ztest.c revision 269416
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, 2014 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 */ 28 29/* 30 * The objective of this program is to provide a DMU/ZAP/SPA stress test 31 * that runs entirely in userland, is easy to use, and easy to extend. 32 * 33 * The overall design of the ztest program is as follows: 34 * 35 * (1) For each major functional area (e.g. adding vdevs to a pool, 36 * creating and destroying datasets, reading and writing objects, etc) 37 * we have a simple routine to test that functionality. These 38 * individual routines do not have to do anything "stressful". 39 * 40 * (2) We turn these simple functionality tests into a stress test by 41 * running them all in parallel, with as many threads as desired, 42 * and spread across as many datasets, objects, and vdevs as desired. 43 * 44 * (3) While all this is happening, we inject faults into the pool to 45 * verify that self-healing data really works. 46 * 47 * (4) Every time we open a dataset, we change its checksum and compression 48 * functions. Thus even individual objects vary from block to block 49 * in which checksum they use and whether they're compressed. 50 * 51 * (5) To verify that we never lose on-disk consistency after a crash, 52 * we run the entire test in a child of the main process. 53 * At random times, the child self-immolates with a SIGKILL. 54 * This is the software equivalent of pulling the power cord. 55 * The parent then runs the test again, using the existing 56 * storage pool, as many times as desired. If backwards compatibility 57 * testing is enabled ztest will sometimes run the "older" version 58 * of ztest after a SIGKILL. 59 * 60 * (6) To verify that we don't have future leaks or temporal incursions, 61 * many of the functional tests record the transaction group number 62 * as part of their data. When reading old data, they verify that 63 * the transaction group number is less than the current, open txg. 64 * If you add a new test, please do this if applicable. 65 * 66 * When run with no arguments, ztest runs for about five minutes and 67 * produces no output if successful. To get a little bit of information, 68 * specify -V. To get more information, specify -VV, and so on. 69 * 70 * To turn this into an overnight stress test, use -T to specify run time. 71 * 72 * You can ask more more vdevs [-v], datasets [-d], or threads [-t] 73 * to increase the pool capacity, fanout, and overall stress level. 74 * 75 * Use the -k option to set the desired frequency of kills. 76 * 77 * When ztest invokes itself it passes all relevant information through a 78 * temporary file which is mmap-ed in the child process. This allows shared 79 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always 80 * stored at offset 0 of this file and contains information on the size and 81 * number of shared structures in the file. The information stored in this file 82 * must remain backwards compatible with older versions of ztest so that 83 * ztest can invoke them during backwards compatibility testing (-B). 84 */ 85 86#include <sys/zfs_context.h> 87#include <sys/spa.h> 88#include <sys/dmu.h> 89#include <sys/txg.h> 90#include <sys/dbuf.h> 91#include <sys/zap.h> 92#include <sys/dmu_objset.h> 93#include <sys/poll.h> 94#include <sys/stat.h> 95#include <sys/time.h> 96#include <sys/wait.h> 97#include <sys/mman.h> 98#include <sys/resource.h> 99#include <sys/zio.h> 100#include <sys/zil.h> 101#include <sys/zil_impl.h> 102#include <sys/vdev_impl.h> 103#include <sys/vdev_file.h> 104#include <sys/spa_impl.h> 105#include <sys/metaslab_impl.h> 106#include <sys/dsl_prop.h> 107#include <sys/dsl_dataset.h> 108#include <sys/dsl_destroy.h> 109#include <sys/dsl_scan.h> 110#include <sys/zio_checksum.h> 111#include <sys/refcount.h> 112#include <sys/zfeature.h> 113#include <sys/dsl_userhold.h> 114#include <stdio.h> 115#include <stdio_ext.h> 116#include <stdlib.h> 117#include <unistd.h> 118#include <signal.h> 119#include <umem.h> 120#include <dlfcn.h> 121#include <ctype.h> 122#include <math.h> 123#include <errno.h> 124#include <sys/fs/zfs.h> 125#include <libnvpair.h> 126 127static int ztest_fd_data = -1; 128static int ztest_fd_rand = -1; 129 130typedef struct ztest_shared_hdr { 131 uint64_t zh_hdr_size; 132 uint64_t zh_opts_size; 133 uint64_t zh_size; 134 uint64_t zh_stats_size; 135 uint64_t zh_stats_count; 136 uint64_t zh_ds_size; 137 uint64_t zh_ds_count; 138} ztest_shared_hdr_t; 139 140static ztest_shared_hdr_t *ztest_shared_hdr; 141 142typedef struct ztest_shared_opts { 143 char zo_pool[MAXNAMELEN]; 144 char zo_dir[MAXNAMELEN]; 145 char zo_alt_ztest[MAXNAMELEN]; 146 char zo_alt_libpath[MAXNAMELEN]; 147 uint64_t zo_vdevs; 148 uint64_t zo_vdevtime; 149 size_t zo_vdev_size; 150 int zo_ashift; 151 int zo_mirrors; 152 int zo_raidz; 153 int zo_raidz_parity; 154 int zo_datasets; 155 int zo_threads; 156 uint64_t zo_passtime; 157 uint64_t zo_killrate; 158 int zo_verbose; 159 int zo_init; 160 uint64_t zo_time; 161 uint64_t zo_maxloops; 162 uint64_t zo_metaslab_gang_bang; 163} ztest_shared_opts_t; 164 165static const ztest_shared_opts_t ztest_opts_defaults = { 166 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' }, 167 .zo_dir = { '/', 't', 'm', 'p', '\0' }, 168 .zo_alt_ztest = { '\0' }, 169 .zo_alt_libpath = { '\0' }, 170 .zo_vdevs = 5, 171 .zo_ashift = SPA_MINBLOCKSHIFT, 172 .zo_mirrors = 2, 173 .zo_raidz = 4, 174 .zo_raidz_parity = 1, 175 .zo_vdev_size = SPA_MINDEVSIZE, 176 .zo_datasets = 7, 177 .zo_threads = 23, 178 .zo_passtime = 60, /* 60 seconds */ 179 .zo_killrate = 70, /* 70% kill rate */ 180 .zo_verbose = 0, 181 .zo_init = 1, 182 .zo_time = 300, /* 5 minutes */ 183 .zo_maxloops = 50, /* max loops during spa_freeze() */ 184 .zo_metaslab_gang_bang = 32 << 10 185}; 186 187extern uint64_t metaslab_gang_bang; 188extern uint64_t metaslab_df_alloc_threshold; 189extern uint64_t zfs_deadman_synctime_ms; 190extern int metaslab_preload_limit; 191 192static ztest_shared_opts_t *ztest_shared_opts; 193static ztest_shared_opts_t ztest_opts; 194 195typedef struct ztest_shared_ds { 196 uint64_t zd_seq; 197} ztest_shared_ds_t; 198 199static ztest_shared_ds_t *ztest_shared_ds; 200#define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d]) 201 202#define BT_MAGIC 0x123456789abcdefULL 203#define MAXFAULTS() \ 204 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1) 205 206enum ztest_io_type { 207 ZTEST_IO_WRITE_TAG, 208 ZTEST_IO_WRITE_PATTERN, 209 ZTEST_IO_WRITE_ZEROES, 210 ZTEST_IO_TRUNCATE, 211 ZTEST_IO_SETATTR, 212 ZTEST_IO_REWRITE, 213 ZTEST_IO_TYPES 214}; 215 216typedef struct ztest_block_tag { 217 uint64_t bt_magic; 218 uint64_t bt_objset; 219 uint64_t bt_object; 220 uint64_t bt_offset; 221 uint64_t bt_gen; 222 uint64_t bt_txg; 223 uint64_t bt_crtxg; 224} ztest_block_tag_t; 225 226typedef struct bufwad { 227 uint64_t bw_index; 228 uint64_t bw_txg; 229 uint64_t bw_data; 230} bufwad_t; 231 232/* 233 * XXX -- fix zfs range locks to be generic so we can use them here. 234 */ 235typedef enum { 236 RL_READER, 237 RL_WRITER, 238 RL_APPEND 239} rl_type_t; 240 241typedef struct rll { 242 void *rll_writer; 243 int rll_readers; 244 mutex_t rll_lock; 245 cond_t rll_cv; 246} rll_t; 247 248typedef struct rl { 249 uint64_t rl_object; 250 uint64_t rl_offset; 251 uint64_t rl_size; 252 rll_t *rl_lock; 253} rl_t; 254 255#define ZTEST_RANGE_LOCKS 64 256#define ZTEST_OBJECT_LOCKS 64 257 258/* 259 * Object descriptor. Used as a template for object lookup/create/remove. 260 */ 261typedef struct ztest_od { 262 uint64_t od_dir; 263 uint64_t od_object; 264 dmu_object_type_t od_type; 265 dmu_object_type_t od_crtype; 266 uint64_t od_blocksize; 267 uint64_t od_crblocksize; 268 uint64_t od_gen; 269 uint64_t od_crgen; 270 char od_name[MAXNAMELEN]; 271} ztest_od_t; 272 273/* 274 * Per-dataset state. 275 */ 276typedef struct ztest_ds { 277 ztest_shared_ds_t *zd_shared; 278 objset_t *zd_os; 279 rwlock_t zd_zilog_lock; 280 zilog_t *zd_zilog; 281 ztest_od_t *zd_od; /* debugging aid */ 282 char zd_name[MAXNAMELEN]; 283 mutex_t zd_dirobj_lock; 284 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS]; 285 rll_t zd_range_lock[ZTEST_RANGE_LOCKS]; 286} ztest_ds_t; 287 288/* 289 * Per-iteration state. 290 */ 291typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id); 292 293typedef struct ztest_info { 294 ztest_func_t *zi_func; /* test function */ 295 uint64_t zi_iters; /* iterations per execution */ 296 uint64_t *zi_interval; /* execute every <interval> seconds */ 297} ztest_info_t; 298 299typedef struct ztest_shared_callstate { 300 uint64_t zc_count; /* per-pass count */ 301 uint64_t zc_time; /* per-pass time */ 302 uint64_t zc_next; /* next time to call this function */ 303} ztest_shared_callstate_t; 304 305static ztest_shared_callstate_t *ztest_shared_callstate; 306#define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c]) 307 308/* 309 * Note: these aren't static because we want dladdr() to work. 310 */ 311ztest_func_t ztest_dmu_read_write; 312ztest_func_t ztest_dmu_write_parallel; 313ztest_func_t ztest_dmu_object_alloc_free; 314ztest_func_t ztest_dmu_commit_callbacks; 315ztest_func_t ztest_zap; 316ztest_func_t ztest_zap_parallel; 317ztest_func_t ztest_zil_commit; 318ztest_func_t ztest_zil_remount; 319ztest_func_t ztest_dmu_read_write_zcopy; 320ztest_func_t ztest_dmu_objset_create_destroy; 321ztest_func_t ztest_dmu_prealloc; 322ztest_func_t ztest_fzap; 323ztest_func_t ztest_dmu_snapshot_create_destroy; 324ztest_func_t ztest_dsl_prop_get_set; 325ztest_func_t ztest_spa_prop_get_set; 326ztest_func_t ztest_spa_create_destroy; 327ztest_func_t ztest_fault_inject; 328ztest_func_t ztest_ddt_repair; 329ztest_func_t ztest_dmu_snapshot_hold; 330ztest_func_t ztest_spa_rename; 331ztest_func_t ztest_scrub; 332ztest_func_t ztest_dsl_dataset_promote_busy; 333ztest_func_t ztest_vdev_attach_detach; 334ztest_func_t ztest_vdev_LUN_growth; 335ztest_func_t ztest_vdev_add_remove; 336ztest_func_t ztest_vdev_aux_add_remove; 337ztest_func_t ztest_split_pool; 338ztest_func_t ztest_reguid; 339ztest_func_t ztest_spa_upgrade; 340 341uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */ 342uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */ 343uint64_t zopt_often = 1ULL * NANOSEC; /* every second */ 344uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */ 345uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */ 346 347ztest_info_t ztest_info[] = { 348 { ztest_dmu_read_write, 1, &zopt_always }, 349 { ztest_dmu_write_parallel, 10, &zopt_always }, 350 { ztest_dmu_object_alloc_free, 1, &zopt_always }, 351 { ztest_dmu_commit_callbacks, 1, &zopt_always }, 352 { ztest_zap, 30, &zopt_always }, 353 { ztest_zap_parallel, 100, &zopt_always }, 354 { ztest_split_pool, 1, &zopt_always }, 355 { ztest_zil_commit, 1, &zopt_incessant }, 356 { ztest_zil_remount, 1, &zopt_sometimes }, 357 { ztest_dmu_read_write_zcopy, 1, &zopt_often }, 358 { ztest_dmu_objset_create_destroy, 1, &zopt_often }, 359 { ztest_dsl_prop_get_set, 1, &zopt_often }, 360 { ztest_spa_prop_get_set, 1, &zopt_sometimes }, 361#if 0 362 { ztest_dmu_prealloc, 1, &zopt_sometimes }, 363#endif 364 { ztest_fzap, 1, &zopt_sometimes }, 365 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes }, 366 { ztest_spa_create_destroy, 1, &zopt_sometimes }, 367 { ztest_fault_inject, 1, &zopt_sometimes }, 368 { ztest_ddt_repair, 1, &zopt_sometimes }, 369 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes }, 370 { ztest_reguid, 1, &zopt_rarely }, 371 { ztest_spa_rename, 1, &zopt_rarely }, 372 { ztest_scrub, 1, &zopt_rarely }, 373 { ztest_spa_upgrade, 1, &zopt_rarely }, 374 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely }, 375 { ztest_vdev_attach_detach, 1, &zopt_sometimes }, 376 { ztest_vdev_LUN_growth, 1, &zopt_rarely }, 377 { ztest_vdev_add_remove, 1, 378 &ztest_opts.zo_vdevtime }, 379 { ztest_vdev_aux_add_remove, 1, 380 &ztest_opts.zo_vdevtime }, 381}; 382 383#define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) 384 385/* 386 * The following struct is used to hold a list of uncalled commit callbacks. 387 * The callbacks are ordered by txg number. 388 */ 389typedef struct ztest_cb_list { 390 mutex_t zcl_callbacks_lock; 391 list_t zcl_callbacks; 392} ztest_cb_list_t; 393 394/* 395 * Stuff we need to share writably between parent and child. 396 */ 397typedef struct ztest_shared { 398 boolean_t zs_do_init; 399 hrtime_t zs_proc_start; 400 hrtime_t zs_proc_stop; 401 hrtime_t zs_thread_start; 402 hrtime_t zs_thread_stop; 403 hrtime_t zs_thread_kill; 404 uint64_t zs_enospc_count; 405 uint64_t zs_vdev_next_leaf; 406 uint64_t zs_vdev_aux; 407 uint64_t zs_alloc; 408 uint64_t zs_space; 409 uint64_t zs_splits; 410 uint64_t zs_mirrors; 411 uint64_t zs_metaslab_sz; 412 uint64_t zs_metaslab_df_alloc_threshold; 413 uint64_t zs_guid; 414} ztest_shared_t; 415 416#define ID_PARALLEL -1ULL 417 418static char ztest_dev_template[] = "%s/%s.%llua"; 419static char ztest_aux_template[] = "%s/%s.%s.%llu"; 420ztest_shared_t *ztest_shared; 421 422static spa_t *ztest_spa = NULL; 423static ztest_ds_t *ztest_ds; 424 425static mutex_t ztest_vdev_lock; 426 427/* 428 * The ztest_name_lock protects the pool and dataset namespace used by 429 * the individual tests. To modify the namespace, consumers must grab 430 * this lock as writer. Grabbing the lock as reader will ensure that the 431 * namespace does not change while the lock is held. 432 */ 433static rwlock_t ztest_name_lock; 434 435static boolean_t ztest_dump_core = B_TRUE; 436static boolean_t ztest_exiting; 437 438/* Global commit callback list */ 439static ztest_cb_list_t zcl; 440 441enum ztest_object { 442 ZTEST_META_DNODE = 0, 443 ZTEST_DIROBJ, 444 ZTEST_OBJECTS 445}; 446 447static void usage(boolean_t) __NORETURN; 448 449/* 450 * These libumem hooks provide a reasonable set of defaults for the allocator's 451 * debugging facilities. 452 */ 453const char * 454_umem_debug_init() 455{ 456 return ("default,verbose"); /* $UMEM_DEBUG setting */ 457} 458 459const char * 460_umem_logging_init(void) 461{ 462 return ("fail,contents"); /* $UMEM_LOGGING setting */ 463} 464 465#define FATAL_MSG_SZ 1024 466 467char *fatal_msg; 468 469static void 470fatal(int do_perror, char *message, ...) 471{ 472 va_list args; 473 int save_errno = errno; 474 char buf[FATAL_MSG_SZ]; 475 476 (void) fflush(stdout); 477 478 va_start(args, message); 479 (void) sprintf(buf, "ztest: "); 480 /* LINTED */ 481 (void) vsprintf(buf + strlen(buf), message, args); 482 va_end(args); 483 if (do_perror) { 484 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), 485 ": %s", strerror(save_errno)); 486 } 487 (void) fprintf(stderr, "%s\n", buf); 488 fatal_msg = buf; /* to ease debugging */ 489 if (ztest_dump_core) 490 abort(); 491 exit(3); 492} 493 494static int 495str2shift(const char *buf) 496{ 497 const char *ends = "BKMGTPEZ"; 498 int i; 499 500 if (buf[0] == '\0') 501 return (0); 502 for (i = 0; i < strlen(ends); i++) { 503 if (toupper(buf[0]) == ends[i]) 504 break; 505 } 506 if (i == strlen(ends)) { 507 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", 508 buf); 509 usage(B_FALSE); 510 } 511 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { 512 return (10*i); 513 } 514 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); 515 usage(B_FALSE); 516 /* NOTREACHED */ 517} 518 519static uint64_t 520nicenumtoull(const char *buf) 521{ 522 char *end; 523 uint64_t val; 524 525 val = strtoull(buf, &end, 0); 526 if (end == buf) { 527 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); 528 usage(B_FALSE); 529 } else if (end[0] == '.') { 530 double fval = strtod(buf, &end); 531 fval *= pow(2, str2shift(end)); 532 if (fval > UINT64_MAX) { 533 (void) fprintf(stderr, "ztest: value too large: %s\n", 534 buf); 535 usage(B_FALSE); 536 } 537 val = (uint64_t)fval; 538 } else { 539 int shift = str2shift(end); 540 if (shift >= 64 || (val << shift) >> shift != val) { 541 (void) fprintf(stderr, "ztest: value too large: %s\n", 542 buf); 543 usage(B_FALSE); 544 } 545 val <<= shift; 546 } 547 return (val); 548} 549 550static void 551usage(boolean_t requested) 552{ 553 const ztest_shared_opts_t *zo = &ztest_opts_defaults; 554 555 char nice_vdev_size[10]; 556 char nice_gang_bang[10]; 557 FILE *fp = requested ? stdout : stderr; 558 559 nicenum(zo->zo_vdev_size, nice_vdev_size); 560 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang); 561 562 (void) fprintf(fp, "Usage: %s\n" 563 "\t[-v vdevs (default: %llu)]\n" 564 "\t[-s size_of_each_vdev (default: %s)]\n" 565 "\t[-a alignment_shift (default: %d)] use 0 for random\n" 566 "\t[-m mirror_copies (default: %d)]\n" 567 "\t[-r raidz_disks (default: %d)]\n" 568 "\t[-R raidz_parity (default: %d)]\n" 569 "\t[-d datasets (default: %d)]\n" 570 "\t[-t threads (default: %d)]\n" 571 "\t[-g gang_block_threshold (default: %s)]\n" 572 "\t[-i init_count (default: %d)] initialize pool i times\n" 573 "\t[-k kill_percentage (default: %llu%%)]\n" 574 "\t[-p pool_name (default: %s)]\n" 575 "\t[-f dir (default: %s)] file directory for vdev files\n" 576 "\t[-V] verbose (use multiple times for ever more blather)\n" 577 "\t[-E] use existing pool instead of creating new one\n" 578 "\t[-T time (default: %llu sec)] total run time\n" 579 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n" 580 "\t[-P passtime (default: %llu sec)] time per pass\n" 581 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n" 582 "\t[-h] (print help)\n" 583 "", 584 zo->zo_pool, 585 (u_longlong_t)zo->zo_vdevs, /* -v */ 586 nice_vdev_size, /* -s */ 587 zo->zo_ashift, /* -a */ 588 zo->zo_mirrors, /* -m */ 589 zo->zo_raidz, /* -r */ 590 zo->zo_raidz_parity, /* -R */ 591 zo->zo_datasets, /* -d */ 592 zo->zo_threads, /* -t */ 593 nice_gang_bang, /* -g */ 594 zo->zo_init, /* -i */ 595 (u_longlong_t)zo->zo_killrate, /* -k */ 596 zo->zo_pool, /* -p */ 597 zo->zo_dir, /* -f */ 598 (u_longlong_t)zo->zo_time, /* -T */ 599 (u_longlong_t)zo->zo_maxloops, /* -F */ 600 (u_longlong_t)zo->zo_passtime); 601 exit(requested ? 0 : 1); 602} 603 604static void 605process_options(int argc, char **argv) 606{ 607 char *path; 608 ztest_shared_opts_t *zo = &ztest_opts; 609 610 int opt; 611 uint64_t value; 612 char altdir[MAXNAMELEN] = { 0 }; 613 614 bcopy(&ztest_opts_defaults, zo, sizeof (*zo)); 615 616 while ((opt = getopt(argc, argv, 617 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) { 618 value = 0; 619 switch (opt) { 620 case 'v': 621 case 's': 622 case 'a': 623 case 'm': 624 case 'r': 625 case 'R': 626 case 'd': 627 case 't': 628 case 'g': 629 case 'i': 630 case 'k': 631 case 'T': 632 case 'P': 633 case 'F': 634 value = nicenumtoull(optarg); 635 } 636 switch (opt) { 637 case 'v': 638 zo->zo_vdevs = value; 639 break; 640 case 's': 641 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value); 642 break; 643 case 'a': 644 zo->zo_ashift = value; 645 break; 646 case 'm': 647 zo->zo_mirrors = value; 648 break; 649 case 'r': 650 zo->zo_raidz = MAX(1, value); 651 break; 652 case 'R': 653 zo->zo_raidz_parity = MIN(MAX(value, 1), 3); 654 break; 655 case 'd': 656 zo->zo_datasets = MAX(1, value); 657 break; 658 case 't': 659 zo->zo_threads = MAX(1, value); 660 break; 661 case 'g': 662 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, 663 value); 664 break; 665 case 'i': 666 zo->zo_init = value; 667 break; 668 case 'k': 669 zo->zo_killrate = value; 670 break; 671 case 'p': 672 (void) strlcpy(zo->zo_pool, optarg, 673 sizeof (zo->zo_pool)); 674 break; 675 case 'f': 676 path = realpath(optarg, NULL); 677 if (path == NULL) { 678 (void) fprintf(stderr, "error: %s: %s\n", 679 optarg, strerror(errno)); 680 usage(B_FALSE); 681 } else { 682 (void) strlcpy(zo->zo_dir, path, 683 sizeof (zo->zo_dir)); 684 } 685 break; 686 case 'V': 687 zo->zo_verbose++; 688 break; 689 case 'E': 690 zo->zo_init = 0; 691 break; 692 case 'T': 693 zo->zo_time = value; 694 break; 695 case 'P': 696 zo->zo_passtime = MAX(1, value); 697 break; 698 case 'F': 699 zo->zo_maxloops = MAX(1, value); 700 break; 701 case 'B': 702 (void) strlcpy(altdir, optarg, sizeof (altdir)); 703 break; 704 case 'h': 705 usage(B_TRUE); 706 break; 707 case '?': 708 default: 709 usage(B_FALSE); 710 break; 711 } 712 } 713 714 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1); 715 716 zo->zo_vdevtime = 717 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs : 718 UINT64_MAX >> 2); 719 720 if (strlen(altdir) > 0) { 721 char *cmd; 722 char *realaltdir; 723 char *bin; 724 char *ztest; 725 char *isa; 726 int isalen; 727 728 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 729 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 730 731 VERIFY(NULL != realpath(getexecname(), cmd)); 732 if (0 != access(altdir, F_OK)) { 733 ztest_dump_core = B_FALSE; 734 fatal(B_TRUE, "invalid alternate ztest path: %s", 735 altdir); 736 } 737 VERIFY(NULL != realpath(altdir, realaltdir)); 738 739 /* 740 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest". 741 * We want to extract <isa> to determine if we should use 742 * 32 or 64 bit binaries. 743 */ 744 bin = strstr(cmd, "/usr/bin/"); 745 ztest = strstr(bin, "/ztest"); 746 isa = bin + 9; 747 isalen = ztest - isa; 748 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest), 749 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa); 750 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath), 751 "%s/usr/lib/%.*s", realaltdir, isalen, isa); 752 753 if (0 != access(zo->zo_alt_ztest, X_OK)) { 754 ztest_dump_core = B_FALSE; 755 fatal(B_TRUE, "invalid alternate ztest: %s", 756 zo->zo_alt_ztest); 757 } else if (0 != access(zo->zo_alt_libpath, X_OK)) { 758 ztest_dump_core = B_FALSE; 759 fatal(B_TRUE, "invalid alternate lib directory %s", 760 zo->zo_alt_libpath); 761 } 762 763 umem_free(cmd, MAXPATHLEN); 764 umem_free(realaltdir, MAXPATHLEN); 765 } 766} 767 768static void 769ztest_kill(ztest_shared_t *zs) 770{ 771 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa)); 772 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa)); 773 774 /* 775 * Before we kill off ztest, make sure that the config is updated. 776 * See comment above spa_config_sync(). 