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) 2013 Steven Hartland. All rights reserved. 26 * Copyright (c) 2014 Integros [integros.com] 27 * Copyright 2017 Joyent, Inc. 28 * Copyright 2017 RackTop Systems. 29 */ 30 31 /* 32 * The objective of this program is to provide a DMU/ZAP/SPA stress test 33 * that runs entirely in userland, is easy to use, and easy to extend. 34 * 35 * The overall design of the ztest program is as follows: 36 * 37 * (1) For each major functional area (e.g. adding vdevs to a pool, 38 * creating and destroying datasets, reading and writing objects, etc) 39 * we have a simple routine to test that functionality. These 40 * individual routines do not have to do anything "stressful". 41 * 42 * (2) We turn these simple functionality tests into a stress test by 43 * running them all in parallel, with as many threads as desired, 44 * and spread across as many datasets, objects, and vdevs as desired. 45 * 46 * (3) While all this is happening, we inject faults into the pool to 47 * verify that self-healing data really works. 48 * 49 * (4) Every time we open a dataset, we change its checksum and compression 50 * functions. Thus even individual objects vary from block to block 51 * in which checksum they use and whether they're compressed. 52 * 53 * (5) To verify that we never lose on-disk consistency after a crash, 54 * we run the entire test in a child of the main process. 55 * At random times, the child self-immolates with a SIGKILL. 56 * This is the software equivalent of pulling the power cord. 57 * The parent then runs the test again, using the existing 58 * storage pool, as many times as desired. If backwards compatibility 59 * testing is enabled ztest will sometimes run the "older" version 60 * of ztest after a SIGKILL. 61 * 62 * (6) To verify that we don't have future leaks or temporal incursions, 63 * many of the functional tests record the transaction group number 64 * as part of their data. When reading old data, they verify that 65 * the transaction group number is less than the current, open txg. 66 * If you add a new test, please do this if applicable. 67 * 68 * When run with no arguments, ztest runs for about five minutes and 69 * produces no output if successful. To get a little bit of information, 70 * specify -V. To get more information, specify -VV, and so on. 71 * 72 * To turn this into an overnight stress test, use -T to specify run time. 73 * 74 * You can ask more more vdevs [-v], datasets [-d], or threads [-t] 75 * to increase the pool capacity, fanout, and overall stress level. 76 * 77 * Use the -k option to set the desired frequency of kills. 78 * 79 * When ztest invokes itself it passes all relevant information through a 80 * temporary file which is mmap-ed in the child process. This allows shared 81 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always 82 * stored at offset 0 of this file and contains information on the size and 83 * number of shared structures in the file. The information stored in this file 84 * must remain backwards compatible with older versions of ztest so that 85 * ztest can invoke them during backwards compatibility testing (-B). 86 */ 87 88 #include <sys/zfs_context.h> 89 #include <sys/spa.h> 90 #include <sys/dmu.h> 91 #include <sys/txg.h> 92 #include <sys/dbuf.h> 93 #include <sys/zap.h> 94 #include <sys/dmu_objset.h> 95 #include <sys/poll.h> 96 #include <sys/stat.h> 97 #include <sys/time.h> 98 #include <sys/wait.h> 99 #include <sys/mman.h> 100 #include <sys/resource.h> 101 #include <sys/zio.h> 102 #include <sys/zil.h> 103 #include <sys/zil_impl.h> 104 #include <sys/vdev_impl.h> 105 #include <sys/vdev_file.h> 106 #include <sys/vdev_initialize.h> 107 #include <sys/spa_impl.h> 108 #include <sys/metaslab_impl.h> 109 #include <sys/dsl_prop.h> 110 #include <sys/dsl_dataset.h> 111 #include <sys/dsl_destroy.h> 112 #include <sys/dsl_scan.h> 113 #include <sys/zio_checksum.h> 114 #include <sys/refcount.h> 115 #include <sys/zfeature.h> 116 #include <sys/dsl_userhold.h> 117 #include <sys/abd.h> 118 #include <stdio.h> 119 #include <stdio_ext.h> 120 #include <stdlib.h> 121 #include <unistd.h> 122 #include <signal.h> 123 #include <umem.h> 124 #include <dlfcn.h> 125 #include <ctype.h> 126 #include <math.h> 127 #include <sys/fs/zfs.h> 128 #include <libnvpair.h> 129 #include <libcmdutils.h> 130 131 static int ztest_fd_data = -1; 132 static int ztest_fd_rand = -1; 133 134 typedef struct ztest_shared_hdr { 135 uint64_t zh_hdr_size; 136 uint64_t zh_opts_size; 137 uint64_t zh_size; 138 uint64_t zh_stats_size; 139 uint64_t zh_stats_count; 140 uint64_t zh_ds_size; 141 uint64_t zh_ds_count; 142 } ztest_shared_hdr_t; 143 144 static ztest_shared_hdr_t *ztest_shared_hdr; 145 146 typedef struct ztest_shared_opts { 147 char zo_pool[ZFS_MAX_DATASET_NAME_LEN]; 148 char zo_dir[ZFS_MAX_DATASET_NAME_LEN]; 149 char zo_alt_ztest[MAXNAMELEN]; 150 char zo_alt_libpath[MAXNAMELEN]; 151 uint64_t zo_vdevs; 152 uint64_t zo_vdevtime; 153 size_t zo_vdev_size; 154 int zo_ashift; 155 int zo_mirrors; 156 int zo_raidz; 157 int zo_raidz_parity; 158 int zo_datasets; 159 int zo_threads; 160 uint64_t zo_passtime; 161 uint64_t zo_killrate; 162 int zo_verbose; 163 int zo_init; 164 uint64_t zo_time; 165 uint64_t zo_maxloops; 166 uint64_t zo_metaslab_force_ganging; 167 } ztest_shared_opts_t; 168 169 static const ztest_shared_opts_t ztest_opts_defaults = { 170 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' }, 171 .zo_dir = { '/', 't', 'm', 'p', '\0' }, 172 .zo_alt_ztest = { '\0' }, 173 .zo_alt_libpath = { '\0' }, 174 .zo_vdevs = 5, 175 .zo_ashift = SPA_MINBLOCKSHIFT, 176 .zo_mirrors = 2, 177 .zo_raidz = 4, 178 .zo_raidz_parity = 1, 179 .zo_vdev_size = SPA_MINDEVSIZE * 4, /* 256m default size */ 180 .zo_datasets = 7, 181 .zo_threads = 23, 182 .zo_passtime = 60, /* 60 seconds */ 183 .zo_killrate = 70, /* 70% kill rate */ 184 .zo_verbose = 0, 185 .zo_init = 1, 186 .zo_time = 300, /* 5 minutes */ 187 .zo_maxloops = 50, /* max loops during spa_freeze() */ 188 .zo_metaslab_force_ganging = 32 << 10 189 }; 190 191 extern uint64_t metaslab_force_ganging; 192 extern uint64_t metaslab_df_alloc_threshold; 193 extern uint64_t zfs_deadman_synctime_ms; 194 extern int metaslab_preload_limit; 195 extern boolean_t zfs_compressed_arc_enabled; 196 extern boolean_t zfs_abd_scatter_enabled; 197 extern int dmu_object_alloc_chunk_shift; 198 extern boolean_t zfs_force_some_double_word_sm_entries; 199 extern unsigned long zfs_reconstruct_indirect_damage_fraction; 200 201 static ztest_shared_opts_t *ztest_shared_opts; 202 static ztest_shared_opts_t ztest_opts; 203 204 typedef struct ztest_shared_ds { 205 uint64_t zd_seq; 206 } ztest_shared_ds_t; 207 208 static 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) 214 215 enum 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 }; 224 225 typedef struct ztest_block_tag { 226 uint64_t bt_magic; 227 uint64_t bt_objset; 228 uint64_t bt_object; 229 uint64_t bt_dnodesize; 230 uint64_t bt_offset; 231 uint64_t bt_gen; 232 uint64_t bt_txg; 233 uint64_t bt_crtxg; 234 } ztest_block_tag_t; 235 236 typedef struct bufwad { 237 uint64_t bw_index; 238 uint64_t bw_txg; 239 uint64_t bw_data; 240 } bufwad_t; 241 242 /* 243 * It would be better to use a rangelock_t per object. Unfortunately 244 * the rangelock_t is not a drop-in replacement for rl_t, because we 245 * still need to map from object ID to rangelock_t. 246 */ 247 typedef enum { 248 RL_READER, 249 RL_WRITER, 250 RL_APPEND 251 } rl_type_t; 252 253 typedef struct rll { 254 void *rll_writer; 255 int rll_readers; 256 kmutex_t rll_lock; 257 kcondvar_t rll_cv; 258 } rll_t; 259 260 typedef struct rl { 261 uint64_t rl_object; 262 uint64_t rl_offset; 263 uint64_t rl_size; 264 rll_t *rl_lock; 265 } rl_t; 266 267 #define ZTEST_RANGE_LOCKS 64 268 #define ZTEST_OBJECT_LOCKS 64 269 270 /* 271 * Object descriptor. Used as a template for object lookup/create/remove. 272 */ 273 typedef struct ztest_od { 274 uint64_t od_dir; 275 uint64_t od_object; 276 dmu_object_type_t od_type; 277 dmu_object_type_t od_crtype; 278 uint64_t od_blocksize; 279 uint64_t od_crblocksize; 280 uint64_t od_crdnodesize; 281 uint64_t od_gen; 282 uint64_t od_crgen; 283 char od_name[ZFS_MAX_DATASET_NAME_LEN]; 284 } ztest_od_t; 285 286 /* 287 * Per-dataset state. 288 */ 289 typedef struct ztest_ds { 290 ztest_shared_ds_t *zd_shared; 291 objset_t *zd_os; 292 krwlock_t zd_zilog_lock; 293 zilog_t *zd_zilog; 294 ztest_od_t *zd_od; /* debugging aid */ 295 char zd_name[ZFS_MAX_DATASET_NAME_LEN]; 296 kmutex_t zd_dirobj_lock; 297 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS]; 298 rll_t zd_range_lock[ZTEST_RANGE_LOCKS]; 299 } ztest_ds_t; 300 301 /* 302 * Per-iteration state. 303 */ 304 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id); 305 306 typedef struct ztest_info { 307 ztest_func_t *zi_func; /* test function */ 308 uint64_t zi_iters; /* iterations per execution */ 309 uint64_t *zi_interval; /* execute every <interval> seconds */ 310 } ztest_info_t; 311 312 typedef struct ztest_shared_callstate { 313 uint64_t zc_count; /* per-pass count */ 314 uint64_t zc_time; /* per-pass time */ 315 uint64_t zc_next; /* next time to call this function */ 316 } ztest_shared_callstate_t; 317 318 static ztest_shared_callstate_t *ztest_shared_callstate; 319 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c]) 320 321 /* 322 * Note: these aren't static because we want dladdr() to work. 323 */ 324 ztest_func_t ztest_dmu_read_write; 325 ztest_func_t ztest_dmu_write_parallel; 326 ztest_func_t ztest_dmu_object_alloc_free; 327 ztest_func_t ztest_dmu_object_next_chunk; 328 ztest_func_t ztest_dmu_commit_callbacks; 329 ztest_func_t ztest_zap; 330 ztest_func_t ztest_zap_parallel; 331 ztest_func_t ztest_zil_commit; 332 ztest_func_t ztest_zil_remount; 333 ztest_func_t ztest_dmu_read_write_zcopy; 334 ztest_func_t ztest_dmu_objset_create_destroy; 335 ztest_func_t ztest_dmu_prealloc; 336 ztest_func_t ztest_fzap; 337 ztest_func_t ztest_dmu_snapshot_create_destroy; 338 ztest_func_t ztest_dsl_prop_get_set; 339 ztest_func_t ztest_spa_prop_get_set; 340 ztest_func_t ztest_spa_create_destroy; 341 ztest_func_t ztest_fault_inject; 342 ztest_func_t ztest_ddt_repair; 343 ztest_func_t ztest_dmu_snapshot_hold; 344 ztest_func_t ztest_scrub; 345 ztest_func_t ztest_dsl_dataset_promote_busy; 346 ztest_func_t ztest_vdev_attach_detach; 347 ztest_func_t ztest_vdev_LUN_growth; 348 ztest_func_t ztest_vdev_add_remove; 349 ztest_func_t ztest_vdev_aux_add_remove; 350 ztest_func_t ztest_split_pool; 351 ztest_func_t ztest_reguid; 352 ztest_func_t ztest_spa_upgrade; 353 ztest_func_t ztest_device_removal; 354 ztest_func_t ztest_remap_blocks; 355 ztest_func_t ztest_spa_checkpoint_create_discard; 356 ztest_func_t ztest_initialize; 357 ztest_func_t ztest_verify_dnode_bt; 358 359 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */ 360 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */ 361 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */ 362 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */ 363 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */ 364 365 ztest_info_t ztest_info[] = { 366 { ztest_dmu_read_write, 1, &zopt_always }, 367 { ztest_dmu_write_parallel, 10, &zopt_always }, 368 { ztest_dmu_object_alloc_free, 1, &zopt_always }, 369 { ztest_dmu_object_next_chunk, 1, &zopt_sometimes }, 370 { ztest_dmu_commit_callbacks, 1, &zopt_always }, 371 { ztest_zap, 30, &zopt_always }, 372 { ztest_zap_parallel, 100, &zopt_always }, 373 { ztest_split_pool, 1, &zopt_always }, 374 { ztest_zil_commit, 1, &zopt_incessant }, 375 { ztest_zil_remount, 1, &zopt_sometimes }, 376 { ztest_dmu_read_write_zcopy, 1, &zopt_often }, 377 { ztest_dmu_objset_create_destroy, 1, &zopt_often }, 378 { ztest_dsl_prop_get_set, 1, &zopt_often }, 379 { ztest_spa_prop_get_set, 1, &zopt_sometimes }, 380 #if 0 381 { ztest_dmu_prealloc, 1, &zopt_sometimes }, 382 #endif 383 { ztest_fzap, 1, &zopt_sometimes }, 384 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes }, 385 { ztest_spa_create_destroy, 1, &zopt_sometimes }, 386 { ztest_fault_inject, 1, &zopt_sometimes }, 387 { ztest_ddt_repair, 1, &zopt_sometimes }, 388 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes }, 389 { ztest_reguid, 1, &zopt_rarely }, 390 { ztest_scrub, 1, &zopt_rarely }, 391 { ztest_spa_upgrade, 1, &zopt_rarely }, 392 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely }, 393 { ztest_vdev_attach_detach, 1, &zopt_sometimes }, 394 { ztest_vdev_LUN_growth, 1, &zopt_rarely }, 395 { ztest_vdev_add_remove, 1, 396 &ztest_opts.zo_vdevtime }, 397 { ztest_vdev_aux_add_remove, 1, 398 &ztest_opts.zo_vdevtime }, 399 { ztest_device_removal, 1, &zopt_sometimes }, 400 { ztest_remap_blocks, 1, &zopt_sometimes }, 401 { ztest_spa_checkpoint_create_discard, 1, &zopt_rarely }, 402 { ztest_initialize, 1, &zopt_sometimes }, 403 { ztest_verify_dnode_bt, 1, &zopt_sometimes } 404 }; 405 406 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) 407 408 /* 409 * The following struct is used to hold a list of uncalled commit callbacks. 410 * The callbacks are ordered by txg number. 411 */ 412 typedef struct ztest_cb_list { 413 kmutex_t zcl_callbacks_lock; 414 list_t zcl_callbacks; 415 } ztest_cb_list_t; 416 417 /* 418 * Stuff we need to share writably between parent and child. 419 */ 420 typedef struct ztest_shared { 421 boolean_t zs_do_init; 422 hrtime_t zs_proc_start; 423 hrtime_t zs_proc_stop; 424 hrtime_t zs_thread_start; 425 hrtime_t zs_thread_stop; 426 hrtime_t zs_thread_kill; 427 uint64_t zs_enospc_count; 428 uint64_t zs_vdev_next_leaf; 429 uint64_t zs_vdev_aux; 430 uint64_t zs_alloc; 431 uint64_t zs_space; 432 uint64_t zs_splits; 433 uint64_t zs_mirrors; 434 uint64_t zs_metaslab_sz; 435 uint64_t zs_metaslab_df_alloc_threshold; 436 uint64_t zs_guid; 437 } ztest_shared_t; 438 439 #define ID_PARALLEL -1ULL 440 441 static char ztest_dev_template[] = "%s/%s.%llua"; 442 static char ztest_aux_template[] = "%s/%s.%s.%llu"; 443 ztest_shared_t *ztest_shared; 444 445 static spa_t *ztest_spa = NULL; 446 static ztest_ds_t *ztest_ds; 447 448 static kmutex_t ztest_vdev_lock; 449 static boolean_t ztest_device_removal_active = B_FALSE; 450 static kmutex_t ztest_checkpoint_lock; 451 452 /* 453 * The ztest_name_lock protects the pool and dataset namespace used by 454 * the individual tests. To modify the namespace, consumers must grab 455 * this lock as writer. Grabbing the lock as reader will ensure that the 456 * namespace does not change while the lock is held. 457 */ 458 static krwlock_t ztest_name_lock; 459 460 static boolean_t ztest_dump_core = B_TRUE; 461 static boolean_t ztest_exiting; 462 463 /* Global commit callback list */ 464 static ztest_cb_list_t zcl; 465 466 enum ztest_object { 467 ZTEST_META_DNODE = 0, 468 ZTEST_DIROBJ, 469 ZTEST_OBJECTS 470 }; 471 472 static void usage(boolean_t) __NORETURN; 473 474 /* 475 * These libumem hooks provide a reasonable set of defaults for the allocator's 476 * debugging facilities. 477 */ 478 const char * 479 _umem_debug_init() 480 { 481 return ("default,verbose"); /* $UMEM_DEBUG setting */ 482 } 483 484 const char * 485 _umem_logging_init(void) 486 { 487 return ("fail,contents"); /* $UMEM_LOGGING setting */ 488 } 489 490 #define FATAL_MSG_SZ 1024 491 492 char *fatal_msg; 493 494 static void 495 fatal(int do_perror, char *message, ...) 496 { 497 va_list args; 498 int save_errno = errno; 499 char buf[FATAL_MSG_SZ]; 500 501 (void) fflush(stdout); 502 503 va_start(args, message); 504 (void) sprintf(buf, "ztest: "); 505 /* LINTED */ 506 (void) vsprintf(buf + strlen(buf), message, args); 507 va_end(args); 508 if (do_perror) { 509 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), 510 ": %s", strerror(save_errno)); 511 } 512 (void) fprintf(stderr, "%s\n", buf); 513 fatal_msg = buf; /* to ease debugging */ 514 if (ztest_dump_core) 515 abort(); 516 exit(3); 517 } 518 519 static int 520 str2shift(const char *buf) 521 { 522 const char *ends = "BKMGTPEZ"; 523 int i; 524 525 if (buf[0] == '\0') 526 return (0); 527 for (i = 0; i < strlen(ends); i++) { 528 if (toupper(buf[0]) == ends[i]) 529 break; 530 } 531 if (i == strlen(ends)) { 532 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", 533 buf); 534 usage(B_FALSE); 535 } 536 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { 537 return (10*i); 538 } 539 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); 540 usage(B_FALSE); 541 /* NOTREACHED */ 542 } 543 544 static uint64_t 545 nicenumtoull(const char *buf) 546 { 547 char *end; 548 uint64_t val; 549 550 val = strtoull(buf, &end, 0); 551 if (end == buf) { 552 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); 553 usage(B_FALSE); 554 } else if (end[0] == '.') { 555 double fval = strtod(buf, &end); 556 fval *= pow(2, str2shift(end)); 557 if (fval > UINT64_MAX) { 558 (void) fprintf(stderr, "ztest: value too large: %s\n", 559 buf); 560 usage(B_FALSE); 561 } 562 val = (uint64_t)fval; 563 } else { 564 int shift = str2shift(end); 565 if (shift >= 64 || (val << shift) >> shift != val) { 566 (void) fprintf(stderr, "ztest: value too large: %s\n", 567 buf); 568 usage(B_FALSE); 569 } 570 val <<= shift; 571 } 572 return (val); 573 } 574 575 static void 576 usage(boolean_t requested) 577 { 578 const ztest_shared_opts_t *zo = &ztest_opts_defaults; 579 580 char nice_vdev_size[NN_NUMBUF_SZ]; 581 char nice_force_ganging[NN_NUMBUF_SZ]; 582 FILE *fp = requested ? stdout : stderr; 583 584 nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size)); 585 nicenum(zo->zo_metaslab_force_ganging, nice_force_ganging, 586 sizeof (nice_force_ganging)); 587 588 (void) fprintf(fp, "Usage: %s\n" 589 "\t[-v vdevs (default: %llu)]\n" 590 "\t[-s size_of_each_vdev (default: %s)]\n" 591 "\t[-a alignment_shift (default: %d)] use 0 for random\n" 592 "\t[-m mirror_copies (default: %d)]\n" 593 "\t[-r raidz_disks (default: %d)]\n" 594 "\t[-R raidz_parity (default: %d)]\n" 595 "\t[-d datasets (default: %d)]\n" 596 "\t[-t threads (default: %d)]\n" 597 "\t[-g gang_block_threshold (default: %s)]\n" 598 "\t[-i init_count (default: %d)] initialize pool i times\n" 599 "\t[-k kill_percentage (default: %llu%%)]\n" 600 "\t[-p pool_name (default: %s)]\n" 601 "\t[-f dir (default: %s)] file directory for vdev files\n" 602 "\t[-V] verbose (use multiple times for ever more blather)\n" 603 "\t[-E] use existing pool instead of creating new one\n" 604 "\t[-T time (default: %llu sec)] total run time\n" 605 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n" 606 "\t[-P passtime (default: %llu sec)] time per pass\n" 607 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n" 608 "\t[-o variable=value] ... set global variable to an unsigned\n" 609 "\t 32-bit integer value\n" 610 "\t[-h] (print help)\n" 611 "", 612 zo->zo_pool, 613 (u_longlong_t)zo->zo_vdevs, /* -v */ 614 nice_vdev_size, /* -s */ 615 zo->zo_ashift, /* -a */ 616 zo->zo_mirrors, /* -m */ 617 zo->zo_raidz, /* -r */ 618 zo->zo_raidz_parity, /* -R */ 619 zo->zo_datasets, /* -d */ 620 zo->zo_threads, /* -t */ 621 nice_force_ganging, /* -g */ 622 zo->zo_init, /* -i */ 623 (u_longlong_t)zo->zo_killrate, /* -k */ 624 zo->zo_pool, /* -p */ 625 zo->zo_dir, /* -f */ 626 (u_longlong_t)zo->zo_time, /* -T */ 627 (u_longlong_t)zo->zo_maxloops, /* -F */ 628 (u_longlong_t)zo->zo_passtime); 629 exit(requested ? 