ztest.c revision 297112
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2011, 2015 by Delphix. All rights reserved. 24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25 * Copyright (c) 2012 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 26 * Copyright (c) 2013 Steven Hartland. All rights reserved. 27 * Copyright (c) 2014 Integros [integros.com] 28 */ 29 30/* 31 * The objective of this program is to provide a DMU/ZAP/SPA stress test 32 * that runs entirely in userland, is easy to use, and easy to extend. 33 * 34 * The overall design of the ztest program is as follows: 35 * 36 * (1) For each major functional area (e.g. adding vdevs to a pool, 37 * creating and destroying datasets, reading and writing objects, etc) 38 * we have a simple routine to test that functionality. These 39 * individual routines do not have to do anything "stressful". 40 * 41 * (2) We turn these simple functionality tests into a stress test by 42 * running them all in parallel, with as many threads as desired, 43 * and spread across as many datasets, objects, and vdevs as desired. 44 * 45 * (3) While all this is happening, we inject faults into the pool to 46 * verify that self-healing data really works. 47 * 48 * (4) Every time we open a dataset, we change its checksum and compression 49 * functions. Thus even individual objects vary from block to block 50 * in which checksum they use and whether they're compressed. 51 * 52 * (5) To verify that we never lose on-disk consistency after a crash, 53 * we run the entire test in a child of the main process. 54 * At random times, the child self-immolates with a SIGKILL. 55 * This is the software equivalent of pulling the power cord. 56 * The parent then runs the test again, using the existing 57 * storage pool, as many times as desired. If backwards compatibility 58 * testing is enabled ztest will sometimes run the "older" version 59 * of ztest after a SIGKILL. 60 * 61 * (6) To verify that we don't have future leaks or temporal incursions, 62 * many of the functional tests record the transaction group number 63 * as part of their data. When reading old data, they verify that 64 * the transaction group number is less than the current, open txg. 65 * If you add a new test, please do this if applicable. 66 * 67 * When run with no arguments, ztest runs for about five minutes and 68 * produces no output if successful. To get a little bit of information, 69 * specify -V. To get more information, specify -VV, and so on. 70 * 71 * To turn this into an overnight stress test, use -T to specify run time. 72 * 73 * You can ask more more vdevs [-v], datasets [-d], or threads [-t] 74 * to increase the pool capacity, fanout, and overall stress level. 75 * 76 * Use the -k option to set the desired frequency of kills. 77 * 78 * When ztest invokes itself it passes all relevant information through a 79 * temporary file which is mmap-ed in the child process. This allows shared 80 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always 81 * stored at offset 0 of this file and contains information on the size and 82 * number of shared structures in the file. The information stored in this file 83 * must remain backwards compatible with older versions of ztest so that 84 * ztest can invoke them during backwards compatibility testing (-B). 85 */ 86 87#include <sys/zfs_context.h> 88#include <sys/spa.h> 89#include <sys/dmu.h> 90#include <sys/txg.h> 91#include <sys/dbuf.h> 92#include <sys/zap.h> 93#include <sys/dmu_objset.h> 94#include <sys/poll.h> 95#include <sys/stat.h> 96#include <sys/time.h> 97#include <sys/wait.h> 98#include <sys/mman.h> 99#include <sys/resource.h> 100#include <sys/zio.h> 101#include <sys/zil.h> 102#include <sys/zil_impl.h> 103#include <sys/vdev_impl.h> 104#include <sys/vdev_file.h> 105#include <sys/spa_impl.h> 106#include <sys/metaslab_impl.h> 107#include <sys/dsl_prop.h> 108#include <sys/dsl_dataset.h> 109#include <sys/dsl_destroy.h> 110#include <sys/dsl_scan.h> 111#include <sys/zio_checksum.h> 112#include <sys/refcount.h> 113#include <sys/zfeature.h> 114#include <sys/dsl_userhold.h> 115#include <stdio.h> 116#include <stdio_ext.h> 117#include <stdlib.h> 118#include <unistd.h> 119#include <signal.h> 120#include <umem.h> 121#include <dlfcn.h> 122#include <ctype.h> 123#include <math.h> 124#include <errno.h> 125#include <sys/fs/zfs.h> 126#include <libnvpair.h> 127 128static int ztest_fd_data = -1; 129static int ztest_fd_rand = -1; 130 131typedef struct ztest_shared_hdr { 132 uint64_t zh_hdr_size; 133 uint64_t zh_opts_size; 134 uint64_t zh_size; 135 uint64_t zh_stats_size; 136 uint64_t zh_stats_count; 137 uint64_t zh_ds_size; 138 uint64_t zh_ds_count; 139} ztest_shared_hdr_t; 140 141static ztest_shared_hdr_t *ztest_shared_hdr; 142 143typedef struct ztest_shared_opts { 144 char zo_pool[MAXNAMELEN]; 145 char zo_dir[MAXNAMELEN]; 146 char zo_alt_ztest[MAXNAMELEN]; 147 char zo_alt_libpath[MAXNAMELEN]; 148 uint64_t zo_vdevs; 149 uint64_t zo_vdevtime; 150 size_t zo_vdev_size; 151 int zo_ashift; 152 int zo_mirrors; 153 int zo_raidz; 154 int zo_raidz_parity; 155 int zo_datasets; 156 int zo_threads; 157 uint64_t zo_passtime; 158 uint64_t zo_killrate; 159 int zo_verbose; 160 int zo_init; 161 uint64_t zo_time; 162 uint64_t zo_maxloops; 163 uint64_t zo_metaslab_gang_bang; 164} ztest_shared_opts_t; 165 166static const ztest_shared_opts_t ztest_opts_defaults = { 167 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' }, 168 .zo_dir = { '/', 't', 'm', 'p', '\0' }, 169 .zo_alt_ztest = { '\0' }, 170 .zo_alt_libpath = { '\0' }, 171 .zo_vdevs = 5, 172 .zo_ashift = SPA_MINBLOCKSHIFT, 173 .zo_mirrors = 2, 174 .zo_raidz = 4, 175 .zo_raidz_parity = 1, 176 .zo_vdev_size = SPA_MINDEVSIZE * 2, 177 .zo_datasets = 7, 178 .zo_threads = 23, 179 .zo_passtime = 60, /* 60 seconds */ 180 .zo_killrate = 70, /* 70% kill rate */ 181 .zo_verbose = 0, 182 .zo_init = 1, 183 .zo_time = 300, /* 5 minutes */ 184 .zo_maxloops = 50, /* max loops during spa_freeze() */ 185 .zo_metaslab_gang_bang = 32 << 10 186}; 187 188extern uint64_t metaslab_gang_bang; 189extern uint64_t metaslab_df_alloc_threshold; 190extern uint64_t zfs_deadman_synctime_ms; 191extern int metaslab_preload_limit; 192 193static ztest_shared_opts_t *ztest_shared_opts; 194static ztest_shared_opts_t ztest_opts; 195 196typedef struct ztest_shared_ds { 197 uint64_t zd_seq; 198} ztest_shared_ds_t; 199 200static ztest_shared_ds_t *ztest_shared_ds; 201#define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d]) 202 203#define BT_MAGIC 0x123456789abcdefULL 204#define MAXFAULTS() \ 205 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1) 206 207enum ztest_io_type { 208 ZTEST_IO_WRITE_TAG, 209 ZTEST_IO_WRITE_PATTERN, 210 ZTEST_IO_WRITE_ZEROES, 211 ZTEST_IO_TRUNCATE, 212 ZTEST_IO_SETATTR, 213 ZTEST_IO_REWRITE, 214 ZTEST_IO_TYPES 215}; 216 217typedef struct ztest_block_tag { 218 uint64_t bt_magic; 219 uint64_t bt_objset; 220 uint64_t bt_object; 221 uint64_t bt_offset; 222 uint64_t bt_gen; 223 uint64_t bt_txg; 224 uint64_t bt_crtxg; 225} ztest_block_tag_t; 226 227typedef struct bufwad { 228 uint64_t bw_index; 229 uint64_t bw_txg; 230 uint64_t bw_data; 231} bufwad_t; 232 233/* 234 * XXX -- fix zfs range locks to be generic so we can use them here. 235 */ 236typedef enum { 237 RL_READER, 238 RL_WRITER, 239 RL_APPEND 240} rl_type_t; 241 242typedef struct rll { 243 void *rll_writer; 244 int rll_readers; 245 mutex_t rll_lock; 246 cond_t rll_cv; 247} rll_t; 248 249typedef struct rl { 250 uint64_t rl_object; 251 uint64_t rl_offset; 252 uint64_t rl_size; 253 rll_t *rl_lock; 254} rl_t; 255 256#define ZTEST_RANGE_LOCKS 64 257#define ZTEST_OBJECT_LOCKS 64 258 259/* 260 * Object descriptor. Used as a template for object lookup/create/remove. 261 */ 262typedef struct ztest_od { 263 uint64_t od_dir; 264 uint64_t od_object; 265 dmu_object_type_t od_type; 266 dmu_object_type_t od_crtype; 267 uint64_t od_blocksize; 268 uint64_t od_crblocksize; 269 uint64_t od_gen; 270 uint64_t od_crgen; 271 char od_name[MAXNAMELEN]; 272} ztest_od_t; 273 274/* 275 * Per-dataset state. 276 */ 277typedef struct ztest_ds { 278 ztest_shared_ds_t *zd_shared; 279 objset_t *zd_os; 280 rwlock_t zd_zilog_lock; 281 zilog_t *zd_zilog; 282 ztest_od_t *zd_od; /* debugging aid */ 283 char zd_name[MAXNAMELEN]; 284 mutex_t zd_dirobj_lock; 285 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS]; 286 rll_t zd_range_lock[ZTEST_RANGE_LOCKS]; 287} ztest_ds_t; 288 289/* 290 * Per-iteration state. 291 */ 292typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id); 293 294typedef struct ztest_info { 295 ztest_func_t *zi_func; /* test function */ 296 uint64_t zi_iters; /* iterations per execution */ 297 uint64_t *zi_interval; /* execute every <interval> seconds */ 298} ztest_info_t; 299 300typedef struct ztest_shared_callstate { 301 uint64_t zc_count; /* per-pass count */ 302 uint64_t zc_time; /* per-pass time */ 303 uint64_t zc_next; /* next time to call this function */ 304} ztest_shared_callstate_t; 305 306static ztest_shared_callstate_t *ztest_shared_callstate; 307#define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c]) 308 309/* 310 * Note: these aren't static because we want dladdr() to work. 311 */ 312ztest_func_t ztest_dmu_read_write; 313ztest_func_t ztest_dmu_write_parallel; 314ztest_func_t ztest_dmu_object_alloc_free; 315ztest_func_t ztest_dmu_commit_callbacks; 316ztest_func_t ztest_zap; 317ztest_func_t ztest_zap_parallel; 318ztest_func_t ztest_zil_commit; 319ztest_func_t ztest_zil_remount; 320ztest_func_t ztest_dmu_read_write_zcopy; 321ztest_func_t ztest_dmu_objset_create_destroy; 322ztest_func_t ztest_dmu_prealloc; 323ztest_func_t ztest_fzap; 324ztest_func_t ztest_dmu_snapshot_create_destroy; 325ztest_func_t ztest_dsl_prop_get_set; 326ztest_func_t ztest_spa_prop_get_set; 327ztest_func_t ztest_spa_create_destroy; 328ztest_func_t ztest_fault_inject; 329ztest_func_t ztest_ddt_repair; 330ztest_func_t ztest_dmu_snapshot_hold; 331ztest_func_t ztest_spa_rename; 332ztest_func_t ztest_scrub; 333ztest_func_t ztest_dsl_dataset_promote_busy; 334ztest_func_t ztest_vdev_attach_detach; 335ztest_func_t ztest_vdev_LUN_growth; 336ztest_func_t ztest_vdev_add_remove; 337ztest_func_t ztest_vdev_aux_add_remove; 338ztest_func_t ztest_split_pool; 339ztest_func_t ztest_reguid; 340ztest_func_t ztest_spa_upgrade; 341 342uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */ 343uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */ 344uint64_t zopt_often = 1ULL * NANOSEC; /* every second */ 345uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */ 346uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */ 347 348ztest_info_t ztest_info[] = { 349 { ztest_dmu_read_write, 1, &zopt_always }, 350 { ztest_dmu_write_parallel, 10, &zopt_always }, 351 { ztest_dmu_object_alloc_free, 1, &zopt_always }, 352 { ztest_dmu_commit_callbacks, 1, &zopt_always }, 353 { ztest_zap, 30, &zopt_always }, 354 { ztest_zap_parallel, 100, &zopt_always }, 355 { ztest_split_pool, 1, &zopt_always }, 356 { ztest_zil_commit, 1, &zopt_incessant }, 357 { ztest_zil_remount, 1, &zopt_sometimes }, 358 { ztest_dmu_read_write_zcopy, 1, &zopt_often }, 359 { ztest_dmu_objset_create_destroy, 1, &zopt_often }, 360 { ztest_dsl_prop_get_set, 1, &zopt_often }, 361 { ztest_spa_prop_get_set, 1, &zopt_sometimes }, 362#if 0 363 { ztest_dmu_prealloc, 1, &zopt_sometimes }, 364#endif 365 { ztest_fzap, 1, &zopt_sometimes }, 366 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes }, 367 { ztest_spa_create_destroy, 1, &zopt_sometimes }, 368 { ztest_fault_inject, 1, &zopt_sometimes }, 369 { ztest_ddt_repair, 1, &zopt_sometimes }, 370 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes }, 371 { ztest_reguid, 1, &zopt_rarely }, 372 { ztest_spa_rename, 1, &zopt_rarely }, 373 { ztest_scrub, 1, &zopt_rarely }, 374 { ztest_spa_upgrade, 1, &zopt_rarely }, 375 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely }, 376 { ztest_vdev_attach_detach, 1, &zopt_sometimes }, 377 { ztest_vdev_LUN_growth, 1, &zopt_rarely }, 378 { ztest_vdev_add_remove, 1, 379 &ztest_opts.zo_vdevtime }, 380 { ztest_vdev_aux_add_remove, 1, 381 &ztest_opts.zo_vdevtime }, 382}; 383 384#define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) 385 386/* 387 * The following struct is used to hold a list of uncalled commit callbacks. 388 * The callbacks are ordered by txg number. 389 */ 390typedef struct ztest_cb_list { 391 mutex_t zcl_callbacks_lock; 392 list_t zcl_callbacks; 393} ztest_cb_list_t; 394 395/* 396 * Stuff we need to share writably between parent and child. 397 */ 398typedef struct ztest_shared { 399 boolean_t zs_do_init; 400 hrtime_t zs_proc_start; 401 hrtime_t zs_proc_stop; 402 hrtime_t zs_thread_start; 403 hrtime_t zs_thread_stop; 404 hrtime_t zs_thread_kill; 405 uint64_t zs_enospc_count; 406 uint64_t zs_vdev_next_leaf; 407 uint64_t zs_vdev_aux; 408 uint64_t zs_alloc; 409 uint64_t zs_space; 410 uint64_t zs_splits; 411 uint64_t zs_mirrors; 412 uint64_t zs_metaslab_sz; 413 uint64_t zs_metaslab_df_alloc_threshold; 414 uint64_t zs_guid; 415} ztest_shared_t; 416 417#define ID_PARALLEL -1ULL 418 419static char ztest_dev_template[] = "%s/%s.%llua"; 420static char ztest_aux_template[] = "%s/%s.%s.%llu"; 421ztest_shared_t *ztest_shared; 422 423static spa_t *ztest_spa = NULL; 424static ztest_ds_t *ztest_ds; 425 426static mutex_t ztest_vdev_lock; 427 428/* 429 * The ztest_name_lock protects the pool and dataset namespace used by 430 * the individual tests. To modify the namespace, consumers must grab 431 * this lock as writer. Grabbing the lock as reader will ensure that the 432 * namespace does not change while the lock is held. 433 */ 434static rwlock_t ztest_name_lock; 435 436static boolean_t ztest_dump_core = B_TRUE; 437static boolean_t ztest_exiting; 438 439/* Global commit callback list */ 440static ztest_cb_list_t zcl; 441 442enum ztest_object { 443 ZTEST_META_DNODE = 0, 444 ZTEST_DIROBJ, 445 ZTEST_OBJECTS 446}; 447 448static void usage(boolean_t) __NORETURN; 449 450/* 451 * These libumem hooks provide a reasonable set of defaults for the allocator's 452 * debugging facilities. 453 */ 454const char * 455_umem_debug_init() 456{ 457 return ("default,verbose"); /* $UMEM_DEBUG setting */ 458} 459 460const char * 461_umem_logging_init(void) 462{ 463 return ("fail,contents"); /* $UMEM_LOGGING setting */ 464} 465 466#define FATAL_MSG_SZ 1024 467 468char *fatal_msg; 469 470static void 471fatal(int do_perror, char *message, ...) 472{ 473 va_list args; 474 int save_errno = errno; 475 char buf[FATAL_MSG_SZ]; 476 477 (void) fflush(stdout); 478 479 va_start(args, message); 480 (void) sprintf(buf, "ztest: "); 481 /* LINTED */ 482 (void) vsprintf(buf + strlen(buf), message, args); 483 va_end(args); 484 if (do_perror) { 485 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), 486 ": %s", strerror(save_errno)); 487 } 488 (void) fprintf(stderr, "%s\n", buf); 489 fatal_msg = buf; /* to ease debugging */ 490 if (ztest_dump_core) 491 abort(); 492 exit(3); 493} 494 495static int 496str2shift(const char *buf) 497{ 498 const char *ends = "BKMGTPEZ"; 499 int i; 500 501 if (buf[0] == '\0') 502 return (0); 503 for (i = 0; i < strlen(ends); i++) { 504 if (toupper(buf[0]) == ends[i]) 505 break; 506 } 507 if (i == strlen(ends)) { 508 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", 509 buf); 510 usage(B_FALSE); 511 } 512 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { 513 return (10*i); 514 } 515 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); 516 usage(B_FALSE); 517 /* NOTREACHED */ 518} 519 520static uint64_t 521nicenumtoull(const char *buf) 522{ 523 char *end; 524 uint64_t val; 525 526 val = strtoull(buf, &end, 0); 527 if (end == buf) { 528 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); 529 usage(B_FALSE); 530 } else if (end[0] == '.') { 531 double fval = strtod(buf, &end); 532 fval *= pow(2, str2shift(end)); 533 if (fval > UINT64_MAX) { 534 (void) fprintf(stderr, "ztest: value too large: %s\n", 535 buf); 536 usage(B_FALSE); 537 } 538 val = (uint64_t)fval; 539 } else { 540 int shift = str2shift(end); 541 if (shift >= 64 || (val << shift) >> shift != val) { 542 (void) fprintf(stderr, "ztest: value too large: %s\n", 543 buf); 544 usage(B_FALSE); 545 } 546 val <<= shift; 547 } 548 return (val); 549} 550 551static void 552usage(boolean_t requested) 553{ 554 const ztest_shared_opts_t *zo = &ztest_opts_defaults; 555 556 char nice_vdev_size[10]; 557 char nice_gang_bang[10]; 558 FILE *fp = requested ? stdout : stderr; 559 560 nicenum(zo->zo_vdev_size, nice_vdev_size); 561 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang); 562 563 (void) fprintf(fp, "Usage: %s\n" 564 "\t[-v vdevs (default: %llu)]\n" 565 "\t[-s size_of_each_vdev (default: %s)]\n" 566 "\t[-a alignment_shift (default: %d)] use 0 for random\n" 567 "\t[-m mirror_copies (default: %d)]\n" 568 "\t[-r raidz_disks (default: %d)]\n" 569 "\t[-R raidz_parity (default: %d)]\n" 570 "\t[-d datasets (default: %d)]\n" 571 "\t[-t threads (default: %d)]\n" 572 "\t[-g gang_block_threshold (default: %s)]\n" 573 "\t[-i init_count (default: %d)] initialize pool i times\n" 574 "\t[-k kill_percentage (default: %llu%%)]\n" 575 "\t[-p pool_name (default: %s)]\n" 576 "\t[-f dir (default: %s)] file directory for vdev files\n" 577 "\t[-V] verbose (use multiple times for ever more blather)\n" 578 "\t[-E] use existing pool instead of creating new one\n" 579 "\t[-T time (default: %llu sec)] total run time\n" 580 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n" 581 "\t[-P passtime (default: %llu sec)] time per pass\n" 582 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n" 583 "\t[-h] (print help)\n" 584 "", 585 zo->zo_pool, 586 (u_longlong_t)zo->zo_vdevs, /* -v */ 587 nice_vdev_size, /* -s */ 588 zo->zo_ashift, /* -a */ 589 zo->zo_mirrors, /* -m */ 590 zo->zo_raidz, /* -r */ 591 zo->zo_raidz_parity, /* -R */ 592 zo->zo_datasets, /* -d */ 593 zo->zo_threads, /* -t */ 594 nice_gang_bang, /* -g */ 595 zo->zo_init, /* -i */ 596 (u_longlong_t)zo->zo_killrate, /* -k */ 597 zo->zo_pool, /* -p */ 598 zo->zo_dir, /* -f */ 599 (u_longlong_t)zo->zo_time, /* -T */ 600 (u_longlong_t)zo->zo_maxloops, /* -F */ 601 (u_longlong_t)zo->zo_passtime); 602 exit(requested ? 0 : 1); 603} 604 605static void 606process_options(int argc, char **argv) 607{ 608 char *path; 609 ztest_shared_opts_t *zo = &ztest_opts; 610 611 int opt; 612 uint64_t value; 613 char altdir[MAXNAMELEN] = { 0 }; 614 615 bcopy(&ztest_opts_defaults, zo, sizeof (*zo)); 616 617 while ((opt = getopt(argc, argv, 618 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) { 619 value = 0; 620 switch (opt) { 621 case 'v': 622 case 's': 623 case 'a': 624 case 'm': 625 case 'r': 626 case 'R': 627 case 'd': 628 case 't': 629 case 'g': 630 case 'i': 631 case 'k': 632 case 'T': 633 case 'P': 634 case 'F': 635 value = nicenumtoull(optarg); 636 } 637 switch (opt) { 638 case 'v': 639 zo->zo_vdevs = value; 640 break; 641 case 's': 642 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value); 643 break; 644 case 'a': 645 zo->zo_ashift = value; 646 break; 647 case 'm': 648 zo->zo_mirrors = value; 649 break; 650 case 'r': 651 zo->zo_raidz = MAX(1, value); 652 break; 653 case 'R': 654 zo->zo_raidz_parity = MIN(MAX(value, 1), 3); 655 break; 656 case 'd': 657 zo->zo_datasets = MAX(1, value); 658 break; 659 case 't': 660 zo->zo_threads = MAX(1, value); 661 break; 662 case 'g': 663 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, 664 value); 665 break; 666 case 'i': 667 zo->zo_init = value; 668 break; 669 case 'k': 670 zo->zo_killrate = value; 671 break; 672 case 'p': 673 (void) strlcpy(zo->zo_pool, optarg, 674 sizeof (zo->zo_pool)); 675 break; 676 case 'f': 677 path = realpath(optarg, NULL); 678 if (path == NULL) { 679 (void) fprintf(stderr, "error: %s: %s\n", 680 optarg, strerror(errno)); 681 usage(B_FALSE); 682 } else { 683 (void) strlcpy(zo->zo_dir, path, 684 sizeof (zo->zo_dir)); 685 } 686 break; 687 case 'V': 688 zo->zo_verbose++; 689 break; 690 case 'E': 691 zo->zo_init = 0; 692 break; 693 case 'T': 694 zo->zo_time = value; 695 break; 696 case 'P': 697 zo->zo_passtime = MAX(1, value); 698 break; 699 case 'F': 700 zo->zo_maxloops = MAX(1, value); 701 break; 702 case 'B': 703 (void) strlcpy(altdir, optarg, sizeof (altdir)); 704 break; 705 case 'h': 706 usage(B_TRUE); 707 break; 708 case '?': 709 default: 710 usage(B_FALSE); 711 break; 712 } 713 } 714 715 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1); 716 717 zo->zo_vdevtime = 718 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs : 719 UINT64_MAX >> 2); 720 721 if (strlen(altdir) > 0) { 722 char *cmd; 723 char *realaltdir; 724 char *bin; 725 char *ztest; 726 char *isa; 727 int isalen; 728 729 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 730 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 731 732 VERIFY(NULL != realpath(getexecname(), cmd)); 733 if (0 != access(altdir, F_OK)) { 734 ztest_dump_core = B_FALSE; 735 fatal(B_TRUE, "invalid alternate ztest path: %s", 736 altdir); 737 } 738 VERIFY(NULL != realpath(altdir, realaltdir)); 739 740 /* 741 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest". 742 * We want to extract <isa> to determine if we should use 743 * 32 or 64 bit binaries. 744 */ 745 bin = strstr(cmd, "/usr/bin/"); 746 ztest = strstr(bin, "/ztest"); 747 isa = bin + 9; 748 isalen = ztest - isa; 749 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest), 750 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa); 751 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath), 752 "%s/usr/lib/%.*s", realaltdir, isalen, isa); 753 754 if (0 != access(zo->zo_alt_ztest, X_OK)) { 755 ztest_dump_core = B_FALSE; 756 fatal(B_TRUE, "invalid alternate ztest: %s", 757 zo->zo_alt_ztest); 758 } else if (0 != access(zo->zo_alt_libpath, X_OK)) { 759 ztest_dump_core = B_FALSE; 760 fatal(B_TRUE, "invalid alternate lib directory %s", 761 zo->zo_alt_libpath); 762 } 763 764 umem_free(cmd, MAXPATHLEN); 765 umem_free(realaltdir, MAXPATHLEN); 766 } 767} 768 769static void 770ztest_kill(ztest_shared_t *zs) 771{ 772 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa)); 773 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa)); 774 775 /* 776 * Before we kill off ztest, make sure that the config is updated. 777 * See comment above spa_config_sync(). 