777 */ 778 mutex_enter(&spa_namespace_lock); 779 spa_config_sync(ztest_spa, B_FALSE, B_FALSE); 780 mutex_exit(&spa_namespace_lock); 781 782 zfs_dbgmsg_print(FTAG); 783 (void) kill(getpid(), SIGKILL); 784} 785 786static uint64_t 787ztest_random(uint64_t range) 788{ 789 uint64_t r; 790 791 ASSERT3S(ztest_fd_rand, >=, 0); 792 793 if (range == 0) 794 return (0); 795 796 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r)) 797 fatal(1, "short read from /dev/urandom"); 798 799 return (r % range); 800} 801 802/* ARGSUSED */ 803static void 804ztest_record_enospc(const char *s) 805{ 806 ztest_shared->zs_enospc_count++; 807} 808 809static uint64_t 810ztest_get_ashift(void) 811{ 812 if (ztest_opts.zo_ashift == 0) 813 return (SPA_MINBLOCKSHIFT + ztest_random(5)); 814 return (ztest_opts.zo_ashift); 815} 816 817static nvlist_t * 818make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift) 819{ 820 char pathbuf[MAXPATHLEN]; 821 uint64_t vdev; 822 nvlist_t *file; 823 824 if (ashift == 0) 825 ashift = ztest_get_ashift(); 826 827 if (path == NULL) { 828 path = pathbuf; 829 830 if (aux != NULL) { 831 vdev = ztest_shared->zs_vdev_aux; 832 (void) snprintf(path, sizeof (pathbuf), 833 ztest_aux_template, ztest_opts.zo_dir, 834 pool == NULL ? ztest_opts.zo_pool : pool, 835 aux, vdev); 836 } else { 837 vdev = ztest_shared->zs_vdev_next_leaf++; 838 (void) snprintf(path, sizeof (pathbuf), 839 ztest_dev_template, ztest_opts.zo_dir, 840 pool == NULL ? ztest_opts.zo_pool : pool, vdev); 841 } 842 } 843 844 if (size != 0) { 845 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666); 846 if (fd == -1) 847 fatal(1, "can't open %s", path); 848 if (ftruncate(fd, size) != 0) 849 fatal(1, "can't ftruncate %s", path); 850 (void) close(fd); 851 } 852 853 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 854 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 855 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0); 856 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 857 858 return (file); 859} 860 861static nvlist_t * 862make_vdev_raidz(char *path, char *aux, char *pool, size_t size, 863 uint64_t ashift, int r) 864{ 865 nvlist_t *raidz, **child; 866 int c; 867 868 if (r < 2) 869 return (make_vdev_file(path, aux, pool, size, ashift)); 870 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); 871 872 for (c = 0; c < r; c++) 873 child[c] = make_vdev_file(path, aux, pool, size, ashift); 874 875 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); 876 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, 877 VDEV_TYPE_RAIDZ) == 0); 878 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, 879 ztest_opts.zo_raidz_parity) == 0); 880 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, 881 child, r) == 0); 882 883 for (c = 0; c < r; c++) 884 nvlist_free(child[c]); 885 886 umem_free(child, r * sizeof (nvlist_t *)); 887 888 return (raidz); 889} 890 891static nvlist_t * 892make_vdev_mirror(char *path, char *aux, char *pool, size_t size, 893 uint64_t ashift, int r, int m) 894{ 895 nvlist_t *mirror, **child; 896 int c; 897 898 if (m < 1) 899 return (make_vdev_raidz(path, aux, pool, size, ashift, r)); 900 901 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); 902 903 for (c = 0; c < m; c++) 904 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r); 905 906 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); 907 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, 908 VDEV_TYPE_MIRROR) == 0); 909 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, 910 child, m) == 0); 911 912 for (c = 0; c < m; c++) 913 nvlist_free(child[c]); 914 915 umem_free(child, m * sizeof (nvlist_t *)); 916 917 return (mirror); 918} 919 920static nvlist_t * 921make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift, 922 int log, int r, int m, int t) 923{ 924 nvlist_t *root, **child; 925 int c; 926 927 ASSERT(t > 0); 928 929 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); 930 931 for (c = 0; c < t; c++) { 932 child[c] = make_vdev_mirror(path, aux, pool, size, ashift, 933 r, m); 934 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, 935 log) == 0); 936 } 937 938 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 939 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 940 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN, 941 child, t) == 0); 942 943 for (c = 0; c < t; c++) 944 nvlist_free(child[c]); 945 946 umem_free(child, t * sizeof (nvlist_t *)); 947 948 return (root); 949} 950 951/* 952 * Find a random spa version. Returns back a random spa version in the 953 * range [initial_version, SPA_VERSION_FEATURES]. 954 */ 955static uint64_t 956ztest_random_spa_version(uint64_t initial_version) 957{ 958 uint64_t version = initial_version; 959 960 if (version <= SPA_VERSION_BEFORE_FEATURES) { 961 version = version + 962 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1); 963 } 964 965 if (version > SPA_VERSION_BEFORE_FEATURES) 966 version = SPA_VERSION_FEATURES; 967 968 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 969 return (version); 970} 971 972/* 973 * Find the largest ashift used 974 */ 975static uint64_t 976ztest_spa_get_ashift() { 977 uint64_t i; 978 uint64_t ashift = SPA_MINBLOCKSHIFT; 979 vdev_t *rvd = ztest_spa->spa_root_vdev; 980 981 for (i = 0; i < rvd->vdev_children; i++) { 982 ashift = MAX(ashift, rvd->vdev_child[i]->vdev_ashift); 983 } 984 return (ashift); 985} 986 987static int 988ztest_random_blocksize(void) 989{ 990 // Choose a block size >= the ashift. 991 uint64_t block_shift = 992 ztest_random(SPA_MAXBLOCKSHIFT - ztest_spa_get_ashift() + 1); 993 return (1 << (SPA_MINBLOCKSHIFT + block_shift)); 994} 995 996static int 997ztest_random_ibshift(void) 998{ 999 return (DN_MIN_INDBLKSHIFT + 1000 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1)); 1001} 1002 1003static uint64_t 1004ztest_random_vdev_top(spa_t *spa, boolean_t log_ok) 1005{ 1006 uint64_t top; 1007 vdev_t *rvd = spa->spa_root_vdev; 1008 vdev_t *tvd; 1009 1010 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1011 1012 do { 1013 top = ztest_random(rvd->vdev_children); 1014 tvd = rvd->vdev_child[top]; 1015 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) || 1016 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL); 1017 1018 return (top); 1019} 1020 1021static uint64_t 1022ztest_random_dsl_prop(zfs_prop_t prop) 1023{ 1024 uint64_t value; 1025 1026 do { 1027 value = zfs_prop_random_value(prop, ztest_random(-1ULL)); 1028 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF); 1029 1030 return (value); 1031} 1032 1033static int 1034ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, 1035 boolean_t inherit) 1036{ 1037 const char *propname = zfs_prop_to_name(prop); 1038 const char *valname; 1039 char setpoint[MAXPATHLEN]; 1040 uint64_t curval; 1041 int error; 1042 1043 error = dsl_prop_set_int(osname, propname, 1044 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value); 1045 1046 if (error == ENOSPC) { 1047 ztest_record_enospc(FTAG); 1048 return (error); 1049 } 1050 ASSERT0(error); 1051 1052 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint)); 1053 1054 if (ztest_opts.zo_verbose >= 6) { 1055 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0); 1056 (void) printf("%s %s = %s at '%s'\n", 1057 osname, propname, valname, setpoint); 1058 } 1059 1060 return (error); 1061} 1062 1063static int 1064ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value) 1065{ 1066 spa_t *spa = ztest_spa; 1067 nvlist_t *props = NULL; 1068 int error; 1069 1070 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); 1071 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0); 1072 1073 error = spa_prop_set(spa, props); 1074 1075 nvlist_free(props); 1076 1077 if (error == ENOSPC) { 1078 ztest_record_enospc(FTAG); 1079 return (error); 1080 } 1081 ASSERT0(error); 1082 1083 return (error); 1084} 1085 1086static void 1087ztest_rll_init(rll_t *rll) 1088{ 1089 rll->rll_writer = NULL; 1090 rll->rll_readers = 0; 1091 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0); 1092 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0); 1093} 1094 1095static void 1096ztest_rll_destroy(rll_t *rll) 1097{ 1098 ASSERT(rll->rll_writer == NULL); 1099 ASSERT(rll->rll_readers == 0); 1100 VERIFY(_mutex_destroy(&rll->rll_lock) == 0); 1101 VERIFY(cond_destroy(&rll->rll_cv) == 0); 1102} 1103 1104static void 1105ztest_rll_lock(rll_t *rll, rl_type_t type) 1106{ 1107 VERIFY(mutex_lock(&rll->rll_lock) == 0); 1108 1109 if (type == RL_READER) { 1110 while (rll->rll_writer != NULL) 1111 (void) cond_wait(&rll->rll_cv, &rll->rll_lock); 1112 rll->rll_readers++; 1113 } else { 1114 while (rll->rll_writer != NULL || rll->rll_readers) 1115 (void) cond_wait(&rll->rll_cv, &rll->rll_lock); 1116 rll->rll_writer = curthread; 1117 } 1118 1119 VERIFY(mutex_unlock(&rll->rll_lock) == 0); 1120} 1121 1122static void 1123ztest_rll_unlock(rll_t *rll) 1124{ 1125 VERIFY(mutex_lock(&rll->rll_lock) == 0); 1126 1127 if (rll->rll_writer) { 1128 ASSERT(rll->rll_readers == 0); 1129 rll->rll_writer = NULL; 1130 } else { 1131 ASSERT(rll->rll_readers != 0); 1132 ASSERT(rll->rll_writer == NULL); 1133 rll->rll_readers--; 1134 } 1135 1136 if (rll->rll_writer == NULL && rll->rll_readers == 0) 1137 VERIFY(cond_broadcast(&rll->rll_cv) == 0); 1138 1139 VERIFY(mutex_unlock(&rll->rll_lock) == 0); 1140} 1141 1142static void 1143ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type) 1144{ 1145 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; 1146 1147 ztest_rll_lock(rll, type); 1148} 1149 1150static void 1151ztest_object_unlock(ztest_ds_t *zd, uint64_t object) 1152{ 1153 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; 1154 1155 ztest_rll_unlock(rll); 1156} 1157 1158static rl_t * 1159ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset, 1160 uint64_t size, rl_type_t type) 1161{ 1162 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1)); 1163 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)]; 1164 rl_t *rl; 1165 1166 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL); 1167 rl->rl_object = object; 1168 rl->rl_offset = offset; 1169 rl->rl_size = size; 1170 rl->rl_lock = rll; 1171 1172 ztest_rll_lock(rll, type); 1173 1174 return (rl); 1175} 1176 1177static void 1178ztest_range_unlock(rl_t *rl) 1179{ 1180 rll_t *rll = rl->rl_lock; 1181 1182 ztest_rll_unlock(rll); 1183 1184 umem_free(rl, sizeof (*rl)); 1185} 1186 1187static void 1188ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os) 1189{ 1190 zd->zd_os = os; 1191 zd->zd_zilog = dmu_objset_zil(os); 1192 zd->zd_shared = szd; 1193 dmu_objset_name(os, zd->zd_name); 1194 1195 if (zd->zd_shared != NULL) 1196 zd->zd_shared->zd_seq = 0; 1197 1198 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0); 1199 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0); 1200 1201 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++) 1202 ztest_rll_init(&zd->zd_object_lock[l]); 1203 1204 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++) 1205 ztest_rll_init(&zd->zd_range_lock[l]); 1206} 1207 1208static void 1209ztest_zd_fini(ztest_ds_t *zd) 1210{ 1211 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0); 1212 1213 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++) 1214 ztest_rll_destroy(&zd->zd_object_lock[l]); 1215 1216 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++) 1217 ztest_rll_destroy(&zd->zd_range_lock[l]); 1218} 1219 1220#define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT) 1221 1222static uint64_t 1223ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag) 1224{ 1225 uint64_t txg; 1226 int error; 1227 1228 /* 1229 * Attempt to assign tx to some transaction group. 1230 */ 1231 error = dmu_tx_assign(tx, txg_how); 1232 if (error) { 1233 if (error == ERESTART) { 1234 ASSERT(txg_how == TXG_NOWAIT); 1235 dmu_tx_wait(tx); 1236 } else { 1237 ASSERT3U(error, ==, ENOSPC); 1238 ztest_record_enospc(tag); 1239 } 1240 dmu_tx_abort(tx); 1241 return (0); 1242 } 1243 txg = dmu_tx_get_txg(tx); 1244 ASSERT(txg != 0); 1245 return (txg); 1246} 1247 1248static void 1249ztest_pattern_set(void *buf, uint64_t size, uint64_t value) 1250{ 1251 uint64_t *ip = buf; 1252 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); 1253 1254 while (ip < ip_end) 1255 *ip++ = value; 1256} 1257 1258static boolean_t 1259ztest_pattern_match(void *buf, uint64_t size, uint64_t value) 1260{ 1261 uint64_t *ip = buf; 1262 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); 1263 uint64_t diff = 0; 1264 1265 while (ip < ip_end) 1266 diff |= (value - *ip++); 1267 1268 return (diff == 0); 1269} 1270 1271static void 1272ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object, 1273 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) 1274{ 1275 bt->bt_magic = BT_MAGIC; 1276 bt->bt_objset = dmu_objset_id(os); 1277 bt->bt_object = object; 1278 bt->bt_offset = offset; 1279 bt->bt_gen = gen; 1280 bt->bt_txg = txg; 1281 bt->bt_crtxg = crtxg; 1282} 1283 1284static void 1285ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object, 1286 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) 1287{ 1288 ASSERT3U(bt->bt_magic, ==, BT_MAGIC); 1289 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os)); 1290 ASSERT3U(bt->bt_object, ==, object); 1291 ASSERT3U(bt->bt_offset, ==, offset); 1292 ASSERT3U(bt->bt_gen, <=, gen); 1293 ASSERT3U(bt->bt_txg, <=, txg); 1294 ASSERT3U(bt->bt_crtxg, ==, crtxg); 1295} 1296 1297static ztest_block_tag_t * 1298ztest_bt_bonus(dmu_buf_t *db) 1299{ 1300 dmu_object_info_t doi; 1301 ztest_block_tag_t *bt; 1302 1303 dmu_object_info_from_db(db, &doi); 1304 ASSERT3U(doi.doi_bonus_size, <=, db->db_size); 1305 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt)); 1306 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt)); 1307 1308 return (bt); 1309} 1310 1311/* 1312 * ZIL logging ops 1313 */ 1314 1315#define lrz_type lr_mode 1316#define lrz_blocksize lr_uid 1317#define lrz_ibshift lr_gid 1318#define lrz_bonustype lr_rdev 1319#define lrz_bonuslen lr_crtime[1] 1320 1321static void 1322ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr) 1323{ 1324 char *name = (void *)(lr + 1); /* name follows lr */ 1325 size_t namesize = strlen(name) + 1; 1326 itx_t *itx; 1327 1328 if (zil_replaying(zd->zd_zilog, tx)) 1329 return; 1330 1331 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize); 1332 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1333 sizeof (*lr) + namesize - sizeof (lr_t)); 1334 1335 zil_itx_assign(zd->zd_zilog, itx, tx); 1336} 1337 1338static void 1339ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object) 1340{ 1341 char *name = (void *)(lr + 1); /* name follows lr */ 1342 size_t namesize = strlen(name) + 1; 1343 itx_t *itx; 1344 1345 if (zil_replaying(zd->zd_zilog, tx)) 1346 return; 1347 1348 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize); 1349 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1350 sizeof (*lr) + namesize - sizeof (lr_t)); 1351 1352 itx->itx_oid = object; 1353 zil_itx_assign(zd->zd_zilog, itx, tx); 1354} 1355 1356static void 1357ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr) 1358{ 1359 itx_t *itx; 1360 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES); 1361 1362 if (zil_replaying(zd->zd_zilog, tx)) 1363 return; 1364 1365 if (lr->lr_length > ZIL_MAX_LOG_DATA) 1366 write_state = WR_INDIRECT; 1367 1368 itx = zil_itx_create(TX_WRITE, 1369 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0)); 1370 1371 if (write_state == WR_COPIED && 1372 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length, 1373 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) { 1374 zil_itx_destroy(itx); 1375 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1376 write_state = WR_NEED_COPY; 1377 } 1378 itx->itx_private = zd; 1379 itx->itx_wr_state = write_state; 1380 itx->itx_sync = (ztest_random(8) == 0); 1381 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0); 1382 1383 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1384 sizeof (*lr) - sizeof (lr_t)); 1385 1386 zil_itx_assign(zd->zd_zilog, itx, tx); 1387} 1388 1389static void 1390ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr) 1391{ 1392 itx_t *itx; 1393 1394 if (zil_replaying(zd->zd_zilog, tx)) 1395 return; 1396 1397 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 1398 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1399 sizeof (*lr) - sizeof (lr_t)); 1400 1401 itx->itx_sync = B_FALSE; 1402 zil_itx_assign(zd->zd_zilog, itx, tx); 1403} 1404 1405static void 1406ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr) 1407{ 1408 itx_t *itx; 1409 1410 if (zil_replaying(zd->zd_zilog, tx)) 1411 return; 1412 1413 itx = zil_itx_create(TX_SETATTR, sizeof (*lr)); 1414 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1415 sizeof (*lr) - sizeof (lr_t)); 1416 1417 itx->itx_sync = B_FALSE; 1418 zil_itx_assign(zd->zd_zilog, itx, tx); 1419} 1420 1421/* 1422 * ZIL replay ops 1423 */ 1424static int 1425ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap) 1426{ 1427 char *name = (void *)(lr + 1); /* name follows lr */ 1428 objset_t *os = zd->zd_os; 1429 ztest_block_tag_t *bbt; 1430 dmu_buf_t *db; 1431 dmu_tx_t *tx; 1432 uint64_t txg; 1433 int error = 0; 1434 1435 if (byteswap) 1436 byteswap_uint64_array(lr, sizeof (*lr)); 1437 1438 ASSERT(lr->lr_doid == ZTEST_DIROBJ); 1439 ASSERT(name[0] != '\0'); 1440 1441 tx = dmu_tx_create(os); 1442 1443 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name); 1444 1445 if (lr->lrz_type == DMU_OT_ZAP_OTHER) { 1446 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); 1447 } else { 1448 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1449 } 1450 1451 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1452 if (txg == 0) 1453 return (ENOSPC); 1454 1455 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid); 1456 1457 if (lr->lrz_type == DMU_OT_ZAP_OTHER) { 1458 if (lr->lr_foid == 0) { 1459 lr->lr_foid = zap_create(os, 1460 lr->lrz_type, lr->lrz_bonustype, 1461 lr->lrz_bonuslen, tx); 1462 } else { 1463 error = zap_create_claim(os, lr->lr_foid, 1464 lr->lrz_type, lr->lrz_bonustype, 1465 lr->lrz_bonuslen, tx); 1466 } 1467 } else { 1468 if (lr->lr_foid == 0) { 1469 lr->lr_foid = dmu_object_alloc(os, 1470 lr->lrz_type, 0, lr->lrz_bonustype, 1471 lr->lrz_bonuslen, tx); 1472 } else { 1473 error = dmu_object_claim(os, lr->lr_foid, 1474 lr->lrz_type, 0, lr->lrz_bonustype, 1475 lr->lrz_bonuslen, tx); 1476 } 1477 } 1478 1479 if (error) { 1480 ASSERT3U(error, ==, EEXIST); 1481 ASSERT(zd->zd_zilog->zl_replay); 1482 dmu_tx_commit(tx); 1483 return (error); 1484 } 1485 1486 ASSERT(lr->lr_foid != 0); 1487 1488 if (lr->lrz_type != DMU_OT_ZAP_OTHER) 1489 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid, 1490 lr->lrz_blocksize, lr->lrz_ibshift, tx)); 1491 1492 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1493 bbt = ztest_bt_bonus(db); 1494 dmu_buf_will_dirty(db, tx); 1495 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg); 1496 dmu_buf_rele(db, FTAG); 1497 1498 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1, 1499 &lr->lr_foid, tx)); 1500 1501 (void) ztest_log_create(zd, tx, lr); 1502 1503 dmu_tx_commit(tx); 1504 1505 return (0); 1506} 1507 1508static int 1509ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap) 1510{ 1511 char *name = (void *)(lr + 1); /* name follows lr */ 1512 objset_t *os = zd->zd_os; 1513 dmu_object_info_t doi; 1514 dmu_tx_t *tx; 1515 uint64_t object, txg; 1516 1517 if (byteswap) 1518 byteswap_uint64_array(lr, sizeof (*lr)); 1519 1520 ASSERT(lr->lr_doid == ZTEST_DIROBJ); 1521 ASSERT(name[0] != '\0'); 1522 1523 VERIFY3U(0, ==, 1524 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object)); 1525 ASSERT(object != 0); 1526 1527 ztest_object_lock(zd, object, RL_WRITER); 1528 1529 VERIFY3U(0, ==, dmu_object_info(os, object, &doi)); 1530 1531 tx = dmu_tx_create(os); 1532 1533 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name); 1534 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 1535 1536 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1537 if (txg == 0) { 1538 ztest_object_unlock(zd, object); 1539 return (ENOSPC); 1540 } 1541 1542 if (doi.doi_type == DMU_OT_ZAP_OTHER) { 1543 VERIFY3U(0, ==, zap_destroy(os, object, tx)); 1544 } else { 1545 VERIFY3U(0, ==, dmu_object_free(os, object, tx)); 1546 } 1547 1548 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx)); 1549 1550 (void) ztest_log_remove(zd, tx, lr, object); 1551 1552 dmu_tx_commit(tx); 1553 1554 ztest_object_unlock(zd, object); 1555 1556 return (0); 1557} 1558 1559static int 1560ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap) 1561{ 1562 objset_t *os = zd->zd_os; 1563 void *data = lr + 1; /* data follows lr */ 1564 uint64_t offset, length; 1565 ztest_block_tag_t *bt = data; 1566 ztest_block_tag_t *bbt; 1567 uint64_t gen, txg, lrtxg, crtxg; 1568 dmu_object_info_t doi; 1569 dmu_tx_t *tx; 1570 dmu_buf_t *db; 1571 arc_buf_t *abuf = NULL; 1572 rl_t *rl; 1573 1574 if (byteswap) 1575 byteswap_uint64_array(lr, sizeof (*lr)); 1576 1577 offset = lr->lr_offset; 1578 length = lr->lr_length; 1579 1580 /* If it's a dmu_sync() block, write the whole block */ 1581 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 1582 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 1583 if (length < blocksize) { 1584 offset -= offset % blocksize; 1585 length = blocksize; 1586 } 1587 } 1588 1589 if (bt->bt_magic == BSWAP_64(BT_MAGIC)) 1590 byteswap_uint64_array(bt, sizeof (*bt)); 1591 1592 if (bt->bt_magic != BT_MAGIC) 1593 bt = NULL; 1594 1595 ztest_object_lock(zd, lr->lr_foid, RL_READER); 1596 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER); 1597 1598 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1599 1600 dmu_object_info_from_db(db, &doi); 1601 1602 bbt = ztest_bt_bonus(db); 1603 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1604 gen = bbt->bt_gen; 1605 crtxg = bbt->bt_crtxg; 1606 lrtxg = lr->lr_common.lrc_txg; 1607 1608 tx = dmu_tx_create(os); 1609 1610 dmu_tx_hold_write(tx, lr->lr_foid, offset, length); 1611 1612 if (ztest_random(8) == 0 && length == doi.doi_data_block_size && 1613 P2PHASE(offset, length) == 0) 1614 abuf = dmu_request_arcbuf(db, length); 1615 1616 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1617 if (txg == 0) { 1618 if (abuf != NULL) 1619 dmu_return_arcbuf(abuf); 1620 dmu_buf_rele(db, FTAG); 1621 ztest_range_unlock(rl); 1622 ztest_object_unlock(zd, lr->lr_foid); 1623 return (ENOSPC); 1624 } 1625 1626 if (bt != NULL) { 1627 /* 1628 * Usually, verify the old data before writing new data -- 1629 * but not always, because we also want to verify correct 1630 * behavior when the data was not recently read into cache. 1631 */ 1632 ASSERT(offset % doi.doi_data_block_size == 0); 1633 if (ztest_random(4) != 0) { 1634 int prefetch = ztest_random(2) ? 1635 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH; 1636 ztest_block_tag_t rbt; 1637 1638 VERIFY(dmu_read(os, lr->lr_foid, offset, 1639 sizeof (rbt), &rbt, prefetch) == 0); 1640 if (rbt.bt_magic == BT_MAGIC) { 1641 ztest_bt_verify(&rbt, os, lr->lr_foid, 1642 offset, gen, txg, crtxg); 1643 } 1644 } 1645 1646 /* 1647 * Writes can appear to be newer than the bonus buffer because 1648 * the ztest_get_data() callback does a dmu_read() of the 1649 * open-context data, which may be different than the data 1650 * as it was when the write was generated. 