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, 2017 by Delphix. All rights reserved. 24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25 * Copyright (c) 2012 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 26 * Copyright (c) 2013 Steven Hartland. All rights reserved. 27 * Copyright (c) 2014 Integros [integros.com] 28 * Copyright 2017 Joyent, Inc. 29 * Copyright 2017 RackTop Systems. 30 */ 31 32/* 33 * The objective of this program is to provide a DMU/ZAP/SPA stress test 34 * that runs entirely in userland, is easy to use, and easy to extend. 35 * 36 * The overall design of the ztest program is as follows: 37 * 38 * (1) For each major functional area (e.g. adding vdevs to a pool, 39 * creating and destroying datasets, reading and writing objects, etc) 40 * we have a simple routine to test that functionality. These 41 * individual routines do not have to do anything "stressful". 42 * 43 * (2) We turn these simple functionality tests into a stress test by 44 * running them all in parallel, with as many threads as desired, 45 * and spread across as many datasets, objects, and vdevs as desired. 46 * 47 * (3) While all this is happening, we inject faults into the pool to 48 * verify that self-healing data really works. 49 * 50 * (4) Every time we open a dataset, we change its checksum and compression 51 * functions. Thus even individual objects vary from block to block 52 * in which checksum they use and whether they're compressed. 53 * 54 * (5) To verify that we never lose on-disk consistency after a crash, 55 * we run the entire test in a child of the main process. 56 * At random times, the child self-immolates with a SIGKILL. 57 * This is the software equivalent of pulling the power cord. 58 * The parent then runs the test again, using the existing 59 * storage pool, as many times as desired. If backwards compatibility 60 * testing is enabled ztest will sometimes run the "older" version 61 * of ztest after a SIGKILL. 62 * 63 * (6) To verify that we don't have future leaks or temporal incursions, 64 * many of the functional tests record the transaction group number 65 * as part of their data. When reading old data, they verify that 66 * the transaction group number is less than the current, open txg. 67 * If you add a new test, please do this if applicable. 68 * 69 * When run with no arguments, ztest runs for about five minutes and 70 * produces no output if successful. To get a little bit of information, 71 * specify -V. To get more information, specify -VV, and so on. 72 * 73 * To turn this into an overnight stress test, use -T to specify run time. 74 * 75 * You can ask more more vdevs [-v], datasets [-d], or threads [-t] 76 * to increase the pool capacity, fanout, and overall stress level. 77 * 78 * Use the -k option to set the desired frequency of kills. 79 * 80 * When ztest invokes itself it passes all relevant information through a 81 * temporary file which is mmap-ed in the child process. This allows shared 82 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always 83 * stored at offset 0 of this file and contains information on the size and 84 * number of shared structures in the file. The information stored in this file 85 * must remain backwards compatible with older versions of ztest so that 86 * ztest can invoke them during backwards compatibility testing (-B). 87 */ 88 89#include <sys/zfs_context.h> 90#include <sys/spa.h> 91#include <sys/dmu.h> 92#include <sys/txg.h> 93#include <sys/dbuf.h> 94#include <sys/zap.h> 95#include <sys/dmu_objset.h> 96#include <sys/poll.h> 97#include <sys/stat.h> 98#include <sys/time.h> 99#include <sys/wait.h> 100#include <sys/mman.h> 101#include <sys/resource.h> 102#include <sys/zio.h> 103#include <sys/zil.h> 104#include <sys/zil_impl.h> 105#include <sys/vdev_impl.h> 106#include <sys/vdev_file.h> 107#include <sys/vdev_initialize.h> 108#include <sys/spa_impl.h> 109#include <sys/metaslab_impl.h> 110#include <sys/dsl_prop.h> 111#include <sys/dsl_dataset.h> 112#include <sys/dsl_destroy.h> 113#include <sys/dsl_scan.h> 114#include <sys/zio_checksum.h> 115#include <sys/refcount.h> 116#include <sys/zfeature.h> 117#include <sys/dsl_userhold.h> 118#include <sys/abd.h> 119#include <stdio.h> 120#include <stdio_ext.h> 121#include <stdlib.h> 122#include <unistd.h> 123#include <signal.h> 124#include <umem.h> 125#include <dlfcn.h> 126#include <ctype.h> 127#include <math.h> 128#include <errno.h> 129#include <sys/fs/zfs.h> 130#include <libnvpair.h> 131#include <libcmdutils.h> 132 133static int ztest_fd_data = -1; 134static int ztest_fd_rand = -1; 135 136typedef struct ztest_shared_hdr { 137 uint64_t zh_hdr_size; 138 uint64_t zh_opts_size; 139 uint64_t zh_size; 140 uint64_t zh_stats_size; 141 uint64_t zh_stats_count; 142 uint64_t zh_ds_size; 143 uint64_t zh_ds_count; 144} ztest_shared_hdr_t; 145 146static ztest_shared_hdr_t *ztest_shared_hdr; 147 148typedef struct ztest_shared_opts { 149 char zo_pool[ZFS_MAX_DATASET_NAME_LEN]; 150 char zo_dir[ZFS_MAX_DATASET_NAME_LEN]; 151 char zo_alt_ztest[MAXNAMELEN]; 152 char zo_alt_libpath[MAXNAMELEN]; 153 uint64_t zo_vdevs; 154 uint64_t zo_vdevtime; 155 size_t zo_vdev_size; 156 int zo_ashift; 157 int zo_mirrors; 158 int zo_raidz; 159 int zo_raidz_parity; 160 int zo_datasets; 161 int zo_threads; 162 uint64_t zo_passtime; 163 uint64_t zo_killrate; 164 int zo_verbose; 165 int zo_init; 166 uint64_t zo_time; 167 uint64_t zo_maxloops; 168 uint64_t zo_metaslab_force_ganging; 169} ztest_shared_opts_t; 170 171static const ztest_shared_opts_t ztest_opts_defaults = { 172 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' }, 173 .zo_dir = { '/', 't', 'm', 'p', '\0' }, 174 .zo_alt_ztest = { '\0' }, 175 .zo_alt_libpath = { '\0' }, 176 .zo_vdevs = 5, 177 .zo_ashift = SPA_MINBLOCKSHIFT, 178 .zo_mirrors = 2, 179 .zo_raidz = 4, 180 .zo_raidz_parity = 1, 181 .zo_vdev_size = SPA_MINDEVSIZE * 4, /* 256m default size */ 182 .zo_datasets = 7, 183 .zo_threads = 23, 184 .zo_passtime = 60, /* 60 seconds */ 185 .zo_killrate = 70, /* 70% kill rate */ 186 .zo_verbose = 0, 187 .zo_init = 1, 188 .zo_time = 300, /* 5 minutes */ 189 .zo_maxloops = 50, /* max loops during spa_freeze() */ 190 .zo_metaslab_force_ganging = 32 << 10 191}; 192 193extern uint64_t metaslab_force_ganging; 194extern uint64_t metaslab_df_alloc_threshold; 195extern uint64_t zfs_deadman_synctime_ms; 196extern int metaslab_preload_limit; 197extern boolean_t zfs_compressed_arc_enabled; 198extern boolean_t zfs_abd_scatter_enabled; 199extern boolean_t zfs_force_some_double_word_sm_entries; 200 201static ztest_shared_opts_t *ztest_shared_opts; 202static ztest_shared_opts_t ztest_opts; 203 204typedef struct ztest_shared_ds { 205 uint64_t zd_seq; 206} ztest_shared_ds_t; 207 208static ztest_shared_ds_t *ztest_shared_ds; 209#define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d]) 210 211#define BT_MAGIC 0x123456789abcdefULL 212#define MAXFAULTS() \ 213 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1) 214 215enum ztest_io_type { 216 ZTEST_IO_WRITE_TAG, 217 ZTEST_IO_WRITE_PATTERN, 218 ZTEST_IO_WRITE_ZEROES, 219 ZTEST_IO_TRUNCATE, 220 ZTEST_IO_SETATTR, 221 ZTEST_IO_REWRITE, 222 ZTEST_IO_TYPES 223}; 224 225typedef struct ztest_block_tag { 226 uint64_t bt_magic; 227 uint64_t bt_objset; 228 uint64_t bt_object; 229 uint64_t bt_offset; 230 uint64_t bt_gen; 231 uint64_t bt_txg; 232 uint64_t bt_crtxg; 233} ztest_block_tag_t; 234 235typedef struct bufwad { 236 uint64_t bw_index; 237 uint64_t bw_txg; 238 uint64_t bw_data; 239} bufwad_t; 240 241/* 242 * XXX -- fix zfs range locks to be generic so we can use them here. 243 */ 244typedef enum { 245 RL_READER, 246 RL_WRITER, 247 RL_APPEND 248} rl_type_t; 249 250typedef struct rll { 251 void *rll_writer; 252 int rll_readers; 253 kmutex_t rll_lock; 254 kcondvar_t rll_cv; 255} rll_t; 256 257typedef struct rl { 258 uint64_t rl_object; 259 uint64_t rl_offset; 260 uint64_t rl_size; 261 rll_t *rl_lock; 262} rl_t; 263 264#define ZTEST_RANGE_LOCKS 64 265#define ZTEST_OBJECT_LOCKS 64 266 267/* 268 * Object descriptor. Used as a template for object lookup/create/remove. 269 */ 270typedef struct ztest_od { 271 uint64_t od_dir; 272 uint64_t od_object; 273 dmu_object_type_t od_type; 274 dmu_object_type_t od_crtype; 275 uint64_t od_blocksize; 276 uint64_t od_crblocksize; 277 uint64_t od_gen; 278 uint64_t od_crgen; 279 char od_name[ZFS_MAX_DATASET_NAME_LEN]; 280} ztest_od_t; 281 282/* 283 * Per-dataset state. 284 */ 285typedef struct ztest_ds { 286 ztest_shared_ds_t *zd_shared; 287 objset_t *zd_os; 288 krwlock_t zd_zilog_lock; 289 zilog_t *zd_zilog; 290 ztest_od_t *zd_od; /* debugging aid */ 291 char zd_name[ZFS_MAX_DATASET_NAME_LEN]; 292 kmutex_t zd_dirobj_lock; 293 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS]; 294 rll_t zd_range_lock[ZTEST_RANGE_LOCKS]; 295} ztest_ds_t; 296 297/* 298 * Per-iteration state. 299 */ 300typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id); 301 302typedef struct ztest_info { 303 ztest_func_t *zi_func; /* test function */ 304 uint64_t zi_iters; /* iterations per execution */ 305 uint64_t *zi_interval; /* execute every <interval> seconds */ 306} ztest_info_t; 307 308typedef struct ztest_shared_callstate { 309 uint64_t zc_count; /* per-pass count */ 310 uint64_t zc_time; /* per-pass time */ 311 uint64_t zc_next; /* next time to call this function */ 312} ztest_shared_callstate_t; 313 314static ztest_shared_callstate_t *ztest_shared_callstate; 315#define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c]) 316 317/* 318 * Note: these aren't static because we want dladdr() to work. 319 */ 320ztest_func_t ztest_dmu_read_write; 321ztest_func_t ztest_dmu_write_parallel; 322ztest_func_t ztest_dmu_object_alloc_free; 323ztest_func_t ztest_dmu_commit_callbacks; 324ztest_func_t ztest_zap; 325ztest_func_t ztest_zap_parallel; 326ztest_func_t ztest_zil_commit; 327ztest_func_t ztest_zil_remount; 328ztest_func_t ztest_dmu_read_write_zcopy; 329ztest_func_t ztest_dmu_objset_create_destroy; 330ztest_func_t ztest_dmu_prealloc; 331ztest_func_t ztest_fzap; 332ztest_func_t ztest_dmu_snapshot_create_destroy; 333ztest_func_t ztest_dsl_prop_get_set; 334ztest_func_t ztest_spa_prop_get_set; 335ztest_func_t ztest_spa_create_destroy; 336ztest_func_t ztest_fault_inject; 337ztest_func_t ztest_ddt_repair; 338ztest_func_t ztest_dmu_snapshot_hold; 339ztest_func_t ztest_spa_rename; 340ztest_func_t ztest_scrub; 341ztest_func_t ztest_dsl_dataset_promote_busy; 342ztest_func_t ztest_vdev_attach_detach; 343ztest_func_t ztest_vdev_LUN_growth; 344ztest_func_t ztest_vdev_add_remove; 345ztest_func_t ztest_vdev_aux_add_remove; 346ztest_func_t ztest_split_pool; 347ztest_func_t ztest_reguid; 348ztest_func_t ztest_spa_upgrade; 349ztest_func_t ztest_device_removal; 350ztest_func_t ztest_remap_blocks; 351ztest_func_t ztest_spa_checkpoint_create_discard; 352ztest_func_t ztest_initialize; 353 354uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */ 355uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */ 356uint64_t zopt_often = 1ULL * NANOSEC; /* every second */ 357uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */ 358uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */ 359 360ztest_info_t ztest_info[] = { 361 { ztest_dmu_read_write, 1, &zopt_always }, 362 { ztest_dmu_write_parallel, 10, &zopt_always }, 363 { ztest_dmu_object_alloc_free, 1, &zopt_always }, 364 { ztest_dmu_commit_callbacks, 1, &zopt_always }, 365 { ztest_zap, 30, &zopt_always }, 366 { ztest_zap_parallel, 100, &zopt_always }, 367 { ztest_split_pool, 1, &zopt_always }, 368 { ztest_zil_commit, 1, &zopt_incessant }, 369 { ztest_zil_remount, 1, &zopt_sometimes }, 370 { ztest_dmu_read_write_zcopy, 1, &zopt_often }, 371 { ztest_dmu_objset_create_destroy, 1, &zopt_often }, 372 { ztest_dsl_prop_get_set, 1, &zopt_often }, 373 { ztest_spa_prop_get_set, 1, &zopt_sometimes }, 374#if 0 375 { ztest_dmu_prealloc, 1, &zopt_sometimes }, 376#endif 377 { ztest_fzap, 1, &zopt_sometimes }, 378 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes }, 379 { ztest_spa_create_destroy, 1, &zopt_sometimes }, 380 { ztest_fault_inject, 1, &zopt_incessant }, 381 { ztest_ddt_repair, 1, &zopt_sometimes }, 382 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes }, 383 { ztest_reguid, 1, &zopt_rarely }, 384 { ztest_spa_rename, 1, &zopt_rarely }, 385 { ztest_scrub, 1, &zopt_often }, 386 { ztest_spa_upgrade, 1, &zopt_rarely }, 387 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely }, 388 { ztest_vdev_attach_detach, 1, &zopt_incessant }, 389 { ztest_vdev_LUN_growth, 1, &zopt_rarely }, 390 { ztest_vdev_add_remove, 1, 391 &ztest_opts.zo_vdevtime }, 392 { ztest_vdev_aux_add_remove, 1, 393 &ztest_opts.zo_vdevtime }, 394 { ztest_device_removal, 1, &zopt_sometimes }, 395 { ztest_remap_blocks, 1, &zopt_sometimes }, 396 { ztest_spa_checkpoint_create_discard, 1, &zopt_rarely }, 397 { ztest_initialize, 1, &zopt_sometimes } 398}; 399 400#define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) 401 402/* 403 * The following struct is used to hold a list of uncalled commit callbacks. 404 * The callbacks are ordered by txg number. 405 */ 406typedef struct ztest_cb_list { 407 kmutex_t zcl_callbacks_lock; 408 list_t zcl_callbacks; 409} ztest_cb_list_t; 410 411/* 412 * Stuff we need to share writably between parent and child. 413 */ 414typedef struct ztest_shared { 415 boolean_t zs_do_init; 416 hrtime_t zs_proc_start; 417 hrtime_t zs_proc_stop; 418 hrtime_t zs_thread_start; 419 hrtime_t zs_thread_stop; 420 hrtime_t zs_thread_kill; 421 uint64_t zs_enospc_count; 422 uint64_t zs_vdev_next_leaf; 423 uint64_t zs_vdev_aux; 424 uint64_t zs_alloc; 425 uint64_t zs_space; 426 uint64_t zs_splits; 427 uint64_t zs_mirrors; 428 uint64_t zs_metaslab_sz; 429 uint64_t zs_metaslab_df_alloc_threshold; 430 uint64_t zs_guid; 431} ztest_shared_t; 432 433#define ID_PARALLEL -1ULL 434 435static char ztest_dev_template[] = "%s/%s.%llua"; 436static char ztest_aux_template[] = "%s/%s.%s.%llu"; 437ztest_shared_t *ztest_shared; 438 439static spa_t *ztest_spa = NULL; 440static ztest_ds_t *ztest_ds; 441 442static kmutex_t ztest_vdev_lock; 443static kmutex_t ztest_checkpoint_lock; 444static boolean_t ztest_device_removal_active = B_FALSE; 445 446/* 447 * The ztest_name_lock protects the pool and dataset namespace used by 448 * the individual tests. To modify the namespace, consumers must grab 449 * this lock as writer. Grabbing the lock as reader will ensure that the 450 * namespace does not change while the lock is held. 451 */ 452static krwlock_t ztest_name_lock; 453 454static boolean_t ztest_dump_core = B_TRUE; 455static boolean_t ztest_exiting; 456 457/* Global commit callback list */ 458static ztest_cb_list_t zcl; 459 460enum ztest_object { 461 ZTEST_META_DNODE = 0, 462 ZTEST_DIROBJ, 463 ZTEST_OBJECTS 464}; 465 466static void usage(boolean_t) __NORETURN; 467 468/* 469 * These libumem hooks provide a reasonable set of defaults for the allocator's 470 * debugging facilities. 471 */ 472const char * 473_umem_debug_init() 474{ 475 return ("default,verbose"); /* $UMEM_DEBUG setting */ 476} 477 478const char * 479_umem_logging_init(void) 480{ 481 return ("fail,contents"); /* $UMEM_LOGGING setting */ 482} 483 484#define FATAL_MSG_SZ 1024 485 486char *fatal_msg; 487 488static void 489fatal(int do_perror, char *message, ...) 490{ 491 va_list args; 492 int save_errno = errno; 493 char buf[FATAL_MSG_SZ]; 494 495 (void) fflush(stdout); 496 497 va_start(args, message); 498 (void) sprintf(buf, "ztest: "); 499 /* LINTED */ 500 (void) vsprintf(buf + strlen(buf), message, args); 501 va_end(args); 502 if (do_perror) { 503 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), 504 ": %s", strerror(save_errno)); 505 } 506 (void) fprintf(stderr, "%s\n", buf); 507 fatal_msg = buf; /* to ease debugging */ 508 if (ztest_dump_core) 509 abort(); 510 exit(3); 511} 512 513static int 514str2shift(const char *buf) 515{ 516 const char *ends = "BKMGTPEZ"; 517 int i; 518 519 if (buf[0] == '\0') 520 return (0); 521 for (i = 0; i < strlen(ends); i++) { 522 if (toupper(buf[0]) == ends[i]) 523 break; 524 } 525 if (i == strlen(ends)) { 526 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", 527 buf); 528 usage(B_FALSE); 529 } 530 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { 531 return (10*i); 532 } 533 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); 534 usage(B_FALSE); 535 /* NOTREACHED */ 536} 537 538static uint64_t 539nicenumtoull(const char *buf) 540{ 541 char *end; 542 uint64_t val; 543 544 val = strtoull(buf, &end, 0); 545 if (end == buf) { 546 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); 547 usage(B_FALSE); 548 } else if (end[0] == '.') { 549 double fval = strtod(buf, &end); 550 fval *= pow(2, str2shift(end)); 551 if (fval > UINT64_MAX) { 552 (void) fprintf(stderr, "ztest: value too large: %s\n", 553 buf); 554 usage(B_FALSE); 555 } 556 val = (uint64_t)fval; 557 } else { 558 int shift = str2shift(end); 559 if (shift >= 64 || (val << shift) >> shift != val) { 560 (void) fprintf(stderr, "ztest: value too large: %s\n", 561 buf); 562 usage(B_FALSE); 563 } 564 val <<= shift; 565 } 566 return (val); 567} 568 569static void 570usage(boolean_t requested) 571{ 572 const ztest_shared_opts_t *zo = &ztest_opts_defaults; 573 574 char nice_vdev_size[NN_NUMBUF_SZ]; 575 char nice_force_ganging[NN_NUMBUF_SZ]; 576 FILE *fp = requested ? stdout : stderr; 577 578 nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size)); 579 nicenum(zo->zo_metaslab_force_ganging, nice_force_ganging, 580 sizeof (nice_force_ganging)); 581 582 (void) fprintf(fp, "Usage: %s\n" 583 "\t[-v vdevs (default: %llu)]\n" 584 "\t[-s size_of_each_vdev (default: %s)]\n" 585 "\t[-a alignment_shift (default: %d)] use 0 for random\n" 586 "\t[-m mirror_copies (default: %d)]\n" 587 "\t[-r raidz_disks (default: %d)]\n" 588 "\t[-R raidz_parity (default: %d)]\n" 589 "\t[-d datasets (default: %d)]\n" 590 "\t[-t threads (default: %d)]\n" 591 "\t[-g gang_block_threshold (default: %s)]\n" 592 "\t[-i init_count (default: %d)] initialize pool i times\n" 593 "\t[-k kill_percentage (default: %llu%%)]\n" 594 "\t[-p pool_name (default: %s)]\n" 595 "\t[-f dir (default: %s)] file directory for vdev files\n" 596 "\t[-V] verbose (use multiple times for ever more blather)\n" 597 "\t[-E] use existing pool instead of creating new one\n" 598 "\t[-T time (default: %llu sec)] total run time\n" 599 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n" 600 "\t[-P passtime (default: %llu sec)] time per pass\n" 601 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n" 602 "\t[-o variable=value] ... set global variable to an unsigned\n" 603 "\t 32-bit integer value\n" 604 "\t[-h] (print help)\n" 605 "", 606 zo->zo_pool, 607 (u_longlong_t)zo->zo_vdevs, /* -v */ 608 nice_vdev_size, /* -s */ 609 zo->zo_ashift, /* -a */ 610 zo->zo_mirrors, /* -m */ 611 zo->zo_raidz, /* -r */ 612 zo->zo_raidz_parity, /* -R */ 613 zo->zo_datasets, /* -d */ 614 zo->zo_threads, /* -t */ 615 nice_force_ganging, /* -g */ 616 zo->zo_init, /* -i */ 617 (u_longlong_t)zo->zo_killrate, /* -k */ 618 zo->zo_pool, /* -p */ 619 zo->zo_dir, /* -f */ 620 (u_longlong_t)zo->zo_time, /* -T */ 621 (u_longlong_t)zo->zo_maxloops, /* -F */ 622 (u_longlong_t)zo->zo_passtime); 623 exit(requested ? 0 : 1); 624} 625 626static void 627process_options(int argc, char **argv) 628{ 629 char *path; 630 ztest_shared_opts_t *zo = &ztest_opts; 631 632 int opt; 633 uint64_t value; 634 char altdir[MAXNAMELEN] = { 0 }; 635 636 bcopy(&ztest_opts_defaults, zo, sizeof (*zo)); 637 638 while ((opt = getopt(argc, argv, 639 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF) { 640 value = 0; 641 switch (opt) { 642 case 'v': 643 case 's': 644 case 'a': 645 case 'm': 646 case 'r': 647 case 'R': 648 case 'd': 649 case 't': 650 case 'g': 651 case 'i': 652 case 'k': 653 case 'T': 654 case 'P': 655 case 'F': 656 value = nicenumtoull(optarg); 657 } 658 switch (opt) { 659 case 'v': 660 zo->zo_vdevs = value; 661 break; 662 case 's': 663 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value); 664 break; 665 case 'a': 666 zo->zo_ashift = value; 667 break; 668 case 'm': 669 zo->zo_mirrors = value; 670 break; 671 case 'r': 672 zo->zo_raidz = MAX(1, value); 673 break; 674 case 'R': 675 zo->zo_raidz_parity = MIN(MAX(value, 1), 3); 676 break; 677 case 'd': 678 zo->zo_datasets = MAX(1, value); 679 break; 680 case 't': 681 zo->zo_threads = MAX(1, value); 682 break; 683 case 'g': 684 zo->zo_metaslab_force_ganging = 685 MAX(SPA_MINBLOCKSIZE << 1, value); 686 break; 687 case 'i': 688 zo->zo_init = value; 689 break; 690 case 'k': 691 zo->zo_killrate = value; 692 break; 693 case 'p': 694 (void) strlcpy(zo->zo_pool, optarg, 695 sizeof (zo->zo_pool)); 696 break; 697 case 'f': 698 path = realpath(optarg, NULL); 699 if (path == NULL) { 700 (void) fprintf(stderr, "error: %s: %s\n", 701 optarg, strerror(errno)); 702 usage(B_FALSE); 703 } else { 704 (void) strlcpy(zo->zo_dir, path, 705 sizeof (zo->zo_dir)); 706 } 707 break; 708 case 'V': 709 zo->zo_verbose++; 710 break; 711 case 'E': 712 zo->zo_init = 0; 713 break; 714 case 'T': 715 zo->zo_time = value; 716 break; 717 case 'P': 718 zo->zo_passtime = MAX(1, value); 719 break; 720 case 'F': 721 zo->zo_maxloops = MAX(1, value); 722 break; 723 case 'B': 724 (void) strlcpy(altdir, optarg, sizeof (altdir)); 725 break; 726 case 'o': 727 if (set_global_var(optarg) != 0) 728 usage(B_FALSE); 729 break; 730 case 'h': 731 usage(B_TRUE); 732 break; 733 case '?': 734 default: 735 usage(B_FALSE); 736 break; 737 } 738 } 739 740 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1); 741 742 zo->zo_vdevtime = 743 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs : 744 UINT64_MAX >> 2); 745 746 if (strlen(altdir) > 0) { 747 char *cmd; 748 char *realaltdir; 749 char *bin; 750 char *ztest; 751 char *isa; 752 int isalen; 753 754 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 755 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 756 757 VERIFY(NULL != realpath(getexecname(), cmd)); 758 if (0 != access(altdir, F_OK)) { 759 ztest_dump_core = B_FALSE; 760 fatal(B_TRUE, "invalid alternate ztest path: %s", 761 altdir); 762 } 763 VERIFY(NULL != realpath(altdir, realaltdir)); 764 765 /* 766 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest". 767 * We want to extract <isa> to determine if we should use 768 * 32 or 64 bit binaries. 769 */ 770 bin = strstr(cmd, "/usr/bin/"); 771 ztest = strstr(bin, "/ztest"); 772 isa = bin + 9; 773 isalen = ztest - isa; 774 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest), 775 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa); 776 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath), 777 "%s/usr/lib/%.*s", realaltdir, isalen, isa); 778 779 if (0 != access(zo->zo_alt_ztest, X_OK)) { 780 ztest_dump_core = B_FALSE; 781 fatal(B_TRUE, "invalid alternate ztest: %s", 782 zo->zo_alt_ztest); 783 } else if (0 != access(zo->zo_alt_libpath, X_OK)) { 784 ztest_dump_core = B_FALSE; 785 fatal(B_TRUE, "invalid alternate lib directory %s", 786 zo->zo_alt_libpath); 787 } 788 789 umem_free(cmd, MAXPATHLEN); 790 umem_free(realaltdir, MAXPATHLEN); 791 } 792} 793 794static void 795ztest_kill(ztest_shared_t *zs) 796{ 797 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa)); 798 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa)); 799 800 /* 801 * Before we kill off ztest, make sure that the config is updated. 802 * See comment above spa_write_cachefile(). 