778 */ 779 mutex_enter(&spa_namespace_lock); 780 spa_config_sync(ztest_spa, B_FALSE, B_FALSE); 781 mutex_exit(&spa_namespace_lock); 782 783 zfs_dbgmsg_print(FTAG); 784 (void) kill(getpid(), SIGKILL); 785} 786 787static uint64_t 788ztest_random(uint64_t range) 789{ 790 uint64_t r; 791 792 ASSERT3S(ztest_fd_rand, >=, 0); 793 794 if (range == 0) 795 return (0); 796 797 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r)) 798 fatal(1, "short read from /dev/urandom"); 799 800 return (r % range); 801} 802 803/* ARGSUSED */ 804static void 805ztest_record_enospc(const char *s) 806{ 807 ztest_shared->zs_enospc_count++; 808} 809 810static uint64_t 811ztest_get_ashift(void) 812{ 813 if (ztest_opts.zo_ashift == 0) 814 return (SPA_MINBLOCKSHIFT + ztest_random(5)); 815 return (ztest_opts.zo_ashift); 816} 817 818static nvlist_t * 819make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift) 820{ 821 char pathbuf[MAXPATHLEN]; 822 uint64_t vdev; 823 nvlist_t *file; 824 825 if (ashift == 0) 826 ashift = ztest_get_ashift(); 827 828 if (path == NULL) { 829 path = pathbuf; 830 831 if (aux != NULL) { 832 vdev = ztest_shared->zs_vdev_aux; 833 (void) snprintf(path, sizeof (pathbuf), 834 ztest_aux_template, ztest_opts.zo_dir, 835 pool == NULL ? ztest_opts.zo_pool : pool, 836 aux, vdev); 837 } else { 838 vdev = ztest_shared->zs_vdev_next_leaf++; 839 (void) snprintf(path, sizeof (pathbuf), 840 ztest_dev_template, ztest_opts.zo_dir, 841 pool == NULL ? ztest_opts.zo_pool : pool, vdev); 842 } 843 } 844 845 if (size != 0) { 846 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666); 847 if (fd == -1) 848 fatal(1, "can't open %s", path); 849 if (ftruncate(fd, size) != 0) 850 fatal(1, "can't ftruncate %s", path); 851 (void) close(fd); 852 } 853 854 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 855 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 856 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0); 857 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 858 859 return (file); 860} 861 862static nvlist_t * 863make_vdev_raidz(char *path, char *aux, char *pool, size_t size, 864 uint64_t ashift, int r) 865{ 866 nvlist_t *raidz, **child; 867 int c; 868 869 if (r < 2) 870 return (make_vdev_file(path, aux, pool, size, ashift)); 871 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); 872 873 for (c = 0; c < r; c++) 874 child[c] = make_vdev_file(path, aux, pool, size, ashift); 875 876 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); 877 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, 878 VDEV_TYPE_RAIDZ) == 0); 879 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, 880 ztest_opts.zo_raidz_parity) == 0); 881 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, 882 child, r) == 0); 883 884 for (c = 0; c < r; c++) 885 nvlist_free(child[c]); 886 887 umem_free(child, r * sizeof (nvlist_t *)); 888 889 return (raidz); 890} 891 892static nvlist_t * 893make_vdev_mirror(char *path, char *aux, char *pool, size_t size, 894 uint64_t ashift, int r, int m) 895{ 896 nvlist_t *mirror, **child; 897 int c; 898 899 if (m < 1) 900 return (make_vdev_raidz(path, aux, pool, size, ashift, r)); 901 902 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); 903 904 for (c = 0; c < m; c++) 905 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r); 906 907 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); 908 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, 909 VDEV_TYPE_MIRROR) == 0); 910 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, 911 child, m) == 0); 912 913 for (c = 0; c < m; c++) 914 nvlist_free(child[c]); 915 916 umem_free(child, m * sizeof (nvlist_t *)); 917 918 return (mirror); 919} 920 921static nvlist_t * 922make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift, 923 int log, int r, int m, int t) 924{ 925 nvlist_t *root, **child; 926 int c; 927 928 ASSERT(t > 0); 929 930 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); 931 932 for (c = 0; c < t; c++) { 933 child[c] = make_vdev_mirror(path, aux, pool, size, ashift, 934 r, m); 935 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, 936 log) == 0); 937 } 938 939 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 940 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 941 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN, 942 child, t) == 0); 943 944 for (c = 0; c < t; c++) 945 nvlist_free(child[c]); 946 947 umem_free(child, t * sizeof (nvlist_t *)); 948 949 return (root); 950} 951 952/* 953 * Find a random spa version. Returns back a random spa version in the 954 * range [initial_version, SPA_VERSION_FEATURES]. 955 */ 956static uint64_t 957ztest_random_spa_version(uint64_t initial_version) 958{ 959 uint64_t version = initial_version; 960 961 if (version <= SPA_VERSION_BEFORE_FEATURES) { 962 version = version + 963 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1); 964 } 965 966 if (version > SPA_VERSION_BEFORE_FEATURES) 967 version = SPA_VERSION_FEATURES; 968 969 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 970 return (version); 971} 972 973static int 974ztest_random_blocksize(void) 975{ 976 uint64_t block_shift; 977 /* 978 * Choose a block size >= the ashift. 979 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks. 980 */ 981 int maxbs = SPA_OLD_MAXBLOCKSHIFT; 982 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE) 983 maxbs = 20; 984 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1); 985 return (1 << (SPA_MINBLOCKSHIFT + block_shift)); 986} 987 988static int 989ztest_random_ibshift(void) 990{ 991 return (DN_MIN_INDBLKSHIFT + 992 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1)); 993} 994 995static uint64_t 996ztest_random_vdev_top(spa_t *spa, boolean_t log_ok) 997{ 998 uint64_t top; 999 vdev_t *rvd = spa->spa_root_vdev; 1000 vdev_t *tvd; 1001 1002 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1003 1004 do { 1005 top = ztest_random(rvd->vdev_children); 1006 tvd = rvd->vdev_child[top]; 1007 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) || 1008 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL); 1009 1010 return (top); 1011} 1012 1013static uint64_t 1014ztest_random_dsl_prop(zfs_prop_t prop) 1015{ 1016 uint64_t value; 1017 1018 do { 1019 value = zfs_prop_random_value(prop, ztest_random(-1ULL)); 1020 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF); 1021 1022 return (value); 1023} 1024 1025static int 1026ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, 1027 boolean_t inherit) 1028{ 1029 const char *propname = zfs_prop_to_name(prop); 1030 const char *valname; 1031 char setpoint[MAXPATHLEN]; 1032 uint64_t curval; 1033 int error; 1034 1035 error = dsl_prop_set_int(osname, propname, 1036 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value); 1037 1038 if (error == ENOSPC) { 1039 ztest_record_enospc(FTAG); 1040 return (error); 1041 } 1042 ASSERT0(error); 1043 1044 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint)); 1045 1046 if (ztest_opts.zo_verbose >= 6) { 1047 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0); 1048 (void) printf("%s %s = %s at '%s'\n", 1049 osname, propname, valname, setpoint); 1050 } 1051 1052 return (error); 1053} 1054 1055static int 1056ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value) 1057{ 1058 spa_t *spa = ztest_spa; 1059 nvlist_t *props = NULL; 1060 int error; 1061 1062 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); 1063 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0); 1064 1065 error = spa_prop_set(spa, props); 1066 1067 nvlist_free(props); 1068 1069 if (error == ENOSPC) { 1070 ztest_record_enospc(FTAG); 1071 return (error); 1072 } 1073 ASSERT0(error); 1074 1075 return (error); 1076} 1077 1078static void 1079ztest_rll_init(rll_t *rll) 1080{ 1081 rll->rll_writer = NULL; 1082 rll->rll_readers = 0; 1083 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0); 1084 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0); 1085} 1086 1087static void 1088ztest_rll_destroy(rll_t *rll) 1089{ 1090 ASSERT(rll->rll_writer == NULL); 1091 ASSERT(rll->rll_readers == 0); 1092 VERIFY(_mutex_destroy(&rll->rll_lock) == 0); 1093 VERIFY(cond_destroy(&rll->rll_cv) == 0); 1094} 1095 1096static void 1097ztest_rll_lock(rll_t *rll, rl_type_t type) 1098{ 1099 VERIFY(mutex_lock(&rll->rll_lock) == 0); 1100 1101 if (type == RL_READER) { 1102 while (rll->rll_writer != NULL) 1103 (void) cond_wait(&rll->rll_cv, &rll->rll_lock); 1104 rll->rll_readers++; 1105 } else { 1106 while (rll->rll_writer != NULL || rll->rll_readers) 1107 (void) cond_wait(&rll->rll_cv, &rll->rll_lock); 1108 rll->rll_writer = curthread; 1109 } 1110 1111 VERIFY(mutex_unlock(&rll->rll_lock) == 0); 1112} 1113 1114static void 1115ztest_rll_unlock(rll_t *rll) 1116{ 1117 VERIFY(mutex_lock(&rll->rll_lock) == 0); 1118 1119 if (rll->rll_writer) { 1120 ASSERT(rll->rll_readers == 0); 1121 rll->rll_writer = NULL; 1122 } else { 1123 ASSERT(rll->rll_readers != 0); 1124 ASSERT(rll->rll_writer == NULL); 1125 rll->rll_readers--; 1126 } 1127 1128 if (rll->rll_writer == NULL && rll->rll_readers == 0) 1129 VERIFY(cond_broadcast(&rll->rll_cv) == 0); 1130 1131 VERIFY(mutex_unlock(&rll->rll_lock) == 0); 1132} 1133 1134static void 1135ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type) 1136{ 1137 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; 1138 1139 ztest_rll_lock(rll, type); 1140} 1141 1142static void 1143ztest_object_unlock(ztest_ds_t *zd, uint64_t object) 1144{ 1145 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; 1146 1147 ztest_rll_unlock(rll); 1148} 1149 1150static rl_t * 1151ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset, 1152 uint64_t size, rl_type_t type) 1153{ 1154 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1)); 1155 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)]; 1156 rl_t *rl; 1157 1158 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL); 1159 rl->rl_object = object; 1160 rl->rl_offset = offset; 1161 rl->rl_size = size; 1162 rl->rl_lock = rll; 1163 1164 ztest_rll_lock(rll, type); 1165 1166 return (rl); 1167} 1168 1169static void 1170ztest_range_unlock(rl_t *rl) 1171{ 1172 rll_t *rll = rl->rl_lock; 1173 1174 ztest_rll_unlock(rll); 1175 1176 umem_free(rl, sizeof (*rl)); 1177} 1178 1179static void 1180ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os) 1181{ 1182 zd->zd_os = os; 1183 zd->zd_zilog = dmu_objset_zil(os); 1184 zd->zd_shared = szd; 1185 dmu_objset_name(os, zd->zd_name); 1186 1187 if (zd->zd_shared != NULL) 1188 zd->zd_shared->zd_seq = 0; 1189 1190 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0); 1191 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0); 1192 1193 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++) 1194 ztest_rll_init(&zd->zd_object_lock[l]); 1195 1196 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++) 1197 ztest_rll_init(&zd->zd_range_lock[l]); 1198} 1199 1200static void 1201ztest_zd_fini(ztest_ds_t *zd) 1202{ 1203 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0); 1204 1205 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++) 1206 ztest_rll_destroy(&zd->zd_object_lock[l]); 1207 1208 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++) 1209 ztest_rll_destroy(&zd->zd_range_lock[l]); 1210} 1211 1212#define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT) 1213 1214static uint64_t 1215ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag) 1216{ 1217 uint64_t txg; 1218 int error; 1219 1220 /* 1221 * Attempt to assign tx to some transaction group. 1222 */ 1223 error = dmu_tx_assign(tx, txg_how); 1224 if (error) { 1225 if (error == ERESTART) { 1226 ASSERT(txg_how == TXG_NOWAIT); 1227 dmu_tx_wait(tx); 1228 } else { 1229 ASSERT3U(error, ==, ENOSPC); 1230 ztest_record_enospc(tag); 1231 } 1232 dmu_tx_abort(tx); 1233 return (0); 1234 } 1235 txg = dmu_tx_get_txg(tx); 1236 ASSERT(txg != 0); 1237 return (txg); 1238} 1239 1240static void 1241ztest_pattern_set(void *buf, uint64_t size, uint64_t value) 1242{ 1243 uint64_t *ip = buf; 1244 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); 1245 1246 while (ip < ip_end) 1247 *ip++ = value; 1248} 1249 1250static boolean_t 1251ztest_pattern_match(void *buf, uint64_t size, uint64_t value) 1252{ 1253 uint64_t *ip = buf; 1254 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); 1255 uint64_t diff = 0; 1256 1257 while (ip < ip_end) 1258 diff |= (value - *ip++); 1259 1260 return (diff == 0); 1261} 1262 1263static void 1264ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object, 1265 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) 1266{ 1267 bt->bt_magic = BT_MAGIC; 1268 bt->bt_objset = dmu_objset_id(os); 1269 bt->bt_object = object; 1270 bt->bt_offset = offset; 1271 bt->bt_gen = gen; 1272 bt->bt_txg = txg; 1273 bt->bt_crtxg = crtxg; 1274} 1275 1276static void 1277ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object, 1278 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) 1279{ 1280 ASSERT3U(bt->bt_magic, ==, BT_MAGIC); 1281 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os)); 1282 ASSERT3U(bt->bt_object, ==, object); 1283 ASSERT3U(bt->bt_offset, ==, offset); 1284 ASSERT3U(bt->bt_gen, <=, gen); 1285 ASSERT3U(bt->bt_txg, <=, txg); 1286 ASSERT3U(bt->bt_crtxg, ==, crtxg); 1287} 1288 1289static ztest_block_tag_t * 1290ztest_bt_bonus(dmu_buf_t *db) 1291{ 1292 dmu_object_info_t doi; 1293 ztest_block_tag_t *bt; 1294 1295 dmu_object_info_from_db(db, &doi); 1296 ASSERT3U(doi.doi_bonus_size, <=, db->db_size); 1297 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt)); 1298 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt)); 1299 1300 return (bt); 1301} 1302 1303/* 1304 * ZIL logging ops 1305 */ 1306 1307#define lrz_type lr_mode 1308#define lrz_blocksize lr_uid 1309#define lrz_ibshift lr_gid 1310#define lrz_bonustype lr_rdev 1311#define lrz_bonuslen lr_crtime[1] 1312 1313static void 1314ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr) 1315{ 1316 char *name = (void *)(lr + 1); /* name follows lr */ 1317 size_t namesize = strlen(name) + 1; 1318 itx_t *itx; 1319 1320 if (zil_replaying(zd->zd_zilog, tx)) 1321 return; 1322 1323 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize); 1324 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1325 sizeof (*lr) + namesize - sizeof (lr_t)); 1326 1327 zil_itx_assign(zd->zd_zilog, itx, tx); 1328} 1329 1330static void 1331ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object) 1332{ 1333 char *name = (void *)(lr + 1); /* name follows lr */ 1334 size_t namesize = strlen(name) + 1; 1335 itx_t *itx; 1336 1337 if (zil_replaying(zd->zd_zilog, tx)) 1338 return; 1339 1340 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize); 1341 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1342 sizeof (*lr) + namesize - sizeof (lr_t)); 1343 1344 itx->itx_oid = object; 1345 zil_itx_assign(zd->zd_zilog, itx, tx); 1346} 1347 1348static void 1349ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr) 1350{ 1351 itx_t *itx; 1352 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES); 1353 1354 if (zil_replaying(zd->zd_zilog, tx)) 1355 return; 1356 1357 if (lr->lr_length > ZIL_MAX_LOG_DATA) 1358 write_state = WR_INDIRECT; 1359 1360 itx = zil_itx_create(TX_WRITE, 1361 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0)); 1362 1363 if (write_state == WR_COPIED && 1364 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length, 1365 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) { 1366 zil_itx_destroy(itx); 1367 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1368 write_state = WR_NEED_COPY; 1369 } 1370 itx->itx_private = zd; 1371 itx->itx_wr_state = write_state; 1372 itx->itx_sync = (ztest_random(8) == 0); 1373 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0); 1374 1375 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1376 sizeof (*lr) - sizeof (lr_t)); 1377 1378 zil_itx_assign(zd->zd_zilog, itx, tx); 1379} 1380 1381static void 1382ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr) 1383{ 1384 itx_t *itx; 1385 1386 if (zil_replaying(zd->zd_zilog, tx)) 1387 return; 1388 1389 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 1390 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1391 sizeof (*lr) - sizeof (lr_t)); 1392 1393 itx->itx_sync = B_FALSE; 1394 zil_itx_assign(zd->zd_zilog, itx, tx); 1395} 1396 1397static void 1398ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr) 1399{ 1400 itx_t *itx; 1401 1402 if (zil_replaying(zd->zd_zilog, tx)) 1403 return; 1404 1405 itx = zil_itx_create(TX_SETATTR, sizeof (*lr)); 1406 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1407 sizeof (*lr) - sizeof (lr_t)); 1408 1409 itx->itx_sync = B_FALSE; 1410 zil_itx_assign(zd->zd_zilog, itx, tx); 1411} 1412 1413/* 1414 * ZIL replay ops 1415 */ 1416static int 1417ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap) 1418{ 1419 char *name = (void *)(lr + 1); /* name follows lr */ 1420 objset_t *os = zd->zd_os; 1421 ztest_block_tag_t *bbt; 1422 dmu_buf_t *db; 1423 dmu_tx_t *tx; 1424 uint64_t txg; 1425 int error = 0; 1426 1427 if (byteswap) 1428 byteswap_uint64_array(lr, sizeof (*lr)); 1429 1430 ASSERT(lr->lr_doid == ZTEST_DIROBJ); 1431 ASSERT(name[0] != '\0'); 1432 1433 tx = dmu_tx_create(os); 1434 1435 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name); 1436 1437 if (lr->lrz_type == DMU_OT_ZAP_OTHER) { 1438 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); 1439 } else { 1440 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1441 } 1442 1443 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1444 if (txg == 0) 1445 return (ENOSPC); 1446 1447 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid); 1448 1449 if (lr->lrz_type == DMU_OT_ZAP_OTHER) { 1450 if (lr->lr_foid == 0) { 1451 lr->lr_foid = zap_create(os, 1452 lr->lrz_type, lr->lrz_bonustype, 1453 lr->lrz_bonuslen, tx); 1454 } else { 1455 error = zap_create_claim(os, lr->lr_foid, 1456 lr->lrz_type, lr->lrz_bonustype, 1457 lr->lrz_bonuslen, tx); 1458 } 1459 } else { 1460 if (lr->lr_foid == 0) { 1461 lr->lr_foid = dmu_object_alloc(os, 1462 lr->lrz_type, 0, lr->lrz_bonustype, 1463 lr->lrz_bonuslen, tx); 1464 } else { 1465 error = dmu_object_claim(os, lr->lr_foid, 1466 lr->lrz_type, 0, lr->lrz_bonustype, 1467 lr->lrz_bonuslen, tx); 1468 } 1469 } 1470 1471 if (error) { 1472 ASSERT3U(error, ==, EEXIST); 1473 ASSERT(zd->zd_zilog->zl_replay); 1474 dmu_tx_commit(tx); 1475 return (error); 1476 } 1477 1478 ASSERT(lr->lr_foid != 0); 1479 1480 if (lr->lrz_type != DMU_OT_ZAP_OTHER) 1481 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid, 1482 lr->lrz_blocksize, lr->lrz_ibshift, tx)); 1483 1484 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1485 bbt = ztest_bt_bonus(db); 1486 dmu_buf_will_dirty(db, tx); 1487 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg); 1488 dmu_buf_rele(db, FTAG); 1489 1490 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1, 1491 &lr->lr_foid, tx)); 1492 1493 (void) ztest_log_create(zd, tx, lr); 1494 1495 dmu_tx_commit(tx); 1496 1497 return (0); 1498} 1499 1500static int 1501ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap) 1502{ 1503 char *name = (void *)(lr + 1); /* name follows lr */ 1504 objset_t *os = zd->zd_os; 1505 dmu_object_info_t doi; 1506 dmu_tx_t *tx; 1507 uint64_t object, txg; 1508 1509 if (byteswap) 1510 byteswap_uint64_array(lr, sizeof (*lr)); 1511 1512 ASSERT(lr->lr_doid == ZTEST_DIROBJ); 1513 ASSERT(name[0] != '\0'); 1514 1515 VERIFY3U(0, ==, 1516 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object)); 1517 ASSERT(object != 0); 1518 1519 ztest_object_lock(zd, object, RL_WRITER); 1520 1521 VERIFY3U(0, ==, dmu_object_info(os, object, &doi)); 1522 1523 tx = dmu_tx_create(os); 1524 1525 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name); 1526 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 1527 1528 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1529 if (txg == 0) { 1530 ztest_object_unlock(zd, object); 1531 return (ENOSPC); 1532 } 1533 1534 if (doi.doi_type == DMU_OT_ZAP_OTHER) { 1535 VERIFY3U(0, ==, zap_destroy(os, object, tx)); 1536 } else { 1537 VERIFY3U(0, ==, dmu_object_free(os, object, tx)); 1538 } 1539 1540 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx)); 1541 1542 (void) ztest_log_remove(zd, tx, lr, object); 1543 1544 dmu_tx_commit(tx); 1545 1546 ztest_object_unlock(zd, object); 1547 1548 return (0); 1549} 1550 1551static int 1552ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap) 1553{ 1554 objset_t *os = zd->zd_os; 1555 void *data = lr + 1; /* data follows lr */ 1556 uint64_t offset, length; 1557 ztest_block_tag_t *bt = data; 1558 ztest_block_tag_t *bbt; 1559 uint64_t gen, txg, lrtxg, crtxg; 1560 dmu_object_info_t doi; 1561 dmu_tx_t *tx; 1562 dmu_buf_t *db; 1563 arc_buf_t *abuf = NULL; 1564 rl_t *rl; 1565 1566 if (byteswap) 1567 byteswap_uint64_array(lr, sizeof (*lr)); 1568 1569 offset = lr->lr_offset; 1570 length = lr->lr_length; 1571 1572 /* If it's a dmu_sync() block, write the whole block */ 1573 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 1574 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 1575 if (length < blocksize) { 1576 offset -= offset % blocksize; 1577 length = blocksize; 1578 } 1579 } 1580 1581 if (bt->bt_magic == BSWAP_64(BT_MAGIC)) 1582 byteswap_uint64_array(bt, sizeof (*bt)); 1583 1584 if (bt->bt_magic != BT_MAGIC) 1585 bt = NULL; 1586 1587 ztest_object_lock(zd, lr->lr_foid, RL_READER); 1588 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER); 1589 1590 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1591 1592 dmu_object_info_from_db(db, &doi); 1593 1594 bbt = ztest_bt_bonus(db); 1595 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1596 gen = bbt->bt_gen; 1597 crtxg = bbt->bt_crtxg; 1598 lrtxg = lr->lr_common.lrc_txg; 1599 1600 tx = dmu_tx_create(os); 1601 1602 dmu_tx_hold_write(tx, lr->lr_foid, offset, length); 1603 1604 if (ztest_random(8) == 0 && length == doi.doi_data_block_size && 1605 P2PHASE(offset, length) == 0) 1606 abuf = dmu_request_arcbuf(db, length); 1607 1608 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1609 if (txg == 0) { 1610 if (abuf != NULL) 1611 dmu_return_arcbuf(abuf); 1612 dmu_buf_rele(db, FTAG); 1613 ztest_range_unlock(rl); 1614 ztest_object_unlock(zd, lr->lr_foid); 1615 return (ENOSPC); 1616 } 1617 1618 if (bt != NULL) { 1619 /* 1620 * Usually, verify the old data before writing new data -- 1621 * but not always, because we also want to verify correct 1622 * behavior when the data was not recently read into cache. 1623 */ 1624 ASSERT(offset % doi.doi_data_block_size == 0); 1625 if (ztest_random(4) != 0) { 1626 int prefetch = ztest_random(2) ? 1627 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH; 1628 ztest_block_tag_t rbt; 1629 1630 VERIFY(dmu_read(os, lr->lr_foid, offset, 1631 sizeof (rbt), &rbt, prefetch) == 0); 1632 if (rbt.bt_magic == BT_MAGIC) { 1633 ztest_bt_verify(&rbt, os, lr->lr_foid, 1634 offset, gen, txg, crtxg); 1635 } 1636 } 1637 1638 /* 1639 * Writes can appear to be newer than the bonus buffer because 1640 * the ztest_get_data() callback does a dmu_read() of the 1641 * open-context data, which may be different than the data 1642 * as it was when the write was generated. 