, tx); 1257 ztest_set_random_blocksize(os, object, tx); 1258 seq = ztest_log_create(zilog, tx, object, 1259 DMU_OT_UINT64_OTHER); 1260 dmu_write(os, object, 0, sizeof (name), name, tx); 1261 dmu_tx_commit(tx); 1262 } 1263 if (ztest_random(5) == 0) { 1264 zil_commit(zilog, seq, object); 1265 } 1266 if (ztest_random(100) == 0) { 1267 error = zil_suspend(zilog); 1268 if (error == 0) { 1269 zil_resume(zilog); 1270 } 1271 } 1272 } 1273 1274 /* 1275 * Verify that we cannot create an existing dataset. 1276 */ 1277 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, NULL, NULL); 1278 if (error != EEXIST) 1279 fatal(0, "created existing dataset, error = %d", error); 1280 1281 /* 1282 * Verify that multiple dataset opens are allowed, but only when 1283 * the new access mode is compatible with the base mode. 1284 * We use a mixture of typed and typeless opens, and when the 1285 * open succeeds, verify that the discovered type is correct. 1286 */ 1287 for (mode = DS_MODE_STANDARD; mode < DS_MODE_LEVELS; mode++) { 1288 objset_t *os2; 1289 error = dmu_objset_open(name, DMU_OST_OTHER, mode, &os2); 1290 expected_error = (basemode + mode < DS_MODE_LEVELS) ? 0 : EBUSY; 1291 if (error != expected_error) 1292 fatal(0, "dmu_objset_open('%s') = %d, expected %d", 1293 name, error, expected_error); 1294 if (error == 0) 1295 dmu_objset_close(os2); 1296 } 1297 1298 zil_close(zilog); 1299 dmu_objset_close(os); 1300 1301 error = dmu_objset_destroy(name); 1302 if (error) 1303 fatal(0, "dmu_objset_destroy(%s) = %d", name, error); 1304 1305 (void) rw_unlock(&ztest_shared->zs_name_lock); 1306} 1307 1308/* 1309 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. 1310 */ 1311void 1312ztest_dmu_snapshot_create_destroy(ztest_args_t *za) 1313{ 1314 int error; 1315 objset_t *os = za->za_os; 1316 char snapname[100]; 1317 char osname[MAXNAMELEN]; 1318 1319 (void) rw_rdlock(&ztest_shared->zs_name_lock); 1320 dmu_objset_name(os, osname); 1321 (void) snprintf(snapname, 100, "%s@%llu", osname, 1322 (u_longlong_t)za->za_instance); 1323 1324 error = dmu_objset_destroy(snapname); 1325 if (error != 0 && error != ENOENT) 1326 fatal(0, "dmu_objset_destroy() = %d", error); 1327 error = dmu_objset_snapshot(osname, strchr(snapname, '@')+1, FALSE); 1328 if (error == ENOSPC) 1329 ztest_record_enospc("dmu_take_snapshot"); 1330 else if (error != 0 && error != EEXIST) 1331 fatal(0, "dmu_take_snapshot() = %d", error); 1332 (void) rw_unlock(&ztest_shared->zs_name_lock); 1333} 1334 1335#define ZTEST_TRAVERSE_BLOCKS 1000 1336 1337static int 1338ztest_blk_cb(traverse_blk_cache_t *bc, spa_t *spa, void *arg) 1339{ 1340 ztest_args_t *za = arg; 1341 zbookmark_t *zb = &bc->bc_bookmark; 1342 blkptr_t *bp = &bc->bc_blkptr; 1343 dnode_phys_t *dnp = bc->bc_dnode; 1344 traverse_handle_t *th = za->za_th; 1345 uint64_t size = BP_GET_LSIZE(bp); 1346 1347 /* 1348 * Level -1 indicates the objset_phys_t or something in its intent log. 1349 */ 1350 if (zb->zb_level == -1) { 1351 if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { 1352 ASSERT3U(zb->zb_object, ==, 0); 1353 ASSERT3U(zb->zb_blkid, ==, 0); 1354 ASSERT3U(size, ==, sizeof (objset_phys_t)); 1355 za->za_zil_seq = 0; 1356 } else if (BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) { 1357 ASSERT3U(zb->zb_object, ==, 0); 1358 ASSERT3U(zb->zb_blkid, >, za->za_zil_seq); 1359 za->za_zil_seq = zb->zb_blkid; 1360 } else { 1361 ASSERT3U(zb->zb_object, !=, 0); /* lr_write_t */ 1362 } 1363 1364 return (0); 1365 } 1366 1367 ASSERT(dnp != NULL); 1368 1369 if (bc->bc_errno) 1370 return (ERESTART); 1371 1372 /* 1373 * Once in a while, abort the traverse. We only do this to odd 1374 * instance numbers to ensure that even ones can run to completion. 1375 */ 1376 if ((za->za_instance & 1) && ztest_random(10000) == 0) 1377 return (EINTR); 1378 1379 if (bp->blk_birth == 0) { 1380 ASSERT(th->th_advance & ADVANCE_HOLES); 1381 return (0); 1382 } 1383 1384 if (zb->zb_level == 0 && !(th->th_advance & ADVANCE_DATA) && 1385 bc == &th->th_cache[ZB_DN_CACHE][0]) { 1386 ASSERT(bc->bc_data == NULL); 1387 return (0); 1388 } 1389 1390 ASSERT(bc->bc_data != NULL); 1391 1392 /* 1393 * This is an expensive question, so don't ask it too often. 1394 */ 1395 if (((za->za_random ^ th->th_callbacks) & 0xff) == 0) { 1396 void *xbuf = umem_alloc(size, UMEM_NOFAIL); 1397 if (arc_tryread(spa, bp, xbuf) == 0) { 1398 ASSERT(bcmp(bc->bc_data, xbuf, size) == 0); 1399 } 1400 umem_free(xbuf, size); 1401 } 1402 1403 if (zb->zb_level > 0) { 1404 ASSERT3U(size, ==, 1ULL << dnp->dn_indblkshift); 1405 return (0); 1406 } 1407 1408 ASSERT(zb->zb_level == 0); 1409 ASSERT3U(size, ==, dnp->dn_datablkszsec << DEV_BSHIFT); 1410 1411 return (0); 1412} 1413 1414/* 1415 * Verify that live pool traversal works. 1416 */ 1417void 1418ztest_traverse(ztest_args_t *za) 1419{ 1420 spa_t *spa = dmu_objset_spa(za->za_os); 1421 traverse_handle_t *th = za->za_th; 1422 int rc, advance; 1423 uint64_t cbstart, cblimit; 1424 1425 if (th == NULL) { 1426 advance = 0; 1427 1428 if (ztest_random(2) == 0) 1429 advance |= ADVANCE_PRE; 1430 1431 if (ztest_random(2) == 0) 1432 advance |= ADVANCE_PRUNE; 1433 1434 if (ztest_random(2) == 0) 1435 advance |= ADVANCE_DATA; 1436 1437 if (ztest_random(2) == 0) 1438 advance |= ADVANCE_HOLES; 1439 1440 if (ztest_random(2) == 0) 1441 advance |= ADVANCE_ZIL; 1442 1443 th = za->za_th = traverse_init(spa, ztest_blk_cb, za, advance, 1444 ZIO_FLAG_CANFAIL); 1445 1446 traverse_add_pool(th, 0, -1ULL); 1447 } 1448 1449 advance = th->th_advance; 1450 cbstart = th->th_callbacks; 1451 cblimit = cbstart + ((advance & ADVANCE_DATA) ? 100 : 1000); 1452 1453 while ((rc = traverse_more(th)) == EAGAIN && th->th_callbacks < cblimit) 1454 continue; 1455 1456 if (zopt_verbose >= 5) 1457 (void) printf("traverse %s%s%s%s %llu blocks to " 1458 "<%llu, %llu, %lld, %llx>%s\n", 1459 (advance & ADVANCE_PRE) ? "pre" : "post", 1460 (advance & ADVANCE_PRUNE) ? "|prune" : "", 1461 (advance & ADVANCE_DATA) ? "|data" : "", 1462 (advance & ADVANCE_HOLES) ? "|holes" : "", 1463 (u_longlong_t)(th->th_callbacks - cbstart), 1464 (u_longlong_t)th->th_lastcb.zb_objset, 1465 (u_longlong_t)th->th_lastcb.zb_object, 1466 (u_longlong_t)th->th_lastcb.zb_level, 1467 (u_longlong_t)th->th_lastcb.zb_blkid, 1468 rc == 0 ? " [done]" : 1469 rc == EINTR ? " [aborted]" : 1470 rc == EAGAIN ? "" : 1471 strerror(rc)); 1472 1473 if (rc != EAGAIN) { 1474 if (rc != 0 && rc != EINTR) 1475 fatal(0, "traverse_more(%p) = %d", th, rc); 1476 traverse_fini(th); 1477 za->za_th = NULL; 1478 } 1479} 1480 1481/* 1482 * Verify that dmu_object_{alloc,free} work as expected. 1483 */ 1484void 1485ztest_dmu_object_alloc_free(ztest_args_t *za) 1486{ 1487 objset_t *os = za->za_os; 1488 dmu_buf_t *db; 1489 dmu_tx_t *tx; 1490 uint64_t batchobj, object, batchsize, endoff, temp; 1491 int b, c, error, bonuslen; 1492 dmu_object_info_t doi; 1493 char osname[MAXNAMELEN]; 1494 1495 dmu_objset_name(os, osname); 1496 1497 endoff = -8ULL; 1498 batchsize = 2; 1499 1500 /* 1501 * Create a batch object if necessary, and record it in the directory. 1502 */ 1503 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 1504 sizeof (uint64_t), &batchobj)); 1505 if (batchobj == 0) { 1506 tx = dmu_tx_create(os); 1507 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, 1508 sizeof (uint64_t)); 1509 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1510 error = dmu_tx_assign(tx, TXG_WAIT); 1511 if (error) { 1512 ztest_record_enospc("create a batch object"); 1513 dmu_tx_abort(tx); 1514 return; 1515 } 1516 batchobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1517 DMU_OT_NONE, 0, tx); 1518 ztest_set_random_blocksize(os, batchobj, tx); 1519 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, 1520 sizeof (uint64_t), &batchobj, tx); 1521 dmu_tx_commit(tx); 1522 } 1523 1524 /* 1525 * Destroy the previous batch of objects. 