803 */ 804 mutex_enter(&spa_namespace_lock); 805 spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE); 806 mutex_exit(&spa_namespace_lock); 807 808 zfs_dbgmsg_print(FTAG); 809 (void) kill(getpid(), SIGKILL); 810} 811 812static uint64_t 813ztest_random(uint64_t range) 814{ 815 uint64_t r; 816 817 ASSERT3S(ztest_fd_rand, >=, 0); 818 819 if (range == 0) 820 return (0); 821 822 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r)) 823 fatal(1, "short read from /dev/urandom"); 824 825 return (r % range); 826} 827 828/* ARGSUSED */ 829static void 830ztest_record_enospc(const char *s) 831{ 832 ztest_shared->zs_enospc_count++; 833} 834 835static uint64_t 836ztest_get_ashift(void) 837{ 838 if (ztest_opts.zo_ashift == 0) 839 return (SPA_MINBLOCKSHIFT + ztest_random(5)); 840 return (ztest_opts.zo_ashift); 841} 842 843static nvlist_t * 844make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift) 845{ 846 char pathbuf[MAXPATHLEN]; 847 uint64_t vdev; 848 nvlist_t *file; 849 850 if (ashift == 0) 851 ashift = ztest_get_ashift(); 852 853 if (path == NULL) { 854 path = pathbuf; 855 856 if (aux != NULL) { 857 vdev = ztest_shared->zs_vdev_aux; 858 (void) snprintf(path, sizeof (pathbuf), 859 ztest_aux_template, ztest_opts.zo_dir, 860 pool == NULL ? ztest_opts.zo_pool : pool, 861 aux, vdev); 862 } else { 863 vdev = ztest_shared->zs_vdev_next_leaf++; 864 (void) snprintf(path, sizeof (pathbuf), 865 ztest_dev_template, ztest_opts.zo_dir, 866 pool == NULL ? ztest_opts.zo_pool : pool, vdev); 867 } 868 } 869 870 if (size != 0) { 871 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666); 872 if (fd == -1) 873 fatal(1, "can't open %s", path); 874 if (ftruncate(fd, size) != 0) 875 fatal(1, "can't ftruncate %s", path); 876 (void) close(fd); 877 } 878 879 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 880 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 881 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0); 882 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 883 884 return (file); 885} 886 887static nvlist_t * 888make_vdev_raidz(char *path, char *aux, char *pool, size_t size, 889 uint64_t ashift, int r) 890{ 891 nvlist_t *raidz, **child; 892 int c; 893 894 if (r < 2) 895 return (make_vdev_file(path, aux, pool, size, ashift)); 896 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); 897 898 for (c = 0; c < r; c++) 899 child[c] = make_vdev_file(path, aux, pool, size, ashift); 900 901 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); 902 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, 903 VDEV_TYPE_RAIDZ) == 0); 904 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, 905 ztest_opts.zo_raidz_parity) == 0); 906 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, 907 child, r) == 0); 908 909 for (c = 0; c < r; c++) 910 nvlist_free(child[c]); 911 912 umem_free(child, r * sizeof (nvlist_t *)); 913 914 return (raidz); 915} 916 917static nvlist_t * 918make_vdev_mirror(char *path, char *aux, char *pool, size_t size, 919 uint64_t ashift, int r, int m) 920{ 921 nvlist_t *mirror, **child; 922 int c; 923 924 if (m < 1) 925 return (make_vdev_raidz(path, aux, pool, size, ashift, r)); 926 927 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); 928 929 for (c = 0; c < m; c++) 930 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r); 931 932 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); 933 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, 934 VDEV_TYPE_MIRROR) == 0); 935 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, 936 child, m) == 0); 937 938 for (c = 0; c < m; c++) 939 nvlist_free(child[c]); 940 941 umem_free(child, m * sizeof (nvlist_t *)); 942 943 return (mirror); 944} 945 946static nvlist_t * 947make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift, 948 int log, int r, int m, int t) 949{ 950 nvlist_t *root, **child; 951 int c; 952 953 ASSERT(t > 0); 954 955 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); 956 957 for (c = 0; c < t; c++) { 958 child[c] = make_vdev_mirror(path, aux, pool, size, ashift, 959 r, m); 960 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, 961 log) == 0); 962 } 963 964 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 965 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 966 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN, 967 child, t) == 0); 968 969 for (c = 0; c < t; c++) 970 nvlist_free(child[c]); 971 972 umem_free(child, t * sizeof (nvlist_t *)); 973 974 return (root); 975} 976 977/* 978 * Find a random spa version. Returns back a random spa version in the 979 * range [initial_version, SPA_VERSION_FEATURES]. 980 */ 981static uint64_t 982ztest_random_spa_version(uint64_t initial_version) 983{ 984 uint64_t version = initial_version; 985 986 if (version <= SPA_VERSION_BEFORE_FEATURES) { 987 version = version + 988 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1); 989 } 990 991 if (version > SPA_VERSION_BEFORE_FEATURES) 992 version = SPA_VERSION_FEATURES; 993 994 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 995 return (version); 996} 997 998static int 999ztest_random_blocksize(void) 1000{ 1001 uint64_t block_shift; 1002 /* 1003 * Choose a block size >= the ashift. 1004 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks. 1005 */ 1006 int maxbs = SPA_OLD_MAXBLOCKSHIFT; 1007 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE) 1008 maxbs = 20; 1009 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1); 1010 return (1 << (SPA_MINBLOCKSHIFT + block_shift)); 1011} 1012 1013static int 1014ztest_random_ibshift(void) 1015{ 1016 return (DN_MIN_INDBLKSHIFT + 1017 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1)); 1018} 1019 1020static uint64_t 1021ztest_random_vdev_top(spa_t *spa, boolean_t log_ok) 1022{ 1023 uint64_t top; 1024 vdev_t *rvd = spa->spa_root_vdev; 1025 vdev_t *tvd; 1026 1027 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1028 1029 do { 1030 top = ztest_random(rvd->vdev_children); 1031 tvd = rvd->vdev_child[top]; 1032 } while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) || 1033 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL); 1034 1035 return (top); 1036} 1037 1038static uint64_t 1039ztest_random_dsl_prop(zfs_prop_t prop) 1040{ 1041 uint64_t value; 1042 1043 do { 1044 value = zfs_prop_random_value(prop, ztest_random(-1ULL)); 1045 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF); 1046 1047 return (value); 1048} 1049 1050static int 1051ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, 1052 boolean_t inherit) 1053{ 1054 const char *propname = zfs_prop_to_name(prop); 1055 const char *valname; 1056 char setpoint[MAXPATHLEN]; 1057 uint64_t curval; 1058 int error; 1059 1060 error = dsl_prop_set_int(osname, propname, 1061 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value); 1062 1063 if (error == ENOSPC) { 1064 ztest_record_enospc(FTAG); 1065 return (error); 1066 } 1067 ASSERT0(error); 1068 1069 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint)); 1070 1071 if (ztest_opts.zo_verbose >= 6) { 1072 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0); 1073 (void) printf("%s %s = %s at '%s'\n", 1074 osname, propname, valname, setpoint); 1075 } 1076 1077 return (error); 1078} 1079 1080static int 1081ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value) 1082{ 1083 spa_t *spa = ztest_spa; 1084 nvlist_t *props = NULL; 1085 int error; 1086 1087 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); 1088 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0); 1089 1090 error = spa_prop_set(spa, props); 1091 1092 nvlist_free(props); 1093 1094 if (error == ENOSPC) { 1095 ztest_record_enospc(FTAG); 1096 return (error); 1097 } 1098 ASSERT0(error); 1099 1100 return (error); 1101} 1102 1103static void 1104ztest_rll_init(rll_t *rll) 1105{ 1106 rll->rll_writer = NULL; 1107 rll->rll_readers = 0; 1108 mutex_init(&rll->rll_lock, NULL, USYNC_THREAD, NULL); 1109 cv_init(&rll->rll_cv, NULL, USYNC_THREAD, NULL); 1110} 1111 1112static void 1113ztest_rll_destroy(rll_t *rll) 1114{ 1115 ASSERT(rll->rll_writer == NULL); 1116 ASSERT(rll->rll_readers == 0); 1117 mutex_destroy(&rll->rll_lock); 1118 cv_destroy(&rll->rll_cv); 1119} 1120 1121static void 1122ztest_rll_lock(rll_t *rll, rl_type_t type) 1123{ 1124 mutex_enter(&rll->rll_lock); 1125 1126 if (type == RL_READER) { 1127 while (rll->rll_writer != NULL) 1128 cv_wait(&rll->rll_cv, &rll->rll_lock); 1129 rll->rll_readers++; 1130 } else { 1131 while (rll->rll_writer != NULL || rll->rll_readers) 1132 cv_wait(&rll->rll_cv, &rll->rll_lock); 1133 rll->rll_writer = curthread; 1134 } 1135 1136 mutex_exit(&rll->rll_lock); 1137} 1138 1139static void 1140ztest_rll_unlock(rll_t *rll) 1141{ 1142 mutex_enter(&rll->rll_lock); 1143 1144 if (rll->rll_writer) { 1145 ASSERT(rll->rll_readers == 0); 1146 rll->rll_writer = NULL; 1147 } else { 1148 ASSERT(rll->rll_readers != 0); 1149 ASSERT(rll->rll_writer == NULL); 1150 rll->rll_readers--; 1151 } 1152 1153 if (rll->rll_writer == NULL && rll->rll_readers == 0) 1154 cv_broadcast(&rll->rll_cv); 1155 1156 mutex_exit(&rll->rll_lock); 1157} 1158 1159static void 1160ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type) 1161{ 1162 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; 1163 1164 ztest_rll_lock(rll, type); 1165} 1166 1167static void 1168ztest_object_unlock(ztest_ds_t *zd, uint64_t object) 1169{ 1170 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; 1171 1172 ztest_rll_unlock(rll); 1173} 1174 1175static rl_t * 1176ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset, 1177 uint64_t size, rl_type_t type) 1178{ 1179 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1)); 1180 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)]; 1181 rl_t *rl; 1182 1183 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL); 1184 rl->rl_object = object; 1185 rl->rl_offset = offset; 1186 rl->rl_size = size; 1187 rl->rl_lock = rll; 1188 1189 ztest_rll_lock(rll, type); 1190 1191 return (rl); 1192} 1193 1194static void 1195ztest_range_unlock(rl_t *rl) 1196{ 1197 rll_t *rll = rl->rl_lock; 1198 1199 ztest_rll_unlock(rll); 1200 1201 umem_free(rl, sizeof (*rl)); 1202} 1203 1204static void 1205ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os) 1206{ 1207 zd->zd_os = os; 1208 zd->zd_zilog = dmu_objset_zil(os); 1209 zd->zd_shared = szd; 1210 dmu_objset_name(os, zd->zd_name); 1211 1212 if (zd->zd_shared != NULL) 1213 zd->zd_shared->zd_seq = 0; 1214 1215 rw_init(&zd->zd_zilog_lock, NULL, USYNC_THREAD, NULL); 1216 mutex_init(&zd->zd_dirobj_lock, NULL, USYNC_THREAD, NULL); 1217 1218 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++) 1219 ztest_rll_init(&zd->zd_object_lock[l]); 1220 1221 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++) 1222 ztest_rll_init(&zd->zd_range_lock[l]); 1223} 1224 1225static void 1226ztest_zd_fini(ztest_ds_t *zd) 1227{ 1228 mutex_destroy(&zd->zd_dirobj_lock); 1229 1230 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++) 1231 ztest_rll_destroy(&zd->zd_object_lock[l]); 1232 1233 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++) 1234 ztest_rll_destroy(&zd->zd_range_lock[l]); 1235} 1236 1237#define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT) 1238 1239static uint64_t 1240ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag) 1241{ 1242 uint64_t txg; 1243 int error; 1244 1245 /* 1246 * Attempt to assign tx to some transaction group. 1247 */ 1248 error = dmu_tx_assign(tx, txg_how); 1249 if (error) { 1250 if (error == ERESTART) { 1251 ASSERT(txg_how == TXG_NOWAIT); 1252 dmu_tx_wait(tx); 1253 } else { 1254 ASSERT3U(error, ==, ENOSPC); 1255 ztest_record_enospc(tag); 1256 } 1257 dmu_tx_abort(tx); 1258 return (0); 1259 } 1260 txg = dmu_tx_get_txg(tx); 1261 ASSERT(txg != 0); 1262 return (txg); 1263} 1264 1265static void 1266ztest_pattern_set(void *buf, uint64_t size, uint64_t value) 1267{ 1268 uint64_t *ip = buf; 1269 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); 1270 1271 while (ip < ip_end) 1272 *ip++ = value; 1273} 1274 1275static boolean_t 1276ztest_pattern_match(void *buf, uint64_t size, uint64_t value) 1277{ 1278 uint64_t *ip = buf; 1279 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); 1280 uint64_t diff = 0; 1281 1282 while (ip < ip_end) 1283 diff |= (value - *ip++); 1284 1285 return (diff == 0); 1286} 1287 1288static void 1289ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object, 1290 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) 1291{ 1292 bt->bt_magic = BT_MAGIC; 1293 bt->bt_objset = dmu_objset_id(os); 1294 bt->bt_object = object; 1295 bt->bt_offset = offset; 1296 bt->bt_gen = gen; 1297 bt->bt_txg = txg; 1298 bt->bt_crtxg = crtxg; 1299} 1300 1301static void 1302ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object, 1303 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) 1304{ 1305 ASSERT3U(bt->bt_magic, ==, BT_MAGIC); 1306 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os)); 1307 ASSERT3U(bt->bt_object, ==, object); 1308 ASSERT3U(bt->bt_offset, ==, offset); 1309 ASSERT3U(bt->bt_gen, <=, gen); 1310 ASSERT3U(bt->bt_txg, <=, txg); 1311 ASSERT3U(bt->bt_crtxg, ==, crtxg); 1312} 1313 1314static ztest_block_tag_t * 1315ztest_bt_bonus(dmu_buf_t *db) 1316{ 1317 dmu_object_info_t doi; 1318 ztest_block_tag_t *bt; 1319 1320 dmu_object_info_from_db(db, &doi); 1321 ASSERT3U(doi.doi_bonus_size, <=, db->db_size); 1322 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt)); 1323 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt)); 1324 1325 return (bt); 1326} 1327 1328/* 1329 * ZIL logging ops 1330 */ 1331 1332#define lrz_type lr_mode 1333#define lrz_blocksize lr_uid 1334#define lrz_ibshift lr_gid 1335#define lrz_bonustype lr_rdev 1336#define lrz_bonuslen lr_crtime[1] 1337 1338static void 1339ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr) 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_CREATE, sizeof (*lr) + namesize); 1349 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1350 sizeof (*lr) + namesize - sizeof (lr_t)); 1351 1352 zil_itx_assign(zd->zd_zilog, itx, tx); 1353} 1354 1355static void 1356ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object) 1357{ 1358 char *name = (void *)(lr + 1); /* name follows lr */ 1359 size_t namesize = strlen(name) + 1; 1360 itx_t *itx; 1361 1362 if (zil_replaying(zd->zd_zilog, tx)) 1363 return; 1364 1365 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize); 1366 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1367 sizeof (*lr) + namesize - sizeof (lr_t)); 1368 1369 itx->itx_oid = object; 1370 zil_itx_assign(zd->zd_zilog, itx, tx); 1371} 1372 1373static void 1374ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr) 1375{ 1376 itx_t *itx; 1377 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES); 1378 1379 if (zil_replaying(zd->zd_zilog, tx)) 1380 return; 1381 1382 if (lr->lr_length > zil_max_log_data(zd->zd_zilog)) 1383 write_state = WR_INDIRECT; 1384 1385 itx = zil_itx_create(TX_WRITE, 1386 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0)); 1387 1388 if (write_state == WR_COPIED && 1389 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length, 1390 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) { 1391 zil_itx_destroy(itx); 1392 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1393 write_state = WR_NEED_COPY; 1394 } 1395 itx->itx_private = zd; 1396 itx->itx_wr_state = write_state; 1397 itx->itx_sync = (ztest_random(8) == 0); 1398 1399 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1400 sizeof (*lr) - sizeof (lr_t)); 1401 1402 zil_itx_assign(zd->zd_zilog, itx, tx); 1403} 1404 1405static void 1406ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_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_TRUNCATE, 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 1421static void 1422ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr) 1423{ 1424 itx_t *itx; 1425 1426 if (zil_replaying(zd->zd_zilog, tx)) 1427 return; 1428 1429 itx = zil_itx_create(TX_SETATTR, sizeof (*lr)); 1430 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1431 sizeof (*lr) - sizeof (lr_t)); 1432 1433 itx->itx_sync = B_FALSE; 1434 zil_itx_assign(zd->zd_zilog, itx, tx); 1435} 1436 1437/* 1438 * ZIL replay ops 1439 */ 1440static int 1441ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap) 1442{ 1443 ztest_ds_t *zd = arg1; 1444 lr_create_t *lr = arg2; 1445 char *name = (void *)(lr + 1); /* name follows lr */ 1446 objset_t *os = zd->zd_os; 1447 ztest_block_tag_t *bbt; 1448 dmu_buf_t *db; 1449 dmu_tx_t *tx; 1450 uint64_t txg; 1451 int error = 0; 1452 1453 if (byteswap) 1454 byteswap_uint64_array(lr, sizeof (*lr)); 1455 1456 ASSERT(lr->lr_doid == ZTEST_DIROBJ); 1457 ASSERT(name[0] != '\0'); 1458 1459 tx = dmu_tx_create(os); 1460 1461 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name); 1462 1463 if (lr->lrz_type == DMU_OT_ZAP_OTHER) { 1464 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); 1465 } else { 1466 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1467 } 1468 1469 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1470 if (txg == 0) 1471 return (ENOSPC); 1472 1473 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid); 1474 1475 if (lr->lrz_type == DMU_OT_ZAP_OTHER) { 1476 if (lr->lr_foid == 0) { 1477 lr->lr_foid = zap_create(os, 1478 lr->lrz_type, lr->lrz_bonustype, 1479 lr->lrz_bonuslen, tx); 1480 } else { 1481 error = zap_create_claim(os, lr->lr_foid, 1482 lr->lrz_type, lr->lrz_bonustype, 1483 lr->lrz_bonuslen, tx); 1484 } 1485 } else { 1486 if (lr->lr_foid == 0) { 1487 lr->lr_foid = dmu_object_alloc(os, 1488 lr->lrz_type, 0, lr->lrz_bonustype, 1489 lr->lrz_bonuslen, tx); 1490 } else { 1491 error = dmu_object_claim(os, lr->lr_foid, 1492 lr->lrz_type, 0, lr->lrz_bonustype, 1493 lr->lrz_bonuslen, tx); 1494 } 1495 } 1496 1497 if (error) { 1498 ASSERT3U(error, ==, EEXIST); 1499 ASSERT(zd->zd_zilog->zl_replay); 1500 dmu_tx_commit(tx); 1501 return (error); 1502 } 1503 1504 ASSERT(lr->lr_foid != 0); 1505 1506 if (lr->lrz_type != DMU_OT_ZAP_OTHER) 1507 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid, 1508 lr->lrz_blocksize, lr->lrz_ibshift, tx)); 1509 1510 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1511 bbt = ztest_bt_bonus(db); 1512 dmu_buf_will_dirty(db, tx); 1513 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg); 1514 dmu_buf_rele(db, FTAG); 1515 1516 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1, 1517 &lr->lr_foid, tx)); 1518 1519 (void) ztest_log_create(zd, tx, lr); 1520 1521 dmu_tx_commit(tx); 1522 1523 return (0); 1524} 1525 1526static int 1527ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap) 1528{ 1529 ztest_ds_t *zd = arg1; 1530 lr_remove_t *lr = arg2; 1531 char *name = (void *)(lr + 1); /* name follows lr */ 1532 objset_t *os = zd->zd_os; 1533 dmu_object_info_t doi; 1534 dmu_tx_t *tx; 1535 uint64_t object, txg; 1536 1537 if (byteswap) 1538 byteswap_uint64_array(lr, sizeof (*lr)); 1539 1540 ASSERT(lr->lr_doid == ZTEST_DIROBJ); 1541 ASSERT(name[0] != '\0'); 1542 1543 VERIFY3U(0, ==, 1544 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object)); 1545 ASSERT(object != 0); 1546 1547 ztest_object_lock(zd, object, RL_WRITER); 1548 1549 VERIFY3U(0, ==, dmu_object_info(os, object, &doi)); 1550 1551 tx = dmu_tx_create(os); 1552 1553 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name); 1554 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 1555 1556 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1557 if (txg == 0) { 1558 ztest_object_unlock(zd, object); 1559 return (ENOSPC); 1560 } 1561 1562 if (doi.doi_type == DMU_OT_ZAP_OTHER) { 1563 VERIFY3U(0, ==, zap_destroy(os, object, tx)); 1564 } else { 1565 VERIFY3U(0, ==, dmu_object_free(os, object, tx)); 1566 } 1567 1568 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx)); 1569 1570 (void) ztest_log_remove(zd, tx, lr, object); 1571 1572 dmu_tx_commit(tx); 1573 1574 ztest_object_unlock(zd, object); 1575 1576 return (0); 1577} 1578 1579static int 1580ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap) 1581{ 1582 ztest_ds_t *zd = arg1; 1583 lr_write_t *lr = arg2; 1584 objset_t *os = zd->zd_os; 1585 void *data = lr + 1; /* data follows lr */ 1586 uint64_t offset, length; 1587 ztest_block_tag_t *bt = data; 1588 ztest_block_tag_t *bbt; 1589 uint64_t gen, txg, lrtxg, crtxg; 1590 dmu_object_info_t doi; 1591 dmu_tx_t *tx; 1592 dmu_buf_t *db; 1593 arc_buf_t *abuf = NULL; 1594 rl_t *rl; 1595 1596 if (byteswap) 1597 byteswap_uint64_array(lr, sizeof (*lr)); 1598 1599 offset = lr->lr_offset; 1600 length = lr->lr_length; 1601 1602 /* If it's a dmu_sync() block, write the whole block */ 1603 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 1604 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 1605 if (length < blocksize) { 1606 offset -= offset % blocksize; 1607 length = blocksize; 1608 } 1609 } 1610 1611 if (bt->bt_magic == BSWAP_64(BT_MAGIC)) 1612 byteswap_uint64_array(bt, sizeof (*bt)); 1613 1614 if (bt->bt_magic != BT_MAGIC) 1615 bt = NULL; 1616 1617 ztest_object_lock(zd, lr->lr_foid, RL_READER); 1618 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER); 1619 1620 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1621 1622 dmu_object_info_from_db(db, &doi); 1623 1624 bbt = ztest_bt_bonus(db); 1625 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1626 gen = bbt->bt_gen; 1627 crtxg = bbt->bt_crtxg; 1628 lrtxg = lr->lr_common.lrc_txg; 1629 1630 tx = dmu_tx_create(os); 1631 1632 dmu_tx_hold_write(tx, lr->lr_foid, offset, length); 1633 1634 if (ztest_random(8) == 0 && length == doi.doi_data_block_size && 1635 P2PHASE(offset, length) == 0) 1636 abuf = dmu_request_arcbuf(db, length); 1637 1638 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1639 if (txg == 0) { 1640 if (abuf != NULL) 1641 dmu_return_arcbuf(abuf); 1642 dmu_buf_rele(db, FTAG); 1643 ztest_range_unlock(rl); 1644 ztest_object_unlock(zd, lr->lr_foid); 1645 return (ENOSPC); 1646 } 1647 1648 if (bt != NULL) { 1649 /* 1650 * Usually, verify the old data before writing new data -- 1651 * but not always, because we also want to verify correct 1652 * behavior when the data was not recently read into cache. 1653 */ 1654 ASSERT(offset % doi.doi_data_block_size == 0); 1655 if (ztest_random(4) != 0) { 1656 int prefetch = ztest_random(2) ? 1657 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH; 1658 ztest_block_tag_t rbt; 1659 1660 VERIFY(dmu_read(os, lr->lr_foid, offset, 1661 sizeof (rbt), &rbt, prefetch) == 0); 1662 if (rbt.bt_magic == BT_MAGIC) { 1663 ztest_bt_verify(&rbt, os, lr->lr_foid, 1664 offset, gen, txg, crtxg); 1665 } 1666 } 1667 1668 /* 1669 * Writes can appear to be newer than the bonus buffer because 1670 * the ztest_get_data() callback does a dmu_read() of the 1671 * open-context data, which may be different than the data 1672 * as it was when the write was generated. 1673 */ 1674 if (zd->zd_zilog->zl_replay) { 1675 ztest_bt_verify(bt, os, lr->lr_foid, offset, 1676 MAX(gen, bt->bt_gen), MAX(txg, lrtxg), 1677 bt->bt_crtxg); 1678 } 1679 1680 /* 1681 * Set the bt's gen/txg to the bonus buffer's gen/txg 1682 * so that all of the usual ASSERTs will work. 1683 */ 1684 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg); 1685 } 1686 1687 if (abuf == NULL) { 1688 dmu_write(os, lr->lr_foid, offset, length, data, tx); 1689 } else { 1690 bcopy(data, abuf->b_data, length); 1691 dmu_assign_arcbuf(db, offset, abuf, tx); 1692 } 1693 1694 (void) ztest_log_write(zd, tx, lr); 1695 1696 dmu_buf_rele(db, FTAG); 1697 1698 dmu_tx_commit(tx); 1699 1700 ztest_range_unlock(rl); 1701 ztest_object_unlock(zd, lr->lr_foid); 1702 1703 return (0); 1704} 1705 1706static int 1707ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap) 1708{ 1709 ztest_ds_t *zd = arg1; 1710 lr_truncate_t *lr = arg2; 1711 objset_t *os = zd->zd_os; 1712 dmu_tx_t *tx; 1713 uint64_t txg; 1714 rl_t *rl; 1715 1716 if (byteswap) 1717 byteswap_uint64_array(lr, sizeof (*lr)); 1718 1719 ztest_object_lock(zd, lr->lr_foid, RL_READER); 1720 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length, 1721 RL_WRITER); 1722 1723 tx = dmu_tx_create(os); 1724 1725 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length); 1726 1727 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1728 if (txg == 0) { 1729 ztest_range_unlock(rl); 1730 ztest_object_unlock(zd, lr->lr_foid); 1731 return (ENOSPC); 1732 } 1733 1734 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset, 1735 lr->lr_length, tx) == 0); 1736 1737 (void) ztest_log_truncate(zd, tx, lr); 1738 1739 dmu_tx_commit(tx); 1740 1741 ztest_range_unlock(rl); 1742 ztest_object_unlock(zd, lr->lr_foid); 1743 1744 return (0); 1745} 1746 1747static int 1748ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap) 1749{ 1750 ztest_ds_t *zd = arg1; 1751 lr_setattr_t *lr = arg2; 1752 objset_t *os = zd->zd_os; 1753 dmu_tx_t *tx; 1754 dmu_buf_t *db; 1755 ztest_block_tag_t *bbt; 1756 uint64_t txg, lrtxg, crtxg; 1757 1758 if (byteswap) 1759 byteswap_uint64_array(lr, sizeof (*lr)); 1760 1761 ztest_object_lock(zd, lr->lr_foid, RL_WRITER); 1762 1763 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1764 1765 tx = dmu_tx_create(os); 1766 dmu_tx_hold_bonus(tx, lr->lr_foid); 1767 1768 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1769 if (txg == 0) { 1770 dmu_buf_rele(db, FTAG); 1771 ztest_object_unlock(zd, lr->lr_foid); 1772 return (ENOSPC); 1773 } 1774 1775 bbt = ztest_bt_bonus(db); 1776 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1777 crtxg = bbt->bt_crtxg; 1778 lrtxg = lr->lr_common.lrc_txg; 1779 1780 if (zd->zd_zilog->zl_replay) { 1781 ASSERT(lr->lr_size != 0); 1782 ASSERT(lr->lr_mode != 0); 1783 ASSERT(lrtxg != 0); 1784 } else { 1785 /* 1786 * Randomly change the size and increment the generation. 