1643 */ 1644 if (zd->zd_zilog->zl_replay) { 1645 ztest_bt_verify(bt, os, lr->lr_foid, offset, 1646 MAX(gen, bt->bt_gen), MAX(txg, lrtxg), 1647 bt->bt_crtxg); 1648 } 1649 1650 /* 1651 * Set the bt's gen/txg to the bonus buffer's gen/txg 1652 * so that all of the usual ASSERTs will work. 1653 */ 1654 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg); 1655 } 1656 1657 if (abuf == NULL) { 1658 dmu_write(os, lr->lr_foid, offset, length, data, tx); 1659 } else { 1660 bcopy(data, abuf->b_data, length); 1661 dmu_assign_arcbuf(db, offset, abuf, tx); 1662 } 1663 1664 (void) ztest_log_write(zd, tx, lr); 1665 1666 dmu_buf_rele(db, FTAG); 1667 1668 dmu_tx_commit(tx); 1669 1670 ztest_range_unlock(rl); 1671 ztest_object_unlock(zd, lr->lr_foid); 1672 1673 return (0); 1674} 1675 1676static int 1677ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap) 1678{ 1679 objset_t *os = zd->zd_os; 1680 dmu_tx_t *tx; 1681 uint64_t txg; 1682 rl_t *rl; 1683 1684 if (byteswap) 1685 byteswap_uint64_array(lr, sizeof (*lr)); 1686 1687 ztest_object_lock(zd, lr->lr_foid, RL_READER); 1688 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length, 1689 RL_WRITER); 1690 1691 tx = dmu_tx_create(os); 1692 1693 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length); 1694 1695 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1696 if (txg == 0) { 1697 ztest_range_unlock(rl); 1698 ztest_object_unlock(zd, lr->lr_foid); 1699 return (ENOSPC); 1700 } 1701 1702 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset, 1703 lr->lr_length, tx) == 0); 1704 1705 (void) ztest_log_truncate(zd, tx, lr); 1706 1707 dmu_tx_commit(tx); 1708 1709 ztest_range_unlock(rl); 1710 ztest_object_unlock(zd, lr->lr_foid); 1711 1712 return (0); 1713} 1714 1715static int 1716ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap) 1717{ 1718 objset_t *os = zd->zd_os; 1719 dmu_tx_t *tx; 1720 dmu_buf_t *db; 1721 ztest_block_tag_t *bbt; 1722 uint64_t txg, lrtxg, crtxg; 1723 1724 if (byteswap) 1725 byteswap_uint64_array(lr, sizeof (*lr)); 1726 1727 ztest_object_lock(zd, lr->lr_foid, RL_WRITER); 1728 1729 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1730 1731 tx = dmu_tx_create(os); 1732 dmu_tx_hold_bonus(tx, lr->lr_foid); 1733 1734 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1735 if (txg == 0) { 1736 dmu_buf_rele(db, FTAG); 1737 ztest_object_unlock(zd, lr->lr_foid); 1738 return (ENOSPC); 1739 } 1740 1741 bbt = ztest_bt_bonus(db); 1742 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1743 crtxg = bbt->bt_crtxg; 1744 lrtxg = lr->lr_common.lrc_txg; 1745 1746 if (zd->zd_zilog->zl_replay) { 1747 ASSERT(lr->lr_size != 0); 1748 ASSERT(lr->lr_mode != 0); 1749 ASSERT(lrtxg != 0); 1750 } else { 1751 /* 1752 * Randomly change the size and increment the generation. 1753 */ 1754 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) * 1755 sizeof (*bbt); 1756 lr->lr_mode = bbt->bt_gen + 1; 1757 ASSERT(lrtxg == 0); 1758 } 1759 1760 /* 1761 * Verify that the current bonus buffer is not newer than our txg. 1762 */ 1763 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, 1764 MAX(txg, lrtxg), crtxg); 1765 1766 dmu_buf_will_dirty(db, tx); 1767 1768 ASSERT3U(lr->lr_size, >=, sizeof (*bbt)); 1769 ASSERT3U(lr->lr_size, <=, db->db_size); 1770 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx)); 1771 bbt = ztest_bt_bonus(db); 1772 1773 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg); 1774 1775 dmu_buf_rele(db, FTAG); 1776 1777 (void) ztest_log_setattr(zd, tx, lr); 1778 1779 dmu_tx_commit(tx); 1780 1781 ztest_object_unlock(zd, lr->lr_foid); 1782 1783 return (0); 1784} 1785 1786zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { 1787 NULL, /* 0 no such transaction type */ 1788 ztest_replay_create, /* TX_CREATE */ 1789 NULL, /* TX_MKDIR */ 1790 NULL, /* TX_MKXATTR */ 1791 NULL, /* TX_SYMLINK */ 1792 ztest_replay_remove, /* TX_REMOVE */ 1793 NULL, /* TX_RMDIR */ 1794 NULL, /* TX_LINK */ 1795 NULL, /* TX_RENAME */ 1796 ztest_replay_write, /* TX_WRITE */ 1797 ztest_replay_truncate, /* TX_TRUNCATE */ 1798 ztest_replay_setattr, /* TX_SETATTR */ 1799 NULL, /* TX_ACL */ 1800 NULL, /* TX_CREATE_ACL */ 1801 NULL, /* TX_CREATE_ATTR */ 1802 NULL, /* TX_CREATE_ACL_ATTR */ 1803 NULL, /* TX_MKDIR_ACL */ 1804 NULL, /* TX_MKDIR_ATTR */ 1805 NULL, /* TX_MKDIR_ACL_ATTR */ 1806 NULL, /* TX_WRITE2 */ 1807}; 1808 1809/* 1810 * ZIL get_data callbacks 1811 */ 1812 1813static void 1814ztest_get_done(zgd_t *zgd, int error) 1815{ 1816 ztest_ds_t *zd = zgd->zgd_private; 1817 uint64_t object = zgd->zgd_rl->rl_object; 1818 1819 if (zgd->zgd_db) 1820 dmu_buf_rele(zgd->zgd_db, zgd); 1821 1822 ztest_range_unlock(zgd->zgd_rl); 1823 ztest_object_unlock(zd, object); 1824 1825 if (error == 0 && zgd->zgd_bp) 1826 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1827 1828 umem_free(zgd, sizeof (*zgd)); 1829} 1830 1831static int 1832ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1833{ 1834 ztest_ds_t *zd = arg; 1835 objset_t *os = zd->zd_os; 1836 uint64_t object = lr->lr_foid; 1837 uint64_t offset = lr->lr_offset; 1838 uint64_t size = lr->lr_length; 1839 blkptr_t *bp = &lr->lr_blkptr; 1840 uint64_t txg = lr->lr_common.lrc_txg; 1841 uint64_t crtxg; 1842 dmu_object_info_t doi; 1843 dmu_buf_t *db; 1844 zgd_t *zgd; 1845 int error; 1846 1847 ztest_object_lock(zd, object, RL_READER); 1848 error = dmu_bonus_hold(os, object, FTAG, &db); 1849 if (error) { 1850 ztest_object_unlock(zd, object); 1851 return (error); 1852 } 1853 1854 crtxg = ztest_bt_bonus(db)->bt_crtxg; 1855 1856 if (crtxg == 0 || crtxg > txg) { 1857 dmu_buf_rele(db, FTAG); 1858 ztest_object_unlock(zd, object); 1859 return (ENOENT); 1860 } 1861 1862 dmu_object_info_from_db(db, &doi); 1863 dmu_buf_rele(db, FTAG); 1864 db = NULL; 1865 1866 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL); 1867 zgd->zgd_zilog = zd->zd_zilog; 1868 zgd->zgd_private = zd; 1869 1870 if (buf != NULL) { /* immediate write */ 1871 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, 1872 RL_READER); 1873 1874 error = dmu_read(os, object, offset, size, buf, 1875 DMU_READ_NO_PREFETCH); 1876 ASSERT(error == 0); 1877 } else { 1878 size = doi.doi_data_block_size; 1879 if (ISP2(size)) { 1880 offset = P2ALIGN(offset, size); 1881 } else { 1882 ASSERT(offset < size); 1883 offset = 0; 1884 } 1885 1886 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, 1887 RL_READER); 1888 1889 error = dmu_buf_hold(os, object, offset, zgd, &db, 1890 DMU_READ_NO_PREFETCH); 1891 1892 if (error == 0) { 1893 blkptr_t *obp = dmu_buf_get_blkptr(db); 1894 if (obp) { 1895 ASSERT(BP_IS_HOLE(bp)); 1896 *bp = *obp; 1897 } 1898 1899 zgd->zgd_db = db; 1900 zgd->zgd_bp = bp; 1901 1902 ASSERT(db->db_offset == offset); 1903 ASSERT(db->db_size == size); 1904 1905 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1906 ztest_get_done, zgd); 1907 1908 if (error == 0) 1909 return (0); 1910 } 1911 } 1912 1913 ztest_get_done(zgd, error); 1914 1915 return (error); 1916} 1917 1918static void * 1919ztest_lr_alloc(size_t lrsize, char *name) 1920{ 1921 char *lr; 1922 size_t namesize = name ? strlen(name) + 1 : 0; 1923 1924 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL); 1925 1926 if (name) 1927 bcopy(name, lr + lrsize, namesize); 1928 1929 return (lr); 1930} 1931 1932void 1933ztest_lr_free(void *lr, size_t lrsize, char *name) 1934{ 1935 size_t namesize = name ? strlen(name) + 1 : 0; 1936 1937 umem_free(lr, lrsize + namesize); 1938} 1939 1940/* 1941 * Lookup a bunch of objects. Returns the number of objects not found. 1942 */ 1943static int 1944ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count) 1945{ 1946 int missing = 0; 1947 int error; 1948 1949 ASSERT(_mutex_held(&zd->zd_dirobj_lock)); 1950 1951 for (int i = 0; i < count; i++, od++) { 1952 od->od_object = 0; 1953 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name, 1954 sizeof (uint64_t), 1, &od->od_object); 1955 if (error) { 1956 ASSERT(error == ENOENT); 1957 ASSERT(od->od_object == 0); 1958 missing++; 1959 } else { 1960 dmu_buf_t *db; 1961 ztest_block_tag_t *bbt; 1962 dmu_object_info_t doi; 1963 1964 ASSERT(od->od_object != 0); 1965 ASSERT(missing == 0); /* there should be no gaps */ 1966 1967 ztest_object_lock(zd, od->od_object, RL_READER); 1968 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os, 1969 od->od_object, FTAG, &db)); 1970 dmu_object_info_from_db(db, &doi); 1971 bbt = ztest_bt_bonus(db); 1972 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1973 od->od_type = doi.doi_type; 1974 od->od_blocksize = doi.doi_data_block_size; 1975 od->od_gen = bbt->bt_gen; 1976 dmu_buf_rele(db, FTAG); 1977 ztest_object_unlock(zd, od->od_object); 1978 } 1979 } 1980 1981 return (missing); 1982} 1983 1984static int 1985ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count) 1986{ 1987 int missing = 0; 1988 1989 ASSERT(_mutex_held(&zd->zd_dirobj_lock)); 1990 1991 for (int i = 0; i < count; i++, od++) { 1992 if (missing) { 1993 od->od_object = 0; 1994 missing++; 1995 continue; 1996 } 1997 1998 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); 1999 2000 lr->lr_doid = od->od_dir; 2001 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */ 2002 lr->lrz_type = od->od_crtype; 2003 lr->lrz_blocksize = od->od_crblocksize; 2004 lr->lrz_ibshift = ztest_random_ibshift(); 2005 lr->lrz_bonustype = DMU_OT_UINT64_OTHER; 2006 lr->lrz_bonuslen = dmu_bonus_max(); 2007 lr->lr_gen = od->od_crgen; 2008 lr->lr_crtime[0] = time(NULL); 2009 2010 if (ztest_replay_create(zd, lr, B_FALSE) != 0) { 2011 ASSERT(missing == 0); 2012 od->od_object = 0; 2013 missing++; 2014 } else { 2015 od->od_object = lr->lr_foid; 2016 od->od_type = od->od_crtype; 2017 od->od_blocksize = od->od_crblocksize; 2018 od->od_gen = od->od_crgen; 2019 ASSERT(od->od_object != 0); 2020 } 2021 2022 ztest_lr_free(lr, sizeof (*lr), od->od_name); 2023 } 2024 2025 return (missing); 2026} 2027 2028static int 2029ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count) 2030{ 2031 int missing = 0; 2032 int error; 2033 2034 ASSERT(_mutex_held(&zd->zd_dirobj_lock)); 2035 2036 od += count - 1; 2037 2038 for (int i = count - 1; i >= 0; i--, od--) { 2039 if (missing) { 2040 missing++; 2041 continue; 2042 } 2043 2044 /* 2045 * No object was found. 2046 */ 2047 if (od->od_object == 0) 2048 continue; 2049 2050 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); 2051 2052 lr->lr_doid = od->od_dir; 2053 2054 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) { 2055 ASSERT3U(error, ==, ENOSPC); 2056 missing++; 2057 } else { 2058 od->od_object = 0; 2059 } 2060 ztest_lr_free(lr, sizeof (*lr), od->od_name); 2061 } 2062 2063 return (missing); 2064} 2065 2066static int 2067ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size, 2068 void *data) 2069{ 2070 lr_write_t *lr; 2071 int error; 2072 2073 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL); 2074 2075 lr->lr_foid = object; 2076 lr->lr_offset = offset; 2077 lr->lr_length = size; 2078 lr->lr_blkoff = 0; 2079 BP_ZERO(&lr->lr_blkptr); 2080 2081 bcopy(data, lr + 1, size); 2082 2083 error = ztest_replay_write(zd, lr, B_FALSE); 2084 2085 ztest_lr_free(lr, sizeof (*lr) + size, NULL); 2086 2087 return (error); 2088} 2089 2090static int 2091ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) 2092{ 2093 lr_truncate_t *lr; 2094 int error; 2095 2096 lr = ztest_lr_alloc(sizeof (*lr), NULL); 2097 2098 lr->lr_foid = object; 2099 lr->lr_offset = offset; 2100 lr->lr_length = size; 2101 2102 error = ztest_replay_truncate(zd, lr, B_FALSE); 2103 2104 ztest_lr_free(lr, sizeof (*lr), NULL); 2105 2106 return (error); 2107} 2108 2109static int 2110ztest_setattr(ztest_ds_t *zd, uint64_t object) 2111{ 2112 lr_setattr_t *lr; 2113 int error; 2114 2115 lr = ztest_lr_alloc(sizeof (*lr), NULL); 2116 2117 lr->lr_foid = object; 2118 lr->lr_size = 0; 2119 lr->lr_mode = 0; 2120 2121 error = ztest_replay_setattr(zd, lr, B_FALSE); 2122 2123 ztest_lr_free(lr, sizeof (*lr), NULL); 2124 2125 return (error); 2126} 2127 2128static void 2129ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) 2130{ 2131 objset_t *os = zd->zd_os; 2132 dmu_tx_t *tx; 2133 uint64_t txg; 2134 rl_t *rl; 2135 2136 txg_wait_synced(dmu_objset_pool(os), 0); 2137 2138 ztest_object_lock(zd, object, RL_READER); 2139 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER); 2140 2141 tx = dmu_tx_create(os); 2142 2143 dmu_tx_hold_write(tx, object, offset, size); 2144 2145 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 2146 2147 if (txg != 0) { 2148 dmu_prealloc(os, object, offset, size, tx); 2149 dmu_tx_commit(tx); 2150 txg_wait_synced(dmu_objset_pool(os), txg); 2151 } else { 2152 (void) dmu_free_long_range(os, object, offset, size); 2153 } 2154 2155 ztest_range_unlock(rl); 2156 ztest_object_unlock(zd, object); 2157} 2158 2159static void 2160ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset) 2161{ 2162 int err; 2163 ztest_block_tag_t wbt; 2164 dmu_object_info_t doi; 2165 enum ztest_io_type io_type; 2166 uint64_t blocksize; 2167 void *data; 2168 2169 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0); 2170 blocksize = doi.doi_data_block_size; 2171 data = umem_alloc(blocksize, UMEM_NOFAIL); 2172 2173 /* 2174 * Pick an i/o type at random, biased toward writing block tags. 2175 */ 2176 io_type = ztest_random(ZTEST_IO_TYPES); 2177 if (ztest_random(2) == 0) 2178 io_type = ZTEST_IO_WRITE_TAG; 2179 2180 (void) rw_rdlock(&zd->zd_zilog_lock); 2181 2182 switch (io_type) { 2183 2184 case ZTEST_IO_WRITE_TAG: 2185 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0); 2186 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt); 2187 break; 2188 2189 case ZTEST_IO_WRITE_PATTERN: 2190 (void) memset(data, 'a' + (object + offset) % 5, blocksize); 2191 if (ztest_random(2) == 0) { 2192 /* 2193 * Induce fletcher2 collisions to ensure that 2194 * zio_ddt_collision() detects and resolves them 2195 * when using fletcher2-verify for deduplication. 2196 */ 2197 ((uint64_t *)data)[0] ^= 1ULL << 63; 2198 ((uint64_t *)data)[4] ^= 1ULL << 63; 2199 } 2200 (void) ztest_write(zd, object, offset, blocksize, data); 2201 break; 2202 2203 case ZTEST_IO_WRITE_ZEROES: 2204 bzero(data, blocksize); 2205 (void) ztest_write(zd, object, offset, blocksize, data); 2206 break; 2207 2208 case ZTEST_IO_TRUNCATE: 2209 (void) ztest_truncate(zd, object, offset, blocksize); 2210 break; 2211 2212 case ZTEST_IO_SETATTR: 2213 (void) ztest_setattr(zd, object); 2214 break; 2215 2216 case ZTEST_IO_REWRITE: 2217 (void) rw_rdlock(&ztest_name_lock); 2218 err = ztest_dsl_prop_set_uint64(zd->zd_name, 2219 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa), 2220 B_FALSE); 2221 VERIFY(err == 0 || err == ENOSPC); 2222 err = ztest_dsl_prop_set_uint64(zd->zd_name, 2223 ZFS_PROP_COMPRESSION, 2224 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), 2225 B_FALSE); 2226 VERIFY(err == 0 || err == ENOSPC); 2227 (void) rw_unlock(&ztest_name_lock); 2228 2229 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data, 2230 DMU_READ_NO_PREFETCH)); 2231 2232 (void) ztest_write(zd, object, offset, blocksize, data); 2233 break; 2234 } 2235 2236 (void) rw_unlock(&zd->zd_zilog_lock); 2237 2238 umem_free(data, blocksize); 2239} 2240 2241/* 2242 * Initialize an object description template. 2243 */ 2244static void 2245ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index, 2246 dmu_object_type_t type, uint64_t blocksize, uint64_t gen) 2247{ 2248 od->od_dir = ZTEST_DIROBJ; 2249 od->od_object = 0; 2250 2251 od->od_crtype = type; 2252 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize(); 2253 od->od_crgen = gen; 2254 2255 od->od_type = DMU_OT_NONE; 2256 od->od_blocksize = 0; 2257 od->od_gen = 0; 2258 2259 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]", 2260 tag, (int64_t)id, index); 2261} 2262 2263/* 2264 * Lookup or create the objects for a test using the od template. 2265 * If the objects do not all exist, or if 'remove' is specified, 2266 * remove any existing objects and create new ones. Otherwise, 2267 * use the existing objects. 2268 */ 2269static int 2270ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove) 2271{ 2272 int count = size / sizeof (*od); 2273 int rv = 0; 2274 2275 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0); 2276 if ((ztest_lookup(zd, od, count) != 0 || remove) && 2277 (ztest_remove(zd, od, count) != 0 || 2278 ztest_create(zd, od, count) != 0)) 2279 rv = -1; 2280 zd->zd_od = od; 2281 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0); 2282 2283 return (rv); 2284} 2285 2286/* ARGSUSED */ 2287void 2288ztest_zil_commit(ztest_ds_t *zd, uint64_t id) 2289{ 2290 zilog_t *zilog = zd->zd_zilog; 2291 2292 (void) rw_rdlock(&zd->zd_zilog_lock); 2293 2294 zil_commit(zilog, ztest_random(ZTEST_OBJECTS)); 2295 2296 /* 2297 * Remember the committed values in zd, which is in parent/child 2298 * shared memory. If we die, the next iteration of ztest_run() 2299 * will verify that the log really does contain this record. 2300 */ 2301 mutex_enter(&zilog->zl_lock); 2302 ASSERT(zd->zd_shared != NULL); 2303 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq); 2304 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq; 2305 mutex_exit(&zilog->zl_lock); 2306 2307 (void) rw_unlock(&zd->zd_zilog_lock); 2308} 2309 2310/* 2311 * This function is designed to simulate the operations that occur during a 2312 * mount/unmount operation. We hold the dataset across these operations in an 2313 * attempt to expose any implicit assumptions about ZIL management. 2314 */ 2315/* ARGSUSED */ 2316void 2317ztest_zil_remount(ztest_ds_t *zd, uint64_t id) 2318{ 2319 objset_t *os = zd->zd_os; 2320 2321 /* 2322 * We grab the zd_dirobj_lock to ensure that no other thread is 2323 * updating the zil (i.e. adding in-memory log records) and the 2324 * zd_zilog_lock to block any I/O. 2325 */ 2326 VERIFY0(mutex_lock(&zd->zd_dirobj_lock)); 2327 (void) rw_wrlock(&zd->zd_zilog_lock); 2328 2329 /* zfsvfs_teardown() */ 2330 zil_close(zd->zd_zilog); 2331 2332 /* zfsvfs_setup() */ 2333 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog); 2334 zil_replay(os, zd, ztest_replay_vector); 2335 2336 (void) rw_unlock(&zd->zd_zilog_lock); 2337 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0); 2338} 2339 2340/* 2341 * Verify that we can't destroy an active pool, create an existing pool, 2342 * or create a pool with a bad vdev spec. 2343 */ 2344/* ARGSUSED */ 2345void 2346ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id) 2347{ 2348 ztest_shared_opts_t *zo = &ztest_opts; 2349 spa_t *spa; 2350 nvlist_t *nvroot; 2351 2352 /* 2353 * Attempt to create using a bad file. 2354 */ 2355 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); 2356 VERIFY3U(ENOENT, ==, 2357 spa_create("ztest_bad_file", nvroot, NULL, NULL)); 2358 nvlist_free(nvroot); 2359 2360 /* 2361 * Attempt to create using a bad mirror. 2362 */ 2363 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1); 2364 VERIFY3U(ENOENT, ==, 2365 spa_create("ztest_bad_mirror", nvroot, NULL, NULL)); 2366 nvlist_free(nvroot); 2367 2368 /* 2369 * Attempt to create an existing pool. It shouldn't matter 2370 * what's in the nvroot; we should fail with EEXIST. 2371 */ 2372 (void) rw_rdlock(&ztest_name_lock); 2373 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); 2374 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL)); 2375 nvlist_free(nvroot); 2376 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG)); 2377 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool)); 2378 spa_close(spa, FTAG); 2379 2380 (void) rw_unlock(&ztest_name_lock); 2381} 2382 2383/* ARGSUSED */ 2384void 2385ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id) 2386{ 2387 spa_t *spa; 2388 uint64_t initial_version = SPA_VERSION_INITIAL; 2389 uint64_t version, newversion; 2390 nvlist_t *nvroot, *props; 2391 char *name; 2392 2393 VERIFY0(mutex_lock(&ztest_vdev_lock)); 2394 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool); 2395 2396 /* 2397 * Clean up from previous runs. 2398 */ 2399 (void) spa_destroy(name); 2400 2401 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0, 2402 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1); 2403 2404 /* 2405 * If we're configuring a RAIDZ device then make sure that the 2406 * the initial version is capable of supporting that feature. 2407 */ 2408 switch (ztest_opts.zo_raidz_parity) { 2409 case 0: 2410 case 1: 2411 initial_version = SPA_VERSION_INITIAL; 2412 break; 2413 case 2: 2414 initial_version = SPA_VERSION_RAIDZ2; 2415 break; 2416 case 3: 2417 initial_version = SPA_VERSION_RAIDZ3; 2418 break; 2419 } 2420 2421 /* 2422 * Create a pool with a spa version that can be upgraded. Pick 2423 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES. 2424 */ 2425 do { 2426 version = ztest_random_spa_version(initial_version); 2427 } while (version > SPA_VERSION_BEFORE_FEATURES); 2428 2429 props = fnvlist_alloc(); 2430 fnvlist_add_uint64(props, 2431 zpool_prop_to_name(ZPOOL_PROP_VERSION), version); 2432 VERIFY0(spa_create(name, nvroot, props, NULL)); 2433 fnvlist_free(nvroot); 2434 fnvlist_free(props); 2435 2436 VERIFY0(spa_open(name, &spa, FTAG)); 2437 VERIFY3U(spa_version(spa), ==, version); 2438 newversion = ztest_random_spa_version(version + 1); 2439 2440 if (ztest_opts.zo_verbose >= 4) { 2441 (void) printf("upgrading spa version from %llu to %llu\n", 2442 (u_longlong_t)version, (u_longlong_t)newversion); 2443 } 2444 2445 spa_upgrade(spa, newversion); 2446 VERIFY3U(spa_version(spa), >, version); 2447 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config, 2448 zpool_prop_to_name(ZPOOL_PROP_VERSION))); 2449 spa_close(spa, FTAG); 2450 2451 strfree(name); 2452 VERIFY0(mutex_unlock(&ztest_vdev_lock)); 2453} 2454 2455static vdev_t * 2456vdev_lookup_by_path(vdev_t *vd, const char *path) 2457{ 2458 vdev_t *mvd; 2459 2460 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0) 2461 return (vd); 2462 2463 for (int c = 0; c < vd->vdev_children; c++) 2464 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != 2465 NULL) 2466 return (mvd); 2467 2468 return (NULL); 2469} 2470 2471/* 2472 * Find the first available hole which can be used as a top-level. 2473 */ 2474int 2475find_vdev_hole(spa_t *spa) 2476{ 2477 vdev_t *rvd = spa->spa_root_vdev; 2478 int c; 2479 2480 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV); 2481 2482 for (c = 0; c < rvd->vdev_children; c++) { 2483 vdev_t *cvd = rvd->vdev_child[c]; 2484 2485 if (cvd->vdev_ishole) 2486 break; 2487 } 2488 return (c); 2489} 2490 2491/* 2492 * Verify that vdev_add() works as expected. 