1526 */ 1527 for (b = 0; b < batchsize; b++) { 1528 VERIFY(0 == dmu_read(os, batchobj, b * sizeof (uint64_t), 1529 sizeof (uint64_t), &object)); 1530 if (object == 0) 1531 continue; 1532 /* 1533 * Read and validate contents. 1534 * We expect the nth byte of the bonus buffer to be n. 1535 */ 1536 VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db)); 1537 1538 dmu_object_info_from_db(db, &doi); 1539 ASSERT(doi.doi_type == DMU_OT_UINT64_OTHER); 1540 ASSERT(doi.doi_bonus_type == DMU_OT_PLAIN_OTHER); 1541 ASSERT3S(doi.doi_physical_blks, >=, 0); 1542 1543 bonuslen = db->db_size; 1544 1545 for (c = 0; c < bonuslen; c++) { 1546 if (((uint8_t *)db->db_data)[c] != 1547 (uint8_t)(c + bonuslen)) { 1548 fatal(0, 1549 "bad bonus: %s, obj %llu, off %d: %u != %u", 1550 osname, object, c, 1551 ((uint8_t *)db->db_data)[c], 1552 (uint8_t)(c + bonuslen)); 1553 } 1554 } 1555 1556 dmu_buf_rele(db, FTAG); 1557 1558 /* 1559 * We expect the word at endoff to be our object number. 1560 */ 1561 VERIFY(0 == dmu_read(os, object, endoff, 1562 sizeof (uint64_t), &temp)); 1563 1564 if (temp != object) { 1565 fatal(0, "bad data in %s, got %llu, expected %llu", 1566 osname, temp, object); 1567 } 1568 1569 /* 1570 * Destroy old object and clear batch entry. 1571 */ 1572 tx = dmu_tx_create(os); 1573 dmu_tx_hold_write(tx, batchobj, 1574 b * sizeof (uint64_t), sizeof (uint64_t)); 1575 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 1576 error = dmu_tx_assign(tx, TXG_WAIT); 1577 if (error) { 1578 ztest_record_enospc("free object"); 1579 dmu_tx_abort(tx); 1580 return; 1581 } 1582 error = dmu_object_free(os, object, tx); 1583 if (error) { 1584 fatal(0, "dmu_object_free('%s', %llu) = %d", 1585 osname, object, error); 1586 } 1587 object = 0; 1588 1589 dmu_object_set_checksum(os, batchobj, 1590 ztest_random_checksum(), tx); 1591 dmu_object_set_compress(os, batchobj, 1592 ztest_random_compress(), tx); 1593 1594 dmu_write(os, batchobj, b * sizeof (uint64_t), 1595 sizeof (uint64_t), &object, tx); 1596 1597 dmu_tx_commit(tx); 1598 } 1599 1600 /* 1601 * Before creating the new batch of objects, generate a bunch of churn. 1602 */ 1603 for (b = ztest_random(100); b > 0; b--) { 1604 tx = dmu_tx_create(os); 1605 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1606 error = dmu_tx_assign(tx, TXG_WAIT); 1607 if (error) { 1608 ztest_record_enospc("churn objects"); 1609 dmu_tx_abort(tx); 1610 return; 1611 } 1612 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1613 DMU_OT_NONE, 0, tx); 1614 ztest_set_random_blocksize(os, object, tx); 1615 error = dmu_object_free(os, object, tx); 1616 if (error) { 1617 fatal(0, "dmu_object_free('%s', %llu) = %d", 1618 osname, object, error); 1619 } 1620 dmu_tx_commit(tx); 1621 } 1622 1623 /* 1624 * Create a new batch of objects with randomly chosen 1625 * blocksizes and record them in the batch directory. 1626 */ 1627 for (b = 0; b < batchsize; b++) { 1628 uint32_t va_blksize; 1629 u_longlong_t va_nblocks; 1630 1631 tx = dmu_tx_create(os); 1632 dmu_tx_hold_write(tx, batchobj, b * sizeof (uint64_t), 1633 sizeof (uint64_t)); 1634 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1635 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, endoff, 1636 sizeof (uint64_t)); 1637 error = dmu_tx_assign(tx, TXG_WAIT); 1638 if (error) { 1639 ztest_record_enospc("create batchobj"); 1640 dmu_tx_abort(tx); 1641 return; 1642 } 1643 bonuslen = (int)ztest_random(dmu_bonus_max()) + 1; 1644 1645 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1646 DMU_OT_PLAIN_OTHER, bonuslen, tx); 1647 1648 ztest_set_random_blocksize(os, object, tx); 1649 1650 dmu_object_set_checksum(os, object, 1651 ztest_random_checksum(), tx); 1652 dmu_object_set_compress(os, object, 1653 ztest_random_compress(), tx); 1654 1655 dmu_write(os, batchobj, b * sizeof (uint64_t), 1656 sizeof (uint64_t), &object, tx); 1657 1658 /* 1659 * Write to both the bonus buffer and the regular data. 1660 */ 1661 VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db)); 1662 ASSERT3U(bonuslen, ==, db->db_size); 1663 1664 dmu_object_size_from_db(db, &va_blksize, &va_nblocks); 1665 ASSERT3S(va_nblocks, >=, 0); 1666 1667 dmu_buf_will_dirty(db, tx); 1668 1669 /* 1670 * See comments above regarding the contents of 1671 * the bonus buffer and the word at endoff. 1672 */ 1673 for (c = 0; c < db->db_size; c++) 1674 ((uint8_t *)db->db_data)[c] = (uint8_t)(c + bonuslen); 1675 1676 dmu_buf_rele(db, FTAG); 1677 1678 /* 1679 * Write to a large offset to increase indirection. 1680 */ 1681 dmu_write(os, object, endoff, sizeof (uint64_t), &object, tx); 1682 1683 dmu_tx_commit(tx); 1684 } 1685} 1686 1687/* 1688 * Verify that dmu_{read,write} work as expected. 1689 */ 1690typedef struct bufwad { 1691 uint64_t bw_index; 1692 uint64_t bw_txg; 1693 uint64_t bw_data; 1694} bufwad_t; 1695 1696typedef struct dmu_read_write_dir { 1697 uint64_t dd_packobj; 1698 uint64_t dd_bigobj; 1699 uint64_t dd_chunk; 1700} dmu_read_write_dir_t; 1701 1702void 1703ztest_dmu_read_write(ztest_args_t *za) 1704{ 1705 objset_t *os = za->za_os; 1706 dmu_read_write_dir_t dd; 1707 dmu_tx_t *tx; 1708 int i, freeit, error; 1709 uint64_t n, s, txg; 1710 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; 1711 uint64_t packoff, packsize, bigoff, bigsize; 1712 uint64_t regions = 997; 1713 uint64_t stride = 123456789ULL; 1714 uint64_t width = 40; 1715 int free_percent = 5; 1716 1717 /* 1718 * This test uses two objects, packobj and bigobj, that are always 1719 * updated together (i.e. in the same tx) so that their contents are 1720 * in sync and can be compared. Their contents relate to each other 1721 * in a simple way: packobj is a dense array of 'bufwad' structures, 1722 * while bigobj is a sparse array of the same bufwads. Specifically, 1723 * for any index n, there are three bufwads that should be identical: 1724 * 1725 * packobj, at offset n * sizeof (bufwad_t) 1726 * bigobj, at the head of the nth chunk 1727 * bigobj, at the tail of the nth chunk 1728 * 1729 * The chunk size is arbitrary. It doesn't have to be a power of two, 1730 * and it doesn't have any relation to the object blocksize. 1731 * The only requirement is that it can hold at least two bufwads. 1732 * 1733 * Normally, we write the bufwad to each of these locations. 1734 * However, free_percent of the time we instead write zeroes to 1735 * packobj and perform a dmu_free_range() on bigobj. By comparing 1736 * bigobj to packobj, we can verify that the DMU is correctly 1737 * tracking which parts of an object are allocated and free, 1738 * and that the contents of the allocated blocks are correct. 1739 */ 1740 1741 /* 1742 * Read the directory info. If it's the first time, set things up. 1743 */ 1744 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 1745 sizeof (dd), &dd)); 1746 if (dd.dd_chunk == 0) { 1747 ASSERT(dd.dd_packobj == 0); 1748 ASSERT(dd.dd_bigobj == 0); 1749 tx = dmu_tx_create(os); 1750 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (dd)); 1751 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1752 error = dmu_tx_assign(tx, TXG_WAIT); 1753 if (error) { 1754 ztest_record_enospc("create r/w directory"); 1755 dmu_tx_abort(tx); 1756 return; 1757 } 1758 1759 dd.dd_packobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1760 DMU_OT_NONE, 0, tx); 1761 dd.dd_bigobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1762 DMU_OT_NONE, 0, tx); 1763 dd.dd_chunk = (1000 + ztest_random(1000)) * sizeof (uint64_t); 1764 1765 ztest_set_random_blocksize(os, dd.dd_packobj, tx); 1766 ztest_set_random_blocksize(os, dd.dd_bigobj, tx); 1767 1768 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (dd), &dd, 1769 tx); 1770 dmu_tx_commit(tx); 1771 } 1772 1773 /* 1774 * Prefetch a random chunk of the big object. 1775 * Our aim here is to get some async reads in flight 1776 * for blocks that we may free below; the DMU should 1777 * handle this race correctly. 