0 : 1); 630 } 631 632 static void 633 process_options(int argc, char **argv) 634 { 635 char *path; 636 ztest_shared_opts_t *zo = &ztest_opts; 637 638 int opt; 639 uint64_t value; 640 char altdir[MAXNAMELEN] = { 0 }; 641 642 bcopy(&ztest_opts_defaults, zo, sizeof (*zo)); 643 644 while ((opt = getopt(argc, argv, 645 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF) { 646 value = 0; 647 switch (opt) { 648 case 'v': 649 case 's': 650 case 'a': 651 case 'm': 652 case 'r': 653 case 'R': 654 case 'd': 655 case 't': 656 case 'g': 657 case 'i': 658 case 'k': 659 case 'T': 660 case 'P': 661 case 'F': 662 value = nicenumtoull(optarg); 663 } 664 switch (opt) { 665 case 'v': 666 zo->zo_vdevs = value; 667 break; 668 case 's': 669 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value); 670 break; 671 case 'a': 672 zo->zo_ashift = value; 673 break; 674 case 'm': 675 zo->zo_mirrors = value; 676 break; 677 case 'r': 678 zo->zo_raidz = MAX(1, value); 679 break; 680 case 'R': 681 zo->zo_raidz_parity = MIN(MAX(value, 1), 3); 682 break; 683 case 'd': 684 zo->zo_datasets = MAX(1, value); 685 break; 686 case 't': 687 zo->zo_threads = MAX(1, value); 688 break; 689 case 'g': 690 zo->zo_metaslab_force_ganging = 691 MAX(SPA_MINBLOCKSIZE << 1, value); 692 break; 693 case 'i': 694 zo->zo_init = value; 695 break; 696 case 'k': 697 zo->zo_killrate = value; 698 break; 699 case 'p': 700 (void) strlcpy(zo->zo_pool, optarg, 701 sizeof (zo->zo_pool)); 702 break; 703 case 'f': 704 path = realpath(optarg, NULL); 705 if (path == NULL) { 706 (void) fprintf(stderr, "error: %s: %s\n", 707 optarg, strerror(errno)); 708 usage(B_FALSE); 709 } else { 710 (void) strlcpy(zo->zo_dir, path, 711 sizeof (zo->zo_dir)); 712 } 713 break; 714 case 'V': 715 zo->zo_verbose++; 716 break; 717 case 'E': 718 zo->zo_init = 0; 719 break; 720 case 'T': 721 zo->zo_time = value; 722 break; 723 case 'P': 724 zo->zo_passtime = MAX(1, value); 725 break; 726 case 'F': 727 zo->zo_maxloops = MAX(1, value); 728 break; 729 case 'B': 730 (void) strlcpy(altdir, optarg, sizeof (altdir)); 731 break; 732 case 'o': 733 if (set_global_var(optarg) != 0) 734 usage(B_FALSE); 735 break; 736 case 'h': 737 usage(B_TRUE); 738 break; 739 case '?': 740 default: 741 usage(B_FALSE); 742 break; 743 } 744 } 745 746 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1); 747 748 zo->zo_vdevtime = 749 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs : 750 UINT64_MAX >> 2); 751 752 if (strlen(altdir) > 0) { 753 char *cmd; 754 char *realaltdir; 755 char *bin; 756 char *ztest; 757 char *isa; 758 int isalen; 759 760 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 761 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 762 763 VERIFY(NULL != realpath(getexecname(), cmd)); 764 if (0 != access(altdir, F_OK)) { 765 ztest_dump_core = B_FALSE; 766 fatal(B_TRUE, "invalid alternate ztest path: %s", 767 altdir); 768 } 769 VERIFY(NULL != realpath(altdir, realaltdir)); 770 771 /* 772 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest". 773 * We want to extract <isa> to determine if we should use 774 * 32 or 64 bit binaries. 775 */ 776 bin = strstr(cmd, "/usr/bin/"); 777 ztest = strstr(bin, "/ztest"); 778 isa = bin + 9; 779 isalen = ztest - isa; 780 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest), 781 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa); 782 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath), 783 "%s/usr/lib/%.*s", realaltdir, isalen, isa); 784 785 if (0 != access(zo->zo_alt_ztest, X_OK)) { 786 ztest_dump_core = B_FALSE; 787 fatal(B_TRUE, "invalid alternate ztest: %s", 788 zo->zo_alt_ztest); 789 } else if (0 != access(zo->zo_alt_libpath, X_OK)) { 790 ztest_dump_core = B_FALSE; 791 fatal(B_TRUE, "invalid alternate lib directory %s", 792 zo->zo_alt_libpath); 793 } 794 795 umem_free(cmd, MAXPATHLEN); 796 umem_free(realaltdir, MAXPATHLEN); 797 } 798 } 799 800 static void 801 ztest_kill(ztest_shared_t *zs) 802 { 803 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa)); 804 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa)); 805 806 /* 807 * Before we kill off ztest, make sure that the config is updated. 808 * See comment above spa_write_cachefile(). 809 */ 810 mutex_enter(&spa_namespace_lock); 811 spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE); 812 mutex_exit(&spa_namespace_lock); 813 814 zfs_dbgmsg_print(FTAG); 815 (void) kill(getpid(), SIGKILL); 816 } 817 818 static uint64_t 819 ztest_random(uint64_t range) 820 { 821 uint64_t r; 822 823 ASSERT3S(ztest_fd_rand, >=, 0); 824 825 if (range == 0) 826 return (0); 827 828 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r)) 829 fatal(1, "short read from /dev/urandom"); 830 831 return (r % range); 832 } 833 834 /* ARGSUSED */ 835 static void 836 ztest_record_enospc(const char *s) 837 { 838 ztest_shared->zs_enospc_count++; 839 } 840 841 static uint64_t 842 ztest_get_ashift(void) 843 { 844 if (ztest_opts.zo_ashift == 0) 845 return (SPA_MINBLOCKSHIFT + ztest_random(5)); 846 return (ztest_opts.zo_ashift); 847 } 848 849 static nvlist_t * 850 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift) 851 { 852 char pathbuf[MAXPATHLEN]; 853 uint64_t vdev; 854 nvlist_t *file; 855 856 if (ashift == 0) 857 ashift = ztest_get_ashift(); 858 859 if (path == NULL) { 860 path = pathbuf; 861 862 if (aux != NULL) { 863 vdev = ztest_shared->zs_vdev_aux; 864 (void) snprintf(path, sizeof (pathbuf), 865 ztest_aux_template, ztest_opts.zo_dir, 866 pool == NULL ? ztest_opts.zo_pool : pool, 867 aux, vdev); 868 } else { 869 vdev = ztest_shared->zs_vdev_next_leaf++; 870 (void) snprintf(path, sizeof (pathbuf), 871 ztest_dev_template, ztest_opts.zo_dir, 872 pool == NULL ? ztest_opts.zo_pool : pool, vdev); 873 } 874 } 875 876 if (size != 0) { 877 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666); 878 if (fd == -1) 879 fatal(1, "can't open %s", path); 880 if (ftruncate(fd, size) != 0) 881 fatal(1, "can't ftruncate %s", path); 882 (void) close(fd); 883 } 884 885 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 886 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 887 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0); 888 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 889 890 return (file); 891 } 892 893 static nvlist_t * 894 make_vdev_raidz(char *path, char *aux, char *pool, size_t size, 895 uint64_t ashift, int r) 896 { 897 nvlist_t *raidz, **child; 898 int c; 899 900 if (r < 2) 901 return (make_vdev_file(path, aux, pool, size, ashift)); 902 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); 903 904 for (c = 0; c < r; c++) 905 child[c] = make_vdev_file(path, aux, pool, size, ashift); 906 907 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); 908 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, 909 VDEV_TYPE_RAIDZ) == 0); 910 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, 911 ztest_opts.zo_raidz_parity) == 0); 912 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, 913 child, r) == 0); 914 915 for (c = 0; c < r; c++) 916 nvlist_free(child[c]); 917 918 umem_free(child, r * sizeof (nvlist_t *)); 919 920 return (raidz); 921 } 922 923 static nvlist_t * 924 make_vdev_mirror(char *path, char *aux, char *pool, size_t size, 925 uint64_t ashift, int r, int m) 926 { 927 nvlist_t *mirror, **child; 928 int c; 929 930 if (m < 1) 931 return (make_vdev_raidz(path, aux, pool, size, ashift, r)); 932 933 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); 934 935 for (c = 0; c < m; c++) 936 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r); 937 938 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); 939 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, 940 VDEV_TYPE_MIRROR) == 0); 941 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, 942 child, m) == 0); 943 944 for (c = 0; c < m; c++) 945 nvlist_free(child[c]); 946 947 umem_free(child, m * sizeof (nvlist_t *)); 948 949 return (mirror); 950 } 951 952 static nvlist_t * 953 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift, 954 int log, int r, int m, int t) 955 { 956 nvlist_t *root, **child; 957 int c; 958 959 ASSERT(t > 0); 960 961 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); 962 963 for (c = 0; c < t; c++) { 964 child[c] = make_vdev_mirror(path, aux, pool, size, ashift, 965 r, m); 966 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, 967 log) == 0); 968 } 969 970 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 971 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 972 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN, 973 child, t) == 0); 974 975 for (c = 0; c < t; c++) 976 nvlist_free(child[c]); 977 978 umem_free(child, t * sizeof (nvlist_t *)); 979 980 return (root); 981 } 982 983 /* 984 * Find a random spa version. Returns back a random spa version in the 985 * range [initial_version, SPA_VERSION_FEATURES]. 986 */ 987 static uint64_t 988 ztest_random_spa_version(uint64_t initial_version) 989 { 990 uint64_t version = initial_version; 991 992 if (version <= SPA_VERSION_BEFORE_FEATURES) { 993 version = version + 994 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1); 995 } 996 997 if (version > SPA_VERSION_BEFORE_FEATURES) 998 version = SPA_VERSION_FEATURES; 999 1000 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 1001 return (version); 1002 } 1003 1004 static int 1005 ztest_random_blocksize(void) 1006 { 1007 uint64_t block_shift; 1008 /* 1009 * Choose a block size >= the ashift. 1010 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks. 1011 */ 1012 int maxbs = SPA_OLD_MAXBLOCKSHIFT; 1013 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE) 1014 maxbs = 20; 1015 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1); 1016 return (1 << (SPA_MINBLOCKSHIFT + block_shift)); 1017 } 1018 1019 static int 1020 ztest_random_dnodesize(void) 1021 { 1022 int slots; 1023 int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT; 1024 1025 if (max_slots == DNODE_MIN_SLOTS) 1026 return (DNODE_MIN_SIZE); 1027 1028 /* 1029 * Weight the random distribution more heavily toward smaller 1030 * dnode sizes since that is more likely to reflect real-world 1031 * usage. 1032 */ 1033 ASSERT3U(max_slots, >, 4); 1034 switch (ztest_random(10)) { 1035 case 0: 1036 slots = 5 + ztest_random(max_slots - 4); 1037 break; 1038 case 1 ... 4: 1039 slots = 2 + ztest_random(3); 1040 break; 1041 default: 1042 slots = 1; 1043 break; 1044 } 1045 1046 return (slots << DNODE_SHIFT); 1047 } 1048 1049 static int 1050 ztest_random_ibshift(void) 1051 { 1052 return (DN_MIN_INDBLKSHIFT + 1053 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1)); 1054 } 1055 1056 static uint64_t 1057 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok) 1058 { 1059 uint64_t top; 1060 vdev_t *rvd = spa->spa_root_vdev; 1061 vdev_t *tvd; 1062 1063 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1064 1065 do { 1066 top = ztest_random(rvd->vdev_children); 1067 tvd = rvd->vdev_child[top]; 1068 } while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) || 1069 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL); 1070 1071 return (top); 1072 } 1073 1074 static uint64_t 1075 ztest_random_dsl_prop(zfs_prop_t prop) 1076 { 1077 uint64_t value; 1078 1079 do { 1080 value = zfs_prop_random_value(prop, ztest_random(-1ULL)); 1081 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF); 1082 1083 return (value); 1084 } 1085 1086 static int 1087 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, 1088 boolean_t inherit) 1089 { 1090 const char *propname = zfs_prop_to_name(prop); 1091 const char *valname; 1092 char setpoint[MAXPATHLEN]; 1093 uint64_t curval; 1094 int error; 1095 1096 error = dsl_prop_set_int(osname, propname, 1097 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value); 1098 1099 if (error == ENOSPC) { 1100 ztest_record_enospc(FTAG); 1101 return (error); 1102 } 1103 ASSERT0(error); 1104 1105 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint)); 1106 1107 if (ztest_opts.zo_verbose >= 6) { 1108 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0); 1109 (void) printf("%s %s = %s at '%s'\n", 1110 osname, propname, valname, setpoint); 1111 } 1112 1113 return (error); 1114 } 1115 1116 static int 1117 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value) 1118 { 1119 spa_t *spa = ztest_spa; 1120 nvlist_t *props = NULL; 1121 int error; 1122 1123 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); 1124 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0); 1125 1126 error = spa_prop_set(spa, props); 1127 1128 nvlist_free(props); 1129 1130 if (error == ENOSPC) { 1131 ztest_record_enospc(FTAG); 1132 return (error); 1133 } 1134 ASSERT0(error); 1135 1136 return (error); 1137 } 1138 1139 static void 1140 ztest_rll_init(rll_t *rll) 1141 { 1142 rll->rll_writer = NULL; 1143 rll->rll_readers = 0; 1144 mutex_init(&rll->rll_lock, NULL, USYNC_THREAD, NULL); 1145 cv_init(&rll->rll_cv, NULL, USYNC_THREAD, NULL); 1146 } 1147 1148 static void 1149 ztest_rll_destroy(rll_t *rll) 1150 { 1151 ASSERT(rll->rll_writer == NULL); 1152 ASSERT(rll->rll_readers == 0); 1153 mutex_destroy(&rll->rll_lock); 1154 cv_destroy(&rll->rll_cv); 1155 } 1156 1157 static void 1158 ztest_rll_lock(rll_t *rll, rl_type_t type) 1159 { 1160 mutex_enter(&rll->rll_lock); 1161 1162 if (type == RL_READER) { 1163 while (rll->rll_writer != NULL) 1164 cv_wait(&rll->rll_cv, &rll->rll_lock); 1165 rll->rll_readers++; 1166 } else { 1167 while (rll->rll_writer != NULL || rll->rll_readers) 1168 cv_wait(&rll->rll_cv, &rll->rll_lock); 1169 rll->rll_writer = curthread; 1170 } 1171 1172 mutex_exit(&rll->rll_lock); 1173 } 1174 1175 static void 1176 ztest_rll_unlock(rll_t *rll) 1177 { 1178 mutex_enter(&rll->rll_lock); 1179 1180 if (rll->rll_writer) { 1181 ASSERT(rll->rll_readers == 0); 1182 rll->rll_writer = NULL; 1183 } else { 1184 ASSERT(rll->rll_readers != 0); 1185 ASSERT(rll->rll_writer == NULL); 1186 rll->rll_readers--; 1187 } 1188 1189 if (rll->rll_writer == NULL && rll->rll_readers == 0) 1190 cv_broadcast(&rll->rll_cv); 1191 1192 mutex_exit(&rll->rll_lock); 1193 } 1194 1195 static void 1196 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type) 1197 { 1198 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; 1199 1200 ztest_rll_lock(rll, type); 1201 } 1202 1203 static void 1204 ztest_object_unlock(ztest_ds_t *zd, uint64_t object) 1205 { 1206 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; 1207 1208 ztest_rll_unlock(rll); 1209 } 1210 1211 static rl_t * 1212 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset, 1213 uint64_t size, rl_type_t type) 1214 { 1215 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1)); 1216 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)]; 1217 rl_t *rl; 1218 1219 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL); 1220 rl->rl_object = object; 1221 rl->rl_offset = offset; 1222 rl->rl_size = size; 1223 rl->rl_lock = rll; 1224 1225 ztest_rll_lock(rll, type); 1226 1227 return (rl); 1228 } 1229 1230 static void 1231 ztest_range_unlock(rl_t *rl) 1232 { 1233 rll_t *rll = rl->rl_lock; 1234 1235 ztest_rll_unlock(rll); 1236 1237 umem_free(rl, sizeof (*rl)); 1238 } 1239 1240 static void 1241 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os) 1242 { 1243 zd->zd_os = os; 1244 zd->zd_zilog = dmu_objset_zil(os); 1245 zd->zd_shared = szd; 1246 dmu_objset_name(os, zd->zd_name); 1247 1248 if (zd->zd_shared != NULL) 1249 zd->zd_shared->zd_seq = 0; 1250 1251 rw_init(&zd->zd_zilog_lock, NULL, USYNC_THREAD, NULL); 1252 mutex_init(&zd->zd_dirobj_lock, NULL, USYNC_THREAD, NULL); 1253 1254 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++) 1255 ztest_rll_init(&zd->zd_object_lock[l]); 1256 1257 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++) 1258 ztest_rll_init(&zd->zd_range_lock[l]); 1259 } 1260 1261 static void 1262 ztest_zd_fini(ztest_ds_t *zd) 1263 { 1264 mutex_destroy(&zd->zd_dirobj_lock); 1265 1266 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++) 1267 ztest_rll_destroy(&zd->zd_object_lock[l]); 1268 1269 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++) 1270 ztest_rll_destroy(&zd->zd_range_lock[l]); 1271 } 1272 1273 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT) 1274 1275 static uint64_t 1276 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag) 1277 { 1278 uint64_t txg; 1279 int error; 1280 1281 /* 1282 * Attempt to assign tx to some transaction group. 