1787 */ 1788 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) * 1789 sizeof (*bbt); 1790 lr->lr_mode = bbt->bt_gen + 1; 1791 ASSERT(lrtxg == 0); 1792 } 1793 1794 /* 1795 * Verify that the current bonus buffer is not newer than our txg. 1796 */ 1797 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, 1798 MAX(txg, lrtxg), crtxg); 1799 1800 dmu_buf_will_dirty(db, tx); 1801 1802 ASSERT3U(lr->lr_size, >=, sizeof (*bbt)); 1803 ASSERT3U(lr->lr_size, <=, db->db_size); 1804 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx)); 1805 bbt = ztest_bt_bonus(db); 1806 1807 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg); 1808 1809 dmu_buf_rele(db, FTAG); 1810 1811 (void) ztest_log_setattr(zd, tx, lr); 1812 1813 dmu_tx_commit(tx); 1814 1815 ztest_object_unlock(zd, lr->lr_foid); 1816 1817 return (0); 1818} 1819 1820zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { 1821 NULL, /* 0 no such transaction type */ 1822 ztest_replay_create, /* TX_CREATE */ 1823 NULL, /* TX_MKDIR */ 1824 NULL, /* TX_MKXATTR */ 1825 NULL, /* TX_SYMLINK */ 1826 ztest_replay_remove, /* TX_REMOVE */ 1827 NULL, /* TX_RMDIR */ 1828 NULL, /* TX_LINK */ 1829 NULL, /* TX_RENAME */ 1830 ztest_replay_write, /* TX_WRITE */ 1831 ztest_replay_truncate, /* TX_TRUNCATE */ 1832 ztest_replay_setattr, /* TX_SETATTR */ 1833 NULL, /* TX_ACL */ 1834 NULL, /* TX_CREATE_ACL */ 1835 NULL, /* TX_CREATE_ATTR */ 1836 NULL, /* TX_CREATE_ACL_ATTR */ 1837 NULL, /* TX_MKDIR_ACL */ 1838 NULL, /* TX_MKDIR_ATTR */ 1839 NULL, /* TX_MKDIR_ACL_ATTR */ 1840 NULL, /* TX_WRITE2 */ 1841}; 1842 1843/* 1844 * ZIL get_data callbacks 1845 */ 1846 1847static void 1848ztest_get_done(zgd_t *zgd, int error) 1849{ 1850 ztest_ds_t *zd = zgd->zgd_private; 1851 uint64_t object = zgd->zgd_rl->rl_object; 1852 1853 if (zgd->zgd_db) 1854 dmu_buf_rele(zgd->zgd_db, zgd); 1855 1856 ztest_range_unlock(zgd->zgd_rl); 1857 ztest_object_unlock(zd, object); 1858 1859 if (error == 0 && zgd->zgd_bp) 1860 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); 1861 1862 umem_free(zgd, sizeof (*zgd)); 1863} 1864 1865static int 1866ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, 1867 zio_t *zio) 1868{ 1869 ztest_ds_t *zd = arg; 1870 objset_t *os = zd->zd_os; 1871 uint64_t object = lr->lr_foid; 1872 uint64_t offset = lr->lr_offset; 1873 uint64_t size = lr->lr_length; 1874 uint64_t txg = lr->lr_common.lrc_txg; 1875 uint64_t crtxg; 1876 dmu_object_info_t doi; 1877 dmu_buf_t *db; 1878 zgd_t *zgd; 1879 int error; 1880 1881 ASSERT3P(lwb, !=, NULL); 1882 ASSERT3P(zio, !=, NULL); 1883 ASSERT3U(size, !=, 0); 1884 1885 ztest_object_lock(zd, object, RL_READER); 1886 error = dmu_bonus_hold(os, object, FTAG, &db); 1887 if (error) { 1888 ztest_object_unlock(zd, object); 1889 return (error); 1890 } 1891 1892 crtxg = ztest_bt_bonus(db)->bt_crtxg; 1893 1894 if (crtxg == 0 || crtxg > txg) { 1895 dmu_buf_rele(db, FTAG); 1896 ztest_object_unlock(zd, object); 1897 return (ENOENT); 1898 } 1899 1900 dmu_object_info_from_db(db, &doi); 1901 dmu_buf_rele(db, FTAG); 1902 db = NULL; 1903 1904 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL); 1905 zgd->zgd_lwb = lwb; 1906 zgd->zgd_private = zd; 1907 1908 if (buf != NULL) { /* immediate write */ 1909 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, 1910 RL_READER); 1911 1912 error = dmu_read(os, object, offset, size, buf, 1913 DMU_READ_NO_PREFETCH); 1914 ASSERT(error == 0); 1915 } else { 1916 size = doi.doi_data_block_size; 1917 if (ISP2(size)) { 1918 offset = P2ALIGN(offset, size); 1919 } else { 1920 ASSERT(offset < size); 1921 offset = 0; 1922 } 1923 1924 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, 1925 RL_READER); 1926 1927 error = dmu_buf_hold(os, object, offset, zgd, &db, 1928 DMU_READ_NO_PREFETCH); 1929 1930 if (error == 0) { 1931 blkptr_t *bp = &lr->lr_blkptr; 1932 1933 zgd->zgd_db = db; 1934 zgd->zgd_bp = bp; 1935 1936 ASSERT(db->db_offset == offset); 1937 ASSERT(db->db_size == size); 1938 1939 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1940 ztest_get_done, zgd); 1941 1942 if (error == 0) 1943 return (0); 1944 } 1945 } 1946 1947 ztest_get_done(zgd, error); 1948 1949 return (error); 1950} 1951 1952static void * 1953ztest_lr_alloc(size_t lrsize, char *name) 1954{ 1955 char *lr; 1956 size_t namesize = name ? strlen(name) + 1 : 0; 1957 1958 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL); 1959 1960 if (name) 1961 bcopy(name, lr + lrsize, namesize); 1962 1963 return (lr); 1964} 1965 1966void 1967ztest_lr_free(void *lr, size_t lrsize, char *name) 1968{ 1969 size_t namesize = name ? strlen(name) + 1 : 0; 1970 1971 umem_free(lr, lrsize + namesize); 1972} 1973 1974/* 1975 * Lookup a bunch of objects. Returns the number of objects not found. 1976 */ 1977static int 1978ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count) 1979{ 1980 int missing = 0; 1981 int error; 1982 1983 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock)); 1984 1985 for (int i = 0; i < count; i++, od++) { 1986 od->od_object = 0; 1987 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name, 1988 sizeof (uint64_t), 1, &od->od_object); 1989 if (error) { 1990 ASSERT(error == ENOENT); 1991 ASSERT(od->od_object == 0); 1992 missing++; 1993 } else { 1994 dmu_buf_t *db; 1995 ztest_block_tag_t *bbt; 1996 dmu_object_info_t doi; 1997 1998 ASSERT(od->od_object != 0); 1999 ASSERT(missing == 0); /* there should be no gaps */ 2000 2001 ztest_object_lock(zd, od->od_object, RL_READER); 2002 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os, 2003 od->od_object, FTAG, &db)); 2004 dmu_object_info_from_db(db, &doi); 2005 bbt = ztest_bt_bonus(db); 2006 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 2007 od->od_type = doi.doi_type; 2008 od->od_blocksize = doi.doi_data_block_size; 2009 od->od_gen = bbt->bt_gen; 2010 dmu_buf_rele(db, FTAG); 2011 ztest_object_unlock(zd, od->od_object); 2012 } 2013 } 2014 2015 return (missing); 2016} 2017 2018static int 2019ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count) 2020{ 2021 int missing = 0; 2022 2023 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock)); 2024 2025 for (int i = 0; i < count; i++, od++) { 2026 if (missing) { 2027 od->od_object = 0; 2028 missing++; 2029 continue; 2030 } 2031 2032 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); 2033 2034 lr->lr_doid = od->od_dir; 2035 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */ 2036 lr->lrz_type = od->od_crtype; 2037 lr->lrz_blocksize = od->od_crblocksize; 2038 lr->lrz_ibshift = ztest_random_ibshift(); 2039 lr->lrz_bonustype = DMU_OT_UINT64_OTHER; 2040 lr->lrz_bonuslen = dmu_bonus_max(); 2041 lr->lr_gen = od->od_crgen; 2042 lr->lr_crtime[0] = time(NULL); 2043 2044 if (ztest_replay_create(zd, lr, B_FALSE) != 0) { 2045 ASSERT(missing == 0); 2046 od->od_object = 0; 2047 missing++; 2048 } else { 2049 od->od_object = lr->lr_foid; 2050 od->od_type = od->od_crtype; 2051 od->od_blocksize = od->od_crblocksize; 2052 od->od_gen = od->od_crgen; 2053 ASSERT(od->od_object != 0); 2054 } 2055 2056 ztest_lr_free(lr, sizeof (*lr), od->od_name); 2057 } 2058 2059 return (missing); 2060} 2061 2062static int 2063ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count) 2064{ 2065 int missing = 0; 2066 int error; 2067 2068 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock)); 2069 2070 od += count - 1; 2071 2072 for (int i = count - 1; i >= 0; i--, od--) { 2073 if (missing) { 2074 missing++; 2075 continue; 2076 } 2077 2078 /* 2079 * No object was found. 2080 */ 2081 if (od->od_object == 0) 2082 continue; 2083 2084 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); 2085 2086 lr->lr_doid = od->od_dir; 2087 2088 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) { 2089 ASSERT3U(error, ==, ENOSPC); 2090 missing++; 2091 } else { 2092 od->od_object = 0; 2093 } 2094 ztest_lr_free(lr, sizeof (*lr), od->od_name); 2095 } 2096 2097 return (missing); 2098} 2099 2100static int 2101ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size, 2102 void *data) 2103{ 2104 lr_write_t *lr; 2105 int error; 2106 2107 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL); 2108 2109 lr->lr_foid = object; 2110 lr->lr_offset = offset; 2111 lr->lr_length = size; 2112 lr->lr_blkoff = 0; 2113 BP_ZERO(&lr->lr_blkptr); 2114 2115 bcopy(data, lr + 1, size); 2116 2117 error = ztest_replay_write(zd, lr, B_FALSE); 2118 2119 ztest_lr_free(lr, sizeof (*lr) + size, NULL); 2120 2121 return (error); 2122} 2123 2124static int 2125ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) 2126{ 2127 lr_truncate_t *lr; 2128 int error; 2129 2130 lr = ztest_lr_alloc(sizeof (*lr), NULL); 2131 2132 lr->lr_foid = object; 2133 lr->lr_offset = offset; 2134 lr->lr_length = size; 2135 2136 error = ztest_replay_truncate(zd, lr, B_FALSE); 2137 2138 ztest_lr_free(lr, sizeof (*lr), NULL); 2139 2140 return (error); 2141} 2142 2143static int 2144ztest_setattr(ztest_ds_t *zd, uint64_t object) 2145{ 2146 lr_setattr_t *lr; 2147 int error; 2148 2149 lr = ztest_lr_alloc(sizeof (*lr), NULL); 2150 2151 lr->lr_foid = object; 2152 lr->lr_size = 0; 2153 lr->lr_mode = 0; 2154 2155 error = ztest_replay_setattr(zd, lr, B_FALSE); 2156 2157 ztest_lr_free(lr, sizeof (*lr), NULL); 2158 2159 return (error); 2160} 2161 2162static void 2163ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) 2164{ 2165 objset_t *os = zd->zd_os; 2166 dmu_tx_t *tx; 2167 uint64_t txg; 2168 rl_t *rl; 2169 2170 txg_wait_synced(dmu_objset_pool(os), 0); 2171 2172 ztest_object_lock(zd, object, RL_READER); 2173 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER); 2174 2175 tx = dmu_tx_create(os); 2176 2177 dmu_tx_hold_write(tx, object, offset, size); 2178 2179 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 2180 2181 if (txg != 0) { 2182 dmu_prealloc(os, object, offset, size, tx); 2183 dmu_tx_commit(tx); 2184 txg_wait_synced(dmu_objset_pool(os), txg); 2185 } else { 2186 (void) dmu_free_long_range(os, object, offset, size); 2187 } 2188 2189 ztest_range_unlock(rl); 2190 ztest_object_unlock(zd, object); 2191} 2192 2193static void 2194ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset) 2195{ 2196 int err; 2197 ztest_block_tag_t wbt; 2198 dmu_object_info_t doi; 2199 enum ztest_io_type io_type; 2200 uint64_t blocksize; 2201 void *data; 2202 2203 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0); 2204 blocksize = doi.doi_data_block_size; 2205 data = umem_alloc(blocksize, UMEM_NOFAIL); 2206 2207 /* 2208 * Pick an i/o type at random, biased toward writing block tags. 2209 */ 2210 io_type = ztest_random(ZTEST_IO_TYPES); 2211 if (ztest_random(2) == 0) 2212 io_type = ZTEST_IO_WRITE_TAG; 2213 2214 rw_enter(&zd->zd_zilog_lock, RW_READER); 2215 2216 switch (io_type) { 2217 2218 case ZTEST_IO_WRITE_TAG: 2219 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0); 2220 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt); 2221 break; 2222 2223 case ZTEST_IO_WRITE_PATTERN: 2224 (void) memset(data, 'a' + (object + offset) % 5, blocksize); 2225 if (ztest_random(2) == 0) { 2226 /* 2227 * Induce fletcher2 collisions to ensure that 2228 * zio_ddt_collision() detects and resolves them 2229 * when using fletcher2-verify for deduplication. 2230 */ 2231 ((uint64_t *)data)[0] ^= 1ULL << 63; 2232 ((uint64_t *)data)[4] ^= 1ULL << 63; 2233 } 2234 (void) ztest_write(zd, object, offset, blocksize, data); 2235 break; 2236 2237 case ZTEST_IO_WRITE_ZEROES: 2238 bzero(data, blocksize); 2239 (void) ztest_write(zd, object, offset, blocksize, data); 2240 break; 2241 2242 case ZTEST_IO_TRUNCATE: 2243 (void) ztest_truncate(zd, object, offset, blocksize); 2244 break; 2245 2246 case ZTEST_IO_SETATTR: 2247 (void) ztest_setattr(zd, object); 2248 break; 2249 2250 case ZTEST_IO_REWRITE: 2251 rw_enter(&ztest_name_lock, RW_READER); 2252 err = ztest_dsl_prop_set_uint64(zd->zd_name, 2253 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa), 2254 B_FALSE); 2255 VERIFY(err == 0 || err == ENOSPC); 2256 err = ztest_dsl_prop_set_uint64(zd->zd_name, 2257 ZFS_PROP_COMPRESSION, 2258 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), 2259 B_FALSE); 2260 VERIFY(err == 0 || err == ENOSPC); 2261 rw_exit(&ztest_name_lock); 2262 2263 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data, 2264 DMU_READ_NO_PREFETCH)); 2265 2266 (void) ztest_write(zd, object, offset, blocksize, data); 2267 break; 2268 } 2269 2270 rw_exit(&zd->zd_zilog_lock); 2271 2272 umem_free(data, blocksize); 2273} 2274 2275/* 2276 * Initialize an object description template. 2277 */ 2278static void 2279ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index, 2280 dmu_object_type_t type, uint64_t blocksize, uint64_t gen) 2281{ 2282 od->od_dir = ZTEST_DIROBJ; 2283 od->od_object = 0; 2284 2285 od->od_crtype = type; 2286 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize(); 2287 od->od_crgen = gen; 2288 2289 od->od_type = DMU_OT_NONE; 2290 od->od_blocksize = 0; 2291 od->od_gen = 0; 2292 2293 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]", 2294 tag, (int64_t)id, index); 2295} 2296 2297/* 2298 * Lookup or create the objects for a test using the od template. 2299 * If the objects do not all exist, or if 'remove' is specified, 2300 * remove any existing objects and create new ones. Otherwise, 2301 * use the existing objects. 2302 */ 2303static int 2304ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove) 2305{ 2306 int count = size / sizeof (*od); 2307 int rv = 0; 2308 2309 mutex_enter(&zd->zd_dirobj_lock); 2310 if ((ztest_lookup(zd, od, count) != 0 || remove) && 2311 (ztest_remove(zd, od, count) != 0 || 2312 ztest_create(zd, od, count) != 0)) 2313 rv = -1; 2314 zd->zd_od = od; 2315 mutex_exit(&zd->zd_dirobj_lock); 2316 2317 return (rv); 2318} 2319 2320/* ARGSUSED */ 2321void 2322ztest_zil_commit(ztest_ds_t *zd, uint64_t id) 2323{ 2324 zilog_t *zilog = zd->zd_zilog; 2325 2326 rw_enter(&zd->zd_zilog_lock, RW_READER); 2327 2328 zil_commit(zilog, ztest_random(ZTEST_OBJECTS)); 2329 2330 /* 2331 * Remember the committed values in zd, which is in parent/child 2332 * shared memory. If we die, the next iteration of ztest_run() 2333 * will verify that the log really does contain this record. 2334 */ 2335 mutex_enter(&zilog->zl_lock); 2336 ASSERT(zd->zd_shared != NULL); 2337 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq); 2338 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq; 2339 mutex_exit(&zilog->zl_lock); 2340 2341 rw_exit(&zd->zd_zilog_lock); 2342} 2343 2344/* 2345 * This function is designed to simulate the operations that occur during a 2346 * mount/unmount operation. We hold the dataset across these operations in an 2347 * attempt to expose any implicit assumptions about ZIL management. 2348 */ 2349/* ARGSUSED */ 2350void 2351ztest_zil_remount(ztest_ds_t *zd, uint64_t id) 2352{ 2353 objset_t *os = zd->zd_os; 2354 2355 /* 2356 * We grab the zd_dirobj_lock to ensure that no other thread is 2357 * updating the zil (i.e. adding in-memory log records) and the 2358 * zd_zilog_lock to block any I/O. 2359 */ 2360 mutex_enter(&zd->zd_dirobj_lock); 2361 rw_enter(&zd->zd_zilog_lock, RW_WRITER); 2362 2363 /* zfsvfs_teardown() */ 2364 zil_close(zd->zd_zilog); 2365 2366 /* zfsvfs_setup() */ 2367 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog); 2368 zil_replay(os, zd, ztest_replay_vector); 2369 2370 rw_exit(&zd->zd_zilog_lock); 2371 mutex_exit(&zd->zd_dirobj_lock); 2372} 2373 2374/* 2375 * Verify that we can't destroy an active pool, create an existing pool, 2376 * or create a pool with a bad vdev spec. 2377 */ 2378/* ARGSUSED */ 2379void 2380ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id) 2381{ 2382 ztest_shared_opts_t *zo = &ztest_opts; 2383 spa_t *spa; 2384 nvlist_t *nvroot; 2385 2386 /* 2387 * Attempt to create using a bad file. 2388 */ 2389 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); 2390 VERIFY3U(ENOENT, ==, 2391 spa_create("ztest_bad_file", nvroot, NULL, NULL)); 2392 nvlist_free(nvroot); 2393 2394 /* 2395 * Attempt to create using a bad mirror. 2396 */ 2397 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1); 2398 VERIFY3U(ENOENT, ==, 2399 spa_create("ztest_bad_mirror", nvroot, NULL, NULL)); 2400 nvlist_free(nvroot); 2401 2402 /* 2403 * Attempt to create an existing pool. It shouldn't matter 2404 * what's in the nvroot; we should fail with EEXIST. 2405 */ 2406 rw_enter(&ztest_name_lock, RW_READER); 2407 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); 2408 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL)); 2409 nvlist_free(nvroot); 2410 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG)); 2411 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool)); 2412 spa_close(spa, FTAG); 2413 2414 rw_exit(&ztest_name_lock); 2415} 2416 2417/* ARGSUSED */ 2418void 2419ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id) 2420{ 2421 spa_t *spa; 2422 uint64_t initial_version = SPA_VERSION_INITIAL; 2423 uint64_t version, newversion; 2424 nvlist_t *nvroot, *props; 2425 char *name; 2426 2427 mutex_enter(&ztest_vdev_lock); 2428 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool); 2429 2430 /* 2431 * Clean up from previous runs. 2432 */ 2433 (void) spa_destroy(name); 2434 2435 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0, 2436 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1); 2437 2438 /* 2439 * If we're configuring a RAIDZ device then make sure that the 2440 * the initial version is capable of supporting that feature. 2441 */ 2442 switch (ztest_opts.zo_raidz_parity) { 2443 case 0: 2444 case 1: 2445 initial_version = SPA_VERSION_INITIAL; 2446 break; 2447 case 2: 2448 initial_version = SPA_VERSION_RAIDZ2; 2449 break; 2450 case 3: 2451 initial_version = SPA_VERSION_RAIDZ3; 2452 break; 2453 } 2454 2455 /* 2456 * Create a pool with a spa version that can be upgraded. Pick 2457 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES. 2458 */ 2459 do { 2460 version = ztest_random_spa_version(initial_version); 2461 } while (version > SPA_VERSION_BEFORE_FEATURES); 2462 2463 props = fnvlist_alloc(); 2464 fnvlist_add_uint64(props, 2465 zpool_prop_to_name(ZPOOL_PROP_VERSION), version); 2466 VERIFY0(spa_create(name, nvroot, props, NULL)); 2467 fnvlist_free(nvroot); 2468 fnvlist_free(props); 2469 2470 VERIFY0(spa_open(name, &spa, FTAG)); 2471 VERIFY3U(spa_version(spa), ==, version); 2472 newversion = ztest_random_spa_version(version + 1); 2473 2474 if (ztest_opts.zo_verbose >= 4) { 2475 (void) printf("upgrading spa version from %llu to %llu\n", 2476 (u_longlong_t)version, (u_longlong_t)newversion); 2477 } 2478 2479 spa_upgrade(spa, newversion); 2480 VERIFY3U(spa_version(spa), >, version); 2481 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config, 2482 zpool_prop_to_name(ZPOOL_PROP_VERSION))); 2483 spa_close(spa, FTAG); 2484 2485 strfree(name); 2486 mutex_exit(&ztest_vdev_lock); 2487} 2488 2489static void 2490ztest_spa_checkpoint(spa_t *spa) 2491{ 2492 ASSERT(MUTEX_HELD(&ztest_checkpoint_lock)); 2493 2494 int error = spa_checkpoint(spa->spa_name); 2495 2496 switch (error) { 2497 case 0: 2498 case ZFS_ERR_DEVRM_IN_PROGRESS: 2499 case ZFS_ERR_DISCARDING_CHECKPOINT: 2500 case ZFS_ERR_CHECKPOINT_EXISTS: 2501 break; 2502 case ENOSPC: 2503 ztest_record_enospc(FTAG); 2504 break; 2505 default: 2506 fatal(0, "spa_checkpoint(%s) = %d", spa->spa_name, error); 2507 } 2508} 2509 2510static void 2511ztest_spa_discard_checkpoint(spa_t *spa) 2512{ 2513 ASSERT(MUTEX_HELD(&ztest_checkpoint_lock)); 2514 2515 int error = spa_checkpoint_discard(spa->spa_name); 2516 2517 switch (error) { 2518 case 0: 2519 case ZFS_ERR_DISCARDING_CHECKPOINT: 2520 case ZFS_ERR_NO_CHECKPOINT: 2521 break; 2522 default: 2523 fatal(0, "spa_discard_checkpoint(%s) = %d", 2524 spa->spa_name, error); 2525 } 2526 2527} 2528 2529/* ARGSUSED */ 2530void 2531ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id) 2532{ 2533 spa_t *spa = ztest_spa; 2534 2535 mutex_enter(&ztest_checkpoint_lock); 2536 if (ztest_random(2) == 0) { 2537 ztest_spa_checkpoint(spa); 2538 } else { 2539 ztest_spa_discard_checkpoint(spa); 2540 } 2541 mutex_exit(&ztest_checkpoint_lock); 2542} 2543 2544 2545static vdev_t * 2546vdev_lookup_by_path(vdev_t *vd, const char *path) 2547{ 2548 vdev_t *mvd; 2549 2550 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0) 2551 return (vd); 2552 2553 for (int c = 0; c < vd->vdev_children; c++) 2554 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != 2555 NULL) 2556 return (mvd); 2557 2558 return (NULL); 2559} 2560 2561/* 2562 * Find the first available hole which can be used as a top-level. 2563 */ 2564int 2565find_vdev_hole(spa_t *spa) 2566{ 2567 vdev_t *rvd = spa->spa_root_vdev; 2568 int c; 2569 2570 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV); 2571 2572 for (c = 0; c < rvd->vdev_children; c++) { 2573 vdev_t *cvd = rvd->vdev_child[c]; 2574 2575 if (cvd->vdev_ishole) 2576 break; 2577 } 2578 return (c); 2579} 2580 2581/* 2582 * Verify that vdev_add() works as expected. 2583 */ 2584/* ARGSUSED */ 2585void 2586ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id) 2587{ 2588 ztest_shared_t *zs = ztest_shared; 2589 spa_t *spa = ztest_spa; 2590 uint64_t leaves; 2591 uint64_t guid; 2592 nvlist_t *nvroot; 2593 int error; 2594 2595 mutex_enter(&ztest_vdev_lock); 2596 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz; 2597 2598 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2599 2600 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves; 2601 2602 /* 2603 * If we have slogs then remove them 1/4 of the time. 2604 */ 2605 if (spa_has_slogs(spa) && ztest_random(4) == 0) { 2606 /* 2607 * Grab the guid from the head of the log class rotor. 2608 */ 2609 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid; 2610 2611 spa_config_exit(spa, SCL_VDEV, FTAG); 2612 2613 /* 2614 * We have to grab the zs_name_lock as writer to 2615 * prevent a race between removing a slog (dmu_objset_find) 2616 * and destroying a dataset. Removing the slog will 2617 * grab a reference on the dataset which may cause 2618 * dmu_objset_destroy() to fail with EBUSY thus 2619 * leaving the dataset in an inconsistent state. 2620 */ 2621 rw_enter(&ztest_name_lock, RW_WRITER); 2622 error = spa_vdev_remove(spa, guid, B_FALSE); 2623 rw_exit(&ztest_name_lock); 2624 2625 switch (error) { 2626 case 0: 2627 case EEXIST: 2628 case ZFS_ERR_CHECKPOINT_EXISTS: 2629 case ZFS_ERR_DISCARDING_CHECKPOINT: 2630 break; 2631 default: 2632 fatal(0, "spa_vdev_remove() = %d", error); 2633 } 2634 } else { 2635 spa_config_exit(spa, SCL_VDEV, FTAG); 2636 2637 /* 2638 * Make 1/4 of the devices be log devices. 2639 */ 2640 nvroot = make_vdev_root(NULL, NULL, NULL, 2641 ztest_opts.zo_vdev_size, 0, 2642 ztest_random(4) == 0, ztest_opts.zo_raidz, 2643 zs->zs_mirrors, 1); 2644 2645 error = spa_vdev_add(spa, nvroot); 2646 nvlist_free(nvroot); 2647 2648 switch (error) { 2649 case 0: 2650 break; 2651 case ENOSPC: 2652 ztest_record_enospc("spa_vdev_add"); 2653 break; 2654 default: 2655 fatal(0, "spa_vdev_add() = %d", error); 2656 } 2657 } 2658 2659 mutex_exit(&ztest_vdev_lock); 2660} 2661 2662/* 2663 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected. 