1778 */ 1779 n = ztest_random(regions) * stride + ztest_random(width); 1780 s = 1 + ztest_random(2 * width - 1); 1781 dmu_prefetch(os, dd.dd_bigobj, n * dd.dd_chunk, s * dd.dd_chunk); 1782 1783 /* 1784 * Pick a random index and compute the offsets into packobj and bigobj. 1785 */ 1786 n = ztest_random(regions) * stride + ztest_random(width); 1787 s = 1 + ztest_random(width - 1); 1788 1789 packoff = n * sizeof (bufwad_t); 1790 packsize = s * sizeof (bufwad_t); 1791 1792 bigoff = n * dd.dd_chunk; 1793 bigsize = s * dd.dd_chunk; 1794 1795 packbuf = umem_alloc(packsize, UMEM_NOFAIL); 1796 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); 1797 1798 /* 1799 * free_percent of the time, free a range of bigobj rather than 1800 * overwriting it. 1801 */ 1802 freeit = (ztest_random(100) < free_percent); 1803 1804 /* 1805 * Read the current contents of our objects. 1806 */ 1807 error = dmu_read(os, dd.dd_packobj, packoff, packsize, packbuf); 1808 ASSERT3U(error, ==, 0); 1809 error = dmu_read(os, dd.dd_bigobj, bigoff, bigsize, bigbuf); 1810 ASSERT3U(error, ==, 0); 1811 1812 /* 1813 * Get a tx for the mods to both packobj and bigobj. 1814 */ 1815 tx = dmu_tx_create(os); 1816 1817 dmu_tx_hold_write(tx, dd.dd_packobj, packoff, packsize); 1818 1819 if (freeit) 1820 dmu_tx_hold_free(tx, dd.dd_bigobj, bigoff, bigsize); 1821 else 1822 dmu_tx_hold_write(tx, dd.dd_bigobj, bigoff, bigsize); 1823 1824 error = dmu_tx_assign(tx, TXG_WAIT); 1825 1826 if (error) { 1827 ztest_record_enospc("dmu r/w range"); 1828 dmu_tx_abort(tx); 1829 umem_free(packbuf, packsize); 1830 umem_free(bigbuf, bigsize); 1831 return; 1832 } 1833 1834 txg = dmu_tx_get_txg(tx); 1835 1836 /* 1837 * For each index from n to n + s, verify that the existing bufwad 1838 * in packobj matches the bufwads at the head and tail of the 1839 * corresponding chunk in bigobj. Then update all three bufwads 1840 * with the new values we want to write out. 1841 */ 1842 for (i = 0; i < s; i++) { 1843 /* LINTED */ 1844 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 1845 /* LINTED */ 1846 bigH = (bufwad_t *)((char *)bigbuf + i * dd.dd_chunk); 1847 /* LINTED */ 1848 bigT = (bufwad_t *)((char *)bigH + dd.dd_chunk) - 1; 1849 1850 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 1851 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 1852 1853 if (pack->bw_txg > txg) 1854 fatal(0, "future leak: got %llx, open txg is %llx", 1855 pack->bw_txg, txg); 1856 1857 if (pack->bw_data != 0 && pack->bw_index != n + i) 1858 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 1859 pack->bw_index, n, i); 1860 1861 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 1862 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 1863 1864 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 1865 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 1866 1867 if (freeit) { 1868 bzero(pack, sizeof (bufwad_t)); 1869 } else { 1870 pack->bw_index = n + i; 1871 pack->bw_txg = txg; 1872 pack->bw_data = 1 + ztest_random(-2ULL); 1873 } 1874 *bigH = *pack; 1875 *bigT = *pack; 1876 } 1877 1878 /* 1879 * We've verified all the old bufwads, and made new ones. 1880 * Now write them out. 1881 */ 1882 dmu_write(os, dd.dd_packobj, packoff, packsize, packbuf, tx); 1883 1884 if (freeit) { 1885 if (zopt_verbose >= 6) { 1886 (void) printf("freeing offset %llx size %llx" 1887 " txg %llx\n", 1888 (u_longlong_t)bigoff, 1889 (u_longlong_t)bigsize, 1890 (u_longlong_t)txg); 1891 } 1892 VERIFY(0 == dmu_free_range(os, dd.dd_bigobj, bigoff, 1893 bigsize, tx)); 1894 } else { 1895 if (zopt_verbose >= 6) { 1896 (void) printf("writing offset %llx size %llx" 1897 " txg %llx\n", 1898 (u_longlong_t)bigoff, 1899 (u_longlong_t)bigsize, 1900 (u_longlong_t)txg); 1901 } 1902 dmu_write(os, dd.dd_bigobj, bigoff, bigsize, bigbuf, tx); 1903 } 1904 1905 dmu_tx_commit(tx); 1906 1907 /* 1908 * Sanity check the stuff we just wrote. 1909 */ 1910 { 1911 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 1912 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 1913 1914 VERIFY(0 == dmu_read(os, dd.dd_packobj, packoff, 1915 packsize, packcheck)); 1916 VERIFY(0 == dmu_read(os, dd.dd_bigobj, bigoff, 1917 bigsize, bigcheck)); 1918 1919 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 1920 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 1921 1922 umem_free(packcheck, packsize); 1923 umem_free(bigcheck, bigsize); 1924 } 1925 1926 umem_free(packbuf, packsize); 1927 umem_free(bigbuf, bigsize); 1928} 1929 1930void 1931ztest_dmu_check_future_leak(objset_t *os, uint64_t txg) 1932{ 1933 dmu_buf_t *db; 1934 ztest_block_tag_t rbt; 1935 1936 if (zopt_verbose >= 3) { 1937 char osname[MAXNAMELEN]; 1938 dmu_objset_name(os, osname); 1939 (void) printf("checking %s for future leaks in txg %lld...\n", 1940 osname, (u_longlong_t)txg); 1941 } 1942 1943 /* 1944 * Make sure that, if there is a write record in the bonus buffer 1945 * of the ZTEST_DIROBJ, that the txg for this record is <= the 1946 * last synced txg of the pool. 1947 */ 1948 1949 VERIFY(0 == dmu_bonus_hold(os, ZTEST_DIROBJ, FTAG, &db)); 1950 ASSERT3U(db->db_size, ==, sizeof (rbt)); 1951 bcopy(db->db_data, &rbt, db->db_size); 1952 if (rbt.bt_objset != 0) { 1953 ASSERT3U(rbt.bt_objset, ==, dmu_objset_id(os)); 1954 ASSERT3U(rbt.bt_object, ==, ZTEST_DIROBJ); 1955 ASSERT3U(rbt.bt_offset, ==, -1ULL); 1956 if (rbt.bt_txg > txg) { 1957 fatal(0, 1958 "future leak: got %llx, last synced txg is %llx", 1959 rbt.bt_txg, txg); 1960 } 1961 } 1962 dmu_buf_rele(db, FTAG); 1963} 1964 1965void 1966ztest_dmu_write_parallel(ztest_args_t *za) 1967{ 1968 objset_t *os = za->za_os; 1969 dmu_tx_t *tx; 1970 dmu_buf_t *db; 1971 int i, b, error, do_free, bs; 1972 uint64_t off, txg_how, txg; 1973 mutex_t *lp; 1974 char osname[MAXNAMELEN]; 1975 char iobuf[SPA_MAXBLOCKSIZE]; 1976 ztest_block_tag_t rbt, wbt; 1977 1978 dmu_objset_name(os, osname); 1979 bs = ZTEST_DIROBJ_BLOCKSIZE; 1980 1981 /* 1982 * Have multiple threads write to large offsets in ZTEST_DIROBJ 1983 * to verify that having multiple threads writing to the same object 1984 * in parallel doesn't cause any trouble. 1985 * Also do parallel writes to the bonus buffer on occasion. 1986 */ 1987 for (i = 0; i < 50; i++) { 1988 b = ztest_random(ZTEST_SYNC_LOCKS); 1989 lp = &ztest_shared->zs_sync_lock[b]; 1990 1991 do_free = (ztest_random(4) == 0); 1992 1993 off = za->za_diroff_shared + ((uint64_t)b << SPA_MAXBLOCKSHIFT); 1994 1995 if (ztest_random(4) == 0) { 1996 /* 1997 * Do the bonus buffer instead of a regular block. 1998 */ 1999 do_free = 0; 2000 off = -1ULL; 2001 } 2002 2003 tx = dmu_tx_create(os); 2004 2005 if (off == -1ULL) 2006 dmu_tx_hold_bonus(tx, ZTEST_DIROBJ); 2007 else if (do_free) 2008 dmu_tx_hold_free(tx, ZTEST_DIROBJ, off, bs); 2009 else 2010 dmu_tx_hold_write(tx, ZTEST_DIROBJ, off, bs); 2011 2012 txg_how = ztest_random(2) == 0 ? TXG_WAIT : TXG_NOWAIT; 2013 error = dmu_tx_assign(tx, txg_how); 2014 if (error) { 2015 if (error == ERESTART) { 2016 ASSERT(txg_how == TXG_NOWAIT); 2017 dmu_tx_wait(tx); 2018 dmu_tx_abort(tx); 2019 continue; 2020 } 2021 dmu_tx_abort(tx); 2022 ztest_record_enospc("dmu write parallel"); 2023 return; 2024 } 2025 txg = dmu_tx_get_txg(tx); 2026 2027 if (do_free) { 2028 (void) mutex_lock(lp); 2029 VERIFY(0 == dmu_free_range(os, ZTEST_DIROBJ, off, 2030 bs, tx)); 2031 (void) mutex_unlock(lp); 2032 dmu_tx_commit(tx); 2033 continue; 2034 } 2035 2036 wbt.bt_objset = dmu_objset_id(os); 2037 wbt.bt_object = ZTEST_DIROBJ; 2038 wbt.bt_offset = off; 2039 wbt.bt_txg = txg; 2040 wbt.bt_thread = za->za_instance; 2041 2042 if (off == -1ULL) { 2043 wbt.bt_seq = 0; 2044 VERIFY(0 == dmu_bonus_hold(os, ZTEST_DIROBJ, 2045 FTAG, &db)); 2046 ASSERT3U(db->db_size, ==, sizeof (wbt)); 2047 bcopy(db->db_data, &rbt, db->db_size); 2048 if (rbt.bt_objset != 0) { 2049 ASSERT3U(rbt.bt_objset, ==, wbt.bt_objset); 2050 