1283 */ 1284 error = dmu_tx_assign(tx, txg_how); 1285 if (error) { 1286 if (error == ERESTART) { 1287 ASSERT(txg_how == TXG_NOWAIT); 1288 dmu_tx_wait(tx); 1289 } else { 1290 ASSERT3U(error, ==, ENOSPC); 1291 ztest_record_enospc(tag); 1292 } 1293 dmu_tx_abort(tx); 1294 return (0); 1295 } 1296 txg = dmu_tx_get_txg(tx); 1297 ASSERT(txg != 0); 1298 return (txg); 1299 } 1300 1301 static void 1302 ztest_pattern_set(void *buf, uint64_t size, uint64_t value) 1303 { 1304 uint64_t *ip = buf; 1305 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); 1306 1307 while (ip < ip_end) 1308 *ip++ = value; 1309 } 1310 1311 static boolean_t 1312 ztest_pattern_match(void *buf, uint64_t size, uint64_t value) 1313 { 1314 uint64_t *ip = buf; 1315 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); 1316 uint64_t diff = 0; 1317 1318 while (ip < ip_end) 1319 diff |= (value - *ip++); 1320 1321 return (diff == 0); 1322 } 1323 1324 static void 1325 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object, 1326 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg, 1327 uint64_t crtxg) 1328 { 1329 bt->bt_magic = BT_MAGIC; 1330 bt->bt_objset = dmu_objset_id(os); 1331 bt->bt_object = object; 1332 bt->bt_dnodesize = dnodesize; 1333 bt->bt_offset = offset; 1334 bt->bt_gen = gen; 1335 bt->bt_txg = txg; 1336 bt->bt_crtxg = crtxg; 1337 } 1338 1339 static void 1340 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object, 1341 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg, 1342 uint64_t crtxg) 1343 { 1344 ASSERT3U(bt->bt_magic, ==, BT_MAGIC); 1345 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os)); 1346 ASSERT3U(bt->bt_object, ==, object); 1347 ASSERT3U(bt->bt_dnodesize, ==, dnodesize); 1348 ASSERT3U(bt->bt_offset, ==, offset); 1349 ASSERT3U(bt->bt_gen, <=, gen); 1350 ASSERT3U(bt->bt_txg, <=, txg); 1351 ASSERT3U(bt->bt_crtxg, ==, crtxg); 1352 } 1353 1354 static ztest_block_tag_t * 1355 ztest_bt_bonus(dmu_buf_t *db) 1356 { 1357 dmu_object_info_t doi; 1358 ztest_block_tag_t *bt; 1359 1360 dmu_object_info_from_db(db, &doi); 1361 ASSERT3U(doi.doi_bonus_size, <=, db->db_size); 1362 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt)); 1363 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt)); 1364 1365 return (bt); 1366 } 1367 1368 /* 1369 * Generate a token to fill up unused bonus buffer space. Try to make 1370 * it unique to the object, generation, and offset to verify that data 1371 * is not getting overwritten by data from other dnodes. 1372 */ 1373 #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \ 1374 (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset)) 1375 1376 /* 1377 * Fill up the unused bonus buffer region before the block tag with a 1378 * verifiable pattern. Filling the whole bonus area with non-zero data 1379 * helps ensure that all dnode traversal code properly skips the 1380 * interior regions of large dnodes. 1381 */ 1382 void 1383 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj, 1384 objset_t *os, uint64_t gen) 1385 { 1386 uint64_t *bonusp; 1387 1388 ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8)); 1389 1390 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) { 1391 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os), 1392 gen, bonusp - (uint64_t *)db->db_data); 1393 *bonusp = token; 1394 } 1395 } 1396 1397 /* 1398 * Verify that the unused area of a bonus buffer is filled with the 1399 * expected tokens. 1400 */ 1401 void 1402 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj, 1403 objset_t *os, uint64_t gen) 1404 { 1405 uint64_t *bonusp; 1406 1407 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) { 1408 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os), 1409 gen, bonusp - (uint64_t *)db->db_data); 1410 VERIFY3U(*bonusp, ==, token); 1411 } 1412 } 1413 1414 /* 1415 * ZIL logging ops 1416 */ 1417 1418 #define lrz_type lr_mode 1419 #define lrz_blocksize lr_uid 1420 #define lrz_ibshift lr_gid 1421 #define lrz_bonustype lr_rdev 1422 #define lrz_dnodesize lr_crtime[1] 1423 1424 static void 1425 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr) 1426 { 1427 char *name = (void *)(lr + 1); /* name follows lr */ 1428 size_t namesize = strlen(name) + 1; 1429 itx_t *itx; 1430 1431 if (zil_replaying(zd->zd_zilog, tx)) 1432 return; 1433 1434 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize); 1435 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1436 sizeof (*lr) + namesize - sizeof (lr_t)); 1437 1438 zil_itx_assign(zd->zd_zilog, itx, tx); 1439 } 1440 1441 static void 1442 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object) 1443 { 1444 char *name = (void *)(lr + 1); /* name follows lr */ 1445 size_t namesize = strlen(name) + 1; 1446 itx_t *itx; 1447 1448 if (zil_replaying(zd->zd_zilog, tx)) 1449 return; 1450 1451 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize); 1452 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1453 sizeof (*lr) + namesize - sizeof (lr_t)); 1454 1455 itx->itx_oid = object; 1456 zil_itx_assign(zd->zd_zilog, itx, tx); 1457 } 1458 1459 static void 1460 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr) 1461 { 1462 itx_t *itx; 1463 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES); 1464 1465 if (zil_replaying(zd->zd_zilog, tx)) 1466 return; 1467 1468 if (lr->lr_length > ZIL_MAX_LOG_DATA) 1469 write_state = WR_INDIRECT; 1470 1471 itx = zil_itx_create(TX_WRITE, 1472 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0)); 1473 1474 if (write_state == WR_COPIED && 1475 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length, 1476 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) { 1477 zil_itx_destroy(itx); 1478 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1479 write_state = WR_NEED_COPY; 1480 } 1481 itx->itx_private = zd; 1482 itx->itx_wr_state = write_state; 1483 itx->itx_sync = (ztest_random(8) == 0); 1484 1485 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1486 sizeof (*lr) - sizeof (lr_t)); 1487 1488 zil_itx_assign(zd->zd_zilog, itx, tx); 1489 } 1490 1491 static void 1492 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr) 1493 { 1494 itx_t *itx; 1495 1496 if (zil_replaying(zd->zd_zilog, tx)) 1497 return; 1498 1499 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 1500 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1501 sizeof (*lr) - sizeof (lr_t)); 1502 1503 itx->itx_sync = B_FALSE; 1504 zil_itx_assign(zd->zd_zilog, itx, tx); 1505 } 1506 1507 static void 1508 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr) 1509 { 1510 itx_t *itx; 1511 1512 if (zil_replaying(zd->zd_zilog, tx)) 1513 return; 1514 1515 itx = zil_itx_create(TX_SETATTR, sizeof (*lr)); 1516 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1517 sizeof (*lr) - sizeof (lr_t)); 1518 1519 itx->itx_sync = B_FALSE; 1520 zil_itx_assign(zd->zd_zilog, itx, tx); 1521 } 1522 1523 /* 1524 * ZIL replay ops 1525 */ 1526 static int 1527 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap) 1528 { 1529 ztest_ds_t *zd = arg1; 1530 lr_create_t *lr = arg2; 1531 char *name = (void *)(lr + 1); /* name follows lr */ 1532 objset_t *os = zd->zd_os; 1533 ztest_block_tag_t *bbt; 1534 dmu_buf_t *db; 1535 dmu_tx_t *tx; 1536 uint64_t txg; 1537 int error = 0; 1538 int bonuslen; 1539 1540 if (byteswap) 1541 byteswap_uint64_array(lr, sizeof (*lr)); 1542 1543 ASSERT(lr->lr_doid == ZTEST_DIROBJ); 1544 ASSERT(name[0] != '\0'); 1545 1546 tx = dmu_tx_create(os); 1547 1548 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name); 1549 1550 if (lr->lrz_type == DMU_OT_ZAP_OTHER) { 1551 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); 1552 } else { 1553 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1554 } 1555 1556 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1557 if (txg == 0) 1558 return (ENOSPC); 1559 1560 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid); 1561 bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize); 1562 1563 if (lr->lrz_type == DMU_OT_ZAP_OTHER) { 1564 if (lr->lr_foid == 0) { 1565 lr->lr_foid = zap_create_dnsize(os, 1566 lr->lrz_type, lr->lrz_bonustype, 1567 bonuslen, lr->lrz_dnodesize, tx); 1568 } else { 1569 error = zap_create_claim_dnsize(os, lr->lr_foid, 1570 lr->lrz_type, lr->lrz_bonustype, 1571 bonuslen, lr->lrz_dnodesize, tx); 1572 } 1573 } else { 1574 if (lr->lr_foid == 0) { 1575 lr->lr_foid = dmu_object_alloc_dnsize(os, 1576 lr->lrz_type, 0, lr->lrz_bonustype, 1577 bonuslen, lr->lrz_dnodesize, tx); 1578 } else { 1579 error = dmu_object_claim_dnsize(os, lr->lr_foid, 1580 lr->lrz_type, 0, lr->lrz_bonustype, 1581 bonuslen, lr->lrz_dnodesize, tx); 1582 } 1583 } 1584 1585 if (error) { 1586 ASSERT3U(error, ==, EEXIST); 1587 ASSERT(zd->zd_zilog->zl_replay); 1588 dmu_tx_commit(tx); 1589 return (error); 1590 } 1591 1592 ASSERT(lr->lr_foid != 0); 1593 1594 if (lr->lrz_type != DMU_OT_ZAP_OTHER) 1595 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid, 1596 lr->lrz_blocksize, lr->lrz_ibshift, tx)); 1597 1598 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1599 bbt = ztest_bt_bonus(db); 1600 dmu_buf_will_dirty(db, tx); 1601 ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL, 1602 lr->lr_gen, txg, txg); 1603 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen); 1604 dmu_buf_rele(db, FTAG); 1605 1606 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1, 1607 &lr->lr_foid, tx)); 1608 1609 (void) ztest_log_create(zd, tx, lr); 1610 1611 dmu_tx_commit(tx); 1612 1613 return (0); 1614 } 1615 1616 static int 1617 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap) 1618 { 1619 ztest_ds_t *zd = arg1; 1620 lr_remove_t *lr = arg2; 1621 char *name = (void *)(lr + 1); /* name follows lr */ 1622 objset_t *os = zd->zd_os; 1623 dmu_object_info_t doi; 1624 dmu_tx_t *tx; 1625 uint64_t object, txg; 1626 1627 if (byteswap) 1628 byteswap_uint64_array(lr, sizeof (*lr)); 1629 1630 ASSERT(lr->lr_doid == ZTEST_DIROBJ); 1631 ASSERT(name[0] != '\0'); 1632 1633 VERIFY3U(0, ==, 1634 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object)); 1635 ASSERT(object != 0); 1636 1637 ztest_object_lock(zd, object, RL_WRITER); 1638 1639 VERIFY3U(0, ==, dmu_object_info(os, object, &doi)); 1640 1641 tx = dmu_tx_create(os); 1642 1643 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name); 1644 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 1645 1646 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1647 if (txg == 0) { 1648 ztest_object_unlock(zd, object); 1649 return (ENOSPC); 1650 } 1651 1652 if (doi.doi_type == DMU_OT_ZAP_OTHER) { 1653 VERIFY3U(0, ==, zap_destroy(os, object, tx)); 1654 } else { 1655 VERIFY3U(0, ==, dmu_object_free(os, object, tx)); 1656 } 1657 1658 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx)); 1659 1660 (void) ztest_log_remove(zd, tx, lr, object); 1661 1662 dmu_tx_commit(tx); 1663 1664 ztest_object_unlock(zd, object); 1665 1666 return (0); 1667 } 1668 1669 static int 1670 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap) 1671 { 1672 ztest_ds_t *zd = arg1; 1673 lr_write_t *lr = arg2; 1674 objset_t *os = zd->zd_os; 1675 void *data = lr + 1; /* data follows lr */ 1676 uint64_t offset, length; 1677 ztest_block_tag_t *bt = data; 1678 ztest_block_tag_t *bbt; 1679 uint64_t gen, txg, lrtxg, crtxg; 1680 dmu_object_info_t doi; 1681 dmu_tx_t *tx; 1682 dmu_buf_t *db; 1683 arc_buf_t *abuf = NULL; 1684 rl_t *rl; 1685 1686 if (byteswap) 1687 byteswap_uint64_array(lr, sizeof (*lr)); 1688 1689 offset = lr->lr_offset; 1690 length = lr->lr_length; 1691 1692 /* If it's a dmu_sync() block, write the whole block */ 1693 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 1694 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 1695 if (length < blocksize) { 1696 offset -= offset % blocksize; 1697 length = blocksize; 1698 } 1699 } 1700 1701 if (bt->bt_magic == BSWAP_64(BT_MAGIC)) 1702 byteswap_uint64_array(bt, sizeof (*bt)); 1703 1704 if (bt->bt_magic != BT_MAGIC) 1705 bt = NULL; 1706 1707 ztest_object_lock(zd, lr->lr_foid, RL_READER); 1708 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER); 1709 1710 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1711 1712 dmu_object_info_from_db(db, &doi); 1713 1714 bbt = ztest_bt_bonus(db); 1715 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1716 gen = bbt->bt_gen; 1717 crtxg = bbt->bt_crtxg; 1718 lrtxg = lr->lr_common.lrc_txg; 1719 1720 tx = dmu_tx_create(os); 1721 1722 dmu_tx_hold_write(tx, lr->lr_foid, offset, length); 1723 1724 if (ztest_random(8) == 0 && length == doi.doi_data_block_size && 1725 P2PHASE(offset, length) == 0) 1726 abuf = dmu_request_arcbuf(db, length); 1727 1728 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1729 if (txg == 0) { 1730 if (abuf != NULL) 1731 dmu_return_arcbuf(abuf); 1732 dmu_buf_rele(db, FTAG); 1733 ztest_range_unlock(rl); 1734 ztest_object_unlock(zd, lr->lr_foid); 1735 return (ENOSPC); 1736 } 1737 1738 if (bt != NULL) { 1739 /* 1740 * Usually, verify the old data before writing new data -- 1741 * but not always, because we also want to verify correct 1742 * behavior when the data was not recently read into cache. 1743 */ 1744 ASSERT(offset % doi.doi_data_block_size == 0); 1745 if (ztest_random(4) != 0) { 1746 int prefetch = ztest_random(2) ? 1747 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH; 1748 ztest_block_tag_t rbt; 1749 1750 VERIFY(dmu_read(os, lr->lr_foid, offset, 1751 sizeof (rbt), &rbt, prefetch) == 0); 1752 if (rbt.bt_magic == BT_MAGIC) { 1753 ztest_bt_verify(&rbt, os, lr->lr_foid, 0, 1754 offset, gen, txg, crtxg); 1755 } 1756 } 1757 1758 /* 1759 * Writes can appear to be newer than the bonus buffer because 1760 * the ztest_get_data() callback does a dmu_read() of the 1761 * open-context data, which may be different than the data 1762 * as it was when the write was generated. 1763 */ 1764 if (zd->zd_zilog->zl_replay) { 1765 ztest_bt_verify(bt, os, lr->lr_foid, 0, offset, 1766 MAX(gen, bt->bt_gen), MAX(txg, lrtxg), 1767 bt->bt_crtxg); 1768 } 1769 1770 /* 1771 * Set the bt's gen/txg to the bonus buffer's gen/txg 1772 * so that all of the usual ASSERTs will work. 1773 */ 1774 ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg, 1775 crtxg); 1776 } 1777 1778 if (abuf == NULL) { 1779 dmu_write(os, lr->lr_foid, offset, length, data, tx); 1780 } else { 1781 bcopy(data, abuf->b_data, length); 1782 dmu_assign_arcbuf(db, offset, abuf, tx); 1783 } 1784 1785 (void) ztest_log_write(zd, tx, lr); 1786 1787 dmu_buf_rele(db, FTAG); 1788 1789 dmu_tx_commit(tx); 1790 1791 ztest_range_unlock(rl); 1792 ztest_object_unlock(zd, lr->lr_foid); 1793 1794 return (0); 1795 } 1796 1797 static int 1798 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap) 1799 { 1800 ztest_ds_t *zd = arg1; 1801 lr_truncate_t *lr = arg2; 1802 objset_t *os = zd->zd_os; 1803 dmu_tx_t *tx; 1804 uint64_t txg; 1805 rl_t *rl; 1806 1807 if (byteswap) 1808 byteswap_uint64_array(lr, sizeof (*lr)); 1809 1810 ztest_object_lock(zd, lr->lr_foid, RL_READER); 1811 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length, 1812 RL_WRITER); 1813 1814 tx = dmu_tx_create(os); 1815 1816 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length); 1817 1818 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1819 if (txg == 0) { 1820 ztest_range_unlock(rl); 1821 ztest_object_unlock(zd, lr->lr_foid); 1822 return (ENOSPC); 1823 } 1824 1825 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset, 1826 lr->lr_length, tx) == 0); 1827 1828 (void) ztest_log_truncate(zd, tx, lr); 1829 1830 dmu_tx_commit(tx); 1831 1832 ztest_range_unlock(rl); 1833 ztest_object_unlock(zd, lr->lr_foid); 1834 1835 return (0); 1836 } 1837 1838 static int 1839 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap) 1840 { 1841 ztest_ds_t *zd = arg1; 1842 lr_setattr_t *lr = arg2; 1843 objset_t *os = zd->zd_os; 1844 dmu_tx_t *tx; 1845 dmu_buf_t *db; 1846 ztest_block_tag_t *bbt; 1847 uint64_t txg, lrtxg, crtxg, dnodesize; 1848 1849 if (byteswap) 1850 byteswap_uint64_array(lr, sizeof (*lr)); 1851 1852 ztest_object_lock(zd, lr->lr_foid, RL_WRITER); 1853 1854 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1855 1856 tx = dmu_tx_create(os); 1857 dmu_tx_hold_bonus(tx, lr->lr_foid); 1858 1859 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1860 if (txg == 0) { 1861 dmu_buf_rele(db, FTAG); 1862 ztest_object_unlock(zd, lr->lr_foid); 1863 return (ENOSPC); 1864 } 1865 1866 bbt = ztest_bt_bonus(db); 1867 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1868 crtxg = bbt->bt_crtxg; 1869 lrtxg = lr->lr_common.lrc_txg; 1870 dnodesize = bbt->bt_dnodesize; 1871 1872 if (zd->zd_zilog->zl_replay) { 1873 ASSERT(lr->lr_size != 0); 1874 ASSERT(lr->lr_mode != 0); 1875 ASSERT(lrtxg != 0); 1876 } else { 1877 /* 1878 * Randomly change the size and increment the generation. 1879 */ 1880 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) * 1881 sizeof (*bbt); 1882 lr->lr_mode = bbt->bt_gen + 1; 1883 ASSERT(lrtxg == 0); 1884 } 1885 1886 /* 1887 * Verify that the current bonus buffer is not newer than our txg. 1888 */ 1889 ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode, 1890 MAX(txg, lrtxg), crtxg); 1891 1892 dmu_buf_will_dirty(db, tx); 1893 1894 ASSERT3U(lr->lr_size, >=, sizeof (*bbt)); 1895 ASSERT3U(lr->lr_size, <=, db->db_size); 1896 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx)); 1897 bbt = ztest_bt_bonus(db); 1898 1899 ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode, 1900 txg, crtxg); 1901 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen); 1902 1903 dmu_buf_rele(db, FTAG); 1904 1905 (void) ztest_log_setattr(zd, tx, lr); 1906 1907 dmu_tx_commit(tx); 1908 1909 ztest_object_unlock(zd, lr->lr_foid); 1910 1911 return (0); 1912 } 1913 1914 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { 1915 NULL, /* 0 no such transaction type */ 1916 ztest_replay_create, /* TX_CREATE */ 1917 NULL, /* TX_MKDIR */ 1918 NULL, /* TX_MKXATTR */ 1919 NULL, /* TX_SYMLINK */ 1920 ztest_replay_remove, /* TX_REMOVE */ 1921 NULL, /* TX_RMDIR */ 1922 NULL, /* TX_LINK */ 1923 NULL, /* TX_RENAME */ 1924 ztest_replay_write, /* TX_WRITE */ 1925 ztest_replay_truncate, /* TX_TRUNCATE */ 1926 ztest_replay_setattr, /* TX_SETATTR */ 1927 NULL, /* TX_ACL */ 1928 NULL, /* TX_CREATE_ACL */ 1929 NULL, /* TX_CREATE_ATTR */ 1930 NULL, /* TX_CREATE_ACL_ATTR */ 1931 NULL, /* TX_MKDIR_ACL */ 1932 NULL, /* TX_MKDIR_ATTR */ 1933 NULL, /* TX_MKDIR_ACL_ATTR */ 1934 NULL, /* TX_WRITE2 */ 1935 }; 1936 1937 /* 1938 * ZIL get_data callbacks 1939 */ 1940 1941 /* ARGSUSED */ 1942 static void 1943 ztest_get_done(zgd_t *zgd, int error) 1944 { 1945 ztest_ds_t *zd = zgd->zgd_private; 1946 uint64_t object = ((rl_t *)zgd->zgd_lr)->rl_object; 1947 1948 if (zgd->zgd_db) 1949 dmu_buf_rele(zgd->zgd_db, zgd); 1950 1951 ztest_range_unlock((rl_t *)zgd->zgd_lr); 1952 ztest_object_unlock(zd, object); 1953 1954 umem_free(zgd, sizeof (*zgd)); 1955 } 1956 1957 static int 1958 ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, 1959 zio_t *zio) 1960 { 1961 ztest_ds_t *zd = arg; 1962 objset_t *os = zd->zd_os; 1963 uint64_t object = lr->lr_foid; 1964 uint64_t offset = lr->lr_offset; 1965 uint64_t size = lr->lr_length; 1966 uint64_t txg = lr->lr_common.lrc_txg; 1967 uint64_t crtxg; 1968 dmu_object_info_t doi; 1969 dmu_buf_t *db; 1970 zgd_t *zgd; 1971 int error; 1972 1973 ASSERT3P(lwb, !