1651 */ 1652 if (zd->zd_zilog->zl_replay) { 1653 ztest_bt_verify(bt, os, lr->lr_foid, offset, 1654 MAX(gen, bt->bt_gen), MAX(txg, lrtxg), 1655 bt->bt_crtxg); 1656 } 1657 1658 /* 1659 * Set the bt's gen/txg to the bonus buffer's gen/txg 1660 * so that all of the usual ASSERTs will work. 1661 */ 1662 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg); 1663 } 1664 1665 if (abuf == NULL) { 1666 dmu_write(os, lr->lr_foid, offset, length, data, tx); 1667 } else { 1668 bcopy(data, abuf->b_data, length); 1669 dmu_assign_arcbuf(db, offset, abuf, tx); 1670 } 1671 1672 (void) ztest_log_write(zd, tx, lr); 1673 1674 dmu_buf_rele(db, FTAG); 1675 1676 dmu_tx_commit(tx); 1677 1678 ztest_range_unlock(rl); 1679 ztest_object_unlock(zd, lr->lr_foid); 1680 1681 return (0); 1682} 1683 1684static int 1685ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap) 1686{ 1687 objset_t *os = zd->zd_os; 1688 dmu_tx_t *tx; 1689 uint64_t txg; 1690 rl_t *rl; 1691 1692 if (byteswap) 1693 byteswap_uint64_array(lr, sizeof (*lr)); 1694 1695 ztest_object_lock(zd, lr->lr_foid, RL_READER); 1696 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length, 1697 RL_WRITER); 1698 1699 tx = dmu_tx_create(os); 1700 1701 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length); 1702 1703 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1704 if (txg == 0) { 1705 ztest_range_unlock(rl); 1706 ztest_object_unlock(zd, lr->lr_foid); 1707 return (ENOSPC); 1708 } 1709 1710 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset, 1711 lr->lr_length, tx) == 0); 1712 1713 (void) ztest_log_truncate(zd, tx, lr); 1714 1715 dmu_tx_commit(tx); 1716 1717 ztest_range_unlock(rl); 1718 ztest_object_unlock(zd, lr->lr_foid); 1719 1720 return (0); 1721} 1722 1723static int 1724ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap) 1725{ 1726 objset_t *os = zd->zd_os; 1727 dmu_tx_t *tx; 1728 dmu_buf_t *db; 1729 ztest_block_tag_t *bbt; 1730 uint64_t txg, lrtxg, crtxg; 1731 1732 if (byteswap) 1733 byteswap_uint64_array(lr, sizeof (*lr)); 1734 1735 ztest_object_lock(zd, lr->lr_foid, RL_WRITER); 1736 1737 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1738 1739 tx = dmu_tx_create(os); 1740 dmu_tx_hold_bonus(tx, lr->lr_foid); 1741 1742 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1743 if (txg == 0) { 1744 dmu_buf_rele(db, FTAG); 1745 ztest_object_unlock(zd, lr->lr_foid); 1746 return (ENOSPC); 1747 } 1748 1749 bbt = ztest_bt_bonus(db); 1750 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1751 crtxg = bbt->bt_crtxg; 1752 lrtxg = lr->lr_common.lrc_txg; 1753 1754 if (zd->zd_zilog->zl_replay) { 1755 ASSERT(lr->lr_size != 0); 1756 ASSERT(lr->lr_mode != 0); 1757 ASSERT(lrtxg != 0); 1758 } else { 1759 /* 1760 * Randomly change the size and increment the generation. 1761 */ 1762 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) * 1763 sizeof (*bbt); 1764 lr->lr_mode = bbt->bt_gen + 1; 1765 ASSERT(lrtxg == 0); 1766 } 1767 1768 /* 1769 * Verify that the current bonus buffer is not newer than our txg. 1770 */ 1771 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, 1772 MAX(txg, lrtxg), crtxg); 1773 1774 dmu_buf_will_dirty(db, tx); 1775 1776 ASSERT3U(lr->lr_size, >=, sizeof (*bbt)); 1777 ASSERT3U(lr->lr_size, <=, db->db_size); 1778 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx)); 1779 bbt = ztest_bt_bonus(db); 1780 1781 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg); 1782 1783 dmu_buf_rele(db, FTAG); 1784 1785 (void) ztest_log_setattr(zd, tx, lr); 1786 1787 dmu_tx_commit(tx); 1788 1789 ztest_object_unlock(zd, lr->lr_foid); 1790 1791 return (0); 1792} 1793 1794zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { 1795 NULL, /* 0 no such transaction type */ 1796 ztest_replay_create, /* TX_CREATE */ 1797 NULL, /* TX_MKDIR */ 1798 NULL, /* TX_MKXATTR */ 1799 NULL, /* TX_SYMLINK */ 1800 ztest_replay_remove, /* TX_REMOVE */ 1801 NULL, /* TX_RMDIR */ 1802 NULL, /* TX_LINK */ 1803 NULL, /* TX_RENAME */ 1804 ztest_replay_write, /* TX_WRITE */ 1805 ztest_replay_truncate, /* TX_TRUNCATE */ 1806 ztest_replay_setattr, /* TX_SETATTR */ 1807 NULL, /* TX_ACL */ 1808 NULL, /* TX_CREATE_ACL */ 1809 NULL, /* TX_CREATE_ATTR */ 1810 NULL, /* TX_CREATE_ACL_ATTR */ 1811 NULL, /* TX_MKDIR_ACL */ 1812 NULL, /* TX_MKDIR_ATTR */ 1813 NULL, /* TX_MKDIR_ACL_ATTR */ 1814 NULL, /* TX_WRITE2 */ 1815}; 1816 1817/* 1818 * ZIL get_data callbacks 1819 */ 1820 1821static void 1822ztest_get_done(zgd_t *zgd, int error) 1823{ 1824 ztest_ds_t *zd = zgd->zgd_private; 1825 uint64_t object = zgd->zgd_rl->rl_object; 1826 1827 if (zgd->zgd_db) 1828 dmu_buf_rele(zgd->zgd_db, zgd); 1829 1830 ztest_range_unlock(zgd->zgd_rl); 1831 ztest_object_unlock(zd, object); 1832 1833 if (error == 0 && zgd->zgd_bp) 1834 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1835 1836 umem_free(zgd, sizeof (*zgd)); 1837} 1838 1839static int 1840ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1841{ 1842 ztest_ds_t *zd = arg; 1843 objset_t *os = zd->zd_os; 1844 uint64_t object = lr->lr_foid; 1845 uint64_t offset = lr->lr_offset; 1846 uint64_t size = lr->lr_length; 1847 blkptr_t *bp = &lr->lr_blkptr; 1848 uint64_t txg = lr->lr_common.lrc_txg; 1849 uint64_t crtxg; 1850 dmu_object_info_t doi; 1851 dmu_buf_t *db; 1852 zgd_t *zgd; 1853 int error; 1854 1855 ztest_object_lock(zd, object, RL_READER); 1856 error = dmu_bonus_hold(os, object, FTAG, &db); 1857 if (error) { 1858 ztest_object_unlock(zd, object); 1859 return (error); 1860 } 1861 1862 crtxg = ztest_bt_bonus(db)->bt_crtxg; 1863 1864 if (crtxg == 0 || crtxg > txg) { 1865 dmu_buf_rele(db, FTAG); 1866 ztest_object_unlock(zd, object); 1867 return (ENOENT); 1868 } 1869 1870 dmu_object_info_from_db(db, &doi); 1871 dmu_buf_rele(db, FTAG); 1872 db = NULL; 1873 1874 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL); 1875 zgd->zgd_zilog = zd->zd_zilog; 1876 zgd->zgd_private = zd; 1877 1878 if (buf != NULL) { /* immediate write */ 1879 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, 1880 RL_READER); 1881 1882 error = dmu_read(os, object, offset, size, buf, 1883 DMU_READ_NO_PREFETCH); 1884 ASSERT(error == 0); 1885 } else { 1886 size = doi.doi_data_block_size; 1887 if (ISP2(size)) { 1888 offset = P2ALIGN(offset, size); 1889 } else { 1890 ASSERT(offset < size); 1891 offset = 0; 1892 } 1893 1894 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, 1895 RL_READER); 1896 1897 error = dmu_buf_hold(os, object, offset, zgd, &db, 1898 DMU_READ_NO_PREFETCH); 1899 1900 if (error == 0) { 1901 blkptr_t *obp = dmu_buf_get_blkptr(db); 1902 if (obp) { 1903 ASSERT(BP_IS_HOLE(bp)); 1904 *bp = *obp; 1905 } 1906 1907 zgd->zgd_db = db; 1908 zgd->zgd_bp = bp; 1909 1910 ASSERT(db->db_offset == offset); 1911 ASSERT(db->db_size == size); 1912 1913 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1914 ztest_get_done, zgd); 1915 1916 if (error == 0) 1917 return (0); 1918 } 1919 } 1920 1921 ztest_get_done(zgd, error); 1922 1923 return (error); 1924} 1925 1926static void * 1927ztest_lr_alloc(size_t lrsize, char *name) 1928{ 1929 char *lr; 1930 size_t namesize = name ? strlen(name) + 1 : 0; 1931 1932 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL); 1933 1934 if (name) 1935 bcopy(name, lr + lrsize, namesize); 1936 1937 return (lr); 1938} 1939 1940void 1941ztest_lr_free(void *lr, size_t lrsize, char *name) 1942{ 1943 size_t namesize = name ? strlen(name) + 1 : 0; 1944 1945 umem_free(lr, lrsize + namesize); 1946} 1947 1948/* 1949 * Lookup a bunch of objects. Returns the number of objects not found. 1950 */ 1951static int 1952ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count) 1953{ 1954 int missing = 0; 1955 int error; 1956 1957 ASSERT(_mutex_held(&zd->zd_dirobj_lock)); 1958 1959 for (int i = 0; i < count; i++, od++) { 1960 od->od_object = 0; 1961 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name, 1962 sizeof (uint64_t), 1, &od->od_object); 1963 if (error) { 1964 ASSERT(error == ENOENT); 1965 ASSERT(od->od_object == 0); 1966 missing++; 1967 } else { 1968 dmu_buf_t *db; 1969 ztest_block_tag_t *bbt; 1970 dmu_object_info_t doi; 1971 1972 ASSERT(od->od_object != 0); 1973 ASSERT(missing == 0); /* there should be no gaps */ 1974 1975 ztest_object_lock(zd, od->od_object, RL_READER); 1976 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os, 1977 od->od_object, FTAG, &db)); 1978 dmu_object_info_from_db(db, &doi); 1979 bbt = ztest_bt_bonus(db); 1980 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1981 od->od_type = doi.doi_type; 1982 od->od_blocksize = doi.doi_data_block_size; 1983 od->od_gen = bbt->bt_gen; 1984 dmu_buf_rele(db, FTAG); 1985 ztest_object_unlock(zd, od->od_object); 1986 } 1987 } 1988 1989 return (missing); 1990} 1991 1992static int 1993ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count) 1994{ 1995 int missing = 0; 1996 1997 ASSERT(_mutex_held(&zd->zd_dirobj_lock)); 1998 1999 for (int i = 0; i < count; i++, od++) { 2000 if (missing) { 2001 od->od_object = 0; 2002 missing++; 2003 continue; 2004 } 2005 2006 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); 2007 2008 lr->lr_doid = od->od_dir; 2009 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */ 2010 lr->lrz_type = od->od_crtype; 2011 lr->lrz_blocksize = od->od_crblocksize; 2012 lr->lrz_ibshift = ztest_random_ibshift(); 2013 lr->lrz_bonustype = DMU_OT_UINT64_OTHER; 2014 lr->lrz_bonuslen = dmu_bonus_max(); 2015 lr->lr_gen = od->od_crgen; 2016 lr->lr_crtime[0] = time(NULL); 2017 2018 if (ztest_replay_create(zd, lr, B_FALSE) != 0) { 2019 ASSERT(missing == 0); 2020 od->od_object = 0; 2021 missing++; 2022 } else { 2023 od->od_object = lr->lr_foid; 2024 od->od_type = od->od_crtype; 2025 od->od_blocksize = od->od_crblocksize; 2026 od->od_gen = od->od_crgen; 2027 ASSERT(od->od_object != 0); 2028 } 2029 2030 ztest_lr_free(lr, sizeof (*lr), od->od_name); 2031 } 2032 2033 return (missing); 2034} 2035 2036static int 2037ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count) 2038{ 2039 int missing = 0; 2040 int error; 2041 2042 ASSERT(_mutex_held(&zd->zd_dirobj_lock)); 2043 2044 od += count - 1; 2045 2046 for (int i = count - 1; i >= 0; i--, od--) { 2047 if (missing) { 2048 missing++; 2049 continue; 2050 } 2051 2052 /* 2053 * No object was found. 2054 */ 2055 if (od->od_object == 0) 2056 continue; 2057 2058 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); 2059 2060 lr->lr_doid = od->od_dir; 2061 2062 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) { 2063 ASSERT3U(error, ==, ENOSPC); 2064 missing++; 2065 } else { 2066 od->od_object = 0; 2067 } 2068 ztest_lr_free(lr, sizeof (*lr), od->od_name); 2069 } 2070 2071 return (missing); 2072} 2073 2074static int 2075ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size, 2076 void *data) 2077{ 2078 lr_write_t *lr; 2079 int error; 2080 2081 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL); 2082 2083 lr->lr_foid = object; 2084 lr->lr_offset = offset; 2085 lr->lr_length = size; 2086 lr->lr_blkoff = 0; 2087 BP_ZERO(&lr->lr_blkptr); 2088 2089 bcopy(data, lr + 1, size); 2090 2091 error = ztest_replay_write(zd, lr, B_FALSE); 2092 2093 ztest_lr_free(lr, sizeof (*lr) + size, NULL); 2094 2095 return (error); 2096} 2097 2098static int 2099ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) 2100{ 2101 lr_truncate_t *lr; 2102 int error; 2103 2104 lr = ztest_lr_alloc(sizeof (*lr), NULL); 2105 2106 lr->lr_foid = object; 2107 lr->lr_offset = offset; 2108 lr->lr_length = size; 2109 2110 error = ztest_replay_truncate(zd, lr, B_FALSE); 2111 2112 ztest_lr_free(lr, sizeof (*lr), NULL); 2113 2114 return (error); 2115} 2116 2117static int 2118ztest_setattr(ztest_ds_t *zd, uint64_t object) 2119{ 2120 lr_setattr_t *lr; 2121 int error; 2122 2123 lr = ztest_lr_alloc(sizeof (*lr), NULL); 2124 2125 lr->lr_foid = object; 2126 lr->lr_size = 0; 2127 lr->lr_mode = 0; 2128 2129 error = ztest_replay_setattr(zd, lr, B_FALSE); 2130 2131 ztest_lr_free(lr, sizeof (*lr), NULL); 2132 2133 return (error); 2134} 2135 2136static void 2137ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) 2138{ 2139 objset_t *os = zd->zd_os; 2140 dmu_tx_t *tx; 2141 uint64_t txg; 2142 rl_t *rl; 2143 2144 txg_wait_synced(dmu_objset_pool(os), 0); 2145 2146 ztest_object_lock(zd, object, RL_READER); 2147 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER); 2148 2149 tx = dmu_tx_create(os); 2150 2151 dmu_tx_hold_write(tx, object, offset, size); 2152 2153 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 2154 2155 if (txg != 0) { 2156 dmu_prealloc(os, object, offset, size, tx); 2157 dmu_tx_commit(tx); 2158 txg_wait_synced(dmu_objset_pool(os), txg); 2159 } else { 2160 (void) dmu_free_long_range(os, object, offset, size); 2161 } 2162 2163 ztest_range_unlock(rl); 2164 ztest_object_unlock(zd, object); 2165} 2166 2167static void 2168ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset) 2169{ 2170 int err; 2171 ztest_block_tag_t wbt; 2172 dmu_object_info_t doi; 2173 enum ztest_io_type io_type; 2174 uint64_t blocksize; 2175 void *data; 2176 2177 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0); 2178 blocksize = doi.doi_data_block_size; 2179 data = umem_alloc(blocksize, UMEM_NOFAIL); 2180 2181 /* 2182 * Pick an i/o type at random, biased toward writing block tags. 2183 */ 2184 io_type = ztest_random(ZTEST_IO_TYPES); 2185 if (ztest_random(2) == 0) 2186 io_type = ZTEST_IO_WRITE_TAG; 2187 2188 (void) rw_rdlock(&zd->zd_zilog_lock); 2189 2190 switch (io_type) { 2191 2192 case ZTEST_IO_WRITE_TAG: 2193 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0); 2194 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt); 2195 break; 2196 2197 case ZTEST_IO_WRITE_PATTERN: 2198 (void) memset(data, 'a' + (object + offset) % 5, blocksize); 2199 if (ztest_random(2) == 0) { 2200 /* 2201 * Induce fletcher2 collisions to ensure that 2202 * zio_ddt_collision() detects and resolves them 2203 * when using fletcher2-verify for deduplication. 2204 */ 2205 ((uint64_t *)data)[0] ^= 1ULL << 63; 2206 ((uint64_t *)data)[4] ^= 1ULL << 63; 2207 } 2208 (void) ztest_write(zd, object, offset, blocksize, data); 2209 break; 2210 2211 case ZTEST_IO_WRITE_ZEROES: 2212 bzero(data, blocksize); 2213 (void) ztest_write(zd, object, offset, blocksize, data); 2214 break; 2215 2216 case ZTEST_IO_TRUNCATE: 2217 (void) ztest_truncate(zd, object, offset, blocksize); 2218 break; 2219 2220 case ZTEST_IO_SETATTR: 2221 (void) ztest_setattr(zd, object); 2222 break; 2223 2224 case ZTEST_IO_REWRITE: 2225 (void) rw_rdlock(&ztest_name_lock); 2226 err = ztest_dsl_prop_set_uint64(zd->zd_name, 2227 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa), 2228 B_FALSE); 2229 VERIFY(err == 0 || err == ENOSPC); 2230 err = ztest_dsl_prop_set_uint64(zd->zd_name, 2231 ZFS_PROP_COMPRESSION, 2232 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), 2233 B_FALSE); 2234 VERIFY(err == 0 || err == ENOSPC); 2235 (void) rw_unlock(&ztest_name_lock); 2236 2237 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data, 2238 DMU_READ_NO_PREFETCH)); 2239 2240 (void) ztest_write(zd, object, offset, blocksize, data); 2241 break; 2242 } 2243 2244 (void) rw_unlock(&zd->zd_zilog_lock); 2245 2246 umem_free(data, blocksize); 2247} 2248 2249/* 2250 * Initialize an object description template. 2251 */ 2252static void 2253ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index, 2254 dmu_object_type_t type, uint64_t blocksize, uint64_t gen) 2255{ 2256 od->od_dir = ZTEST_DIROBJ; 2257 od->od_object = 0; 2258 2259 od->od_crtype = type; 2260 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize(); 2261 od->od_crgen = gen; 2262 2263 od->od_type = DMU_OT_NONE; 2264 od->od_blocksize = 0; 2265 od->od_gen = 0; 2266 2267 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]", 2268 tag, (int64_t)id, index); 2269} 2270 2271/* 2272 * Lookup or create the objects for a test using the od template. 2273 * If the objects do not all exist, or if 'remove' is specified, 2274 * remove any existing objects and create new ones. Otherwise, 2275 * use the existing objects. 2276 */ 2277static int 2278ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove) 2279{ 2280 int count = size / sizeof (*od); 2281 int rv = 0; 2282 2283 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0); 2284 if ((ztest_lookup(zd, od, count) != 0 || remove) && 2285 (ztest_remove(zd, od, count) != 0 || 2286 ztest_create(zd, od, count) != 0)) 2287 rv = -1; 2288 zd->zd_od = od; 2289 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0); 2290 2291 return (rv); 2292} 2293 2294/* ARGSUSED */ 2295void 2296ztest_zil_commit(ztest_ds_t *zd, uint64_t id) 2297{ 2298 zilog_t *zilog = zd->zd_zilog; 2299 2300 (void) rw_rdlock(&zd->zd_zilog_lock); 2301 2302 zil_commit(zilog, ztest_random(ZTEST_OBJECTS)); 2303 2304 /* 2305 * Remember the committed values in zd, which is in parent/child 2306 * shared memory. If we die, the next iteration of ztest_run() 2307 * will verify that the log really does contain this record. 2308 */ 2309 mutex_enter(&zilog->zl_lock); 2310 ASSERT(zd->zd_shared != NULL); 2311 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq); 2312 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq; 2313 mutex_exit(&zilog->zl_lock); 2314 2315 (void) rw_unlock(&zd->zd_zilog_lock); 2316} 2317 2318/* 2319 * This function is designed to simulate the operations that occur during a 2320 * mount/unmount operation. We hold the dataset across these operations in an 2321 * attempt to expose any implicit assumptions about ZIL management. 2322 */ 2323/* ARGSUSED */ 2324void 2325ztest_zil_remount(ztest_ds_t *zd, uint64_t id) 2326{ 2327 objset_t *os = zd->zd_os; 2328 2329 /* 2330 * We grab the zd_dirobj_lock to ensure that no other thread is 2331 * updating the zil (i.e. adding in-memory log records) and the 2332 * zd_zilog_lock to block any I/O. 2333 */ 2334 VERIFY0(mutex_lock(&zd->zd_dirobj_lock)); 2335 (void) rw_wrlock(&zd->zd_zilog_lock); 2336 2337 /* zfsvfs_teardown() */ 2338 zil_close(zd->zd_zilog); 2339 2340 /* zfsvfs_setup() */ 2341 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog); 2342 zil_replay(os, zd, ztest_replay_vector); 2343 2344 (void) rw_unlock(&zd->zd_zilog_lock); 2345 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0); 2346} 2347 2348/* 2349 * Verify that we can't destroy an active pool, create an existing pool, 2350 * or create a pool with a bad vdev spec. 2351 */ 2352/* ARGSUSED */ 2353void 2354ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id) 2355{ 2356 ztest_shared_opts_t *zo = &ztest_opts; 2357 spa_t *spa; 2358 nvlist_t *nvroot; 2359 2360 /* 2361 * Attempt to create using a bad file. 2362 */ 2363 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); 2364 VERIFY3U(ENOENT, ==, 2365 spa_create("ztest_bad_file", nvroot, NULL, NULL)); 2366 nvlist_free(nvroot); 2367 2368 /* 2369 * Attempt to create using a bad mirror. 2370 */ 2371 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1); 2372 VERIFY3U(ENOENT, ==, 2373 spa_create("ztest_bad_mirror", nvroot, NULL, NULL)); 2374 nvlist_free(nvroot); 2375 2376 /* 2377 * Attempt to create an existing pool. It shouldn't matter 2378 * what's in the nvroot; we should fail with EEXIST. 2379 */ 2380 (void) rw_rdlock(&ztest_name_lock); 2381 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); 2382 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL)); 2383 nvlist_free(nvroot); 2384 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG)); 2385 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool)); 2386 spa_close(spa, FTAG); 2387 2388 (void) rw_unlock(&ztest_name_lock); 2389} 2390 2391/* ARGSUSED */ 2392void 2393ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id) 2394{ 2395 spa_t *spa; 2396 uint64_t initial_version = SPA_VERSION_INITIAL; 2397 uint64_t version, newversion; 2398 nvlist_t *nvroot, *props; 2399 char *name; 2400 2401 VERIFY0(mutex_lock(&ztest_vdev_lock)); 2402 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool); 2403 2404 /* 2405 * Clean up from previous runs. 2406 */ 2407 (void) spa_destroy(name); 2408 2409 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0, 2410 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1); 2411 2412 /* 2413 * If we're configuring a RAIDZ device then make sure that the 2414 * the initial version is capable of supporting that feature. 2415 */ 2416 switch (ztest_opts.zo_raidz_parity) { 2417 case 0: 2418 case 1: 2419 initial_version = SPA_VERSION_INITIAL; 2420 break; 2421 case 2: 2422 initial_version = SPA_VERSION_RAIDZ2; 2423 break; 2424 case 3: 2425 initial_version = SPA_VERSION_RAIDZ3; 2426 break; 2427 } 2428 2429 /* 2430 * Create a pool with a spa version that can be upgraded. Pick 2431 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES. 2432 */ 2433 do { 2434 version = ztest_random_spa_version(initial_version); 2435 } while (version > SPA_VERSION_BEFORE_FEATURES); 2436 2437 props = fnvlist_alloc(); 2438 fnvlist_add_uint64(props, 2439 zpool_prop_to_name(ZPOOL_PROP_VERSION), version); 2440 VERIFY0(spa_create(name, nvroot, props, NULL)); 2441 fnvlist_free(nvroot); 2442 fnvlist_free(props); 2443 2444 VERIFY0(spa_open(name, &spa, FTAG)); 2445 VERIFY3U(spa_version(spa), ==, version); 2446 newversion = ztest_random_spa_version(version + 1); 2447 2448 if (ztest_opts.zo_verbose >= 4) { 2449 (void) printf("upgrading spa version from %llu to %llu\n", 2450 (u_longlong_t)version, (u_longlong_t)newversion); 2451 } 2452 2453 spa_upgrade(spa, newversion); 2454 VERIFY3U(spa_version(spa), >, version); 2455 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config, 2456 zpool_prop_to_name(ZPOOL_PROP_VERSION))); 2457 spa_close(spa, FTAG); 2458 2459 strfree(name); 2460 VERIFY0(mutex_unlock(&ztest_vdev_lock)); 2461} 2462 2463static vdev_t * 2464vdev_lookup_by_path(vdev_t *vd, const char *path) 2465{ 2466 vdev_t *mvd; 2467 2468 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0) 2469 return (vd); 2470 2471 for (int c = 0; c < vd->vdev_children; c++) 2472 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != 2473 NULL) 2474 return (mvd); 2475 2476 return (NULL); 2477} 2478 2479/* 2480 * Find the first available hole which can be used as a top-level. 2481 */ 2482int 2483find_vdev_hole(spa_t *spa) 2484{ 2485 vdev_t *rvd = spa->spa_root_vdev; 2486 int c; 2487 2488 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV); 2489 2490 for (c = 0; c < rvd->vdev_children; c++) { 2491 vdev_t *cvd = rvd->vdev_child[c]; 2492 2493 if (cvd->vdev_ishole) 2494 break; 2495 } 2496 return (c); 2497} 2498 2499/* 2500 * Verify that vdev_add() works as expected. 