ASSERT3U(rbt.bt_object, ==, wbt.bt_object); 2051 ASSERT3U(rbt.bt_offset, ==, wbt.bt_offset); 2052 ASSERT3U(rbt.bt_txg, <=, wbt.bt_txg); 2053 } 2054 dmu_buf_will_dirty(db, tx); 2055 bcopy(&wbt, db->db_data, db->db_size); 2056 dmu_buf_rele(db, FTAG); 2057 dmu_tx_commit(tx); 2058 continue; 2059 } 2060 2061 (void) mutex_lock(lp); 2062 2063 wbt.bt_seq = ztest_shared->zs_seq[b]++; 2064 2065 dmu_write(os, ZTEST_DIROBJ, off, sizeof (wbt), &wbt, tx); 2066 2067 (void) mutex_unlock(lp); 2068 2069 if (ztest_random(100) == 0) 2070 (void) poll(NULL, 0, 1); /* open dn_notxholds window */ 2071 2072 dmu_tx_commit(tx); 2073 2074 if (ztest_random(1000) == 0) 2075 txg_wait_synced(dmu_objset_pool(os), txg); 2076 2077 if (ztest_random(2) == 0) { 2078 blkptr_t blk = { 0 }; 2079 uint64_t blkoff; 2080 zbookmark_t zb; 2081 2082 (void) mutex_lock(lp); 2083 blkoff = P2ALIGN_TYPED(off, bs, uint64_t); 2084 error = dmu_buf_hold(os, 2085 ZTEST_DIROBJ, blkoff, FTAG, &db); 2086 if (error) { 2087 dprintf("dmu_buf_hold(%s, %d, %llx) = %d\n", 2088 osname, ZTEST_DIROBJ, blkoff, error); 2089 (void) mutex_unlock(lp); 2090 continue; 2091 } 2092 blkoff = off - blkoff; 2093 error = dmu_sync(NULL, db, &blk, txg, NULL, NULL); 2094 dmu_buf_rele(db, FTAG); 2095 (void) mutex_unlock(lp); 2096 if (error) { 2097 dprintf("dmu_sync(%s, %d, %llx) = %d\n", 2098 osname, ZTEST_DIROBJ, off, error); 2099 continue; 2100 } 2101 2102 if (blk.blk_birth == 0) { /* concurrent free */ 2103 continue; 2104 } 2105 txg_suspend(dmu_objset_pool(os)); 2106 2107 ASSERT(blk.blk_fill == 1); 2108 ASSERT3U(BP_GET_TYPE(&blk), ==, DMU_OT_UINT64_OTHER); 2109 ASSERT3U(BP_GET_LEVEL(&blk), ==, 0); 2110 ASSERT3U(BP_GET_LSIZE(&blk), ==, bs); 2111 2112 /* 2113 * Read the block that dmu_sync() returned to 2114 * make sure its contents match what we wrote. 2115 * We do this while still txg_suspend()ed to ensure 2116 * that the block can't be reused before we read it. 2117 */ 2118 zb.zb_objset = dmu_objset_id(os); 2119 zb.zb_object = ZTEST_DIROBJ; 2120 zb.zb_level = 0; 2121 zb.zb_blkid = off / bs; 2122 error = zio_wait(zio_read(NULL, dmu_objset_spa(os), 2123 &blk, iobuf, bs, NULL, NULL, 2124 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_MUSTSUCCEED, &zb)); 2125 ASSERT(error == 0); 2126 2127 txg_resume(dmu_objset_pool(os)); 2128 2129 bcopy(&iobuf[blkoff], &rbt, sizeof (rbt)); 2130 2131 if (rbt.bt_objset == 0) /* concurrent free */ 2132 continue; 2133 2134 ASSERT3U(rbt.bt_objset, ==, wbt.bt_objset); 2135 ASSERT3U(rbt.bt_object, ==, wbt.bt_object); 2136 ASSERT3U(rbt.bt_offset, ==, wbt.bt_offset); 2137 2138 /* 2139 * The semantic of dmu_sync() is that we always 2140 * push the most recent version of the data, 2141 * so in the face of concurrent updates we may 2142 * see a newer version of the block. That's OK. 2143 */ 2144 ASSERT3U(rbt.bt_txg, >=, wbt.bt_txg); 2145 if (rbt.bt_thread == wbt.bt_thread) 2146 ASSERT3U(rbt.bt_seq, ==, wbt.bt_seq); 2147 else 2148 ASSERT3U(rbt.bt_seq, >, wbt.bt_seq); 2149 } 2150 } 2151} 2152 2153/* 2154 * Verify that zap_{create,destroy,add,remove,update} work as expected. 2155 */ 2156#define ZTEST_ZAP_MIN_INTS 1 2157#define ZTEST_ZAP_MAX_INTS 4 2158#define ZTEST_ZAP_MAX_PROPS 1000 2159 2160void 2161ztest_zap(ztest_args_t *za) 2162{ 2163 objset_t *os = za->za_os; 2164 uint64_t object; 2165 uint64_t txg, last_txg; 2166 uint64_t value[ZTEST_ZAP_MAX_INTS]; 2167 uint64_t zl_ints, zl_intsize, prop; 2168 int i, ints; 2169 int iters = 100; 2170 dmu_tx_t *tx; 2171 char propname[100], txgname[100]; 2172 int error; 2173 char osname[MAXNAMELEN]; 2174 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; 2175 2176 dmu_objset_name(os, osname); 2177 2178 /* 2179 * Create a new object if necessary, and record it in the directory. 2180 */ 2181 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 2182 sizeof (uint64_t), &object)); 2183 2184 if (object == 0) { 2185 tx = dmu_tx_create(os); 2186 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, 2187 sizeof (uint64_t)); 2188 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL); 2189 error = dmu_tx_assign(tx, TXG_WAIT); 2190 if (error) { 2191 ztest_record_enospc("create zap test obj"); 2192 dmu_tx_abort(tx); 2193 return; 2194 } 2195 object = zap_create(os, DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx); 2196 if (error) { 2197 fatal(0, "zap_create('%s', %llu) = %d", 2198 osname, object, error); 2199 } 2200 ASSERT(object != 0); 2201 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, 2202 sizeof (uint64_t), &object, tx); 2203 /* 2204 * Generate a known hash collision, and verify that 2205 * we can lookup and remove both entries. 2206 */ 2207 for (i = 0; i < 2; i++) { 2208 value[i] = i; 2209 error = zap_add(os, object, hc[i], sizeof (uint64_t), 2210 1, &value[i], tx); 2211 ASSERT3U(error, ==, 0); 2212 } 2213 for (i = 0; i < 2; i++) { 2214 error = zap_add(os, object, hc[i], sizeof (uint64_t), 2215 1, &value[i], tx); 2216 ASSERT3U(error, ==, EEXIST); 2217 error = zap_length(os, object, hc[i], 2218 &zl_intsize, &zl_ints); 2219 ASSERT3U(error, ==, 0); 2220 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2221 ASSERT3U(zl_ints, ==, 1); 2222 } 2223 for (i = 0; i < 2; i++) { 2224 error = zap_remove(os, object, hc[i], tx); 2225 ASSERT3U(error, ==, 0); 2226 } 2227 2228 dmu_tx_commit(tx); 2229 } 2230 2231 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); 2232 2233 while (--iters >= 0) { 2234 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 2235 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 2236 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 2237 bzero(value, sizeof (value)); 2238 last_txg = 0; 2239 2240 /* 2241 * If these zap entries already exist, validate their contents. 2242 */ 2243 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 2244 if (error == 0) { 2245 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2246 ASSERT3U(zl_ints, ==, 1); 2247 2248 error = zap_lookup(os, object, txgname, zl_intsize, 2249 zl_ints, &last_txg); 2250 2251 ASSERT3U(error, ==, 0); 2252 2253 error = zap_length(os, object, propname, &zl_intsize, 2254 &zl_ints); 2255 2256 ASSERT3U(error, ==, 0); 2257 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2258 ASSERT3U(zl_ints, ==, ints); 2259 2260 error = zap_lookup(os, object, propname, zl_intsize, 2261 zl_ints, value); 2262 2263 ASSERT3U(error, ==, 0); 2264 2265 for (i = 0; i < ints; i++) { 2266 ASSERT3U(value[i], ==, last_txg + object + i); 2267 } 2268 } else { 2269 ASSERT3U(error, ==, ENOENT); 2270 } 2271 2272 /* 2273 * Atomically update two entries in our zap object. 2274 * The first is named txg_%llu, and contains the txg 2275 * in which the property was last updated. The second 2276 * is named prop_%llu, and the nth element of its value 2277 * should be txg + object + n. 2278 */ 2279 tx = dmu_tx_create(os); 2280 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2281 error = dmu_tx_assign(tx, TXG_WAIT); 2282 if (error) { 2283 ztest_record_enospc("create zap entry"); 2284 dmu_tx_abort(tx); 2285 return; 2286 } 2287 txg = dmu_tx_get_txg(tx); 2288 2289 if (last_txg > txg) 2290 fatal(0, "zap future leak: old %llu new %llu", 2291 last_txg, txg); 2292 2293 for (i = 0; i < ints; i++) 2294 value[i] = txg + object + i; 2295 2296 error = zap_update(os, object, txgname, sizeof (uint64_t), 2297 1, &txg, tx); 2298 if (error) 2299 fatal(0, "zap_update('%s', %llu, '%s') = %d", 2300 osname, object, txgname, error); 2301 2302 error = zap_update(os, object, propname, sizeof (uint64_t), 2303 ints, value, tx); 2304 if (error) 2305 fatal(0, "zap_update('%s', %llu, '%s') = %d", 2306 osname, object, propname, error); 2307 2308 dmu_tx_commit(tx); 2309 2310 /* 2311 * Remove a random pair of entries. 