2664 */ 2665/* ARGSUSED */ 2666void 2667ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id) 2668{ 2669 ztest_shared_t *zs = ztest_shared; 2670 spa_t *spa = ztest_spa; 2671 vdev_t *rvd = spa->spa_root_vdev; 2672 spa_aux_vdev_t *sav; 2673 char *aux; 2674 uint64_t guid = 0; 2675 int error; 2676 2677 if (ztest_random(2) == 0) { 2678 sav = &spa->spa_spares; 2679 aux = ZPOOL_CONFIG_SPARES; 2680 } else { 2681 sav = &spa->spa_l2cache; 2682 aux = ZPOOL_CONFIG_L2CACHE; 2683 } 2684 2685 mutex_enter(&ztest_vdev_lock); 2686 2687 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2688 2689 if (sav->sav_count != 0 && ztest_random(4) == 0) { 2690 /* 2691 * Pick a random device to remove. 2692 */ 2693 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid; 2694 } else { 2695 /* 2696 * Find an unused device we can add. 2697 */ 2698 zs->zs_vdev_aux = 0; 2699 for (;;) { 2700 char path[MAXPATHLEN]; 2701 int c; 2702 (void) snprintf(path, sizeof (path), ztest_aux_template, 2703 ztest_opts.zo_dir, ztest_opts.zo_pool, aux, 2704 zs->zs_vdev_aux); 2705 for (c = 0; c < sav->sav_count; c++) 2706 if (strcmp(sav->sav_vdevs[c]->vdev_path, 2707 path) == 0) 2708 break; 2709 if (c == sav->sav_count && 2710 vdev_lookup_by_path(rvd, path) == NULL) 2711 break; 2712 zs->zs_vdev_aux++; 2713 } 2714 } 2715 2716 spa_config_exit(spa, SCL_VDEV, FTAG); 2717 2718 if (guid == 0) { 2719 /* 2720 * Add a new device. 2721 */ 2722 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL, 2723 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1); 2724 error = spa_vdev_add(spa, nvroot); 2725 2726 switch (error) { 2727 case 0: 2728 break; 2729 default: 2730 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error); 2731 } 2732 nvlist_free(nvroot); 2733 } else { 2734 /* 2735 * Remove an existing device. Sometimes, dirty its 2736 * vdev state first to make sure we handle removal 2737 * of devices that have pending state changes. 2738 */ 2739 if (ztest_random(2) == 0) 2740 (void) vdev_online(spa, guid, 0, NULL); 2741 2742 error = spa_vdev_remove(spa, guid, B_FALSE); 2743 2744 switch (error) { 2745 case 0: 2746 case EBUSY: 2747 case ZFS_ERR_CHECKPOINT_EXISTS: 2748 case ZFS_ERR_DISCARDING_CHECKPOINT: 2749 break; 2750 default: 2751 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error); 2752 } 2753 } 2754 2755 mutex_exit(&ztest_vdev_lock); 2756} 2757 2758/* 2759 * split a pool if it has mirror tlvdevs 2760 */ 2761/* ARGSUSED */ 2762void 2763ztest_split_pool(ztest_ds_t *zd, uint64_t id) 2764{ 2765 ztest_shared_t *zs = ztest_shared; 2766 spa_t *spa = ztest_spa; 2767 vdev_t *rvd = spa->spa_root_vdev; 2768 nvlist_t *tree, **child, *config, *split, **schild; 2769 uint_t c, children, schildren = 0, lastlogid = 0; 2770 int error = 0; 2771 2772 mutex_enter(&ztest_vdev_lock); 2773 2774 /* ensure we have a useable config; mirrors of raidz aren't supported */ 2775 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) { 2776 mutex_exit(&ztest_vdev_lock); 2777 return; 2778 } 2779 2780 /* clean up the old pool, if any */ 2781 (void) spa_destroy("splitp"); 2782 2783 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2784 2785 /* generate a config from the existing config */ 2786 mutex_enter(&spa->spa_props_lock); 2787 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE, 2788 &tree) == 0); 2789 mutex_exit(&spa->spa_props_lock); 2790 2791 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child, 2792 &children) == 0); 2793 2794 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *)); 2795 for (c = 0; c < children; c++) { 2796 vdev_t *tvd = rvd->vdev_child[c]; 2797 nvlist_t **mchild; 2798 uint_t mchildren; 2799 2800 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) { 2801 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME, 2802 0) == 0); 2803 VERIFY(nvlist_add_string(schild[schildren], 2804 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); 2805 VERIFY(nvlist_add_uint64(schild[schildren], 2806 ZPOOL_CONFIG_IS_HOLE, 1) == 0); 2807 if (lastlogid == 0) 2808 lastlogid = schildren; 2809 ++schildren; 2810 continue; 2811 } 2812 lastlogid = 0; 2813 VERIFY(nvlist_lookup_nvlist_array(child[c], 2814 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0); 2815 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0); 2816 } 2817 2818 /* OK, create a config that can be used to split */ 2819 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0); 2820 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE, 2821 VDEV_TYPE_ROOT) == 0); 2822 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild, 2823 lastlogid != 0 ? lastlogid : schildren) == 0); 2824 2825 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0); 2826 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0); 2827 2828 for (c = 0; c < schildren; c++) 2829 nvlist_free(schild[c]); 2830 free(schild); 2831 nvlist_free(split); 2832 2833 spa_config_exit(spa, SCL_VDEV, FTAG); 2834 2835 rw_enter(&ztest_name_lock, RW_WRITER); 2836 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE); 2837 rw_exit(&ztest_name_lock); 2838 2839 nvlist_free(config); 2840 2841 if (error == 0) { 2842 (void) printf("successful split - results:\n"); 2843 mutex_enter(&spa_namespace_lock); 2844 show_pool_stats(spa); 2845 show_pool_stats(spa_lookup("splitp")); 2846 mutex_exit(&spa_namespace_lock); 2847 ++zs->zs_splits; 2848 --zs->zs_mirrors; 2849 } 2850 mutex_exit(&ztest_vdev_lock); 2851} 2852 2853/* 2854 * Verify that we can attach and detach devices. 2855 */ 2856/* ARGSUSED */ 2857void 2858ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id) 2859{ 2860 ztest_shared_t *zs = ztest_shared; 2861 spa_t *spa = ztest_spa; 2862 spa_aux_vdev_t *sav = &spa->spa_spares; 2863 vdev_t *rvd = spa->spa_root_vdev; 2864 vdev_t *oldvd, *newvd, *pvd; 2865 nvlist_t *root; 2866 uint64_t leaves; 2867 uint64_t leaf, top; 2868 uint64_t ashift = ztest_get_ashift(); 2869 uint64_t oldguid, pguid; 2870 uint64_t oldsize, newsize; 2871 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; 2872 int replacing; 2873 int oldvd_has_siblings = B_FALSE; 2874 int newvd_is_spare = B_FALSE; 2875 int oldvd_is_log; 2876 int error, expected_error; 2877 2878 mutex_enter(&ztest_vdev_lock); 2879 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; 2880 2881 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2882 2883 /* 2884 * If a vdev is in the process of being removed, its removal may 2885 * finish while we are in progress, leading to an unexpected error 2886 * value. Don't bother trying to attach while we are in the middle 2887 * of removal. 2888 */ 2889 if (ztest_device_removal_active) { 2890 spa_config_exit(spa, SCL_ALL, FTAG); 2891 mutex_exit(&ztest_vdev_lock); 2892 return; 2893 } 2894 2895 /* 2896 * Decide whether to do an attach or a replace. 2897 */ 2898 replacing = ztest_random(2); 2899 2900 /* 2901 * Pick a random top-level vdev. 2902 */ 2903 top = ztest_random_vdev_top(spa, B_TRUE); 2904 2905 /* 2906 * Pick a random leaf within it. 2907 */ 2908 leaf = ztest_random(leaves); 2909 2910 /* 2911 * Locate this vdev. 2912 */ 2913 oldvd = rvd->vdev_child[top]; 2914 if (zs->zs_mirrors >= 1) { 2915 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops); 2916 ASSERT(oldvd->vdev_children >= zs->zs_mirrors); 2917 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz]; 2918 } 2919 if (ztest_opts.zo_raidz > 1) { 2920 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops); 2921 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz); 2922 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz]; 2923 } 2924 2925 /* 2926 * If we're already doing an attach or replace, oldvd may be a 2927 * mirror vdev -- in which case, pick a random child. 2928 */ 2929 while (oldvd->vdev_children != 0) { 2930 oldvd_has_siblings = B_TRUE; 2931 ASSERT(oldvd->vdev_children >= 2); 2932 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)]; 2933 } 2934 2935 oldguid = oldvd->vdev_guid; 2936 oldsize = vdev_get_min_asize(oldvd); 2937 oldvd_is_log = oldvd->vdev_top->vdev_islog; 2938 (void) strcpy(oldpath, oldvd->vdev_path); 2939 pvd = oldvd->vdev_parent; 2940 pguid = pvd->vdev_guid; 2941 2942 /* 2943 * If oldvd has siblings, then half of the time, detach it. 2944 */ 2945 if (oldvd_has_siblings && ztest_random(2) == 0) { 2946 spa_config_exit(spa, SCL_ALL, FTAG); 2947 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE); 2948 if (error != 0 && error != ENODEV && error != EBUSY && 2949 error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS && 2950 error != ZFS_ERR_DISCARDING_CHECKPOINT) 2951 fatal(0, "detach (%s) returned %d", oldpath, error); 2952 mutex_exit(&ztest_vdev_lock); 2953 return; 2954 } 2955 2956 /* 2957 * For the new vdev, choose with equal probability between the two 2958 * standard paths (ending in either 'a' or 'b') or a random hot spare. 2959 */ 2960 if (sav->sav_count != 0 && ztest_random(3) == 0) { 2961 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)]; 2962 newvd_is_spare = B_TRUE; 2963 (void) strcpy(newpath, newvd->vdev_path); 2964 } else { 2965 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template, 2966 ztest_opts.zo_dir, ztest_opts.zo_pool, 2967 top * leaves + leaf); 2968 if (ztest_random(2) == 0) 2969 newpath[strlen(newpath) - 1] = 'b'; 2970 newvd = vdev_lookup_by_path(rvd, newpath); 2971 } 2972 2973 if (newvd) { 2974 /* 2975 * Reopen to ensure the vdev's asize field isn't stale. 2976 */ 2977 vdev_reopen(newvd); 2978 newsize = vdev_get_min_asize(newvd); 2979 } else { 2980 /* 2981 * Make newsize a little bigger or smaller than oldsize. 2982 * If it's smaller, the attach should fail. 2983 * If it's larger, and we're doing a replace, 2984 * we should get dynamic LUN growth when we're done. 2985 */ 2986 newsize = 10 * oldsize / (9 + ztest_random(3)); 2987 } 2988 2989 /* 2990 * If pvd is not a mirror or root, the attach should fail with ENOTSUP, 2991 * unless it's a replace; in that case any non-replacing parent is OK. 2992 * 2993 * If newvd is already part of the pool, it should fail with EBUSY. 2994 * 2995 * If newvd is too small, it should fail with EOVERFLOW. 2996 */ 2997 if (pvd->vdev_ops != &vdev_mirror_ops && 2998 pvd->vdev_ops != &vdev_root_ops && (!replacing || 2999 pvd->vdev_ops == &vdev_replacing_ops || 3000 pvd->vdev_ops == &vdev_spare_ops)) 3001 expected_error = ENOTSUP; 3002 else if (newvd_is_spare && (!replacing || oldvd_is_log)) 3003 expected_error = ENOTSUP; 3004 else if (newvd == oldvd) 3005 expected_error = replacing ? 0 : EBUSY; 3006 else if (vdev_lookup_by_path(rvd, newpath) != NULL) 3007 expected_error = EBUSY; 3008 else if (newsize < oldsize) 3009 expected_error = EOVERFLOW; 3010 else if (ashift > oldvd->vdev_top->vdev_ashift) 3011 expected_error = EDOM; 3012 else 3013 expected_error = 0; 3014 3015 spa_config_exit(spa, SCL_ALL, FTAG); 3016 3017 /* 3018 * Build the nvlist describing newpath. 3019 */ 3020 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0, 3021 ashift, 0, 0, 0, 1); 3022 3023 error = spa_vdev_attach(spa, oldguid, root, replacing); 3024 3025 nvlist_free(root); 3026 3027 /* 3028 * If our parent was the replacing vdev, but the replace completed, 3029 * then instead of failing with ENOTSUP we may either succeed, 3030 * fail with ENODEV, or fail with EOVERFLOW. 3031 */ 3032 if (expected_error == ENOTSUP && 3033 (error == 0 || error == ENODEV || error == EOVERFLOW)) 3034 expected_error = error; 3035 3036 /* 3037 * If someone grew the LUN, the replacement may be too small. 3038 */ 3039 if (error == EOVERFLOW || error == EBUSY) 3040 expected_error = error; 3041 3042 if (error == ZFS_ERR_CHECKPOINT_EXISTS || 3043 error == ZFS_ERR_DISCARDING_CHECKPOINT) 3044 expected_error = error; 3045 3046 /* XXX workaround 6690467 */ 3047 if (error != expected_error && expected_error != EBUSY) { 3048 fatal(0, "attach (%s %llu, %s %llu, %d) " 3049 "returned %d, expected %d", 3050 oldpath, oldsize, newpath, 3051 newsize, replacing, error, expected_error); 3052 } 3053 3054 mutex_exit(&ztest_vdev_lock); 3055} 3056 3057/* ARGSUSED */ 3058void 3059ztest_device_removal(ztest_ds_t *zd, uint64_t id) 3060{ 3061 spa_t *spa = ztest_spa; 3062 vdev_t *vd; 3063 uint64_t guid; 3064 int error; 3065 3066 mutex_enter(&ztest_vdev_lock); 3067 3068 if (ztest_device_removal_active) { 3069 mutex_exit(&ztest_vdev_lock); 3070 return; 3071 } 3072 3073 /* 3074 * Remove a random top-level vdev and wait for removal to finish. 3075 */ 3076 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 3077 vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE)); 3078 guid = vd->vdev_guid; 3079 spa_config_exit(spa, SCL_VDEV, FTAG); 3080 3081 error = spa_vdev_remove(spa, guid, B_FALSE); 3082 if (error == 0) { 3083 ztest_device_removal_active = B_TRUE; 3084 mutex_exit(&ztest_vdev_lock); 3085 3086 while (spa->spa_vdev_removal != NULL) 3087 txg_wait_synced(spa_get_dsl(spa), 0); 3088 } else { 3089 mutex_exit(&ztest_vdev_lock); 3090 return; 3091 } 3092 3093 /* 3094 * The pool needs to be scrubbed after completing device removal. 3095 * Failure to do so may result in checksum errors due to the 3096 * strategy employed by ztest_fault_inject() when selecting which 3097 * offset are redundant and can be damaged. 3098 */ 3099 error = spa_scan(spa, POOL_SCAN_SCRUB); 3100 if (error == 0) { 3101 while (dsl_scan_scrubbing(spa_get_dsl(spa))) 3102 txg_wait_synced(spa_get_dsl(spa), 0); 3103 } 3104 3105 mutex_enter(&ztest_vdev_lock); 3106 ztest_device_removal_active = B_FALSE; 3107 mutex_exit(&ztest_vdev_lock); 3108} 3109 3110/* 3111 * Callback function which expands the physical size of the vdev. 3112 */ 3113vdev_t * 3114grow_vdev(vdev_t *vd, void *arg) 3115{ 3116 spa_t *spa = vd->vdev_spa; 3117 size_t *newsize = arg; 3118 size_t fsize; 3119 int fd; 3120 3121 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); 3122 ASSERT(vd->vdev_ops->vdev_op_leaf); 3123 3124 if ((fd = open(vd->vdev_path, O_RDWR)) == -1) 3125 return (vd); 3126 3127 fsize = lseek(fd, 0, SEEK_END); 3128 (void) ftruncate(fd, *newsize); 3129 3130 if (ztest_opts.zo_verbose >= 6) { 3131 (void) printf("%s grew from %lu to %lu bytes\n", 3132 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize); 3133 } 3134 (void) close(fd); 3135 return (NULL); 3136} 3137 3138/* 3139 * Callback function which expands a given vdev by calling vdev_online(). 3140 */ 3141/* ARGSUSED */ 3142vdev_t * 3143online_vdev(vdev_t *vd, void *arg) 3144{ 3145 spa_t *spa = vd->vdev_spa; 3146 vdev_t *tvd = vd->vdev_top; 3147 uint64_t guid = vd->vdev_guid; 3148 uint64_t generation = spa->spa_config_generation + 1; 3149 vdev_state_t newstate = VDEV_STATE_UNKNOWN; 3150 int error; 3151 3152 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); 3153 ASSERT(vd->vdev_ops->vdev_op_leaf); 3154 3155 /* Calling vdev_online will initialize the new metaslabs */ 3156 spa_config_exit(spa, SCL_STATE, spa); 3157 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate); 3158 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3159 3160 /* 3161 * If vdev_online returned an error or the underlying vdev_open 3162 * failed then we abort the expand. The only way to know that 3163 * vdev_open fails is by checking the returned newstate. 3164 */ 3165 if (error || newstate != VDEV_STATE_HEALTHY) { 3166 if (ztest_opts.zo_verbose >= 5) { 3167 (void) printf("Unable to expand vdev, state %llu, " 3168 "error %d\n", (u_longlong_t)newstate, error); 3169 } 3170 return (vd); 3171 } 3172 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY); 3173 3174 /* 3175 * Since we dropped the lock we need to ensure that we're 3176 * still talking to the original vdev. It's possible this 3177 * vdev may have been detached/replaced while we were 3178 * trying to online it. 3179 */ 3180 if (generation != spa->spa_config_generation) { 3181 if (ztest_opts.zo_verbose >= 5) { 3182 (void) printf("vdev configuration has changed, " 3183 "guid %llu, state %llu, expected gen %llu, " 3184 "got gen %llu\n", 3185 (u_longlong_t)guid, 3186 (u_longlong_t)tvd->vdev_state, 3187 (u_longlong_t)generation, 3188 (u_longlong_t)spa->spa_config_generation); 3189 } 3190 return (vd); 3191 } 3192 return (NULL); 3193} 3194 3195/* 3196 * Traverse the vdev tree calling the supplied function. 3197 * We continue to walk the tree until we either have walked all 3198 * children or we receive a non-NULL return from the callback. 3199 * If a NULL callback is passed, then we just return back the first 3200 * leaf vdev we encounter. 3201 */ 3202vdev_t * 3203vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg) 3204{ 3205 if (vd->vdev_ops->vdev_op_leaf) { 3206 if (func == NULL) 3207 return (vd); 3208 else 3209 return (func(vd, arg)); 3210 } 3211 3212 for (uint_t c = 0; c < vd->vdev_children; c++) { 3213 vdev_t *cvd = vd->vdev_child[c]; 3214 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL) 3215 return (cvd); 3216 } 3217 return (NULL); 3218} 3219 3220/* 3221 * Verify that dynamic LUN growth works as expected. 3222 */ 3223/* ARGSUSED */ 3224void 3225ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id) 3226{ 3227 spa_t *spa = ztest_spa; 3228 vdev_t *vd, *tvd; 3229 metaslab_class_t *mc; 3230 metaslab_group_t *mg; 3231 size_t psize, newsize; 3232 uint64_t top; 3233 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count; 3234 3235 mutex_enter(&ztest_checkpoint_lock); 3236 mutex_enter(&ztest_vdev_lock); 3237 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3238 3239 /* 3240 * If there is a vdev removal in progress, it could complete while 3241 * we are running, in which case we would not be able to verify 3242 * that the metaslab_class space increased (because it decreases 3243 * when the device removal completes). 3244 */ 3245 if (ztest_device_removal_active) { 3246 spa_config_exit(spa, SCL_STATE, spa); 3247 mutex_exit(&ztest_vdev_lock); 3248 mutex_exit(&ztest_checkpoint_lock); 3249 return; 3250 } 3251 3252 top = ztest_random_vdev_top(spa, B_TRUE); 3253 3254 tvd = spa->spa_root_vdev->vdev_child[top]; 3255 mg = tvd->vdev_mg; 3256 mc = mg->mg_class; 3257 old_ms_count = tvd->vdev_ms_count; 3258 old_class_space = metaslab_class_get_space(mc); 3259 3260 /* 3261 * Determine the size of the first leaf vdev associated with 3262 * our top-level device. 3263 */ 3264 vd = vdev_walk_tree(tvd, NULL, NULL); 3265 ASSERT3P(vd, !=, NULL); 3266 ASSERT(vd->vdev_ops->vdev_op_leaf); 3267 3268 psize = vd->vdev_psize; 3269 3270 /* 3271 * We only try to expand the vdev if it's healthy, less than 4x its 3272 * original size, and it has a valid psize. 3273 */ 3274 if (tvd->vdev_state != VDEV_STATE_HEALTHY || 3275 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) { 3276 spa_config_exit(spa, SCL_STATE, spa); 3277 mutex_exit(&ztest_vdev_lock); 3278 mutex_exit(&ztest_checkpoint_lock); 3279 return; 3280 } 3281 ASSERT(psize > 0); 3282 newsize = psize + psize / 8; 3283 ASSERT3U(newsize, >, psize); 3284 3285 if (ztest_opts.zo_verbose >= 6) { 3286 (void) printf("Expanding LUN %s from %lu to %lu\n", 3287 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize); 3288 } 3289 3290 /* 3291 * Growing the vdev is a two step process: 3292 * 1). expand the physical size (i.e. relabel) 3293 * 2). online the vdev to create the new metaslabs 3294 */ 3295 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL || 3296 vdev_walk_tree(tvd, online_vdev, NULL) != NULL || 3297 tvd->vdev_state != VDEV_STATE_HEALTHY) { 3298 if (ztest_opts.zo_verbose >= 5) { 3299 (void) printf("Could not expand LUN because " 3300 "the vdev configuration changed.\n"); 3301 } 3302 spa_config_exit(spa, SCL_STATE, spa); 3303 mutex_exit(&ztest_vdev_lock); 3304 mutex_exit(&ztest_checkpoint_lock); 3305 return; 3306 } 3307 3308 spa_config_exit(spa, SCL_STATE, spa); 3309 3310 /* 3311 * Expanding the LUN will update the config asynchronously, 3312 * thus we must wait for the async thread to complete any 3313 * pending tasks before proceeding. 3314 */ 3315 for (;;) { 3316 boolean_t done; 3317 mutex_enter(&spa->spa_async_lock); 3318 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks); 3319 mutex_exit(&spa->spa_async_lock); 3320 if (done) 3321 break; 3322 txg_wait_synced(spa_get_dsl(spa), 0); 3323 (void) poll(NULL, 0, 100); 3324 } 3325 3326 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3327 3328 tvd = spa->spa_root_vdev->vdev_child[top]; 3329 new_ms_count = tvd->vdev_ms_count; 3330 new_class_space = metaslab_class_get_space(mc); 3331 3332 if (tvd->vdev_mg != mg || mg->mg_class != mc) { 3333 if (ztest_opts.zo_verbose >= 5) { 3334 (void) printf("Could not verify LUN expansion due to " 3335 "intervening vdev offline or remove.\n"); 3336 } 3337 spa_config_exit(spa, SCL_STATE, spa); 3338 mutex_exit(&ztest_vdev_lock); 3339 mutex_exit(&ztest_checkpoint_lock); 3340 return; 3341 } 3342 3343 /* 3344 * Make sure we were able to grow the vdev. 3345 */ 3346 if (new_ms_count <= old_ms_count) { 3347 fatal(0, "LUN expansion failed: ms_count %llu < %llu\n", 3348 old_ms_count, new_ms_count); 3349 } 3350 3351 /* 3352 * Make sure we were able to grow the pool. 3353 */ 3354 if (new_class_space <= old_class_space) { 3355 fatal(0, "LUN expansion failed: class_space %llu < %llu\n", 3356 old_class_space, new_class_space); 3357 } 3358 3359 if (ztest_opts.zo_verbose >= 5) { 3360 char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ]; 3361 3362 nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf)); 3363 nicenum(new_class_space, newnumbuf, sizeof (newnumbuf)); 3364 (void) printf("%s grew from %s to %s\n", 3365 spa->spa_name, oldnumbuf, newnumbuf); 3366 } 3367 3368 spa_config_exit(spa, SCL_STATE, spa); 3369 mutex_exit(&ztest_vdev_lock); 3370 mutex_exit(&ztest_checkpoint_lock); 3371} 3372 3373/* 3374 * Verify that dmu_objset_{create,destroy,open,close} work as expected. 3375 */ 3376/* ARGSUSED */ 3377static void 3378ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 3379{ 3380 /* 3381 * Create the objects common to all ztest datasets. 3382 */ 3383 VERIFY(zap_create_claim(os, ZTEST_DIROBJ, 3384 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 3385} 3386 3387static int 3388ztest_dataset_create(char *dsname) 3389{ 3390 uint64_t zilset = ztest_random(100); 3391 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, 3392 ztest_objset_create_cb, NULL); 3393 3394 if (err || zilset < 80) 3395 return (err); 3396 3397 if (ztest_opts.zo_verbose >= 6) 3398 (void) printf("Setting dataset %s to sync always\n", dsname); 3399 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC, 3400 ZFS_SYNC_ALWAYS, B_FALSE)); 3401} 3402 3403/* ARGSUSED */ 3404static int 3405ztest_objset_destroy_cb(const char *name, void *arg) 3406{ 3407 objset_t *os; 3408 dmu_object_info_t doi; 3409 int error; 3410 3411 /* 3412 * Verify that the dataset contains a directory object. 3413 */ 3414 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os)); 3415 error = dmu_object_info(os, ZTEST_DIROBJ, &doi); 3416 if (error != ENOENT) { 3417 /* We could have crashed in the middle of destroying it */ 3418 ASSERT0(error); 3419 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER); 3420 ASSERT3S(doi.doi_physical_blocks_512, >=, 0); 3421 } 3422 dmu_objset_disown(os, FTAG); 3423 3424 /* 3425 * Destroy the dataset. 3426 */ 3427 if (strchr(name, '@') != NULL) { 3428 VERIFY0(dsl_destroy_snapshot(name, B_FALSE)); 3429 } else { 3430 VERIFY0(dsl_destroy_head(name)); 3431 } 3432 return (0); 3433} 3434 3435static boolean_t 3436ztest_snapshot_create(char *osname, uint64_t id) 3437{ 3438 char snapname[ZFS_MAX_DATASET_NAME_LEN]; 3439 int error; 3440 3441 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id); 3442 3443 error = dmu_objset_snapshot_one(osname, snapname); 3444 if (error == ENOSPC) { 3445 ztest_record_enospc(FTAG); 3446 return (B_FALSE); 3447 } 3448 if (error != 0 && error != EEXIST) { 3449 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname, 3450 snapname, error); 3451 } 3452 return (B_TRUE); 3453} 3454 3455static boolean_t 3456ztest_snapshot_destroy(char *osname, uint64_t id) 3457{ 3458 char snapname[ZFS_MAX_DATASET_NAME_LEN]; 3459 int error; 3460 3461 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname, 3462 (u_longlong_t)id); 3463 3464 error = dsl_destroy_snapshot(snapname, B_FALSE); 3465 if (error != 0 && error != ENOENT) 3466 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error); 3467 return (B_TRUE); 3468} 3469 3470/* ARGSUSED */ 3471void 3472ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) 3473{ 3474 ztest_ds_t zdtmp; 3475 int iters; 3476 int error; 3477 objset_t *os, *os2; 3478 char name[ZFS_MAX_DATASET_NAME_LEN]; 3479 zilog_t *zilog; 3480 3481 rw_enter(&ztest_name_lock, RW_READER); 3482 3483 (void) snprintf(name, sizeof (name), "%s/temp_%llu", 3484 ztest_opts.zo_pool, (u_longlong_t)id); 3485 3486 /* 3487 * If this dataset exists from a previous run, process its replay log 3488 * half of the time. If we don't replay it, then dmu_objset_destroy() 3489 * (invoked from ztest_objset_destroy_cb()) should just throw it away. 3490 */ 3491 if (ztest_random(2) == 0 && 3492 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) { 3493 ztest_zd_init(&zdtmp, NULL, os); 3494 zil_replay(os, &zdtmp, ztest_replay_vector); 3495 ztest_zd_fini(&zdtmp); 3496 dmu_objset_disown(os, FTAG); 3497 } 3498 3499 /* 3500 * There may be an old instance of the dataset we're about to 3501 * create lying around from a previous run. If so, destroy it 3502 * and all of its snapshots. 