=, NULL); 1974 ASSERT3P(zio, !=, NULL); 1975 ASSERT3U(size, !=, 0); 1976 1977 ztest_object_lock(zd, object, RL_READER); 1978 error = dmu_bonus_hold(os, object, FTAG, &db); 1979 if (error) { 1980 ztest_object_unlock(zd, object); 1981 return (error); 1982 } 1983 1984 crtxg = ztest_bt_bonus(db)->bt_crtxg; 1985 1986 if (crtxg == 0 || crtxg > txg) { 1987 dmu_buf_rele(db, FTAG); 1988 ztest_object_unlock(zd, object); 1989 return (ENOENT); 1990 } 1991 1992 dmu_object_info_from_db(db, &doi); 1993 dmu_buf_rele(db, FTAG); 1994 db = NULL; 1995 1996 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL); 1997 zgd->zgd_lwb = lwb; 1998 zgd->zgd_private = zd; 1999 2000 if (buf != NULL) { /* immediate write */ 2001 zgd->zgd_lr = (struct locked_range *)ztest_range_lock(zd, 2002 object, offset, size, RL_READER); 2003 2004 error = dmu_read(os, object, offset, size, buf, 2005 DMU_READ_NO_PREFETCH); 2006 ASSERT(error == 0); 2007 } else { 2008 size = doi.doi_data_block_size; 2009 if (ISP2(size)) { 2010 offset = P2ALIGN(offset, size); 2011 } else { 2012 ASSERT(offset < size); 2013 offset = 0; 2014 } 2015 2016 zgd->zgd_lr = (struct locked_range *)ztest_range_lock(zd, 2017 object, offset, size, RL_READER); 2018 2019 error = dmu_buf_hold(os, object, offset, zgd, &db, 2020 DMU_READ_NO_PREFETCH); 2021 2022 if (error == 0) { 2023 blkptr_t *bp = &lr->lr_blkptr; 2024 2025 zgd->zgd_db = db; 2026 zgd->zgd_bp = bp; 2027 2028 ASSERT(db->db_offset == offset); 2029 ASSERT(db->db_size == size); 2030 2031 error = dmu_sync(zio, lr->lr_common.lrc_txg, 2032 ztest_get_done, zgd); 2033 2034 if (error == 0) 2035 return (0); 2036 } 2037 } 2038 2039 ztest_get_done(zgd, error); 2040 2041 return (error); 2042 } 2043 2044 static void * 2045 ztest_lr_alloc(size_t lrsize, char *name) 2046 { 2047 char *lr; 2048 size_t namesize = name ? strlen(name) + 1 : 0; 2049 2050 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL); 2051 2052 if (name) 2053 bcopy(name, lr + lrsize, namesize); 2054 2055 return (lr); 2056 } 2057 2058 void 2059 ztest_lr_free(void *lr, size_t lrsize, char *name) 2060 { 2061 size_t namesize = name ? strlen(name) + 1 : 0; 2062 2063 umem_free(lr, lrsize + namesize); 2064 } 2065 2066 /* 2067 * Lookup a bunch of objects. Returns the number of objects not found. 2068 */ 2069 static int 2070 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count) 2071 { 2072 int missing = 0; 2073 int error; 2074 2075 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock)); 2076 2077 for (int i = 0; i < count; i++, od++) { 2078 od->od_object = 0; 2079 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name, 2080 sizeof (uint64_t), 1, &od->od_object); 2081 if (error) { 2082 ASSERT(error == ENOENT); 2083 ASSERT(od->od_object == 0); 2084 missing++; 2085 } else { 2086 dmu_buf_t *db; 2087 ztest_block_tag_t *bbt; 2088 dmu_object_info_t doi; 2089 2090 ASSERT(od->od_object != 0); 2091 ASSERT(missing == 0); /* there should be no gaps */ 2092 2093 ztest_object_lock(zd, od->od_object, RL_READER); 2094 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os, 2095 od->od_object, FTAG, &db)); 2096 dmu_object_info_from_db(db, &doi); 2097 bbt = ztest_bt_bonus(db); 2098 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 2099 od->od_type = doi.doi_type; 2100 od->od_blocksize = doi.doi_data_block_size; 2101 od->od_gen = bbt->bt_gen; 2102 dmu_buf_rele(db, FTAG); 2103 ztest_object_unlock(zd, od->od_object); 2104 } 2105 } 2106 2107 return (missing); 2108 } 2109 2110 static int 2111 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count) 2112 { 2113 int missing = 0; 2114 2115 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock)); 2116 2117 for (int i = 0; i < count; i++, od++) { 2118 if (missing) { 2119 od->od_object = 0; 2120 missing++; 2121 continue; 2122 } 2123 2124 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); 2125 2126 lr->lr_doid = od->od_dir; 2127 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */ 2128 lr->lrz_type = od->od_crtype; 2129 lr->lrz_blocksize = od->od_crblocksize; 2130 lr->lrz_ibshift = ztest_random_ibshift(); 2131 lr->lrz_bonustype = DMU_OT_UINT64_OTHER; 2132 lr->lrz_dnodesize = od->od_crdnodesize; 2133 lr->lr_gen = od->od_crgen; 2134 lr->lr_crtime[0] = time(NULL); 2135 2136 if (ztest_replay_create(zd, lr, B_FALSE) != 0) { 2137 ASSERT(missing == 0); 2138 od->od_object = 0; 2139 missing++; 2140 } else { 2141 od->od_object = lr->lr_foid; 2142 od->od_type = od->od_crtype; 2143 od->od_blocksize = od->od_crblocksize; 2144 od->od_gen = od->od_crgen; 2145 ASSERT(od->od_object != 0); 2146 } 2147 2148 ztest_lr_free(lr, sizeof (*lr), od->od_name); 2149 } 2150 2151 return (missing); 2152 } 2153 2154 static int 2155 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count) 2156 { 2157 int missing = 0; 2158 int error; 2159 2160 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock)); 2161 2162 od += count - 1; 2163 2164 for (int i = count - 1; i >= 0; i--, od--) { 2165 if (missing) { 2166 missing++; 2167 continue; 2168 } 2169 2170 /* 2171 * No object was found. 2172 */ 2173 if (od->od_object == 0) 2174 continue; 2175 2176 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); 2177 2178 lr->lr_doid = od->od_dir; 2179 2180 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) { 2181 ASSERT3U(error, ==, ENOSPC); 2182 missing++; 2183 } else { 2184 od->od_object = 0; 2185 } 2186 ztest_lr_free(lr, sizeof (*lr), od->od_name); 2187 } 2188 2189 return (missing); 2190 } 2191 2192 static int 2193 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size, 2194 void *data) 2195 { 2196 lr_write_t *lr; 2197 int error; 2198 2199 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL); 2200 2201 lr->lr_foid = object; 2202 lr->lr_offset = offset; 2203 lr->lr_length = size; 2204 lr->lr_blkoff = 0; 2205 BP_ZERO(&lr->lr_blkptr); 2206 2207 bcopy(data, lr + 1, size); 2208 2209 error = ztest_replay_write(zd, lr, B_FALSE); 2210 2211 ztest_lr_free(lr, sizeof (*lr) + size, NULL); 2212 2213 return (error); 2214 } 2215 2216 static int 2217 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) 2218 { 2219 lr_truncate_t *lr; 2220 int error; 2221 2222 lr = ztest_lr_alloc(sizeof (*lr), NULL); 2223 2224 lr->lr_foid = object; 2225 lr->lr_offset = offset; 2226 lr->lr_length = size; 2227 2228 error = ztest_replay_truncate(zd, lr, B_FALSE); 2229 2230 ztest_lr_free(lr, sizeof (*lr), NULL); 2231 2232 return (error); 2233 } 2234 2235 static int 2236 ztest_setattr(ztest_ds_t *zd, uint64_t object) 2237 { 2238 lr_setattr_t *lr; 2239 int error; 2240 2241 lr = ztest_lr_alloc(sizeof (*lr), NULL); 2242 2243 lr->lr_foid = object; 2244 lr->lr_size = 0; 2245 lr->lr_mode = 0; 2246 2247 error = ztest_replay_setattr(zd, lr, B_FALSE); 2248 2249 ztest_lr_free(lr, sizeof (*lr), NULL); 2250 2251 return (error); 2252 } 2253 2254 static void 2255 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) 2256 { 2257 objset_t *os = zd->zd_os; 2258 dmu_tx_t *tx; 2259 uint64_t txg; 2260 rl_t *rl; 2261 2262 txg_wait_synced(dmu_objset_pool(os), 0); 2263 2264 ztest_object_lock(zd, object, RL_READER); 2265 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER); 2266 2267 tx = dmu_tx_create(os); 2268 2269 dmu_tx_hold_write(tx, object, offset, size); 2270 2271 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 2272 2273 if (txg != 0) { 2274 dmu_prealloc(os, object, offset, size, tx); 2275 dmu_tx_commit(tx); 2276 txg_wait_synced(dmu_objset_pool(os), txg); 2277 } else { 2278 (void) dmu_free_long_range(os, object, offset, size); 2279 } 2280 2281 ztest_range_unlock(rl); 2282 ztest_object_unlock(zd, object); 2283 } 2284 2285 static void 2286 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset) 2287 { 2288 int err; 2289 ztest_block_tag_t wbt; 2290 dmu_object_info_t doi; 2291 enum ztest_io_type io_type; 2292 uint64_t blocksize; 2293 void *data; 2294 2295 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0); 2296 blocksize = doi.doi_data_block_size; 2297 data = umem_alloc(blocksize, UMEM_NOFAIL); 2298 2299 /* 2300 * Pick an i/o type at random, biased toward writing block tags. 2301 */ 2302 io_type = ztest_random(ZTEST_IO_TYPES); 2303 if (ztest_random(2) == 0) 2304 io_type = ZTEST_IO_WRITE_TAG; 2305 2306 rw_enter(&zd->zd_zilog_lock, RW_READER); 2307 2308 switch (io_type) { 2309 2310 case ZTEST_IO_WRITE_TAG: 2311 ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize, 2312 offset, 0, 0, 0); 2313 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt); 2314 break; 2315 2316 case ZTEST_IO_WRITE_PATTERN: 2317 (void) memset(data, 'a' + (object + offset) % 5, blocksize); 2318 if (ztest_random(2) == 0) { 2319 /* 2320 * Induce fletcher2 collisions to ensure that 2321 * zio_ddt_collision() detects and resolves them 2322 * when using fletcher2-verify for deduplication. 2323 */ 2324 ((uint64_t *)data)[0] ^= 1ULL << 63; 2325 ((uint64_t *)data)[4] ^= 1ULL << 63; 2326 } 2327 (void) ztest_write(zd, object, offset, blocksize, data); 2328 break; 2329 2330 case ZTEST_IO_WRITE_ZEROES: 2331 bzero(data, blocksize); 2332 (void) ztest_write(zd, object, offset, blocksize, data); 2333 break; 2334 2335 case ZTEST_IO_TRUNCATE: 2336 (void) ztest_truncate(zd, object, offset, blocksize); 2337 break; 2338 2339 case ZTEST_IO_SETATTR: 2340 (void) ztest_setattr(zd, object); 2341 break; 2342 2343 case ZTEST_IO_REWRITE: 2344 rw_enter(&ztest_name_lock, RW_READER); 2345 err = ztest_dsl_prop_set_uint64(zd->zd_name, 2346 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa), 2347 B_FALSE); 2348 VERIFY(err == 0 || err == ENOSPC); 2349 err = ztest_dsl_prop_set_uint64(zd->zd_name, 2350 ZFS_PROP_COMPRESSION, 2351 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), 2352 B_FALSE); 2353 VERIFY(err == 0 || err == ENOSPC); 2354 rw_exit(&ztest_name_lock); 2355 2356 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data, 2357 DMU_READ_NO_PREFETCH)); 2358 2359 (void) ztest_write(zd, object, offset, blocksize, data); 2360 break; 2361 } 2362 2363 rw_exit(&zd->zd_zilog_lock); 2364 2365 umem_free(data, blocksize); 2366 } 2367 2368 /* 2369 * Initialize an object description template. 2370 */ 2371 static void 2372 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index, 2373 dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize, 2374 uint64_t gen) 2375 { 2376 od->od_dir = ZTEST_DIROBJ; 2377 od->od_object = 0; 2378 2379 od->od_crtype = type; 2380 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize(); 2381 od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize(); 2382 od->od_crgen = gen; 2383 2384 od->od_type = DMU_OT_NONE; 2385 od->od_blocksize = 0; 2386 od->od_gen = 0; 2387 2388 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]", 2389 tag, (int64_t)id, index); 2390 } 2391 2392 /* 2393 * Lookup or create the objects for a test using the od template. 2394 * If the objects do not all exist, or if 'remove' is specified, 2395 * remove any existing objects and create new ones. Otherwise, 2396 * use the existing objects. 2397 */ 2398 static int 2399 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove) 2400 { 2401 int count = size / sizeof (*od); 2402 int rv = 0; 2403 2404 mutex_enter(&zd->zd_dirobj_lock); 2405 if ((ztest_lookup(zd, od, count) != 0 || remove) && 2406 (ztest_remove(zd, od, count) != 0 || 2407 ztest_create(zd, od, count) != 0)) 2408 rv = -1; 2409 zd->zd_od = od; 2410 mutex_exit(&zd->zd_dirobj_lock); 2411 2412 return (rv); 2413 } 2414 2415 /* ARGSUSED */ 2416 void 2417 ztest_zil_commit(ztest_ds_t *zd, uint64_t id) 2418 { 2419 zilog_t *zilog = zd->zd_zilog; 2420 2421 rw_enter(&zd->zd_zilog_lock, RW_READER); 2422 2423 zil_commit(zilog, ztest_random(ZTEST_OBJECTS)); 2424 2425 /* 2426 * Remember the committed values in zd, which is in parent/child 2427 * shared memory. If we die, the next iteration of ztest_run() 2428 * will verify that the log really does contain this record. 2429 */ 2430 mutex_enter(&zilog->zl_lock); 2431 ASSERT(zd->zd_shared != NULL); 2432 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq); 2433 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq; 2434 mutex_exit(&zilog->zl_lock); 2435 2436 rw_exit(&zd->zd_zilog_lock); 2437 } 2438 2439 /* 2440 * This function is designed to simulate the operations that occur during a 2441 * mount/unmount operation. We hold the dataset across these operations in an 2442 * attempt to expose any implicit assumptions about ZIL management. 2443 */ 2444 /* ARGSUSED */ 2445 void 2446 ztest_zil_remount(ztest_ds_t *zd, uint64_t id) 2447 { 2448 objset_t *os = zd->zd_os; 2449 2450 /* 2451 * We grab the zd_dirobj_lock to ensure that no other thread is 2452 * updating the zil (i.e. adding in-memory log records) and the 2453 * zd_zilog_lock to block any I/O. 2454 */ 2455 mutex_enter(&zd->zd_dirobj_lock); 2456 rw_enter(&zd->zd_zilog_lock, RW_WRITER); 2457 2458 /* zfsvfs_teardown() */ 2459 zil_close(zd->zd_zilog); 2460 2461 /* zfsvfs_setup() */ 2462 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog); 2463 zil_replay(os, zd, ztest_replay_vector); 2464 2465 rw_exit(&zd->zd_zilog_lock); 2466 mutex_exit(&zd->zd_dirobj_lock); 2467 } 2468 2469 /* 2470 * Verify that we can't destroy an active pool, create an existing pool, 2471 * or create a pool with a bad vdev spec. 2472 */ 2473 /* ARGSUSED */ 2474 void 2475 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id) 2476 { 2477 ztest_shared_opts_t *zo = &ztest_opts; 2478 spa_t *spa; 2479 nvlist_t *nvroot; 2480 2481 /* 2482 * Attempt to create using a bad file. 2483 */ 2484 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); 2485 VERIFY3U(ENOENT, ==, 2486 spa_create("ztest_bad_file", nvroot, NULL, NULL)); 2487 nvlist_free(nvroot); 2488 2489 /* 2490 * Attempt to create using a bad mirror. 2491 */ 2492 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1); 2493 VERIFY3U(ENOENT, ==, 2494 spa_create("ztest_bad_mirror", nvroot, NULL, NULL)); 2495 nvlist_free(nvroot); 2496 2497 /* 2498 * Attempt to create an existing pool. It shouldn't matter 2499 * what's in the nvroot; we should fail with EEXIST. 2500 */ 2501 rw_enter(&ztest_name_lock, RW_READER); 2502 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); 2503 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL)); 2504 nvlist_free(nvroot); 2505 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG)); 2506 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool)); 2507 spa_close(spa, FTAG); 2508 2509 rw_exit(&ztest_name_lock); 2510 } 2511 2512 /* ARGSUSED */ 2513 void 2514 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id) 2515 { 2516 spa_t *spa; 2517 uint64_t initial_version = SPA_VERSION_INITIAL; 2518 uint64_t version, newversion; 2519 nvlist_t *nvroot, *props; 2520 char *name; 2521 2522 mutex_enter(&ztest_vdev_lock); 2523 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool); 2524 2525 /* 2526 * Clean up from previous runs. 2527 */ 2528 (void) spa_destroy(name); 2529 2530 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0, 2531 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1); 2532 2533 /* 2534 * If we're configuring a RAIDZ device then make sure that the 2535 * the initial version is capable of supporting that feature. 2536 */ 2537 switch (ztest_opts.zo_raidz_parity) { 2538 case 0: 2539 case 1: 2540 initial_version = SPA_VERSION_INITIAL; 2541 break; 2542 case 2: 2543 initial_version = SPA_VERSION_RAIDZ2; 2544 break; 2545 case 3: 2546 initial_version = SPA_VERSION_RAIDZ3; 2547 break; 2548 } 2549 2550 /* 2551 * Create a pool with a spa version that can be upgraded. Pick 2552 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES. 2553 */ 2554 do { 2555 version = ztest_random_spa_version(initial_version); 2556 } while (version > SPA_VERSION_BEFORE_FEATURES); 2557 2558 props = fnvlist_alloc(); 2559 fnvlist_add_uint64(props, 2560 zpool_prop_to_name(ZPOOL_PROP_VERSION), version); 2561 VERIFY0(spa_create(name, nvroot, props, NULL)); 2562 fnvlist_free(nvroot); 2563 fnvlist_free(props); 2564 2565 VERIFY0(spa_open(name, &spa, FTAG)); 2566 VERIFY3U(spa_version(spa), ==, version); 2567 newversion = ztest_random_spa_version(version + 1); 2568 2569 if (ztest_opts.zo_verbose >= 4) { 2570 (void) printf("upgrading spa version from %llu to %llu\n", 2571 (u_longlong_t)version, (u_longlong_t)newversion); 2572 } 2573 2574 spa_upgrade(spa, newversion); 2575 VERIFY3U(spa_version(spa), >, version); 2576 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config, 2577 zpool_prop_to_name(ZPOOL_PROP_VERSION))); 2578 spa_close(spa, FTAG); 2579 2580 strfree(name); 2581 mutex_exit(&ztest_vdev_lock); 2582 } 2583 2584 static void 2585 ztest_spa_checkpoint(spa_t *spa) 2586 { 2587 ASSERT(MUTEX_HELD(&ztest_checkpoint_lock)); 2588 2589 int error = spa_checkpoint(spa->spa_name); 2590 2591 switch (error) { 2592 case 0: 2593 case ZFS_ERR_DEVRM_IN_PROGRESS: 2594 case ZFS_ERR_DISCARDING_CHECKPOINT: 2595 case ZFS_ERR_CHECKPOINT_EXISTS: 2596 break; 2597 case ENOSPC: 2598 ztest_record_enospc(FTAG); 2599 break; 2600 default: 2601 fatal(0, "spa_checkpoint(%s) = %d", spa->spa_name, error); 2602 } 2603 } 2604 2605 static void 2606 ztest_spa_discard_checkpoint(spa_t *spa) 2607 { 2608 ASSERT(MUTEX_HELD(&ztest_checkpoint_lock)); 2609 2610 int error = spa_checkpoint_discard(spa->spa_name); 2611 2612 switch (error) { 2613 case 0: 2614 case ZFS_ERR_DISCARDING_CHECKPOINT: 2615 case ZFS_ERR_NO_CHECKPOINT: 2616 break; 2617 default: 2618 fatal(0, "spa_discard_checkpoint(%s) = %d", 2619 spa->spa_name, error); 2620 } 2621 2622 } 2623 2624 /* ARGSUSED */ 2625 void 2626 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id) 2627 { 2628 spa_t *spa = ztest_spa; 2629 2630 mutex_enter(&ztest_checkpoint_lock); 2631 if (ztest_random(2) == 0) { 2632 ztest_spa_checkpoint(spa); 2633 } else { 2634 ztest_spa_discard_checkpoint(spa); 2635 } 2636 mutex_exit(&ztest_checkpoint_lock); 2637 } 2638 2639 2640 static vdev_t * 2641 vdev_lookup_by_path(vdev_t *vd, const char *path) 2642 { 2643 vdev_t *mvd; 2644 2645 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0) 2646 return (vd); 2647 2648 for (int c = 0; c < vd->vdev_children; c++) 2649 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != 2650 NULL) 2651 return (mvd); 2652 2653 return (NULL); 2654 } 2655 2656 /* 2657 * Find the first available hole which can be used as a top-level. 2658 */ 2659 int 2660 find_vdev_hole(spa_t *spa) 2661 { 2662 vdev_t *rvd = spa->spa_root_vdev; 2663 int c; 2664 2665 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV); 2666 2667 for (c = 0; c < rvd->vdev_children; c++) { 2668 vdev_t *cvd = rvd->vdev_child[c]; 2669 2670 if (cvd->vdev_ishole) 2671 break; 2672 } 2673 return (c); 2674 } 2675 2676 /* 2677 * Verify that vdev_add() works as expected. 2678 */ 2679 /* ARGSUSED */ 2680 void 2681 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id) 2682 { 2683 ztest_shared_t *zs = ztest_shared; 2684 spa_t *spa = ztest_spa; 2685 uint64_t leaves; 2686 uint64_t guid; 2687 nvlist_t *nvroot; 2688 int error; 2689 2690 mutex_enter(&ztest_vdev_lock); 2691 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz; 2692 2693 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2694 2695 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves; 2696 2697 /* 2698 * If we have slogs then remove them 1/4 of the time. 2699 */ 2700 if (spa_has_slogs(spa) && ztest_random(4) == 0) { 2701 /* 2702 * Grab the guid from the head of the log class rotor. 2703 */ 2704 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid; 2705 2706 spa_config_exit(spa, SCL_VDEV, FTAG); 2707 2708 /* 2709 * We have to grab the zs_name_lock as writer to 2710 * prevent a race between removing a slog (dmu_objset_find) 2711 * and destroying a dataset. Removing the slog will 2712 * grab a reference on the dataset which may cause 2713 * dmu_objset_destroy() to fail with EBUSY thus 2714 * leaving the dataset in an inconsistent state. 2715 */ 2716 rw_enter(&ztest_name_lock, RW_WRITER); 2717 error = spa_vdev_remove(spa, guid, B_FALSE); 2718 rw_exit(&ztest_name_lock); 2719 2720 switch (error) { 2721 case 0: 2722 case EEXIST: 2723 case ZFS_ERR_CHECKPOINT_EXISTS: 2724 case ZFS_ERR_DISCARDING_CHECKPOINT: 2725 break; 2726 default: 2727 fatal(0, "spa_vdev_remove() = %d", error); 2728 } 2729 } else { 2730 spa_config_exit(spa, SCL_VDEV, FTAG); 2731 2732 /* 2733 * Make 1/4 of the devices be log devices. 2734 */ 2735 nvroot = make_vdev_root(NULL, NULL, NULL, 2736 ztest_opts.zo_vdev_size, 0, 2737 ztest_random(4) == 0, ztest_opts.zo_raidz, 2738 zs->zs_mirrors, 1); 2739 2740 error = spa_vdev_add(spa, nvroot); 2741 nvlist_free(nvroot); 2742 2743 switch (error) { 2744 case 0: 2745 break; 2746 case ENOSPC: 2747 ztest_record_enospc("spa_vdev_add"); 2748 break; 2749 default: 2750 fatal(0, "spa_vdev_add() = %d", error); 2751 } 2752 } 2753 2754 mutex_exit(&ztest_vdev_lock); 2755 } 2756 2757 /* 2758 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected. 2759 */ 2760 /* ARGSUSED */ 2761 void 2762 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id) 2763 { 2764 ztest_shared_t *zs = ztest_shared; 2765 spa_t *spa = ztest_spa; 2766 vdev_t *rvd = spa->spa_root_vdev; 2767 spa_aux_vdev_t *sav; 2768 char *aux; 2769 uint64_t guid = 0; 2770 int error; 2771 2772 if (ztest_random(2) == 0) { 2773 sav = &spa->spa_spares; 2774 aux = ZPOOL_CONFIG_SPARES; 2775 } else { 2776 sav = &spa->spa_l2cache; 2777 aux = ZPOOL_CONFIG_L2CACHE; 2778 } 2779 2780 mutex_enter(&ztest_vdev_lock); 2781 2782 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2783 2784 if (sav->sav_count != 0 && ztest_random(4) == 0) { 2785 /* 2786 * Pick a random device to remove. 2787 */ 2788 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid; 2789 } else { 2790 /* 2791 * Find an unused device we can add. 2792 */ 2793 zs->zs_vdev_aux = 0; 2794 for (;;) { 2795 char path[MAXPATHLEN]; 2796 int c; 2797 (void) snprintf(path, sizeof (path), ztest_aux_template, 2798 ztest_opts.zo_dir, ztest_opts.zo_pool, aux, 2799 zs->zs_vdev_aux); 2800 for (c = 0; c < sav->sav_count; c++) 2801 if (strcmp(sav->sav_vdevs[c]->vdev_path, 2802 path) == 0) 2803 break; 2804 if (c == sav->sav_count && 2805 vdev_lookup_by_path(rvd, path) == NULL) 2806 break; 2807 zs->zs_vdev_aux++; 2808 } 2809 } 2810 2811 spa_config_exit(spa, SCL_VDEV, FTAG); 2812 2813 if (guid == 0) { 2814 /* 2815 * Add a new device. 2816 */ 2817 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL, 2818 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1); 2819 error = spa_vdev_add(spa, nvroot); 2820 2821 switch (error) { 2822 case 0: 2823 break; 2824 default: 2825 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error); 2826 } 2827 nvlist_free(nvroot); 2828 } else { 2829 /* 2830 * Remove an existing device. Sometimes, dirty its 2831 * vdev state first to make sure we handle removal 2832 * of devices that have pending state changes. 2833 */ 2834 if (ztest_random(2) == 0) 2835 (void) vdev_online(spa, guid, 0, NULL); 2836 2837 error = spa_vdev_remove(spa, guid, B_FALSE); 2838 2839 switch (error) { 2840 case 0: 2841 case EBUSY: 2842 case ZFS_ERR_CHECKPOINT_EXISTS: 2843 case ZFS_ERR_DISCARDING_CHECKPOINT: 2844 break; 2845 default: 2846 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error); 2847 } 2848 } 2849 2850 mutex_exit(&ztest_vdev_lock); 2851 } 2852 2853 /* 2854 * split a pool if it has mirror tlvdevs 2855 */ 2856 /* ARGSUSED */ 2857 void 2858 ztest_split_pool(ztest_ds_t *zd, uint64_t id) 2859 { 2860 ztest_shared_t *zs = ztest_shared; 2861 spa_t *spa = ztest_spa; 2862 vdev_t *rvd = spa->spa_root_vdev; 2863 nvlist_t *tree, **child, *config, *split, **schild; 2864 uint_t c, children, schildren = 0, lastlogid = 0; 2865 int error = 0; 2866 2867 mutex_enter(&ztest_vdev_lock); 2868 2869 /* ensure we have a useable config; mirrors of raidz aren't supported */ 2870 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) { 2871 mutex_exit(&ztest_vdev_lock); 2872 return; 2873 } 2874 2875 /* clean up the old pool, if any */ 2876 (void) spa_destroy("splitp"); 2877 2878 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2879 2880 /* generate a config from the existing config */ 2881 mutex_enter(&spa->spa_props_lock); 2882 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE, 2883 &tree) == 0); 2884 mutex_exit(&spa->spa_props_lock); 2885 2886 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child, 2887 &children) == 0); 2888 2889 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *)); 2890 for (c = 0; c < children; c++) { 2891 vdev_t *tvd = rvd->vdev_child[c]; 2892 nvlist_t **mchild; 2893 uint_t mchildren; 2894 2895 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) { 2896 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME, 2897 0) == 0); 2898 VERIFY(nvlist_add_string(schild[schildren], 2899 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); 2900 VERIFY(nvlist_add_uint64(schild[schildren], 2901 ZPOOL_CONFIG_IS_HOLE, 1) == 0); 2902 if (lastlogid == 0) 2903 lastlogid = schildren; 2904 ++schildren; 2905 continue; 2906 } 2907 lastlogid = 0; 2908 VERIFY(nvlist_lookup_nvlist_array(child[c], 2909 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0); 2910 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0); 2911 } 2912 2913 /* OK, create a config that can be used to split */ 2914 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0); 2915 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE, 2916 VDEV_TYPE_ROOT) == 0); 2917 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild, 2918 lastlogid != 0 ? lastlogid : schildren) == 0); 2919 2920 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0); 2921 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0); 2922 2923 for (c = 0; c < schildren; c++) 2924 nvlist_free(schild[c]); 2925 free(schild); 2926 nvlist_free(split); 2927 2928 spa_config_exit(spa, SCL_VDEV, FTAG); 2929 2930 rw_enter(&ztest_name_lock, RW_WRITER); 2931 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE); 2932 rw_exit(&ztest_name_lock); 2933 2934 nvlist_free(config); 2935 2936 if (error == 0) { 2937 (void) printf("successful split - results:\n"); 2938 mutex_enter(&spa_namespace_lock); 2939 show_pool_stats(spa); 2940 show_pool_stats(spa_lookup("splitp")); 2941 mutex_exit(&spa_namespace_lock); 2942 ++zs->zs_splits; 2943 --zs->zs_mirrors; 2944 } 2945 mutex_exit(&ztest_vdev_lock); 2946 } 2947 2948 /* 2949 * Verify that we can attach and detach devices. 2950 */ 2951 /* ARGSUSED */ 2952 void 2953 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id) 2954 { 2955 ztest_shared_t *zs = ztest_shared; 2956 spa_t *spa = ztest_spa; 2957 spa_aux_vdev_t *sav = &spa->spa_spares; 2958 vdev_t *rvd = spa->spa_root_vdev; 2959 vdev_t *oldvd, *newvd, *pvd; 2960 nvlist_t *root; 2961 uint64_t leaves; 2962 uint64_t leaf, top; 2963 uint64_t ashift = ztest_get_ashift(); 2964 uint64_t oldguid, pguid; 2965 uint64_t oldsize, newsize; 2966 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; 2967 int replacing; 2968 int oldvd_has_siblings = B_FALSE; 2969 int newvd_is_spare = B_FALSE; 2970 int oldvd_is_log; 2971 int error, expected_error; 2972 2973 mutex_enter(&ztest_vdev_lock); 2974 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; 2975 2976 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2977 2978 /* 2979 * If a vdev is in the process of being removed, its removal may 2980 * finish while we are in progress, leading to an unexpected error 2981 * value. Don't bother trying to attach while we are in the middle 2982 * of removal. 2983 */ 2984 if (ztest_device_removal_active) { 2985 spa_config_exit(spa, SCL_ALL, FTAG); 2986 mutex_exit(&ztest_vdev_lock); 2987 return; 2988 } 2989 2990 /* 2991 * Decide whether to do an attach or a replace. 2992 */ 2993 replacing = ztest_random(2); 2994 2995 /* 2996 * Pick a random top-level vdev. 2997 */ 2998 top = ztest_random_vdev_top(spa, B_TRUE); 2999 3000 /* 3001 * Pick a random leaf within it. 3002 */ 3003 leaf = ztest_random(leaves); 3004 3005 /* 3006 * Locate this vdev. 3007 */ 3008 oldvd = rvd->vdev_child[top]; 3009 if (zs->zs_mirrors >= 1) { 3010 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops); 3011 ASSERT(oldvd->vdev_children >= zs->zs_mirrors); 3012 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz]; 3013 } 3014 if (ztest_opts.zo_raidz > 1) { 3015 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops); 3016 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz); 3017 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz]; 3018 } 3019 3020 /* 3021 * If we're already doing an attach or replace, oldvd may be a 3022 * mirror vdev -- in which case, pick a random child. 3023 */ 3024 while (oldvd->vdev_children != 0) { 3025 oldvd_has_siblings = B_TRUE; 3026 ASSERT(oldvd->vdev_children >= 2); 3027 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)]; 3028 } 3029 3030 oldguid = oldvd->vdev_guid; 3031 oldsize = vdev_get_min_asize(oldvd); 3032 oldvd_is_log = oldvd->vdev_top->vdev_islog; 3033 (void) strcpy(oldpath, oldvd->vdev_path); 3034 pvd = oldvd->vdev_parent; 3035 pguid = pvd->vdev_guid; 3036 3037 /* 3038 * If oldvd has siblings, then half of the time, detach it. 3039 */ 3040 if (oldvd_has_siblings && ztest_random(2) == 0) { 3041 spa_config_exit(spa, SCL_ALL, FTAG); 3042 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE); 3043 if (error != 0 && error != ENODEV && error != EBUSY && 3044 error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS && 3045 error != ZFS_ERR_DISCARDING_CHECKPOINT) 3046 fatal(0, "detach (%s) returned %d", oldpath, error); 3047 mutex_exit(&ztest_vdev_lock); 3048 return; 3049 } 3050 3051 /* 3052 * For the new vdev, choose with equal probability between the two 3053 * standard paths (ending in either 'a' or 'b') or a random hot spare. 3054 */ 3055 if (sav->sav_count != 0 && ztest_random(3) == 0) { 3056 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)]; 3057 newvd_is_spare = B_TRUE; 3058 (void) strcpy(newpath, newvd->vdev_path); 3059 } else { 3060 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template, 3061 ztest_opts.zo_dir, ztest_opts.zo_pool, 3062 top * leaves + leaf); 3063 if (ztest_random(2) == 0) 3064 newpath[strlen(newpath) - 1] = 'b'; 3065 newvd = vdev_lookup_by_path(rvd, newpath); 3066 } 3067 3068 if (newvd) { 3069 /* 3070 * Reopen to ensure the vdev's asize field isn't stale. 3071 */ 3072 vdev_reopen(newvd); 3073 newsize = vdev_get_min_asize(newvd); 3074 } else { 3075 /* 3076 * Make newsize a little bigger or smaller than oldsize. 3077 * If it's smaller, the attach should fail. 3078 * If it's larger, and we're doing a replace, 3079 * we should get dynamic LUN growth when we're done. 3080 */ 3081 newsize = 10 * oldsize / (9 + ztest_random(3)); 3082 } 3083 3084 /* 3085 * If pvd is not a mirror or root, the attach should fail with ENOTSUP, 3086 * unless it's a replace; in that case any non-replacing parent is OK. 3087 * 3088 * If newvd is already part of the pool, it should fail with EBUSY. 3089 * 3090 * If newvd is too small, it should fail with EOVERFLOW. 3091 */ 3092 if (pvd->vdev_ops != &vdev_mirror_ops && 3093 pvd->vdev_ops != &vdev_root_ops && (!replacing || 3094 pvd->vdev_ops == &vdev_replacing_ops || 3095 pvd->vdev_ops == &vdev_spare_ops)) 3096 expected_error = ENOTSUP; 3097 else if (newvd_is_spare && (!replacing || oldvd_is_log)) 3098 expected_error = ENOTSUP; 3099 else if (newvd == oldvd) 3100 expected_error = replacing ? 0 : EBUSY; 3101 else if (vdev_lookup_by_path(rvd, newpath) != NULL) 3102 expected_error = EBUSY; 3103 else if (newsize < oldsize) 3104 expected_error = EOVERFLOW; 3105 else if (ashift > oldvd->vdev_top->vdev_ashift) 3106 expected_error = EDOM; 3107 else 3108 expected_error = 0; 3109 3110 spa_config_exit(spa, SCL_ALL, FTAG); 3111 3112 /* 3113 * Build the nvlist describing newpath. 3114 */ 3115 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0, 3116 ashift, 0, 0, 0, 1); 3117 3118 error = spa_vdev_attach(spa, oldguid, root, replacing); 3119 3120 nvlist_free(root); 3121 3122 /* 3123 * If our parent was the replacing vdev, but the replace completed, 3124 * then instead of failing with ENOTSUP we may either succeed, 3125 * fail with ENODEV, or fail with EOVERFLOW. 3126 */ 3127 if (expected_error == ENOTSUP && 3128 (error == 0 || error == ENODEV || error == EOVERFLOW)) 3129 expected_error = error; 3130 3131 /* 3132 * If someone grew the LUN, the replacement may be too small. 3133 */ 3134 if (error == EOVERFLOW || error == EBUSY) 3135 expected_error = error; 3136 3137 if (error == ZFS_ERR_CHECKPOINT_EXISTS || 3138 error == ZFS_ERR_DISCARDING_CHECKPOINT) 3139 expected_error = error; 3140 3141 /* XXX workaround 6690467 */ 3142 if (error != expected_error && expected_error != EBUSY) { 3143 fatal(0, "attach (%s %llu, %s %llu, %d) " 3144 "returned %d, expected %d", 3145 oldpath, oldsize, newpath, 3146 newsize, replacing, error, expected_error); 3147 } 3148 3149 mutex_exit(&ztest_vdev_lock); 3150 } 3151 3152 /* ARGSUSED */ 3153 void 3154 ztest_device_removal(ztest_ds_t *zd, uint64_t id) 3155 { 3156 spa_t *spa = ztest_spa; 3157 vdev_t *vd; 3158 uint64_t guid; 3159 int error; 3160 3161 mutex_enter(&ztest_vdev_lock); 3162 3163 if (ztest_device_removal_active) { 3164 mutex_exit(&ztest_vdev_lock); 3165 return; 3166 } 3167 3168 /* 3169 * Remove a random top-level vdev and wait for removal to finish. 3170 */ 3171 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 3172 vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE)); 3173 guid = vd->vdev_guid; 3174 spa_config_exit(spa, SCL_VDEV, FTAG); 3175 3176 error = spa_vdev_remove(spa, guid, B_FALSE); 3177 if (error == 0) { 3178 ztest_device_removal_active = B_TRUE; 3179 mutex_exit(&ztest_vdev_lock); 3180 3181 while (spa->spa_vdev_removal != NULL) 3182 txg_wait_synced(spa_get_dsl(spa), 0); 3183 } else { 3184 mutex_exit(&ztest_vdev_lock); 3185 return; 3186 } 3187 3188 /* 3189 * The pool needs to be scrubbed after completing device removal. 3190 * Failure to do so may result in checksum errors due to the 3191 * strategy employed by ztest_fault_inject() when selecting which 3192 * offset are redundant and can be damaged. 3193 */ 3194 error = spa_scan(spa, POOL_SCAN_SCRUB); 3195 if (error == 0) { 3196 while (dsl_scan_scrubbing(spa_get_dsl(spa))) 3197 txg_wait_synced(spa_get_dsl(spa), 0); 3198 } 3199 3200 mutex_enter(&ztest_vdev_lock); 3201 ztest_device_removal_active = B_FALSE; 3202 mutex_exit(&ztest_vdev_lock); 3203 } 3204 3205 /* 3206 * Callback function which expands the physical size of the vdev. 3207 */ 3208 vdev_t * 3209 grow_vdev(vdev_t *vd, void *arg) 3210 { 3211 spa_t *spa = vd->vdev_spa; 3212 size_t *newsize = arg; 3213 size_t fsize; 3214 int fd; 3215 3216 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); 3217 ASSERT(vd->vdev_ops->vdev_op_leaf); 3218 3219 if ((fd = open(vd->vdev_path, O_RDWR)) == -1) 3220 return (vd); 3221 3222 fsize = lseek(fd, 0, SEEK_END); 3223 (void) ftruncate(fd, *newsize); 3224 3225 if (ztest_opts.zo_verbose >= 6) { 3226 (void) printf("%s grew from %lu to %lu bytes\n", 3227 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize); 3228 } 3229 (void) close(fd); 3230 return (NULL); 3231 } 3232 3233 /* 3234 * Callback function which expands a given vdev by calling vdev_online(). 