2312 */ 2313 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 2314 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 2315 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 2316 2317 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 2318 2319 if (error == ENOENT) 2320 continue; 2321 2322 ASSERT3U(error, ==, 0); 2323 2324 tx = dmu_tx_create(os); 2325 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2326 error = dmu_tx_assign(tx, TXG_WAIT); 2327 if (error) { 2328 ztest_record_enospc("remove zap entry"); 2329 dmu_tx_abort(tx); 2330 return; 2331 } 2332 error = zap_remove(os, object, txgname, tx); 2333 if (error) 2334 fatal(0, "zap_remove('%s', %llu, '%s') = %d", 2335 osname, object, txgname, error); 2336 2337 error = zap_remove(os, object, propname, tx); 2338 if (error) 2339 fatal(0, "zap_remove('%s', %llu, '%s') = %d", 2340 osname, object, propname, error); 2341 2342 dmu_tx_commit(tx); 2343 } 2344 2345 /* 2346 * Once in a while, destroy the object. 2347 */ 2348 if (ztest_random(100) != 0) 2349 return; 2350 2351 tx = dmu_tx_create(os); 2352 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t)); 2353 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 2354 error = dmu_tx_assign(tx, TXG_WAIT); 2355 if (error) { 2356 ztest_record_enospc("destroy zap object"); 2357 dmu_tx_abort(tx); 2358 return; 2359 } 2360 error = zap_destroy(os, object, tx); 2361 if (error) 2362 fatal(0, "zap_destroy('%s', %llu) = %d", 2363 osname, object, error); 2364 object = 0; 2365 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t), 2366 &object, tx); 2367 dmu_tx_commit(tx); 2368} 2369 2370void 2371ztest_zap_parallel(ztest_args_t *za) 2372{ 2373 objset_t *os = za->za_os; 2374 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; 2375 int iters = 100; 2376 dmu_tx_t *tx; 2377 int i, namelen, error; 2378 char name[20], string_value[20]; 2379 void *data; 2380 2381 while (--iters >= 0) { 2382 /* 2383 * Generate a random name of the form 'xxx.....' where each 2384 * x is a random printable character and the dots are dots. 2385 * There are 94 such characters, and the name length goes from 2386 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. 2387 */ 2388 namelen = ztest_random(sizeof (name) - 5) + 5 + 1; 2389 2390 for (i = 0; i < 3; i++) 2391 name[i] = '!' + ztest_random('~' - '!' + 1); 2392 for (; i < namelen - 1; i++) 2393 name[i] = '.'; 2394 name[i] = '\0'; 2395 2396 if (ztest_random(2) == 0) 2397 object = ZTEST_MICROZAP_OBJ; 2398 else 2399 object = ZTEST_FATZAP_OBJ; 2400 2401 if ((namelen & 1) || object == ZTEST_MICROZAP_OBJ) { 2402 wsize = sizeof (txg); 2403 wc = 1; 2404 data = &txg; 2405 } else { 2406 wsize = 1; 2407 wc = namelen; 2408 data = string_value; 2409 } 2410 2411 count = -1ULL; 2412 VERIFY(zap_count(os, object, &count) == 0); 2413 ASSERT(count != -1ULL); 2414 2415 /* 2416 * Select an operation: length, lookup, add, update, remove. 2417 */ 2418 i = ztest_random(5); 2419 2420 if (i >= 2) { 2421 tx = dmu_tx_create(os); 2422 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2423 error = dmu_tx_assign(tx, TXG_WAIT); 2424 if (error) { 2425 ztest_record_enospc("zap parallel"); 2426 dmu_tx_abort(tx); 2427 return; 2428 } 2429 txg = dmu_tx_get_txg(tx); 2430 bcopy(name, string_value, namelen); 2431 } else { 2432 tx = NULL; 2433 txg = 0; 2434 bzero(string_value, namelen); 2435 } 2436 2437 switch (i) { 2438 2439 case 0: 2440 error = zap_length(os, object, name, &zl_wsize, &zl_wc); 2441 if (error == 0) { 2442 ASSERT3U(wsize, ==, zl_wsize); 2443 ASSERT3U(wc, ==, zl_wc); 2444 } else { 2445 ASSERT3U(error, ==, ENOENT); 2446 } 2447 break; 2448 2449 case 1: 2450 error = zap_lookup(os, object, name, wsize, wc, data); 2451 if (error == 0) { 2452 if (data == string_value && 2453 bcmp(name, data, namelen) != 0) 2454 fatal(0, "name '%s' != val '%s' len %d", 2455 name, data, namelen); 2456 } else { 2457 ASSERT3U(error, ==, ENOENT); 2458 } 2459 break; 2460 2461 case 2: 2462 error = zap_add(os, object, name, wsize, wc, data, tx); 2463 ASSERT(error == 0 || error == EEXIST); 2464 break; 2465 2466 case 3: 2467 VERIFY(zap_update(os, object, name, wsize, wc, 2468 data, tx) == 0); 2469 break; 2470 2471 case 4: 2472 error = zap_remove(os, object, name, tx); 2473 ASSERT(error == 0 || error == ENOENT); 2474 break; 2475 } 2476 2477 if (tx != NULL) 2478 dmu_tx_commit(tx); 2479 } 2480} 2481 2482void 2483ztest_dsl_prop_get_set(ztest_args_t *za) 2484{ 2485 objset_t *os = za->za_os; 2486 int i, inherit; 2487 uint64_t value; 2488 const char *prop, *valname; 2489 char setpoint[MAXPATHLEN]; 2490 char osname[MAXNAMELEN]; 2491 int error; 2492 2493 (void) rw_rdlock(&ztest_shared->zs_name_lock); 2494 2495 dmu_objset_name(os, osname); 2496 2497 for (i = 0; i < 2; i++) { 2498 if (i == 0) { 2499 prop = "checksum"; 2500 value = ztest_random_checksum(); 2501 inherit = (value == ZIO_CHECKSUM_INHERIT); 2502 } else { 2503 prop = "compression"; 2504 value = ztest_random_compress(); 2505 inherit = (value == ZIO_COMPRESS_INHERIT); 2506 } 2507 2508 error = dsl_prop_set(osname, prop, sizeof (value), 2509 !inherit, &value); 2510 2511 if (error == ENOSPC) { 2512 ztest_record_enospc("dsl_prop_set"); 2513 break; 2514 } 2515 2516 ASSERT3U(error, ==, 0); 2517 2518 VERIFY3U(dsl_prop_get(osname, prop, sizeof (value), 2519 1, &value, setpoint), ==, 0); 2520 2521 if (i == 0) 2522 valname = zio_checksum_table[value].ci_name; 2523 else 2524 valname = zio_compress_table[value].ci_name; 2525 2526 if (zopt_verbose >= 6) { 2527 (void) printf("%s %s = %s for '%s'\n", 2528 osname, prop, valname, setpoint); 2529 } 2530 } 2531 2532 (void) rw_unlock(&ztest_shared->zs_name_lock); 2533} 2534 2535static void 2536ztest_error_setup(vdev_t *vd, int mode, int mask, uint64_t arg) 2537{ 2538 int c; 2539 2540 for (c = 0; c < vd->vdev_children; c++) 2541 ztest_error_setup(vd->vdev_child[c], mode, mask, arg); 2542 2543 if (vd->vdev_path != NULL) { 2544 vd->vdev_fault_mode = mode; 2545 vd->vdev_fault_mask = mask; 2546 vd->vdev_fault_arg = arg; 2547 } 2548} 2549 2550/* 2551 * Inject random faults into the on-disk data. 2552 */ 2553void 2554ztest_fault_inject(ztest_args_t *za) 2555{ 2556 int fd; 2557 uint64_t offset; 2558 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 2559 uint64_t bad = 0x1990c0ffeedecadeULL; 2560 uint64_t top, leaf; 2561 char path0[MAXPATHLEN]; 2562 char pathrand[MAXPATHLEN]; 2563 size_t fsize; 2564 spa_t *spa = dmu_objset_spa(za->za_os); 2565 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ 2566 int iters = 1000; 2567 vdev_t *vd0; 2568 uint64_t guid0 = 0; 2569 2570 /* 2571 * We can't inject faults when we have no fault tolerance. 2572 */ 2573 if (zopt_maxfaults == 0) 2574 return; 2575 2576 ASSERT(leaves >= 2); 2577 2578 /* 2579 * Pick a random top-level vdev. 2580 */ 2581 spa_config_enter(spa, RW_READER, FTAG); 2582 top = ztest_random(spa->spa_root_vdev->vdev_children); 2583 spa_config_exit(spa, FTAG); 2584 2585 /* 2586 * Pick a random leaf. 2587 */ 2588 leaf = ztest_random(leaves); 2589 2590 /* 2591 * Generate paths to the first two leaves in this top-level vdev, 2592 * and to the random leaf we selected. We'll induce transient 2593 * I/O errors and random online/offline activity on leaf 0, 2594 * and we'll write random garbage to the randomly chosen leaf. 2595 */ 2596 (void) snprintf(path0, sizeof (path0), 2597 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + 0); 2598 (void) snprintf(pathrand, sizeof (pathrand), 2599 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf); 2600 2601 dprintf("damaging %s and %s\n", path0, pathrand); 2602 2603 spa_config_enter(spa, RW_READER, FTAG); 2604 2605 /* 2606 * If we can tolerate two or more faults, make vd0 fail randomly. 2607 */ 2608 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); 2609 if (vd0 != NULL && zopt_maxfaults >= 2) { 2610 guid0 = vd0->vdev_guid; 2611 ztest_error_setup(vd0, VDEV_FAULT_COUNT, 2612 (1U << ZIO_TYPE_READ) | (1U << ZIO_TYPE_WRITE), 100); 2613 } 2614 2615 spa_config_exit(spa, FTAG); 2616 2617 /* 2618 * If we can tolerate two or more faults, randomly online/offline vd0. 