3503 */ 3504 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, 3505 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 3506 3507 /* 3508 * Verify that the destroyed dataset is no longer in the namespace. 3509 */ 3510 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, 3511 FTAG, &os)); 3512 3513 /* 3514 * Verify that we can create a new dataset. 3515 */ 3516 error = ztest_dataset_create(name); 3517 if (error) { 3518 if (error == ENOSPC) { 3519 ztest_record_enospc(FTAG); 3520 rw_exit(&ztest_name_lock); 3521 return; 3522 } 3523 fatal(0, "dmu_objset_create(%s) = %d", name, error); 3524 } 3525 3526 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); 3527 3528 ztest_zd_init(&zdtmp, NULL, os); 3529 3530 /* 3531 * Open the intent log for it. 3532 */ 3533 zilog = zil_open(os, ztest_get_data); 3534 3535 /* 3536 * Put some objects in there, do a little I/O to them, 3537 * and randomly take a couple of snapshots along the way. 3538 */ 3539 iters = ztest_random(5); 3540 for (int i = 0; i < iters; i++) { 3541 ztest_dmu_object_alloc_free(&zdtmp, id); 3542 if (ztest_random(iters) == 0) 3543 (void) ztest_snapshot_create(name, i); 3544 } 3545 3546 /* 3547 * Verify that we cannot create an existing dataset. 3548 */ 3549 VERIFY3U(EEXIST, ==, 3550 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL)); 3551 3552 /* 3553 * Verify that we can hold an objset that is also owned. 3554 */ 3555 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2)); 3556 dmu_objset_rele(os2, FTAG); 3557 3558 /* 3559 * Verify that we cannot own an objset that is already owned. 3560 */ 3561 VERIFY3U(EBUSY, ==, 3562 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2)); 3563 3564 zil_close(zilog); 3565 dmu_objset_disown(os, FTAG); 3566 ztest_zd_fini(&zdtmp); 3567 3568 rw_exit(&ztest_name_lock); 3569} 3570 3571/* 3572 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. 3573 */ 3574void 3575ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id) 3576{ 3577 rw_enter(&ztest_name_lock, RW_READER); 3578 (void) ztest_snapshot_destroy(zd->zd_name, id); 3579 (void) ztest_snapshot_create(zd->zd_name, id); 3580 rw_exit(&ztest_name_lock); 3581} 3582 3583/* 3584 * Cleanup non-standard snapshots and clones. 3585 */ 3586void 3587ztest_dsl_dataset_cleanup(char *osname, uint64_t id) 3588{ 3589 char snap1name[ZFS_MAX_DATASET_NAME_LEN]; 3590 char clone1name[ZFS_MAX_DATASET_NAME_LEN]; 3591 char snap2name[ZFS_MAX_DATASET_NAME_LEN]; 3592 char clone2name[ZFS_MAX_DATASET_NAME_LEN]; 3593 char snap3name[ZFS_MAX_DATASET_NAME_LEN]; 3594 int error; 3595 3596 (void) snprintf(snap1name, sizeof (snap1name), 3597 "%s@s1_%llu", osname, id); 3598 (void) snprintf(clone1name, sizeof (clone1name), 3599 "%s/c1_%llu", osname, id); 3600 (void) snprintf(snap2name, sizeof (snap2name), 3601 "%s@s2_%llu", clone1name, id); 3602 (void) snprintf(clone2name, sizeof (clone2name), 3603 "%s/c2_%llu", osname, id); 3604 (void) snprintf(snap3name, sizeof (snap3name), 3605 "%s@s3_%llu", clone1name, id); 3606 3607 error = dsl_destroy_head(clone2name); 3608 if (error && error != ENOENT) 3609 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error); 3610 error = dsl_destroy_snapshot(snap3name, B_FALSE); 3611 if (error && error != ENOENT) 3612 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error); 3613 error = dsl_destroy_snapshot(snap2name, B_FALSE); 3614 if (error && error != ENOENT) 3615 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error); 3616 error = dsl_destroy_head(clone1name); 3617 if (error && error != ENOENT) 3618 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error); 3619 error = dsl_destroy_snapshot(snap1name, B_FALSE); 3620 if (error && error != ENOENT) 3621 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error); 3622} 3623 3624/* 3625 * Verify dsl_dataset_promote handles EBUSY 3626 */ 3627void 3628ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) 3629{ 3630 objset_t *os; 3631 char snap1name[ZFS_MAX_DATASET_NAME_LEN]; 3632 char clone1name[ZFS_MAX_DATASET_NAME_LEN]; 3633 char snap2name[ZFS_MAX_DATASET_NAME_LEN]; 3634 char clone2name[ZFS_MAX_DATASET_NAME_LEN]; 3635 char snap3name[ZFS_MAX_DATASET_NAME_LEN]; 3636 char *osname = zd->zd_name; 3637 int error; 3638 3639 rw_enter(&ztest_name_lock, RW_READER); 3640 3641 ztest_dsl_dataset_cleanup(osname, id); 3642 3643 (void) snprintf(snap1name, sizeof (snap1name), 3644 "%s@s1_%llu", osname, id); 3645 (void) snprintf(clone1name, sizeof (clone1name), 3646 "%s/c1_%llu", osname, id); 3647 (void) snprintf(snap2name, sizeof (snap2name), 3648 "%s@s2_%llu", clone1name, id); 3649 (void) snprintf(clone2name, sizeof (clone2name), 3650 "%s/c2_%llu", osname, id); 3651 (void) snprintf(snap3name, sizeof (snap3name), 3652 "%s@s3_%llu", clone1name, id); 3653 3654 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1); 3655 if (error && error != EEXIST) { 3656 if (error == ENOSPC) { 3657 ztest_record_enospc(FTAG); 3658 goto out; 3659 } 3660 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error); 3661 } 3662 3663 error = dmu_objset_clone(clone1name, snap1name); 3664 if (error) { 3665 if (error == ENOSPC) { 3666 ztest_record_enospc(FTAG); 3667 goto out; 3668 } 3669 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error); 3670 } 3671 3672 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1); 3673 if (error && error != EEXIST) { 3674 if (error == ENOSPC) { 3675 ztest_record_enospc(FTAG); 3676 goto out; 3677 } 3678 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error); 3679 } 3680 3681 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1); 3682 if (error && error != EEXIST) { 3683 if (error == ENOSPC) { 3684 ztest_record_enospc(FTAG); 3685 goto out; 3686 } 3687 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); 3688 } 3689 3690 error = dmu_objset_clone(clone2name, snap3name); 3691 if (error) { 3692 if (error == ENOSPC) { 3693 ztest_record_enospc(FTAG); 3694 goto out; 3695 } 3696 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error); 3697 } 3698 3699 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os); 3700 if (error) 3701 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error); 3702 error = dsl_dataset_promote(clone2name, NULL); 3703 if (error == ENOSPC) { 3704 dmu_objset_disown(os, FTAG); 3705 ztest_record_enospc(FTAG); 3706 goto out; 3707 } 3708 if (error != EBUSY) 3709 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name, 3710 error); 3711 dmu_objset_disown(os, FTAG); 3712 3713out: 3714 ztest_dsl_dataset_cleanup(osname, id); 3715 3716 rw_exit(&ztest_name_lock); 3717} 3718 3719/* 3720 * Verify that dmu_object_{alloc,free} work as expected. 3721 */ 3722void 3723ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id) 3724{ 3725 ztest_od_t od[4]; 3726 int batchsize = sizeof (od) / sizeof (od[0]); 3727 3728 for (int b = 0; b < batchsize; b++) 3729 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0); 3730 3731 /* 3732 * Destroy the previous batch of objects, create a new batch, 3733 * and do some I/O on the new objects. 3734 */ 3735 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0) 3736 return; 3737 3738 while (ztest_random(4 * batchsize) != 0) 3739 ztest_io(zd, od[ztest_random(batchsize)].od_object, 3740 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 3741} 3742 3743/* 3744 * Verify that dmu_{read,write} work as expected. 3745 */ 3746void 3747ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id) 3748{ 3749 objset_t *os = zd->zd_os; 3750 ztest_od_t od[2]; 3751 dmu_tx_t *tx; 3752 int i, freeit, error; 3753 uint64_t n, s, txg; 3754 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; 3755 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; 3756 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t); 3757 uint64_t regions = 997; 3758 uint64_t stride = 123456789ULL; 3759 uint64_t width = 40; 3760 int free_percent = 5; 3761 3762 /* 3763 * This test uses two objects, packobj and bigobj, that are always 3764 * updated together (i.e. in the same tx) so that their contents are 3765 * in sync and can be compared. Their contents relate to each other 3766 * in a simple way: packobj is a dense array of 'bufwad' structures, 3767 * while bigobj is a sparse array of the same bufwads. Specifically, 3768 * for any index n, there are three bufwads that should be identical: 3769 * 3770 * packobj, at offset n * sizeof (bufwad_t) 3771 * bigobj, at the head of the nth chunk 3772 * bigobj, at the tail of the nth chunk 3773 * 3774 * The chunk size is arbitrary. It doesn't have to be a power of two, 3775 * and it doesn't have any relation to the object blocksize. 3776 * The only requirement is that it can hold at least two bufwads. 3777 * 3778 * Normally, we write the bufwad to each of these locations. 3779 * However, free_percent of the time we instead write zeroes to 3780 * packobj and perform a dmu_free_range() on bigobj. By comparing 3781 * bigobj to packobj, we can verify that the DMU is correctly 3782 * tracking which parts of an object are allocated and free, 3783 * and that the contents of the allocated blocks are correct. 3784 */ 3785 3786 /* 3787 * Read the directory info. If it's the first time, set things up. 3788 */ 3789 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize); 3790 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); 3791 3792 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 3793 return; 3794 3795 bigobj = od[0].od_object; 3796 packobj = od[1].od_object; 3797 chunksize = od[0].od_gen; 3798 ASSERT(chunksize == od[1].od_gen); 3799 3800 /* 3801 * Prefetch a random chunk of the big object. 3802 * Our aim here is to get some async reads in flight 3803 * for blocks that we may free below; the DMU should 3804 * handle this race correctly. 3805 */ 3806 n = ztest_random(regions) * stride + ztest_random(width); 3807 s = 1 + ztest_random(2 * width - 1); 3808 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize, 3809 ZIO_PRIORITY_SYNC_READ); 3810 3811 /* 3812 * Pick a random index and compute the offsets into packobj and bigobj. 3813 */ 3814 n = ztest_random(regions) * stride + ztest_random(width); 3815 s = 1 + ztest_random(width - 1); 3816 3817 packoff = n * sizeof (bufwad_t); 3818 packsize = s * sizeof (bufwad_t); 3819 3820 bigoff = n * chunksize; 3821 bigsize = s * chunksize; 3822 3823 packbuf = umem_alloc(packsize, UMEM_NOFAIL); 3824 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); 3825 3826 /* 3827 * free_percent of the time, free a range of bigobj rather than 3828 * overwriting it. 3829 */ 3830 freeit = (ztest_random(100) < free_percent); 3831 3832 /* 3833 * Read the current contents of our objects. 3834 */ 3835 error = dmu_read(os, packobj, packoff, packsize, packbuf, 3836 DMU_READ_PREFETCH); 3837 ASSERT0(error); 3838 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf, 3839 DMU_READ_PREFETCH); 3840 ASSERT0(error); 3841 3842 /* 3843 * Get a tx for the mods to both packobj and bigobj. 3844 */ 3845 tx = dmu_tx_create(os); 3846 3847 dmu_tx_hold_write(tx, packobj, packoff, packsize); 3848 3849 if (freeit) 3850 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize); 3851 else 3852 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); 3853 3854 /* This accounts for setting the checksum/compression. */ 3855 dmu_tx_hold_bonus(tx, bigobj); 3856 3857 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 3858 if (txg == 0) { 3859 umem_free(packbuf, packsize); 3860 umem_free(bigbuf, bigsize); 3861 return; 3862 } 3863 3864 enum zio_checksum cksum; 3865 do { 3866 cksum = (enum zio_checksum) 3867 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM); 3868 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS); 3869 dmu_object_set_checksum(os, bigobj, cksum, tx); 3870 3871 enum zio_compress comp; 3872 do { 3873 comp = (enum zio_compress) 3874 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION); 3875 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS); 3876 dmu_object_set_compress(os, bigobj, comp, tx); 3877 3878 /* 3879 * For each index from n to n + s, verify that the existing bufwad 3880 * in packobj matches the bufwads at the head and tail of the 3881 * corresponding chunk in bigobj. Then update all three bufwads 3882 * with the new values we want to write out. 3883 */ 3884 for (i = 0; i < s; i++) { 3885 /* LINTED */ 3886 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 3887 /* LINTED */ 3888 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); 3889 /* LINTED */ 3890 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; 3891 3892 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 3893 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 3894 3895 if (pack->bw_txg > txg) 3896 fatal(0, "future leak: got %llx, open txg is %llx", 3897 pack->bw_txg, txg); 3898 3899 if (pack->bw_data != 0 && pack->bw_index != n + i) 3900 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 3901 pack->bw_index, n, i); 3902 3903 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 3904 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 3905 3906 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 3907 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 3908 3909 if (freeit) { 3910 bzero(pack, sizeof (bufwad_t)); 3911 } else { 3912 pack->bw_index = n + i; 3913 pack->bw_txg = txg; 3914 pack->bw_data = 1 + ztest_random(-2ULL); 3915 } 3916 *bigH = *pack; 3917 *bigT = *pack; 3918 } 3919 3920 /* 3921 * We've verified all the old bufwads, and made new ones. 3922 * Now write them out. 3923 */ 3924 dmu_write(os, packobj, packoff, packsize, packbuf, tx); 3925 3926 if (freeit) { 3927 if (ztest_opts.zo_verbose >= 7) { 3928 (void) printf("freeing offset %llx size %llx" 3929 " txg %llx\n", 3930 (u_longlong_t)bigoff, 3931 (u_longlong_t)bigsize, 3932 (u_longlong_t)txg); 3933 } 3934 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx)); 3935 } else { 3936 if (ztest_opts.zo_verbose >= 7) { 3937 (void) printf("writing offset %llx size %llx" 3938 " txg %llx\n", 3939 (u_longlong_t)bigoff, 3940 (u_longlong_t)bigsize, 3941 (u_longlong_t)txg); 3942 } 3943 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx); 3944 } 3945 3946 dmu_tx_commit(tx); 3947 3948 /* 3949 * Sanity check the stuff we just wrote. 3950 */ 3951 { 3952 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 3953 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 3954 3955 VERIFY(0 == dmu_read(os, packobj, packoff, 3956 packsize, packcheck, DMU_READ_PREFETCH)); 3957 VERIFY(0 == dmu_read(os, bigobj, bigoff, 3958 bigsize, bigcheck, DMU_READ_PREFETCH)); 3959 3960 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 3961 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 3962 3963 umem_free(packcheck, packsize); 3964 umem_free(bigcheck, bigsize); 3965 } 3966 3967 umem_free(packbuf, packsize); 3968 umem_free(bigbuf, bigsize); 3969} 3970 3971void 3972compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf, 3973 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg) 3974{ 3975 uint64_t i; 3976 bufwad_t *pack; 3977 bufwad_t *bigH; 3978 bufwad_t *bigT; 3979 3980 /* 3981 * For each index from n to n + s, verify that the existing bufwad 3982 * in packobj matches the bufwads at the head and tail of the 3983 * corresponding chunk in bigobj. Then update all three bufwads 3984 * with the new values we want to write out. 3985 */ 3986 for (i = 0; i < s; i++) { 3987 /* LINTED */ 3988 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 3989 /* LINTED */ 3990 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); 3991 /* LINTED */ 3992 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; 3993 3994 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 3995 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 3996 3997 if (pack->bw_txg > txg) 3998 fatal(0, "future leak: got %llx, open txg is %llx", 3999 pack->bw_txg, txg); 4000 4001 if (pack->bw_data != 0 && pack->bw_index != n + i) 4002 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 4003 pack->bw_index, n, i); 4004 4005 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 4006 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 4007 4008 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 4009 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 4010 4011 pack->bw_index = n + i; 4012 pack->bw_txg = txg; 4013 pack->bw_data = 1 + ztest_random(-2ULL); 4014 4015 *bigH = *pack; 4016 *bigT = *pack; 4017 } 4018} 4019 4020void 4021ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id) 4022{ 4023 objset_t *os = zd->zd_os; 4024 ztest_od_t od[2]; 4025 dmu_tx_t *tx; 4026 uint64_t i; 4027 int error; 4028 uint64_t n, s, txg; 4029 bufwad_t *packbuf, *bigbuf; 4030 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; 4031 uint64_t blocksize = ztest_random_blocksize(); 4032 uint64_t chunksize = blocksize; 4033 uint64_t regions = 997; 4034 uint64_t stride = 123456789ULL; 4035 uint64_t width = 9; 4036 dmu_buf_t *bonus_db; 4037 arc_buf_t **bigbuf_arcbufs; 4038 dmu_object_info_t doi; 4039 4040 /* 4041 * This test uses two objects, packobj and bigobj, that are always 4042 * updated together (i.e. in the same tx) so that their contents are 4043 * in sync and can be compared. Their contents relate to each other 4044 * in a simple way: packobj is a dense array of 'bufwad' structures, 4045 * while bigobj is a sparse array of the same bufwads. Specifically, 4046 * for any index n, there are three bufwads that should be identical: 4047 * 4048 * packobj, at offset n * sizeof (bufwad_t) 4049 * bigobj, at the head of the nth chunk 4050 * bigobj, at the tail of the nth chunk 4051 * 4052 * The chunk size is set equal to bigobj block size so that 4053 * dmu_assign_arcbuf() can be tested for object updates. 4054 */ 4055 4056 /* 4057 * Read the directory info. If it's the first time, set things up. 4058 */ 4059 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 4060 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); 4061 4062 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4063 return; 4064 4065 bigobj = od[0].od_object; 4066 packobj = od[1].od_object; 4067 blocksize = od[0].od_blocksize; 4068 chunksize = blocksize; 4069 ASSERT(chunksize == od[1].od_gen); 4070 4071 VERIFY(dmu_object_info(os, bigobj, &doi) == 0); 4072 VERIFY(ISP2(doi.doi_data_block_size)); 4073 VERIFY(chunksize == doi.doi_data_block_size); 4074 VERIFY(chunksize >= 2 * sizeof (bufwad_t)); 4075 4076 /* 4077 * Pick a random index and compute the offsets into packobj and bigobj. 4078 */ 4079 n = ztest_random(regions) * stride + ztest_random(width); 4080 s = 1 + ztest_random(width - 1); 4081 4082 packoff = n * sizeof (bufwad_t); 4083 packsize = s * sizeof (bufwad_t); 4084 4085 bigoff = n * chunksize; 4086 bigsize = s * chunksize; 4087 4088 packbuf = umem_zalloc(packsize, UMEM_NOFAIL); 4089 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL); 4090 4091 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db)); 4092 4093 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL); 4094 4095 /* 4096 * Iteration 0 test zcopy for DB_UNCACHED dbufs. 4097 * Iteration 1 test zcopy to already referenced dbufs. 4098 * Iteration 2 test zcopy to dirty dbuf in the same txg. 4099 * Iteration 3 test zcopy to dbuf dirty in previous txg. 4100 * Iteration 4 test zcopy when dbuf is no longer dirty. 4101 * Iteration 5 test zcopy when it can't be done. 4102 * Iteration 6 one more zcopy write. 4103 */ 4104 for (i = 0; i < 7; i++) { 4105 uint64_t j; 4106 uint64_t off; 4107 4108 /* 4109 * In iteration 5 (i == 5) use arcbufs 4110 * that don't match bigobj blksz to test 4111 * dmu_assign_arcbuf() when it can't directly 4112 * assign an arcbuf to a dbuf. 4113 */ 4114 for (j = 0; j < s; j++) { 4115 if (i != 5) { 4116 bigbuf_arcbufs[j] = 4117 dmu_request_arcbuf(bonus_db, chunksize); 4118 } else { 4119 bigbuf_arcbufs[2 * j] = 4120 dmu_request_arcbuf(bonus_db, chunksize / 2); 4121 bigbuf_arcbufs[2 * j + 1] = 4122 dmu_request_arcbuf(bonus_db, chunksize / 2); 4123 } 4124 } 4125 4126 /* 4127 * Get a tx for the mods to both packobj and bigobj. 4128 */ 4129 tx = dmu_tx_create(os); 4130 4131 dmu_tx_hold_write(tx, packobj, packoff, packsize); 4132 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); 4133 4134 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4135 if (txg == 0) { 4136 umem_free(packbuf, packsize); 4137 umem_free(bigbuf, bigsize); 4138 for (j = 0; j < s; j++) { 4139 if (i != 5) { 4140 dmu_return_arcbuf(bigbuf_arcbufs[j]); 4141 } else { 4142 dmu_return_arcbuf( 4143 bigbuf_arcbufs[2 * j]); 4144 dmu_return_arcbuf( 4145 bigbuf_arcbufs[2 * j + 1]); 4146 } 4147 } 4148 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); 4149 dmu_buf_rele(bonus_db, FTAG); 4150 return; 4151 } 4152 4153 /* 4154 * 50% of the time don't read objects in the 1st iteration to 4155 * test dmu_assign_arcbuf() for the case when there're no 4156 * existing dbufs for the specified offsets. 4157 */ 4158 if (i != 0 || ztest_random(2) != 0) { 4159 error = dmu_read(os, packobj, packoff, 4160 packsize, packbuf, DMU_READ_PREFETCH); 4161 ASSERT0(error); 4162 error = dmu_read(os, bigobj, bigoff, bigsize, 4163 bigbuf, DMU_READ_PREFETCH); 4164 ASSERT0(error); 4165 } 4166 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize, 4167 n, chunksize, txg); 4168 4169 /* 4170 * We've verified all the old bufwads, and made new ones. 4171 * Now write them out. 4172 */ 4173 dmu_write(os, packobj, packoff, packsize, packbuf, tx); 4174 if (ztest_opts.zo_verbose >= 7) { 4175 (void) printf("writing offset %llx size %llx" 4176 " txg %llx\n", 4177 (u_longlong_t)bigoff, 4178 (u_longlong_t)bigsize, 4179 (u_longlong_t)txg); 4180 } 4181 for (off = bigoff, j = 0; j < s; j++, off += chunksize) { 4182 dmu_buf_t *dbt; 4183 if (i != 5) { 4184 bcopy((caddr_t)bigbuf + (off - bigoff), 4185 bigbuf_arcbufs[j]->b_data, chunksize); 4186 } else { 4187 bcopy((caddr_t)bigbuf + (off - bigoff), 4188 bigbuf_arcbufs[2 * j]->b_data, 4189 chunksize / 2); 4190 bcopy((caddr_t)bigbuf + (off - bigoff) + 4191 chunksize / 2, 4192 bigbuf_arcbufs[2 * j + 1]->b_data, 4193 chunksize / 2); 4194 } 4195 4196 if (i == 1) { 4197 VERIFY(dmu_buf_hold(os, bigobj, off, 4198 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0); 4199 } 4200 if (i != 5) { 4201 dmu_assign_arcbuf(bonus_db, off, 4202 bigbuf_arcbufs[j], tx); 4203 } else { 4204 dmu_assign_arcbuf(bonus_db, off, 4205 bigbuf_arcbufs[2 * j], tx); 4206 dmu_assign_arcbuf(bonus_db, 4207 off + chunksize / 2, 4208 bigbuf_arcbufs[2 * j + 1], tx); 4209 } 4210 if (i == 1) { 4211 dmu_buf_rele(dbt, FTAG); 4212 } 4213 } 4214 dmu_tx_commit(tx); 4215 4216 /* 4217 * Sanity check the stuff we just wrote. 