2501 */ 2502/* ARGSUSED */ 2503void 2504ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id) 2505{ 2506 ztest_shared_t *zs = ztest_shared; 2507 spa_t *spa = ztest_spa; 2508 uint64_t leaves; 2509 uint64_t guid; 2510 nvlist_t *nvroot; 2511 int error; 2512 2513 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2514 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz; 2515 2516 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2517 2518 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves; 2519 2520 /* 2521 * If we have slogs then remove them 1/4 of the time. 2522 */ 2523 if (spa_has_slogs(spa) && ztest_random(4) == 0) { 2524 /* 2525 * Grab the guid from the head of the log class rotor. 2526 */ 2527 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid; 2528 2529 spa_config_exit(spa, SCL_VDEV, FTAG); 2530 2531 /* 2532 * We have to grab the zs_name_lock as writer to 2533 * prevent a race between removing a slog (dmu_objset_find) 2534 * and destroying a dataset. Removing the slog will 2535 * grab a reference on the dataset which may cause 2536 * dmu_objset_destroy() to fail with EBUSY thus 2537 * leaving the dataset in an inconsistent state. 2538 */ 2539 VERIFY(rw_wrlock(&ztest_name_lock) == 0); 2540 error = spa_vdev_remove(spa, guid, B_FALSE); 2541 VERIFY(rw_unlock(&ztest_name_lock) == 0); 2542 2543 if (error && error != EEXIST) 2544 fatal(0, "spa_vdev_remove() = %d", error); 2545 } else { 2546 spa_config_exit(spa, SCL_VDEV, FTAG); 2547 2548 /* 2549 * Make 1/4 of the devices be log devices. 2550 */ 2551 nvroot = make_vdev_root(NULL, NULL, NULL, 2552 ztest_opts.zo_vdev_size, 0, 2553 ztest_random(4) == 0, ztest_opts.zo_raidz, 2554 zs->zs_mirrors, 1); 2555 2556 error = spa_vdev_add(spa, nvroot); 2557 nvlist_free(nvroot); 2558 2559 if (error == ENOSPC) 2560 ztest_record_enospc("spa_vdev_add"); 2561 else if (error != 0) 2562 fatal(0, "spa_vdev_add() = %d", error); 2563 } 2564 2565 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2566} 2567 2568/* 2569 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected. 2570 */ 2571/* ARGSUSED */ 2572void 2573ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id) 2574{ 2575 ztest_shared_t *zs = ztest_shared; 2576 spa_t *spa = ztest_spa; 2577 vdev_t *rvd = spa->spa_root_vdev; 2578 spa_aux_vdev_t *sav; 2579 char *aux; 2580 uint64_t guid = 0; 2581 int error; 2582 2583 if (ztest_random(2) == 0) { 2584 sav = &spa->spa_spares; 2585 aux = ZPOOL_CONFIG_SPARES; 2586 } else { 2587 sav = &spa->spa_l2cache; 2588 aux = ZPOOL_CONFIG_L2CACHE; 2589 } 2590 2591 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2592 2593 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2594 2595 if (sav->sav_count != 0 && ztest_random(4) == 0) { 2596 /* 2597 * Pick a random device to remove. 2598 */ 2599 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid; 2600 } else { 2601 /* 2602 * Find an unused device we can add. 2603 */ 2604 zs->zs_vdev_aux = 0; 2605 for (;;) { 2606 char path[MAXPATHLEN]; 2607 int c; 2608 (void) snprintf(path, sizeof (path), ztest_aux_template, 2609 ztest_opts.zo_dir, ztest_opts.zo_pool, aux, 2610 zs->zs_vdev_aux); 2611 for (c = 0; c < sav->sav_count; c++) 2612 if (strcmp(sav->sav_vdevs[c]->vdev_path, 2613 path) == 0) 2614 break; 2615 if (c == sav->sav_count && 2616 vdev_lookup_by_path(rvd, path) == NULL) 2617 break; 2618 zs->zs_vdev_aux++; 2619 } 2620 } 2621 2622 spa_config_exit(spa, SCL_VDEV, FTAG); 2623 2624 if (guid == 0) { 2625 /* 2626 * Add a new device. 2627 */ 2628 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL, 2629 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1); 2630 error = spa_vdev_add(spa, nvroot); 2631 if (error != 0) 2632 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error); 2633 nvlist_free(nvroot); 2634 } else { 2635 /* 2636 * Remove an existing device. Sometimes, dirty its 2637 * vdev state first to make sure we handle removal 2638 * of devices that have pending state changes. 2639 */ 2640 if (ztest_random(2) == 0) 2641 (void) vdev_online(spa, guid, 0, NULL); 2642 2643 error = spa_vdev_remove(spa, guid, B_FALSE); 2644 if (error != 0 && error != EBUSY) 2645 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error); 2646 } 2647 2648 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2649} 2650 2651/* 2652 * split a pool if it has mirror tlvdevs 2653 */ 2654/* ARGSUSED */ 2655void 2656ztest_split_pool(ztest_ds_t *zd, uint64_t id) 2657{ 2658 ztest_shared_t *zs = ztest_shared; 2659 spa_t *spa = ztest_spa; 2660 vdev_t *rvd = spa->spa_root_vdev; 2661 nvlist_t *tree, **child, *config, *split, **schild; 2662 uint_t c, children, schildren = 0, lastlogid = 0; 2663 int error = 0; 2664 2665 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2666 2667 /* ensure we have a useable config; mirrors of raidz aren't supported */ 2668 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) { 2669 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2670 return; 2671 } 2672 2673 /* clean up the old pool, if any */ 2674 (void) spa_destroy("splitp"); 2675 2676 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2677 2678 /* generate a config from the existing config */ 2679 mutex_enter(&spa->spa_props_lock); 2680 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE, 2681 &tree) == 0); 2682 mutex_exit(&spa->spa_props_lock); 2683 2684 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child, 2685 &children) == 0); 2686 2687 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *)); 2688 for (c = 0; c < children; c++) { 2689 vdev_t *tvd = rvd->vdev_child[c]; 2690 nvlist_t **mchild; 2691 uint_t mchildren; 2692 2693 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) { 2694 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME, 2695 0) == 0); 2696 VERIFY(nvlist_add_string(schild[schildren], 2697 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); 2698 VERIFY(nvlist_add_uint64(schild[schildren], 2699 ZPOOL_CONFIG_IS_HOLE, 1) == 0); 2700 if (lastlogid == 0) 2701 lastlogid = schildren; 2702 ++schildren; 2703 continue; 2704 } 2705 lastlogid = 0; 2706 VERIFY(nvlist_lookup_nvlist_array(child[c], 2707 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0); 2708 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0); 2709 } 2710 2711 /* OK, create a config that can be used to split */ 2712 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0); 2713 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE, 2714 VDEV_TYPE_ROOT) == 0); 2715 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild, 2716 lastlogid != 0 ? lastlogid : schildren) == 0); 2717 2718 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0); 2719 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0); 2720 2721 for (c = 0; c < schildren; c++) 2722 nvlist_free(schild[c]); 2723 free(schild); 2724 nvlist_free(split); 2725 2726 spa_config_exit(spa, SCL_VDEV, FTAG); 2727 2728 (void) rw_wrlock(&ztest_name_lock); 2729 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE); 2730 (void) rw_unlock(&ztest_name_lock); 2731 2732 nvlist_free(config); 2733 2734 if (error == 0) { 2735 (void) printf("successful split - results:\n"); 2736 mutex_enter(&spa_namespace_lock); 2737 show_pool_stats(spa); 2738 show_pool_stats(spa_lookup("splitp")); 2739 mutex_exit(&spa_namespace_lock); 2740 ++zs->zs_splits; 2741 --zs->zs_mirrors; 2742 } 2743 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2744 2745} 2746 2747/* 2748 * Verify that we can attach and detach devices. 2749 */ 2750/* ARGSUSED */ 2751void 2752ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id) 2753{ 2754 ztest_shared_t *zs = ztest_shared; 2755 spa_t *spa = ztest_spa; 2756 spa_aux_vdev_t *sav = &spa->spa_spares; 2757 vdev_t *rvd = spa->spa_root_vdev; 2758 vdev_t *oldvd, *newvd, *pvd; 2759 nvlist_t *root; 2760 uint64_t leaves; 2761 uint64_t leaf, top; 2762 uint64_t ashift = ztest_get_ashift(); 2763 uint64_t oldguid, pguid; 2764 uint64_t oldsize, newsize; 2765 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; 2766 int replacing; 2767 int oldvd_has_siblings = B_FALSE; 2768 int newvd_is_spare = B_FALSE; 2769 int oldvd_is_log; 2770 int error, expected_error; 2771 2772 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2773 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; 2774 2775 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2776 2777 /* 2778 * Decide whether to do an attach or a replace. 2779 */ 2780 replacing = ztest_random(2); 2781 2782 /* 2783 * Pick a random top-level vdev. 2784 */ 2785 top = ztest_random_vdev_top(spa, B_TRUE); 2786 2787 /* 2788 * Pick a random leaf within it. 2789 */ 2790 leaf = ztest_random(leaves); 2791 2792 /* 2793 * Locate this vdev. 2794 */ 2795 oldvd = rvd->vdev_child[top]; 2796 if (zs->zs_mirrors >= 1) { 2797 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops); 2798 ASSERT(oldvd->vdev_children >= zs->zs_mirrors); 2799 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz]; 2800 } 2801 if (ztest_opts.zo_raidz > 1) { 2802 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops); 2803 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz); 2804 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz]; 2805 } 2806 2807 /* 2808 * If we're already doing an attach or replace, oldvd may be a 2809 * mirror vdev -- in which case, pick a random child. 2810 */ 2811 while (oldvd->vdev_children != 0) { 2812 oldvd_has_siblings = B_TRUE; 2813 ASSERT(oldvd->vdev_children >= 2); 2814 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)]; 2815 } 2816 2817 oldguid = oldvd->vdev_guid; 2818 oldsize = vdev_get_min_asize(oldvd); 2819 oldvd_is_log = oldvd->vdev_top->vdev_islog; 2820 (void) strcpy(oldpath, oldvd->vdev_path); 2821 pvd = oldvd->vdev_parent; 2822 pguid = pvd->vdev_guid; 2823 2824 /* 2825 * If oldvd has siblings, then half of the time, detach it. 2826 */ 2827 if (oldvd_has_siblings && ztest_random(2) == 0) { 2828 spa_config_exit(spa, SCL_VDEV, FTAG); 2829 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE); 2830 if (error != 0 && error != ENODEV && error != EBUSY && 2831 error != ENOTSUP) 2832 fatal(0, "detach (%s) returned %d", oldpath, error); 2833 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2834 return; 2835 } 2836 2837 /* 2838 * For the new vdev, choose with equal probability between the two 2839 * standard paths (ending in either 'a' or 'b') or a random hot spare. 2840 */ 2841 if (sav->sav_count != 0 && ztest_random(3) == 0) { 2842 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)]; 2843 newvd_is_spare = B_TRUE; 2844 (void) strcpy(newpath, newvd->vdev_path); 2845 } else { 2846 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template, 2847 ztest_opts.zo_dir, ztest_opts.zo_pool, 2848 top * leaves + leaf); 2849 if (ztest_random(2) == 0) 2850 newpath[strlen(newpath) - 1] = 'b'; 2851 newvd = vdev_lookup_by_path(rvd, newpath); 2852 } 2853 2854 if (newvd) { 2855 newsize = vdev_get_min_asize(newvd); 2856 } else { 2857 /* 2858 * Make newsize a little bigger or smaller than oldsize. 2859 * If it's smaller, the attach should fail. 2860 * If it's larger, and we're doing a replace, 2861 * we should get dynamic LUN growth when we're done. 2862 */ 2863 newsize = 10 * oldsize / (9 + ztest_random(3)); 2864 } 2865 2866 /* 2867 * If pvd is not a mirror or root, the attach should fail with ENOTSUP, 2868 * unless it's a replace; in that case any non-replacing parent is OK. 2869 * 2870 * If newvd is already part of the pool, it should fail with EBUSY. 2871 * 2872 * If newvd is too small, it should fail with EOVERFLOW. 2873 */ 2874 if (pvd->vdev_ops != &vdev_mirror_ops && 2875 pvd->vdev_ops != &vdev_root_ops && (!replacing || 2876 pvd->vdev_ops == &vdev_replacing_ops || 2877 pvd->vdev_ops == &vdev_spare_ops)) 2878 expected_error = ENOTSUP; 2879 else if (newvd_is_spare && (!replacing || oldvd_is_log)) 2880 expected_error = ENOTSUP; 2881 else if (newvd == oldvd) 2882 expected_error = replacing ? 0 : EBUSY; 2883 else if (vdev_lookup_by_path(rvd, newpath) != NULL) 2884 expected_error = EBUSY; 2885 else if (newsize < oldsize) 2886 expected_error = EOVERFLOW; 2887 else if (ashift > oldvd->vdev_top->vdev_ashift) 2888 expected_error = EDOM; 2889 else 2890 expected_error = 0; 2891 2892 spa_config_exit(spa, SCL_VDEV, FTAG); 2893 2894 /* 2895 * Build the nvlist describing newpath. 2896 */ 2897 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0, 2898 ashift, 0, 0, 0, 1); 2899 2900 error = spa_vdev_attach(spa, oldguid, root, replacing); 2901 2902 nvlist_free(root); 2903 2904 /* 2905 * If our parent was the replacing vdev, but the replace completed, 2906 * then instead of failing with ENOTSUP we may either succeed, 2907 * fail with ENODEV, or fail with EOVERFLOW. 2908 */ 2909 if (expected_error == ENOTSUP && 2910 (error == 0 || error == ENODEV || error == EOVERFLOW)) 2911 expected_error = error; 2912 2913 /* 2914 * If someone grew the LUN, the replacement may be too small. 2915 */ 2916 if (error == EOVERFLOW || error == EBUSY) 2917 expected_error = error; 2918 2919 /* XXX workaround 6690467 */ 2920 if (error != expected_error && expected_error != EBUSY) { 2921 fatal(0, "attach (%s %llu, %s %llu, %d) " 2922 "returned %d, expected %d", 2923 oldpath, oldsize, newpath, 2924 newsize, replacing, error, expected_error); 2925 } 2926 2927 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2928} 2929 2930/* 2931 * Callback function which expands the physical size of the vdev. 2932 */ 2933vdev_t * 2934grow_vdev(vdev_t *vd, void *arg) 2935{ 2936 spa_t *spa = vd->vdev_spa; 2937 size_t *newsize = arg; 2938 size_t fsize; 2939 int fd; 2940 2941 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); 2942 ASSERT(vd->vdev_ops->vdev_op_leaf); 2943 2944 if ((fd = open(vd->vdev_path, O_RDWR)) == -1) 2945 return (vd); 2946 2947 fsize = lseek(fd, 0, SEEK_END); 2948 (void) ftruncate(fd, *newsize); 2949 2950 if (ztest_opts.zo_verbose >= 6) { 2951 (void) printf("%s grew from %lu to %lu bytes\n", 2952 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize); 2953 } 2954 (void) close(fd); 2955 return (NULL); 2956} 2957 2958/* 2959 * Callback function which expands a given vdev by calling vdev_online(). 2960 */ 2961/* ARGSUSED */ 2962vdev_t * 2963online_vdev(vdev_t *vd, void *arg) 2964{ 2965 spa_t *spa = vd->vdev_spa; 2966 vdev_t *tvd = vd->vdev_top; 2967 uint64_t guid = vd->vdev_guid; 2968 uint64_t generation = spa->spa_config_generation + 1; 2969 vdev_state_t newstate = VDEV_STATE_UNKNOWN; 2970 int error; 2971 2972 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); 2973 ASSERT(vd->vdev_ops->vdev_op_leaf); 2974 2975 /* Calling vdev_online will initialize the new metaslabs */ 2976 spa_config_exit(spa, SCL_STATE, spa); 2977 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate); 2978 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 2979 2980 /* 2981 * If vdev_online returned an error or the underlying vdev_open 2982 * failed then we abort the expand. The only way to know that 2983 * vdev_open fails is by checking the returned newstate. 2984 */ 2985 if (error || newstate != VDEV_STATE_HEALTHY) { 2986 if (ztest_opts.zo_verbose >= 5) { 2987 (void) printf("Unable to expand vdev, state %llu, " 2988 "error %d\n", (u_longlong_t)newstate, error); 2989 } 2990 return (vd); 2991 } 2992 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY); 2993 2994 /* 2995 * Since we dropped the lock we need to ensure that we're 2996 * still talking to the original vdev. It's possible this 2997 * vdev may have been detached/replaced while we were 2998 * trying to online it. 2999 */ 3000 if (generation != spa->spa_config_generation) { 3001 if (ztest_opts.zo_verbose >= 5) { 3002 (void) printf("vdev configuration has changed, " 3003 "guid %llu, state %llu, expected gen %llu, " 3004 "got gen %llu\n", 3005 (u_longlong_t)guid, 3006 (u_longlong_t)tvd->vdev_state, 3007 (u_longlong_t)generation, 3008 (u_longlong_t)spa->spa_config_generation); 3009 } 3010 return (vd); 3011 } 3012 return (NULL); 3013} 3014 3015/* 3016 * Traverse the vdev tree calling the supplied function. 3017 * We continue to walk the tree until we either have walked all 3018 * children or we receive a non-NULL return from the callback. 3019 * If a NULL callback is passed, then we just return back the first 3020 * leaf vdev we encounter. 3021 */ 3022vdev_t * 3023vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg) 3024{ 3025 if (vd->vdev_ops->vdev_op_leaf) { 3026 if (func == NULL) 3027 return (vd); 3028 else 3029 return (func(vd, arg)); 3030 } 3031 3032 for (uint_t c = 0; c < vd->vdev_children; c++) { 3033 vdev_t *cvd = vd->vdev_child[c]; 3034 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL) 3035 return (cvd); 3036 } 3037 return (NULL); 3038} 3039 3040/* 3041 * Verify that dynamic LUN growth works as expected. 3042 */ 3043/* ARGSUSED */ 3044void 3045ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id) 3046{ 3047 spa_t *spa = ztest_spa; 3048 vdev_t *vd, *tvd; 3049 metaslab_class_t *mc; 3050 metaslab_group_t *mg; 3051 size_t psize, newsize; 3052 uint64_t top; 3053 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count; 3054 3055 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 3056 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3057 3058 top = ztest_random_vdev_top(spa, B_TRUE); 3059 3060 tvd = spa->spa_root_vdev->vdev_child[top]; 3061 mg = tvd->vdev_mg; 3062 mc = mg->mg_class; 3063 old_ms_count = tvd->vdev_ms_count; 3064 old_class_space = metaslab_class_get_space(mc); 3065 3066 /* 3067 * Determine the size of the first leaf vdev associated with 3068 * our top-level device. 3069 */ 3070 vd = vdev_walk_tree(tvd, NULL, NULL); 3071 ASSERT3P(vd, !=, NULL); 3072 ASSERT(vd->vdev_ops->vdev_op_leaf); 3073 3074 psize = vd->vdev_psize; 3075 3076 /* 3077 * We only try to expand the vdev if it's healthy, less than 4x its 3078 * original size, and it has a valid psize. 3079 */ 3080 if (tvd->vdev_state != VDEV_STATE_HEALTHY || 3081 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) { 3082 spa_config_exit(spa, SCL_STATE, spa); 3083 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3084 return; 3085 } 3086 ASSERT(psize > 0); 3087 newsize = psize + psize / 8; 3088 ASSERT3U(newsize, >, psize); 3089 3090 if (ztest_opts.zo_verbose >= 6) { 3091 (void) printf("Expanding LUN %s from %lu to %lu\n", 3092 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize); 3093 } 3094 3095 /* 3096 * Growing the vdev is a two step process: 3097 * 1). expand the physical size (i.e. relabel) 3098 * 2). online the vdev to create the new metaslabs 3099 */ 3100 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL || 3101 vdev_walk_tree(tvd, online_vdev, NULL) != NULL || 3102 tvd->vdev_state != VDEV_STATE_HEALTHY) { 3103 if (ztest_opts.zo_verbose >= 5) { 3104 (void) printf("Could not expand LUN because " 3105 "the vdev configuration changed.\n"); 3106 } 3107 spa_config_exit(spa, SCL_STATE, spa); 3108 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3109 return; 3110 } 3111 3112 spa_config_exit(spa, SCL_STATE, spa); 3113 3114 /* 3115 * Expanding the LUN will update the config asynchronously, 3116 * thus we must wait for the async thread to complete any 3117 * pending tasks before proceeding. 3118 */ 3119 for (;;) { 3120 boolean_t done; 3121 mutex_enter(&spa->spa_async_lock); 3122 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks); 3123 mutex_exit(&spa->spa_async_lock); 3124 if (done) 3125 break; 3126 txg_wait_synced(spa_get_dsl(spa), 0); 3127 (void) poll(NULL, 0, 100); 3128 } 3129 3130 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3131 3132 tvd = spa->spa_root_vdev->vdev_child[top]; 3133 new_ms_count = tvd->vdev_ms_count; 3134 new_class_space = metaslab_class_get_space(mc); 3135 3136 if (tvd->vdev_mg != mg || mg->mg_class != mc) { 3137 if (ztest_opts.zo_verbose >= 5) { 3138 (void) printf("Could not verify LUN expansion due to " 3139 "intervening vdev offline or remove.\n"); 3140 } 3141 spa_config_exit(spa, SCL_STATE, spa); 3142 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3143 return; 3144 } 3145 3146 /* 3147 * Make sure we were able to grow the vdev. 3148 */ 3149 if (new_ms_count <= old_ms_count) 3150 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n", 3151 old_ms_count, new_ms_count); 3152 3153 /* 3154 * Make sure we were able to grow the pool. 3155 */ 3156 if (new_class_space <= old_class_space) 3157 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n", 3158 old_class_space, new_class_space); 3159 3160 if (ztest_opts.zo_verbose >= 5) { 3161 char oldnumbuf[6], newnumbuf[6]; 3162 3163 nicenum(old_class_space, oldnumbuf); 3164 nicenum(new_class_space, newnumbuf); 3165 (void) printf("%s grew from %s to %s\n", 3166 spa->spa_name, oldnumbuf, newnumbuf); 3167 } 3168 3169 spa_config_exit(spa, SCL_STATE, spa); 3170 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3171} 3172 3173/* 3174 * Verify that dmu_objset_{create,destroy,open,close} work as expected. 3175 */ 3176/* ARGSUSED */ 3177static void 3178ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 3179{ 3180 /* 3181 * Create the objects common to all ztest datasets. 3182 */ 3183 VERIFY(zap_create_claim(os, ZTEST_DIROBJ, 3184 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 3185} 3186 3187static int 3188ztest_dataset_create(char *dsname) 3189{ 3190 uint64_t zilset = ztest_random(100); 3191 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, 3192 ztest_objset_create_cb, NULL); 3193 3194 if (err || zilset < 80) 3195 return (err); 3196 3197 if (ztest_opts.zo_verbose >= 6) 3198 (void) printf("Setting dataset %s to sync always\n", dsname); 3199 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC, 3200 ZFS_SYNC_ALWAYS, B_FALSE)); 3201} 3202 3203/* ARGSUSED */ 3204static int 3205ztest_objset_destroy_cb(const char *name, void *arg) 3206{ 3207 objset_t *os; 3208 dmu_object_info_t doi; 3209 int error; 3210 3211 /* 3212 * Verify that the dataset contains a directory object. 3213 */ 3214 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os)); 3215 error = dmu_object_info(os, ZTEST_DIROBJ, &doi); 3216 if (error != ENOENT) { 3217 /* We could have crashed in the middle of destroying it */ 3218 ASSERT0(error); 3219 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER); 3220 ASSERT3S(doi.doi_physical_blocks_512, >=, 0); 3221 } 3222 dmu_objset_disown(os, FTAG); 3223 3224 /* 3225 * Destroy the dataset. 