2619 */ 2620 if (zopt_maxfaults >= 2 && guid0 != 0) { 2621 if (ztest_random(10) < 6) 2622 (void) vdev_offline(spa, guid0, B_TRUE); 2623 else 2624 (void) vdev_online(spa, guid0); 2625 } 2626 2627 /* 2628 * We have at least single-fault tolerance, so inject data corruption. 2629 */ 2630 fd = open(pathrand, O_RDWR); 2631 2632 if (fd == -1) /* we hit a gap in the device namespace */ 2633 return; 2634 2635 fsize = lseek(fd, 0, SEEK_END); 2636 2637 while (--iters != 0) { 2638 offset = ztest_random(fsize / (leaves << bshift)) * 2639 (leaves << bshift) + (leaf << bshift) + 2640 (ztest_random(1ULL << (bshift - 1)) & -8ULL); 2641 2642 if (offset >= fsize) 2643 continue; 2644 2645 if (zopt_verbose >= 6) 2646 (void) printf("injecting bad word into %s," 2647 " offset 0x%llx\n", pathrand, (u_longlong_t)offset); 2648 2649 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) 2650 fatal(1, "can't inject bad word at 0x%llx in %s", 2651 offset, pathrand); 2652 } 2653 2654 (void) close(fd); 2655} 2656 2657/* 2658 * Scrub the pool. 2659 */ 2660void 2661ztest_scrub(ztest_args_t *za) 2662{ 2663 spa_t *spa = dmu_objset_spa(za->za_os); 2664 2665 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE); 2666 (void) poll(NULL, 0, 1000); /* wait a second, then force a restart */ 2667 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE); 2668} 2669 2670/* 2671 * Rename the pool to a different name and then rename it back. 2672 */ 2673void 2674ztest_spa_rename(ztest_args_t *za) 2675{ 2676 char *oldname, *newname; 2677 int error; 2678 spa_t *spa; 2679 2680 (void) rw_wrlock(&ztest_shared->zs_name_lock); 2681 2682 oldname = za->za_pool; 2683 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); 2684 (void) strcpy(newname, oldname); 2685 (void) strcat(newname, "_tmp"); 2686 2687 /* 2688 * Do the rename 2689 */ 2690 error = spa_rename(oldname, newname); 2691 if (error) 2692 fatal(0, "spa_rename('%s', '%s') = %d", oldname, 2693 newname, error); 2694 2695 /* 2696 * Try to open it under the old name, which shouldn't exist 2697 */ 2698 error = spa_open(oldname, &spa, FTAG); 2699 if (error != ENOENT) 2700 fatal(0, "spa_open('%s') = %d", oldname, error); 2701 2702 /* 2703 * Open it under the new name and make sure it's still the same spa_t. 2704 */ 2705 error = spa_open(newname, &spa, FTAG); 2706 if (error != 0) 2707 fatal(0, "spa_open('%s') = %d", newname, error); 2708 2709 ASSERT(spa == dmu_objset_spa(za->za_os)); 2710 spa_close(spa, FTAG); 2711 2712 /* 2713 * Rename it back to the original 2714 */ 2715 error = spa_rename(newname, oldname); 2716 if (error) 2717 fatal(0, "spa_rename('%s', '%s') = %d", newname, 2718 oldname, error); 2719 2720 /* 2721 * Make sure it can still be opened 2722 */ 2723 error = spa_open(oldname, &spa, FTAG); 2724 if (error != 0) 2725 fatal(0, "spa_open('%s') = %d", oldname, error); 2726 2727 ASSERT(spa == dmu_objset_spa(za->za_os)); 2728 spa_close(spa, FTAG); 2729 2730 umem_free(newname, strlen(newname) + 1); 2731 2732 (void) rw_unlock(&ztest_shared->zs_name_lock); 2733} 2734 2735 2736/* 2737 * Completely obliterate one disk. 2738 */ 2739static void 2740ztest_obliterate_one_disk(uint64_t vdev) 2741{ 2742 int fd; 2743 char dev_name[MAXPATHLEN], copy_name[MAXPATHLEN]; 2744 size_t fsize; 2745 2746 if (zopt_maxfaults < 2) 2747 return; 2748 2749 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 2750 (void) snprintf(copy_name, MAXPATHLEN, "%s.old", dev_name); 2751 2752 fd = open(dev_name, O_RDWR); 2753 2754 if (fd == -1) 2755 fatal(1, "can't open %s", dev_name); 2756 2757 /* 2758 * Determine the size. 2759 */ 2760 fsize = lseek(fd, 0, SEEK_END); 2761 2762 (void) close(fd); 2763 2764 /* 2765 * Rename the old device to dev_name.old (useful for debugging). 2766 */ 2767 VERIFY(rename(dev_name, copy_name) == 0); 2768 2769 /* 2770 * Create a new one. 2771 */ 2772 VERIFY((fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666)) >= 0); 2773 VERIFY(ftruncate(fd, fsize) == 0); 2774 (void) close(fd); 2775} 2776 2777static void 2778ztest_replace_one_disk(spa_t *spa, uint64_t vdev) 2779{ 2780 char dev_name[MAXPATHLEN]; 2781 nvlist_t *file, *root; 2782 int error; 2783 uint64_t guid; 2784 uint64_t ashift = ztest_get_ashift(); 2785 vdev_t *vd; 2786 2787 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 2788 2789 /* 2790 * Build the nvlist describing dev_name. 2791 */ 2792 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 2793 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 2794 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0); 2795 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 2796 2797 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 2798 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 2799 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 2800 &file, 1) == 0); 2801 2802 spa_config_enter(spa, RW_READER, FTAG); 2803 if ((vd = vdev_lookup_by_path(spa->spa_root_vdev, dev_name)) == NULL) 2804 guid = 0; 2805 else 2806 guid = vd->vdev_guid; 2807 spa_config_exit(spa, FTAG); 2808 error = spa_vdev_attach(spa, guid, root, B_TRUE); 2809 if (error != 0 && 2810 error != EBUSY && 2811 error != ENOTSUP && 2812 error != ENODEV && 2813 error != EDOM) 2814 fatal(0, "spa_vdev_attach(in-place) = %d", error); 2815 2816 nvlist_free(file); 2817 nvlist_free(root); 2818} 2819 2820static void 2821ztest_verify_blocks(char *pool) 2822{ 2823 int status; 2824 char zdb[MAXPATHLEN + MAXNAMELEN + 20]; 2825 char zbuf[1024]; 2826 char *bin; 2827 FILE *fp; 2828 2829 if (realpath(progname, zdb) == NULL) 2830 assert(!"realpath() failed"); 2831 2832 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ 2833 bin = strstr(zdb, "/usr/bin/"); 2834 if (bin == NULL) 2835 bin = zdb; 2836 /* LINTED */ 2837 (void) sprintf(bin, "/usr/sbin/zdb -bc%s%s -U -O %s %s", 2838 zopt_verbose >= 3 ? "s" : "", 2839 zopt_verbose >= 4 ? "v" : "", 2840 ztest_random(2) == 0 ? "pre" : "post", pool); 2841 2842 if (zopt_verbose >= 5) 2843 (void) printf("Executing %s\n", strstr(zdb, "zdb ")); 2844 2845 fp = popen(zdb, "r"); 2846 assert(fp != NULL); 2847 2848 while (fgets(zbuf, sizeof (zbuf), fp) != NULL) 2849 if (zopt_verbose >= 3) 2850 (void) printf("%s", zbuf); 2851 2852 status = pclose(fp); 2853 2854 if (status == 0) 2855 return; 2856 2857 ztest_dump_core = 0; 2858 if (WIFEXITED(status)) 2859 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); 2860 else 2861 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); 2862} 2863 2864static void 2865ztest_walk_pool_directory(char *header) 2866{ 2867 spa_t *spa = NULL; 2868 2869 if (zopt_verbose >= 6) 2870 (void) printf("%s\n", header); 2871 2872 mutex_enter(&spa_namespace_lock); 2873 while ((spa = spa_next(spa)) != NULL) 2874 if (zopt_verbose >= 6) 2875 (void) printf("\t%s\n", spa_name(spa)); 2876 mutex_exit(&spa_namespace_lock); 2877} 2878 2879static void 2880ztest_spa_import_export(char *oldname, char *newname) 2881{ 2882 nvlist_t *config; 2883 uint64_t pool_guid; 2884 spa_t *spa; 2885 int error; 2886 2887 if (zopt_verbose >= 4) { 2888 (void) printf("import/export: old = %s, new = %s\n", 2889 oldname, newname); 2890 } 2891 2892 /* 2893 * Clean up from previous runs. 2894 */ 2895 (void) spa_destroy(newname); 2896 2897 /* 2898 * Get the pool's configuration and guid. 2899 */ 2900 error = spa_open(oldname, &spa, FTAG); 2901 if (error) 2902 fatal(0, "spa_open('%s') = %d", oldname, error); 2903 2904 pool_guid = spa_guid(spa); 2905 spa_close(spa, FTAG); 2906 2907 ztest_walk_pool_directory("pools before export"); 2908 2909 /* 2910 * Export it. 2911 */ 2912 error = spa_export(oldname, &config); 2913 if (error) 2914 fatal(0, "spa_export('%s') = %d", oldname, error); 2915 2916 ztest_walk_pool_directory("pools after export"); 2917 2918 /* 2919 * Import it under the new name. 2920 */ 2921 error = spa_import(newname, config, NULL); 2922 if (error) 2923 fatal(0, "spa_import('%s') = %d", newname, error); 2924 2925 ztest_walk_pool_directory("pools after import"); 2926 2927 /* 2928 * Try to import it again -- should fail with EEXIST. 