4218 */ 4219 { 4220 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 4221 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 4222 4223 VERIFY(0 == dmu_read(os, packobj, packoff, 4224 packsize, packcheck, DMU_READ_PREFETCH)); 4225 VERIFY(0 == dmu_read(os, bigobj, bigoff, 4226 bigsize, bigcheck, DMU_READ_PREFETCH)); 4227 4228 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 4229 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 4230 4231 umem_free(packcheck, packsize); 4232 umem_free(bigcheck, bigsize); 4233 } 4234 if (i == 2) { 4235 txg_wait_open(dmu_objset_pool(os), 0); 4236 } else if (i == 3) { 4237 txg_wait_synced(dmu_objset_pool(os), 0); 4238 } 4239 } 4240 4241 dmu_buf_rele(bonus_db, FTAG); 4242 umem_free(packbuf, packsize); 4243 umem_free(bigbuf, bigsize); 4244 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); 4245} 4246 4247/* ARGSUSED */ 4248void 4249ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id) 4250{ 4251 ztest_od_t od[1]; 4252 uint64_t offset = (1ULL << (ztest_random(20) + 43)) + 4253 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 4254 4255 /* 4256 * Have multiple threads write to large offsets in an object 4257 * to verify that parallel writes to an object -- even to the 4258 * same blocks within the object -- doesn't cause any trouble. 4259 */ 4260 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 4261 4262 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4263 return; 4264 4265 while (ztest_random(10) != 0) 4266 ztest_io(zd, od[0].od_object, offset); 4267} 4268 4269void 4270ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id) 4271{ 4272 ztest_od_t od[1]; 4273 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) + 4274 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 4275 uint64_t count = ztest_random(20) + 1; 4276 uint64_t blocksize = ztest_random_blocksize(); 4277 void *data; 4278 4279 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 4280 4281 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4282 return; 4283 4284 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0) 4285 return; 4286 4287 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize); 4288 4289 data = umem_zalloc(blocksize, UMEM_NOFAIL); 4290 4291 while (ztest_random(count) != 0) { 4292 uint64_t randoff = offset + (ztest_random(count) * blocksize); 4293 if (ztest_write(zd, od[0].od_object, randoff, blocksize, 4294 data) != 0) 4295 break; 4296 while (ztest_random(4) != 0) 4297 ztest_io(zd, od[0].od_object, randoff); 4298 } 4299 4300 umem_free(data, blocksize); 4301} 4302 4303/* 4304 * Verify that zap_{create,destroy,add,remove,update} work as expected. 4305 */ 4306#define ZTEST_ZAP_MIN_INTS 1 4307#define ZTEST_ZAP_MAX_INTS 4 4308#define ZTEST_ZAP_MAX_PROPS 1000 4309 4310void 4311ztest_zap(ztest_ds_t *zd, uint64_t id) 4312{ 4313 objset_t *os = zd->zd_os; 4314 ztest_od_t od[1]; 4315 uint64_t object; 4316 uint64_t txg, last_txg; 4317 uint64_t value[ZTEST_ZAP_MAX_INTS]; 4318 uint64_t zl_ints, zl_intsize, prop; 4319 int i, ints; 4320 dmu_tx_t *tx; 4321 char propname[100], txgname[100]; 4322 int error; 4323 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; 4324 4325 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); 4326 4327 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4328 return; 4329 4330 object = od[0].od_object; 4331 4332 /* 4333 * Generate a known hash collision, and verify that 4334 * we can lookup and remove both entries. 4335 */ 4336 tx = dmu_tx_create(os); 4337 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4338 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4339 if (txg == 0) 4340 return; 4341 for (i = 0; i < 2; i++) { 4342 value[i] = i; 4343 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t), 4344 1, &value[i], tx)); 4345 } 4346 for (i = 0; i < 2; i++) { 4347 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i], 4348 sizeof (uint64_t), 1, &value[i], tx)); 4349 VERIFY3U(0, ==, 4350 zap_length(os, object, hc[i], &zl_intsize, &zl_ints)); 4351 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4352 ASSERT3U(zl_ints, ==, 1); 4353 } 4354 for (i = 0; i < 2; i++) { 4355 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx)); 4356 } 4357 dmu_tx_commit(tx); 4358 4359 /* 4360 * Generate a buch of random entries. 4361 */ 4362 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); 4363 4364 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 4365 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 4366 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 4367 bzero(value, sizeof (value)); 4368 last_txg = 0; 4369 4370 /* 4371 * If these zap entries already exist, validate their contents. 4372 */ 4373 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 4374 if (error == 0) { 4375 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4376 ASSERT3U(zl_ints, ==, 1); 4377 4378 VERIFY(zap_lookup(os, object, txgname, zl_intsize, 4379 zl_ints, &last_txg) == 0); 4380 4381 VERIFY(zap_length(os, object, propname, &zl_intsize, 4382 &zl_ints) == 0); 4383 4384 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4385 ASSERT3U(zl_ints, ==, ints); 4386 4387 VERIFY(zap_lookup(os, object, propname, zl_intsize, 4388 zl_ints, value) == 0); 4389 4390 for (i = 0; i < ints; i++) { 4391 ASSERT3U(value[i], ==, last_txg + object + i); 4392 } 4393 } else { 4394 ASSERT3U(error, ==, ENOENT); 4395 } 4396 4397 /* 4398 * Atomically update two entries in our zap object. 4399 * The first is named txg_%llu, and contains the txg 4400 * in which the property was last updated. The second 4401 * is named prop_%llu, and the nth element of its value 4402 * should be txg + object + n. 4403 */ 4404 tx = dmu_tx_create(os); 4405 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4406 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4407 if (txg == 0) 4408 return; 4409 4410 if (last_txg > txg) 4411 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); 4412 4413 for (i = 0; i < ints; i++) 4414 value[i] = txg + object + i; 4415 4416 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t), 4417 1, &txg, tx)); 4418 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t), 4419 ints, value, tx)); 4420 4421 dmu_tx_commit(tx); 4422 4423 /* 4424 * Remove a random pair of entries. 4425 */ 4426 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 4427 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 4428 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 4429 4430 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 4431 4432 if (error == ENOENT) 4433 return; 4434 4435 ASSERT0(error); 4436 4437 tx = dmu_tx_create(os); 4438 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4439 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4440 if (txg == 0) 4441 return; 4442 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx)); 4443 VERIFY3U(0, ==, zap_remove(os, object, propname, tx)); 4444 dmu_tx_commit(tx); 4445} 4446 4447/* 4448 * Testcase to test the upgrading of a microzap to fatzap. 4449 */ 4450void 4451ztest_fzap(ztest_ds_t *zd, uint64_t id) 4452{ 4453 objset_t *os = zd->zd_os; 4454 ztest_od_t od[1]; 4455 uint64_t object, txg; 4456 4457 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); 4458 4459 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4460 return; 4461 4462 object = od[0].od_object; 4463 4464 /* 4465 * Add entries to this ZAP and make sure it spills over 4466 * and gets upgraded to a fatzap. Also, since we are adding 4467 * 2050 entries we should see ptrtbl growth and leaf-block split. 4468 */ 4469 for (int i = 0; i < 2050; i++) { 4470 char name[ZFS_MAX_DATASET_NAME_LEN]; 4471 uint64_t value = i; 4472 dmu_tx_t *tx; 4473 int error; 4474 4475 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu", 4476 id, value); 4477 4478 tx = dmu_tx_create(os); 4479 dmu_tx_hold_zap(tx, object, B_TRUE, name); 4480 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4481 if (txg == 0) 4482 return; 4483 error = zap_add(os, object, name, sizeof (uint64_t), 1, 4484 &value, tx); 4485 ASSERT(error == 0 || error == EEXIST); 4486 dmu_tx_commit(tx); 4487 } 4488} 4489 4490/* ARGSUSED */ 4491void 4492ztest_zap_parallel(ztest_ds_t *zd, uint64_t id) 4493{ 4494 objset_t *os = zd->zd_os; 4495 ztest_od_t od[1]; 4496 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; 4497 dmu_tx_t *tx; 4498 int i, namelen, error; 4499 int micro = ztest_random(2); 4500 char name[20], string_value[20]; 4501 void *data; 4502 4503 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0); 4504 4505 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4506 return; 4507 4508 object = od[0].od_object; 4509 4510 /* 4511 * Generate a random name of the form 'xxx.....' where each 4512 * x is a random printable character and the dots are dots. 4513 * There are 94 such characters, and the name length goes from 4514 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. 4515 */ 4516 namelen = ztest_random(sizeof (name) - 5) + 5 + 1; 4517 4518 for (i = 0; i < 3; i++) 4519 name[i] = '!' + ztest_random('~' - '!' + 1); 4520 for (; i < namelen - 1; i++) 4521 name[i] = '.'; 4522 name[i] = '\0'; 4523 4524 if ((namelen & 1) || micro) { 4525 wsize = sizeof (txg); 4526 wc = 1; 4527 data = &txg; 4528 } else { 4529 wsize = 1; 4530 wc = namelen; 4531 data = string_value; 4532 } 4533 4534 count = -1ULL; 4535 VERIFY0(zap_count(os, object, &count)); 4536 ASSERT(count != -1ULL); 4537 4538 /* 4539 * Select an operation: length, lookup, add, update, remove. 4540 */ 4541 i = ztest_random(5); 4542 4543 if (i >= 2) { 4544 tx = dmu_tx_create(os); 4545 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4546 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4547 if (txg == 0) 4548 return; 4549 bcopy(name, string_value, namelen); 4550 } else { 4551 tx = NULL; 4552 txg = 0; 4553 bzero(string_value, namelen); 4554 } 4555 4556 switch (i) { 4557 4558 case 0: 4559 error = zap_length(os, object, name, &zl_wsize, &zl_wc); 4560 if (error == 0) { 4561 ASSERT3U(wsize, ==, zl_wsize); 4562 ASSERT3U(wc, ==, zl_wc); 4563 } else { 4564 ASSERT3U(error, ==, ENOENT); 4565 } 4566 break; 4567 4568 case 1: 4569 error = zap_lookup(os, object, name, wsize, wc, data); 4570 if (error == 0) { 4571 if (data == string_value && 4572 bcmp(name, data, namelen) != 0) 4573 fatal(0, "name '%s' != val '%s' len %d", 4574 name, data, namelen); 4575 } else { 4576 ASSERT3U(error, ==, ENOENT); 4577 } 4578 break; 4579 4580 case 2: 4581 error = zap_add(os, object, name, wsize, wc, data, tx); 4582 ASSERT(error == 0 || error == EEXIST); 4583 break; 4584 4585 case 3: 4586 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); 4587 break; 4588 4589 case 4: 4590 error = zap_remove(os, object, name, tx); 4591 ASSERT(error == 0 || error == ENOENT); 4592 break; 4593 } 4594 4595 if (tx != NULL) 4596 dmu_tx_commit(tx); 4597} 4598 4599/* 4600 * Commit callback data. 4601 */ 4602typedef struct ztest_cb_data { 4603 list_node_t zcd_node; 4604 uint64_t zcd_txg; 4605 int zcd_expected_err; 4606 boolean_t zcd_added; 4607 boolean_t zcd_called; 4608 spa_t *zcd_spa; 4609} ztest_cb_data_t; 4610 4611/* This is the actual commit callback function */ 4612static void 4613ztest_commit_callback(void *arg, int error) 4614{ 4615 ztest_cb_data_t *data = arg; 4616 uint64_t synced_txg; 4617 4618 VERIFY(data != NULL); 4619 VERIFY3S(data->zcd_expected_err, ==, error); 4620 VERIFY(!data->zcd_called); 4621 4622 synced_txg = spa_last_synced_txg(data->zcd_spa); 4623 if (data->zcd_txg > synced_txg) 4624 fatal(0, "commit callback of txg %" PRIu64 " called prematurely" 4625 ", last synced txg = %" PRIu64 "\n", data->zcd_txg, 4626 synced_txg); 4627 4628 data->zcd_called = B_TRUE; 4629 4630 if (error == ECANCELED) { 4631 ASSERT0(data->zcd_txg); 4632 ASSERT(!data->zcd_added); 4633 4634 /* 4635 * The private callback data should be destroyed here, but 4636 * since we are going to check the zcd_called field after 4637 * dmu_tx_abort(), we will destroy it there. 4638 */ 4639 return; 4640 } 4641 4642 /* Was this callback added to the global callback list? */ 4643 if (!data->zcd_added) 4644 goto out; 4645 4646 ASSERT3U(data->zcd_txg, !=, 0); 4647 4648 /* Remove our callback from the list */ 4649 mutex_enter(&zcl.zcl_callbacks_lock); 4650 list_remove(&zcl.zcl_callbacks, data); 4651 mutex_exit(&zcl.zcl_callbacks_lock); 4652 4653out: 4654 umem_free(data, sizeof (ztest_cb_data_t)); 4655} 4656 4657/* Allocate and initialize callback data structure */ 4658static ztest_cb_data_t * 4659ztest_create_cb_data(objset_t *os, uint64_t txg) 4660{ 4661 ztest_cb_data_t *cb_data; 4662 4663 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL); 4664 4665 cb_data->zcd_txg = txg; 4666 cb_data->zcd_spa = dmu_objset_spa(os); 4667 4668 return (cb_data); 4669} 4670 4671/* 4672 * If a number of txgs equal to this threshold have been created after a commit 4673 * callback has been registered but not called, then we assume there is an 4674 * implementation bug. 4675 */ 4676#define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2) 4677 4678/* 4679 * Commit callback test. 4680 */ 4681void 4682ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id) 4683{ 4684 objset_t *os = zd->zd_os; 4685 ztest_od_t od[1]; 4686 dmu_tx_t *tx; 4687 ztest_cb_data_t *cb_data[3], *tmp_cb; 4688 uint64_t old_txg, txg; 4689 int i, error; 4690 4691 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 4692 4693 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4694 return; 4695 4696 tx = dmu_tx_create(os); 4697 4698 cb_data[0] = ztest_create_cb_data(os, 0); 4699 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]); 4700 4701 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t)); 4702 4703 /* Every once in a while, abort the transaction on purpose */ 4704 if (ztest_random(100) == 0) 4705 error = -1; 4706 4707 if (!error) 4708 error = dmu_tx_assign(tx, TXG_NOWAIT); 4709 4710 txg = error ? 0 : dmu_tx_get_txg(tx); 4711 4712 cb_data[0]->zcd_txg = txg; 4713 cb_data[1] = ztest_create_cb_data(os, txg); 4714 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]); 4715 4716 if (error) { 4717 /* 4718 * It's not a strict requirement to call the registered 4719 * callbacks from inside dmu_tx_abort(), but that's what 4720 * it's supposed to happen in the current implementation 4721 * so we will check for that. 4722 */ 4723 for (i = 0; i < 2; i++) { 4724 cb_data[i]->zcd_expected_err = ECANCELED; 4725 VERIFY(!cb_data[i]->zcd_called); 4726 } 4727 4728 dmu_tx_abort(tx); 4729 4730 for (i = 0; i < 2; i++) { 4731 VERIFY(cb_data[i]->zcd_called); 4732 umem_free(cb_data[i], sizeof (ztest_cb_data_t)); 4733 } 4734 4735 return; 4736 } 4737 4738 cb_data[2] = ztest_create_cb_data(os, txg); 4739 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]); 4740 4741 /* 4742 * Read existing data to make sure there isn't a future leak. 4743 */ 4744 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t), 4745 &old_txg, DMU_READ_PREFETCH)); 4746 4747 if (old_txg > txg) 4748 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64, 4749 old_txg, txg); 4750 4751 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx); 4752 4753 mutex_enter(&zcl.zcl_callbacks_lock); 4754 4755 /* 4756 * Since commit callbacks don't have any ordering requirement and since 4757 * it is theoretically possible for a commit callback to be called 4758 * after an arbitrary amount of time has elapsed since its txg has been 4759 * synced, it is difficult to reliably determine whether a commit 4760 * callback hasn't been called due to high load or due to a flawed 4761 * implementation. 4762 * 4763 * In practice, we will assume that if after a certain number of txgs a 4764 * commit callback hasn't been called, then most likely there's an 4765 * implementation bug.. 4766 */ 4767 tmp_cb = list_head(&zcl.zcl_callbacks); 4768 if (tmp_cb != NULL && 4769 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) { 4770 fatal(0, "Commit callback threshold exceeded, oldest txg: %" 4771 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg); 4772 } 4773 4774 /* 4775 * Let's find the place to insert our callbacks. 4776 * 4777 * Even though the list is ordered by txg, it is possible for the 4778 * insertion point to not be the end because our txg may already be 4779 * quiescing at this point and other callbacks in the open txg 4780 * (from other objsets) may have sneaked in. 4781 */ 4782 tmp_cb = list_tail(&zcl.zcl_callbacks); 4783 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg) 4784 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb); 4785 4786 /* Add the 3 callbacks to the list */ 4787 for (i = 0; i < 3; i++) { 4788 if (tmp_cb == NULL) 4789 list_insert_head(&zcl.zcl_callbacks, cb_data[i]); 4790 else 4791 list_insert_after(&zcl.zcl_callbacks, tmp_cb, 4792 cb_data[i]); 4793 4794 cb_data[i]->zcd_added = B_TRUE; 4795 VERIFY(!cb_data[i]->zcd_called); 4796 4797 tmp_cb = cb_data[i]; 4798 } 4799 4800 mutex_exit(&zcl.zcl_callbacks_lock); 4801 4802 dmu_tx_commit(tx); 4803} 4804 4805/* ARGSUSED */ 4806void 4807ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id) 4808{ 4809 zfs_prop_t proplist[] = { 4810 ZFS_PROP_CHECKSUM, 4811 ZFS_PROP_COMPRESSION, 4812 ZFS_PROP_COPIES, 4813 ZFS_PROP_DEDUP 4814 }; 4815 4816 rw_enter(&ztest_name_lock, RW_READER); 4817 4818 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) 4819 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p], 4820 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2)); 4821 4822 rw_exit(&ztest_name_lock); 4823} 4824 4825/* ARGSUSED */ 4826void 4827ztest_remap_blocks(ztest_ds_t *zd, uint64_t id) 4828{ 4829 rw_enter(&ztest_name_lock, RW_READER); 4830 4831 int error = dmu_objset_remap_indirects(zd->zd_name); 4832 if (error == ENOSPC) 4833 error = 0; 4834 ASSERT0(error); 4835 4836 rw_exit(&ztest_name_lock); 4837} 4838 4839/* ARGSUSED */ 4840void 4841ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id) 4842{ 4843 nvlist_t *props = NULL; 4844 4845 rw_enter(&ztest_name_lock, RW_READER); 4846 4847 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO, 4848 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN)); 4849 4850 VERIFY0(spa_prop_get(ztest_spa, &props)); 4851 4852 if (ztest_opts.zo_verbose >= 6) 4853 dump_nvlist(props, 4); 4854 4855 nvlist_free(props); 4856 4857 rw_exit(&ztest_name_lock); 4858} 4859 4860static int 4861user_release_one(const char *snapname, const char *holdname) 4862{ 4863 nvlist_t *snaps, *holds; 4864 int error; 4865 4866 snaps = fnvlist_alloc(); 4867 holds = fnvlist_alloc(); 4868 fnvlist_add_boolean(holds, holdname); 4869 fnvlist_add_nvlist(snaps, snapname, holds); 4870 fnvlist_free(holds); 4871 error = dsl_dataset_user_release(snaps, NULL); 4872 fnvlist_free(snaps); 4873 return (error); 4874} 4875 4876/* 4877 * Test snapshot hold/release and deferred destroy. 4878 */ 4879void 4880ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) 4881{ 4882 int error; 4883 objset_t *os = zd->zd_os; 4884 objset_t *origin; 4885 char snapname[100]; 4886 char fullname[100]; 4887 char clonename[100]; 4888 char tag[100]; 4889 char osname[ZFS_MAX_DATASET_NAME_LEN]; 4890 nvlist_t *holds; 4891 4892 rw_enter(&ztest_name_lock, RW_READER); 4893 4894 dmu_objset_name(os, osname); 4895 4896 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id); 4897 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname); 4898 (void) snprintf(clonename, sizeof (clonename), 4899 "%s/ch1_%llu", osname, id); 4900 (void) snprintf(tag, sizeof (tag), "tag_%llu", id); 4901 4902 /* 4903 * Clean up from any previous run. 4904 */ 4905 error = dsl_destroy_head(clonename); 4906 if (error != ENOENT) 4907 ASSERT0(error); 4908 error = user_release_one(fullname, tag); 4909 if (error != ESRCH && error != ENOENT) 4910 ASSERT0(error); 4911 error = dsl_destroy_snapshot(fullname, B_FALSE); 4912 if (error != ENOENT) 4913 ASSERT0(error); 4914 4915 /* 4916 * Create snapshot, clone it, mark snap for deferred destroy, 4917 * destroy clone, verify snap was also destroyed. 4918 */ 4919 error = dmu_objset_snapshot_one(osname, snapname); 4920 if (error) { 4921 if (error == ENOSPC) { 4922 ztest_record_enospc("dmu_objset_snapshot"); 4923 goto out; 4924 } 4925 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); 4926 } 4927 4928 error = dmu_objset_clone(clonename, fullname); 4929 if (error) { 4930 if (error == ENOSPC) { 4931 ztest_record_enospc("dmu_objset_clone"); 4932 goto out; 4933 } 4934 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error); 4935 } 4936 4937 error = dsl_destroy_snapshot(fullname, B_TRUE); 4938 if (error) { 4939 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", 4940 fullname, error); 4941 } 4942 4943 error = dsl_destroy_head(clonename); 4944 if (error) 4945 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error); 4946 4947 error = dmu_objset_hold(fullname, FTAG, &origin); 4948 if (error != ENOENT) 4949 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); 4950 4951 /* 4952 * Create snapshot, add temporary hold, verify that we can't 4953 * destroy a held snapshot, mark for deferred destroy, 4954 * release hold, verify snapshot was destroyed. 4955 */ 4956 error = dmu_objset_snapshot_one(osname, snapname); 4957 if (error) { 4958 if (error == ENOSPC) { 4959 ztest_record_enospc("dmu_objset_snapshot"); 4960 goto out; 4961 } 4962 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); 4963 } 4964 4965 holds = fnvlist_alloc(); 4966 fnvlist_add_string(holds, fullname, tag); 4967 error = dsl_dataset_user_hold(holds, 0, NULL); 4968 fnvlist_free(holds); 4969 4970 if (error == ENOSPC) { 4971 ztest_record_enospc("dsl_dataset_user_hold"); 4972 goto out; 4973 } else if (error) { 4974 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u", 4975 fullname, tag, error); 4976 } 4977 4978 error = dsl_destroy_snapshot(fullname, B_FALSE); 4979 if (error != EBUSY) { 4980 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d", 4981 fullname, error); 4982 } 4983 4984 error = dsl_destroy_snapshot(fullname, B_TRUE); 4985 if (error) { 4986 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", 4987 fullname, error); 4988 } 4989 4990 error = user_release_one(fullname, tag); 4991 if (error) 4992 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error); 4993 4994 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT); 4995 4996out: 4997 rw_exit(&ztest_name_lock); 4998} 4999 5000/* 5001 * Inject random faults into the on-disk data. 5002 */ 5003/* ARGSUSED */ 5004void 5005ztest_fault_inject(ztest_ds_t *zd, uint64_t id) 5006{ 5007 ztest_shared_t *zs = ztest_shared; 5008 spa_t *spa = ztest_spa; 5009 int fd; 5010 uint64_t offset; 5011 uint64_t leaves; 5012 uint64_t bad = 0x1990c0ffeedecadeULL; 5013 uint64_t top, leaf; 5014 char path0[MAXPATHLEN]; 5015 char pathrand[MAXPATHLEN]; 5016 size_t fsize; 5017 int bshift = SPA_MAXBLOCKSHIFT + 2; 5018 int iters = 1000; 5019 int maxfaults; 5020 int mirror_save; 5021 vdev_t *vd0 = NULL; 5022 uint64_t guid0 = 0; 5023 boolean_t islog = B_FALSE; 5024 5025 mutex_enter(&ztest_vdev_lock); 5026 5027 /* 5028 * Device removal is in progress, fault injection must be disabled 5029 * until it completes and the pool is scrubbed. The fault injection 5030 * strategy for damaging blocks does not take in to account evacuated 5031 * blocks which may have already been damaged. 5032 */ 5033 if (ztest_device_removal_active) { 5034 mutex_exit(&ztest_vdev_lock); 5035 return; 5036 } 5037 5038 maxfaults = MAXFAULTS(); 5039 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; 5040 mirror_save = zs->zs_mirrors; 5041 mutex_exit(&ztest_vdev_lock); 5042 5043 ASSERT(leaves >= 1); 5044 5045 /* 5046 * Grab the name lock as reader. There are some operations 5047 * which don't like to have their vdevs changed while 5048 * they are in progress (i.e. spa_change_guid). Those 5049 * operations will have grabbed the name lock as writer. 5050 */ 5051 rw_enter(&ztest_name_lock, RW_READER); 5052 5053 /* 5054 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd. 5055 */ 5056 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 5057 5058 if (ztest_random(2) == 0) { 5059 /* 5060 * Inject errors on a normal data device or slog device. 5061 */ 5062 top = ztest_random_vdev_top(spa, B_TRUE); 5063 leaf = ztest_random(leaves) + zs->zs_splits; 5064 5065 /* 5066 * Generate paths to the first leaf in this top-level vdev, 5067 * and to the random leaf we selected. We'll induce transient 5068 * write failures and random online/offline activity on leaf 0, 5069 * and we'll write random garbage to the randomly chosen leaf. 5070 */ 5071 (void) snprintf(path0, sizeof (path0), ztest_dev_template, 5072 ztest_opts.zo_dir, ztest_opts.zo_pool, 5073 top * leaves + zs->zs_splits); 5074 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template, 5075 ztest_opts.zo_dir, ztest_opts.zo_pool, 5076 top * leaves + leaf); 5077 5078 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); 5079 if (vd0 != NULL && vd0->vdev_top->vdev_islog) 5080 islog = B_TRUE; 5081 5082 /* 5083 * If the top-level vdev needs to be resilvered 5084 * then we only allow faults on the device that is 5085 * resilvering. 