2493 */ 2494/* ARGSUSED */ 2495void 2496ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id) 2497{ 2498 ztest_shared_t *zs = ztest_shared; 2499 spa_t *spa = ztest_spa; 2500 uint64_t leaves; 2501 uint64_t guid; 2502 nvlist_t *nvroot; 2503 int error; 2504 2505 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2506 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz; 2507 2508 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2509 2510 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves; 2511 2512 /* 2513 * If we have slogs then remove them 1/4 of the time. 2514 */ 2515 if (spa_has_slogs(spa) && ztest_random(4) == 0) { 2516 /* 2517 * Grab the guid from the head of the log class rotor. 2518 */ 2519 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid; 2520 2521 spa_config_exit(spa, SCL_VDEV, FTAG); 2522 2523 /* 2524 * We have to grab the zs_name_lock as writer to 2525 * prevent a race between removing a slog (dmu_objset_find) 2526 * and destroying a dataset. Removing the slog will 2527 * grab a reference on the dataset which may cause 2528 * dmu_objset_destroy() to fail with EBUSY thus 2529 * leaving the dataset in an inconsistent state. 2530 */ 2531 VERIFY(rw_wrlock(&ztest_name_lock) == 0); 2532 error = spa_vdev_remove(spa, guid, B_FALSE); 2533 VERIFY(rw_unlock(&ztest_name_lock) == 0); 2534 2535 if (error && error != EEXIST) 2536 fatal(0, "spa_vdev_remove() = %d", error); 2537 } else { 2538 spa_config_exit(spa, SCL_VDEV, FTAG); 2539 2540 /* 2541 * Make 1/4 of the devices be log devices. 2542 */ 2543 nvroot = make_vdev_root(NULL, NULL, NULL, 2544 ztest_opts.zo_vdev_size, 0, 2545 ztest_random(4) == 0, ztest_opts.zo_raidz, 2546 zs->zs_mirrors, 1); 2547 2548 error = spa_vdev_add(spa, nvroot); 2549 nvlist_free(nvroot); 2550 2551 if (error == ENOSPC) 2552 ztest_record_enospc("spa_vdev_add"); 2553 else if (error != 0) 2554 fatal(0, "spa_vdev_add() = %d", error); 2555 } 2556 2557 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2558} 2559 2560/* 2561 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected. 2562 */ 2563/* ARGSUSED */ 2564void 2565ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id) 2566{ 2567 ztest_shared_t *zs = ztest_shared; 2568 spa_t *spa = ztest_spa; 2569 vdev_t *rvd = spa->spa_root_vdev; 2570 spa_aux_vdev_t *sav; 2571 char *aux; 2572 uint64_t guid = 0; 2573 int error; 2574 2575 if (ztest_random(2) == 0) { 2576 sav = &spa->spa_spares; 2577 aux = ZPOOL_CONFIG_SPARES; 2578 } else { 2579 sav = &spa->spa_l2cache; 2580 aux = ZPOOL_CONFIG_L2CACHE; 2581 } 2582 2583 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2584 2585 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2586 2587 if (sav->sav_count != 0 && ztest_random(4) == 0) { 2588 /* 2589 * Pick a random device to remove. 2590 */ 2591 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid; 2592 } else { 2593 /* 2594 * Find an unused device we can add. 2595 */ 2596 zs->zs_vdev_aux = 0; 2597 for (;;) { 2598 char path[MAXPATHLEN]; 2599 int c; 2600 (void) snprintf(path, sizeof (path), ztest_aux_template, 2601 ztest_opts.zo_dir, ztest_opts.zo_pool, aux, 2602 zs->zs_vdev_aux); 2603 for (c = 0; c < sav->sav_count; c++) 2604 if (strcmp(sav->sav_vdevs[c]->vdev_path, 2605 path) == 0) 2606 break; 2607 if (c == sav->sav_count && 2608 vdev_lookup_by_path(rvd, path) == NULL) 2609 break; 2610 zs->zs_vdev_aux++; 2611 } 2612 } 2613 2614 spa_config_exit(spa, SCL_VDEV, FTAG); 2615 2616 if (guid == 0) { 2617 /* 2618 * Add a new device. 2619 */ 2620 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL, 2621 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1); 2622 error = spa_vdev_add(spa, nvroot); 2623 if (error != 0) 2624 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error); 2625 nvlist_free(nvroot); 2626 } else { 2627 /* 2628 * Remove an existing device. Sometimes, dirty its 2629 * vdev state first to make sure we handle removal 2630 * of devices that have pending state changes. 2631 */ 2632 if (ztest_random(2) == 0) 2633 (void) vdev_online(spa, guid, 0, NULL); 2634 2635 error = spa_vdev_remove(spa, guid, B_FALSE); 2636 if (error != 0 && error != EBUSY) 2637 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error); 2638 } 2639 2640 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2641} 2642 2643/* 2644 * split a pool if it has mirror tlvdevs 2645 */ 2646/* ARGSUSED */ 2647void 2648ztest_split_pool(ztest_ds_t *zd, uint64_t id) 2649{ 2650 ztest_shared_t *zs = ztest_shared; 2651 spa_t *spa = ztest_spa; 2652 vdev_t *rvd = spa->spa_root_vdev; 2653 nvlist_t *tree, **child, *config, *split, **schild; 2654 uint_t c, children, schildren = 0, lastlogid = 0; 2655 int error = 0; 2656 2657 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2658 2659 /* ensure we have a useable config; mirrors of raidz aren't supported */ 2660 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) { 2661 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2662 return; 2663 } 2664 2665 /* clean up the old pool, if any */ 2666 (void) spa_destroy("splitp"); 2667 2668 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2669 2670 /* generate a config from the existing config */ 2671 mutex_enter(&spa->spa_props_lock); 2672 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE, 2673 &tree) == 0); 2674 mutex_exit(&spa->spa_props_lock); 2675 2676 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child, 2677 &children) == 0); 2678 2679 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *)); 2680 for (c = 0; c < children; c++) { 2681 vdev_t *tvd = rvd->vdev_child[c]; 2682 nvlist_t **mchild; 2683 uint_t mchildren; 2684 2685 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) { 2686 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME, 2687 0) == 0); 2688 VERIFY(nvlist_add_string(schild[schildren], 2689 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); 2690 VERIFY(nvlist_add_uint64(schild[schildren], 2691 ZPOOL_CONFIG_IS_HOLE, 1) == 0); 2692 if (lastlogid == 0) 2693 lastlogid = schildren; 2694 ++schildren; 2695 continue; 2696 } 2697 lastlogid = 0; 2698 VERIFY(nvlist_lookup_nvlist_array(child[c], 2699 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0); 2700 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0); 2701 } 2702 2703 /* OK, create a config that can be used to split */ 2704 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0); 2705 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE, 2706 VDEV_TYPE_ROOT) == 0); 2707 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild, 2708 lastlogid != 0 ? lastlogid : schildren) == 0); 2709 2710 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0); 2711 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0); 2712 2713 for (c = 0; c < schildren; c++) 2714 nvlist_free(schild[c]); 2715 free(schild); 2716 nvlist_free(split); 2717 2718 spa_config_exit(spa, SCL_VDEV, FTAG); 2719 2720 (void) rw_wrlock(&ztest_name_lock); 2721 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE); 2722 (void) rw_unlock(&ztest_name_lock); 2723 2724 nvlist_free(config); 2725 2726 if (error == 0) { 2727 (void) printf("successful split - results:\n"); 2728 mutex_enter(&spa_namespace_lock); 2729 show_pool_stats(spa); 2730 show_pool_stats(spa_lookup("splitp")); 2731 mutex_exit(&spa_namespace_lock); 2732 ++zs->zs_splits; 2733 --zs->zs_mirrors; 2734 } 2735 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2736 2737} 2738 2739/* 2740 * Verify that we can attach and detach devices. 2741 */ 2742/* ARGSUSED */ 2743void 2744ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id) 2745{ 2746 ztest_shared_t *zs = ztest_shared; 2747 spa_t *spa = ztest_spa; 2748 spa_aux_vdev_t *sav = &spa->spa_spares; 2749 vdev_t *rvd = spa->spa_root_vdev; 2750 vdev_t *oldvd, *newvd, *pvd; 2751 nvlist_t *root; 2752 uint64_t leaves; 2753 uint64_t leaf, top; 2754 uint64_t ashift = ztest_get_ashift(); 2755 uint64_t oldguid, pguid; 2756 uint64_t oldsize, newsize; 2757 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; 2758 int replacing; 2759 int oldvd_has_siblings = B_FALSE; 2760 int newvd_is_spare = B_FALSE; 2761 int oldvd_is_log; 2762 int error, expected_error; 2763 2764 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2765 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; 2766 2767 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2768 2769 /* 2770 * Decide whether to do an attach or a replace. 2771 */ 2772 replacing = ztest_random(2); 2773 2774 /* 2775 * Pick a random top-level vdev. 2776 */ 2777 top = ztest_random_vdev_top(spa, B_TRUE); 2778 2779 /* 2780 * Pick a random leaf within it. 2781 */ 2782 leaf = ztest_random(leaves); 2783 2784 /* 2785 * Locate this vdev. 2786 */ 2787 oldvd = rvd->vdev_child[top]; 2788 if (zs->zs_mirrors >= 1) { 2789 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops); 2790 ASSERT(oldvd->vdev_children >= zs->zs_mirrors); 2791 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz]; 2792 } 2793 if (ztest_opts.zo_raidz > 1) { 2794 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops); 2795 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz); 2796 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz]; 2797 } 2798 2799 /* 2800 * If we're already doing an attach or replace, oldvd may be a 2801 * mirror vdev -- in which case, pick a random child. 2802 */ 2803 while (oldvd->vdev_children != 0) { 2804 oldvd_has_siblings = B_TRUE; 2805 ASSERT(oldvd->vdev_children >= 2); 2806 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)]; 2807 } 2808 2809 oldguid = oldvd->vdev_guid; 2810 oldsize = vdev_get_min_asize(oldvd); 2811 oldvd_is_log = oldvd->vdev_top->vdev_islog; 2812 (void) strcpy(oldpath, oldvd->vdev_path); 2813 pvd = oldvd->vdev_parent; 2814 pguid = pvd->vdev_guid; 2815 2816 /* 2817 * If oldvd has siblings, then half of the time, detach it. 2818 */ 2819 if (oldvd_has_siblings && ztest_random(2) == 0) { 2820 spa_config_exit(spa, SCL_VDEV, FTAG); 2821 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE); 2822 if (error != 0 && error != ENODEV && error != EBUSY && 2823 error != ENOTSUP) 2824 fatal(0, "detach (%s) returned %d", oldpath, error); 2825 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2826 return; 2827 } 2828 2829 /* 2830 * For the new vdev, choose with equal probability between the two 2831 * standard paths (ending in either 'a' or 'b') or a random hot spare. 2832 */ 2833 if (sav->sav_count != 0 && ztest_random(3) == 0) { 2834 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)]; 2835 newvd_is_spare = B_TRUE; 2836 (void) strcpy(newpath, newvd->vdev_path); 2837 } else { 2838 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template, 2839 ztest_opts.zo_dir, ztest_opts.zo_pool, 2840 top * leaves + leaf); 2841 if (ztest_random(2) == 0) 2842 newpath[strlen(newpath) - 1] = 'b'; 2843 newvd = vdev_lookup_by_path(rvd, newpath); 2844 } 2845 2846 if (newvd) { 2847 newsize = vdev_get_min_asize(newvd); 2848 } else { 2849 /* 2850 * Make newsize a little bigger or smaller than oldsize. 2851 * If it's smaller, the attach should fail. 2852 * If it's larger, and we're doing a replace, 2853 * we should get dynamic LUN growth when we're done. 2854 */ 2855 newsize = 10 * oldsize / (9 + ztest_random(3)); 2856 } 2857 2858 /* 2859 * If pvd is not a mirror or root, the attach should fail with ENOTSUP, 2860 * unless it's a replace; in that case any non-replacing parent is OK. 2861 * 2862 * If newvd is already part of the pool, it should fail with EBUSY. 2863 * 2864 * If newvd is too small, it should fail with EOVERFLOW. 2865 */ 2866 if (pvd->vdev_ops != &vdev_mirror_ops && 2867 pvd->vdev_ops != &vdev_root_ops && (!replacing || 2868 pvd->vdev_ops == &vdev_replacing_ops || 2869 pvd->vdev_ops == &vdev_spare_ops)) 2870 expected_error = ENOTSUP; 2871 else if (newvd_is_spare && (!replacing || oldvd_is_log)) 2872 expected_error = ENOTSUP; 2873 else if (newvd == oldvd) 2874 expected_error = replacing ? 0 : EBUSY; 2875 else if (vdev_lookup_by_path(rvd, newpath) != NULL) 2876 expected_error = EBUSY; 2877 else if (newsize < oldsize) 2878 expected_error = EOVERFLOW; 2879 else if (ashift > oldvd->vdev_top->vdev_ashift) 2880 expected_error = EDOM; 2881 else 2882 expected_error = 0; 2883 2884 spa_config_exit(spa, SCL_VDEV, FTAG); 2885 2886 /* 2887 * Build the nvlist describing newpath. 2888 */ 2889 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0, 2890 ashift, 0, 0, 0, 1); 2891 2892 error = spa_vdev_attach(spa, oldguid, root, replacing); 2893 2894 nvlist_free(root); 2895 2896 /* 2897 * If our parent was the replacing vdev, but the replace completed, 2898 * then instead of failing with ENOTSUP we may either succeed, 2899 * fail with ENODEV, or fail with EOVERFLOW. 2900 */ 2901 if (expected_error == ENOTSUP && 2902 (error == 0 || error == ENODEV || error == EOVERFLOW)) 2903 expected_error = error; 2904 2905 /* 2906 * If someone grew the LUN, the replacement may be too small. 2907 */ 2908 if (error == EOVERFLOW || error == EBUSY) 2909 expected_error = error; 2910 2911 /* XXX workaround 6690467 */ 2912 if (error != expected_error && expected_error != EBUSY) { 2913 fatal(0, "attach (%s %llu, %s %llu, %d) " 2914 "returned %d, expected %d", 2915 oldpath, oldsize, newpath, 2916 newsize, replacing, error, expected_error); 2917 } 2918 2919 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2920} 2921 2922/* 2923 * Callback function which expands the physical size of the vdev. 2924 */ 2925vdev_t * 2926grow_vdev(vdev_t *vd, void *arg) 2927{ 2928 spa_t *spa = vd->vdev_spa; 2929 size_t *newsize = arg; 2930 size_t fsize; 2931 int fd; 2932 2933 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); 2934 ASSERT(vd->vdev_ops->vdev_op_leaf); 2935 2936 if ((fd = open(vd->vdev_path, O_RDWR)) == -1) 2937 return (vd); 2938 2939 fsize = lseek(fd, 0, SEEK_END); 2940 (void) ftruncate(fd, *newsize); 2941 2942 if (ztest_opts.zo_verbose >= 6) { 2943 (void) printf("%s grew from %lu to %lu bytes\n", 2944 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize); 2945 } 2946 (void) close(fd); 2947 return (NULL); 2948} 2949 2950/* 2951 * Callback function which expands a given vdev by calling vdev_online(). 2952 */ 2953/* ARGSUSED */ 2954vdev_t * 2955online_vdev(vdev_t *vd, void *arg) 2956{ 2957 spa_t *spa = vd->vdev_spa; 2958 vdev_t *tvd = vd->vdev_top; 2959 uint64_t guid = vd->vdev_guid; 2960 uint64_t generation = spa->spa_config_generation + 1; 2961 vdev_state_t newstate = VDEV_STATE_UNKNOWN; 2962 int error; 2963 2964 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); 2965 ASSERT(vd->vdev_ops->vdev_op_leaf); 2966 2967 /* Calling vdev_online will initialize the new metaslabs */ 2968 spa_config_exit(spa, SCL_STATE, spa); 2969 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate); 2970 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 2971 2972 /* 2973 * If vdev_online returned an error or the underlying vdev_open 2974 * failed then we abort the expand. The only way to know that 2975 * vdev_open fails is by checking the returned newstate. 2976 */ 2977 if (error || newstate != VDEV_STATE_HEALTHY) { 2978 if (ztest_opts.zo_verbose >= 5) { 2979 (void) printf("Unable to expand vdev, state %llu, " 2980 "error %d\n", (u_longlong_t)newstate, error); 2981 } 2982 return (vd); 2983 } 2984 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY); 2985 2986 /* 2987 * Since we dropped the lock we need to ensure that we're 2988 * still talking to the original vdev. It's possible this 2989 * vdev may have been detached/replaced while we were 2990 * trying to online it. 2991 */ 2992 if (generation != spa->spa_config_generation) { 2993 if (ztest_opts.zo_verbose >= 5) { 2994 (void) printf("vdev configuration has changed, " 2995 "guid %llu, state %llu, expected gen %llu, " 2996 "got gen %llu\n", 2997 (u_longlong_t)guid, 2998 (u_longlong_t)tvd->vdev_state, 2999 (u_longlong_t)generation, 3000 (u_longlong_t)spa->spa_config_generation); 3001 } 3002 return (vd); 3003 } 3004 return (NULL); 3005} 3006 3007/* 3008 * Traverse the vdev tree calling the supplied function. 3009 * We continue to walk the tree until we either have walked all 3010 * children or we receive a non-NULL return from the callback. 3011 * If a NULL callback is passed, then we just return back the first 3012 * leaf vdev we encounter. 3013 */ 3014vdev_t * 3015vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg) 3016{ 3017 if (vd->vdev_ops->vdev_op_leaf) { 3018 if (func == NULL) 3019 return (vd); 3020 else 3021 return (func(vd, arg)); 3022 } 3023 3024 for (uint_t c = 0; c < vd->vdev_children; c++) { 3025 vdev_t *cvd = vd->vdev_child[c]; 3026 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL) 3027 return (cvd); 3028 } 3029 return (NULL); 3030} 3031 3032/* 3033 * Verify that dynamic LUN growth works as expected. 3034 */ 3035/* ARGSUSED */ 3036void 3037ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id) 3038{ 3039 spa_t *spa = ztest_spa; 3040 vdev_t *vd, *tvd; 3041 metaslab_class_t *mc; 3042 metaslab_group_t *mg; 3043 size_t psize, newsize; 3044 uint64_t top; 3045 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count; 3046 3047 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 3048 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3049 3050 top = ztest_random_vdev_top(spa, B_TRUE); 3051 3052 tvd = spa->spa_root_vdev->vdev_child[top]; 3053 mg = tvd->vdev_mg; 3054 mc = mg->mg_class; 3055 old_ms_count = tvd->vdev_ms_count; 3056 old_class_space = metaslab_class_get_space(mc); 3057 3058 /* 3059 * Determine the size of the first leaf vdev associated with 3060 * our top-level device. 3061 */ 3062 vd = vdev_walk_tree(tvd, NULL, NULL); 3063 ASSERT3P(vd, !=, NULL); 3064 ASSERT(vd->vdev_ops->vdev_op_leaf); 3065 3066 psize = vd->vdev_psize; 3067 3068 /* 3069 * We only try to expand the vdev if it's healthy, less than 4x its 3070 * original size, and it has a valid psize. 3071 */ 3072 if (tvd->vdev_state != VDEV_STATE_HEALTHY || 3073 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) { 3074 spa_config_exit(spa, SCL_STATE, spa); 3075 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3076 return; 3077 } 3078 ASSERT(psize > 0); 3079 newsize = psize + psize / 8; 3080 ASSERT3U(newsize, >, psize); 3081 3082 if (ztest_opts.zo_verbose >= 6) { 3083 (void) printf("Expanding LUN %s from %lu to %lu\n", 3084 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize); 3085 } 3086 3087 /* 3088 * Growing the vdev is a two step process: 3089 * 1). expand the physical size (i.e. relabel) 3090 * 2). online the vdev to create the new metaslabs 3091 */ 3092 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL || 3093 vdev_walk_tree(tvd, online_vdev, NULL) != NULL || 3094 tvd->vdev_state != VDEV_STATE_HEALTHY) { 3095 if (ztest_opts.zo_verbose >= 5) { 3096 (void) printf("Could not expand LUN because " 3097 "the vdev configuration changed.\n"); 3098 } 3099 spa_config_exit(spa, SCL_STATE, spa); 3100 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3101 return; 3102 } 3103 3104 spa_config_exit(spa, SCL_STATE, spa); 3105 3106 /* 3107 * Expanding the LUN will update the config asynchronously, 3108 * thus we must wait for the async thread to complete any 3109 * pending tasks before proceeding. 3110 */ 3111 for (;;) { 3112 boolean_t done; 3113 mutex_enter(&spa->spa_async_lock); 3114 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks); 3115 mutex_exit(&spa->spa_async_lock); 3116 if (done) 3117 break; 3118 txg_wait_synced(spa_get_dsl(spa), 0); 3119 (void) poll(NULL, 0, 100); 3120 } 3121 3122 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3123 3124 tvd = spa->spa_root_vdev->vdev_child[top]; 3125 new_ms_count = tvd->vdev_ms_count; 3126 new_class_space = metaslab_class_get_space(mc); 3127 3128 if (tvd->vdev_mg != mg || mg->mg_class != mc) { 3129 if (ztest_opts.zo_verbose >= 5) { 3130 (void) printf("Could not verify LUN expansion due to " 3131 "intervening vdev offline or remove.\n"); 3132 } 3133 spa_config_exit(spa, SCL_STATE, spa); 3134 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3135 return; 3136 } 3137 3138 /* 3139 * Make sure we were able to grow the vdev. 3140 */ 3141 if (new_ms_count <= old_ms_count) 3142 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n", 3143 old_ms_count, new_ms_count); 3144 3145 /* 3146 * Make sure we were able to grow the pool. 3147 */ 3148 if (new_class_space <= old_class_space) 3149 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n", 3150 old_class_space, new_class_space); 3151 3152 if (ztest_opts.zo_verbose >= 5) { 3153 char oldnumbuf[6], newnumbuf[6]; 3154 3155 nicenum(old_class_space, oldnumbuf); 3156 nicenum(new_class_space, newnumbuf); 3157 (void) printf("%s grew from %s to %s\n", 3158 spa->spa_name, oldnumbuf, newnumbuf); 3159 } 3160 3161 spa_config_exit(spa, SCL_STATE, spa); 3162 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3163} 3164 3165/* 3166 * Verify that dmu_objset_{create,destroy,open,close} work as expected. 3167 */ 3168/* ARGSUSED */ 3169static void 3170ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 3171{ 3172 /* 3173 * Create the objects common to all ztest datasets. 3174 */ 3175 VERIFY(zap_create_claim(os, ZTEST_DIROBJ, 3176 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 3177} 3178 3179static int 3180ztest_dataset_create(char *dsname) 3181{ 3182 uint64_t zilset = ztest_random(100); 3183 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, 3184 ztest_objset_create_cb, NULL); 3185 3186 if (err || zilset < 80) 3187 return (err); 3188 3189 if (ztest_opts.zo_verbose >= 6) 3190 (void) printf("Setting dataset %s to sync always\n", dsname); 3191 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC, 3192 ZFS_SYNC_ALWAYS, B_FALSE)); 3193} 3194 3195/* ARGSUSED */ 3196static int 3197ztest_objset_destroy_cb(const char *name, void *arg) 3198{ 3199 objset_t *os; 3200 dmu_object_info_t doi; 3201 int error; 3202 3203 /* 3204 * Verify that the dataset contains a directory object. 3205 */ 3206 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os)); 3207 error = dmu_object_info(os, ZTEST_DIROBJ, &doi); 3208 if (error != ENOENT) { 3209 /* We could have crashed in the middle of destroying it */ 3210 ASSERT0(error); 3211 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER); 3212 ASSERT3S(doi.doi_physical_blocks_512, >=, 0); 3213 } 3214 dmu_objset_disown(os, FTAG); 3215 3216 /* 3217 * Destroy the dataset. 