2929 */ 2930 error = spa_import(newname, config, NULL); 2931 if (error != EEXIST) 2932 fatal(0, "spa_import('%s') twice", newname); 2933 2934 /* 2935 * Try to import it under a different name -- should fail with EEXIST. 2936 */ 2937 error = spa_import(oldname, config, NULL); 2938 if (error != EEXIST) 2939 fatal(0, "spa_import('%s') under multiple names", newname); 2940 2941 /* 2942 * Verify that the pool is no longer visible under the old name. 2943 */ 2944 error = spa_open(oldname, &spa, FTAG); 2945 if (error != ENOENT) 2946 fatal(0, "spa_open('%s') = %d", newname, error); 2947 2948 /* 2949 * Verify that we can open and close the pool using the new name. 2950 */ 2951 error = spa_open(newname, &spa, FTAG); 2952 if (error) 2953 fatal(0, "spa_open('%s') = %d", newname, error); 2954 ASSERT(pool_guid == spa_guid(spa)); 2955 spa_close(spa, FTAG); 2956 2957 nvlist_free(config); 2958} 2959 2960static void * 2961ztest_thread(void *arg) 2962{ 2963 ztest_args_t *za = arg; 2964 ztest_shared_t *zs = ztest_shared; 2965 hrtime_t now, functime; 2966 ztest_info_t *zi; 2967 int f; 2968 2969 while ((now = gethrtime()) < za->za_stop) { 2970 /* 2971 * See if it's time to force a crash. 2972 */ 2973 if (now > za->za_kill) { 2974 dmu_tx_t *tx; 2975 uint64_t txg; 2976 2977 mutex_enter(&spa_namespace_lock); 2978 tx = dmu_tx_create(za->za_os); 2979 VERIFY(0 == dmu_tx_assign(tx, TXG_NOWAIT)); 2980 txg = dmu_tx_get_txg(tx); 2981 dmu_tx_commit(tx); 2982 zs->zs_txg = txg; 2983 if (zopt_verbose >= 3) 2984 (void) printf( 2985 "killing process after txg %lld\n", 2986 (u_longlong_t)txg); 2987 txg_wait_synced(dmu_objset_pool(za->za_os), txg); 2988 zs->zs_alloc = spa_get_alloc(dmu_objset_spa(za->za_os)); 2989 zs->zs_space = spa_get_space(dmu_objset_spa(za->za_os)); 2990 (void) kill(getpid(), SIGKILL); 2991 } 2992 2993 /* 2994 * Pick a random function. 2995 */ 2996 f = ztest_random(ZTEST_FUNCS); 2997 zi = &zs->zs_info[f]; 2998 2999 /* 3000 * Decide whether to call it, based on the requested frequency. 3001 */ 3002 if (zi->zi_call_target == 0 || 3003 (double)zi->zi_call_total / zi->zi_call_target > 3004 (double)(now - zs->zs_start_time) / (zopt_time * NANOSEC)) 3005 continue; 3006 3007 atomic_add_64(&zi->zi_calls, 1); 3008 atomic_add_64(&zi->zi_call_total, 1); 3009 3010 za->za_diroff = (za->za_instance * ZTEST_FUNCS + f) * 3011 ZTEST_DIRSIZE; 3012 za->za_diroff_shared = (1ULL << 63); 3013 3014 ztest_dmu_write_parallel(za); 3015 3016 zi->zi_func(za); 3017 3018 functime = gethrtime() - now; 3019 3020 atomic_add_64(&zi->zi_call_time, functime); 3021 3022 if (zopt_verbose >= 4) { 3023 Dl_info dli; 3024 (void) dladdr((void *)zi->zi_func, &dli); 3025 (void) printf("%6.2f sec in %s\n", 3026 (double)functime / NANOSEC, dli.dli_sname); 3027 } 3028 3029 /* 3030 * If we're getting ENOSPC with some regularity, stop. 3031 */ 3032 if (zs->zs_enospc_count > 10) 3033 break; 3034 } 3035 3036 return (NULL); 3037} 3038 3039/* 3040 * Kick off threads to run tests on all datasets in parallel. 3041 */ 3042static void 3043ztest_run(char *pool) 3044{ 3045 int t, d, error; 3046 ztest_shared_t *zs = ztest_shared; 3047 ztest_args_t *za; 3048 spa_t *spa; 3049 char name[100]; 3050 3051 (void) _mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL); 3052 (void) rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL); 3053 3054 for (t = 0; t < ZTEST_SYNC_LOCKS; t++) 3055 (void) _mutex_init(&zs->zs_sync_lock[t], USYNC_THREAD, NULL); 3056 3057 /* 3058 * Destroy one disk before we even start. 3059 * It's mirrored, so everything should work just fine. 3060 * This makes us exercise fault handling very early in spa_load(). 3061 */ 3062 ztest_obliterate_one_disk(0); 3063 3064 /* 3065 * Verify that the sum of the sizes of all blocks in the pool 3066 * equals the SPA's allocated space total. 3067 */ 3068 ztest_verify_blocks(pool); 3069 3070 /* 3071 * Kick off a replacement of the disk we just obliterated. 3072 */ 3073 kernel_init(FREAD | FWRITE); 3074 error = spa_open(pool, &spa, FTAG); 3075 if (error) 3076 fatal(0, "spa_open(%s) = %d", pool, error); 3077 ztest_replace_one_disk(spa, 0); 3078 if (zopt_verbose >= 5) 3079 show_pool_stats(spa); 3080 spa_close(spa, FTAG); 3081 kernel_fini(); 3082 3083 kernel_init(FREAD | FWRITE); 3084 3085 /* 3086 * Verify that we can export the pool and reimport it under a 3087 * different name. 3088 */ 3089 if (ztest_random(2) == 0) { 3090 (void) snprintf(name, 100, "%s_import", pool); 3091 ztest_spa_import_export(pool, name); 3092 ztest_spa_import_export(name, pool); 3093 } 3094 3095 /* 3096 * Verify that we can loop over all pools. 3097 */ 3098 mutex_enter(&spa_namespace_lock); 3099 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) { 3100 if (zopt_verbose > 3) { 3101 (void) printf("spa_next: found %s\n", spa_name(spa)); 3102 } 3103 } 3104 mutex_exit(&spa_namespace_lock); 3105 3106 /* 3107 * Open our pool. 3108 */ 3109 error = spa_open(pool, &spa, FTAG); 3110 if (error) 3111 fatal(0, "spa_open() = %d", error); 3112 3113 /* 3114 * Verify that we can safely inquire about about any object, 3115 * whether it's allocated or not. To make it interesting, 3116 * we probe a 5-wide window around each power of two. 3117 * This hits all edge cases, including zero and the max. 3118 */ 3119 for (t = 0; t < 64; t++) { 3120 for (d = -5; d <= 5; d++) { 3121 error = dmu_object_info(spa->spa_meta_objset, 3122 (1ULL << t) + d, NULL); 3123 ASSERT(error == 0 || error == ENOENT || 3124 error == EINVAL); 3125 } 3126 } 3127 3128 /* 3129 * Now kick off all the tests that run in parallel. 3130 */ 3131 zs->zs_enospc_count = 0; 3132 3133 za = umem_zalloc(zopt_threads * sizeof (ztest_args_t), UMEM_NOFAIL); 3134 3135 if (zopt_verbose >= 4) 3136 (void) printf("starting main threads...\n"); 3137 3138 za[0].za_start = gethrtime(); 3139 za[0].za_stop = za[0].za_start + zopt_passtime * NANOSEC; 3140 za[0].za_stop = MIN(za[0].za_stop, zs->zs_stop_time); 3141 za[0].za_kill = za[0].za_stop; 3142 if (ztest_random(100) < zopt_killrate) 3143 za[0].za_kill -= ztest_random(zopt_passtime * NANOSEC); 3144 3145 for (t = 0; t < zopt_threads; t++) { 3146 d = t % zopt_datasets; 3147 if (t < zopt_datasets) { 3148 ztest_replay_t zr; 3149 int test_future = FALSE; 3150 (void) rw_rdlock(&ztest_shared->zs_name_lock); 3151 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d); 3152 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 3153 ztest_create_cb, NULL); 3154 if (error == EEXIST) { 3155 test_future = TRUE; 3156 } else if (error != 0) { 3157 if (error == ENOSPC) { 3158 zs->zs_enospc_count++; 3159 (void) rw_unlock( 3160 &ztest_shared->zs_name_lock); 3161 break; 3162 } 3163 fatal(0, "dmu_objset_create(%s) = %d", 3164 name, error); 3165 } 3166 error = dmu_objset_open(name, DMU_OST_OTHER, 3167 DS_MODE_STANDARD, &za[d].za_os); 3168 if (error) 3169 fatal(0, "dmu_objset_open('%s') = %d", 3170 name, error); 3171 (void) rw_unlock(&ztest_shared->zs_name_lock); 3172 if (test_future && ztest_shared->zs_txg > 0) 3173 ztest_dmu_check_future_leak(za[d].za_os, 3174 ztest_shared->zs_txg); 3175 zr.zr_os = za[d].za_os; 3176 zil_replay(zr.zr_os, &zr, &zr.zr_assign, 3177 ztest_replay_vector); 3178 za[d].za_zilog = zil_open(za[d].za_os, NULL); 3179 } 3180 za[t].za_pool = spa_strdup(pool); 3181 za[t].za_os = za[d].za_os; 3182 za[t].za_zilog = za[d].za_zilog; 3183 za[t].za_instance = t; 3184 za[t].za_random = ztest_random(-1ULL); 3185 za[t].za_start = za[0].za_start; 3186 za[t].za_stop = za[0].za_stop; 3187 za[t].za_kill = za[0].za_kill; 3188 3189 error = thr_create(0, 0, ztest_thread, &za[t], THR_BOUND, 3190 &za[t].za_thread); 3191 if (error) 3192 fatal(0, "can't create thread %d: error %d", 3193 t, error); 3194 } 3195 ztest_shared->zs_txg = 0; 3196 3197 while (--t >= 0) { 3198 error = thr_join(za[t].za_thread, NULL, NULL); 3199 if (error) 3200 fatal(0, "thr_join(%d) = %d", t, error); 3201 if (za[t].za_th) 3202 traverse_fini(za[t].za_th); 3203 if (t < zopt_datasets) { 3204 zil_close(za[t].za_zilog); 3205 dmu_objset_close(za[t].za_os); 3206 } 3207 spa_strfree(za[t].za_pool); 3208 } 3209 3210 umem_free(za, zopt_threads * sizeof (ztest_args_t)); 3211 3212 if (zopt_verbose >= 3) 3213 show_pool_stats(spa); 3214 3215 txg_wait_synced(spa_get_dsl(spa), 0); 3216 3217 zs->zs_alloc = spa_get_alloc(spa); 3218 zs->zs_space = spa_get_space(spa); 3219 3220 /* 3221 * Did we have out-of-space errors? If so, destroy a random objset. 