3235 */ 3236 /* ARGSUSED */ 3237 vdev_t * 3238 online_vdev(vdev_t *vd, void *arg) 3239 { 3240 spa_t *spa = vd->vdev_spa; 3241 vdev_t *tvd = vd->vdev_top; 3242 uint64_t guid = vd->vdev_guid; 3243 uint64_t generation = spa->spa_config_generation + 1; 3244 vdev_state_t newstate = VDEV_STATE_UNKNOWN; 3245 int error; 3246 3247 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); 3248 ASSERT(vd->vdev_ops->vdev_op_leaf); 3249 3250 /* Calling vdev_online will initialize the new metaslabs */ 3251 spa_config_exit(spa, SCL_STATE, spa); 3252 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate); 3253 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3254 3255 /* 3256 * If vdev_online returned an error or the underlying vdev_open 3257 * failed then we abort the expand. The only way to know that 3258 * vdev_open fails is by checking the returned newstate. 3259 */ 3260 if (error || newstate != VDEV_STATE_HEALTHY) { 3261 if (ztest_opts.zo_verbose >= 5) { 3262 (void) printf("Unable to expand vdev, state %llu, " 3263 "error %d\n", (u_longlong_t)newstate, error); 3264 } 3265 return (vd); 3266 } 3267 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY); 3268 3269 /* 3270 * Since we dropped the lock we need to ensure that we're 3271 * still talking to the original vdev. It's possible this 3272 * vdev may have been detached/replaced while we were 3273 * trying to online it. 3274 */ 3275 if (generation != spa->spa_config_generation) { 3276 if (ztest_opts.zo_verbose >= 5) { 3277 (void) printf("vdev configuration has changed, " 3278 "guid %llu, state %llu, expected gen %llu, " 3279 "got gen %llu\n", 3280 (u_longlong_t)guid, 3281 (u_longlong_t)tvd->vdev_state, 3282 (u_longlong_t)generation, 3283 (u_longlong_t)spa->spa_config_generation); 3284 } 3285 return (vd); 3286 } 3287 return (NULL); 3288 } 3289 3290 /* 3291 * Traverse the vdev tree calling the supplied function. 3292 * We continue to walk the tree until we either have walked all 3293 * children or we receive a non-NULL return from the callback. 3294 * If a NULL callback is passed, then we just return back the first 3295 * leaf vdev we encounter. 3296 */ 3297 vdev_t * 3298 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg) 3299 { 3300 if (vd->vdev_ops->vdev_op_leaf) { 3301 if (func == NULL) 3302 return (vd); 3303 else 3304 return (func(vd, arg)); 3305 } 3306 3307 for (uint_t c = 0; c < vd->vdev_children; c++) { 3308 vdev_t *cvd = vd->vdev_child[c]; 3309 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL) 3310 return (cvd); 3311 } 3312 return (NULL); 3313 } 3314 3315 /* 3316 * Verify that dynamic LUN growth works as expected. 3317 */ 3318 /* ARGSUSED */ 3319 void 3320 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id) 3321 { 3322 spa_t *spa = ztest_spa; 3323 vdev_t *vd, *tvd; 3324 metaslab_class_t *mc; 3325 metaslab_group_t *mg; 3326 size_t psize, newsize; 3327 uint64_t top; 3328 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count; 3329 3330 mutex_enter(&ztest_checkpoint_lock); 3331 mutex_enter(&ztest_vdev_lock); 3332 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3333 3334 /* 3335 * If there is a vdev removal in progress, it could complete while 3336 * we are running, in which case we would not be able to verify 3337 * that the metaslab_class space increased (because it decreases 3338 * when the device removal completes). 3339 */ 3340 if (ztest_device_removal_active) { 3341 spa_config_exit(spa, SCL_STATE, spa); 3342 mutex_exit(&ztest_vdev_lock); 3343 mutex_exit(&ztest_checkpoint_lock); 3344 return; 3345 } 3346 3347 top = ztest_random_vdev_top(spa, B_TRUE); 3348 3349 tvd = spa->spa_root_vdev->vdev_child[top]; 3350 mg = tvd->vdev_mg; 3351 mc = mg->mg_class; 3352 old_ms_count = tvd->vdev_ms_count; 3353 old_class_space = metaslab_class_get_space(mc); 3354 3355 /* 3356 * Determine the size of the first leaf vdev associated with 3357 * our top-level device. 3358 */ 3359 vd = vdev_walk_tree(tvd, NULL, NULL); 3360 ASSERT3P(vd, !=, NULL); 3361 ASSERT(vd->vdev_ops->vdev_op_leaf); 3362 3363 psize = vd->vdev_psize; 3364 3365 /* 3366 * We only try to expand the vdev if it's healthy, less than 4x its 3367 * original size, and it has a valid psize. 3368 */ 3369 if (tvd->vdev_state != VDEV_STATE_HEALTHY || 3370 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) { 3371 spa_config_exit(spa, SCL_STATE, spa); 3372 mutex_exit(&ztest_vdev_lock); 3373 mutex_exit(&ztest_checkpoint_lock); 3374 return; 3375 } 3376 ASSERT(psize > 0); 3377 newsize = psize + psize / 8; 3378 ASSERT3U(newsize, >, psize); 3379 3380 if (ztest_opts.zo_verbose >= 6) { 3381 (void) printf("Expanding LUN %s from %lu to %lu\n", 3382 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize); 3383 } 3384 3385 /* 3386 * Growing the vdev is a two step process: 3387 * 1). expand the physical size (i.e. relabel) 3388 * 2). online the vdev to create the new metaslabs 3389 */ 3390 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL || 3391 vdev_walk_tree(tvd, online_vdev, NULL) != NULL || 3392 tvd->vdev_state != VDEV_STATE_HEALTHY) { 3393 if (ztest_opts.zo_verbose >= 5) { 3394 (void) printf("Could not expand LUN because " 3395 "the vdev configuration changed.\n"); 3396 } 3397 spa_config_exit(spa, SCL_STATE, spa); 3398 mutex_exit(&ztest_vdev_lock); 3399 mutex_exit(&ztest_checkpoint_lock); 3400 return; 3401 } 3402 3403 spa_config_exit(spa, SCL_STATE, spa); 3404 3405 /* 3406 * Expanding the LUN will update the config asynchronously, 3407 * thus we must wait for the async thread to complete any 3408 * pending tasks before proceeding. 3409 */ 3410 for (;;) { 3411 boolean_t done; 3412 mutex_enter(&spa->spa_async_lock); 3413 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks); 3414 mutex_exit(&spa->spa_async_lock); 3415 if (done) 3416 break; 3417 txg_wait_synced(spa_get_dsl(spa), 0); 3418 (void) poll(NULL, 0, 100); 3419 } 3420 3421 spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3422 3423 tvd = spa->spa_root_vdev->vdev_child[top]; 3424 new_ms_count = tvd->vdev_ms_count; 3425 new_class_space = metaslab_class_get_space(mc); 3426 3427 if (tvd->vdev_mg != mg || mg->mg_class != mc) { 3428 if (ztest_opts.zo_verbose >= 5) { 3429 (void) printf("Could not verify LUN expansion due to " 3430 "intervening vdev offline or remove.\n"); 3431 } 3432 spa_config_exit(spa, SCL_STATE, spa); 3433 mutex_exit(&ztest_vdev_lock); 3434 mutex_exit(&ztest_checkpoint_lock); 3435 return; 3436 } 3437 3438 /* 3439 * Make sure we were able to grow the vdev. 3440 */ 3441 if (new_ms_count <= old_ms_count) { 3442 fatal(0, "LUN expansion failed: ms_count %llu < %llu\n", 3443 old_ms_count, new_ms_count); 3444 } 3445 3446 /* 3447 * Make sure we were able to grow the pool. 3448 */ 3449 if (new_class_space <= old_class_space) { 3450 fatal(0, "LUN expansion failed: class_space %llu < %llu\n", 3451 old_class_space, new_class_space); 3452 } 3453 3454 if (ztest_opts.zo_verbose >= 5) { 3455 char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ]; 3456 3457 nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf)); 3458 nicenum(new_class_space, newnumbuf, sizeof (newnumbuf)); 3459 (void) printf("%s grew from %s to %s\n", 3460 spa->spa_name, oldnumbuf, newnumbuf); 3461 } 3462 3463 spa_config_exit(spa, SCL_STATE, spa); 3464 mutex_exit(&ztest_vdev_lock); 3465 mutex_exit(&ztest_checkpoint_lock); 3466 } 3467 3468 /* 3469 * Verify that dmu_objset_{create,destroy,open,close} work as expected. 3470 */ 3471 /* ARGSUSED */ 3472 static void 3473 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 3474 { 3475 /* 3476 * Create the objects common to all ztest datasets. 3477 */ 3478 VERIFY(zap_create_claim(os, ZTEST_DIROBJ, 3479 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 3480 } 3481 3482 static int 3483 ztest_dataset_create(char *dsname) 3484 { 3485 uint64_t zilset = ztest_random(100); 3486 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, 3487 ztest_objset_create_cb, NULL); 3488 3489 if (err || zilset < 80) 3490 return (err); 3491 3492 if (ztest_opts.zo_verbose >= 6) 3493 (void) printf("Setting dataset %s to sync always\n", dsname); 3494 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC, 3495 ZFS_SYNC_ALWAYS, B_FALSE)); 3496 } 3497 3498 /* ARGSUSED */ 3499 static int 3500 ztest_objset_destroy_cb(const char *name, void *arg) 3501 { 3502 objset_t *os; 3503 dmu_object_info_t doi; 3504 int error; 3505 3506 /* 3507 * Verify that the dataset contains a directory object. 3508 */ 3509 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os)); 3510 error = dmu_object_info(os, ZTEST_DIROBJ, &doi); 3511 if (error != ENOENT) { 3512 /* We could have crashed in the middle of destroying it */ 3513 ASSERT0(error); 3514 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER); 3515 ASSERT3S(doi.doi_physical_blocks_512, >=, 0); 3516 } 3517 dmu_objset_disown(os, FTAG); 3518 3519 /* 3520 * Destroy the dataset. 3521 */ 3522 if (strchr(name, '@') != NULL) { 3523 VERIFY0(dsl_destroy_snapshot(name, B_TRUE)); 3524 } else { 3525 error = dsl_destroy_head(name); 3526 /* There could be a hold on this dataset */ 3527 if (error != EBUSY) 3528 ASSERT0(error); 3529 } 3530 return (0); 3531 } 3532 3533 static boolean_t 3534 ztest_snapshot_create(char *osname, uint64_t id) 3535 { 3536 char snapname[ZFS_MAX_DATASET_NAME_LEN]; 3537 int error; 3538 3539 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id); 3540 3541 error = dmu_objset_snapshot_one(osname, snapname); 3542 if (error == ENOSPC) { 3543 ztest_record_enospc(FTAG); 3544 return (B_FALSE); 3545 } 3546 if (error != 0 && error != EEXIST) { 3547 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname, 3548 snapname, error); 3549 } 3550 return (B_TRUE); 3551 } 3552 3553 static boolean_t 3554 ztest_snapshot_destroy(char *osname, uint64_t id) 3555 { 3556 char snapname[ZFS_MAX_DATASET_NAME_LEN]; 3557 int error; 3558 3559 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname, 3560 (u_longlong_t)id); 3561 3562 error = dsl_destroy_snapshot(snapname, B_FALSE); 3563 if (error != 0 && error != ENOENT) 3564 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error); 3565 return (B_TRUE); 3566 } 3567 3568 /* ARGSUSED */ 3569 void 3570 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) 3571 { 3572 ztest_ds_t zdtmp; 3573 int iters; 3574 int error; 3575 objset_t *os, *os2; 3576 char name[ZFS_MAX_DATASET_NAME_LEN]; 3577 zilog_t *zilog; 3578 3579 rw_enter(&ztest_name_lock, RW_READER); 3580 3581 (void) snprintf(name, sizeof (name), "%s/temp_%llu", 3582 ztest_opts.zo_pool, (u_longlong_t)id); 3583 3584 /* 3585 * If this dataset exists from a previous run, process its replay log 3586 * half of the time. If we don't replay it, then dmu_objset_destroy() 3587 * (invoked from ztest_objset_destroy_cb()) should just throw it away. 3588 */ 3589 if (ztest_random(2) == 0 && 3590 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) { 3591 ztest_zd_init(&zdtmp, NULL, os); 3592 zil_replay(os, &zdtmp, ztest_replay_vector); 3593 ztest_zd_fini(&zdtmp); 3594 dmu_objset_disown(os, FTAG); 3595 } 3596 3597 /* 3598 * There may be an old instance of the dataset we're about to 3599 * create lying around from a previous run. If so, destroy it 3600 * and all of its snapshots. 3601 */ 3602 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, 3603 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 3604 3605 /* 3606 * Verify that the destroyed dataset is no longer in the namespace. 3607 */ 3608 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, 3609 FTAG, &os)); 3610 3611 /* 3612 * Verify that we can create a new dataset. 3613 */ 3614 error = ztest_dataset_create(name); 3615 if (error) { 3616 if (error == ENOSPC) { 3617 ztest_record_enospc(FTAG); 3618 rw_exit(&ztest_name_lock); 3619 return; 3620 } 3621 fatal(0, "dmu_objset_create(%s) = %d", name, error); 3622 } 3623 3624 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); 3625 3626 ztest_zd_init(&zdtmp, NULL, os); 3627 3628 /* 3629 * Open the intent log for it. 3630 */ 3631 zilog = zil_open(os, ztest_get_data); 3632 3633 /* 3634 * Put some objects in there, do a little I/O to them, 3635 * and randomly take a couple of snapshots along the way. 3636 */ 3637 iters = ztest_random(5); 3638 for (int i = 0; i < iters; i++) { 3639 ztest_dmu_object_alloc_free(&zdtmp, id); 3640 if (ztest_random(iters) == 0) 3641 (void) ztest_snapshot_create(name, i); 3642 } 3643 3644 /* 3645 * Verify that we cannot create an existing dataset. 3646 */ 3647 VERIFY3U(EEXIST, ==, 3648 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL)); 3649 3650 /* 3651 * Verify that we can hold an objset that is also owned. 3652 */ 3653 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2)); 3654 dmu_objset_rele(os2, FTAG); 3655 3656 /* 3657 * Verify that we cannot own an objset that is already owned. 3658 */ 3659 VERIFY3U(EBUSY, ==, 3660 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2)); 3661 3662 zil_close(zilog); 3663 dmu_objset_disown(os, FTAG); 3664 ztest_zd_fini(&zdtmp); 3665 3666 rw_exit(&ztest_name_lock); 3667 } 3668 3669 /* 3670 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. 3671 */ 3672 void 3673 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id) 3674 { 3675 rw_enter(&ztest_name_lock, RW_READER); 3676 (void) ztest_snapshot_destroy(zd->zd_name, id); 3677 (void) ztest_snapshot_create(zd->zd_name, id); 3678 rw_exit(&ztest_name_lock); 3679 } 3680 3681 /* 3682 * Cleanup non-standard snapshots and clones. 3683 */ 3684 void 3685 ztest_dsl_dataset_cleanup(char *osname, uint64_t id) 3686 { 3687 char snap1name[ZFS_MAX_DATASET_NAME_LEN]; 3688 char clone1name[ZFS_MAX_DATASET_NAME_LEN]; 3689 char snap2name[ZFS_MAX_DATASET_NAME_LEN]; 3690 char clone2name[ZFS_MAX_DATASET_NAME_LEN]; 3691 char snap3name[ZFS_MAX_DATASET_NAME_LEN]; 3692 int error; 3693 3694 (void) snprintf(snap1name, sizeof (snap1name), 3695 "%s@s1_%llu", osname, id); 3696 (void) snprintf(clone1name, sizeof (clone1name), 3697 "%s/c1_%llu", osname, id); 3698 (void) snprintf(snap2name, sizeof (snap2name), 3699 "%s@s2_%llu", clone1name, id); 3700 (void) snprintf(clone2name, sizeof (clone2name), 3701 "%s/c2_%llu", osname, id); 3702 (void) snprintf(snap3name, sizeof (snap3name), 3703 "%s@s3_%llu", clone1name, id); 3704 3705 error = dsl_destroy_head(clone2name); 3706 if (error && error != ENOENT) 3707 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error); 3708 error = dsl_destroy_snapshot(snap3name, B_FALSE); 3709 if (error && error != ENOENT) 3710 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error); 3711 error = dsl_destroy_snapshot(snap2name, B_FALSE); 3712 if (error && error != ENOENT) 3713 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error); 3714 error = dsl_destroy_head(clone1name); 3715 if (error && error != ENOENT) 3716 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error); 3717 error = dsl_destroy_snapshot(snap1name, B_FALSE); 3718 if (error && error != ENOENT) 3719 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error); 3720 } 3721 3722 /* 3723 * Verify dsl_dataset_promote handles EBUSY 3724 */ 3725 void 3726 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) 3727 { 3728 objset_t *os; 3729 char snap1name[ZFS_MAX_DATASET_NAME_LEN]; 3730 char clone1name[ZFS_MAX_DATASET_NAME_LEN]; 3731 char snap2name[ZFS_MAX_DATASET_NAME_LEN]; 3732 char clone2name[ZFS_MAX_DATASET_NAME_LEN]; 3733 char snap3name[ZFS_MAX_DATASET_NAME_LEN]; 3734 char *osname = zd->zd_name; 3735 int error; 3736 3737 rw_enter(&ztest_name_lock, RW_READER); 3738 3739 ztest_dsl_dataset_cleanup(osname, id); 3740 3741 (void) snprintf(snap1name, sizeof (snap1name), 3742 "%s@s1_%llu", osname, id); 3743 (void) snprintf(clone1name, sizeof (clone1name), 3744 "%s/c1_%llu", osname, id); 3745 (void) snprintf(snap2name, sizeof (snap2name), 3746 "%s@s2_%llu", clone1name, id); 3747 (void) snprintf(clone2name, sizeof (clone2name), 3748 "%s/c2_%llu", osname, id); 3749 (void) snprintf(snap3name, sizeof (snap3name), 3750 "%s@s3_%llu", clone1name, id); 3751 3752 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1); 3753 if (error && error != EEXIST) { 3754 if (error == ENOSPC) { 3755 ztest_record_enospc(FTAG); 3756 goto out; 3757 } 3758 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error); 3759 } 3760 3761 error = dmu_objset_clone(clone1name, snap1name); 3762 if (error) { 3763 if (error == ENOSPC) { 3764 ztest_record_enospc(FTAG); 3765 goto out; 3766 } 3767 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error); 3768 } 3769 3770 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1); 3771 if (error && error != EEXIST) { 3772 if (error == ENOSPC) { 3773 ztest_record_enospc(FTAG); 3774 goto out; 3775 } 3776 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error); 3777 } 3778 3779 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1); 3780 if (error && error != EEXIST) { 3781 if (error == ENOSPC) { 3782 ztest_record_enospc(FTAG); 3783 goto out; 3784 } 3785 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); 3786 } 3787 3788 error = dmu_objset_clone(clone2name, snap3name); 3789 if (error) { 3790 if (error == ENOSPC) { 3791 ztest_record_enospc(FTAG); 3792 goto out; 3793 } 3794 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error); 3795 } 3796 3797 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os); 3798 if (error) 3799 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error); 3800 error = dsl_dataset_promote(clone2name, NULL); 3801 if (error == ENOSPC) { 3802 dmu_objset_disown(os, FTAG); 3803 ztest_record_enospc(FTAG); 3804 goto out; 3805 } 3806 if (error != EBUSY) 3807 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name, 3808 error); 3809 dmu_objset_disown(os, FTAG); 3810 3811 out: 3812 ztest_dsl_dataset_cleanup(osname, id); 3813 3814 rw_exit(&ztest_name_lock); 3815 } 3816 3817 /* 3818 * Verify that dmu_object_{alloc,free} work as expected. 