5086 */ 5087 if (vd0 != NULL && maxfaults != 1 && 5088 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) || 5089 vd0->vdev_resilver_txg != 0)) { 5090 /* 5091 * Make vd0 explicitly claim to be unreadable, 5092 * or unwriteable, or reach behind its back 5093 * and close the underlying fd. We can do this if 5094 * maxfaults == 0 because we'll fail and reexecute, 5095 * and we can do it if maxfaults >= 2 because we'll 5096 * have enough redundancy. If maxfaults == 1, the 5097 * combination of this with injection of random data 5098 * corruption below exceeds the pool's fault tolerance. 5099 */ 5100 vdev_file_t *vf = vd0->vdev_tsd; 5101 5102 zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d", 5103 (long long)vd0->vdev_id, (int)maxfaults); 5104 5105 if (vf != NULL && ztest_random(3) == 0) { 5106 (void) close(vf->vf_vnode->v_fd); 5107 vf->vf_vnode->v_fd = -1; 5108 } else if (ztest_random(2) == 0) { 5109 vd0->vdev_cant_read = B_TRUE; 5110 } else { 5111 vd0->vdev_cant_write = B_TRUE; 5112 } 5113 guid0 = vd0->vdev_guid; 5114 } 5115 } else { 5116 /* 5117 * Inject errors on an l2cache device. 5118 */ 5119 spa_aux_vdev_t *sav = &spa->spa_l2cache; 5120 5121 if (sav->sav_count == 0) { 5122 spa_config_exit(spa, SCL_STATE, FTAG); 5123 rw_exit(&ztest_name_lock); 5124 return; 5125 } 5126 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)]; 5127 guid0 = vd0->vdev_guid; 5128 (void) strcpy(path0, vd0->vdev_path); 5129 (void) strcpy(pathrand, vd0->vdev_path); 5130 5131 leaf = 0; 5132 leaves = 1; 5133 maxfaults = INT_MAX; /* no limit on cache devices */ 5134 } 5135 5136 spa_config_exit(spa, SCL_STATE, FTAG); 5137 rw_exit(&ztest_name_lock); 5138 5139 /* 5140 * If we can tolerate two or more faults, or we're dealing 5141 * with a slog, randomly online/offline vd0. 5142 */ 5143 if ((maxfaults >= 2 || islog) && guid0 != 0) { 5144 if (ztest_random(10) < 6) { 5145 int flags = (ztest_random(2) == 0 ? 5146 ZFS_OFFLINE_TEMPORARY : 0); 5147 5148 /* 5149 * We have to grab the zs_name_lock as writer to 5150 * prevent a race between offlining a slog and 5151 * destroying a dataset. Offlining the slog will 5152 * grab a reference on the dataset which may cause 5153 * dmu_objset_destroy() to fail with EBUSY thus 5154 * leaving the dataset in an inconsistent state. 5155 */ 5156 if (islog) 5157 rw_enter(&ztest_name_lock, RW_WRITER); 5158 5159 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY); 5160 5161 if (islog) 5162 rw_exit(&ztest_name_lock); 5163 } else { 5164 /* 5165 * Ideally we would like to be able to randomly 5166 * call vdev_[on|off]line without holding locks 5167 * to force unpredictable failures but the side 5168 * effects of vdev_[on|off]line prevent us from 5169 * doing so. We grab the ztest_vdev_lock here to 5170 * prevent a race between injection testing and 5171 * aux_vdev removal. 5172 */ 5173 mutex_enter(&ztest_vdev_lock); 5174 (void) vdev_online(spa, guid0, 0, NULL); 5175 mutex_exit(&ztest_vdev_lock); 5176 } 5177 } 5178 5179 if (maxfaults == 0) 5180 return; 5181 5182 /* 5183 * We have at least single-fault tolerance, so inject data corruption. 5184 */ 5185 fd = open(pathrand, O_RDWR); 5186 5187 if (fd == -1) /* we hit a gap in the device namespace */ 5188 return; 5189 5190 fsize = lseek(fd, 0, SEEK_END); 5191 5192 while (--iters != 0) { 5193 /* 5194 * The offset must be chosen carefully to ensure that 5195 * we do not inject a given logical block with errors 5196 * on two different leaf devices, because ZFS can not 5197 * tolerate that (if maxfaults==1). 5198 * 5199 * We divide each leaf into chunks of size 5200 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk 5201 * there is a series of ranges to which we can inject errors. 5202 * Each range can accept errors on only a single leaf vdev. 5203 * The error injection ranges are separated by ranges 5204 * which we will not inject errors on any device (DMZs). 5205 * Each DMZ must be large enough such that a single block 5206 * can not straddle it, so that a single block can not be 5207 * a target in two different injection ranges (on different 5208 * leaf vdevs). 5209 * 5210 * For example, with 3 leaves, each chunk looks like: 5211 * 0 to 32M: injection range for leaf 0 5212 * 32M to 64M: DMZ - no injection allowed 5213 * 64M to 96M: injection range for leaf 1 5214 * 96M to 128M: DMZ - no injection allowed 5215 * 128M to 160M: injection range for leaf 2 5216 * 160M to 192M: DMZ - no injection allowed 5217 */ 5218 offset = ztest_random(fsize / (leaves << bshift)) * 5219 (leaves << bshift) + (leaf << bshift) + 5220 (ztest_random(1ULL << (bshift - 1)) & -8ULL); 5221 5222 /* 5223 * Only allow damage to the labels at one end of the vdev. 5224 * 5225 * If all labels are damaged, the device will be totally 5226 * inaccessible, which will result in loss of data, 5227 * because we also damage (parts of) the other side of 5228 * the mirror/raidz. 5229 * 5230 * Additionally, we will always have both an even and an 5231 * odd label, so that we can handle crashes in the 5232 * middle of vdev_config_sync(). 5233 */ 5234 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE) 5235 continue; 5236 5237 /* 5238 * The two end labels are stored at the "end" of the disk, but 5239 * the end of the disk (vdev_psize) is aligned to 5240 * sizeof (vdev_label_t). 5241 */ 5242 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t)); 5243 if ((leaf & 1) == 1 && 5244 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE) 5245 continue; 5246 5247 mutex_enter(&ztest_vdev_lock); 5248 if (mirror_save != zs->zs_mirrors) { 5249 mutex_exit(&ztest_vdev_lock); 5250 (void) close(fd); 5251 return; 5252 } 5253 5254 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) 5255 fatal(1, "can't inject bad word at 0x%llx in %s", 5256 offset, pathrand); 5257 5258 mutex_exit(&ztest_vdev_lock); 5259 5260 if (ztest_opts.zo_verbose >= 7) 5261 (void) printf("injected bad word into %s," 5262 " offset 0x%llx\n", pathrand, (u_longlong_t)offset); 5263 } 5264 5265 (void) close(fd); 5266} 5267 5268/* 5269 * Verify that DDT repair works as expected. 5270 */ 5271void 5272ztest_ddt_repair(ztest_ds_t *zd, uint64_t id) 5273{ 5274 ztest_shared_t *zs = ztest_shared; 5275 spa_t *spa = ztest_spa; 5276 objset_t *os = zd->zd_os; 5277 ztest_od_t od[1]; 5278 uint64_t object, blocksize, txg, pattern, psize; 5279 enum zio_checksum checksum = spa_dedup_checksum(spa); 5280 dmu_buf_t *db; 5281 dmu_tx_t *tx; 5282 abd_t *abd; 5283 blkptr_t blk; 5284 int copies = 2 * ZIO_DEDUPDITTO_MIN; 5285 5286 blocksize = ztest_random_blocksize(); 5287 blocksize = MIN(blocksize, 2048); /* because we write so many */ 5288 5289 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 5290 5291 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 5292 return; 5293 5294 /* 5295 * Take the name lock as writer to prevent anyone else from changing 5296 * the pool and dataset properies we need to maintain during this test. 5297 */ 5298 rw_enter(&ztest_name_lock, RW_WRITER); 5299 5300 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum, 5301 B_FALSE) != 0 || 5302 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1, 5303 B_FALSE) != 0) { 5304 rw_exit(&ztest_name_lock); 5305 return; 5306 } 5307 5308 dmu_objset_stats_t dds; 5309 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 5310 dmu_objset_fast_stat(os, &dds); 5311 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 5312 5313 object = od[0].od_object; 5314 blocksize = od[0].od_blocksize; 5315 pattern = zs->zs_guid ^ dds.dds_guid; 5316 5317 ASSERT(object != 0); 5318 5319 tx = dmu_tx_create(os); 5320 dmu_tx_hold_write(tx, object, 0, copies * blocksize); 5321 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 5322 if (txg == 0) { 5323 rw_exit(&ztest_name_lock); 5324 return; 5325 } 5326 5327 /* 5328 * Write all the copies of our block. 5329 */ 5330 for (int i = 0; i < copies; i++) { 5331 uint64_t offset = i * blocksize; 5332 int error = dmu_buf_hold(os, object, offset, FTAG, &db, 5333 DMU_READ_NO_PREFETCH); 5334 if (error != 0) { 5335 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u", 5336 os, (long long)object, (long long) offset, error); 5337 } 5338 ASSERT(db->db_offset == offset); 5339 ASSERT(db->db_size == blocksize); 5340 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) || 5341 ztest_pattern_match(db->db_data, db->db_size, 0ULL)); 5342 dmu_buf_will_fill(db, tx); 5343 ztest_pattern_set(db->db_data, db->db_size, pattern); 5344 dmu_buf_rele(db, FTAG); 5345 } 5346 5347 dmu_tx_commit(tx); 5348 txg_wait_synced(spa_get_dsl(spa), txg); 5349 5350 /* 5351 * Find out what block we got. 5352 */ 5353 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db, 5354 DMU_READ_NO_PREFETCH)); 5355 blk = *((dmu_buf_impl_t *)db)->db_blkptr; 5356 dmu_buf_rele(db, FTAG); 5357 5358 /* 5359 * Damage the block. Dedup-ditto will save us when we read it later. 5360 */ 5361 psize = BP_GET_PSIZE(&blk); 5362 abd = abd_alloc_linear(psize, B_TRUE); 5363 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern); 5364 5365 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk, 5366 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE, 5367 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL)); 5368 5369 abd_free(abd); 5370 5371 rw_exit(&ztest_name_lock); 5372} 5373 5374/* 5375 * Scrub the pool. 5376 */ 5377/* ARGSUSED */ 5378void 5379ztest_scrub(ztest_ds_t *zd, uint64_t id) 5380{ 5381 spa_t *spa = ztest_spa; 5382 5383 /* 5384 * Scrub in progress by device removal. 5385 */ 5386 if (ztest_device_removal_active) 5387 return; 5388 5389 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5390 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */ 5391 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5392} 5393 5394/* 5395 * Change the guid for the pool. 5396 */ 5397/* ARGSUSED */ 5398void 5399ztest_reguid(ztest_ds_t *zd, uint64_t id) 5400{ 5401 spa_t *spa = ztest_spa; 5402 uint64_t orig, load; 5403 int error; 5404 5405 orig = spa_guid(spa); 5406 load = spa_load_guid(spa); 5407 5408 rw_enter(&ztest_name_lock, RW_WRITER); 5409 error = spa_change_guid(spa); 5410 rw_exit(&ztest_name_lock); 5411 5412 if (error != 0) 5413 return; 5414 5415 if (ztest_opts.zo_verbose >= 4) { 5416 (void) printf("Changed guid old %llu -> %llu\n", 5417 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa)); 5418 } 5419 5420 VERIFY3U(orig, !=, spa_guid(spa)); 5421 VERIFY3U(load, ==, spa_load_guid(spa)); 5422} 5423 5424/* 5425 * Rename the pool to a different name and then rename it back. 5426 */ 5427/* ARGSUSED */ 5428void 5429ztest_spa_rename(ztest_ds_t *zd, uint64_t id) 5430{ 5431 char *oldname, *newname; 5432 spa_t *spa; 5433 5434 rw_enter(&ztest_name_lock, RW_WRITER); 5435 5436 oldname = ztest_opts.zo_pool; 5437 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); 5438 (void) strcpy(newname, oldname); 5439 (void) strcat(newname, "_tmp"); 5440 5441 /* 5442 * Do the rename 5443 */ 5444 VERIFY3U(0, ==, spa_rename(oldname, newname)); 5445 5446 /* 5447 * Try to open it under the old name, which shouldn't exist 5448 */ 5449 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); 5450 5451 /* 5452 * Open it under the new name and make sure it's still the same spa_t. 5453 */ 5454 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); 5455 5456 ASSERT(spa == ztest_spa); 5457 spa_close(spa, FTAG); 5458 5459 /* 5460 * Rename it back to the original 5461 */ 5462 VERIFY3U(0, ==, spa_rename(newname, oldname)); 5463 5464 /* 5465 * Make sure it can still be opened 5466 */ 5467 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); 5468 5469 ASSERT(spa == ztest_spa); 5470 spa_close(spa, FTAG); 5471 5472 umem_free(newname, strlen(newname) + 1); 5473 5474 rw_exit(&ztest_name_lock); 5475} 5476 5477static vdev_t * 5478ztest_random_concrete_vdev_leaf(vdev_t *vd) 5479{ 5480 if (vd == NULL) 5481 return (NULL); 5482 5483 if (vd->vdev_children == 0) 5484 return (vd); 5485 5486 vdev_t *eligible[vd->vdev_children]; 5487 int eligible_idx = 0, i; 5488 for (i = 0; i < vd->vdev_children; i++) { 5489 vdev_t *cvd = vd->vdev_child[i]; 5490 if (cvd->vdev_top->vdev_removing) 5491 continue; 5492 if (cvd->vdev_children > 0 || 5493 (vdev_is_concrete(cvd) && !cvd->vdev_detached)) { 5494 eligible[eligible_idx++] = cvd; 5495 } 5496 } 5497 VERIFY(eligible_idx > 0); 5498 5499 uint64_t child_no = ztest_random(eligible_idx); 5500 return (ztest_random_concrete_vdev_leaf(eligible[child_no])); 5501} 5502 5503/* ARGSUSED */ 5504void 5505ztest_initialize(ztest_ds_t *zd, uint64_t id) 5506{ 5507 spa_t *spa = ztest_spa; 5508 int error = 0; 5509 5510 mutex_enter(&ztest_vdev_lock); 5511 5512 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 5513 5514 /* Random leaf vdev */ 5515 vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev); 5516 if (rand_vd == NULL) { 5517 spa_config_exit(spa, SCL_VDEV, FTAG); 5518 mutex_exit(&ztest_vdev_lock); 5519 return; 5520 } 5521 5522 /* 5523 * The random vdev we've selected may change as soon as we 5524 * drop the spa_config_lock. We create local copies of things 5525 * we're interested in. 5526 */ 5527 uint64_t guid = rand_vd->vdev_guid; 5528 char *path = strdup(rand_vd->vdev_path); 5529 boolean_t active = rand_vd->vdev_initialize_thread != NULL; 5530 5531 zfs_dbgmsg("vd %p, guid %llu", rand_vd, guid); 5532 spa_config_exit(spa, SCL_VDEV, FTAG); 5533 5534 uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS); 5535 error = spa_vdev_initialize(spa, guid, cmd); 5536 switch (cmd) { 5537 case POOL_INITIALIZE_CANCEL: 5538 if (ztest_opts.zo_verbose >= 4) { 5539 (void) printf("Cancel initialize %s", path); 5540 if (!active) 5541 (void) printf(" failed (no initialize active)"); 5542 (void) printf("\n"); 5543 } 5544 break; 5545 case POOL_INITIALIZE_DO: 5546 if (ztest_opts.zo_verbose >= 4) { 5547 (void) printf("Start initialize %s", path); 5548 if (active && error == 0) 5549 (void) printf(" failed (already active)"); 5550 else if (error != 0) 5551 (void) printf(" failed (error %d)", error); 5552 (void) printf("\n"); 5553 } 5554 break; 5555 case POOL_INITIALIZE_SUSPEND: 5556 if (ztest_opts.zo_verbose >= 4) { 5557 (void) printf("Suspend initialize %s", path); 5558 if (!active) 5559 (void) printf(" failed (no initialize active)"); 5560 (void) printf("\n"); 5561 } 5562 break; 5563 } 5564 free(path); 5565 mutex_exit(&ztest_vdev_lock); 5566} 5567 5568/* 5569 * Verify pool integrity by running zdb. 5570 */ 5571static void 5572ztest_run_zdb(char *pool) 5573{ 5574 int status; 5575 char zdb[MAXPATHLEN + MAXNAMELEN + 20]; 5576 char zbuf[1024]; 5577 char *bin; 5578 char *ztest; 5579 char *isa; 5580 int isalen; 5581 FILE *fp; 5582 5583 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb)); 5584 5585 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ 5586 bin = strstr(zdb, "/usr/bin/"); 5587 ztest = strstr(bin, "/ztest"); 5588 isa = bin + 8; 5589 isalen = ztest - isa; 5590 isa = strdup(isa); 5591 /* LINTED */ 5592 (void) sprintf(bin, 5593 "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s %s", 5594 isalen, 5595 isa, 5596 ztest_opts.zo_verbose >= 3 ? "s" : "", 5597 ztest_opts.zo_verbose >= 4 ? "v" : "", 5598 spa_config_path, 5599 pool); 5600 free(isa); 5601 5602 if (ztest_opts.zo_verbose >= 5) 5603 (void) printf("Executing %s\n", strstr(zdb, "zdb ")); 5604 5605 fp = popen(zdb, "r"); 5606 assert(fp != NULL); 5607 5608 while (fgets(zbuf, sizeof (zbuf), fp) != NULL) 5609 if (ztest_opts.zo_verbose >= 3) 5610 (void) printf("%s", zbuf); 5611 5612 status = pclose(fp); 5613 5614 if (status == 0) 5615 return; 5616 5617 ztest_dump_core = 0; 5618 if (WIFEXITED(status)) 5619 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); 5620 else 5621 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); 5622} 5623 5624static void 5625ztest_walk_pool_directory(char *header) 5626{ 5627 spa_t *spa = NULL; 5628 5629 if (ztest_opts.zo_verbose >= 6) 5630 (void) printf("%s\n", header); 5631 5632 mutex_enter(&spa_namespace_lock); 5633 while ((spa = spa_next(spa)) != NULL) 5634 if (ztest_opts.zo_verbose >= 6) 5635 (void) printf("\t%s\n", spa_name(spa)); 5636 mutex_exit(&spa_namespace_lock); 5637} 5638 5639static void 5640ztest_spa_import_export(char *oldname, char *newname) 5641{ 5642 nvlist_t *config, *newconfig; 5643 uint64_t pool_guid; 5644 spa_t *spa; 5645 int error; 5646 5647 if (ztest_opts.zo_verbose >= 4) { 5648 (void) printf("import/export: old = %s, new = %s\n", 5649 oldname, newname); 5650 } 5651 5652 /* 5653 * Clean up from previous runs. 5654 */ 5655 (void) spa_destroy(newname); 5656 5657 /* 5658 * Get the pool's configuration and guid. 5659 */ 5660 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); 5661 5662 /* 5663 * Kick off a scrub to tickle scrub/export races. 5664 */ 5665 if (ztest_random(2) == 0) 5666 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5667 5668 pool_guid = spa_guid(spa); 5669 spa_close(spa, FTAG); 5670 5671 ztest_walk_pool_directory("pools before export"); 5672 5673 /* 5674 * Export it. 5675 */ 5676 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE)); 5677 5678 ztest_walk_pool_directory("pools after export"); 5679 5680 /* 5681 * Try to import it. 5682 */ 5683 newconfig = spa_tryimport(config); 5684 ASSERT(newconfig != NULL); 5685 nvlist_free(newconfig); 5686 5687 /* 5688 * Import it under the new name. 5689 */ 5690 error = spa_import(newname, config, NULL, 0); 5691 if (error != 0) { 5692 dump_nvlist(config, 0); 5693 fatal(B_FALSE, "couldn't import pool %s as %s: error %u", 5694 oldname, newname, error); 5695 } 5696 5697 ztest_walk_pool_directory("pools after import"); 5698 5699 /* 5700 * Try to import it again -- should fail with EEXIST. 5701 */ 5702 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0)); 5703 5704 /* 5705 * Try to import it under a different name -- should fail with EEXIST. 5706 */ 5707 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0)); 5708 5709 /* 5710 * Verify that the pool is no longer visible under the old name. 5711 */ 5712 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); 5713 5714 /* 5715 * Verify that we can open and close the pool using the new name. 5716 */ 5717 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); 5718 ASSERT(pool_guid == spa_guid(spa)); 5719 spa_close(spa, FTAG); 5720 5721 nvlist_free(config); 5722} 5723 5724static void 5725ztest_resume(spa_t *spa) 5726{ 5727 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6) 5728 (void) printf("resuming from suspended state\n"); 5729 spa_vdev_state_enter(spa, SCL_NONE); 5730 vdev_clear(spa, NULL); 5731 (void) spa_vdev_state_exit(spa, NULL, 0); 5732 (void) zio_resume(spa); 5733} 5734 5735static void * 5736ztest_resume_thread(void *arg) 5737{ 5738 spa_t *spa = arg; 5739 5740 while (!ztest_exiting) { 5741 if (spa_suspended(spa)) 5742 ztest_resume(spa); 5743 (void) poll(NULL, 0, 100); 5744 5745 /* 5746 * Periodically change the zfs_compressed_arc_enabled setting. 5747 */ 5748 if (ztest_random(10) == 0) 5749 zfs_compressed_arc_enabled = ztest_random(2); 5750 5751 /* 5752 * Periodically change the zfs_abd_scatter_enabled setting. 5753 */ 5754 if (ztest_random(10) == 0) 5755 zfs_abd_scatter_enabled = ztest_random(2); 5756 } 5757 return (NULL); 5758} 5759 5760static void * 5761ztest_deadman_thread(void *arg) 5762{ 5763 ztest_shared_t *zs = arg; 5764 spa_t *spa = ztest_spa; 5765 hrtime_t delta, total = 0; 5766 5767 for (;;) { 5768 delta = zs->zs_thread_stop - zs->zs_thread_start + 5769 MSEC2NSEC(zfs_deadman_synctime_ms); 5770 5771 (void) poll(NULL, 0, (int)NSEC2MSEC(delta)); 5772 5773 /* 5774 * If the pool is suspended then fail immediately. Otherwise, 5775 * check to see if the pool is making any progress. If 5776 * vdev_deadman() discovers that there hasn't been any recent 5777 * I/Os then it will end up aborting the tests. 5778 */ 5779 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) { 5780 fatal(0, "aborting test after %llu seconds because " 5781 "pool has transitioned to a suspended state.", 5782 zfs_deadman_synctime_ms / 1000); 5783 return (NULL); 5784 } 5785 vdev_deadman(spa->spa_root_vdev); 5786 5787 total += zfs_deadman_synctime_ms/1000; 5788 (void) printf("ztest has been running for %lld seconds\n", 5789 total); 5790 } 5791} 5792 5793static void 5794ztest_execute(int test, ztest_info_t *zi, uint64_t id) 5795{ 5796 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets]; 5797 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test); 5798 hrtime_t functime = gethrtime(); 5799 5800 for (int i = 0; i < zi->zi_iters; i++) 5801 zi->zi_func(zd, id); 5802 5803 functime = gethrtime() - functime; 5804 5805 atomic_add_64(&zc->zc_count, 1); 5806 atomic_add_64(&zc->zc_time, functime); 5807 5808 if (ztest_opts.zo_verbose >= 4) { 5809 Dl_info dli; 5810 (void) dladdr((void *)zi->zi_func, &dli); 5811 (void) printf("%6.2f sec in %s\n", 5812 (double)functime / NANOSEC, dli.dli_sname); 5813 } 5814} 5815 5816static void * 5817ztest_thread(void *arg) 5818{ 5819 int rand; 5820 uint64_t id = (uintptr_t)arg; 5821 ztest_shared_t *zs = ztest_shared; 5822 uint64_t call_next; 5823 hrtime_t now; 5824 ztest_info_t *zi; 5825 ztest_shared_callstate_t *zc; 5826 5827 while ((now = gethrtime()) < zs->zs_thread_stop) { 5828 /* 5829 * See if it's time to force a crash. 5830 */ 5831 if (now > zs->zs_thread_kill) 5832 ztest_kill(zs); 5833 5834 /* 5835 * If we're getting ENOSPC with some regularity, stop. 5836 */ 5837 if (zs->zs_enospc_count > 10) 5838 break; 5839 5840 /* 5841 * Pick a random function to execute. 5842 */ 5843 rand = ztest_random(ZTEST_FUNCS); 5844 zi = &ztest_info[rand]; 5845 zc = ZTEST_GET_SHARED_CALLSTATE(rand); 5846 call_next = zc->zc_next; 5847 5848 if (now >= call_next && 5849 atomic_cas_64(&zc->zc_next, call_next, call_next + 5850 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) { 5851 ztest_execute(rand, zi, id); 5852 } 5853 } 5854 5855 return (NULL); 5856} 5857 5858static void 5859ztest_dataset_name(char *dsname, char *pool, int d) 5860{ 5861 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d); 5862} 5863 5864static void 5865ztest_dataset_destroy(int d) 5866{ 5867 char name[ZFS_MAX_DATASET_NAME_LEN]; 5868 5869 ztest_dataset_name(name, ztest_opts.zo_pool, d); 5870 5871 if (ztest_opts.zo_verbose >= 3) 5872 (void) printf("Destroying %s to free up space\n", name); 5873 5874 /* 5875 * Cleanup any non-standard clones and snapshots. In general, 5876 * ztest thread t operates on dataset (t % zopt_datasets), 5877 * so there may be more than one thing to clean up. 5878 */ 5879 for (int t = d; t < ztest_opts.zo_threads; 5880 t += ztest_opts.zo_datasets) { 5881 ztest_dsl_dataset_cleanup(name, t); 5882 } 5883 5884 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, 5885 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); 5886} 5887 5888static void 5889ztest_dataset_dirobj_verify(ztest_ds_t *zd) 5890{ 5891 uint64_t usedobjs, dirobjs, scratch; 5892 5893 /* 5894 * ZTEST_DIROBJ is the object directory for the entire dataset. 5895 * Therefore, the number of objects in use should equal the 5896 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself. 5897 * If not, we have an object leak. 5898 * 5899 * Note that we can only check this in ztest_dataset_open(), 5900 * when the open-context and syncing-context values agree. 5901 * That's because zap_count() returns the open-context value, 5902 * while dmu_objset_space() returns the rootbp fill count. 