3218 */ 3219 if (strchr(name, '@') != NULL) { 3220 VERIFY0(dsl_destroy_snapshot(name, B_FALSE)); 3221 } else { 3222 VERIFY0(dsl_destroy_head(name)); 3223 } 3224 return (0); 3225} 3226 3227static boolean_t 3228ztest_snapshot_create(char *osname, uint64_t id) 3229{ 3230 char snapname[MAXNAMELEN]; 3231 int error; 3232 3233 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id); 3234 3235 error = dmu_objset_snapshot_one(osname, snapname); 3236 if (error == ENOSPC) { 3237 ztest_record_enospc(FTAG); 3238 return (B_FALSE); 3239 } 3240 if (error != 0 && error != EEXIST) { 3241 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname, 3242 snapname, error); 3243 } 3244 return (B_TRUE); 3245} 3246 3247static boolean_t 3248ztest_snapshot_destroy(char *osname, uint64_t id) 3249{ 3250 char snapname[MAXNAMELEN]; 3251 int error; 3252 3253 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, 3254 (u_longlong_t)id); 3255 3256 error = dsl_destroy_snapshot(snapname, B_FALSE); 3257 if (error != 0 && error != ENOENT) 3258 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error); 3259 return (B_TRUE); 3260} 3261 3262/* ARGSUSED */ 3263void 3264ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) 3265{ 3266 ztest_ds_t zdtmp; 3267 int iters; 3268 int error; 3269 objset_t *os, *os2; 3270 char name[MAXNAMELEN]; 3271 zilog_t *zilog; 3272 3273 (void) rw_rdlock(&ztest_name_lock); 3274 3275 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu", 3276 ztest_opts.zo_pool, (u_longlong_t)id); 3277 3278 /* 3279 * If this dataset exists from a previous run, process its replay log 3280 * half of the time. If we don't replay it, then dmu_objset_destroy() 3281 * (invoked from ztest_objset_destroy_cb()) should just throw it away. 3282 */ 3283 if (ztest_random(2) == 0 && 3284 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) { 3285 ztest_zd_init(&zdtmp, NULL, os); 3286 zil_replay(os, &zdtmp, ztest_replay_vector); 3287 ztest_zd_fini(&zdtmp); 3288 dmu_objset_disown(os, FTAG); 3289 } 3290 3291 /* 3292 * There may be an old instance of the dataset we're about to 3293 * create lying around from a previous run. If so, destroy it 3294 * and all of its snapshots. 3295 */ 3296 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, 3297 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 3298 3299 /* 3300 * Verify that the destroyed dataset is no longer in the namespace. 3301 */ 3302 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, 3303 FTAG, &os)); 3304 3305 /* 3306 * Verify that we can create a new dataset. 3307 */ 3308 error = ztest_dataset_create(name); 3309 if (error) { 3310 if (error == ENOSPC) { 3311 ztest_record_enospc(FTAG); 3312 (void) rw_unlock(&ztest_name_lock); 3313 return; 3314 } 3315 fatal(0, "dmu_objset_create(%s) = %d", name, error); 3316 } 3317 3318 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); 3319 3320 ztest_zd_init(&zdtmp, NULL, os); 3321 3322 /* 3323 * Open the intent log for it. 3324 */ 3325 zilog = zil_open(os, ztest_get_data); 3326 3327 /* 3328 * Put some objects in there, do a little I/O to them, 3329 * and randomly take a couple of snapshots along the way. 3330 */ 3331 iters = ztest_random(5); 3332 for (int i = 0; i < iters; i++) { 3333 ztest_dmu_object_alloc_free(&zdtmp, id); 3334 if (ztest_random(iters) == 0) 3335 (void) ztest_snapshot_create(name, i); 3336 } 3337 3338 /* 3339 * Verify that we cannot create an existing dataset. 3340 */ 3341 VERIFY3U(EEXIST, ==, 3342 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL)); 3343 3344 /* 3345 * Verify that we can hold an objset that is also owned. 3346 */ 3347 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2)); 3348 dmu_objset_rele(os2, FTAG); 3349 3350 /* 3351 * Verify that we cannot own an objset that is already owned. 3352 */ 3353 VERIFY3U(EBUSY, ==, 3354 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2)); 3355 3356 zil_close(zilog); 3357 dmu_objset_disown(os, FTAG); 3358 ztest_zd_fini(&zdtmp); 3359 3360 (void) rw_unlock(&ztest_name_lock); 3361} 3362 3363/* 3364 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. 3365 */ 3366void 3367ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id) 3368{ 3369 (void) rw_rdlock(&ztest_name_lock); 3370 (void) ztest_snapshot_destroy(zd->zd_name, id); 3371 (void) ztest_snapshot_create(zd->zd_name, id); 3372 (void) rw_unlock(&ztest_name_lock); 3373} 3374 3375/* 3376 * Cleanup non-standard snapshots and clones. 3377 */ 3378void 3379ztest_dsl_dataset_cleanup(char *osname, uint64_t id) 3380{ 3381 char snap1name[MAXNAMELEN]; 3382 char clone1name[MAXNAMELEN]; 3383 char snap2name[MAXNAMELEN]; 3384 char clone2name[MAXNAMELEN]; 3385 char snap3name[MAXNAMELEN]; 3386 int error; 3387 3388 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id); 3389 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id); 3390 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id); 3391 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id); 3392 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id); 3393 3394 error = dsl_destroy_head(clone2name); 3395 if (error && error != ENOENT) 3396 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error); 3397 error = dsl_destroy_snapshot(snap3name, B_FALSE); 3398 if (error && error != ENOENT) 3399 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error); 3400 error = dsl_destroy_snapshot(snap2name, B_FALSE); 3401 if (error && error != ENOENT) 3402 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error); 3403 error = dsl_destroy_head(clone1name); 3404 if (error && error != ENOENT) 3405 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error); 3406 error = dsl_destroy_snapshot(snap1name, B_FALSE); 3407 if (error && error != ENOENT) 3408 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error); 3409} 3410 3411/* 3412 * Verify dsl_dataset_promote handles EBUSY 3413 */ 3414void 3415ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) 3416{ 3417 objset_t *os; 3418 char snap1name[MAXNAMELEN]; 3419 char clone1name[MAXNAMELEN]; 3420 char snap2name[MAXNAMELEN]; 3421 char clone2name[MAXNAMELEN]; 3422 char snap3name[MAXNAMELEN]; 3423 char *osname = zd->zd_name; 3424 int error; 3425 3426 (void) rw_rdlock(&ztest_name_lock); 3427 3428 ztest_dsl_dataset_cleanup(osname, id); 3429 3430 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id); 3431 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id); 3432 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id); 3433 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id); 3434 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id); 3435 3436 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1); 3437 if (error && error != EEXIST) { 3438 if (error == ENOSPC) { 3439 ztest_record_enospc(FTAG); 3440 goto out; 3441 } 3442 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error); 3443 } 3444 3445 error = dmu_objset_clone(clone1name, snap1name); 3446 if (error) { 3447 if (error == ENOSPC) { 3448 ztest_record_enospc(FTAG); 3449 goto out; 3450 } 3451 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error); 3452 } 3453 3454 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1); 3455 if (error && error != EEXIST) { 3456 if (error == ENOSPC) { 3457 ztest_record_enospc(FTAG); 3458 goto out; 3459 } 3460 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error); 3461 } 3462 3463 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1); 3464 if (error && error != EEXIST) { 3465 if (error == ENOSPC) { 3466 ztest_record_enospc(FTAG); 3467 goto out; 3468 } 3469 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); 3470 } 3471 3472 error = dmu_objset_clone(clone2name, snap3name); 3473 if (error) { 3474 if (error == ENOSPC) { 3475 ztest_record_enospc(FTAG); 3476 goto out; 3477 } 3478 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error); 3479 } 3480 3481 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os); 3482 if (error) 3483 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error); 3484 error = dsl_dataset_promote(clone2name, NULL); 3485 if (error == ENOSPC) { 3486 dmu_objset_disown(os, FTAG); 3487 ztest_record_enospc(FTAG); 3488 goto out; 3489 } 3490 if (error != EBUSY) 3491 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name, 3492 error); 3493 dmu_objset_disown(os, FTAG); 3494 3495out: 3496 ztest_dsl_dataset_cleanup(osname, id); 3497 3498 (void) rw_unlock(&ztest_name_lock); 3499} 3500 3501/* 3502 * Verify that dmu_object_{alloc,free} work as expected. 3503 */ 3504void 3505ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id) 3506{ 3507 ztest_od_t od[4]; 3508 int batchsize = sizeof (od) / sizeof (od[0]); 3509 3510 for (int b = 0; b < batchsize; b++) 3511 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0); 3512 3513 /* 3514 * Destroy the previous batch of objects, create a new batch, 3515 * and do some I/O on the new objects. 3516 */ 3517 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0) 3518 return; 3519 3520 while (ztest_random(4 * batchsize) != 0) 3521 ztest_io(zd, od[ztest_random(batchsize)].od_object, 3522 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 3523} 3524 3525/* 3526 * Verify that dmu_{read,write} work as expected. 3527 */ 3528void 3529ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id) 3530{ 3531 objset_t *os = zd->zd_os; 3532 ztest_od_t od[2]; 3533 dmu_tx_t *tx; 3534 int i, freeit, error; 3535 uint64_t n, s, txg; 3536 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; 3537 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; 3538 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t); 3539 uint64_t regions = 997; 3540 uint64_t stride = 123456789ULL; 3541 uint64_t width = 40; 3542 int free_percent = 5; 3543 3544 /* 3545 * This test uses two objects, packobj and bigobj, that are always 3546 * updated together (i.e. in the same tx) so that their contents are 3547 * in sync and can be compared. Their contents relate to each other 3548 * in a simple way: packobj is a dense array of 'bufwad' structures, 3549 * while bigobj is a sparse array of the same bufwads. Specifically, 3550 * for any index n, there are three bufwads that should be identical: 3551 * 3552 * packobj, at offset n * sizeof (bufwad_t) 3553 * bigobj, at the head of the nth chunk 3554 * bigobj, at the tail of the nth chunk 3555 * 3556 * The chunk size is arbitrary. It doesn't have to be a power of two, 3557 * and it doesn't have any relation to the object blocksize. 3558 * The only requirement is that it can hold at least two bufwads. 3559 * 3560 * Normally, we write the bufwad to each of these locations. 3561 * However, free_percent of the time we instead write zeroes to 3562 * packobj and perform a dmu_free_range() on bigobj. By comparing 3563 * bigobj to packobj, we can verify that the DMU is correctly 3564 * tracking which parts of an object are allocated and free, 3565 * and that the contents of the allocated blocks are correct. 3566 */ 3567 3568 /* 3569 * Read the directory info. If it's the first time, set things up. 3570 */ 3571 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize); 3572 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); 3573 3574 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 3575 return; 3576 3577 bigobj = od[0].od_object; 3578 packobj = od[1].od_object; 3579 chunksize = od[0].od_gen; 3580 ASSERT(chunksize == od[1].od_gen); 3581 3582 /* 3583 * Prefetch a random chunk of the big object. 3584 * Our aim here is to get some async reads in flight 3585 * for blocks that we may free below; the DMU should 3586 * handle this race correctly. 3587 */ 3588 n = ztest_random(regions) * stride + ztest_random(width); 3589 s = 1 + ztest_random(2 * width - 1); 3590 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize, 3591 ZIO_PRIORITY_SYNC_READ); 3592 3593 /* 3594 * Pick a random index and compute the offsets into packobj and bigobj. 3595 */ 3596 n = ztest_random(regions) * stride + ztest_random(width); 3597 s = 1 + ztest_random(width - 1); 3598 3599 packoff = n * sizeof (bufwad_t); 3600 packsize = s * sizeof (bufwad_t); 3601 3602 bigoff = n * chunksize; 3603 bigsize = s * chunksize; 3604 3605 packbuf = umem_alloc(packsize, UMEM_NOFAIL); 3606 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); 3607 3608 /* 3609 * free_percent of the time, free a range of bigobj rather than 3610 * overwriting it. 3611 */ 3612 freeit = (ztest_random(100) < free_percent); 3613 3614 /* 3615 * Read the current contents of our objects. 3616 */ 3617 error = dmu_read(os, packobj, packoff, packsize, packbuf, 3618 DMU_READ_PREFETCH); 3619 ASSERT0(error); 3620 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf, 3621 DMU_READ_PREFETCH); 3622 ASSERT0(error); 3623 3624 /* 3625 * Get a tx for the mods to both packobj and bigobj. 3626 */ 3627 tx = dmu_tx_create(os); 3628 3629 dmu_tx_hold_write(tx, packobj, packoff, packsize); 3630 3631 if (freeit) 3632 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize); 3633 else 3634 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); 3635 3636 /* This accounts for setting the checksum/compression. */ 3637 dmu_tx_hold_bonus(tx, bigobj); 3638 3639 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 3640 if (txg == 0) { 3641 umem_free(packbuf, packsize); 3642 umem_free(bigbuf, bigsize); 3643 return; 3644 } 3645 3646 enum zio_checksum cksum; 3647 do { 3648 cksum = (enum zio_checksum) 3649 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM); 3650 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS); 3651 dmu_object_set_checksum(os, bigobj, cksum, tx); 3652 3653 enum zio_compress comp; 3654 do { 3655 comp = (enum zio_compress) 3656 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION); 3657 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS); 3658 dmu_object_set_compress(os, bigobj, comp, tx); 3659 3660 /* 3661 * For each index from n to n + s, verify that the existing bufwad 3662 * in packobj matches the bufwads at the head and tail of the 3663 * corresponding chunk in bigobj. Then update all three bufwads 3664 * with the new values we want to write out. 3665 */ 3666 for (i = 0; i < s; i++) { 3667 /* LINTED */ 3668 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 3669 /* LINTED */ 3670 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); 3671 /* LINTED */ 3672 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; 3673 3674 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 3675 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 3676 3677 if (pack->bw_txg > txg) 3678 fatal(0, "future leak: got %llx, open txg is %llx", 3679 pack->bw_txg, txg); 3680 3681 if (pack->bw_data != 0 && pack->bw_index != n + i) 3682 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 3683 pack->bw_index, n, i); 3684 3685 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 3686 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 3687 3688 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 3689 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 3690 3691 if (freeit) { 3692 bzero(pack, sizeof (bufwad_t)); 3693 } else { 3694 pack->bw_index = n + i; 3695 pack->bw_txg = txg; 3696 pack->bw_data = 1 + ztest_random(-2ULL); 3697 } 3698 *bigH = *pack; 3699 *bigT = *pack; 3700 } 3701 3702 /* 3703 * We've verified all the old bufwads, and made new ones. 3704 * Now write them out. 3705 */ 3706 dmu_write(os, packobj, packoff, packsize, packbuf, tx); 3707 3708 if (freeit) { 3709 if (ztest_opts.zo_verbose >= 7) { 3710 (void) printf("freeing offset %llx size %llx" 3711 " txg %llx\n", 3712 (u_longlong_t)bigoff, 3713 (u_longlong_t)bigsize, 3714 (u_longlong_t)txg); 3715 } 3716 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx)); 3717 } else { 3718 if (ztest_opts.zo_verbose >= 7) { 3719 (void) printf("writing offset %llx size %llx" 3720 " txg %llx\n", 3721 (u_longlong_t)bigoff, 3722 (u_longlong_t)bigsize, 3723 (u_longlong_t)txg); 3724 } 3725 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx); 3726 } 3727 3728 dmu_tx_commit(tx); 3729 3730 /* 3731 * Sanity check the stuff we just wrote. 3732 */ 3733 { 3734 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 3735 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 3736 3737 VERIFY(0 == dmu_read(os, packobj, packoff, 3738 packsize, packcheck, DMU_READ_PREFETCH)); 3739 VERIFY(0 == dmu_read(os, bigobj, bigoff, 3740 bigsize, bigcheck, DMU_READ_PREFETCH)); 3741 3742 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 3743 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 3744 3745 umem_free(packcheck, packsize); 3746 umem_free(bigcheck, bigsize); 3747 } 3748 3749 umem_free(packbuf, packsize); 3750 umem_free(bigbuf, bigsize); 3751} 3752 3753void 3754compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf, 3755 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg) 3756{ 3757 uint64_t i; 3758 bufwad_t *pack; 3759 bufwad_t *bigH; 3760 bufwad_t *bigT; 3761 3762 /* 3763 * For each index from n to n + s, verify that the existing bufwad 3764 * in packobj matches the bufwads at the head and tail of the 3765 * corresponding chunk in bigobj. Then update all three bufwads 3766 * with the new values we want to write out. 3767 */ 3768 for (i = 0; i < s; i++) { 3769 /* LINTED */ 3770 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 3771 /* LINTED */ 3772 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); 3773 /* LINTED */ 3774 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; 3775 3776 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 3777 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 3778 3779 if (pack->bw_txg > txg) 3780 fatal(0, "future leak: got %llx, open txg is %llx", 3781 pack->bw_txg, txg); 3782 3783 if (pack->bw_data != 0 && pack->bw_index != n + i) 3784 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 3785 pack->bw_index, n, i); 3786 3787 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 3788 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 3789 3790 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 3791 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 3792 3793 pack->bw_index = n + i; 3794 pack->bw_txg = txg; 3795 pack->bw_data = 1 + ztest_random(-2ULL); 3796 3797 *bigH = *pack; 3798 *bigT = *pack; 3799 } 3800} 3801 3802void 3803ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id) 3804{ 3805 objset_t *os = zd->zd_os; 3806 ztest_od_t od[2]; 3807 dmu_tx_t *tx; 3808 uint64_t i; 3809 int error; 3810 uint64_t n, s, txg; 3811 bufwad_t *packbuf, *bigbuf; 3812 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; 3813 uint64_t blocksize = ztest_random_blocksize(); 3814 uint64_t chunksize = blocksize; 3815 uint64_t regions = 997; 3816 uint64_t stride = 123456789ULL; 3817 uint64_t width = 9; 3818 dmu_buf_t *bonus_db; 3819 arc_buf_t **bigbuf_arcbufs; 3820 dmu_object_info_t doi; 3821 3822 /* 3823 * This test uses two objects, packobj and bigobj, that are always 3824 * updated together (i.e. in the same tx) so that their contents are 3825 * in sync and can be compared. Their contents relate to each other 3826 * in a simple way: packobj is a dense array of 'bufwad' structures, 3827 * while bigobj is a sparse array of the same bufwads. Specifically, 3828 * for any index n, there are three bufwads that should be identical: 3829 * 3830 * packobj, at offset n * sizeof (bufwad_t) 3831 * bigobj, at the head of the nth chunk 3832 * bigobj, at the tail of the nth chunk 3833 * 3834 * The chunk size is set equal to bigobj block size so that 3835 * dmu_assign_arcbuf() can be tested for object updates. 3836 */ 3837 3838 /* 3839 * Read the directory info. If it's the first time, set things up. 3840 */ 3841 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 3842 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); 3843 3844 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 3845 return; 3846 3847 bigobj = od[0].od_object; 3848 packobj = od[1].od_object; 3849 blocksize = od[0].od_blocksize; 3850 chunksize = blocksize; 3851 ASSERT(chunksize == od[1].od_gen); 3852 3853 VERIFY(dmu_object_info(os, bigobj, &doi) == 0); 3854 VERIFY(ISP2(doi.doi_data_block_size)); 3855 VERIFY(chunksize == doi.doi_data_block_size); 3856 VERIFY(chunksize >= 2 * sizeof (bufwad_t)); 3857 3858 /* 3859 * Pick a random index and compute the offsets into packobj and bigobj. 3860 */ 3861 n = ztest_random(regions) * stride + ztest_random(width); 3862 s = 1 + ztest_random(width - 1); 3863 3864 packoff = n * sizeof (bufwad_t); 3865 packsize = s * sizeof (bufwad_t); 3866 3867 bigoff = n * chunksize; 3868 bigsize = s * chunksize; 3869 3870 packbuf = umem_zalloc(packsize, UMEM_NOFAIL); 3871 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL); 3872 3873 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db)); 3874 3875 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL); 3876 3877 /* 3878 * Iteration 0 test zcopy for DB_UNCACHED dbufs. 3879 * Iteration 1 test zcopy to already referenced dbufs. 3880 * Iteration 2 test zcopy to dirty dbuf in the same txg. 3881 * Iteration 3 test zcopy to dbuf dirty in previous txg. 3882 * Iteration 4 test zcopy when dbuf is no longer dirty. 3883 * Iteration 5 test zcopy when it can't be done. 3884 * Iteration 6 one more zcopy write. 3885 */ 3886 for (i = 0; i < 7; i++) { 3887 uint64_t j; 3888 uint64_t off; 3889 3890 /* 3891 * In iteration 5 (i == 5) use arcbufs 3892 * that don't match bigobj blksz to test 3893 * dmu_assign_arcbuf() when it can't directly 3894 * assign an arcbuf to a dbuf. 3895 */ 3896 for (j = 0; j < s; j++) { 3897 if (i != 5) { 3898 bigbuf_arcbufs[j] = 3899 dmu_request_arcbuf(bonus_db, chunksize); 3900 } else { 3901 bigbuf_arcbufs[2 * j] = 3902 dmu_request_arcbuf(bonus_db, chunksize / 2); 3903 bigbuf_arcbufs[2 * j + 1] = 3904 dmu_request_arcbuf(bonus_db, chunksize / 2); 3905 } 3906 } 3907 3908 /* 3909 * Get a tx for the mods to both packobj and bigobj. 3910 */ 3911 tx = dmu_tx_create(os); 3912 3913 dmu_tx_hold_write(tx, packobj, packoff, packsize); 3914 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); 3915 3916 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 3917 if (txg == 0) { 3918 umem_free(packbuf, packsize); 3919 umem_free(bigbuf, bigsize); 3920 for (j = 0; j < s; j++) { 3921 if (i != 5) { 3922 dmu_return_arcbuf(bigbuf_arcbufs[j]); 3923 } else { 3924 dmu_return_arcbuf( 3925 bigbuf_arcbufs[2 * j]); 3926 dmu_return_arcbuf( 3927 bigbuf_arcbufs[2 * j + 1]); 3928 } 3929 } 3930 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); 3931 dmu_buf_rele(bonus_db, FTAG); 3932 return; 3933 } 3934 3935 /* 3936 * 50% of the time don't read objects in the 1st iteration to 3937 * test dmu_assign_arcbuf() for the case when there're no 3938 * existing dbufs for the specified offsets. 