3819 */ 3820 void 3821 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id) 3822 { 3823 ztest_od_t od[4]; 3824 int batchsize = sizeof (od) / sizeof (od[0]); 3825 3826 for (int b = 0; b < batchsize; b++) { 3827 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 3828 0, 0, 0); 3829 } 3830 3831 /* 3832 * Destroy the previous batch of objects, create a new batch, 3833 * and do some I/O on the new objects. 3834 */ 3835 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0) 3836 return; 3837 3838 while (ztest_random(4 * batchsize) != 0) 3839 ztest_io(zd, od[ztest_random(batchsize)].od_object, 3840 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 3841 } 3842 3843 /* 3844 * Rewind the global allocator to verify object allocation backfilling. 3845 */ 3846 void 3847 ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id) 3848 { 3849 objset_t *os = zd->zd_os; 3850 int dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift; 3851 uint64_t object; 3852 3853 /* 3854 * Rewind the global allocator randomly back to a lower object number 3855 * to force backfilling and reclamation of recently freed dnodes. 3856 */ 3857 mutex_enter(&os->os_obj_lock); 3858 object = ztest_random(os->os_obj_next_chunk); 3859 os->os_obj_next_chunk = P2ALIGN(object, dnodes_per_chunk); 3860 mutex_exit(&os->os_obj_lock); 3861 } 3862 3863 /* 3864 * Verify that dmu_{read,write} work as expected. 3865 */ 3866 void 3867 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id) 3868 { 3869 objset_t *os = zd->zd_os; 3870 ztest_od_t od[2]; 3871 dmu_tx_t *tx; 3872 int i, freeit, error; 3873 uint64_t n, s, txg; 3874 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; 3875 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; 3876 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t); 3877 uint64_t regions = 997; 3878 uint64_t stride = 123456789ULL; 3879 uint64_t width = 40; 3880 int free_percent = 5; 3881 3882 /* 3883 * This test uses two objects, packobj and bigobj, that are always 3884 * updated together (i.e. in the same tx) so that their contents are 3885 * in sync and can be compared. Their contents relate to each other 3886 * in a simple way: packobj is a dense array of 'bufwad' structures, 3887 * while bigobj is a sparse array of the same bufwads. Specifically, 3888 * for any index n, there are three bufwads that should be identical: 3889 * 3890 * packobj, at offset n * sizeof (bufwad_t) 3891 * bigobj, at the head of the nth chunk 3892 * bigobj, at the tail of the nth chunk 3893 * 3894 * The chunk size is arbitrary. It doesn't have to be a power of two, 3895 * and it doesn't have any relation to the object blocksize. 3896 * The only requirement is that it can hold at least two bufwads. 3897 * 3898 * Normally, we write the bufwad to each of these locations. 3899 * However, free_percent of the time we instead write zeroes to 3900 * packobj and perform a dmu_free_range() on bigobj. By comparing 3901 * bigobj to packobj, we can verify that the DMU is correctly 3902 * tracking which parts of an object are allocated and free, 3903 * and that the contents of the allocated blocks are correct. 3904 */ 3905 3906 /* 3907 * Read the directory info. If it's the first time, set things up. 3908 */ 3909 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 3910 chunksize); 3911 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0, 3912 chunksize); 3913 3914 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 3915 return; 3916 3917 bigobj = od[0].od_object; 3918 packobj = od[1].od_object; 3919 chunksize = od[0].od_gen; 3920 ASSERT(chunksize == od[1].od_gen); 3921 3922 /* 3923 * Prefetch a random chunk of the big object. 3924 * Our aim here is to get some async reads in flight 3925 * for blocks that we may free below; the DMU should 3926 * handle this race correctly. 3927 */ 3928 n = ztest_random(regions) * stride + ztest_random(width); 3929 s = 1 + ztest_random(2 * width - 1); 3930 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize, 3931 ZIO_PRIORITY_SYNC_READ); 3932 3933 /* 3934 * Pick a random index and compute the offsets into packobj and bigobj. 3935 */ 3936 n = ztest_random(regions) * stride + ztest_random(width); 3937 s = 1 + ztest_random(width - 1); 3938 3939 packoff = n * sizeof (bufwad_t); 3940 packsize = s * sizeof (bufwad_t); 3941 3942 bigoff = n * chunksize; 3943 bigsize = s * chunksize; 3944 3945 packbuf = umem_alloc(packsize, UMEM_NOFAIL); 3946 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); 3947 3948 /* 3949 * free_percent of the time, free a range of bigobj rather than 3950 * overwriting it. 3951 */ 3952 freeit = (ztest_random(100) < free_percent); 3953 3954 /* 3955 * Read the current contents of our objects. 3956 */ 3957 error = dmu_read(os, packobj, packoff, packsize, packbuf, 3958 DMU_READ_PREFETCH); 3959 ASSERT0(error); 3960 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf, 3961 DMU_READ_PREFETCH); 3962 ASSERT0(error); 3963 3964 /* 3965 * Get a tx for the mods to both packobj and bigobj. 3966 */ 3967 tx = dmu_tx_create(os); 3968 3969 dmu_tx_hold_write(tx, packobj, packoff, packsize); 3970 3971 if (freeit) 3972 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize); 3973 else 3974 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); 3975 3976 /* This accounts for setting the checksum/compression. */ 3977 dmu_tx_hold_bonus(tx, bigobj); 3978 3979 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 3980 if (txg == 0) { 3981 umem_free(packbuf, packsize); 3982 umem_free(bigbuf, bigsize); 3983 return; 3984 } 3985 3986 enum zio_checksum cksum; 3987 do { 3988 cksum = (enum zio_checksum) 3989 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM); 3990 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS); 3991 dmu_object_set_checksum(os, bigobj, cksum, tx); 3992 3993 enum zio_compress comp; 3994 do { 3995 comp = (enum zio_compress) 3996 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION); 3997 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS); 3998 dmu_object_set_compress(os, bigobj, comp, tx); 3999 4000 /* 4001 * For each index from n to n + s, verify that the existing bufwad 4002 * in packobj matches the bufwads at the head and tail of the 4003 * corresponding chunk in bigobj. Then update all three bufwads 4004 * with the new values we want to write out. 4005 */ 4006 for (i = 0; i < s; i++) { 4007 /* LINTED */ 4008 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 4009 /* LINTED */ 4010 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); 4011 /* LINTED */ 4012 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; 4013 4014 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 4015 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 4016 4017 if (pack->bw_txg > txg) 4018 fatal(0, "future leak: got %llx, open txg is %llx", 4019 pack->bw_txg, txg); 4020 4021 if (pack->bw_data != 0 && pack->bw_index != n + i) 4022 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 4023 pack->bw_index, n, i); 4024 4025 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 4026 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 4027 4028 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 4029 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 4030 4031 if (freeit) { 4032 bzero(pack, sizeof (bufwad_t)); 4033 } else { 4034 pack->bw_index = n + i; 4035 pack->bw_txg = txg; 4036 pack->bw_data = 1 + ztest_random(-2ULL); 4037 } 4038 *bigH = *pack; 4039 *bigT = *pack; 4040 } 4041 4042 /* 4043 * We've verified all the old bufwads, and made new ones. 4044 * Now write them out. 4045 */ 4046 dmu_write(os, packobj, packoff, packsize, packbuf, tx); 4047 4048 if (freeit) { 4049 if (ztest_opts.zo_verbose >= 7) { 4050 (void) printf("freeing offset %llx size %llx" 4051 " txg %llx\n", 4052 (u_longlong_t)bigoff, 4053 (u_longlong_t)bigsize, 4054 (u_longlong_t)txg); 4055 } 4056 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx)); 4057 } else { 4058 if (ztest_opts.zo_verbose >= 7) { 4059 (void) printf("writing offset %llx size %llx" 4060 " txg %llx\n", 4061 (u_longlong_t)bigoff, 4062 (u_longlong_t)bigsize, 4063 (u_longlong_t)txg); 4064 } 4065 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx); 4066 } 4067 4068 dmu_tx_commit(tx); 4069 4070 /* 4071 * Sanity check the stuff we just wrote. 4072 */ 4073 { 4074 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 4075 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 4076 4077 VERIFY(0 == dmu_read(os, packobj, packoff, 4078 packsize, packcheck, DMU_READ_PREFETCH)); 4079 VERIFY(0 == dmu_read(os, bigobj, bigoff, 4080 bigsize, bigcheck, DMU_READ_PREFETCH)); 4081 4082 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 4083 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 4084 4085 umem_free(packcheck, packsize); 4086 umem_free(bigcheck, bigsize); 4087 } 4088 4089 umem_free(packbuf, packsize); 4090 umem_free(bigbuf, bigsize); 4091 } 4092 4093 void 4094 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf, 4095 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg) 4096 { 4097 uint64_t i; 4098 bufwad_t *pack; 4099 bufwad_t *bigH; 4100 bufwad_t *bigT; 4101 4102 /* 4103 * For each index from n to n + s, verify that the existing bufwad 4104 * in packobj matches the bufwads at the head and tail of the 4105 * corresponding chunk in bigobj. Then update all three bufwads 4106 * with the new values we want to write out. 4107 */ 4108 for (i = 0; i < s; i++) { 4109 /* LINTED */ 4110 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 4111 /* LINTED */ 4112 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); 4113 /* LINTED */ 4114 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; 4115 4116 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 4117 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 4118 4119 if (pack->bw_txg > txg) 4120 fatal(0, "future leak: got %llx, open txg is %llx", 4121 pack->bw_txg, txg); 4122 4123 if (pack->bw_data != 0 && pack->bw_index != n + i) 4124 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 4125 pack->bw_index, n, i); 4126 4127 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 4128 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 4129 4130 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 4131 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 4132 4133 pack->bw_index = n + i; 4134 pack->bw_txg = txg; 4135 pack->bw_data = 1 + ztest_random(-2ULL); 4136 4137 *bigH = *pack; 4138 *bigT = *pack; 4139 } 4140 } 4141 4142 void 4143 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id) 4144 { 4145 objset_t *os = zd->zd_os; 4146 ztest_od_t od[2]; 4147 dmu_tx_t *tx; 4148 uint64_t i; 4149 int error; 4150 uint64_t n, s, txg; 4151 bufwad_t *packbuf, *bigbuf; 4152 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; 4153 uint64_t blocksize = ztest_random_blocksize(); 4154 uint64_t chunksize = blocksize; 4155 uint64_t regions = 997; 4156 uint64_t stride = 123456789ULL; 4157 uint64_t width = 9; 4158 dmu_buf_t *bonus_db; 4159 arc_buf_t **bigbuf_arcbufs; 4160 dmu_object_info_t doi; 4161 4162 /* 4163 * This test uses two objects, packobj and bigobj, that are always 4164 * updated together (i.e. in the same tx) so that their contents are 4165 * in sync and can be compared. Their contents relate to each other 4166 * in a simple way: packobj is a dense array of 'bufwad' structures, 4167 * while bigobj is a sparse array of the same bufwads. Specifically, 4168 * for any index n, there are three bufwads that should be identical: 4169 * 4170 * packobj, at offset n * sizeof (bufwad_t) 4171 * bigobj, at the head of the nth chunk 4172 * bigobj, at the tail of the nth chunk 4173 * 4174 * The chunk size is set equal to bigobj block size so that 4175 * dmu_assign_arcbuf() can be tested for object updates. 4176 */ 4177 4178 /* 4179 * Read the directory info. If it's the first time, set things up. 4180 */ 4181 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 4182 0, 0); 4183 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0, 4184 chunksize); 4185 4186 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4187 return; 4188 4189 bigobj = od[0].od_object; 4190 packobj = od[1].od_object; 4191 blocksize = od[0].od_blocksize; 4192 chunksize = blocksize; 4193 ASSERT(chunksize == od[1].od_gen); 4194 4195 VERIFY(dmu_object_info(os, bigobj, &doi) == 0); 4196 VERIFY(ISP2(doi.doi_data_block_size)); 4197 VERIFY(chunksize == doi.doi_data_block_size); 4198 VERIFY(chunksize >= 2 * sizeof (bufwad_t)); 4199 4200 /* 4201 * Pick a random index and compute the offsets into packobj and bigobj. 4202 */ 4203 n = ztest_random(regions) * stride + ztest_random(width); 4204 s = 1 + ztest_random(width - 1); 4205 4206 packoff = n * sizeof (bufwad_t); 4207 packsize = s * sizeof (bufwad_t); 4208 4209 bigoff = n * chunksize; 4210 bigsize = s * chunksize; 4211 4212 packbuf = umem_zalloc(packsize, UMEM_NOFAIL); 4213 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL); 4214 4215 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db)); 4216 4217 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL); 4218 4219 /* 4220 * Iteration 0 test zcopy for DB_UNCACHED dbufs. 4221 * Iteration 1 test zcopy to already referenced dbufs. 4222 * Iteration 2 test zcopy to dirty dbuf in the same txg. 4223 * Iteration 3 test zcopy to dbuf dirty in previous txg. 4224 * Iteration 4 test zcopy when dbuf is no longer dirty. 4225 * Iteration 5 test zcopy when it can't be done. 4226 * Iteration 6 one more zcopy write. 4227 */ 4228 for (i = 0; i < 7; i++) { 4229 uint64_t j; 4230 uint64_t off; 4231 4232 /* 4233 * In iteration 5 (i == 5) use arcbufs 4234 * that don't match bigobj blksz to test 4235 * dmu_assign_arcbuf() when it can't directly 4236 * assign an arcbuf to a dbuf. 4237 */ 4238 for (j = 0; j < s; j++) { 4239 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) { 4240 bigbuf_arcbufs[j] = 4241 dmu_request_arcbuf(bonus_db, chunksize); 4242 } else { 4243 bigbuf_arcbufs[2 * j] = 4244 dmu_request_arcbuf(bonus_db, chunksize / 2); 4245 bigbuf_arcbufs[2 * j + 1] = 4246 dmu_request_arcbuf(bonus_db, chunksize / 2); 4247 } 4248 } 4249 4250 /* 4251 * Get a tx for the mods to both packobj and bigobj. 4252 */ 4253 tx = dmu_tx_create(os); 4254 4255 dmu_tx_hold_write(tx, packobj, packoff, packsize); 4256 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); 4257 4258 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4259 if (txg == 0) { 4260 umem_free(packbuf, packsize); 4261 umem_free(bigbuf, bigsize); 4262 for (j = 0; j < s; j++) { 4263 if (i != 5 || 4264 chunksize < (SPA_MINBLOCKSIZE * 2)) { 4265 dmu_return_arcbuf(bigbuf_arcbufs[j]); 4266 } else { 4267 dmu_return_arcbuf( 4268 bigbuf_arcbufs[2 * j]); 4269 dmu_return_arcbuf( 4270 bigbuf_arcbufs[2 * j + 1]); 4271 } 4272 } 4273 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); 4274 dmu_buf_rele(bonus_db, FTAG); 4275 return; 4276 } 4277 4278 /* 4279 * 50% of the time don't read objects in the 1st iteration to 4280 * test dmu_assign_arcbuf() for the case when there're no 4281 * existing dbufs for the specified offsets. 4282 */ 4283 if (i != 0 || ztest_random(2) != 0) { 4284 error = dmu_read(os, packobj, packoff, 4285 packsize, packbuf, DMU_READ_PREFETCH); 4286 ASSERT0(error); 4287 error = dmu_read(os, bigobj, bigoff, bigsize, 4288 bigbuf, DMU_READ_PREFETCH); 4289 ASSERT0(error); 4290 } 4291 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize, 4292 n, chunksize, txg); 4293 4294 /* 4295 * We've verified all the old bufwads, and made new ones. 4296 * Now write them out. 4297 */ 4298 dmu_write(os, packobj, packoff, packsize, packbuf, tx); 4299 if (ztest_opts.zo_verbose >= 7) { 4300 (void) printf("writing offset %llx size %llx" 4301 " txg %llx\n", 4302 (u_longlong_t)bigoff, 4303 (u_longlong_t)bigsize, 4304 (u_longlong_t)txg); 4305 } 4306 for (off = bigoff, j = 0; j < s; j++, off += chunksize) { 4307 dmu_buf_t *dbt; 4308 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) { 4309 bcopy((caddr_t)bigbuf + (off - bigoff), 4310 bigbuf_arcbufs[j]->b_data, chunksize); 4311 } else { 4312 bcopy((caddr_t)bigbuf + (off - bigoff), 4313 bigbuf_arcbufs[2 * j]->b_data, 4314 chunksize / 2); 4315 bcopy((caddr_t)bigbuf + (off - bigoff) + 4316 chunksize / 2, 4317 bigbuf_arcbufs[2 * j + 1]->b_data, 4318 chunksize / 2); 4319 } 4320 4321 if (i == 1) { 4322 VERIFY(dmu_buf_hold(os, bigobj, off, 4323 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0); 4324 } 4325 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) { 4326 dmu_assign_arcbuf(bonus_db, off, 4327 bigbuf_arcbufs[j], tx); 4328 } else { 4329 dmu_assign_arcbuf(bonus_db, off, 4330 bigbuf_arcbufs[2 * j], tx); 4331 dmu_assign_arcbuf(bonus_db, 4332 off + chunksize / 2, 4333 bigbuf_arcbufs[2 * j + 1], tx); 4334 } 4335 if (i == 1) { 4336 dmu_buf_rele(dbt, FTAG); 4337 } 4338 } 4339 dmu_tx_commit(tx); 4340 4341 /* 4342 * Sanity check the stuff we just wrote. 4343 */ 4344 { 4345 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 4346 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 4347 4348 VERIFY(0 == dmu_read(os, packobj, packoff, 4349 packsize, packcheck, DMU_READ_PREFETCH)); 4350 VERIFY(0 == dmu_read(os, bigobj, bigoff, 4351 bigsize, bigcheck, DMU_READ_PREFETCH)); 4352 4353 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 4354 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 4355 4356 umem_free(packcheck, packsize); 4357 umem_free(bigcheck, bigsize); 4358 } 4359 if (i == 2) { 4360 txg_wait_open(dmu_objset_pool(os), 0); 4361 } else if (i == 3) { 4362 txg_wait_synced(dmu_objset_pool(os), 0); 4363 } 4364 } 4365 4366 dmu_buf_rele(bonus_db, FTAG); 4367 umem_free(packbuf, packsize); 4368 umem_free(bigbuf, bigsize); 4369 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); 4370 } 4371 4372 /* ARGSUSED */ 4373 void 4374 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id) 4375 { 4376 ztest_od_t od[1]; 4377 uint64_t offset = (1ULL << (ztest_random(20) + 43)) + 4378 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 4379 4380 /* 4381 * Have multiple threads write to large offsets in an object 4382 * to verify that parallel writes to an object -- even to the 4383 * same blocks within the object -- doesn't cause any trouble. 