5903 */ 5904 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs)); 5905 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch); 5906 ASSERT3U(dirobjs + 1, ==, usedobjs); 5907} 5908 5909static int 5910ztest_dataset_open(int d) 5911{ 5912 ztest_ds_t *zd = &ztest_ds[d]; 5913 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq; 5914 objset_t *os; 5915 zilog_t *zilog; 5916 char name[ZFS_MAX_DATASET_NAME_LEN]; 5917 int error; 5918 5919 ztest_dataset_name(name, ztest_opts.zo_pool, d); 5920 5921 rw_enter(&ztest_name_lock, RW_READER); 5922 5923 error = ztest_dataset_create(name); 5924 if (error == ENOSPC) { 5925 rw_exit(&ztest_name_lock); 5926 ztest_record_enospc(FTAG); 5927 return (error); 5928 } 5929 ASSERT(error == 0 || error == EEXIST); 5930 5931 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os)); 5932 rw_exit(&ztest_name_lock); 5933 5934 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os); 5935 5936 zilog = zd->zd_zilog; 5937 5938 if (zilog->zl_header->zh_claim_lr_seq != 0 && 5939 zilog->zl_header->zh_claim_lr_seq < committed_seq) 5940 fatal(0, "missing log records: claimed %llu < committed %llu", 5941 zilog->zl_header->zh_claim_lr_seq, committed_seq); 5942 5943 ztest_dataset_dirobj_verify(zd); 5944 5945 zil_replay(os, zd, ztest_replay_vector); 5946 5947 ztest_dataset_dirobj_verify(zd); 5948 5949 if (ztest_opts.zo_verbose >= 6) 5950 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n", 5951 zd->zd_name, 5952 (u_longlong_t)zilog->zl_parse_blk_count, 5953 (u_longlong_t)zilog->zl_parse_lr_count, 5954 (u_longlong_t)zilog->zl_replaying_seq); 5955 5956 zilog = zil_open(os, ztest_get_data); 5957 5958 if (zilog->zl_replaying_seq != 0 && 5959 zilog->zl_replaying_seq < committed_seq) 5960 fatal(0, "missing log records: replayed %llu < committed %llu", 5961 zilog->zl_replaying_seq, committed_seq); 5962 5963 return (0); 5964} 5965 5966static void 5967ztest_dataset_close(int d) 5968{ 5969 ztest_ds_t *zd = &ztest_ds[d]; 5970 5971 zil_close(zd->zd_zilog); 5972 dmu_objset_disown(zd->zd_os, zd); 5973 5974 ztest_zd_fini(zd); 5975} 5976 5977/* 5978 * Kick off threads to run tests on all datasets in parallel. 5979 */ 5980static void 5981ztest_run(ztest_shared_t *zs) 5982{ 5983 thread_t *tid; 5984 spa_t *spa; 5985 objset_t *os; 5986 thread_t resume_tid; 5987 int error; 5988 5989 ztest_exiting = B_FALSE; 5990 5991 /* 5992 * Initialize parent/child shared state. 5993 */ 5994 mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL); 5995 mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL); 5996 rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL); 5997 5998 zs->zs_thread_start = gethrtime(); 5999 zs->zs_thread_stop = 6000 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC; 6001 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop); 6002 zs->zs_thread_kill = zs->zs_thread_stop; 6003 if (ztest_random(100) < ztest_opts.zo_killrate) { 6004 zs->zs_thread_kill -= 6005 ztest_random(ztest_opts.zo_passtime * NANOSEC); 6006 } 6007 6008 mutex_init(&zcl.zcl_callbacks_lock, NULL, USYNC_THREAD, NULL); 6009 6010 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t), 6011 offsetof(ztest_cb_data_t, zcd_node)); 6012 6013 /* 6014 * Open our pool. 6015 */ 6016 kernel_init(FREAD | FWRITE); 6017 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG)); 6018 metaslab_preload_limit = ztest_random(20) + 1; 6019 ztest_spa = spa; 6020 6021 dmu_objset_stats_t dds; 6022 VERIFY0(dmu_objset_own(ztest_opts.zo_pool, 6023 DMU_OST_ANY, B_TRUE, FTAG, &os)); 6024 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 6025 dmu_objset_fast_stat(os, &dds); 6026 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 6027 zs->zs_guid = dds.dds_guid; 6028 dmu_objset_disown(os, FTAG); 6029 6030 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN; 6031 6032 /* 6033 * We don't expect the pool to suspend unless maxfaults == 0, 6034 * in which case ztest_fault_inject() temporarily takes away 6035 * the only valid replica. 6036 */ 6037 if (MAXFAULTS() == 0) 6038 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT; 6039 else 6040 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC; 6041 6042 /* 6043 * Create a thread to periodically resume suspended I/O. 6044 */ 6045 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND, 6046 &resume_tid) == 0); 6047 6048 /* 6049 * Create a deadman thread to abort() if we hang. 6050 */ 6051 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND, 6052 NULL) == 0); 6053 6054 /* 6055 * Verify that we can safely inquire about any object, 6056 * whether it's allocated or not. To make it interesting, 6057 * we probe a 5-wide window around each power of two. 6058 * This hits all edge cases, including zero and the max. 6059 */ 6060 for (int t = 0; t < 64; t++) { 6061 for (int d = -5; d <= 5; d++) { 6062 error = dmu_object_info(spa->spa_meta_objset, 6063 (1ULL << t) + d, NULL); 6064 ASSERT(error == 0 || error == ENOENT || 6065 error == EINVAL); 6066 } 6067 } 6068 6069 /* 6070 * If we got any ENOSPC errors on the previous run, destroy something. 6071 */ 6072 if (zs->zs_enospc_count != 0) { 6073 int d = ztest_random(ztest_opts.zo_datasets); 6074 ztest_dataset_destroy(d); 6075 } 6076 zs->zs_enospc_count = 0; 6077 6078 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t), 6079 UMEM_NOFAIL); 6080 6081 if (ztest_opts.zo_verbose >= 4) 6082 (void) printf("starting main threads...\n"); 6083 6084 /* 6085 * Kick off all the tests that run in parallel. 6086 */ 6087 for (int t = 0; t < ztest_opts.zo_threads; t++) { 6088 if (t < ztest_opts.zo_datasets && 6089 ztest_dataset_open(t) != 0) 6090 return; 6091 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t, 6092 THR_BOUND, &tid[t]) == 0); 6093 } 6094 6095 /* 6096 * Wait for all of the tests to complete. We go in reverse order 6097 * so we don't close datasets while threads are still using them. 6098 */ 6099 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) { 6100 VERIFY(thr_join(tid[t], NULL, NULL) == 0); 6101 if (t < ztest_opts.zo_datasets) 6102 ztest_dataset_close(t); 6103 } 6104 6105 txg_wait_synced(spa_get_dsl(spa), 0); 6106 6107 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 6108 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa)); 6109 zfs_dbgmsg_print(FTAG); 6110 6111 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t)); 6112 6113 /* Kill the resume thread */ 6114 ztest_exiting = B_TRUE; 6115 VERIFY(thr_join(resume_tid, NULL, NULL) == 0); 6116 ztest_resume(spa); 6117 6118 /* 6119 * Right before closing the pool, kick off a bunch of async I/O; 6120 * spa_close() should wait for it to complete. 6121 */ 6122 for (uint64_t object = 1; object < 50; object++) { 6123 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20, 6124 ZIO_PRIORITY_SYNC_READ); 6125 } 6126 6127 spa_close(spa, FTAG); 6128 6129 /* 6130 * Verify that we can loop over all pools. 6131 */ 6132 mutex_enter(&spa_namespace_lock); 6133 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) 6134 if (ztest_opts.zo_verbose > 3) 6135 (void) printf("spa_next: found %s\n", spa_name(spa)); 6136 mutex_exit(&spa_namespace_lock); 6137 6138 /* 6139 * Verify that we can export the pool and reimport it under a 6140 * different name. 6141 */ 6142 if (ztest_random(2) == 0) { 6143 char name[ZFS_MAX_DATASET_NAME_LEN]; 6144 (void) snprintf(name, sizeof (name), "%s_import", 6145 ztest_opts.zo_pool); 6146 ztest_spa_import_export(ztest_opts.zo_pool, name); 6147 ztest_spa_import_export(name, ztest_opts.zo_pool); 6148 } 6149 6150 kernel_fini(); 6151 6152 list_destroy(&zcl.zcl_callbacks); 6153 6154 mutex_destroy(&zcl.zcl_callbacks_lock); 6155 6156 rw_destroy(&ztest_name_lock); 6157 mutex_destroy(&ztest_vdev_lock); 6158 mutex_destroy(&ztest_checkpoint_lock); 6159} 6160 6161static void 6162ztest_freeze(void) 6163{ 6164 ztest_ds_t *zd = &ztest_ds[0]; 6165 spa_t *spa; 6166 int numloops = 0; 6167 6168 if (ztest_opts.zo_verbose >= 3) 6169 (void) printf("testing spa_freeze()...\n"); 6170 6171 kernel_init(FREAD | FWRITE); 6172 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 6173 VERIFY3U(0, ==, ztest_dataset_open(0)); 6174 ztest_spa = spa; 6175 6176 /* 6177 * Force the first log block to be transactionally allocated. 6178 * We have to do this before we freeze the pool -- otherwise 6179 * the log chain won't be anchored. 6180 */ 6181 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) { 6182 ztest_dmu_object_alloc_free(zd, 0); 6183 zil_commit(zd->zd_zilog, 0); 6184 } 6185 6186 txg_wait_synced(spa_get_dsl(spa), 0); 6187 6188 /* 6189 * Freeze the pool. This stops spa_sync() from doing anything, 6190 * so that the only way to record changes from now on is the ZIL. 6191 */ 6192 spa_freeze(spa); 6193 6194 /* 6195 * Because it is hard to predict how much space a write will actually 6196 * require beforehand, we leave ourselves some fudge space to write over 6197 * capacity. 6198 */ 6199 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2; 6200 6201 /* 6202 * Run tests that generate log records but don't alter the pool config 6203 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc). 6204 * We do a txg_wait_synced() after each iteration to force the txg 6205 * to increase well beyond the last synced value in the uberblock. 6206 * The ZIL should be OK with that. 6207 * 6208 * Run a random number of times less than zo_maxloops and ensure we do 6209 * not run out of space on the pool. 6210 */ 6211 while (ztest_random(10) != 0 && 6212 numloops++ < ztest_opts.zo_maxloops && 6213 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) { 6214 ztest_od_t od; 6215 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 6216 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE)); 6217 ztest_io(zd, od.od_object, 6218 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 6219 txg_wait_synced(spa_get_dsl(spa), 0); 6220 } 6221 6222 /* 6223 * Commit all of the changes we just generated. 6224 */ 6225 zil_commit(zd->zd_zilog, 0); 6226 txg_wait_synced(spa_get_dsl(spa), 0); 6227 6228 /* 6229 * Close our dataset and close the pool. 6230 */ 6231 ztest_dataset_close(0); 6232 spa_close(spa, FTAG); 6233 kernel_fini(); 6234 6235 /* 6236 * Open and close the pool and dataset to induce log replay. 6237 */ 6238 kernel_init(FREAD | FWRITE); 6239 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 6240 ASSERT(spa_freeze_txg(spa) == UINT64_MAX); 6241 VERIFY3U(0, ==, ztest_dataset_open(0)); 6242 ztest_dataset_close(0); 6243 6244 ztest_spa = spa; 6245 txg_wait_synced(spa_get_dsl(spa), 0); 6246 ztest_reguid(NULL, 0); 6247 6248 spa_close(spa, FTAG); 6249 kernel_fini(); 6250} 6251 6252void 6253print_time(hrtime_t t, char *timebuf) 6254{ 6255 hrtime_t s = t / NANOSEC; 6256 hrtime_t m = s / 60; 6257 hrtime_t h = m / 60; 6258 hrtime_t d = h / 24; 6259 6260 s -= m * 60; 6261 m -= h * 60; 6262 h -= d * 24; 6263 6264 timebuf[0] = '\0'; 6265 6266 if (d) 6267 (void) sprintf(timebuf, 6268 "%llud%02lluh%02llum%02llus", d, h, m, s); 6269 else if (h) 6270 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); 6271 else if (m) 6272 (void) sprintf(timebuf, "%llum%02llus", m, s); 6273 else 6274 (void) sprintf(timebuf, "%llus", s); 6275} 6276 6277static nvlist_t * 6278make_random_props() 6279{ 6280 nvlist_t *props; 6281 6282 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); 6283 if (ztest_random(2) == 0) 6284 return (props); 6285 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0); 6286 6287 return (props); 6288} 6289 6290/* 6291 * Create a storage pool with the given name and initial vdev size. 6292 * Then test spa_freeze() functionality. 6293 */ 6294static void 6295ztest_init(ztest_shared_t *zs) 6296{ 6297 spa_t *spa; 6298 nvlist_t *nvroot, *props; 6299 6300 mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL); 6301 mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL); 6302 rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL); 6303 6304 kernel_init(FREAD | FWRITE); 6305 6306 /* 6307 * Create the storage pool. 6308 */ 6309 (void) spa_destroy(ztest_opts.zo_pool); 6310 ztest_shared->zs_vdev_next_leaf = 0; 6311 zs->zs_splits = 0; 6312 zs->zs_mirrors = ztest_opts.zo_mirrors; 6313 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0, 6314 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1); 6315 props = make_random_props(); 6316 for (int i = 0; i < SPA_FEATURES; i++) { 6317 char buf[1024]; 6318 (void) snprintf(buf, sizeof (buf), "feature@%s", 6319 spa_feature_table[i].fi_uname); 6320 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0)); 6321 } 6322 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL)); 6323 nvlist_free(nvroot); 6324 nvlist_free(props); 6325 6326 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 6327 zs->zs_metaslab_sz = 6328 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift; 6329 6330 spa_close(spa, FTAG); 6331 6332 kernel_fini(); 6333 6334 ztest_run_zdb(ztest_opts.zo_pool); 6335 6336 ztest_freeze(); 6337 6338 ztest_run_zdb(ztest_opts.zo_pool); 6339 6340 rw_destroy(&ztest_name_lock); 6341 mutex_destroy(&ztest_vdev_lock); 6342 mutex_destroy(&ztest_checkpoint_lock); 6343} 6344 6345static void 6346setup_data_fd(void) 6347{ 6348 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX"; 6349 6350 ztest_fd_data = mkstemp(ztest_name_data); 6351 ASSERT3S(ztest_fd_data, >=, 0); 6352 (void) unlink(ztest_name_data); 6353} 6354 6355 6356static int 6357shared_data_size(ztest_shared_hdr_t *hdr) 6358{ 6359 int size; 6360 6361 size = hdr->zh_hdr_size; 6362 size += hdr->zh_opts_size; 6363 size += hdr->zh_size; 6364 size += hdr->zh_stats_size * hdr->zh_stats_count; 6365 size += hdr->zh_ds_size * hdr->zh_ds_count; 6366 6367 return (size); 6368} 6369 6370static void 6371setup_hdr(void) 6372{ 6373 int size; 6374 ztest_shared_hdr_t *hdr; 6375 6376 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), 6377 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); 6378 ASSERT(hdr != MAP_FAILED); 6379 6380 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t))); 6381 6382 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t); 6383 hdr->zh_opts_size = sizeof (ztest_shared_opts_t); 6384 hdr->zh_size = sizeof (ztest_shared_t); 6385 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t); 6386 hdr->zh_stats_count = ZTEST_FUNCS; 6387 hdr->zh_ds_size = sizeof (ztest_shared_ds_t); 6388 hdr->zh_ds_count = ztest_opts.zo_datasets; 6389 6390 size = shared_data_size(hdr); 6391 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size)); 6392 6393 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); 6394} 6395 6396static void 6397setup_data(void) 6398{ 6399 int size, offset; 6400 ztest_shared_hdr_t *hdr; 6401 uint8_t *buf; 6402 6403 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), 6404 PROT_READ, MAP_SHARED, ztest_fd_data, 0); 6405 ASSERT(hdr != MAP_FAILED); 6406 6407 size = shared_data_size(hdr); 6408 6409 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); 6410 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()), 6411 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); 6412 ASSERT(hdr != MAP_FAILED); 6413 buf = (uint8_t *)hdr; 6414 6415 offset = hdr->zh_hdr_size; 6416 ztest_shared_opts = (void *)&buf[offset]; 6417 offset += hdr->zh_opts_size; 6418 ztest_shared = (void *)&buf[offset]; 6419 offset += hdr->zh_size; 6420 ztest_shared_callstate = (void *)&buf[offset]; 6421 offset += hdr->zh_stats_size * hdr->zh_stats_count; 6422 ztest_shared_ds = (void *)&buf[offset]; 6423} 6424 6425static boolean_t 6426exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp) 6427{ 6428 pid_t pid; 6429 int status; 6430 char *cmdbuf = NULL; 6431 6432 pid = fork(); 6433 6434 if (cmd == NULL) { 6435 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 6436 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN); 6437 cmd = cmdbuf; 6438 } 6439 6440 if (pid == -1) 6441 fatal(1, "fork failed"); 6442 6443 if (pid == 0) { /* child */ 6444 char *emptyargv[2] = { cmd, NULL }; 6445 char fd_data_str[12]; 6446 6447 struct rlimit rl = { 1024, 1024 }; 6448 (void) setrlimit(RLIMIT_NOFILE, &rl); 6449 6450 (void) close(ztest_fd_rand); 6451 VERIFY3U(11, >=, 6452 snprintf(fd_data_str, 12, "%d", ztest_fd_data)); 6453 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1)); 6454 6455 (void) enable_extended_FILE_stdio(-1, -1); 6456 if (libpath != NULL) 6457 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1)); 6458#ifdef illumos 6459 (void) execv(cmd, emptyargv); 6460#else 6461 (void) execvp(cmd, emptyargv); 6462#endif 6463 ztest_dump_core = B_FALSE; 6464 fatal(B_TRUE, "exec failed: %s", cmd); 6465 } 6466 6467 if (cmdbuf != NULL) { 6468 umem_free(cmdbuf, MAXPATHLEN); 6469 cmd = NULL; 6470 } 6471 6472 while (waitpid(pid, &status, 0) != pid) 6473 continue; 6474 if (statusp != NULL) 6475 *statusp = status; 6476 6477 if (WIFEXITED(status)) { 6478 if (WEXITSTATUS(status) != 0) { 6479 (void) fprintf(stderr, "child exited with code %d\n", 6480 WEXITSTATUS(status)); 6481 exit(2); 6482 } 6483 return (B_FALSE); 6484 } else if (WIFSIGNALED(status)) { 6485 if (!ignorekill || WTERMSIG(status) != SIGKILL) { 6486 (void) fprintf(stderr, "child died with signal %d\n", 6487 WTERMSIG(status)); 6488 exit(3); 6489 } 6490 return (B_TRUE); 6491 } else { 6492 (void) fprintf(stderr, "something strange happened to child\n"); 6493 exit(4); 6494 /* NOTREACHED */ 6495 } 6496} 6497 6498static void 6499ztest_run_init(void) 6500{ 6501 ztest_shared_t *zs = ztest_shared; 6502 6503 ASSERT(ztest_opts.zo_init != 0); 6504 6505 /* 6506 * Blow away any existing copy of zpool.cache 6507 */ 6508 (void) remove(spa_config_path); 6509 6510 /* 6511 * Create and initialize our storage pool. 6512 */ 6513 for (int i = 1; i <= ztest_opts.zo_init; i++) { 6514 bzero(zs, sizeof (ztest_shared_t)); 6515 if (ztest_opts.zo_verbose >= 3 && 6516 ztest_opts.zo_init != 1) { 6517 (void) printf("ztest_init(), pass %d\n", i); 6518 } 6519 ztest_init(zs); 6520 } 6521} 6522 6523int 6524main(int argc, char **argv) 6525{ 6526 int kills = 0; 6527 int iters = 0; 6528 int older = 0; 6529 int newer = 0; 6530 ztest_shared_t *zs; 6531 ztest_info_t *zi; 6532 ztest_shared_callstate_t *zc; 6533 char timebuf[100]; 6534 char numbuf[NN_NUMBUF_SZ]; 6535 spa_t *spa; 6536 char *cmd; 6537 boolean_t hasalt; 6538 char *fd_data_str = getenv("ZTEST_FD_DATA"); 6539 6540 (void) setvbuf(stdout, NULL, _IOLBF, 0); 6541 6542 dprintf_setup(&argc, argv); 6543 zfs_deadman_synctime_ms = 300000; 6544 /* 6545 * As two-word space map entries may not come up often (especially 6546 * if pool and vdev sizes are small) we want to force at least some 6547 * of them so the feature get tested. 6548 */ 6549 zfs_force_some_double_word_sm_entries = B_TRUE; 6550 6551 ztest_fd_rand = open("/dev/urandom", O_RDONLY); 6552 ASSERT3S(ztest_fd_rand, >=, 0); 6553 6554 if (!fd_data_str) { 6555 process_options(argc, argv); 6556 6557 setup_data_fd(); 6558 setup_hdr(); 6559 setup_data(); 6560 bcopy(&ztest_opts, ztest_shared_opts, 6561 sizeof (*ztest_shared_opts)); 6562 } else { 6563 ztest_fd_data = atoi(fd_data_str); 6564 setup_data(); 6565 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts)); 6566 } 6567 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count); 6568 6569 /* Override location of zpool.cache */ 6570 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache", 6571 ztest_opts.zo_dir), !=, -1); 6572 6573 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t), 6574 UMEM_NOFAIL); 6575 zs = ztest_shared; 6576 6577 if (fd_data_str) { 6578 metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging; 6579 metaslab_df_alloc_threshold = 6580 zs->zs_metaslab_df_alloc_threshold; 6581 6582 if (zs->zs_do_init) 6583 ztest_run_init(); 6584 else 6585 ztest_run(zs); 6586 exit(0); 6587 } 6588 6589 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0); 6590 6591 if (ztest_opts.zo_verbose >= 1) { 6592 (void) printf("%llu vdevs, %d datasets, %d threads," 6593 " %llu seconds...\n", 6594 (u_longlong_t)ztest_opts.zo_vdevs, 6595 ztest_opts.zo_datasets, 6596 ztest_opts.zo_threads, 6597 (u_longlong_t)ztest_opts.zo_time); 6598 } 6599 6600 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); 6601 (void) strlcpy(cmd, getexecname(), MAXNAMELEN); 6602 6603 zs->zs_do_init = B_TRUE; 6604 if (strlen(ztest_opts.zo_alt_ztest) != 0) { 6605 if (ztest_opts.zo_verbose >= 1) { 6606 (void) printf("Executing older ztest for " 6607 "initialization: %s\n", ztest_opts.zo_alt_ztest); 6608 } 6609 VERIFY(!exec_child(ztest_opts.zo_alt_ztest, 6610 ztest_opts.zo_alt_libpath, B_FALSE, NULL)); 6611 } else { 6612 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL)); 6613 } 6614 zs->zs_do_init = B_FALSE; 6615 6616 zs->zs_proc_start = gethrtime(); 6617 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC; 6618 6619 for (int f = 0; f < ZTEST_FUNCS; f++) { 6620 zi = &ztest_info[f]; 6621 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6622 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop) 6623 zc->zc_next = UINT64_MAX; 6624 else 6625 zc->zc_next = zs->zs_proc_start + 6626 ztest_random(2 * zi->zi_interval[0] + 1); 6627 } 6628 6629 /* 6630 * Run the tests in a loop. These tests include fault injection 6631 * to verify that self-healing data works, and forced crashes 6632 * to verify that we never lose on-disk consistency. 6633 */ 6634 while (gethrtime() < zs->zs_proc_stop) { 6635 int status; 6636 boolean_t killed; 6637 6638 /* 6639 * Initialize the workload counters for each function. 6640 */ 6641 for (int f = 0; f < ZTEST_FUNCS; f++) { 6642 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6643 zc->zc_count = 0; 6644 zc->zc_time = 0; 6645 } 6646 6647 /* Set the allocation switch size */ 6648 zs->zs_metaslab_df_alloc_threshold = 6649 ztest_random(zs->zs_metaslab_sz / 4) + 1; 6650 6651 if (!hasalt || ztest_random(2) == 0) { 6652 if (hasalt && ztest_opts.zo_verbose >= 1) { 6653 (void) printf("Executing newer ztest: %s\n", 6654 cmd); 6655 } 6656 newer++; 6657 killed = exec_child(cmd, NULL, B_TRUE, &status); 6658 } else { 6659 if (hasalt && ztest_opts.zo_verbose >= 1) { 6660 (void) printf("Executing older ztest: %s\n", 6661 ztest_opts.zo_alt_ztest); 6662 } 6663 older++; 6664 killed = exec_child(ztest_opts.zo_alt_ztest, 6665 ztest_opts.zo_alt_libpath, B_TRUE, &status); 6666 } 6667 6668 if (killed) 6669 kills++; 6670 iters++; 6671 6672 if (ztest_opts.zo_verbose >= 1) { 6673 hrtime_t now = gethrtime(); 6674 6675 now = MIN(now, zs->zs_proc_stop); 6676 print_time(zs->zs_proc_stop - now, timebuf); 6677 nicenum(zs->zs_space, numbuf, sizeof (numbuf)); 6678 6679 (void) printf("Pass %3d, %8s, %3llu ENOSPC, " 6680 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", 6681 iters, 6682 WIFEXITED(status) ? "Complete" : "SIGKILL", 6683 (u_longlong_t)zs->zs_enospc_count, 6684 100.0 * zs->zs_alloc / zs->zs_space, 6685 numbuf, 6686 100.0 * (now - zs->zs_proc_start) / 6687 (ztest_opts.zo_time * NANOSEC), timebuf); 6688 } 6689 6690 if (ztest_opts.zo_verbose >= 2) { 6691 (void) printf("\nWorkload summary:\n\n"); 6692 (void) printf("%7s %9s %s\n", 6693 "Calls", "Time", "Function"); 6694 (void) printf("%7s %9s %s\n", 6695 "-----", "----", "--------"); 6696 for (int f = 0; f < ZTEST_FUNCS; f++) { 6697 Dl_info dli; 6698 6699 zi = &ztest_info[f]; 6700 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6701 print_time(zc->zc_time, timebuf); 6702 (void) dladdr((void *)zi->zi_func, &dli); 6703 (void) printf("%7llu %9s %s\n", 6704 (u_longlong_t)zc->zc_count, timebuf, 6705 dli.dli_sname); 6706 } 6707 (void) printf("\n"); 6708 } 6709 6710 /* 6711 * It's possible that we killed a child during a rename test, 6712 * in which case we'll have a 'ztest_tmp' pool lying around 6713 * instead of 'ztest'. Do a blind rename in case this happened. 6714 */ 6715 kernel_init(FREAD); 6716 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) { 6717 spa_close(spa, FTAG); 6718 } else { 6719 char tmpname[ZFS_MAX_DATASET_NAME_LEN]; 6720 kernel_fini(); 6721 kernel_init(FREAD | FWRITE); 6722 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp", 6723 ztest_opts.zo_pool); 6724 (void) spa_rename(tmpname, ztest_opts.zo_pool); 6725 } 6726 kernel_fini(); 6727 6728 ztest_run_zdb(ztest_opts.zo_pool); 6729 } 6730 6731 if (ztest_opts.zo_verbose >= 1) { 6732 if (hasalt) { 6733 (void) printf("%d runs of older ztest: %s\n", older, 6734 ztest_opts.zo_alt_ztest); 6735 (void) printf("%d runs of newer ztest: %s\n", newer, 6736 cmd); 6737 } 6738 (void) printf("%d killed, %d completed, %.0f%% kill rate\n", 6739 kills, iters - kills, (100.0 * kills) / MAX(1, iters)); 6740 } 6741 6742 umem_free(cmd, MAXNAMELEN); 6743 6744 return (0); 6745} 6746