3939 */ 3940 if (i != 0 || ztest_random(2) != 0) { 3941 error = dmu_read(os, packobj, packoff, 3942 packsize, packbuf, DMU_READ_PREFETCH); 3943 ASSERT0(error); 3944 error = dmu_read(os, bigobj, bigoff, bigsize, 3945 bigbuf, DMU_READ_PREFETCH); 3946 ASSERT0(error); 3947 } 3948 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize, 3949 n, chunksize, txg); 3950 3951 /* 3952 * We've verified all the old bufwads, and made new ones. 3953 * Now write them out. 3954 */ 3955 dmu_write(os, packobj, packoff, packsize, packbuf, tx); 3956 if (ztest_opts.zo_verbose >= 7) { 3957 (void) printf("writing offset %llx size %llx" 3958 " txg %llx\n", 3959 (u_longlong_t)bigoff, 3960 (u_longlong_t)bigsize, 3961 (u_longlong_t)txg); 3962 } 3963 for (off = bigoff, j = 0; j < s; j++, off += chunksize) { 3964 dmu_buf_t *dbt; 3965 if (i != 5) { 3966 bcopy((caddr_t)bigbuf + (off - bigoff), 3967 bigbuf_arcbufs[j]->b_data, chunksize); 3968 } else { 3969 bcopy((caddr_t)bigbuf + (off - bigoff), 3970 bigbuf_arcbufs[2 * j]->b_data, 3971 chunksize / 2); 3972 bcopy((caddr_t)bigbuf + (off - bigoff) + 3973 chunksize / 2, 3974 bigbuf_arcbufs[2 * j + 1]->b_data, 3975 chunksize / 2); 3976 } 3977 3978 if (i == 1) { 3979 VERIFY(dmu_buf_hold(os, bigobj, off, 3980 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0); 3981 } 3982 if (i != 5) { 3983 dmu_assign_arcbuf(bonus_db, off, 3984 bigbuf_arcbufs[j], tx); 3985 } else { 3986 dmu_assign_arcbuf(bonus_db, off, 3987 bigbuf_arcbufs[2 * j], tx); 3988 dmu_assign_arcbuf(bonus_db, 3989 off + chunksize / 2, 3990 bigbuf_arcbufs[2 * j + 1], tx); 3991 } 3992 if (i == 1) { 3993 dmu_buf_rele(dbt, FTAG); 3994 } 3995 } 3996 dmu_tx_commit(tx); 3997 3998 /* 3999 * Sanity check the stuff we just wrote. 4000 */ 4001 { 4002 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 4003 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 4004 4005 VERIFY(0 == dmu_read(os, packobj, packoff, 4006 packsize, packcheck, DMU_READ_PREFETCH)); 4007 VERIFY(0 == dmu_read(os, bigobj, bigoff, 4008 bigsize, bigcheck, DMU_READ_PREFETCH)); 4009 4010 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 4011 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 4012 4013 umem_free(packcheck, packsize); 4014 umem_free(bigcheck, bigsize); 4015 } 4016 if (i == 2) { 4017 txg_wait_open(dmu_objset_pool(os), 0); 4018 } else if (i == 3) { 4019 txg_wait_synced(dmu_objset_pool(os), 0); 4020 } 4021 } 4022 4023 dmu_buf_rele(bonus_db, FTAG); 4024 umem_free(packbuf, packsize); 4025 umem_free(bigbuf, bigsize); 4026 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); 4027} 4028 4029/* ARGSUSED */ 4030void 4031ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id) 4032{ 4033 ztest_od_t od[1]; 4034 uint64_t offset = (1ULL << (ztest_random(20) + 43)) + 4035 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 4036 4037 /* 4038 * Have multiple threads write to large offsets in an object 4039 * to verify that parallel writes to an object -- even to the 4040 * same blocks within the object -- doesn't cause any trouble. 4041 */ 4042 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 4043 4044 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4045 return; 4046 4047 while (ztest_random(10) != 0) 4048 ztest_io(zd, od[0].od_object, offset); 4049} 4050 4051void 4052ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id) 4053{ 4054 ztest_od_t od[1]; 4055 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) + 4056 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 4057 uint64_t count = ztest_random(20) + 1; 4058 uint64_t blocksize = ztest_random_blocksize(); 4059 void *data; 4060 4061 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 4062 4063 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4064 return; 4065 4066 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0) 4067 return; 4068 4069 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize); 4070 4071 data = umem_zalloc(blocksize, UMEM_NOFAIL); 4072 4073 while (ztest_random(count) != 0) { 4074 uint64_t randoff = offset + (ztest_random(count) * blocksize); 4075 if (ztest_write(zd, od[0].od_object, randoff, blocksize, 4076 data) != 0) 4077 break; 4078 while (ztest_random(4) != 0) 4079 ztest_io(zd, od[0].od_object, randoff); 4080 } 4081 4082 umem_free(data, blocksize); 4083} 4084 4085/* 4086 * Verify that zap_{create,destroy,add,remove,update} work as expected. 4087 */ 4088#define ZTEST_ZAP_MIN_INTS 1 4089#define ZTEST_ZAP_MAX_INTS 4 4090#define ZTEST_ZAP_MAX_PROPS 1000 4091 4092void 4093ztest_zap(ztest_ds_t *zd, uint64_t id) 4094{ 4095 objset_t *os = zd->zd_os; 4096 ztest_od_t od[1]; 4097 uint64_t object; 4098 uint64_t txg, last_txg; 4099 uint64_t value[ZTEST_ZAP_MAX_INTS]; 4100 uint64_t zl_ints, zl_intsize, prop; 4101 int i, ints; 4102 dmu_tx_t *tx; 4103 char propname[100], txgname[100]; 4104 int error; 4105 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; 4106 4107 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); 4108 4109 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4110 return; 4111 4112 object = od[0].od_object; 4113 4114 /* 4115 * Generate a known hash collision, and verify that 4116 * we can lookup and remove both entries. 4117 */ 4118 tx = dmu_tx_create(os); 4119 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4120 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4121 if (txg == 0) 4122 return; 4123 for (i = 0; i < 2; i++) { 4124 value[i] = i; 4125 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t), 4126 1, &value[i], tx)); 4127 } 4128 for (i = 0; i < 2; i++) { 4129 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i], 4130 sizeof (uint64_t), 1, &value[i], tx)); 4131 VERIFY3U(0, ==, 4132 zap_length(os, object, hc[i], &zl_intsize, &zl_ints)); 4133 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4134 ASSERT3U(zl_ints, ==, 1); 4135 } 4136 for (i = 0; i < 2; i++) { 4137 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx)); 4138 } 4139 dmu_tx_commit(tx); 4140 4141 /* 4142 * Generate a buch of random entries. 4143 */ 4144 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); 4145 4146 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 4147 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 4148 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 4149 bzero(value, sizeof (value)); 4150 last_txg = 0; 4151 4152 /* 4153 * If these zap entries already exist, validate their contents. 4154 */ 4155 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 4156 if (error == 0) { 4157 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4158 ASSERT3U(zl_ints, ==, 1); 4159 4160 VERIFY(zap_lookup(os, object, txgname, zl_intsize, 4161 zl_ints, &last_txg) == 0); 4162 4163 VERIFY(zap_length(os, object, propname, &zl_intsize, 4164 &zl_ints) == 0); 4165 4166 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4167 ASSERT3U(zl_ints, ==, ints); 4168 4169 VERIFY(zap_lookup(os, object, propname, zl_intsize, 4170 zl_ints, value) == 0); 4171 4172 for (i = 0; i < ints; i++) { 4173 ASSERT3U(value[i], ==, last_txg + object + i); 4174 } 4175 } else { 4176 ASSERT3U(error, ==, ENOENT); 4177 } 4178 4179 /* 4180 * Atomically update two entries in our zap object. 4181 * The first is named txg_%llu, and contains the txg 4182 * in which the property was last updated. The second 4183 * is named prop_%llu, and the nth element of its value 4184 * should be txg + object + n. 4185 */ 4186 tx = dmu_tx_create(os); 4187 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4188 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4189 if (txg == 0) 4190 return; 4191 4192 if (last_txg > txg) 4193 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); 4194 4195 for (i = 0; i < ints; i++) 4196 value[i] = txg + object + i; 4197 4198 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t), 4199 1, &txg, tx)); 4200 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t), 4201 ints, value, tx)); 4202 4203 dmu_tx_commit(tx); 4204 4205 /* 4206 * Remove a random pair of entries. 4207 */ 4208 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 4209 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 4210 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 4211 4212 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 4213 4214 if (error == ENOENT) 4215 return; 4216 4217 ASSERT0(error); 4218 4219 tx = dmu_tx_create(os); 4220 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4221 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4222 if (txg == 0) 4223 return; 4224 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx)); 4225 VERIFY3U(0, ==, zap_remove(os, object, propname, tx)); 4226 dmu_tx_commit(tx); 4227} 4228 4229/* 4230 * Testcase to test the upgrading of a microzap to fatzap. 4231 */ 4232void 4233ztest_fzap(ztest_ds_t *zd, uint64_t id) 4234{ 4235 objset_t *os = zd->zd_os; 4236 ztest_od_t od[1]; 4237 uint64_t object, txg; 4238 4239 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); 4240 4241 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4242 return; 4243 4244 object = od[0].od_object; 4245 4246 /* 4247 * Add entries to this ZAP and make sure it spills over 4248 * and gets upgraded to a fatzap. Also, since we are adding 4249 * 2050 entries we should see ptrtbl growth and leaf-block split. 4250 */ 4251 for (int i = 0; i < 2050; i++) { 4252 char name[MAXNAMELEN]; 4253 uint64_t value = i; 4254 dmu_tx_t *tx; 4255 int error; 4256 4257 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu", 4258 id, value); 4259 4260 tx = dmu_tx_create(os); 4261 dmu_tx_hold_zap(tx, object, B_TRUE, name); 4262 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4263 if (txg == 0) 4264 return; 4265 error = zap_add(os, object, name, sizeof (uint64_t), 1, 4266 &value, tx); 4267 ASSERT(error == 0 || error == EEXIST); 4268 dmu_tx_commit(tx); 4269 } 4270} 4271 4272/* ARGSUSED */ 4273void 4274ztest_zap_parallel(ztest_ds_t *zd, uint64_t id) 4275{ 4276 objset_t *os = zd->zd_os; 4277 ztest_od_t od[1]; 4278 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; 4279 dmu_tx_t *tx; 4280 int i, namelen, error; 4281 int micro = ztest_random(2); 4282 char name[20], string_value[20]; 4283 void *data; 4284 4285 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0); 4286 4287 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4288 return; 4289 4290 object = od[0].od_object; 4291 4292 /* 4293 * Generate a random name of the form 'xxx.....' where each 4294 * x is a random printable character and the dots are dots. 4295 * There are 94 such characters, and the name length goes from 4296 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. 4297 */ 4298 namelen = ztest_random(sizeof (name) - 5) + 5 + 1; 4299 4300 for (i = 0; i < 3; i++) 4301 name[i] = '!' + ztest_random('~' - '!' + 1); 4302 for (; i < namelen - 1; i++) 4303 name[i] = '.'; 4304 name[i] = '\0'; 4305 4306 if ((namelen & 1) || micro) { 4307 wsize = sizeof (txg); 4308 wc = 1; 4309 data = &txg; 4310 } else { 4311 wsize = 1; 4312 wc = namelen; 4313 data = string_value; 4314 } 4315 4316 count = -1ULL; 4317 VERIFY0(zap_count(os, object, &count)); 4318 ASSERT(count != -1ULL); 4319 4320 /* 4321 * Select an operation: length, lookup, add, update, remove. 4322 */ 4323 i = ztest_random(5); 4324 4325 if (i >= 2) { 4326 tx = dmu_tx_create(os); 4327 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4328 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4329 if (txg == 0) 4330 return; 4331 bcopy(name, string_value, namelen); 4332 } else { 4333 tx = NULL; 4334 txg = 0; 4335 bzero(string_value, namelen); 4336 } 4337 4338 switch (i) { 4339 4340 case 0: 4341 error = zap_length(os, object, name, &zl_wsize, &zl_wc); 4342 if (error == 0) { 4343 ASSERT3U(wsize, ==, zl_wsize); 4344 ASSERT3U(wc, ==, zl_wc); 4345 } else { 4346 ASSERT3U(error, ==, ENOENT); 4347 } 4348 break; 4349 4350 case 1: 4351 error = zap_lookup(os, object, name, wsize, wc, data); 4352 if (error == 0) { 4353 if (data == string_value && 4354 bcmp(name, data, namelen) != 0) 4355 fatal(0, "name '%s' != val '%s' len %d", 4356 name, data, namelen); 4357 } else { 4358 ASSERT3U(error, ==, ENOENT); 4359 } 4360 break; 4361 4362 case 2: 4363 error = zap_add(os, object, name, wsize, wc, data, tx); 4364 ASSERT(error == 0 || error == EEXIST); 4365 break; 4366 4367 case 3: 4368 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); 4369 break; 4370 4371 case 4: 4372 error = zap_remove(os, object, name, tx); 4373 ASSERT(error == 0 || error == ENOENT); 4374 break; 4375 } 4376 4377 if (tx != NULL) 4378 dmu_tx_commit(tx); 4379} 4380 4381/* 4382 * Commit callback data. 4383 */ 4384typedef struct ztest_cb_data { 4385 list_node_t zcd_node; 4386 uint64_t zcd_txg; 4387 int zcd_expected_err; 4388 boolean_t zcd_added; 4389 boolean_t zcd_called; 4390 spa_t *zcd_spa; 4391} ztest_cb_data_t; 4392 4393/* This is the actual commit callback function */ 4394static void 4395ztest_commit_callback(void *arg, int error) 4396{ 4397 ztest_cb_data_t *data = arg; 4398 uint64_t synced_txg; 4399 4400 VERIFY(data != NULL); 4401 VERIFY3S(data->zcd_expected_err, ==, error); 4402 VERIFY(!data->zcd_called); 4403 4404 synced_txg = spa_last_synced_txg(data->zcd_spa); 4405 if (data->zcd_txg > synced_txg) 4406 fatal(0, "commit callback of txg %" PRIu64 " called prematurely" 4407 ", last synced txg = %" PRIu64 "\n", data->zcd_txg, 4408 synced_txg); 4409 4410 data->zcd_called = B_TRUE; 4411 4412 if (error == ECANCELED) { 4413 ASSERT0(data->zcd_txg); 4414 ASSERT(!data->zcd_added); 4415 4416 /* 4417 * The private callback data should be destroyed here, but 4418 * since we are going to check the zcd_called field after 4419 * dmu_tx_abort(), we will destroy it there. 4420 */ 4421 return; 4422 } 4423 4424 /* Was this callback added to the global callback list? */ 4425 if (!data->zcd_added) 4426 goto out; 4427 4428 ASSERT3U(data->zcd_txg, !=, 0); 4429 4430 /* Remove our callback from the list */ 4431 (void) mutex_lock(&zcl.zcl_callbacks_lock); 4432 list_remove(&zcl.zcl_callbacks, data); 4433 (void) mutex_unlock(&zcl.zcl_callbacks_lock); 4434 4435out: 4436 umem_free(data, sizeof (ztest_cb_data_t)); 4437} 4438 4439/* Allocate and initialize callback data structure */ 4440static ztest_cb_data_t * 4441ztest_create_cb_data(objset_t *os, uint64_t txg) 4442{ 4443 ztest_cb_data_t *cb_data; 4444 4445 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL); 4446 4447 cb_data->zcd_txg = txg; 4448 cb_data->zcd_spa = dmu_objset_spa(os); 4449 4450 return (cb_data); 4451} 4452 4453/* 4454 * If a number of txgs equal to this threshold have been created after a commit 4455 * callback has been registered but not called, then we assume there is an 4456 * implementation bug. 4457 */ 4458#define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2) 4459 4460/* 4461 * Commit callback test. 4462 */ 4463void 4464ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id) 4465{ 4466 objset_t *os = zd->zd_os; 4467 ztest_od_t od[1]; 4468 dmu_tx_t *tx; 4469 ztest_cb_data_t *cb_data[3], *tmp_cb; 4470 uint64_t old_txg, txg; 4471 int i, error; 4472 4473 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 4474 4475 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4476 return; 4477 4478 tx = dmu_tx_create(os); 4479 4480 cb_data[0] = ztest_create_cb_data(os, 0); 4481 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]); 4482 4483 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t)); 4484 4485 /* Every once in a while, abort the transaction on purpose */ 4486 if (ztest_random(100) == 0) 4487 error = -1; 4488 4489 if (!error) 4490 error = dmu_tx_assign(tx, TXG_NOWAIT); 4491 4492 txg = error ? 0 : dmu_tx_get_txg(tx); 4493 4494 cb_data[0]->zcd_txg = txg; 4495 cb_data[1] = ztest_create_cb_data(os, txg); 4496 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]); 4497 4498 if (error) { 4499 /* 4500 * It's not a strict requirement to call the registered 4501 * callbacks from inside dmu_tx_abort(), but that's what 4502 * it's supposed to happen in the current implementation 4503 * so we will check for that. 4504 */ 4505 for (i = 0; i < 2; i++) { 4506 cb_data[i]->zcd_expected_err = ECANCELED; 4507 VERIFY(!cb_data[i]->zcd_called); 4508 } 4509 4510 dmu_tx_abort(tx); 4511 4512 for (i = 0; i < 2; i++) { 4513 VERIFY(cb_data[i]->zcd_called); 4514 umem_free(cb_data[i], sizeof (ztest_cb_data_t)); 4515 } 4516 4517 return; 4518 } 4519 4520 cb_data[2] = ztest_create_cb_data(os, txg); 4521 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]); 4522 4523 /* 4524 * Read existing data to make sure there isn't a future leak. 4525 */ 4526 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t), 4527 &old_txg, DMU_READ_PREFETCH)); 4528 4529 if (old_txg > txg) 4530 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64, 4531 old_txg, txg); 4532 4533 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx); 4534 4535 (void) mutex_lock(&zcl.zcl_callbacks_lock); 4536 4537 /* 4538 * Since commit callbacks don't have any ordering requirement and since 4539 * it is theoretically possible for a commit callback to be called 4540 * after an arbitrary amount of time has elapsed since its txg has been 4541 * synced, it is difficult to reliably determine whether a commit 4542 * callback hasn't been called due to high load or due to a flawed 4543 * implementation. 4544 * 4545 * In practice, we will assume that if after a certain number of txgs a 4546 * commit callback hasn't been called, then most likely there's an 4547 * implementation bug.. 4548 */ 4549 tmp_cb = list_head(&zcl.zcl_callbacks); 4550 if (tmp_cb != NULL && 4551 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) { 4552 fatal(0, "Commit callback threshold exceeded, oldest txg: %" 4553 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg); 4554 } 4555 4556 /* 4557 * Let's find the place to insert our callbacks. 4558 * 4559 * Even though the list is ordered by txg, it is possible for the 4560 * insertion point to not be the end because our txg may already be 4561 * quiescing at this point and other callbacks in the open txg 4562 * (from other objsets) may have sneaked in. 4563 */ 4564 tmp_cb = list_tail(&zcl.zcl_callbacks); 4565 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg) 4566 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb); 4567 4568 /* Add the 3 callbacks to the list */ 4569 for (i = 0; i < 3; i++) { 4570 if (tmp_cb == NULL) 4571 list_insert_head(&zcl.zcl_callbacks, cb_data[i]); 4572 else 4573 list_insert_after(&zcl.zcl_callbacks, tmp_cb, 4574 cb_data[i]); 4575 4576 cb_data[i]->zcd_added = B_TRUE; 4577 VERIFY(!cb_data[i]->zcd_called); 4578 4579 tmp_cb = cb_data[i]; 4580 } 4581 4582 (void) mutex_unlock(&zcl.zcl_callbacks_lock); 4583 4584 dmu_tx_commit(tx); 4585} 4586 4587/* ARGSUSED */ 4588void 4589ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id) 4590{ 4591 zfs_prop_t proplist[] = { 4592 ZFS_PROP_CHECKSUM, 4593 ZFS_PROP_COMPRESSION, 4594 ZFS_PROP_COPIES, 4595 ZFS_PROP_DEDUP 4596 }; 4597 4598 (void) rw_rdlock(&ztest_name_lock); 4599 4600 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) 4601 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p], 4602 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2)); 4603 4604 (void) rw_unlock(&ztest_name_lock); 4605} 4606 4607/* ARGSUSED */ 4608void 4609ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id) 4610{ 4611 nvlist_t *props = NULL; 4612 4613 (void) rw_rdlock(&ztest_name_lock); 4614 4615 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO, 4616 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN)); 4617 4618 VERIFY0(spa_prop_get(ztest_spa, &props)); 4619 4620 if (ztest_opts.zo_verbose >= 6) 4621 dump_nvlist(props, 4); 4622 4623 nvlist_free(props); 4624 4625 (void) rw_unlock(&ztest_name_lock); 4626} 4627 4628static int 4629user_release_one(const char *snapname, const char *holdname) 4630{ 4631 nvlist_t *snaps, *holds; 4632 int error; 4633 4634 snaps = fnvlist_alloc(); 4635 holds = fnvlist_alloc(); 4636 fnvlist_add_boolean(holds, holdname); 4637 fnvlist_add_nvlist(snaps, snapname, holds); 4638 fnvlist_free(holds); 4639 error = dsl_dataset_user_release(snaps, NULL); 4640 fnvlist_free(snaps); 4641 return (error); 4642} 4643 4644/* 4645 * Test snapshot hold/release and deferred destroy. 4646 */ 4647void 4648ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) 4649{ 4650 int error; 4651 objset_t *os = zd->zd_os; 4652 objset_t *origin; 4653 char snapname[100]; 4654 char fullname[100]; 4655 char clonename[100]; 4656 char tag[100]; 4657 char osname[MAXNAMELEN]; 4658 nvlist_t *holds; 4659 4660 (void) rw_rdlock(&ztest_name_lock); 4661 4662 dmu_objset_name(os, osname); 4663 4664 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id); 4665 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname); 4666 (void) snprintf(clonename, sizeof (clonename), 4667 "%s/ch1_%llu", osname, id); 4668 (void) snprintf(tag, sizeof (tag), "tag_%llu", id); 4669 4670 /* 4671 * Clean up from any previous run. 4672 */ 4673 error = dsl_destroy_head(clonename); 4674 if (error != ENOENT) 4675 ASSERT0(error); 4676 error = user_release_one(fullname, tag); 4677 if (error != ESRCH && error != ENOENT) 4678 ASSERT0(error); 4679 error = dsl_destroy_snapshot(fullname, B_FALSE); 4680 if (error != ENOENT) 4681 ASSERT0(error); 4682 4683 /* 4684 * Create snapshot, clone it, mark snap for deferred destroy, 4685 * destroy clone, verify snap was also destroyed. 4686 */ 4687 error = dmu_objset_snapshot_one(osname, snapname); 4688 if (error) { 4689 if (error == ENOSPC) { 4690 ztest_record_enospc("dmu_objset_snapshot"); 4691 goto out; 4692 } 4693 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); 4694 } 4695 4696 error = dmu_objset_clone(clonename, fullname); 4697 if (error) { 4698 if (error == ENOSPC) { 4699 ztest_record_enospc("dmu_objset_clone"); 4700 goto out; 4701 } 4702 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error); 4703 } 4704 4705 error = dsl_destroy_snapshot(fullname, B_TRUE); 4706 if (error) { 4707 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", 4708 fullname, error); 4709 } 4710 4711 error = dsl_destroy_head(clonename); 4712 if (error) 4713 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error); 4714 4715 error = dmu_objset_hold(fullname, FTAG, &origin); 4716 if (error != ENOENT) 4717 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); 4718 4719 /* 4720 * Create snapshot, add temporary hold, verify that we can't 4721 * destroy a held snapshot, mark for deferred destroy, 4722 * release hold, verify snapshot was destroyed. 