3222 */ 3223 if (zs->zs_enospc_count != 0) { 3224 (void) rw_rdlock(&ztest_shared->zs_name_lock); 3225 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, 3226 (int)ztest_random(zopt_datasets)); 3227 if (zopt_verbose >= 3) 3228 (void) printf("Destroying %s to free up space\n", name); 3229 (void) dmu_objset_find(name, ztest_destroy_cb, NULL, 3230 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); 3231 (void) rw_unlock(&ztest_shared->zs_name_lock); 3232 } 3233 3234 txg_wait_synced(spa_get_dsl(spa), 0); 3235 3236 /* 3237 * Right before closing the pool, kick off a bunch of async I/O; 3238 * spa_close() should wait for it to complete. 3239 */ 3240 for (t = 1; t < 50; t++) 3241 dmu_prefetch(spa->spa_meta_objset, t, 0, 1 << 15); 3242 3243 spa_close(spa, FTAG); 3244 3245 kernel_fini(); 3246} 3247 3248void 3249print_time(hrtime_t t, char *timebuf) 3250{ 3251 hrtime_t s = t / NANOSEC; 3252 hrtime_t m = s / 60; 3253 hrtime_t h = m / 60; 3254 hrtime_t d = h / 24; 3255 3256 s -= m * 60; 3257 m -= h * 60; 3258 h -= d * 24; 3259 3260 timebuf[0] = '\0'; 3261 3262 if (d) 3263 (void) sprintf(timebuf, 3264 "%llud%02lluh%02llum%02llus", d, h, m, s); 3265 else if (h) 3266 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); 3267 else if (m) 3268 (void) sprintf(timebuf, "%llum%02llus", m, s); 3269 else 3270 (void) sprintf(timebuf, "%llus", s); 3271} 3272 3273/* 3274 * Create a storage pool with the given name and initial vdev size. 3275 * Then create the specified number of datasets in the pool. 3276 */ 3277static void 3278ztest_init(char *pool) 3279{ 3280 spa_t *spa; 3281 int error; 3282 nvlist_t *nvroot; 3283 3284 kernel_init(FREAD | FWRITE); 3285 3286 /* 3287 * Create the storage pool. 3288 */ 3289 (void) spa_destroy(pool); 3290 ztest_shared->zs_vdev_primaries = 0; 3291 nvroot = make_vdev_root(zopt_vdev_size, zopt_raidz, zopt_mirrors, 1); 3292 error = spa_create(pool, nvroot, NULL); 3293 nvlist_free(nvroot); 3294 3295 if (error) 3296 fatal(0, "spa_create() = %d", error); 3297 error = spa_open(pool, &spa, FTAG); 3298 if (error) 3299 fatal(0, "spa_open() = %d", error); 3300 3301 if (zopt_verbose >= 3) 3302 show_pool_stats(spa); 3303 3304 spa_close(spa, FTAG); 3305 3306 kernel_fini(); 3307} 3308 3309int 3310main(int argc, char **argv) 3311{ 3312 int kills = 0; 3313 int iters = 0; 3314 int i, f; 3315 ztest_shared_t *zs; 3316 ztest_info_t *zi; 3317 char timebuf[100]; 3318 char numbuf[6]; 3319 3320 (void) setvbuf(stdout, NULL, _IOLBF, 0); 3321 3322 /* Override location of zpool.cache */ 3323 spa_config_dir = "/tmp"; 3324 3325 ztest_random_fd = open("/dev/urandom", O_RDONLY); 3326 3327 process_options(argc, argv); 3328 3329 argc -= optind; 3330 argv += optind; 3331 3332 dprintf_setup(&argc, argv); 3333 3334 /* 3335 * Blow away any existing copy of zpool.cache 3336 */ 3337 if (zopt_init != 0) 3338 (void) remove("/tmp/zpool.cache"); 3339 3340 zs = ztest_shared = (void *)mmap(0, 3341 P2ROUNDUP(sizeof (ztest_shared_t), getpagesize()), 3342 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); 3343 3344 if (zopt_verbose >= 1) { 3345 (void) printf("%llu vdevs, %d datasets, %d threads," 3346 " %llu seconds...\n", 3347 (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads, 3348 (u_longlong_t)zopt_time); 3349 } 3350 3351 /* 3352 * Create and initialize our storage pool. 3353 */ 3354 for (i = 1; i <= zopt_init; i++) { 3355 bzero(zs, sizeof (ztest_shared_t)); 3356 if (zopt_verbose >= 3 && zopt_init != 1) 3357 (void) printf("ztest_init(), pass %d\n", i); 3358 ztest_init(zopt_pool); 3359 } 3360 3361 /* 3362 * Initialize the call targets for each function. 3363 */ 3364 for (f = 0; f < ZTEST_FUNCS; f++) { 3365 zi = &zs->zs_info[f]; 3366 3367 *zi = ztest_info[f]; 3368 3369 if (*zi->zi_interval == 0) 3370 zi->zi_call_target = UINT64_MAX; 3371 else 3372 zi->zi_call_target = zopt_time / *zi->zi_interval; 3373 } 3374 3375 zs->zs_start_time = gethrtime(); 3376 zs->zs_stop_time = zs->zs_start_time + zopt_time * NANOSEC; 3377 3378 /* 3379 * Run the tests in a loop. These tests include fault injection 3380 * to verify that self-healing data works, and forced crashes 3381 * to verify that we never lose on-disk consistency. 3382 */ 3383 while (gethrtime() < zs->zs_stop_time) { 3384 int status; 3385 pid_t pid; 3386 char *tmp; 3387 3388 /* 3389 * Initialize the workload counters for each function. 3390 */ 3391 for (f = 0; f < ZTEST_FUNCS; f++) { 3392 zi = &zs->zs_info[f]; 3393 zi->zi_calls = 0; 3394 zi->zi_call_time = 0; 3395 } 3396 3397 pid = fork(); 3398 3399 if (pid == -1) 3400 fatal(1, "fork failed"); 3401 3402 if (pid == 0) { /* child */ 3403 struct rlimit rl = { 1024, 1024 }; 3404 (void) setrlimit(RLIMIT_NOFILE, &rl); 3405 (void) enable_extended_FILE_stdio(-1, -1); 3406 ztest_run(zopt_pool); 3407 exit(0); 3408 } 3409 3410 while (waitpid(pid, &status, 0) != pid) 3411 continue; 3412 3413 if (WIFEXITED(status)) { 3414 if (WEXITSTATUS(status) != 0) { 3415 (void) fprintf(stderr, 3416 "child exited with code %d\n", 3417 WEXITSTATUS(status)); 3418 exit(2); 3419 } 3420 } else if (WIFSIGNALED(status)) { 3421 if (WTERMSIG(status) != SIGKILL) { 3422 (void) fprintf(stderr, 3423 "child died with signal %d\n", 3424 WTERMSIG(status)); 3425 exit(3); 3426 } 3427 kills++; 3428 } else { 3429 (void) fprintf(stderr, "something strange happened " 3430 "to child\n"); 3431 exit(4); 3432 } 3433 3434 iters++; 3435 3436 if (zopt_verbose >= 1) { 3437 hrtime_t now = gethrtime(); 3438 3439 now = MIN(now, zs->zs_stop_time); 3440 print_time(zs->zs_stop_time - now, timebuf); 3441 nicenum(zs->zs_space, numbuf); 3442 3443 (void) printf("Pass %3d, %8s, %3llu ENOSPC, " 3444 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", 3445 iters, 3446 WIFEXITED(status) ? "Complete" : "SIGKILL", 3447 (u_longlong_t)zs->zs_enospc_count, 3448 100.0 * zs->zs_alloc / zs->zs_space, 3449 numbuf, 3450 100.0 * (now - zs->zs_start_time) / 3451 (zopt_time * NANOSEC), timebuf); 3452 } 3453 3454 if (zopt_verbose >= 2) { 3455 (void) printf("\nWorkload summary:\n\n"); 3456 (void) printf("%7s %9s %s\n", 3457 "Calls", "Time", "Function"); 3458 (void) printf("%7s %9s %s\n", 3459 "-----", "----", "--------"); 3460 for (f = 0; f < ZTEST_FUNCS; f++) { 3461 Dl_info dli; 3462 3463 zi = &zs->zs_info[f]; 3464 print_time(zi->zi_call_time, timebuf); 3465 (void) dladdr((void *)zi->zi_func, &dli); 3466 (void) printf("%7llu %9s %s\n", 3467 (u_longlong_t)zi->zi_calls, timebuf, 3468 dli.dli_sname); 3469 } 3470 (void) printf("\n"); 3471 } 3472 3473 /* 3474 * It's possible that we killed a child during a rename test, in 3475 * which case we'll have a 'ztest_tmp' pool lying around instead 3476 * of 'ztest'. Do a blind rename in case this happened. 3477 */ 3478 tmp = umem_alloc(strlen(zopt_pool) + 5, UMEM_NOFAIL); 3479 (void) strcpy(tmp, zopt_pool); 3480 (void) strcat(tmp, "_tmp"); 3481 kernel_init(FREAD | FWRITE); 3482 (void) spa_rename(tmp, zopt_pool); 3483 kernel_fini(); 3484 umem_free(tmp, strlen(tmp) + 1); 3485 } 3486 3487 ztest_verify_blocks(zopt_pool); 3488 3489 if (zopt_verbose >= 1) { 3490 (void) printf("%d killed, %d completed, %.0f%% kill rate\n", 3491 kills, iters - kills, (100.0 * kills) / MAX(1, iters)); 3492 } 3493 3494 return (0); 3495} 3496