3226 */ 3227 if (strchr(name, '@') != NULL) { 3228 VERIFY0(dsl_destroy_snapshot(name, B_FALSE)); 3229 } else { 3230 VERIFY0(dsl_destroy_head(name)); 3231 } 3232 return (0); 3233} 3234 3235static boolean_t 3236ztest_snapshot_create(char *osname, uint64_t id) 3237{ 3238 char snapname[MAXNAMELEN]; 3239 int error; 3240 3241 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id); 3242 3243 error = dmu_objset_snapshot_one(osname, snapname); 3244 if (error == ENOSPC) { 3245 ztest_record_enospc(FTAG); 3246 return (B_FALSE); 3247 } 3248 if (error != 0 && error != EEXIST) { 3249 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname, 3250 snapname, error); 3251 } 3252 return (B_TRUE); 3253} 3254 3255static boolean_t 3256ztest_snapshot_destroy(char *osname, uint64_t id) 3257{ 3258 char snapname[MAXNAMELEN]; 3259 int error; 3260 3261 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, 3262 (u_longlong_t)id); 3263 3264 error = dsl_destroy_snapshot(snapname, B_FALSE); 3265 if (error != 0 && error != ENOENT) 3266 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error); 3267 return (B_TRUE); 3268} 3269 3270/* ARGSUSED */ 3271void 3272ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) 3273{ 3274 ztest_ds_t zdtmp; 3275 int iters; 3276 int error; 3277 objset_t *os, *os2; 3278 char name[MAXNAMELEN]; 3279 zilog_t *zilog; 3280 3281 (void) rw_rdlock(&ztest_name_lock); 3282 3283 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu", 3284 ztest_opts.zo_pool, (u_longlong_t)id); 3285 3286 /* 3287 * If this dataset exists from a previous run, process its replay log 3288 * half of the time. If we don't replay it, then dmu_objset_destroy() 3289 * (invoked from ztest_objset_destroy_cb()) should just throw it away. 3290 */ 3291 if (ztest_random(2) == 0 && 3292 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) { 3293 ztest_zd_init(&zdtmp, NULL, os); 3294 zil_replay(os, &zdtmp, ztest_replay_vector); 3295 ztest_zd_fini(&zdtmp); 3296 dmu_objset_disown(os, FTAG); 3297 } 3298 3299 /* 3300 * There may be an old instance of the dataset we're about to 3301 * create lying around from a previous run. If so, destroy it 3302 * and all of its snapshots. 3303 */ 3304 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, 3305 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 3306 3307 /* 3308 * Verify that the destroyed dataset is no longer in the namespace. 3309 */ 3310 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, 3311 FTAG, &os)); 3312 3313 /* 3314 * Verify that we can create a new dataset. 3315 */ 3316 error = ztest_dataset_create(name); 3317 if (error) { 3318 if (error == ENOSPC) { 3319 ztest_record_enospc(FTAG); 3320 (void) rw_unlock(&ztest_name_lock); 3321 return; 3322 } 3323 fatal(0, "dmu_objset_create(%s) = %d", name, error); 3324 } 3325 3326 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); 3327 3328 ztest_zd_init(&zdtmp, NULL, os); 3329 3330 /* 3331 * Open the intent log for it. 3332 */ 3333 zilog = zil_open(os, ztest_get_data); 3334 3335 /* 3336 * Put some objects in there, do a little I/O to them, 3337 * and randomly take a couple of snapshots along the way. 3338 */ 3339 iters = ztest_random(5); 3340 for (int i = 0; i < iters; i++) { 3341 ztest_dmu_object_alloc_free(&zdtmp, id); 3342 if (ztest_random(iters) == 0) 3343 (void) ztest_snapshot_create(name, i); 3344 } 3345 3346 /* 3347 * Verify that we cannot create an existing dataset. 3348 */ 3349 VERIFY3U(EEXIST, ==, 3350 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL)); 3351 3352 /* 3353 * Verify that we can hold an objset that is also owned. 3354 */ 3355 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2)); 3356 dmu_objset_rele(os2, FTAG); 3357 3358 /* 3359 * Verify that we cannot own an objset that is already owned. 3360 */ 3361 VERIFY3U(EBUSY, ==, 3362 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2)); 3363 3364 zil_close(zilog); 3365 dmu_objset_disown(os, FTAG); 3366 ztest_zd_fini(&zdtmp); 3367 3368 (void) rw_unlock(&ztest_name_lock); 3369} 3370 3371/* 3372 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. 3373 */ 3374void 3375ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id) 3376{ 3377 (void) rw_rdlock(&ztest_name_lock); 3378 (void) ztest_snapshot_destroy(zd->zd_name, id); 3379 (void) ztest_snapshot_create(zd->zd_name, id); 3380 (void) rw_unlock(&ztest_name_lock); 3381} 3382 3383/* 3384 * Cleanup non-standard snapshots and clones. 3385 */ 3386void 3387ztest_dsl_dataset_cleanup(char *osname, uint64_t id) 3388{ 3389 char snap1name[MAXNAMELEN]; 3390 char clone1name[MAXNAMELEN]; 3391 char snap2name[MAXNAMELEN]; 3392 char clone2name[MAXNAMELEN]; 3393 char snap3name[MAXNAMELEN]; 3394 int error; 3395 3396 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id); 3397 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id); 3398 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id); 3399 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id); 3400 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id); 3401 3402 error = dsl_destroy_head(clone2name); 3403 if (error && error != ENOENT) 3404 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error); 3405 error = dsl_destroy_snapshot(snap3name, B_FALSE); 3406 if (error && error != ENOENT) 3407 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error); 3408 error = dsl_destroy_snapshot(snap2name, B_FALSE); 3409 if (error && error != ENOENT) 3410 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error); 3411 error = dsl_destroy_head(clone1name); 3412 if (error && error != ENOENT) 3413 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error); 3414 error = dsl_destroy_snapshot(snap1name, B_FALSE); 3415 if (error && error != ENOENT) 3416 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error); 3417} 3418 3419/* 3420 * Verify dsl_dataset_promote handles EBUSY 3421 */ 3422void 3423ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) 3424{ 3425 objset_t *os; 3426 char snap1name[MAXNAMELEN]; 3427 char clone1name[MAXNAMELEN]; 3428 char snap2name[MAXNAMELEN]; 3429 char clone2name[MAXNAMELEN]; 3430 char snap3name[MAXNAMELEN]; 3431 char *osname = zd->zd_name; 3432 int error; 3433 3434 (void) rw_rdlock(&ztest_name_lock); 3435 3436 ztest_dsl_dataset_cleanup(osname, id); 3437 3438 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id); 3439 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id); 3440 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id); 3441 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id); 3442 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id); 3443 3444 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1); 3445 if (error && error != EEXIST) { 3446 if (error == ENOSPC) { 3447 ztest_record_enospc(FTAG); 3448 goto out; 3449 } 3450 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error); 3451 } 3452 3453 error = dmu_objset_clone(clone1name, snap1name); 3454 if (error) { 3455 if (error == ENOSPC) { 3456 ztest_record_enospc(FTAG); 3457 goto out; 3458 } 3459 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error); 3460 } 3461 3462 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1); 3463 if (error && error != EEXIST) { 3464 if (error == ENOSPC) { 3465 ztest_record_enospc(FTAG); 3466 goto out; 3467 } 3468 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error); 3469 } 3470 3471 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1); 3472 if (error && error != EEXIST) { 3473 if (error == ENOSPC) { 3474 ztest_record_enospc(FTAG); 3475 goto out; 3476 } 3477 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); 3478 } 3479 3480 error = dmu_objset_clone(clone2name, snap3name); 3481 if (error) { 3482 if (error == ENOSPC) { 3483 ztest_record_enospc(FTAG); 3484 goto out; 3485 } 3486 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error); 3487 } 3488 3489 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os); 3490 if (error) 3491 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error); 3492 error = dsl_dataset_promote(clone2name, NULL); 3493 if (error == ENOSPC) { 3494 dmu_objset_disown(os, FTAG); 3495 ztest_record_enospc(FTAG); 3496 goto out; 3497 } 3498 if (error != EBUSY) 3499 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name, 3500 error); 3501 dmu_objset_disown(os, FTAG); 3502 3503out: 3504 ztest_dsl_dataset_cleanup(osname, id); 3505 3506 (void) rw_unlock(&ztest_name_lock); 3507} 3508 3509/* 3510 * Verify that dmu_object_{alloc,free} work as expected. 3511 */ 3512void 3513ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id) 3514{ 3515 ztest_od_t od[4]; 3516 int batchsize = sizeof (od) / sizeof (od[0]); 3517 3518 for (int b = 0; b < batchsize; b++) 3519 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0); 3520 3521 /* 3522 * Destroy the previous batch of objects, create a new batch, 3523 * and do some I/O on the new objects. 3524 */ 3525 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0) 3526 return; 3527 3528 while (ztest_random(4 * batchsize) != 0) 3529 ztest_io(zd, od[ztest_random(batchsize)].od_object, 3530 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 3531} 3532 3533/* 3534 * Verify that dmu_{read,write} work as expected. 3535 */ 3536void 3537ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id) 3538{ 3539 objset_t *os = zd->zd_os; 3540 ztest_od_t od[2]; 3541 dmu_tx_t *tx; 3542 int i, freeit, error; 3543 uint64_t n, s, txg; 3544 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; 3545 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; 3546 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t); 3547 uint64_t regions = 997; 3548 uint64_t stride = 123456789ULL; 3549 uint64_t width = 40; 3550 int free_percent = 5; 3551 3552 /* 3553 * This test uses two objects, packobj and bigobj, that are always 3554 * updated together (i.e. in the same tx) so that their contents are 3555 * in sync and can be compared. Their contents relate to each other 3556 * in a simple way: packobj is a dense array of 'bufwad' structures, 3557 * while bigobj is a sparse array of the same bufwads. Specifically, 3558 * for any index n, there are three bufwads that should be identical: 3559 * 3560 * packobj, at offset n * sizeof (bufwad_t) 3561 * bigobj, at the head of the nth chunk 3562 * bigobj, at the tail of the nth chunk 3563 * 3564 * The chunk size is arbitrary. It doesn't have to be a power of two, 3565 * and it doesn't have any relation to the object blocksize. 3566 * The only requirement is that it can hold at least two bufwads. 3567 * 3568 * Normally, we write the bufwad to each of these locations. 3569 * However, free_percent of the time we instead write zeroes to 3570 * packobj and perform a dmu_free_range() on bigobj. By comparing 3571 * bigobj to packobj, we can verify that the DMU is correctly 3572 * tracking which parts of an object are allocated and free, 3573 * and that the contents of the allocated blocks are correct. 3574 */ 3575 3576 /* 3577 * Read the directory info. If it's the first time, set things up. 3578 */ 3579 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize); 3580 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); 3581 3582 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 3583 return; 3584 3585 bigobj = od[0].od_object; 3586 packobj = od[1].od_object; 3587 chunksize = od[0].od_gen; 3588 ASSERT(chunksize == od[1].od_gen); 3589 3590 /* 3591 * Prefetch a random chunk of the big object. 3592 * Our aim here is to get some async reads in flight 3593 * for blocks that we may free below; the DMU should 3594 * handle this race correctly. 3595 */ 3596 n = ztest_random(regions) * stride + ztest_random(width); 3597 s = 1 + ztest_random(2 * width - 1); 3598 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize); 3599 3600 /* 3601 * Pick a random index and compute the offsets into packobj and bigobj. 3602 */ 3603 n = ztest_random(regions) * stride + ztest_random(width); 3604 s = 1 + ztest_random(width - 1); 3605 3606 packoff = n * sizeof (bufwad_t); 3607 packsize = s * sizeof (bufwad_t); 3608 3609 bigoff = n * chunksize; 3610 bigsize = s * chunksize; 3611 3612 packbuf = umem_alloc(packsize, UMEM_NOFAIL); 3613 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); 3614 3615 /* 3616 * free_percent of the time, free a range of bigobj rather than 3617 * overwriting it. 3618 */ 3619 freeit = (ztest_random(100) < free_percent); 3620 3621 /* 3622 * Read the current contents of our objects. 3623 */ 3624 error = dmu_read(os, packobj, packoff, packsize, packbuf, 3625 DMU_READ_PREFETCH); 3626 ASSERT0(error); 3627 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf, 3628 DMU_READ_PREFETCH); 3629 ASSERT0(error); 3630 3631 /* 3632 * Get a tx for the mods to both packobj and bigobj. 3633 */ 3634 tx = dmu_tx_create(os); 3635 3636 dmu_tx_hold_write(tx, packobj, packoff, packsize); 3637 3638 if (freeit) 3639 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize); 3640 else 3641 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); 3642 3643 /* This accounts for setting the checksum/compression. */ 3644 dmu_tx_hold_bonus(tx, bigobj); 3645 3646 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 3647 if (txg == 0) { 3648 umem_free(packbuf, packsize); 3649 umem_free(bigbuf, bigsize); 3650 return; 3651 } 3652 3653 enum zio_checksum cksum; 3654 do { 3655 cksum = (enum zio_checksum) 3656 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM); 3657 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS); 3658 dmu_object_set_checksum(os, bigobj, cksum, tx); 3659 3660 enum zio_compress comp; 3661 do { 3662 comp = (enum zio_compress) 3663 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION); 3664 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS); 3665 dmu_object_set_compress(os, bigobj, comp, tx); 3666 3667 /* 3668 * For each index from n to n + s, verify that the existing bufwad 3669 * in packobj matches the bufwads at the head and tail of the 3670 * corresponding chunk in bigobj. Then update all three bufwads 3671 * with the new values we want to write out. 3672 */ 3673 for (i = 0; i < s; i++) { 3674 /* LINTED */ 3675 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 3676 /* LINTED */ 3677 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); 3678 /* LINTED */ 3679 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; 3680 3681 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 3682 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 3683 3684 if (pack->bw_txg > txg) 3685 fatal(0, "future leak: got %llx, open txg is %llx", 3686 pack->bw_txg, txg); 3687 3688 if (pack->bw_data != 0 && pack->bw_index != n + i) 3689 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 3690 pack->bw_index, n, i); 3691 3692 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 3693 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 3694 3695 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 3696 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 3697 3698 if (freeit) { 3699 bzero(pack, sizeof (bufwad_t)); 3700 } else { 3701 pack->bw_index = n + i; 3702 pack->bw_txg = txg; 3703 pack->bw_data = 1 + ztest_random(-2ULL); 3704 } 3705 *bigH = *pack; 3706 *bigT = *pack; 3707 } 3708 3709 /* 3710 * We've verified all the old bufwads, and made new ones. 3711 * Now write them out. 3712 */ 3713 dmu_write(os, packobj, packoff, packsize, packbuf, tx); 3714 3715 if (freeit) { 3716 if (ztest_opts.zo_verbose >= 7) { 3717 (void) printf("freeing offset %llx size %llx" 3718 " txg %llx\n", 3719 (u_longlong_t)bigoff, 3720 (u_longlong_t)bigsize, 3721 (u_longlong_t)txg); 3722 } 3723 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx)); 3724 } else { 3725 if (ztest_opts.zo_verbose >= 7) { 3726 (void) printf("writing offset %llx size %llx" 3727 " txg %llx\n", 3728 (u_longlong_t)bigoff, 3729 (u_longlong_t)bigsize, 3730 (u_longlong_t)txg); 3731 } 3732 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx); 3733 } 3734 3735 dmu_tx_commit(tx); 3736 3737 /* 3738 * Sanity check the stuff we just wrote. 3739 */ 3740 { 3741 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 3742 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 3743 3744 VERIFY(0 == dmu_read(os, packobj, packoff, 3745 packsize, packcheck, DMU_READ_PREFETCH)); 3746 VERIFY(0 == dmu_read(os, bigobj, bigoff, 3747 bigsize, bigcheck, DMU_READ_PREFETCH)); 3748 3749 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 3750 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 3751 3752 umem_free(packcheck, packsize); 3753 umem_free(bigcheck, bigsize); 3754 } 3755 3756 umem_free(packbuf, packsize); 3757 umem_free(bigbuf, bigsize); 3758} 3759 3760void 3761compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf, 3762 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg) 3763{ 3764 uint64_t i; 3765 bufwad_t *pack; 3766 bufwad_t *bigH; 3767 bufwad_t *bigT; 3768 3769 /* 3770 * For each index from n to n + s, verify that the existing bufwad 3771 * in packobj matches the bufwads at the head and tail of the 3772 * corresponding chunk in bigobj. Then update all three bufwads 3773 * with the new values we want to write out. 3774 */ 3775 for (i = 0; i < s; i++) { 3776 /* LINTED */ 3777 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 3778 /* LINTED */ 3779 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); 3780 /* LINTED */ 3781 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; 3782 3783 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 3784 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 3785 3786 if (pack->bw_txg > txg) 3787 fatal(0, "future leak: got %llx, open txg is %llx", 3788 pack->bw_txg, txg); 3789 3790 if (pack->bw_data != 0 && pack->bw_index != n + i) 3791 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 3792 pack->bw_index, n, i); 3793 3794 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 3795 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 3796 3797 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 3798 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 3799 3800 pack->bw_index = n + i; 3801 pack->bw_txg = txg; 3802 pack->bw_data = 1 + ztest_random(-2ULL); 3803 3804 *bigH = *pack; 3805 *bigT = *pack; 3806 } 3807} 3808 3809void 3810ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id) 3811{ 3812 objset_t *os = zd->zd_os; 3813 ztest_od_t od[2]; 3814 dmu_tx_t *tx; 3815 uint64_t i; 3816 int error; 3817 uint64_t n, s, txg; 3818 bufwad_t *packbuf, *bigbuf; 3819 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; 3820 uint64_t blocksize = ztest_random_blocksize(); 3821 uint64_t chunksize = blocksize; 3822 uint64_t regions = 997; 3823 uint64_t stride = 123456789ULL; 3824 uint64_t width = 9; 3825 dmu_buf_t *bonus_db; 3826 arc_buf_t **bigbuf_arcbufs; 3827 dmu_object_info_t doi; 3828 3829 /* 3830 * This test uses two objects, packobj and bigobj, that are always 3831 * updated together (i.e. in the same tx) so that their contents are 3832 * in sync and can be compared. Their contents relate to each other 3833 * in a simple way: packobj is a dense array of 'bufwad' structures, 3834 * while bigobj is a sparse array of the same bufwads. Specifically, 3835 * for any index n, there are three bufwads that should be identical: 3836 * 3837 * packobj, at offset n * sizeof (bufwad_t) 3838 * bigobj, at the head of the nth chunk 3839 * bigobj, at the tail of the nth chunk 3840 * 3841 * The chunk size is set equal to bigobj block size so that 3842 * dmu_assign_arcbuf() can be tested for object updates. 3843 */ 3844 3845 /* 3846 * Read the directory info. If it's the first time, set things up. 3847 */ 3848 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 3849 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); 3850 3851 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 3852 return; 3853 3854 bigobj = od[0].od_object; 3855 packobj = od[1].od_object; 3856 blocksize = od[0].od_blocksize; 3857 chunksize = blocksize; 3858 ASSERT(chunksize == od[1].od_gen); 3859 3860 VERIFY(dmu_object_info(os, bigobj, &doi) == 0); 3861 VERIFY(ISP2(doi.doi_data_block_size)); 3862 VERIFY(chunksize == doi.doi_data_block_size); 3863 VERIFY(chunksize >= 2 * sizeof (bufwad_t)); 3864 3865 /* 3866 * Pick a random index and compute the offsets into packobj and bigobj. 3867 */ 3868 n = ztest_random(regions) * stride + ztest_random(width); 3869 s = 1 + ztest_random(width - 1); 3870 3871 packoff = n * sizeof (bufwad_t); 3872 packsize = s * sizeof (bufwad_t); 3873 3874 bigoff = n * chunksize; 3875 bigsize = s * chunksize; 3876 3877 packbuf = umem_zalloc(packsize, UMEM_NOFAIL); 3878 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL); 3879 3880 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db)); 3881 3882 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL); 3883 3884 /* 3885 * Iteration 0 test zcopy for DB_UNCACHED dbufs. 3886 * Iteration 1 test zcopy to already referenced dbufs. 3887 * Iteration 2 test zcopy to dirty dbuf in the same txg. 3888 * Iteration 3 test zcopy to dbuf dirty in previous txg. 3889 * Iteration 4 test zcopy when dbuf is no longer dirty. 3890 * Iteration 5 test zcopy when it can't be done. 3891 * Iteration 6 one more zcopy write. 3892 */ 3893 for (i = 0; i < 7; i++) { 3894 uint64_t j; 3895 uint64_t off; 3896 3897 /* 3898 * In iteration 5 (i == 5) use arcbufs 3899 * that don't match bigobj blksz to test 3900 * dmu_assign_arcbuf() when it can't directly 3901 * assign an arcbuf to a dbuf. 3902 */ 3903 for (j = 0; j < s; j++) { 3904 if (i != 5) { 3905 bigbuf_arcbufs[j] = 3906 dmu_request_arcbuf(bonus_db, chunksize); 3907 } else { 3908 bigbuf_arcbufs[2 * j] = 3909 dmu_request_arcbuf(bonus_db, chunksize / 2); 3910 bigbuf_arcbufs[2 * j + 1] = 3911 dmu_request_arcbuf(bonus_db, chunksize / 2); 3912 } 3913 } 3914 3915 /* 3916 * Get a tx for the mods to both packobj and bigobj. 3917 */ 3918 tx = dmu_tx_create(os); 3919 3920 dmu_tx_hold_write(tx, packobj, packoff, packsize); 3921 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); 3922 3923 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 3924 if (txg == 0) { 3925 umem_free(packbuf, packsize); 3926 umem_free(bigbuf, bigsize); 3927 for (j = 0; j < s; j++) { 3928 if (i != 5) { 3929 dmu_return_arcbuf(bigbuf_arcbufs[j]); 3930 } else { 3931 dmu_return_arcbuf( 3932 bigbuf_arcbufs[2 * j]); 3933 dmu_return_arcbuf( 3934 bigbuf_arcbufs[2 * j + 1]); 3935 } 3936 } 3937 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); 3938 dmu_buf_rele(bonus_db, FTAG); 3939 return; 3940 } 3941 3942 /* 3943 * 50% of the time don't read objects in the 1st iteration to 3944 * test dmu_assign_arcbuf() for the case when there're no 3945 * existing dbufs for the specified offsets. 