4723 */ 4724 error = dmu_objset_snapshot_one(osname, snapname); 4725 if (error) { 4726 if (error == ENOSPC) { 4727 ztest_record_enospc("dmu_objset_snapshot"); 4728 goto out; 4729 } 4730 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); 4731 } 4732 4733 holds = fnvlist_alloc(); 4734 fnvlist_add_string(holds, fullname, tag); 4735 error = dsl_dataset_user_hold(holds, 0, NULL); 4736 fnvlist_free(holds); 4737 4738 if (error == ENOSPC) { 4739 ztest_record_enospc("dsl_dataset_user_hold"); 4740 goto out; 4741 } else if (error) { 4742 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u", 4743 fullname, tag, error); 4744 } 4745 4746 error = dsl_destroy_snapshot(fullname, B_FALSE); 4747 if (error != EBUSY) { 4748 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d", 4749 fullname, error); 4750 } 4751 4752 error = dsl_destroy_snapshot(fullname, B_TRUE); 4753 if (error) { 4754 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", 4755 fullname, error); 4756 } 4757 4758 error = user_release_one(fullname, tag); 4759 if (error) 4760 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error); 4761 4762 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT); 4763 4764out: 4765 (void) rw_unlock(&ztest_name_lock); 4766} 4767 4768/* 4769 * Inject random faults into the on-disk data. 4770 */ 4771/* ARGSUSED */ 4772void 4773ztest_fault_inject(ztest_ds_t *zd, uint64_t id) 4774{ 4775 ztest_shared_t *zs = ztest_shared; 4776 spa_t *spa = ztest_spa; 4777 int fd; 4778 uint64_t offset; 4779 uint64_t leaves; 4780 uint64_t bad = 0x1990c0ffeedecadeULL; 4781 uint64_t top, leaf; 4782 char path0[MAXPATHLEN]; 4783 char pathrand[MAXPATHLEN]; 4784 size_t fsize; 4785 int bshift = SPA_OLD_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ 4786 int iters = 1000; 4787 int maxfaults; 4788 int mirror_save; 4789 vdev_t *vd0 = NULL; 4790 uint64_t guid0 = 0; 4791 boolean_t islog = B_FALSE; 4792 4793 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 4794 maxfaults = MAXFAULTS(); 4795 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; 4796 mirror_save = zs->zs_mirrors; 4797 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4798 4799 ASSERT(leaves >= 1); 4800 4801 /* 4802 * Grab the name lock as reader. There are some operations 4803 * which don't like to have their vdevs changed while 4804 * they are in progress (i.e. spa_change_guid). Those 4805 * operations will have grabbed the name lock as writer. 4806 */ 4807 (void) rw_rdlock(&ztest_name_lock); 4808 4809 /* 4810 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd. 4811 */ 4812 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 4813 4814 if (ztest_random(2) == 0) { 4815 /* 4816 * Inject errors on a normal data device or slog device. 4817 */ 4818 top = ztest_random_vdev_top(spa, B_TRUE); 4819 leaf = ztest_random(leaves) + zs->zs_splits; 4820 4821 /* 4822 * Generate paths to the first leaf in this top-level vdev, 4823 * and to the random leaf we selected. We'll induce transient 4824 * write failures and random online/offline activity on leaf 0, 4825 * and we'll write random garbage to the randomly chosen leaf. 4826 */ 4827 (void) snprintf(path0, sizeof (path0), ztest_dev_template, 4828 ztest_opts.zo_dir, ztest_opts.zo_pool, 4829 top * leaves + zs->zs_splits); 4830 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template, 4831 ztest_opts.zo_dir, ztest_opts.zo_pool, 4832 top * leaves + leaf); 4833 4834 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); 4835 if (vd0 != NULL && vd0->vdev_top->vdev_islog) 4836 islog = B_TRUE; 4837 4838 /* 4839 * If the top-level vdev needs to be resilvered 4840 * then we only allow faults on the device that is 4841 * resilvering. 4842 */ 4843 if (vd0 != NULL && maxfaults != 1 && 4844 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) || 4845 vd0->vdev_resilver_txg != 0)) { 4846 /* 4847 * Make vd0 explicitly claim to be unreadable, 4848 * or unwriteable, or reach behind its back 4849 * and close the underlying fd. We can do this if 4850 * maxfaults == 0 because we'll fail and reexecute, 4851 * and we can do it if maxfaults >= 2 because we'll 4852 * have enough redundancy. If maxfaults == 1, the 4853 * combination of this with injection of random data 4854 * corruption below exceeds the pool's fault tolerance. 4855 */ 4856 vdev_file_t *vf = vd0->vdev_tsd; 4857 4858 if (vf != NULL && ztest_random(3) == 0) { 4859 (void) close(vf->vf_vnode->v_fd); 4860 vf->vf_vnode->v_fd = -1; 4861 } else if (ztest_random(2) == 0) { 4862 vd0->vdev_cant_read = B_TRUE; 4863 } else { 4864 vd0->vdev_cant_write = B_TRUE; 4865 } 4866 guid0 = vd0->vdev_guid; 4867 } 4868 } else { 4869 /* 4870 * Inject errors on an l2cache device. 4871 */ 4872 spa_aux_vdev_t *sav = &spa->spa_l2cache; 4873 4874 if (sav->sav_count == 0) { 4875 spa_config_exit(spa, SCL_STATE, FTAG); 4876 (void) rw_unlock(&ztest_name_lock); 4877 return; 4878 } 4879 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)]; 4880 guid0 = vd0->vdev_guid; 4881 (void) strcpy(path0, vd0->vdev_path); 4882 (void) strcpy(pathrand, vd0->vdev_path); 4883 4884 leaf = 0; 4885 leaves = 1; 4886 maxfaults = INT_MAX; /* no limit on cache devices */ 4887 } 4888 4889 spa_config_exit(spa, SCL_STATE, FTAG); 4890 (void) rw_unlock(&ztest_name_lock); 4891 4892 /* 4893 * If we can tolerate two or more faults, or we're dealing 4894 * with a slog, randomly online/offline vd0. 4895 */ 4896 if ((maxfaults >= 2 || islog) && guid0 != 0) { 4897 if (ztest_random(10) < 6) { 4898 int flags = (ztest_random(2) == 0 ? 4899 ZFS_OFFLINE_TEMPORARY : 0); 4900 4901 /* 4902 * We have to grab the zs_name_lock as writer to 4903 * prevent a race between offlining a slog and 4904 * destroying a dataset. Offlining the slog will 4905 * grab a reference on the dataset which may cause 4906 * dmu_objset_destroy() to fail with EBUSY thus 4907 * leaving the dataset in an inconsistent state. 4908 */ 4909 if (islog) 4910 (void) rw_wrlock(&ztest_name_lock); 4911 4912 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY); 4913 4914 if (islog) 4915 (void) rw_unlock(&ztest_name_lock); 4916 } else { 4917 /* 4918 * Ideally we would like to be able to randomly 4919 * call vdev_[on|off]line without holding locks 4920 * to force unpredictable failures but the side 4921 * effects of vdev_[on|off]line prevent us from 4922 * doing so. We grab the ztest_vdev_lock here to 4923 * prevent a race between injection testing and 4924 * aux_vdev removal. 4925 */ 4926 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 4927 (void) vdev_online(spa, guid0, 0, NULL); 4928 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4929 } 4930 } 4931 4932 if (maxfaults == 0) 4933 return; 4934 4935 /* 4936 * We have at least single-fault tolerance, so inject data corruption. 4937 */ 4938 fd = open(pathrand, O_RDWR); 4939 4940 if (fd == -1) /* we hit a gap in the device namespace */ 4941 return; 4942 4943 fsize = lseek(fd, 0, SEEK_END); 4944 4945 while (--iters != 0) { 4946 offset = ztest_random(fsize / (leaves << bshift)) * 4947 (leaves << bshift) + (leaf << bshift) + 4948 (ztest_random(1ULL << (bshift - 1)) & -8ULL); 4949 4950 if (offset >= fsize) 4951 continue; 4952 4953 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 4954 if (mirror_save != zs->zs_mirrors) { 4955 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4956 (void) close(fd); 4957 return; 4958 } 4959 4960 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) 4961 fatal(1, "can't inject bad word at 0x%llx in %s", 4962 offset, pathrand); 4963 4964 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4965 4966 if (ztest_opts.zo_verbose >= 7) 4967 (void) printf("injected bad word into %s," 4968 " offset 0x%llx\n", pathrand, (u_longlong_t)offset); 4969 } 4970 4971 (void) close(fd); 4972} 4973 4974/* 4975 * Verify that DDT repair works as expected. 4976 */ 4977void 4978ztest_ddt_repair(ztest_ds_t *zd, uint64_t id) 4979{ 4980 ztest_shared_t *zs = ztest_shared; 4981 spa_t *spa = ztest_spa; 4982 objset_t *os = zd->zd_os; 4983 ztest_od_t od[1]; 4984 uint64_t object, blocksize, txg, pattern, psize; 4985 enum zio_checksum checksum = spa_dedup_checksum(spa); 4986 dmu_buf_t *db; 4987 dmu_tx_t *tx; 4988 void *buf; 4989 blkptr_t blk; 4990 int copies = 2 * ZIO_DEDUPDITTO_MIN; 4991 4992 blocksize = ztest_random_blocksize(); 4993 blocksize = MIN(blocksize, 2048); /* because we write so many */ 4994 4995 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 4996 4997 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4998 return; 4999 5000 /* 5001 * Take the name lock as writer to prevent anyone else from changing 5002 * the pool and dataset properies we need to maintain during this test. 5003 */ 5004 (void) rw_wrlock(&ztest_name_lock); 5005 5006 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum, 5007 B_FALSE) != 0 || 5008 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1, 5009 B_FALSE) != 0) { 5010 (void) rw_unlock(&ztest_name_lock); 5011 return; 5012 } 5013 5014 object = od[0].od_object; 5015 blocksize = od[0].od_blocksize; 5016 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os); 5017 5018 ASSERT(object != 0); 5019 5020 tx = dmu_tx_create(os); 5021 dmu_tx_hold_write(tx, object, 0, copies * blocksize); 5022 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 5023 if (txg == 0) { 5024 (void) rw_unlock(&ztest_name_lock); 5025 return; 5026 } 5027 5028 /* 5029 * Write all the copies of our block. 5030 */ 5031 for (int i = 0; i < copies; i++) { 5032 uint64_t offset = i * blocksize; 5033 int error = dmu_buf_hold(os, object, offset, FTAG, &db, 5034 DMU_READ_NO_PREFETCH); 5035 if (error != 0) { 5036 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u", 5037 os, (long long)object, (long long) offset, error); 5038 } 5039 ASSERT(db->db_offset == offset); 5040 ASSERT(db->db_size == blocksize); 5041 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) || 5042 ztest_pattern_match(db->db_data, db->db_size, 0ULL)); 5043 dmu_buf_will_fill(db, tx); 5044 ztest_pattern_set(db->db_data, db->db_size, pattern); 5045 dmu_buf_rele(db, FTAG); 5046 } 5047 5048 dmu_tx_commit(tx); 5049 txg_wait_synced(spa_get_dsl(spa), txg); 5050 5051 /* 5052 * Find out what block we got. 5053 */ 5054 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db, 5055 DMU_READ_NO_PREFETCH)); 5056 blk = *((dmu_buf_impl_t *)db)->db_blkptr; 5057 dmu_buf_rele(db, FTAG); 5058 5059 /* 5060 * Damage the block. Dedup-ditto will save us when we read it later. 5061 */ 5062 psize = BP_GET_PSIZE(&blk); 5063 buf = zio_buf_alloc(psize); 5064 ztest_pattern_set(buf, psize, ~pattern); 5065 5066 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk, 5067 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE, 5068 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL)); 5069 5070 zio_buf_free(buf, psize); 5071 5072 (void) rw_unlock(&ztest_name_lock); 5073} 5074 5075/* 5076 * Scrub the pool. 5077 */ 5078/* ARGSUSED */ 5079void 5080ztest_scrub(ztest_ds_t *zd, uint64_t id) 5081{ 5082 spa_t *spa = ztest_spa; 5083 5084 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5085 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */ 5086 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5087} 5088 5089/* 5090 * Change the guid for the pool. 5091 */ 5092/* ARGSUSED */ 5093void 5094ztest_reguid(ztest_ds_t *zd, uint64_t id) 5095{ 5096 spa_t *spa = ztest_spa; 5097 uint64_t orig, load; 5098 int error; 5099 5100 orig = spa_guid(spa); 5101 load = spa_load_guid(spa); 5102 5103 (void) rw_wrlock(&ztest_name_lock); 5104 error = spa_change_guid(spa); 5105 (void) rw_unlock(&ztest_name_lock); 5106 5107 if (error != 0) 5108 return; 5109 5110 if (ztest_opts.zo_verbose >= 4) { 5111 (void) printf("Changed guid old %llu -> %llu\n", 5112 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa)); 5113 } 5114 5115 VERIFY3U(orig, !=, spa_guid(spa)); 5116 VERIFY3U(load, ==, spa_load_guid(spa)); 5117} 5118 5119/* 5120 * Rename the pool to a different name and then rename it back. 5121 */ 5122/* ARGSUSED */ 5123void 5124ztest_spa_rename(ztest_ds_t *zd, uint64_t id) 5125{ 5126 char *oldname, *newname; 5127 spa_t *spa; 5128 5129 (void) rw_wrlock(&ztest_name_lock); 5130 5131 oldname = ztest_opts.zo_pool; 5132 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); 5133 (void) strcpy(newname, oldname); 5134 (void) strcat(newname, "_tmp"); 5135 5136 /* 5137 * Do the rename 5138 */ 5139 VERIFY3U(0, ==, spa_rename(oldname, newname)); 5140 5141 /* 5142 * Try to open it under the old name, which shouldn't exist 5143 */ 5144 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); 5145 5146 /* 5147 * Open it under the new name and make sure it's still the same spa_t. 5148 */ 5149 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); 5150 5151 ASSERT(spa == ztest_spa); 5152 spa_close(spa, FTAG); 5153 5154 /* 5155 * Rename it back to the original 5156 */ 5157 VERIFY3U(0, ==, spa_rename(newname, oldname)); 5158 5159 /* 5160 * Make sure it can still be opened 5161 */ 5162 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); 5163 5164 ASSERT(spa == ztest_spa); 5165 spa_close(spa, FTAG); 5166 5167 umem_free(newname, strlen(newname) + 1); 5168 5169 (void) rw_unlock(&ztest_name_lock); 5170} 5171 5172/* 5173 * Verify pool integrity by running zdb. 5174 */ 5175static void 5176ztest_run_zdb(char *pool) 5177{ 5178 int status; 5179 char zdb[MAXPATHLEN + MAXNAMELEN + 20]; 5180 char zbuf[1024]; 5181 char *bin; 5182 char *ztest; 5183 char *isa; 5184 int isalen; 5185 FILE *fp; 5186 5187 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb)); 5188 5189 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ 5190 bin = strstr(zdb, "/usr/bin/"); 5191 ztest = strstr(bin, "/ztest"); 5192 isa = bin + 8; 5193 isalen = ztest - isa; 5194 isa = strdup(isa); 5195 /* LINTED */ 5196 (void) sprintf(bin, 5197 "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s", 5198 isalen, 5199 isa, 5200 ztest_opts.zo_verbose >= 3 ? "s" : "", 5201 ztest_opts.zo_verbose >= 4 ? "v" : "", 5202 spa_config_path, 5203 pool); 5204 free(isa); 5205 5206 if (ztest_opts.zo_verbose >= 5) 5207 (void) printf("Executing %s\n", strstr(zdb, "zdb ")); 5208 5209 fp = popen(zdb, "r"); 5210 assert(fp != NULL); 5211 5212 while (fgets(zbuf, sizeof (zbuf), fp) != NULL) 5213 if (ztest_opts.zo_verbose >= 3) 5214 (void) printf("%s", zbuf); 5215 5216 status = pclose(fp); 5217 5218 if (status == 0) 5219 return; 5220 5221 ztest_dump_core = 0; 5222 if (WIFEXITED(status)) 5223 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); 5224 else 5225 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); 5226} 5227 5228static void 5229ztest_walk_pool_directory(char *header) 5230{ 5231 spa_t *spa = NULL; 5232 5233 if (ztest_opts.zo_verbose >= 6) 5234 (void) printf("%s\n", header); 5235 5236 mutex_enter(&spa_namespace_lock); 5237 while ((spa = spa_next(spa)) != NULL) 5238 if (ztest_opts.zo_verbose >= 6) 5239 (void) printf("\t%s\n", spa_name(spa)); 5240 mutex_exit(&spa_namespace_lock); 5241} 5242 5243static void 5244ztest_spa_import_export(char *oldname, char *newname) 5245{ 5246 nvlist_t *config, *newconfig; 5247 uint64_t pool_guid; 5248 spa_t *spa; 5249 int error; 5250 5251 if (ztest_opts.zo_verbose >= 4) { 5252 (void) printf("import/export: old = %s, new = %s\n", 5253 oldname, newname); 5254 } 5255 5256 /* 5257 * Clean up from previous runs. 5258 */ 5259 (void) spa_destroy(newname); 5260 5261 /* 5262 * Get the pool's configuration and guid. 5263 */ 5264 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); 5265 5266 /* 5267 * Kick off a scrub to tickle scrub/export races. 5268 */ 5269 if (ztest_random(2) == 0) 5270 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5271 5272 pool_guid = spa_guid(spa); 5273 spa_close(spa, FTAG); 5274 5275 ztest_walk_pool_directory("pools before export"); 5276 5277 /* 5278 * Export it. 5279 */ 5280 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE)); 5281 5282 ztest_walk_pool_directory("pools after export"); 5283 5284 /* 5285 * Try to import it. 5286 */ 5287 newconfig = spa_tryimport(config); 5288 ASSERT(newconfig != NULL); 5289 nvlist_free(newconfig); 5290 5291 /* 5292 * Import it under the new name. 5293 */ 5294 error = spa_import(newname, config, NULL, 0); 5295 if (error != 0) { 5296 dump_nvlist(config, 0); 5297 fatal(B_FALSE, "couldn't import pool %s as %s: error %u", 5298 oldname, newname, error); 5299 } 5300 5301 ztest_walk_pool_directory("pools after import"); 5302 5303 /* 5304 * Try to import it again -- should fail with EEXIST. 5305 */ 5306 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0)); 5307 5308 /* 5309 * Try to import it under a different name -- should fail with EEXIST. 5310 */ 5311 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0)); 5312 5313 /* 5314 * Verify that the pool is no longer visible under the old name. 5315 */ 5316 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); 5317 5318 /* 5319 * Verify that we can open and close the pool using the new name. 5320 */ 5321 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); 5322 ASSERT(pool_guid == spa_guid(spa)); 5323 spa_close(spa, FTAG); 5324 5325 nvlist_free(config); 5326} 5327 5328static void 5329ztest_resume(spa_t *spa) 5330{ 5331 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6) 5332 (void) printf("resuming from suspended state\n"); 5333 spa_vdev_state_enter(spa, SCL_NONE); 5334 vdev_clear(spa, NULL); 5335 (void) spa_vdev_state_exit(spa, NULL, 0); 5336 (void) zio_resume(spa); 5337} 5338 5339static void * 5340ztest_resume_thread(void *arg) 5341{ 5342 spa_t *spa = arg; 5343 5344 while (!ztest_exiting) { 5345 if (spa_suspended(spa)) 5346 ztest_resume(spa); 5347 (void) poll(NULL, 0, 100); 5348 } 5349 return (NULL); 5350} 5351 5352static void * 5353ztest_deadman_thread(void *arg) 5354{ 5355 ztest_shared_t *zs = arg; 5356 spa_t *spa = ztest_spa; 5357 hrtime_t delta, total = 0; 5358 5359 for (;;) { 5360 delta = zs->zs_thread_stop - zs->zs_thread_start + 5361 MSEC2NSEC(zfs_deadman_synctime_ms); 5362 5363 (void) poll(NULL, 0, (int)NSEC2MSEC(delta)); 5364 5365 /* 5366 * If the pool is suspended then fail immediately. Otherwise, 5367 * check to see if the pool is making any progress. If 5368 * vdev_deadman() discovers that there hasn't been any recent 5369 * I/Os then it will end up aborting the tests. 5370 */ 5371 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) { 5372 fatal(0, "aborting test after %llu seconds because " 5373 "pool has transitioned to a suspended state.", 5374 zfs_deadman_synctime_ms / 1000); 5375 return (NULL); 5376 } 5377 vdev_deadman(spa->spa_root_vdev); 5378 5379 total += zfs_deadman_synctime_ms/1000; 5380 (void) printf("ztest has been running for %lld seconds\n", 5381 total); 5382 } 5383} 5384 5385static void 5386ztest_execute(int test, ztest_info_t *zi, uint64_t id) 5387{ 5388 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets]; 5389 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test); 5390 hrtime_t functime = gethrtime(); 5391 5392 for (int i = 0; i < zi->zi_iters; i++) 5393 zi->zi_func(zd, id); 5394 5395 functime = gethrtime() - functime; 5396 5397 atomic_add_64(&zc->zc_count, 1); 5398 atomic_add_64(&zc->zc_time, functime); 5399 5400 if (ztest_opts.zo_verbose >= 4) { 5401 Dl_info dli; 5402 (void) dladdr((void *)zi->zi_func, &dli); 5403 (void) printf("%6.2f sec in %s\n", 5404 (double)functime / NANOSEC, dli.dli_sname); 5405 } 5406} 5407 5408static void * 5409ztest_thread(void *arg) 5410{ 5411 int rand; 5412 uint64_t id = (uintptr_t)arg; 5413 ztest_shared_t *zs = ztest_shared; 5414 uint64_t call_next; 5415 hrtime_t now; 5416 ztest_info_t *zi; 5417 ztest_shared_callstate_t *zc; 5418 5419 while ((now = gethrtime()) < zs->zs_thread_stop) { 5420 /* 5421 * See if it's time to force a crash. 5422 */ 5423 if (now > zs->zs_thread_kill) 5424 ztest_kill(zs); 5425 5426 /* 5427 * If we're getting ENOSPC with some regularity, stop. 5428 */ 5429 if (zs->zs_enospc_count > 10) 5430 break; 5431 5432 /* 5433 * Pick a random function to execute. 5434 */ 5435 rand = ztest_random(ZTEST_FUNCS); 5436 zi = &ztest_info[rand]; 5437 zc = ZTEST_GET_SHARED_CALLSTATE(rand); 5438 call_next = zc->zc_next; 5439 5440 if (now >= call_next && 5441 atomic_cas_64(&zc->zc_next, call_next, call_next + 5442 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) { 5443 ztest_execute(rand, zi, id); 5444 } 5445 } 5446 5447 return (NULL); 5448} 5449 5450static void 5451ztest_dataset_name(char *dsname, char *pool, int d) 5452{ 5453 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d); 5454} 5455 5456static void 5457ztest_dataset_destroy(int d) 5458{ 5459 char name[MAXNAMELEN]; 5460 5461 ztest_dataset_name(name, ztest_opts.zo_pool, d); 5462 5463 if (ztest_opts.zo_verbose >= 3) 5464 (void) printf("Destroying %s to free up space\n", name); 5465 5466 /* 5467 * Cleanup any non-standard clones and snapshots. In general, 5468 * ztest thread t operates on dataset (t % zopt_datasets), 5469 * so there may be more than one thing to clean up. 5470 */ 5471 for (int t = d; t < ztest_opts.zo_threads; 5472 t += ztest_opts.zo_datasets) { 5473 ztest_dsl_dataset_cleanup(name, t); 5474 } 5475 5476 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, 5477 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); 5478} 5479 5480static void 5481ztest_dataset_dirobj_verify(ztest_ds_t *zd) 5482{ 5483 uint64_t usedobjs, dirobjs, scratch; 5484 5485 /* 5486 * ZTEST_DIROBJ is the object directory for the entire dataset. 5487 * Therefore, the number of objects in use should equal the 5488 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself. 5489 * If not, we have an object leak. 5490 * 5491 * Note that we can only check this in ztest_dataset_open(), 5492 * when the open-context and syncing-context values agree. 5493 * That's because zap_count() returns the open-context value, 5494 * while dmu_objset_space() returns the rootbp fill count. 5495 */ 5496 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs)); 5497 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch); 5498 ASSERT3U(dirobjs + 1, ==, usedobjs); 5499} 5500 5501static int 5502ztest_dataset_open(int d) 5503{ 5504 ztest_ds_t *zd = &ztest_ds[d]; 5505 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq; 5506 objset_t *os; 5507 zilog_t *zilog; 5508 char name[MAXNAMELEN]; 5509 int error; 5510 5511 ztest_dataset_name(name, ztest_opts.zo_pool, d); 5512 5513 (void) rw_rdlock(&ztest_name_lock); 5514 5515 error = ztest_dataset_create(name); 5516 if (error == ENOSPC) { 5517 (void) rw_unlock(&ztest_name_lock); 5518 ztest_record_enospc(FTAG); 5519 return (error); 5520 } 5521 ASSERT(error == 0 || error == EEXIST); 5522 5523 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os)); 5524 (void) rw_unlock(&ztest_name_lock); 5525 5526 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os); 5527 5528 zilog = zd->zd_zilog; 5529 5530 if (zilog->zl_header->zh_claim_lr_seq != 0 && 5531 zilog->zl_header->zh_claim_lr_seq < committed_seq) 5532 fatal(0, "missing log records: claimed %llu < committed %llu", 5533 zilog->zl_header->zh_claim_lr_seq, committed_seq); 5534 5535 ztest_dataset_dirobj_verify(zd); 5536 5537 zil_replay(os, zd, ztest_replay_vector); 5538 5539 ztest_dataset_dirobj_verify(zd); 5540 5541 if (ztest_opts.zo_verbose >= 6) 5542 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n", 5543 zd->zd_name, 5544 (u_longlong_t)zilog->zl_parse_blk_count, 5545 (u_longlong_t)zilog->zl_parse_lr_count, 5546 (u_longlong_t)zilog->zl_replaying_seq); 5547 5548 zilog = zil_open(os, ztest_get_data); 5549 5550 if (zilog->zl_replaying_seq != 0 && 5551 zilog->zl_replaying_seq < committed_seq) 5552 fatal(0, "missing log records: replayed %llu < committed %llu", 5553 zilog->zl_replaying_seq, committed_seq); 5554 5555 return (0); 5556} 5557 5558static void 5559ztest_dataset_close(int d) 5560{ 5561 ztest_ds_t *zd = &ztest_ds[d]; 5562 5563 zil_close(zd->zd_zilog); 5564 dmu_objset_disown(zd->zd_os, zd); 5565 5566 ztest_zd_fini(zd); 5567} 5568 5569/* 5570 * Kick off threads to run tests on all datasets in parallel. 