3946 */ 3947 if (i != 0 || ztest_random(2) != 0) { 3948 error = dmu_read(os, packobj, packoff, 3949 packsize, packbuf, DMU_READ_PREFETCH); 3950 ASSERT0(error); 3951 error = dmu_read(os, bigobj, bigoff, bigsize, 3952 bigbuf, DMU_READ_PREFETCH); 3953 ASSERT0(error); 3954 } 3955 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize, 3956 n, chunksize, txg); 3957 3958 /* 3959 * We've verified all the old bufwads, and made new ones. 3960 * Now write them out. 3961 */ 3962 dmu_write(os, packobj, packoff, packsize, packbuf, tx); 3963 if (ztest_opts.zo_verbose >= 7) { 3964 (void) printf("writing offset %llx size %llx" 3965 " txg %llx\n", 3966 (u_longlong_t)bigoff, 3967 (u_longlong_t)bigsize, 3968 (u_longlong_t)txg); 3969 } 3970 for (off = bigoff, j = 0; j < s; j++, off += chunksize) { 3971 dmu_buf_t *dbt; 3972 if (i != 5) { 3973 bcopy((caddr_t)bigbuf + (off - bigoff), 3974 bigbuf_arcbufs[j]->b_data, chunksize); 3975 } else { 3976 bcopy((caddr_t)bigbuf + (off - bigoff), 3977 bigbuf_arcbufs[2 * j]->b_data, 3978 chunksize / 2); 3979 bcopy((caddr_t)bigbuf + (off - bigoff) + 3980 chunksize / 2, 3981 bigbuf_arcbufs[2 * j + 1]->b_data, 3982 chunksize / 2); 3983 } 3984 3985 if (i == 1) { 3986 VERIFY(dmu_buf_hold(os, bigobj, off, 3987 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0); 3988 } 3989 if (i != 5) { 3990 dmu_assign_arcbuf(bonus_db, off, 3991 bigbuf_arcbufs[j], tx); 3992 } else { 3993 dmu_assign_arcbuf(bonus_db, off, 3994 bigbuf_arcbufs[2 * j], tx); 3995 dmu_assign_arcbuf(bonus_db, 3996 off + chunksize / 2, 3997 bigbuf_arcbufs[2 * j + 1], tx); 3998 } 3999 if (i == 1) { 4000 dmu_buf_rele(dbt, FTAG); 4001 } 4002 } 4003 dmu_tx_commit(tx); 4004 4005 /* 4006 * Sanity check the stuff we just wrote. 4007 */ 4008 { 4009 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 4010 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 4011 4012 VERIFY(0 == dmu_read(os, packobj, packoff, 4013 packsize, packcheck, DMU_READ_PREFETCH)); 4014 VERIFY(0 == dmu_read(os, bigobj, bigoff, 4015 bigsize, bigcheck, DMU_READ_PREFETCH)); 4016 4017 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 4018 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 4019 4020 umem_free(packcheck, packsize); 4021 umem_free(bigcheck, bigsize); 4022 } 4023 if (i == 2) { 4024 txg_wait_open(dmu_objset_pool(os), 0); 4025 } else if (i == 3) { 4026 txg_wait_synced(dmu_objset_pool(os), 0); 4027 } 4028 } 4029 4030 dmu_buf_rele(bonus_db, FTAG); 4031 umem_free(packbuf, packsize); 4032 umem_free(bigbuf, bigsize); 4033 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); 4034} 4035 4036/* ARGSUSED */ 4037void 4038ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id) 4039{ 4040 ztest_od_t od[1]; 4041 uint64_t offset = (1ULL << (ztest_random(20) + 43)) + 4042 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 4043 4044 /* 4045 * Have multiple threads write to large offsets in an object 4046 * to verify that parallel writes to an object -- even to the 4047 * same blocks within the object -- doesn't cause any trouble. 4048 */ 4049 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 4050 4051 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4052 return; 4053 4054 while (ztest_random(10) != 0) 4055 ztest_io(zd, od[0].od_object, offset); 4056} 4057 4058void 4059ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id) 4060{ 4061 ztest_od_t od[1]; 4062 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) + 4063 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 4064 uint64_t count = ztest_random(20) + 1; 4065 uint64_t blocksize = ztest_random_blocksize(); 4066 void *data; 4067 4068 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 4069 4070 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4071 return; 4072 4073 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0) 4074 return; 4075 4076 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize); 4077 4078 data = umem_zalloc(blocksize, UMEM_NOFAIL); 4079 4080 while (ztest_random(count) != 0) { 4081 uint64_t randoff = offset + (ztest_random(count) * blocksize); 4082 if (ztest_write(zd, od[0].od_object, randoff, blocksize, 4083 data) != 0) 4084 break; 4085 while (ztest_random(4) != 0) 4086 ztest_io(zd, od[0].od_object, randoff); 4087 } 4088 4089 umem_free(data, blocksize); 4090} 4091 4092/* 4093 * Verify that zap_{create,destroy,add,remove,update} work as expected. 4094 */ 4095#define ZTEST_ZAP_MIN_INTS 1 4096#define ZTEST_ZAP_MAX_INTS 4 4097#define ZTEST_ZAP_MAX_PROPS 1000 4098 4099void 4100ztest_zap(ztest_ds_t *zd, uint64_t id) 4101{ 4102 objset_t *os = zd->zd_os; 4103 ztest_od_t od[1]; 4104 uint64_t object; 4105 uint64_t txg, last_txg; 4106 uint64_t value[ZTEST_ZAP_MAX_INTS]; 4107 uint64_t zl_ints, zl_intsize, prop; 4108 int i, ints; 4109 dmu_tx_t *tx; 4110 char propname[100], txgname[100]; 4111 int error; 4112 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; 4113 4114 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); 4115 4116 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4117 return; 4118 4119 object = od[0].od_object; 4120 4121 /* 4122 * Generate a known hash collision, and verify that 4123 * we can lookup and remove both entries. 4124 */ 4125 tx = dmu_tx_create(os); 4126 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4127 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4128 if (txg == 0) 4129 return; 4130 for (i = 0; i < 2; i++) { 4131 value[i] = i; 4132 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t), 4133 1, &value[i], tx)); 4134 } 4135 for (i = 0; i < 2; i++) { 4136 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i], 4137 sizeof (uint64_t), 1, &value[i], tx)); 4138 VERIFY3U(0, ==, 4139 zap_length(os, object, hc[i], &zl_intsize, &zl_ints)); 4140 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4141 ASSERT3U(zl_ints, ==, 1); 4142 } 4143 for (i = 0; i < 2; i++) { 4144 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx)); 4145 } 4146 dmu_tx_commit(tx); 4147 4148 /* 4149 * Generate a buch of random entries. 4150 */ 4151 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); 4152 4153 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 4154 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 4155 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 4156 bzero(value, sizeof (value)); 4157 last_txg = 0; 4158 4159 /* 4160 * If these zap entries already exist, validate their contents. 4161 */ 4162 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 4163 if (error == 0) { 4164 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4165 ASSERT3U(zl_ints, ==, 1); 4166 4167 VERIFY(zap_lookup(os, object, txgname, zl_intsize, 4168 zl_ints, &last_txg) == 0); 4169 4170 VERIFY(zap_length(os, object, propname, &zl_intsize, 4171 &zl_ints) == 0); 4172 4173 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4174 ASSERT3U(zl_ints, ==, ints); 4175 4176 VERIFY(zap_lookup(os, object, propname, zl_intsize, 4177 zl_ints, value) == 0); 4178 4179 for (i = 0; i < ints; i++) { 4180 ASSERT3U(value[i], ==, last_txg + object + i); 4181 } 4182 } else { 4183 ASSERT3U(error, ==, ENOENT); 4184 } 4185 4186 /* 4187 * Atomically update two entries in our zap object. 4188 * The first is named txg_%llu, and contains the txg 4189 * in which the property was last updated. The second 4190 * is named prop_%llu, and the nth element of its value 4191 * should be txg + object + n. 4192 */ 4193 tx = dmu_tx_create(os); 4194 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4195 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4196 if (txg == 0) 4197 return; 4198 4199 if (last_txg > txg) 4200 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); 4201 4202 for (i = 0; i < ints; i++) 4203 value[i] = txg + object + i; 4204 4205 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t), 4206 1, &txg, tx)); 4207 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t), 4208 ints, value, tx)); 4209 4210 dmu_tx_commit(tx); 4211 4212 /* 4213 * Remove a random pair of entries. 4214 */ 4215 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 4216 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 4217 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 4218 4219 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 4220 4221 if (error == ENOENT) 4222 return; 4223 4224 ASSERT0(error); 4225 4226 tx = dmu_tx_create(os); 4227 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4228 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4229 if (txg == 0) 4230 return; 4231 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx)); 4232 VERIFY3U(0, ==, zap_remove(os, object, propname, tx)); 4233 dmu_tx_commit(tx); 4234} 4235 4236/* 4237 * Testcase to test the upgrading of a microzap to fatzap. 4238 */ 4239void 4240ztest_fzap(ztest_ds_t *zd, uint64_t id) 4241{ 4242 objset_t *os = zd->zd_os; 4243 ztest_od_t od[1]; 4244 uint64_t object, txg; 4245 4246 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); 4247 4248 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4249 return; 4250 4251 object = od[0].od_object; 4252 4253 /* 4254 * Add entries to this ZAP and make sure it spills over 4255 * and gets upgraded to a fatzap. Also, since we are adding 4256 * 2050 entries we should see ptrtbl growth and leaf-block split. 4257 */ 4258 for (int i = 0; i < 2050; i++) { 4259 char name[MAXNAMELEN]; 4260 uint64_t value = i; 4261 dmu_tx_t *tx; 4262 int error; 4263 4264 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu", 4265 id, value); 4266 4267 tx = dmu_tx_create(os); 4268 dmu_tx_hold_zap(tx, object, B_TRUE, name); 4269 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4270 if (txg == 0) 4271 return; 4272 error = zap_add(os, object, name, sizeof (uint64_t), 1, 4273 &value, tx); 4274 ASSERT(error == 0 || error == EEXIST); 4275 dmu_tx_commit(tx); 4276 } 4277} 4278 4279/* ARGSUSED */ 4280void 4281ztest_zap_parallel(ztest_ds_t *zd, uint64_t id) 4282{ 4283 objset_t *os = zd->zd_os; 4284 ztest_od_t od[1]; 4285 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; 4286 dmu_tx_t *tx; 4287 int i, namelen, error; 4288 int micro = ztest_random(2); 4289 char name[20], string_value[20]; 4290 void *data; 4291 4292 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0); 4293 4294 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4295 return; 4296 4297 object = od[0].od_object; 4298 4299 /* 4300 * Generate a random name of the form 'xxx.....' where each 4301 * x is a random printable character and the dots are dots. 4302 * There are 94 such characters, and the name length goes from 4303 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. 4304 */ 4305 namelen = ztest_random(sizeof (name) - 5) + 5 + 1; 4306 4307 for (i = 0; i < 3; i++) 4308 name[i] = '!' + ztest_random('~' - '!' + 1); 4309 for (; i < namelen - 1; i++) 4310 name[i] = '.'; 4311 name[i] = '\0'; 4312 4313 if ((namelen & 1) || micro) { 4314 wsize = sizeof (txg); 4315 wc = 1; 4316 data = &txg; 4317 } else { 4318 wsize = 1; 4319 wc = namelen; 4320 data = string_value; 4321 } 4322 4323 count = -1ULL; 4324 VERIFY0(zap_count(os, object, &count)); 4325 ASSERT(count != -1ULL); 4326 4327 /* 4328 * Select an operation: length, lookup, add, update, remove. 4329 */ 4330 i = ztest_random(5); 4331 4332 if (i >= 2) { 4333 tx = dmu_tx_create(os); 4334 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4335 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4336 if (txg == 0) 4337 return; 4338 bcopy(name, string_value, namelen); 4339 } else { 4340 tx = NULL; 4341 txg = 0; 4342 bzero(string_value, namelen); 4343 } 4344 4345 switch (i) { 4346 4347 case 0: 4348 error = zap_length(os, object, name, &zl_wsize, &zl_wc); 4349 if (error == 0) { 4350 ASSERT3U(wsize, ==, zl_wsize); 4351 ASSERT3U(wc, ==, zl_wc); 4352 } else { 4353 ASSERT3U(error, ==, ENOENT); 4354 } 4355 break; 4356 4357 case 1: 4358 error = zap_lookup(os, object, name, wsize, wc, data); 4359 if (error == 0) { 4360 if (data == string_value && 4361 bcmp(name, data, namelen) != 0) 4362 fatal(0, "name '%s' != val '%s' len %d", 4363 name, data, namelen); 4364 } else { 4365 ASSERT3U(error, ==, ENOENT); 4366 } 4367 break; 4368 4369 case 2: 4370 error = zap_add(os, object, name, wsize, wc, data, tx); 4371 ASSERT(error == 0 || error == EEXIST); 4372 break; 4373 4374 case 3: 4375 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); 4376 break; 4377 4378 case 4: 4379 error = zap_remove(os, object, name, tx); 4380 ASSERT(error == 0 || error == ENOENT); 4381 break; 4382 } 4383 4384 if (tx != NULL) 4385 dmu_tx_commit(tx); 4386} 4387 4388/* 4389 * Commit callback data. 4390 */ 4391typedef struct ztest_cb_data { 4392 list_node_t zcd_node; 4393 uint64_t zcd_txg; 4394 int zcd_expected_err; 4395 boolean_t zcd_added; 4396 boolean_t zcd_called; 4397 spa_t *zcd_spa; 4398} ztest_cb_data_t; 4399 4400/* This is the actual commit callback function */ 4401static void 4402ztest_commit_callback(void *arg, int error) 4403{ 4404 ztest_cb_data_t *data = arg; 4405 uint64_t synced_txg; 4406 4407 VERIFY(data != NULL); 4408 VERIFY3S(data->zcd_expected_err, ==, error); 4409 VERIFY(!data->zcd_called); 4410 4411 synced_txg = spa_last_synced_txg(data->zcd_spa); 4412 if (data->zcd_txg > synced_txg) 4413 fatal(0, "commit callback of txg %" PRIu64 " called prematurely" 4414 ", last synced txg = %" PRIu64 "\n", data->zcd_txg, 4415 synced_txg); 4416 4417 data->zcd_called = B_TRUE; 4418 4419 if (error == ECANCELED) { 4420 ASSERT0(data->zcd_txg); 4421 ASSERT(!data->zcd_added); 4422 4423 /* 4424 * The private callback data should be destroyed here, but 4425 * since we are going to check the zcd_called field after 4426 * dmu_tx_abort(), we will destroy it there. 4427 */ 4428 return; 4429 } 4430 4431 /* Was this callback added to the global callback list? */ 4432 if (!data->zcd_added) 4433 goto out; 4434 4435 ASSERT3U(data->zcd_txg, !=, 0); 4436 4437 /* Remove our callback from the list */ 4438 (void) mutex_lock(&zcl.zcl_callbacks_lock); 4439 list_remove(&zcl.zcl_callbacks, data); 4440 (void) mutex_unlock(&zcl.zcl_callbacks_lock); 4441 4442out: 4443 umem_free(data, sizeof (ztest_cb_data_t)); 4444} 4445 4446/* Allocate and initialize callback data structure */ 4447static ztest_cb_data_t * 4448ztest_create_cb_data(objset_t *os, uint64_t txg) 4449{ 4450 ztest_cb_data_t *cb_data; 4451 4452 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL); 4453 4454 cb_data->zcd_txg = txg; 4455 cb_data->zcd_spa = dmu_objset_spa(os); 4456 4457 return (cb_data); 4458} 4459 4460/* 4461 * If a number of txgs equal to this threshold have been created after a commit 4462 * callback has been registered but not called, then we assume there is an 4463 * implementation bug. 4464 */ 4465#define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2) 4466 4467/* 4468 * Commit callback test. 4469 */ 4470void 4471ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id) 4472{ 4473 objset_t *os = zd->zd_os; 4474 ztest_od_t od[1]; 4475 dmu_tx_t *tx; 4476 ztest_cb_data_t *cb_data[3], *tmp_cb; 4477 uint64_t old_txg, txg; 4478 int i, error; 4479 4480 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 4481 4482 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4483 return; 4484 4485 tx = dmu_tx_create(os); 4486 4487 cb_data[0] = ztest_create_cb_data(os, 0); 4488 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]); 4489 4490 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t)); 4491 4492 /* Every once in a while, abort the transaction on purpose */ 4493 if (ztest_random(100) == 0) 4494 error = -1; 4495 4496 if (!error) 4497 error = dmu_tx_assign(tx, TXG_NOWAIT); 4498 4499 txg = error ? 0 : dmu_tx_get_txg(tx); 4500 4501 cb_data[0]->zcd_txg = txg; 4502 cb_data[1] = ztest_create_cb_data(os, txg); 4503 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]); 4504 4505 if (error) { 4506 /* 4507 * It's not a strict requirement to call the registered 4508 * callbacks from inside dmu_tx_abort(), but that's what 4509 * it's supposed to happen in the current implementation 4510 * so we will check for that. 4511 */ 4512 for (i = 0; i < 2; i++) { 4513 cb_data[i]->zcd_expected_err = ECANCELED; 4514 VERIFY(!cb_data[i]->zcd_called); 4515 } 4516 4517 dmu_tx_abort(tx); 4518 4519 for (i = 0; i < 2; i++) { 4520 VERIFY(cb_data[i]->zcd_called); 4521 umem_free(cb_data[i], sizeof (ztest_cb_data_t)); 4522 } 4523 4524 return; 4525 } 4526 4527 cb_data[2] = ztest_create_cb_data(os, txg); 4528 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]); 4529 4530 /* 4531 * Read existing data to make sure there isn't a future leak. 4532 */ 4533 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t), 4534 &old_txg, DMU_READ_PREFETCH)); 4535 4536 if (old_txg > txg) 4537 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64, 4538 old_txg, txg); 4539 4540 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx); 4541 4542 (void) mutex_lock(&zcl.zcl_callbacks_lock); 4543 4544 /* 4545 * Since commit callbacks don't have any ordering requirement and since 4546 * it is theoretically possible for a commit callback to be called 4547 * after an arbitrary amount of time has elapsed since its txg has been 4548 * synced, it is difficult to reliably determine whether a commit 4549 * callback hasn't been called due to high load or due to a flawed 4550 * implementation. 4551 * 4552 * In practice, we will assume that if after a certain number of txgs a 4553 * commit callback hasn't been called, then most likely there's an 4554 * implementation bug.. 4555 */ 4556 tmp_cb = list_head(&zcl.zcl_callbacks); 4557 if (tmp_cb != NULL && 4558 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) { 4559 fatal(0, "Commit callback threshold exceeded, oldest txg: %" 4560 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg); 4561 } 4562 4563 /* 4564 * Let's find the place to insert our callbacks. 4565 * 4566 * Even though the list is ordered by txg, it is possible for the 4567 * insertion point to not be the end because our txg may already be 4568 * quiescing at this point and other callbacks in the open txg 4569 * (from other objsets) may have sneaked in. 4570 */ 4571 tmp_cb = list_tail(&zcl.zcl_callbacks); 4572 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg) 4573 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb); 4574 4575 /* Add the 3 callbacks to the list */ 4576 for (i = 0; i < 3; i++) { 4577 if (tmp_cb == NULL) 4578 list_insert_head(&zcl.zcl_callbacks, cb_data[i]); 4579 else 4580 list_insert_after(&zcl.zcl_callbacks, tmp_cb, 4581 cb_data[i]); 4582 4583 cb_data[i]->zcd_added = B_TRUE; 4584 VERIFY(!cb_data[i]->zcd_called); 4585 4586 tmp_cb = cb_data[i]; 4587 } 4588 4589 (void) mutex_unlock(&zcl.zcl_callbacks_lock); 4590 4591 dmu_tx_commit(tx); 4592} 4593 4594/* ARGSUSED */ 4595void 4596ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id) 4597{ 4598 zfs_prop_t proplist[] = { 4599 ZFS_PROP_CHECKSUM, 4600 ZFS_PROP_COMPRESSION, 4601 ZFS_PROP_COPIES, 4602 ZFS_PROP_DEDUP 4603 }; 4604 4605 (void) rw_rdlock(&ztest_name_lock); 4606 4607 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) 4608 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p], 4609 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2)); 4610 4611 (void) rw_unlock(&ztest_name_lock); 4612} 4613 4614/* ARGSUSED */ 4615void 4616ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id) 4617{ 4618 nvlist_t *props = NULL; 4619 4620 (void) rw_rdlock(&ztest_name_lock); 4621 4622 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO, 4623 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN)); 4624 4625 VERIFY0(spa_prop_get(ztest_spa, &props)); 4626 4627 if (ztest_opts.zo_verbose >= 6) 4628 dump_nvlist(props, 4); 4629 4630 nvlist_free(props); 4631 4632 (void) rw_unlock(&ztest_name_lock); 4633} 4634 4635static int 4636user_release_one(const char *snapname, const char *holdname) 4637{ 4638 nvlist_t *snaps, *holds; 4639 int error; 4640 4641 snaps = fnvlist_alloc(); 4642 holds = fnvlist_alloc(); 4643 fnvlist_add_boolean(holds, holdname); 4644 fnvlist_add_nvlist(snaps, snapname, holds); 4645 fnvlist_free(holds); 4646 error = dsl_dataset_user_release(snaps, NULL); 4647 fnvlist_free(snaps); 4648 return (error); 4649} 4650 4651/* 4652 * Test snapshot hold/release and deferred destroy. 4653 */ 4654void 4655ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) 4656{ 4657 int error; 4658 objset_t *os = zd->zd_os; 4659 objset_t *origin; 4660 char snapname[100]; 4661 char fullname[100]; 4662 char clonename[100]; 4663 char tag[100]; 4664 char osname[MAXNAMELEN]; 4665 nvlist_t *holds; 4666 4667 (void) rw_rdlock(&ztest_name_lock); 4668 4669 dmu_objset_name(os, osname); 4670 4671 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id); 4672 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname); 4673 (void) snprintf(clonename, sizeof (clonename), 4674 "%s/ch1_%llu", osname, id); 4675 (void) snprintf(tag, sizeof (tag), "tag_%llu", id); 4676 4677 /* 4678 * Clean up from any previous run. 4679 */ 4680 error = dsl_destroy_head(clonename); 4681 if (error != ENOENT) 4682 ASSERT0(error); 4683 error = user_release_one(fullname, tag); 4684 if (error != ESRCH && error != ENOENT) 4685 ASSERT0(error); 4686 error = dsl_destroy_snapshot(fullname, B_FALSE); 4687 if (error != ENOENT) 4688 ASSERT0(error); 4689 4690 /* 4691 * Create snapshot, clone it, mark snap for deferred destroy, 4692 * destroy clone, verify snap was also destroyed. 4693 */ 4694 error = dmu_objset_snapshot_one(osname, snapname); 4695 if (error) { 4696 if (error == ENOSPC) { 4697 ztest_record_enospc("dmu_objset_snapshot"); 4698 goto out; 4699 } 4700 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); 4701 } 4702 4703 error = dmu_objset_clone(clonename, fullname); 4704 if (error) { 4705 if (error == ENOSPC) { 4706 ztest_record_enospc("dmu_objset_clone"); 4707 goto out; 4708 } 4709 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error); 4710 } 4711 4712 error = dsl_destroy_snapshot(fullname, B_TRUE); 4713 if (error) { 4714 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", 4715 fullname, error); 4716 } 4717 4718 error = dsl_destroy_head(clonename); 4719 if (error) 4720 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error); 4721 4722 error = dmu_objset_hold(fullname, FTAG, &origin); 4723 if (error != ENOENT) 4724 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); 4725 4726 /* 4727 * Create snapshot, add temporary hold, verify that we can't 4728 * destroy a held snapshot, mark for deferred destroy, 4729 * release hold, verify snapshot was destroyed. 4730 */ 4731 error = dmu_objset_snapshot_one(osname, snapname); 4732 if (error) { 4733 if (error == ENOSPC) { 4734 ztest_record_enospc("dmu_objset_snapshot"); 4735 goto out; 4736 } 4737 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); 4738 } 4739 4740 holds = fnvlist_alloc(); 4741 fnvlist_add_string(holds, fullname, tag); 4742 error = dsl_dataset_user_hold(holds, 0, NULL); 4743 fnvlist_free(holds); 4744 4745 if (error == ENOSPC) { 4746 ztest_record_enospc("dsl_dataset_user_hold"); 4747 goto out; 4748 } else if (error) { 4749 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u", 4750 fullname, tag, error); 4751 } 4752 4753 error = dsl_destroy_snapshot(fullname, B_FALSE); 4754 if (error != EBUSY) { 4755 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d", 4756 fullname, error); 4757 } 4758 4759 error = dsl_destroy_snapshot(fullname, B_TRUE); 4760 if (error) { 4761 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", 4762 fullname, error); 4763 } 4764 4765 error = user_release_one(fullname, tag); 4766 if (error) 4767 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error); 4768 4769 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT); 4770 4771out: 4772 (void) rw_unlock(&ztest_name_lock); 4773} 4774 4775/* 4776 * Inject random faults into the on-disk data. 4777 */ 4778/* ARGSUSED */ 4779void 4780ztest_fault_inject(ztest_ds_t *zd, uint64_t id) 4781{ 4782 ztest_shared_t *zs = ztest_shared; 4783 spa_t *spa = ztest_spa; 4784 int fd; 4785 uint64_t offset; 4786 uint64_t leaves; 4787 uint64_t bad = 0x1990c0ffeedecadeULL; 4788 uint64_t top, leaf; 4789 char path0[MAXPATHLEN]; 4790 char pathrand[MAXPATHLEN]; 4791 size_t fsize; 4792 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ 4793 int iters = 1000; 4794 int maxfaults; 4795 int mirror_save; 4796 vdev_t *vd0 = NULL; 4797 uint64_t guid0 = 0; 4798 boolean_t islog = B_FALSE; 4799 4800 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 4801 maxfaults = MAXFAULTS(); 4802 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; 4803 mirror_save = zs->zs_mirrors; 4804 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4805 4806 ASSERT(leaves >= 1); 4807 4808 /* 4809 * Grab the name lock as reader. There are some operations 4810 * which don't like to have their vdevs changed while 4811 * they are in progress (i.e. spa_change_guid). Those 4812 * operations will have grabbed the name lock as writer. 4813 */ 4814 (void) rw_rdlock(&ztest_name_lock); 4815 4816 /* 4817 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd. 4818 */ 4819 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 4820 4821 if (ztest_random(2) == 0) { 4822 /* 4823 * Inject errors on a normal data device or slog device. 4824 */ 4825 top = ztest_random_vdev_top(spa, B_TRUE); 4826 leaf = ztest_random(leaves) + zs->zs_splits; 4827 4828 /* 4829 * Generate paths to the first leaf in this top-level vdev, 4830 * and to the random leaf we selected. We'll induce transient 4831 * write failures and random online/offline activity on leaf 0, 4832 * and we'll write random garbage to the randomly chosen leaf. 4833 */ 4834 (void) snprintf(path0, sizeof (path0), ztest_dev_template, 4835 ztest_opts.zo_dir, ztest_opts.zo_pool, 4836 top * leaves + zs->zs_splits); 4837 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template, 4838 ztest_opts.zo_dir, ztest_opts.zo_pool, 4839 top * leaves + leaf); 4840 4841 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); 4842 if (vd0 != NULL && vd0->vdev_top->vdev_islog) 4843 islog = B_TRUE; 4844 4845 /* 4846 * If the top-level vdev needs to be resilvered 4847 * then we only allow faults on the device that is 4848 * resilvering. 4849 */ 4850 if (vd0 != NULL && maxfaults != 1 && 4851 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) || 4852 vd0->vdev_resilver_txg != 0)) { 4853 /* 4854 * Make vd0 explicitly claim to be unreadable, 4855 * or unwriteable, or reach behind its back 4856 * and close the underlying fd. We can do this if 4857 * maxfaults == 0 because we'll fail and reexecute, 4858 * and we can do it if maxfaults >= 2 because we'll 4859 * have enough redundancy. If maxfaults == 1, the 4860 * combination of this with injection of random data 4861 * corruption below exceeds the pool's fault tolerance. 4862 */ 4863 vdev_file_t *vf = vd0->vdev_tsd; 4864 4865 if (vf != NULL && ztest_random(3) == 0) { 4866 (void) close(vf->vf_vnode->v_fd); 4867 vf->vf_vnode->v_fd = -1; 4868 } else if (ztest_random(2) == 0) { 4869 vd0->vdev_cant_read = B_TRUE; 4870 } else { 4871 vd0->vdev_cant_write = B_TRUE; 4872 } 4873 guid0 = vd0->vdev_guid; 4874 } 4875 } else { 4876 /* 4877 * Inject errors on an l2cache device. 4878 */ 4879 spa_aux_vdev_t *sav = &spa->spa_l2cache; 4880 4881 if (sav->sav_count == 0) { 4882 spa_config_exit(spa, SCL_STATE, FTAG); 4883 (void) rw_unlock(&ztest_name_lock); 4884 return; 4885 } 4886 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)]; 4887 guid0 = vd0->vdev_guid; 4888 (void) strcpy(path0, vd0->vdev_path); 4889 (void) strcpy(pathrand, vd0->vdev_path); 4890 4891 leaf = 0; 4892 leaves = 1; 4893 maxfaults = INT_MAX; /* no limit on cache devices */ 4894 } 4895 4896 spa_config_exit(spa, SCL_STATE, FTAG); 4897 (void) rw_unlock(&ztest_name_lock); 4898 4899 /* 4900 * If we can tolerate two or more faults, or we're dealing 4901 * with a slog, randomly online/offline vd0. 4902 */ 4903 if ((maxfaults >= 2 || islog) && guid0 != 0) { 4904 if (ztest_random(10) < 6) { 4905 int flags = (ztest_random(2) == 0 ? 4906 ZFS_OFFLINE_TEMPORARY : 0); 4907 4908 /* 4909 * We have to grab the zs_name_lock as writer to 4910 * prevent a race between offlining a slog and 4911 * destroying a dataset. Offlining the slog will 4912 * grab a reference on the dataset which may cause 4913 * dmu_objset_destroy() to fail with EBUSY thus 4914 * leaving the dataset in an inconsistent state. 4915 */ 4916 if (islog) 4917 (void) rw_wrlock(&ztest_name_lock); 4918 4919 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY); 4920 4921 if (islog) 4922 (void) rw_unlock(&ztest_name_lock); 4923 } else { 4924 /* 4925 * Ideally we would like to be able to randomly 4926 * call vdev_[on|off]line without holding locks 4927 * to force unpredictable failures but the side 4928 * effects of vdev_[on|off]line prevent us from 4929 * doing so. We grab the ztest_vdev_lock here to 4930 * prevent a race between injection testing and 4931 * aux_vdev removal. 4932 */ 4933 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 4934 (void) vdev_online(spa, guid0, 0, NULL); 4935 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4936 } 4937 } 4938 4939 if (maxfaults == 0) 4940 return; 4941 4942 /* 4943 * We have at least single-fault tolerance, so inject data corruption. 4944 */ 4945 fd = open(pathrand, O_RDWR); 4946 4947 if (fd == -1) /* we hit a gap in the device namespace */ 4948 return; 4949 4950 fsize = lseek(fd, 0, SEEK_END); 4951 4952 while (--iters != 0) { 4953 offset = ztest_random(fsize / (leaves << bshift)) * 4954 (leaves << bshift) + (leaf << bshift) + 4955 (ztest_random(1ULL << (bshift - 1)) & -8ULL); 4956 4957 if (offset >= fsize) 4958 continue; 4959 4960 VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 4961 if (mirror_save != zs->zs_mirrors) { 4962 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4963 (void) close(fd); 4964 return; 4965 } 4966 4967 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) 4968 fatal(1, "can't inject bad word at 0x%llx in %s", 4969 offset, pathrand); 4970 4971 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4972 4973 if (ztest_opts.zo_verbose >= 7) 4974 (void) printf("injected bad word into %s," 4975 " offset 0x%llx\n", pathrand, (u_longlong_t)offset); 4976 } 4977 4978 (void) close(fd); 4979} 4980 4981/* 4982 * Verify that DDT repair works as expected. 4983 */ 4984void 4985ztest_ddt_repair(ztest_ds_t *zd, uint64_t id) 4986{ 4987 ztest_shared_t *zs = ztest_shared; 4988 spa_t *spa = ztest_spa; 4989 objset_t *os = zd->zd_os; 4990 ztest_od_t od[1]; 4991 uint64_t object, blocksize, txg, pattern, psize; 4992 enum zio_checksum checksum = spa_dedup_checksum(spa); 4993 dmu_buf_t *db; 4994 dmu_tx_t *tx; 4995 void *buf; 4996 blkptr_t blk; 4997 int copies = 2 * ZIO_DEDUPDITTO_MIN; 4998 4999 blocksize = ztest_random_blocksize(); 5000 blocksize = MIN(blocksize, 2048); /* because we write so many */ 5001 5002 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 5003 5004 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 5005 return; 5006 5007 /* 5008 * Take the name lock as writer to prevent anyone else from changing 5009 * the pool and dataset properies we need to maintain during this test. 5010 */ 5011 (void) rw_wrlock(&ztest_name_lock); 5012 5013 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum, 5014 B_FALSE) != 0 || 5015 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1, 5016 B_FALSE) != 0) { 5017 (void) rw_unlock(&ztest_name_lock); 5018 return; 5019 } 5020 5021 object = od[0].od_object; 5022 blocksize = od[0].od_blocksize; 5023 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os); 5024 5025 ASSERT(object != 0); 5026 5027 tx = dmu_tx_create(os); 5028 dmu_tx_hold_write(tx, object, 0, copies * blocksize); 5029 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 5030 if (txg == 0) { 5031 (void) rw_unlock(&ztest_name_lock); 5032 return; 5033 } 5034 5035 /* 5036 * Write all the copies of our block. 5037 */ 5038 for (int i = 0; i < copies; i++) { 5039 uint64_t offset = i * blocksize; 5040 int error = dmu_buf_hold(os, object, offset, FTAG, &db, 5041 DMU_READ_NO_PREFETCH); 5042 if (error != 0) { 5043 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u", 5044 os, (long long)object, (long long) offset, error); 5045 } 5046 ASSERT(db->db_offset == offset); 5047 ASSERT(db->db_size == blocksize); 5048 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) || 5049 ztest_pattern_match(db->db_data, db->db_size, 0ULL)); 5050 dmu_buf_will_fill(db, tx); 5051 ztest_pattern_set(db->db_data, db->db_size, pattern); 5052 dmu_buf_rele(db, FTAG); 5053 } 5054 5055 dmu_tx_commit(tx); 5056 txg_wait_synced(spa_get_dsl(spa), txg); 5057 5058 /* 5059 * Find out what block we got. 5060 */ 5061 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db, 5062 DMU_READ_NO_PREFETCH)); 5063 blk = *((dmu_buf_impl_t *)db)->db_blkptr; 5064 dmu_buf_rele(db, FTAG); 5065 5066 /* 5067 * Damage the block. Dedup-ditto will save us when we read it later. 5068 */ 5069 psize = BP_GET_PSIZE(&blk); 5070 buf = zio_buf_alloc(psize); 5071 ztest_pattern_set(buf, psize, ~pattern); 5072 5073 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk, 5074 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE, 5075 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL)); 5076 5077 zio_buf_free(buf, psize); 5078 5079 (void) rw_unlock(&ztest_name_lock); 5080} 5081 5082/* 5083 * Scrub the pool. 5084 */ 5085/* ARGSUSED */ 5086void 5087ztest_scrub(ztest_ds_t *zd, uint64_t id) 5088{ 5089 spa_t *spa = ztest_spa; 5090 5091 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5092 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */ 5093 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5094} 5095 5096/* 5097 * Change the guid for the pool. 5098 */ 5099/* ARGSUSED */ 5100void 5101ztest_reguid(ztest_ds_t *zd, uint64_t id) 5102{ 5103 spa_t *spa = ztest_spa; 5104 uint64_t orig, load; 5105 int error; 5106 5107 orig = spa_guid(spa); 5108 load = spa_load_guid(spa); 5109 5110 (void) rw_wrlock(&ztest_name_lock); 5111 error = spa_change_guid(spa); 5112 (void) rw_unlock(&ztest_name_lock); 5113 5114 if (error != 0) 5115 return; 5116 5117 if (ztest_opts.zo_verbose >= 4) { 5118 (void) printf("Changed guid old %llu -> %llu\n", 5119 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa)); 5120 } 5121 5122 VERIFY3U(orig, !=, spa_guid(spa)); 5123 VERIFY3U(load, ==, spa_load_guid(spa)); 5124} 5125 5126/* 5127 * Rename the pool to a different name and then rename it back. 5128 */ 5129/* ARGSUSED */ 5130void 5131ztest_spa_rename(ztest_ds_t *zd, uint64_t id) 5132{ 5133 char *oldname, *newname; 5134 spa_t *spa; 5135 5136 (void) rw_wrlock(&ztest_name_lock); 5137 5138 oldname = ztest_opts.zo_pool; 5139 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); 5140 (void) strcpy(newname, oldname); 5141 (void) strcat(newname, "_tmp"); 5142 5143 /* 5144 * Do the rename 5145 */ 5146 VERIFY3U(0, ==, spa_rename(oldname, newname)); 5147 5148 /* 5149 * Try to open it under the old name, which shouldn't exist 5150 */ 5151 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); 5152 5153 /* 5154 * Open it under the new name and make sure it's still the same spa_t. 5155 */ 5156 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); 5157 5158 ASSERT(spa == ztest_spa); 5159 spa_close(spa, FTAG); 5160 5161 /* 5162 * Rename it back to the original 5163 */ 5164 VERIFY3U(0, ==, spa_rename(newname, oldname)); 5165 5166 /* 5167 * Make sure it can still be opened 5168 */ 5169 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); 5170 5171 ASSERT(spa == ztest_spa); 5172 spa_close(spa, FTAG); 5173 5174 umem_free(newname, strlen(newname) + 1); 5175 5176 (void) rw_unlock(&ztest_name_lock); 5177} 5178 5179/* 5180 * Verify pool integrity by running zdb. 5181 */ 5182static void 5183ztest_run_zdb(char *pool) 5184{ 5185 int status; 5186 char zdb[MAXPATHLEN + MAXNAMELEN + 20]; 5187 char zbuf[1024]; 5188 char *bin; 5189 char *ztest; 5190 char *isa; 5191 int isalen; 5192 FILE *fp; 5193 5194 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb)); 5195 5196 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ 5197 bin = strstr(zdb, "/usr/bin/"); 5198 ztest = strstr(bin, "/ztest"); 5199 isa = bin + 8; 5200 isalen = ztest - isa; 5201 isa = strdup(isa); 5202 /* LINTED */ 5203 (void) sprintf(bin, 5204 "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s", 5205 isalen, 5206 isa, 5207 ztest_opts.zo_verbose >= 3 ? "s" : "", 5208 ztest_opts.zo_verbose >= 4 ? "v" : "", 5209 spa_config_path, 5210 pool); 5211 free(isa); 5212 5213 if (ztest_opts.zo_verbose >= 5) 5214 (void) printf("Executing %s\n", strstr(zdb, "zdb ")); 5215 5216 fp = popen(zdb, "r"); 5217 assert(fp != NULL); 5218 5219 while (fgets(zbuf, sizeof (zbuf), fp) != NULL) 5220 if (ztest_opts.zo_verbose >= 3) 5221 (void) printf("%s", zbuf); 5222 5223 status = pclose(fp); 5224 5225 if (status == 0) 5226 return; 5227 5228 ztest_dump_core = 0; 5229 if (WIFEXITED(status)) 5230 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); 5231 else 5232 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); 5233} 5234 5235static void 5236ztest_walk_pool_directory(char *header) 5237{ 5238 spa_t *spa = NULL; 5239 5240 if (ztest_opts.zo_verbose >= 6) 5241 (void) printf("%s\n", header); 5242 5243 mutex_enter(&spa_namespace_lock); 5244 while ((spa = spa_next(spa)) != NULL) 5245 if (ztest_opts.zo_verbose >= 6) 5246 (void) printf("\t%s\n", spa_name(spa)); 5247 mutex_exit(&spa_namespace_lock); 5248} 5249 5250static void 5251ztest_spa_import_export(char *oldname, char *newname) 5252{ 5253 nvlist_t *config, *newconfig; 5254 uint64_t pool_guid; 5255 spa_t *spa; 5256 int error; 5257 5258 if (ztest_opts.zo_verbose >= 4) { 5259 (void) printf("import/export: old = %s, new = %s\n", 5260 oldname, newname); 5261 } 5262 5263 /* 5264 * Clean up from previous runs. 5265 */ 5266 (void) spa_destroy(newname); 5267 5268 /* 5269 * Get the pool's configuration and guid. 5270 */ 5271 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); 5272 5273 /* 5274 * Kick off a scrub to tickle scrub/export races. 5275 */ 5276 if (ztest_random(2) == 0) 5277 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5278 5279 pool_guid = spa_guid(spa); 5280 spa_close(spa, FTAG); 5281 5282 ztest_walk_pool_directory("pools before export"); 5283 5284 /* 5285 * Export it. 5286 */ 5287 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE)); 5288 5289 ztest_walk_pool_directory("pools after export"); 5290 5291 /* 5292 * Try to import it. 5293 */ 5294 newconfig = spa_tryimport(config); 5295 ASSERT(newconfig != NULL); 5296 nvlist_free(newconfig); 5297 5298 /* 5299 * Import it under the new name. 5300 */ 5301 error = spa_import(newname, config, NULL, 0); 5302 if (error != 0) { 5303 dump_nvlist(config, 0); 5304 fatal(B_FALSE, "couldn't import pool %s as %s: error %u", 5305 oldname, newname, error); 5306 } 5307 5308 ztest_walk_pool_directory("pools after import"); 5309 5310 /* 5311 * Try to import it again -- should fail with EEXIST. 5312 */ 5313 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0)); 5314 5315 /* 5316 * Try to import it under a different name -- should fail with EEXIST. 5317 */ 5318 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0)); 5319 5320 /* 5321 * Verify that the pool is no longer visible under the old name. 5322 */ 5323 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); 5324 5325 /* 5326 * Verify that we can open and close the pool using the new name. 5327 */ 5328 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); 5329 ASSERT(pool_guid == spa_guid(spa)); 5330 spa_close(spa, FTAG); 5331 5332 nvlist_free(config); 5333} 5334 5335static void 5336ztest_resume(spa_t *spa) 5337{ 5338 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6) 5339 (void) printf("resuming from suspended state\n"); 5340 spa_vdev_state_enter(spa, SCL_NONE); 5341 vdev_clear(spa, NULL); 5342 (void) spa_vdev_state_exit(spa, NULL, 0); 5343 (void) zio_resume(spa); 5344} 5345 5346static void * 5347ztest_resume_thread(void *arg) 5348{ 5349 spa_t *spa = arg; 5350 5351 while (!ztest_exiting) { 5352 if (spa_suspended(spa)) 5353 ztest_resume(spa); 5354 (void) poll(NULL, 0, 100); 5355 } 5356 return (NULL); 5357} 5358 5359static void * 5360ztest_deadman_thread(void *arg) 5361{ 5362 ztest_shared_t *zs = arg; 5363 spa_t *spa = ztest_spa; 5364 hrtime_t delta, total = 0; 5365 5366 for (;;) { 5367 delta = zs->zs_thread_stop - zs->zs_thread_start + 5368 MSEC2NSEC(zfs_deadman_synctime_ms); 5369 5370 (void) poll(NULL, 0, (int)NSEC2MSEC(delta)); 5371 5372 /* 5373 * If the pool is suspended then fail immediately. Otherwise, 5374 * check to see if the pool is making any progress. If 5375 * vdev_deadman() discovers that there hasn't been any recent 5376 * I/Os then it will end up aborting the tests. 5377 */ 5378 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) { 5379 fatal(0, "aborting test after %llu seconds because " 5380 "pool has transitioned to a suspended state.", 5381 zfs_deadman_synctime_ms / 1000); 5382 return (NULL); 5383 } 5384 vdev_deadman(spa->spa_root_vdev); 5385 5386 total += zfs_deadman_synctime_ms/1000; 5387 (void) printf("ztest has been running for %lld seconds\n", 5388 total); 5389 } 5390} 5391 5392static void 5393ztest_execute(int test, ztest_info_t *zi, uint64_t id) 5394{ 5395 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets]; 5396 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test); 5397 hrtime_t functime = gethrtime(); 5398 5399 for (int i = 0; i < zi->zi_iters; i++) 5400 zi->zi_func(zd, id); 5401 5402 functime = gethrtime() - functime; 5403 5404 atomic_add_64(&zc->zc_count, 1); 5405 atomic_add_64(&zc->zc_time, functime); 5406 5407 if (ztest_opts.zo_verbose >= 4) { 5408 Dl_info dli; 5409 (void) dladdr((void *)zi->zi_func, &dli); 5410 (void) printf("%6.2f sec in %s\n", 5411 (double)functime / NANOSEC, dli.dli_sname); 5412 } 5413} 5414 5415static void * 5416ztest_thread(void *arg) 5417{ 5418 int rand; 5419 uint64_t id = (uintptr_t)arg; 5420 ztest_shared_t *zs = ztest_shared; 5421 uint64_t call_next; 5422 hrtime_t now; 5423 ztest_info_t *zi; 5424 ztest_shared_callstate_t *zc; 5425 5426 while ((now = gethrtime()) < zs->zs_thread_stop) { 5427 /* 5428 * See if it's time to force a crash. 5429 */ 5430 if (now > zs->zs_thread_kill) 5431 ztest_kill(zs); 5432 5433 /* 5434 * If we're getting ENOSPC with some regularity, stop. 5435 */ 5436 if (zs->zs_enospc_count > 10) 5437 break; 5438 5439 /* 5440 * Pick a random function to execute. 5441 */ 5442 rand = ztest_random(ZTEST_FUNCS); 5443 zi = &ztest_info[rand]; 5444 zc = ZTEST_GET_SHARED_CALLSTATE(rand); 5445 call_next = zc->zc_next; 5446 5447 if (now >= call_next && 5448 atomic_cas_64(&zc->zc_next, call_next, call_next + 5449 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) { 5450 ztest_execute(rand, zi, id); 5451 } 5452 } 5453 5454 return (NULL); 5455} 5456 5457static void 5458ztest_dataset_name(char *dsname, char *pool, int d) 5459{ 5460 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d); 5461} 5462 5463static void 5464ztest_dataset_destroy(int d) 5465{ 5466 char name[MAXNAMELEN]; 5467 5468 ztest_dataset_name(name, ztest_opts.zo_pool, d); 5469 5470 if (ztest_opts.zo_verbose >= 3) 5471 (void) printf("Destroying %s to free up space\n", name); 5472 5473 /* 5474 * Cleanup any non-standard clones and snapshots. In general, 5475 * ztest thread t operates on dataset (t % zopt_datasets), 5476 * so there may be more than one thing to clean up. 5477 */ 5478 for (int t = d; t < ztest_opts.zo_threads; 5479 t += ztest_opts.zo_datasets) { 5480 ztest_dsl_dataset_cleanup(name, t); 5481 } 5482 5483 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, 5484 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); 5485} 5486 5487static void 5488ztest_dataset_dirobj_verify(ztest_ds_t *zd) 5489{ 5490 uint64_t usedobjs, dirobjs, scratch; 5491 5492 /* 5493 * ZTEST_DIROBJ is the object directory for the entire dataset. 5494 * Therefore, the number of objects in use should equal the 5495 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself. 5496 * If not, we have an object leak. 5497 * 5498 * Note that we can only check this in ztest_dataset_open(), 5499 * when the open-context and syncing-context values agree. 5500 * That's because zap_count() returns the open-context value, 5501 * while dmu_objset_space() returns the rootbp fill count. 5502 */ 5503 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs)); 5504 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch); 5505 ASSERT3U(dirobjs + 1, ==, usedobjs); 5506} 5507 5508static int 5509ztest_dataset_open(int d) 5510{ 5511 ztest_ds_t *zd = &ztest_ds[d]; 5512 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq; 5513 objset_t *os; 5514 zilog_t *zilog; 5515 char name[MAXNAMELEN]; 5516 int error; 5517 5518 ztest_dataset_name(name, ztest_opts.zo_pool, d); 5519 5520 (void) rw_rdlock(&ztest_name_lock); 5521 5522 error = ztest_dataset_create(name); 5523 if (error == ENOSPC) { 5524 (void) rw_unlock(&ztest_name_lock); 5525 ztest_record_enospc(FTAG); 5526 return (error); 5527 } 5528 ASSERT(error == 0 || error == EEXIST); 5529 5530 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os)); 5531 (void) rw_unlock(&ztest_name_lock); 5532 5533 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os); 5534 5535 zilog = zd->zd_zilog; 5536 5537 if (zilog->zl_header->zh_claim_lr_seq != 0 && 5538 zilog->zl_header->zh_claim_lr_seq < committed_seq) 5539 fatal(0, "missing log records: claimed %llu < committed %llu", 5540 zilog->zl_header->zh_claim_lr_seq, committed_seq); 5541 5542 ztest_dataset_dirobj_verify(zd); 5543 5544 zil_replay(os, zd, ztest_replay_vector); 5545 5546 ztest_dataset_dirobj_verify(zd); 5547 5548 if (ztest_opts.zo_verbose >= 6) 5549 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n", 5550 zd->zd_name, 5551 (u_longlong_t)zilog->zl_parse_blk_count, 5552 (u_longlong_t)zilog->zl_parse_lr_count, 5553 (u_longlong_t)zilog->zl_replaying_seq); 5554 5555 zilog = zil_open(os, ztest_get_data); 5556 5557 if (zilog->zl_replaying_seq != 0 && 5558 zilog->zl_replaying_seq < committed_seq) 5559 fatal(0, "missing log records: replayed %llu < committed %llu", 5560 zilog->zl_replaying_seq, committed_seq); 5561 5562 return (0); 5563} 5564 5565static void 5566ztest_dataset_close(int d) 5567{ 5568 ztest_ds_t *zd = &ztest_ds[d]; 5569 5570 zil_close(zd->zd_zilog); 5571 dmu_objset_disown(zd->zd_os, zd); 5572 5573 ztest_zd_fini(zd); 5574} 5575 5576/* 5577 * Kick off threads to run tests on all datasets in parallel. 