5571 */ 5572static void 5573ztest_run(ztest_shared_t *zs) 5574{ 5575 thread_t *tid; 5576 spa_t *spa; 5577 objset_t *os; 5578 thread_t resume_tid; 5579 int error; 5580 5581 ztest_exiting = B_FALSE; 5582 5583 /* 5584 * Initialize parent/child shared state. 5585 */ 5586 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0); 5587 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0); 5588 5589 zs->zs_thread_start = gethrtime(); 5590 zs->zs_thread_stop = 5591 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC; 5592 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop); 5593 zs->zs_thread_kill = zs->zs_thread_stop; 5594 if (ztest_random(100) < ztest_opts.zo_killrate) { 5595 zs->zs_thread_kill -= 5596 ztest_random(ztest_opts.zo_passtime * NANOSEC); 5597 } 5598 5599 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL); 5600 5601 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t), 5602 offsetof(ztest_cb_data_t, zcd_node)); 5603 5604 /* 5605 * Open our pool. 5606 */ 5607 kernel_init(FREAD | FWRITE); 5608 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5609 spa->spa_debug = B_TRUE; 5610 metaslab_preload_limit = ztest_random(20) + 1; 5611 ztest_spa = spa; 5612 5613 VERIFY0(dmu_objset_own(ztest_opts.zo_pool, 5614 DMU_OST_ANY, B_TRUE, FTAG, &os)); 5615 zs->zs_guid = dmu_objset_fsid_guid(os); 5616 dmu_objset_disown(os, FTAG); 5617 5618 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN; 5619 5620 /* 5621 * We don't expect the pool to suspend unless maxfaults == 0, 5622 * in which case ztest_fault_inject() temporarily takes away 5623 * the only valid replica. 5624 */ 5625 if (MAXFAULTS() == 0) 5626 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT; 5627 else 5628 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC; 5629 5630 /* 5631 * Create a thread to periodically resume suspended I/O. 5632 */ 5633 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND, 5634 &resume_tid) == 0); 5635 5636 /* 5637 * Create a deadman thread to abort() if we hang. 5638 */ 5639 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND, 5640 NULL) == 0); 5641 5642 /* 5643 * Verify that we can safely inquire about about any object, 5644 * whether it's allocated or not. To make it interesting, 5645 * we probe a 5-wide window around each power of two. 5646 * This hits all edge cases, including zero and the max. 5647 */ 5648 for (int t = 0; t < 64; t++) { 5649 for (int d = -5; d <= 5; d++) { 5650 error = dmu_object_info(spa->spa_meta_objset, 5651 (1ULL << t) + d, NULL); 5652 ASSERT(error == 0 || error == ENOENT || 5653 error == EINVAL); 5654 } 5655 } 5656 5657 /* 5658 * If we got any ENOSPC errors on the previous run, destroy something. 5659 */ 5660 if (zs->zs_enospc_count != 0) { 5661 int d = ztest_random(ztest_opts.zo_datasets); 5662 ztest_dataset_destroy(d); 5663 } 5664 zs->zs_enospc_count = 0; 5665 5666 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t), 5667 UMEM_NOFAIL); 5668 5669 if (ztest_opts.zo_verbose >= 4) 5670 (void) printf("starting main threads...\n"); 5671 5672 /* 5673 * Kick off all the tests that run in parallel. 5674 */ 5675 for (int t = 0; t < ztest_opts.zo_threads; t++) { 5676 if (t < ztest_opts.zo_datasets && 5677 ztest_dataset_open(t) != 0) 5678 return; 5679 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t, 5680 THR_BOUND, &tid[t]) == 0); 5681 } 5682 5683 /* 5684 * Wait for all of the tests to complete. We go in reverse order 5685 * so we don't close datasets while threads are still using them. 5686 */ 5687 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) { 5688 VERIFY(thr_join(tid[t], NULL, NULL) == 0); 5689 if (t < ztest_opts.zo_datasets) 5690 ztest_dataset_close(t); 5691 } 5692 5693 txg_wait_synced(spa_get_dsl(spa), 0); 5694 5695 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 5696 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa)); 5697 zfs_dbgmsg_print(FTAG); 5698 5699 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t)); 5700 5701 /* Kill the resume thread */ 5702 ztest_exiting = B_TRUE; 5703 VERIFY(thr_join(resume_tid, NULL, NULL) == 0); 5704 ztest_resume(spa); 5705 5706 /* 5707 * Right before closing the pool, kick off a bunch of async I/O; 5708 * spa_close() should wait for it to complete. 5709 */ 5710 for (uint64_t object = 1; object < 50; object++) { 5711 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20, 5712 ZIO_PRIORITY_SYNC_READ); 5713 } 5714 5715 spa_close(spa, FTAG); 5716 5717 /* 5718 * Verify that we can loop over all pools. 5719 */ 5720 mutex_enter(&spa_namespace_lock); 5721 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) 5722 if (ztest_opts.zo_verbose > 3) 5723 (void) printf("spa_next: found %s\n", spa_name(spa)); 5724 mutex_exit(&spa_namespace_lock); 5725 5726 /* 5727 * Verify that we can export the pool and reimport it under a 5728 * different name. 5729 */ 5730 if (ztest_random(2) == 0) { 5731 char name[MAXNAMELEN]; 5732 (void) snprintf(name, MAXNAMELEN, "%s_import", 5733 ztest_opts.zo_pool); 5734 ztest_spa_import_export(ztest_opts.zo_pool, name); 5735 ztest_spa_import_export(name, ztest_opts.zo_pool); 5736 } 5737 5738 kernel_fini(); 5739 5740 list_destroy(&zcl.zcl_callbacks); 5741 5742 (void) _mutex_destroy(&zcl.zcl_callbacks_lock); 5743 5744 (void) rwlock_destroy(&ztest_name_lock); 5745 (void) _mutex_destroy(&ztest_vdev_lock); 5746} 5747 5748static void 5749ztest_freeze(void) 5750{ 5751 ztest_ds_t *zd = &ztest_ds[0]; 5752 spa_t *spa; 5753 int numloops = 0; 5754 5755 if (ztest_opts.zo_verbose >= 3) 5756 (void) printf("testing spa_freeze()...\n"); 5757 5758 kernel_init(FREAD | FWRITE); 5759 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5760 VERIFY3U(0, ==, ztest_dataset_open(0)); 5761 spa->spa_debug = B_TRUE; 5762 ztest_spa = spa; 5763 5764 /* 5765 * Force the first log block to be transactionally allocated. 5766 * We have to do this before we freeze the pool -- otherwise 5767 * the log chain won't be anchored. 5768 */ 5769 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) { 5770 ztest_dmu_object_alloc_free(zd, 0); 5771 zil_commit(zd->zd_zilog, 0); 5772 } 5773 5774 txg_wait_synced(spa_get_dsl(spa), 0); 5775 5776 /* 5777 * Freeze the pool. This stops spa_sync() from doing anything, 5778 * so that the only way to record changes from now on is the ZIL. 5779 */ 5780 spa_freeze(spa); 5781 5782 /* 5783 * Because it is hard to predict how much space a write will actually 5784 * require beforehand, we leave ourselves some fudge space to write over 5785 * capacity. 5786 */ 5787 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2; 5788 5789 /* 5790 * Run tests that generate log records but don't alter the pool config 5791 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc). 5792 * We do a txg_wait_synced() after each iteration to force the txg 5793 * to increase well beyond the last synced value in the uberblock. 5794 * The ZIL should be OK with that. 5795 * 5796 * Run a random number of times less than zo_maxloops and ensure we do 5797 * not run out of space on the pool. 5798 */ 5799 while (ztest_random(10) != 0 && 5800 numloops++ < ztest_opts.zo_maxloops && 5801 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) { 5802 ztest_od_t od; 5803 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 5804 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE)); 5805 ztest_io(zd, od.od_object, 5806 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 5807 txg_wait_synced(spa_get_dsl(spa), 0); 5808 } 5809 5810 /* 5811 * Commit all of the changes we just generated. 5812 */ 5813 zil_commit(zd->zd_zilog, 0); 5814 txg_wait_synced(spa_get_dsl(spa), 0); 5815 5816 /* 5817 * Close our dataset and close the pool. 5818 */ 5819 ztest_dataset_close(0); 5820 spa_close(spa, FTAG); 5821 kernel_fini(); 5822 5823 /* 5824 * Open and close the pool and dataset to induce log replay. 5825 */ 5826 kernel_init(FREAD | FWRITE); 5827 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5828 ASSERT(spa_freeze_txg(spa) == UINT64_MAX); 5829 VERIFY3U(0, ==, ztest_dataset_open(0)); 5830 ztest_dataset_close(0); 5831 5832 spa->spa_debug = B_TRUE; 5833 ztest_spa = spa; 5834 txg_wait_synced(spa_get_dsl(spa), 0); 5835 ztest_reguid(NULL, 0); 5836 5837 spa_close(spa, FTAG); 5838 kernel_fini(); 5839} 5840 5841void 5842print_time(hrtime_t t, char *timebuf) 5843{ 5844 hrtime_t s = t / NANOSEC; 5845 hrtime_t m = s / 60; 5846 hrtime_t h = m / 60; 5847 hrtime_t d = h / 24; 5848 5849 s -= m * 60; 5850 m -= h * 60; 5851 h -= d * 24; 5852 5853 timebuf[0] = '\0'; 5854 5855 if (d) 5856 (void) sprintf(timebuf, 5857 "%llud%02lluh%02llum%02llus", d, h, m, s); 5858 else if (h) 5859 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); 5860 else if (m) 5861 (void) sprintf(timebuf, "%llum%02llus", m, s); 5862 else 5863 (void) sprintf(timebuf, "%llus", s); 5864} 5865 5866static nvlist_t * 5867make_random_props() 5868{ 5869 nvlist_t *props; 5870 5871 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); 5872 if (ztest_random(2) == 0) 5873 return (props); 5874 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0); 5875 5876 return (props); 5877} 5878 5879/* 5880 * Create a storage pool with the given name and initial vdev size. 5881 * Then test spa_freeze() functionality. 5882 */ 5883static void 5884ztest_init(ztest_shared_t *zs) 5885{ 5886 spa_t *spa; 5887 nvlist_t *nvroot, *props; 5888 5889 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0); 5890 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0); 5891 5892 kernel_init(FREAD | FWRITE); 5893 5894 /* 5895 * Create the storage pool. 5896 */ 5897 (void) spa_destroy(ztest_opts.zo_pool); 5898 ztest_shared->zs_vdev_next_leaf = 0; 5899 zs->zs_splits = 0; 5900 zs->zs_mirrors = ztest_opts.zo_mirrors; 5901 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0, 5902 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1); 5903 props = make_random_props(); 5904 for (int i = 0; i < SPA_FEATURES; i++) { 5905 char buf[1024]; 5906 (void) snprintf(buf, sizeof (buf), "feature@%s", 5907 spa_feature_table[i].fi_uname); 5908 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0)); 5909 } 5910 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL)); 5911 nvlist_free(nvroot); 5912 nvlist_free(props); 5913 5914 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5915 zs->zs_metaslab_sz = 5916 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift; 5917 5918 spa_close(spa, FTAG); 5919 5920 kernel_fini(); 5921 5922 ztest_run_zdb(ztest_opts.zo_pool); 5923 5924 ztest_freeze(); 5925 5926 ztest_run_zdb(ztest_opts.zo_pool); 5927 5928 (void) rwlock_destroy(&ztest_name_lock); 5929 (void) _mutex_destroy(&ztest_vdev_lock); 5930} 5931 5932static void 5933setup_data_fd(void) 5934{ 5935 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX"; 5936 5937 ztest_fd_data = mkstemp(ztest_name_data); 5938 ASSERT3S(ztest_fd_data, >=, 0); 5939 (void) unlink(ztest_name_data); 5940} 5941 5942 5943static int 5944shared_data_size(ztest_shared_hdr_t *hdr) 5945{ 5946 int size; 5947 5948 size = hdr->zh_hdr_size; 5949 size += hdr->zh_opts_size; 5950 size += hdr->zh_size; 5951 size += hdr->zh_stats_size * hdr->zh_stats_count; 5952 size += hdr->zh_ds_size * hdr->zh_ds_count; 5953 5954 return (size); 5955} 5956 5957static void 5958setup_hdr(void) 5959{ 5960 int size; 5961 ztest_shared_hdr_t *hdr; 5962 5963 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), 5964 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); 5965 ASSERT(hdr != MAP_FAILED); 5966 5967 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t))); 5968 5969 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t); 5970 hdr->zh_opts_size = sizeof (ztest_shared_opts_t); 5971 hdr->zh_size = sizeof (ztest_shared_t); 5972 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t); 5973 hdr->zh_stats_count = ZTEST_FUNCS; 5974 hdr->zh_ds_size = sizeof (ztest_shared_ds_t); 5975 hdr->zh_ds_count = ztest_opts.zo_datasets; 5976 5977 size = shared_data_size(hdr); 5978 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size)); 5979 5980 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); 5981} 5982 5983static void 5984setup_data(void) 5985{ 5986 int size, offset; 5987 ztest_shared_hdr_t *hdr; 5988 uint8_t *buf; 5989 5990 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), 5991 PROT_READ, MAP_SHARED, ztest_fd_data, 0); 5992 ASSERT(hdr != MAP_FAILED); 5993 5994 size = shared_data_size(hdr); 5995 5996 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); 5997 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()), 5998 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); 5999 ASSERT(hdr != MAP_FAILED); 6000 buf = (uint8_t *)hdr; 6001 6002 offset = hdr->zh_hdr_size; 6003 ztest_shared_opts = (void *)&buf[offset]; 6004 offset += hdr->zh_opts_size; 6005 ztest_shared = (void *)&buf[offset]; 6006 offset += hdr->zh_size; 6007 ztest_shared_callstate = (void *)&buf[offset]; 6008 offset += hdr->zh_stats_size * hdr->zh_stats_count; 6009 ztest_shared_ds = (void *)&buf[offset]; 6010} 6011 6012static boolean_t 6013exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp) 6014{ 6015 pid_t pid; 6016 int status; 6017 char *cmdbuf = NULL; 6018 6019 pid = fork(); 6020 6021 if (cmd == NULL) { 6022 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 6023 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN); 6024 cmd = cmdbuf; 6025 } 6026 6027 if (pid == -1) 6028 fatal(1, "fork failed"); 6029 6030 if (pid == 0) { /* child */ 6031 char *emptyargv[2] = { cmd, NULL }; 6032 char fd_data_str[12]; 6033 6034 struct rlimit rl = { 1024, 1024 }; 6035 (void) setrlimit(RLIMIT_NOFILE, &rl); 6036 6037 (void) close(ztest_fd_rand); 6038 VERIFY3U(11, >=, 6039 snprintf(fd_data_str, 12, "%d", ztest_fd_data)); 6040 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1)); 6041 6042 (void) enable_extended_FILE_stdio(-1, -1); 6043 if (libpath != NULL) 6044 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1)); 6045#ifdef illumos 6046 (void) execv(cmd, emptyargv); 6047#else 6048 (void) execvp(cmd, emptyargv); 6049#endif 6050 ztest_dump_core = B_FALSE; 6051 fatal(B_TRUE, "exec failed: %s", cmd); 6052 } 6053 6054 if (cmdbuf != NULL) { 6055 umem_free(cmdbuf, MAXPATHLEN); 6056 cmd = NULL; 6057 } 6058 6059 while (waitpid(pid, &status, 0) != pid) 6060 continue; 6061 if (statusp != NULL) 6062 *statusp = status; 6063 6064 if (WIFEXITED(status)) { 6065 if (WEXITSTATUS(status) != 0) { 6066 (void) fprintf(stderr, "child exited with code %d\n", 6067 WEXITSTATUS(status)); 6068 exit(2); 6069 } 6070 return (B_FALSE); 6071 } else if (WIFSIGNALED(status)) { 6072 if (!ignorekill || WTERMSIG(status) != SIGKILL) { 6073 (void) fprintf(stderr, "child died with signal %d\n", 6074 WTERMSIG(status)); 6075 exit(3); 6076 } 6077 return (B_TRUE); 6078 } else { 6079 (void) fprintf(stderr, "something strange happened to child\n"); 6080 exit(4); 6081 /* NOTREACHED */ 6082 } 6083} 6084 6085static void 6086ztest_run_init(void) 6087{ 6088 ztest_shared_t *zs = ztest_shared; 6089 6090 ASSERT(ztest_opts.zo_init != 0); 6091 6092 /* 6093 * Blow away any existing copy of zpool.cache 6094 */ 6095 (void) remove(spa_config_path); 6096 6097 /* 6098 * Create and initialize our storage pool. 6099 */ 6100 for (int i = 1; i <= ztest_opts.zo_init; i++) { 6101 bzero(zs, sizeof (ztest_shared_t)); 6102 if (ztest_opts.zo_verbose >= 3 && 6103 ztest_opts.zo_init != 1) { 6104 (void) printf("ztest_init(), pass %d\n", i); 6105 } 6106 ztest_init(zs); 6107 } 6108} 6109 6110int 6111main(int argc, char **argv) 6112{ 6113 int kills = 0; 6114 int iters = 0; 6115 int older = 0; 6116 int newer = 0; 6117 ztest_shared_t *zs; 6118 ztest_info_t *zi; 6119 ztest_shared_callstate_t *zc; 6120 char timebuf[100]; 6121 char numbuf[6]; 6122 spa_t *spa; 6123 char *cmd; 6124 boolean_t hasalt; 6125 char *fd_data_str = getenv("ZTEST_FD_DATA"); 6126 6127 (void) setvbuf(stdout, NULL, _IOLBF, 0); 6128 6129 dprintf_setup(&argc, argv); 6130 zfs_deadman_synctime_ms = 300000; 6131 6132 ztest_fd_rand = open("/dev/urandom", O_RDONLY); 6133 ASSERT3S(ztest_fd_rand, >=, 0); 6134 6135 if (!fd_data_str) { 6136 process_options(argc, argv); 6137 6138 setup_data_fd(); 6139 setup_hdr(); 6140 setup_data(); 6141 bcopy(&ztest_opts, ztest_shared_opts, 6142 sizeof (*ztest_shared_opts)); 6143 } else { 6144 ztest_fd_data = atoi(fd_data_str); 6145 setup_data(); 6146 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts)); 6147 } 6148 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count); 6149 6150 /* Override location of zpool.cache */ 6151 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache", 6152 ztest_opts.zo_dir), !=, -1); 6153 6154 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t), 6155 UMEM_NOFAIL); 6156 zs = ztest_shared; 6157 6158 if (fd_data_str) { 6159 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang; 6160 metaslab_df_alloc_threshold = 6161 zs->zs_metaslab_df_alloc_threshold; 6162 6163 if (zs->zs_do_init) 6164 ztest_run_init(); 6165 else 6166 ztest_run(zs); 6167 exit(0); 6168 } 6169 6170 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0); 6171 6172 if (ztest_opts.zo_verbose >= 1) { 6173 (void) printf("%llu vdevs, %d datasets, %d threads," 6174 " %llu seconds...\n", 6175 (u_longlong_t)ztest_opts.zo_vdevs, 6176 ztest_opts.zo_datasets, 6177 ztest_opts.zo_threads, 6178 (u_longlong_t)ztest_opts.zo_time); 6179 } 6180 6181 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); 6182 (void) strlcpy(cmd, getexecname(), MAXNAMELEN); 6183 6184 zs->zs_do_init = B_TRUE; 6185 if (strlen(ztest_opts.zo_alt_ztest) != 0) { 6186 if (ztest_opts.zo_verbose >= 1) { 6187 (void) printf("Executing older ztest for " 6188 "initialization: %s\n", ztest_opts.zo_alt_ztest); 6189 } 6190 VERIFY(!exec_child(ztest_opts.zo_alt_ztest, 6191 ztest_opts.zo_alt_libpath, B_FALSE, NULL)); 6192 } else { 6193 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL)); 6194 } 6195 zs->zs_do_init = B_FALSE; 6196 6197 zs->zs_proc_start = gethrtime(); 6198 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC; 6199 6200 for (int f = 0; f < ZTEST_FUNCS; f++) { 6201 zi = &ztest_info[f]; 6202 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6203 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop) 6204 zc->zc_next = UINT64_MAX; 6205 else 6206 zc->zc_next = zs->zs_proc_start + 6207 ztest_random(2 * zi->zi_interval[0] + 1); 6208 } 6209 6210 /* 6211 * Run the tests in a loop. These tests include fault injection 6212 * to verify that self-healing data works, and forced crashes 6213 * to verify that we never lose on-disk consistency. 6214 */ 6215 while (gethrtime() < zs->zs_proc_stop) { 6216 int status; 6217 boolean_t killed; 6218 6219 /* 6220 * Initialize the workload counters for each function. 6221 */ 6222 for (int f = 0; f < ZTEST_FUNCS; f++) { 6223 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6224 zc->zc_count = 0; 6225 zc->zc_time = 0; 6226 } 6227 6228 /* Set the allocation switch size */ 6229 zs->zs_metaslab_df_alloc_threshold = 6230 ztest_random(zs->zs_metaslab_sz / 4) + 1; 6231 6232 if (!hasalt || ztest_random(2) == 0) { 6233 if (hasalt && ztest_opts.zo_verbose >= 1) { 6234 (void) printf("Executing newer ztest: %s\n", 6235 cmd); 6236 } 6237 newer++; 6238 killed = exec_child(cmd, NULL, B_TRUE, &status); 6239 } else { 6240 if (hasalt && ztest_opts.zo_verbose >= 1) { 6241 (void) printf("Executing older ztest: %s\n", 6242 ztest_opts.zo_alt_ztest); 6243 } 6244 older++; 6245 killed = exec_child(ztest_opts.zo_alt_ztest, 6246 ztest_opts.zo_alt_libpath, B_TRUE, &status); 6247 } 6248 6249 if (killed) 6250 kills++; 6251 iters++; 6252 6253 if (ztest_opts.zo_verbose >= 1) { 6254 hrtime_t now = gethrtime(); 6255 6256 now = MIN(now, zs->zs_proc_stop); 6257 print_time(zs->zs_proc_stop - now, timebuf); 6258 nicenum(zs->zs_space, numbuf); 6259 6260 (void) printf("Pass %3d, %8s, %3llu ENOSPC, " 6261 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", 6262 iters, 6263 WIFEXITED(status) ? "Complete" : "SIGKILL", 6264 (u_longlong_t)zs->zs_enospc_count, 6265 100.0 * zs->zs_alloc / zs->zs_space, 6266 numbuf, 6267 100.0 * (now - zs->zs_proc_start) / 6268 (ztest_opts.zo_time * NANOSEC), timebuf); 6269 } 6270 6271 if (ztest_opts.zo_verbose >= 2) { 6272 (void) printf("\nWorkload summary:\n\n"); 6273 (void) printf("%7s %9s %s\n", 6274 "Calls", "Time", "Function"); 6275 (void) printf("%7s %9s %s\n", 6276 "-----", "----", "--------"); 6277 for (int f = 0; f < ZTEST_FUNCS; f++) { 6278 Dl_info dli; 6279 6280 zi = &ztest_info[f]; 6281 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6282 print_time(zc->zc_time, timebuf); 6283 (void) dladdr((void *)zi->zi_func, &dli); 6284 (void) printf("%7llu %9s %s\n", 6285 (u_longlong_t)zc->zc_count, timebuf, 6286 dli.dli_sname); 6287 } 6288 (void) printf("\n"); 6289 } 6290 6291 /* 6292 * It's possible that we killed a child during a rename test, 6293 * in which case we'll have a 'ztest_tmp' pool lying around 6294 * instead of 'ztest'. Do a blind rename in case this happened. 6295 */ 6296 kernel_init(FREAD); 6297 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) { 6298 spa_close(spa, FTAG); 6299 } else { 6300 char tmpname[MAXNAMELEN]; 6301 kernel_fini(); 6302 kernel_init(FREAD | FWRITE); 6303 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp", 6304 ztest_opts.zo_pool); 6305 (void) spa_rename(tmpname, ztest_opts.zo_pool); 6306 } 6307 kernel_fini(); 6308 6309 ztest_run_zdb(ztest_opts.zo_pool); 6310 } 6311 6312 if (ztest_opts.zo_verbose >= 1) { 6313 if (hasalt) { 6314 (void) printf("%d runs of older ztest: %s\n", older, 6315 ztest_opts.zo_alt_ztest); 6316 (void) printf("%d runs of newer ztest: %s\n", newer, 6317 cmd); 6318 } 6319 (void) printf("%d killed, %d completed, %.0f%% kill rate\n", 6320 kills, iters - kills, (100.0 * kills) / MAX(1, iters)); 6321 } 6322 6323 umem_free(cmd, MAXNAMELEN); 6324 6325 return (0); 6326} 6327