5578 */ 5579static void 5580ztest_run(ztest_shared_t *zs) 5581{ 5582 thread_t *tid; 5583 spa_t *spa; 5584 objset_t *os; 5585 thread_t resume_tid; 5586 int error; 5587 5588 ztest_exiting = B_FALSE; 5589 5590 /* 5591 * Initialize parent/child shared state. 5592 */ 5593 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0); 5594 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0); 5595 5596 zs->zs_thread_start = gethrtime(); 5597 zs->zs_thread_stop = 5598 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC; 5599 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop); 5600 zs->zs_thread_kill = zs->zs_thread_stop; 5601 if (ztest_random(100) < ztest_opts.zo_killrate) { 5602 zs->zs_thread_kill -= 5603 ztest_random(ztest_opts.zo_passtime * NANOSEC); 5604 } 5605 5606 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL); 5607 5608 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t), 5609 offsetof(ztest_cb_data_t, zcd_node)); 5610 5611 /* 5612 * Open our pool. 5613 */ 5614 kernel_init(FREAD | FWRITE); 5615 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5616 spa->spa_debug = B_TRUE; 5617 metaslab_preload_limit = ztest_random(20) + 1; 5618 ztest_spa = spa; 5619 5620 VERIFY0(dmu_objset_own(ztest_opts.zo_pool, 5621 DMU_OST_ANY, B_TRUE, FTAG, &os)); 5622 zs->zs_guid = dmu_objset_fsid_guid(os); 5623 dmu_objset_disown(os, FTAG); 5624 5625 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN; 5626 5627 /* 5628 * We don't expect the pool to suspend unless maxfaults == 0, 5629 * in which case ztest_fault_inject() temporarily takes away 5630 * the only valid replica. 5631 */ 5632 if (MAXFAULTS() == 0) 5633 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT; 5634 else 5635 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC; 5636 5637 /* 5638 * Create a thread to periodically resume suspended I/O. 5639 */ 5640 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND, 5641 &resume_tid) == 0); 5642 5643 /* 5644 * Create a deadman thread to abort() if we hang. 5645 */ 5646 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND, 5647 NULL) == 0); 5648 5649 /* 5650 * Verify that we can safely inquire about about any object, 5651 * whether it's allocated or not. To make it interesting, 5652 * we probe a 5-wide window around each power of two. 5653 * This hits all edge cases, including zero and the max. 5654 */ 5655 for (int t = 0; t < 64; t++) { 5656 for (int d = -5; d <= 5; d++) { 5657 error = dmu_object_info(spa->spa_meta_objset, 5658 (1ULL << t) + d, NULL); 5659 ASSERT(error == 0 || error == ENOENT || 5660 error == EINVAL); 5661 } 5662 } 5663 5664 /* 5665 * If we got any ENOSPC errors on the previous run, destroy something. 5666 */ 5667 if (zs->zs_enospc_count != 0) { 5668 int d = ztest_random(ztest_opts.zo_datasets); 5669 ztest_dataset_destroy(d); 5670 } 5671 zs->zs_enospc_count = 0; 5672 5673 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t), 5674 UMEM_NOFAIL); 5675 5676 if (ztest_opts.zo_verbose >= 4) 5677 (void) printf("starting main threads...\n"); 5678 5679 /* 5680 * Kick off all the tests that run in parallel. 5681 */ 5682 for (int t = 0; t < ztest_opts.zo_threads; t++) { 5683 if (t < ztest_opts.zo_datasets && 5684 ztest_dataset_open(t) != 0) 5685 return; 5686 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t, 5687 THR_BOUND, &tid[t]) == 0); 5688 } 5689 5690 /* 5691 * Wait for all of the tests to complete. We go in reverse order 5692 * so we don't close datasets while threads are still using them. 5693 */ 5694 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) { 5695 VERIFY(thr_join(tid[t], NULL, NULL) == 0); 5696 if (t < ztest_opts.zo_datasets) 5697 ztest_dataset_close(t); 5698 } 5699 5700 txg_wait_synced(spa_get_dsl(spa), 0); 5701 5702 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 5703 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa)); 5704 zfs_dbgmsg_print(FTAG); 5705 5706 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t)); 5707 5708 /* Kill the resume thread */ 5709 ztest_exiting = B_TRUE; 5710 VERIFY(thr_join(resume_tid, NULL, NULL) == 0); 5711 ztest_resume(spa); 5712 5713 /* 5714 * Right before closing the pool, kick off a bunch of async I/O; 5715 * spa_close() should wait for it to complete. 5716 */ 5717 for (uint64_t object = 1; object < 50; object++) 5718 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20); 5719 5720 spa_close(spa, FTAG); 5721 5722 /* 5723 * Verify that we can loop over all pools. 5724 */ 5725 mutex_enter(&spa_namespace_lock); 5726 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) 5727 if (ztest_opts.zo_verbose > 3) 5728 (void) printf("spa_next: found %s\n", spa_name(spa)); 5729 mutex_exit(&spa_namespace_lock); 5730 5731 /* 5732 * Verify that we can export the pool and reimport it under a 5733 * different name. 5734 */ 5735 if (ztest_random(2) == 0) { 5736 char name[MAXNAMELEN]; 5737 (void) snprintf(name, MAXNAMELEN, "%s_import", 5738 ztest_opts.zo_pool); 5739 ztest_spa_import_export(ztest_opts.zo_pool, name); 5740 ztest_spa_import_export(name, ztest_opts.zo_pool); 5741 } 5742 5743 kernel_fini(); 5744 5745 list_destroy(&zcl.zcl_callbacks); 5746 5747 (void) _mutex_destroy(&zcl.zcl_callbacks_lock); 5748 5749 (void) rwlock_destroy(&ztest_name_lock); 5750 (void) _mutex_destroy(&ztest_vdev_lock); 5751} 5752 5753static void 5754ztest_freeze(void) 5755{ 5756 ztest_ds_t *zd = &ztest_ds[0]; 5757 spa_t *spa; 5758 int numloops = 0; 5759 5760 if (ztest_opts.zo_verbose >= 3) 5761 (void) printf("testing spa_freeze()...\n"); 5762 5763 kernel_init(FREAD | FWRITE); 5764 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5765 VERIFY3U(0, ==, ztest_dataset_open(0)); 5766 spa->spa_debug = B_TRUE; 5767 ztest_spa = spa; 5768 5769 /* 5770 * Force the first log block to be transactionally allocated. 5771 * We have to do this before we freeze the pool -- otherwise 5772 * the log chain won't be anchored. 5773 */ 5774 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) { 5775 ztest_dmu_object_alloc_free(zd, 0); 5776 zil_commit(zd->zd_zilog, 0); 5777 } 5778 5779 txg_wait_synced(spa_get_dsl(spa), 0); 5780 5781 /* 5782 * Freeze the pool. This stops spa_sync() from doing anything, 5783 * so that the only way to record changes from now on is the ZIL. 5784 */ 5785 spa_freeze(spa); 5786 5787 /* 5788 * Because it is hard to predict how much space a write will actually 5789 * require beforehand, we leave ourselves some fudge space to write over 5790 * capacity. 5791 */ 5792 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2; 5793 5794 /* 5795 * Run tests that generate log records but don't alter the pool config 5796 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc). 5797 * We do a txg_wait_synced() after each iteration to force the txg 5798 * to increase well beyond the last synced value in the uberblock. 5799 * The ZIL should be OK with that. 5800 * 5801 * Run a random number of times less than zo_maxloops and ensure we do 5802 * not run out of space on the pool. 5803 */ 5804 while (ztest_random(10) != 0 && 5805 numloops++ < ztest_opts.zo_maxloops && 5806 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) { 5807 ztest_od_t od; 5808 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 5809 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE)); 5810 ztest_io(zd, od.od_object, 5811 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 5812 txg_wait_synced(spa_get_dsl(spa), 0); 5813 } 5814 5815 /* 5816 * Commit all of the changes we just generated. 5817 */ 5818 zil_commit(zd->zd_zilog, 0); 5819 txg_wait_synced(spa_get_dsl(spa), 0); 5820 5821 /* 5822 * Close our dataset and close the pool. 5823 */ 5824 ztest_dataset_close(0); 5825 spa_close(spa, FTAG); 5826 kernel_fini(); 5827 5828 /* 5829 * Open and close the pool and dataset to induce log replay. 5830 */ 5831 kernel_init(FREAD | FWRITE); 5832 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5833 ASSERT(spa_freeze_txg(spa) == UINT64_MAX); 5834 VERIFY3U(0, ==, ztest_dataset_open(0)); 5835 ztest_dataset_close(0); 5836 5837 spa->spa_debug = B_TRUE; 5838 ztest_spa = spa; 5839 txg_wait_synced(spa_get_dsl(spa), 0); 5840 ztest_reguid(NULL, 0); 5841 5842 spa_close(spa, FTAG); 5843 kernel_fini(); 5844} 5845 5846void 5847print_time(hrtime_t t, char *timebuf) 5848{ 5849 hrtime_t s = t / NANOSEC; 5850 hrtime_t m = s / 60; 5851 hrtime_t h = m / 60; 5852 hrtime_t d = h / 24; 5853 5854 s -= m * 60; 5855 m -= h * 60; 5856 h -= d * 24; 5857 5858 timebuf[0] = '\0'; 5859 5860 if (d) 5861 (void) sprintf(timebuf, 5862 "%llud%02lluh%02llum%02llus", d, h, m, s); 5863 else if (h) 5864 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); 5865 else if (m) 5866 (void) sprintf(timebuf, "%llum%02llus", m, s); 5867 else 5868 (void) sprintf(timebuf, "%llus", s); 5869} 5870 5871static nvlist_t * 5872make_random_props() 5873{ 5874 nvlist_t *props; 5875 5876 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); 5877 if (ztest_random(2) == 0) 5878 return (props); 5879 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0); 5880 5881 return (props); 5882} 5883 5884/* 5885 * Create a storage pool with the given name and initial vdev size. 5886 * Then test spa_freeze() functionality. 5887 */ 5888static void 5889ztest_init(ztest_shared_t *zs) 5890{ 5891 spa_t *spa; 5892 nvlist_t *nvroot, *props; 5893 5894 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0); 5895 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0); 5896 5897 kernel_init(FREAD | FWRITE); 5898 5899 /* 5900 * Create the storage pool. 5901 */ 5902 (void) spa_destroy(ztest_opts.zo_pool); 5903 ztest_shared->zs_vdev_next_leaf = 0; 5904 zs->zs_splits = 0; 5905 zs->zs_mirrors = ztest_opts.zo_mirrors; 5906 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0, 5907 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1); 5908 props = make_random_props(); 5909 for (int i = 0; i < SPA_FEATURES; i++) { 5910 char buf[1024]; 5911 (void) snprintf(buf, sizeof (buf), "feature@%s", 5912 spa_feature_table[i].fi_uname); 5913 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0)); 5914 } 5915 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL)); 5916 nvlist_free(nvroot); 5917 5918 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5919 zs->zs_metaslab_sz = 5920 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift; 5921 5922 spa_close(spa, FTAG); 5923 5924 kernel_fini(); 5925 5926 ztest_run_zdb(ztest_opts.zo_pool); 5927 5928 ztest_freeze(); 5929 5930 ztest_run_zdb(ztest_opts.zo_pool); 5931 5932 (void) rwlock_destroy(&ztest_name_lock); 5933 (void) _mutex_destroy(&ztest_vdev_lock); 5934} 5935 5936static void 5937setup_data_fd(void) 5938{ 5939 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX"; 5940 5941 ztest_fd_data = mkstemp(ztest_name_data); 5942 ASSERT3S(ztest_fd_data, >=, 0); 5943 (void) unlink(ztest_name_data); 5944} 5945 5946 5947static int 5948shared_data_size(ztest_shared_hdr_t *hdr) 5949{ 5950 int size; 5951 5952 size = hdr->zh_hdr_size; 5953 size += hdr->zh_opts_size; 5954 size += hdr->zh_size; 5955 size += hdr->zh_stats_size * hdr->zh_stats_count; 5956 size += hdr->zh_ds_size * hdr->zh_ds_count; 5957 5958 return (size); 5959} 5960 5961static void 5962setup_hdr(void) 5963{ 5964 int size; 5965 ztest_shared_hdr_t *hdr; 5966 5967 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), 5968 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); 5969 ASSERT(hdr != MAP_FAILED); 5970 5971 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t))); 5972 5973 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t); 5974 hdr->zh_opts_size = sizeof (ztest_shared_opts_t); 5975 hdr->zh_size = sizeof (ztest_shared_t); 5976 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t); 5977 hdr->zh_stats_count = ZTEST_FUNCS; 5978 hdr->zh_ds_size = sizeof (ztest_shared_ds_t); 5979 hdr->zh_ds_count = ztest_opts.zo_datasets; 5980 5981 size = shared_data_size(hdr); 5982 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size)); 5983 5984 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); 5985} 5986 5987static void 5988setup_data(void) 5989{ 5990 int size, offset; 5991 ztest_shared_hdr_t *hdr; 5992 uint8_t *buf; 5993 5994 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), 5995 PROT_READ, MAP_SHARED, ztest_fd_data, 0); 5996 ASSERT(hdr != MAP_FAILED); 5997 5998 size = shared_data_size(hdr); 5999 6000 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); 6001 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()), 6002 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); 6003 ASSERT(hdr != MAP_FAILED); 6004 buf = (uint8_t *)hdr; 6005 6006 offset = hdr->zh_hdr_size; 6007 ztest_shared_opts = (void *)&buf[offset]; 6008 offset += hdr->zh_opts_size; 6009 ztest_shared = (void *)&buf[offset]; 6010 offset += hdr->zh_size; 6011 ztest_shared_callstate = (void *)&buf[offset]; 6012 offset += hdr->zh_stats_size * hdr->zh_stats_count; 6013 ztest_shared_ds = (void *)&buf[offset]; 6014} 6015 6016static boolean_t 6017exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp) 6018{ 6019 pid_t pid; 6020 int status; 6021 char *cmdbuf = NULL; 6022 6023 pid = fork(); 6024 6025 if (cmd == NULL) { 6026 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 6027 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN); 6028 cmd = cmdbuf; 6029 } 6030 6031 if (pid == -1) 6032 fatal(1, "fork failed"); 6033 6034 if (pid == 0) { /* child */ 6035 char *emptyargv[2] = { cmd, NULL }; 6036 char fd_data_str[12]; 6037 6038 struct rlimit rl = { 1024, 1024 }; 6039 (void) setrlimit(RLIMIT_NOFILE, &rl); 6040 6041 (void) close(ztest_fd_rand); 6042 VERIFY3U(11, >=, 6043 snprintf(fd_data_str, 12, "%d", ztest_fd_data)); 6044 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1)); 6045 6046 (void) enable_extended_FILE_stdio(-1, -1); 6047 if (libpath != NULL) 6048 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1)); 6049#ifdef illumos 6050 (void) execv(cmd, emptyargv); 6051#else 6052 (void) execvp(cmd, emptyargv); 6053#endif 6054 ztest_dump_core = B_FALSE; 6055 fatal(B_TRUE, "exec failed: %s", cmd); 6056 } 6057 6058 if (cmdbuf != NULL) { 6059 umem_free(cmdbuf, MAXPATHLEN); 6060 cmd = NULL; 6061 } 6062 6063 while (waitpid(pid, &status, 0) != pid) 6064 continue; 6065 if (statusp != NULL) 6066 *statusp = status; 6067 6068 if (WIFEXITED(status)) { 6069 if (WEXITSTATUS(status) != 0) { 6070 (void) fprintf(stderr, "child exited with code %d\n", 6071 WEXITSTATUS(status)); 6072 exit(2); 6073 } 6074 return (B_FALSE); 6075 } else if (WIFSIGNALED(status)) { 6076 if (!ignorekill || WTERMSIG(status) != SIGKILL) { 6077 (void) fprintf(stderr, "child died with signal %d\n", 6078 WTERMSIG(status)); 6079 exit(3); 6080 } 6081 return (B_TRUE); 6082 } else { 6083 (void) fprintf(stderr, "something strange happened to child\n"); 6084 exit(4); 6085 /* NOTREACHED */ 6086 } 6087} 6088 6089static void 6090ztest_run_init(void) 6091{ 6092 ztest_shared_t *zs = ztest_shared; 6093 6094 ASSERT(ztest_opts.zo_init != 0); 6095 6096 /* 6097 * Blow away any existing copy of zpool.cache 6098 */ 6099 (void) remove(spa_config_path); 6100 6101 /* 6102 * Create and initialize our storage pool. 6103 */ 6104 for (int i = 1; i <= ztest_opts.zo_init; i++) { 6105 bzero(zs, sizeof (ztest_shared_t)); 6106 if (ztest_opts.zo_verbose >= 3 && 6107 ztest_opts.zo_init != 1) { 6108 (void) printf("ztest_init(), pass %d\n", i); 6109 } 6110 ztest_init(zs); 6111 } 6112} 6113 6114int 6115main(int argc, char **argv) 6116{ 6117 int kills = 0; 6118 int iters = 0; 6119 int older = 0; 6120 int newer = 0; 6121 ztest_shared_t *zs; 6122 ztest_info_t *zi; 6123 ztest_shared_callstate_t *zc; 6124 char timebuf[100]; 6125 char numbuf[6]; 6126 spa_t *spa; 6127 char *cmd; 6128 boolean_t hasalt; 6129 char *fd_data_str = getenv("ZTEST_FD_DATA"); 6130 6131 (void) setvbuf(stdout, NULL, _IOLBF, 0); 6132 6133 dprintf_setup(&argc, argv); 6134 zfs_deadman_synctime_ms = 300000; 6135 6136 ztest_fd_rand = open("/dev/urandom", O_RDONLY); 6137 ASSERT3S(ztest_fd_rand, >=, 0); 6138 6139 if (!fd_data_str) { 6140 process_options(argc, argv); 6141 6142 setup_data_fd(); 6143 setup_hdr(); 6144 setup_data(); 6145 bcopy(&ztest_opts, ztest_shared_opts, 6146 sizeof (*ztest_shared_opts)); 6147 } else { 6148 ztest_fd_data = atoi(fd_data_str); 6149 setup_data(); 6150 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts)); 6151 } 6152 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count); 6153 6154 /* Override location of zpool.cache */ 6155 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache", 6156 ztest_opts.zo_dir), !=, -1); 6157 6158 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t), 6159 UMEM_NOFAIL); 6160 zs = ztest_shared; 6161 6162 if (fd_data_str) { 6163 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang; 6164 metaslab_df_alloc_threshold = 6165 zs->zs_metaslab_df_alloc_threshold; 6166 6167 if (zs->zs_do_init) 6168 ztest_run_init(); 6169 else 6170 ztest_run(zs); 6171 exit(0); 6172 } 6173 6174 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0); 6175 6176 if (ztest_opts.zo_verbose >= 1) { 6177 (void) printf("%llu vdevs, %d datasets, %d threads," 6178 " %llu seconds...\n", 6179 (u_longlong_t)ztest_opts.zo_vdevs, 6180 ztest_opts.zo_datasets, 6181 ztest_opts.zo_threads, 6182 (u_longlong_t)ztest_opts.zo_time); 6183 } 6184 6185 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); 6186 (void) strlcpy(cmd, getexecname(), MAXNAMELEN); 6187 6188 zs->zs_do_init = B_TRUE; 6189 if (strlen(ztest_opts.zo_alt_ztest) != 0) { 6190 if (ztest_opts.zo_verbose >= 1) { 6191 (void) printf("Executing older ztest for " 6192 "initialization: %s\n", ztest_opts.zo_alt_ztest); 6193 } 6194 VERIFY(!exec_child(ztest_opts.zo_alt_ztest, 6195 ztest_opts.zo_alt_libpath, B_FALSE, NULL)); 6196 } else { 6197 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL)); 6198 } 6199 zs->zs_do_init = B_FALSE; 6200 6201 zs->zs_proc_start = gethrtime(); 6202 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC; 6203 6204 for (int f = 0; f < ZTEST_FUNCS; f++) { 6205 zi = &ztest_info[f]; 6206 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6207 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop) 6208 zc->zc_next = UINT64_MAX; 6209 else 6210 zc->zc_next = zs->zs_proc_start + 6211 ztest_random(2 * zi->zi_interval[0] + 1); 6212 } 6213 6214 /* 6215 * Run the tests in a loop. These tests include fault injection 6216 * to verify that self-healing data works, and forced crashes 6217 * to verify that we never lose on-disk consistency. 6218 */ 6219 while (gethrtime() < zs->zs_proc_stop) { 6220 int status; 6221 boolean_t killed; 6222 6223 /* 6224 * Initialize the workload counters for each function. 6225 */ 6226 for (int f = 0; f < ZTEST_FUNCS; f++) { 6227 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6228 zc->zc_count = 0; 6229 zc->zc_time = 0; 6230 } 6231 6232 /* Set the allocation switch size */ 6233 zs->zs_metaslab_df_alloc_threshold = 6234 ztest_random(zs->zs_metaslab_sz / 4) + 1; 6235 6236 if (!hasalt || ztest_random(2) == 0) { 6237 if (hasalt && ztest_opts.zo_verbose >= 1) { 6238 (void) printf("Executing newer ztest: %s\n", 6239 cmd); 6240 } 6241 newer++; 6242 killed = exec_child(cmd, NULL, B_TRUE, &status); 6243 } else { 6244 if (hasalt && ztest_opts.zo_verbose >= 1) { 6245 (void) printf("Executing older ztest: %s\n", 6246 ztest_opts.zo_alt_ztest); 6247 } 6248 older++; 6249 killed = exec_child(ztest_opts.zo_alt_ztest, 6250 ztest_opts.zo_alt_libpath, B_TRUE, &status); 6251 } 6252 6253 if (killed) 6254 kills++; 6255 iters++; 6256 6257 if (ztest_opts.zo_verbose >= 1) { 6258 hrtime_t now = gethrtime(); 6259 6260 now = MIN(now, zs->zs_proc_stop); 6261 print_time(zs->zs_proc_stop - now, timebuf); 6262 nicenum(zs->zs_space, numbuf); 6263 6264 (void) printf("Pass %3d, %8s, %3llu ENOSPC, " 6265 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", 6266 iters, 6267 WIFEXITED(status) ? "Complete" : "SIGKILL", 6268 (u_longlong_t)zs->zs_enospc_count, 6269 100.0 * zs->zs_alloc / zs->zs_space, 6270 numbuf, 6271 100.0 * (now - zs->zs_proc_start) / 6272 (ztest_opts.zo_time * NANOSEC), timebuf); 6273 } 6274 6275 if (ztest_opts.zo_verbose >= 2) { 6276 (void) printf("\nWorkload summary:\n\n"); 6277 (void) printf("%7s %9s %s\n", 6278 "Calls", "Time", "Function"); 6279 (void) printf("%7s %9s %s\n", 6280 "-----", "----", "--------"); 6281 for (int f = 0; f < ZTEST_FUNCS; f++) { 6282 Dl_info dli; 6283 6284 zi = &ztest_info[f]; 6285 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6286 print_time(zc->zc_time, timebuf); 6287 (void) dladdr((void *)zi->zi_func, &dli); 6288 (void) printf("%7llu %9s %s\n", 6289 (u_longlong_t)zc->zc_count, timebuf, 6290 dli.dli_sname); 6291 } 6292 (void) printf("\n"); 6293 } 6294 6295 /* 6296 * It's possible that we killed a child during a rename test, 6297 * in which case we'll have a 'ztest_tmp' pool lying around 6298 * instead of 'ztest'. Do a blind rename in case this happened. 6299 */ 6300 kernel_init(FREAD); 6301 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) { 6302 spa_close(spa, FTAG); 6303 } else { 6304 char tmpname[MAXNAMELEN]; 6305 kernel_fini(); 6306 kernel_init(FREAD | FWRITE); 6307 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp", 6308 ztest_opts.zo_pool); 6309 (void) spa_rename(tmpname, ztest_opts.zo_pool); 6310 } 6311 kernel_fini(); 6312 6313 ztest_run_zdb(ztest_opts.zo_pool); 6314 } 6315 6316 if (ztest_opts.zo_verbose >= 1) { 6317 if (hasalt) { 6318 (void) printf("%d runs of older ztest: %s\n", older, 6319 ztest_opts.zo_alt_ztest); 6320 (void) printf("%d runs of newer ztest: %s\n", newer, 6321 cmd); 6322 } 6323 (void) printf("%d killed, %d completed, %.0f%% kill rate\n", 6324 kills, iters - kills, (100.0 * kills) / MAX(1, iters)); 6325 } 6326 6327 umem_free(cmd, MAXNAMELEN); 6328 6329 return (0); 6330} 6331