4384 */ 4385 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 4386 0, 0, 0); 4387 4388 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4389 return; 4390 4391 while (ztest_random(10) != 0) 4392 ztest_io(zd, od[0].od_object, offset); 4393 } 4394 4395 void 4396 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id) 4397 { 4398 ztest_od_t od[1]; 4399 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) + 4400 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 4401 uint64_t count = ztest_random(20) + 1; 4402 uint64_t blocksize = ztest_random_blocksize(); 4403 void *data; 4404 4405 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 4406 0, 0); 4407 4408 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4409 return; 4410 4411 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0) 4412 return; 4413 4414 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize); 4415 4416 data = umem_zalloc(blocksize, UMEM_NOFAIL); 4417 4418 while (ztest_random(count) != 0) { 4419 uint64_t randoff = offset + (ztest_random(count) * blocksize); 4420 if (ztest_write(zd, od[0].od_object, randoff, blocksize, 4421 data) != 0) 4422 break; 4423 while (ztest_random(4) != 0) 4424 ztest_io(zd, od[0].od_object, randoff); 4425 } 4426 4427 umem_free(data, blocksize); 4428 } 4429 4430 /* 4431 * Verify that zap_{create,destroy,add,remove,update} work as expected. 4432 */ 4433 #define ZTEST_ZAP_MIN_INTS 1 4434 #define ZTEST_ZAP_MAX_INTS 4 4435 #define ZTEST_ZAP_MAX_PROPS 1000 4436 4437 void 4438 ztest_zap(ztest_ds_t *zd, uint64_t id) 4439 { 4440 objset_t *os = zd->zd_os; 4441 ztest_od_t od[1]; 4442 uint64_t object; 4443 uint64_t txg, last_txg; 4444 uint64_t value[ZTEST_ZAP_MAX_INTS]; 4445 uint64_t zl_ints, zl_intsize, prop; 4446 int i, ints; 4447 dmu_tx_t *tx; 4448 char propname[100], txgname[100]; 4449 int error; 4450 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; 4451 4452 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0); 4453 4454 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4455 return; 4456 4457 object = od[0].od_object; 4458 4459 /* 4460 * Generate a known hash collision, and verify that 4461 * we can lookup and remove both entries. 4462 */ 4463 tx = dmu_tx_create(os); 4464 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4465 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4466 if (txg == 0) 4467 return; 4468 for (i = 0; i < 2; i++) { 4469 value[i] = i; 4470 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t), 4471 1, &value[i], tx)); 4472 } 4473 for (i = 0; i < 2; i++) { 4474 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i], 4475 sizeof (uint64_t), 1, &value[i], tx)); 4476 VERIFY3U(0, ==, 4477 zap_length(os, object, hc[i], &zl_intsize, &zl_ints)); 4478 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4479 ASSERT3U(zl_ints, ==, 1); 4480 } 4481 for (i = 0; i < 2; i++) { 4482 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx)); 4483 } 4484 dmu_tx_commit(tx); 4485 4486 /* 4487 * Generate a buch of random entries. 4488 */ 4489 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); 4490 4491 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 4492 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 4493 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 4494 bzero(value, sizeof (value)); 4495 last_txg = 0; 4496 4497 /* 4498 * If these zap entries already exist, validate their contents. 4499 */ 4500 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 4501 if (error == 0) { 4502 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4503 ASSERT3U(zl_ints, ==, 1); 4504 4505 VERIFY(zap_lookup(os, object, txgname, zl_intsize, 4506 zl_ints, &last_txg) == 0); 4507 4508 VERIFY(zap_length(os, object, propname, &zl_intsize, 4509 &zl_ints) == 0); 4510 4511 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4512 ASSERT3U(zl_ints, ==, ints); 4513 4514 VERIFY(zap_lookup(os, object, propname, zl_intsize, 4515 zl_ints, value) == 0); 4516 4517 for (i = 0; i < ints; i++) { 4518 ASSERT3U(value[i], ==, last_txg + object + i); 4519 } 4520 } else { 4521 ASSERT3U(error, ==, ENOENT); 4522 } 4523 4524 /* 4525 * Atomically update two entries in our zap object. 4526 * The first is named txg_%llu, and contains the txg 4527 * in which the property was last updated. The second 4528 * is named prop_%llu, and the nth element of its value 4529 * should be txg + object + n. 4530 */ 4531 tx = dmu_tx_create(os); 4532 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4533 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4534 if (txg == 0) 4535 return; 4536 4537 if (last_txg > txg) 4538 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); 4539 4540 for (i = 0; i < ints; i++) 4541 value[i] = txg + object + i; 4542 4543 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t), 4544 1, &txg, tx)); 4545 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t), 4546 ints, value, tx)); 4547 4548 dmu_tx_commit(tx); 4549 4550 /* 4551 * Remove a random pair of entries. 4552 */ 4553 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 4554 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 4555 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 4556 4557 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 4558 4559 if (error == ENOENT) 4560 return; 4561 4562 ASSERT0(error); 4563 4564 tx = dmu_tx_create(os); 4565 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4566 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4567 if (txg == 0) 4568 return; 4569 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx)); 4570 VERIFY3U(0, ==, zap_remove(os, object, propname, tx)); 4571 dmu_tx_commit(tx); 4572 } 4573 4574 /* 4575 * Testcase to test the upgrading of a microzap to fatzap. 4576 */ 4577 void 4578 ztest_fzap(ztest_ds_t *zd, uint64_t id) 4579 { 4580 objset_t *os = zd->zd_os; 4581 ztest_od_t od[1]; 4582 uint64_t object, txg; 4583 4584 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0); 4585 4586 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4587 return; 4588 4589 object = od[0].od_object; 4590 4591 /* 4592 * Add entries to this ZAP and make sure it spills over 4593 * and gets upgraded to a fatzap. Also, since we are adding 4594 * 2050 entries we should see ptrtbl growth and leaf-block split. 4595 */ 4596 for (int i = 0; i < 2050; i++) { 4597 char name[ZFS_MAX_DATASET_NAME_LEN]; 4598 uint64_t value = i; 4599 dmu_tx_t *tx; 4600 int error; 4601 4602 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu", 4603 id, value); 4604 4605 tx = dmu_tx_create(os); 4606 dmu_tx_hold_zap(tx, object, B_TRUE, name); 4607 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4608 if (txg == 0) 4609 return; 4610 error = zap_add(os, object, name, sizeof (uint64_t), 1, 4611 &value, tx); 4612 ASSERT(error == 0 || error == EEXIST); 4613 dmu_tx_commit(tx); 4614 } 4615 } 4616 4617 /* ARGSUSED */ 4618 void 4619 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id) 4620 { 4621 objset_t *os = zd->zd_os; 4622 ztest_od_t od[1]; 4623 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; 4624 dmu_tx_t *tx; 4625 int i, namelen, error; 4626 int micro = ztest_random(2); 4627 char name[20], string_value[20]; 4628 void *data; 4629 4630 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 4631 0, 0, 0); 4632 4633 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4634 return; 4635 4636 object = od[0].od_object; 4637 4638 /* 4639 * Generate a random name of the form 'xxx.....' where each 4640 * x is a random printable character and the dots are dots. 4641 * There are 94 such characters, and the name length goes from 4642 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. 4643 */ 4644 namelen = ztest_random(sizeof (name) - 5) + 5 + 1; 4645 4646 for (i = 0; i < 3; i++) 4647 name[i] = '!' + ztest_random('~' - '!' + 1); 4648 for (; i < namelen - 1; i++) 4649 name[i] = '.'; 4650 name[i] = '\0'; 4651 4652 if ((namelen & 1) || micro) { 4653 wsize = sizeof (txg); 4654 wc = 1; 4655 data = &txg; 4656 } else { 4657 wsize = 1; 4658 wc = namelen; 4659 data = string_value; 4660 } 4661 4662 count = -1ULL; 4663 VERIFY0(zap_count(os, object, &count)); 4664 ASSERT(count != -1ULL); 4665 4666 /* 4667 * Select an operation: length, lookup, add, update, remove. 4668 */ 4669 i = ztest_random(5); 4670 4671 if (i >= 2) { 4672 tx = dmu_tx_create(os); 4673 dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4674 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4675 if (txg == 0) 4676 return; 4677 bcopy(name, string_value, namelen); 4678 } else { 4679 tx = NULL; 4680 txg = 0; 4681 bzero(string_value, namelen); 4682 } 4683 4684 switch (i) { 4685 4686 case 0: 4687 error = zap_length(os, object, name, &zl_wsize, &zl_wc); 4688 if (error == 0) { 4689 ASSERT3U(wsize, ==, zl_wsize); 4690 ASSERT3U(wc, ==, zl_wc); 4691 } else { 4692 ASSERT3U(error, ==, ENOENT); 4693 } 4694 break; 4695 4696 case 1: 4697 error = zap_lookup(os, object, name, wsize, wc, data); 4698 if (error == 0) { 4699 if (data == string_value && 4700 bcmp(name, data, namelen) != 0) 4701 fatal(0, "name '%s' != val '%s' len %d", 4702 name, data, namelen); 4703 } else { 4704 ASSERT3U(error, ==, ENOENT); 4705 } 4706 break; 4707 4708 case 2: 4709 error = zap_add(os, object, name, wsize, wc, data, tx); 4710 ASSERT(error == 0 || error == EEXIST); 4711 break; 4712 4713 case 3: 4714 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); 4715 break; 4716 4717 case 4: 4718 error = zap_remove(os, object, name, tx); 4719 ASSERT(error == 0 || error == ENOENT); 4720 break; 4721 } 4722 4723 if (tx != NULL) 4724 dmu_tx_commit(tx); 4725 } 4726 4727 /* 4728 * Commit callback data. 4729 */ 4730 typedef struct ztest_cb_data { 4731 list_node_t zcd_node; 4732 uint64_t zcd_txg; 4733 int zcd_expected_err; 4734 boolean_t zcd_added; 4735 boolean_t zcd_called; 4736 spa_t *zcd_spa; 4737 } ztest_cb_data_t; 4738 4739 /* This is the actual commit callback function */ 4740 static void 4741 ztest_commit_callback(void *arg, int error) 4742 { 4743 ztest_cb_data_t *data = arg; 4744 uint64_t synced_txg; 4745 4746 VERIFY(data != NULL); 4747 VERIFY3S(data->zcd_expected_err, ==, error); 4748 VERIFY(!data->zcd_called); 4749 4750 synced_txg = spa_last_synced_txg(data->zcd_spa); 4751 if (data->zcd_txg > synced_txg) 4752 fatal(0, "commit callback of txg %" PRIu64 " called prematurely" 4753 ", last synced txg = %" PRIu64 "\n", data->zcd_txg, 4754 synced_txg); 4755 4756 data->zcd_called = B_TRUE; 4757 4758 if (error == ECANCELED) { 4759 ASSERT0(data->zcd_txg); 4760 ASSERT(!data->zcd_added); 4761 4762 /* 4763 * The private callback data should be destroyed here, but 4764 * since we are going to check the zcd_called field after 4765 * dmu_tx_abort(), we will destroy it there. 4766 */ 4767 return; 4768 } 4769 4770 /* Was this callback added to the global callback list? */ 4771 if (!data->zcd_added) 4772 goto out; 4773 4774 ASSERT3U(data->zcd_txg, !=, 0); 4775 4776 /* Remove our callback from the list */ 4777 mutex_enter(&zcl.zcl_callbacks_lock); 4778 list_remove(&zcl.zcl_callbacks, data); 4779 mutex_exit(&zcl.zcl_callbacks_lock); 4780 4781 out: 4782 umem_free(data, sizeof (ztest_cb_data_t)); 4783 } 4784 4785 /* Allocate and initialize callback data structure */ 4786 static ztest_cb_data_t * 4787 ztest_create_cb_data(objset_t *os, uint64_t txg) 4788 { 4789 ztest_cb_data_t *cb_data; 4790 4791 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL); 4792 4793 cb_data->zcd_txg = txg; 4794 cb_data->zcd_spa = dmu_objset_spa(os); 4795 4796 return (cb_data); 4797 } 4798 4799 /* 4800 * If a number of txgs equal to this threshold have been created after a commit 4801 * callback has been registered but not called, then we assume there is an 4802 * implementation bug. 4803 */ 4804 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2) 4805 4806 /* 4807 * Commit callback test. 4808 */ 4809 void 4810 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id) 4811 { 4812 objset_t *os = zd->zd_os; 4813 ztest_od_t od[1]; 4814 dmu_tx_t *tx; 4815 ztest_cb_data_t *cb_data[3], *tmp_cb; 4816 uint64_t old_txg, txg; 4817 int i, error; 4818 4819 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0); 4820 4821 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4822 return; 4823 4824 tx = dmu_tx_create(os); 4825 4826 cb_data[0] = ztest_create_cb_data(os, 0); 4827 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]); 4828 4829 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t)); 4830 4831 /* Every once in a while, abort the transaction on purpose */ 4832 if (ztest_random(100) == 0) 4833 error = -1; 4834 4835 if (!error) 4836 error = dmu_tx_assign(tx, TXG_NOWAIT); 4837 4838 txg = error ? 0 : dmu_tx_get_txg(tx); 4839 4840 cb_data[0]->zcd_txg = txg; 4841 cb_data[1] = ztest_create_cb_data(os, txg); 4842 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]); 4843 4844 if (error) { 4845 /* 4846 * It's not a strict requirement to call the registered 4847 * callbacks from inside dmu_tx_abort(), but that's what 4848 * it's supposed to happen in the current implementation 4849 * so we will check for that. 4850 */ 4851 for (i = 0; i < 2; i++) { 4852 cb_data[i]->zcd_expected_err = ECANCELED; 4853 VERIFY(!cb_data[i]->zcd_called); 4854 } 4855 4856 dmu_tx_abort(tx); 4857 4858 for (i = 0; i < 2; i++) { 4859 VERIFY(cb_data[i]->zcd_called); 4860 umem_free(cb_data[i], sizeof (ztest_cb_data_t)); 4861 } 4862 4863 return; 4864 } 4865 4866 cb_data[2] = ztest_create_cb_data(os, txg); 4867 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]); 4868 4869 /* 4870 * Read existing data to make sure there isn't a future leak. 4871 */ 4872 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t), 4873 &old_txg, DMU_READ_PREFETCH)); 4874 4875 if (old_txg > txg) 4876 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64, 4877 old_txg, txg); 4878 4879 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx); 4880 4881 mutex_enter(&zcl.zcl_callbacks_lock); 4882 4883 /* 4884 * Since commit callbacks don't have any ordering requirement and since 4885 * it is theoretically possible for a commit callback to be called 4886 * after an arbitrary amount of time has elapsed since its txg has been 4887 * synced, it is difficult to reliably determine whether a commit 4888 * callback hasn't been called due to high load or due to a flawed 4889 * implementation. 4890 * 4891 * In practice, we will assume that if after a certain number of txgs a 4892 * commit callback hasn't been called, then most likely there's an 4893 * implementation bug.. 4894 */ 4895 tmp_cb = list_head(&zcl.zcl_callbacks); 4896 if (tmp_cb != NULL && 4897 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) { 4898 fatal(0, "Commit callback threshold exceeded, oldest txg: %" 4899 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg); 4900 } 4901 4902 /* 4903 * Let's find the place to insert our callbacks. 4904 * 4905 * Even though the list is ordered by txg, it is possible for the 4906 * insertion point to not be the end because our txg may already be 4907 * quiescing at this point and other callbacks in the open txg 4908 * (from other objsets) may have sneaked in. 4909 */ 4910 tmp_cb = list_tail(&zcl.zcl_callbacks); 4911 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg) 4912 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb); 4913 4914 /* Add the 3 callbacks to the list */ 4915 for (i = 0; i < 3; i++) { 4916 if (tmp_cb == NULL) 4917 list_insert_head(&zcl.zcl_callbacks, cb_data[i]); 4918 else 4919 list_insert_after(&zcl.zcl_callbacks, tmp_cb, 4920 cb_data[i]); 4921 4922 cb_data[i]->zcd_added = B_TRUE; 4923 VERIFY(!cb_data[i]->zcd_called); 4924 4925 tmp_cb = cb_data[i]; 4926 } 4927 4928 mutex_exit(&zcl.zcl_callbacks_lock); 4929 4930 dmu_tx_commit(tx); 4931 } 4932 4933 /* 4934 * Visit each object in the dataset. Verify that its properties 4935 * are consistent what was stored in the block tag when it was created, 4936 * and that its unused bonus buffer space has not been overwritten. 4937 */ 4938 void 4939 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id) 4940 { 4941 objset_t *os = zd->zd_os; 4942 uint64_t obj; 4943 int err = 0; 4944 4945 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 4946 ztest_block_tag_t *bt = NULL; 4947 dmu_object_info_t doi; 4948 dmu_buf_t *db; 4949 4950 if (dmu_bonus_hold(os, obj, FTAG, &db) != 0) 4951 continue; 4952 4953 dmu_object_info_from_db(db, &doi); 4954 if (doi.doi_bonus_size >= sizeof (*bt)) 4955 bt = ztest_bt_bonus(db); 4956 4957 if (bt && bt->bt_magic == BT_MAGIC) { 4958 ztest_bt_verify(bt, os, obj, doi.doi_dnodesize, 4959 bt->bt_offset, bt->bt_gen, bt->bt_txg, 4960 bt->bt_crtxg); 4961 ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen); 4962 } 4963 4964 dmu_buf_rele(db, FTAG); 4965 } 4966 } 4967 4968 /* ARGSUSED */ 4969 void 4970 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id) 4971 { 4972 zfs_prop_t proplist[] = { 4973 ZFS_PROP_CHECKSUM, 4974 ZFS_PROP_COMPRESSION, 4975 ZFS_PROP_COPIES, 4976 ZFS_PROP_DEDUP 4977 }; 4978 4979 rw_enter(&ztest_name_lock, RW_READER); 4980 4981 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) 4982 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p], 4983 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2)); 4984 4985 rw_exit(&ztest_name_lock); 4986 } 4987 4988 /* ARGSUSED */ 4989 void 4990 ztest_remap_blocks(ztest_ds_t *zd, uint64_t id) 4991 { 4992 rw_enter(&ztest_name_lock, RW_READER); 4993 4994 int error = dmu_objset_remap_indirects(zd->zd_name); 4995 if (error == ENOSPC) 4996 error = 0; 4997 ASSERT0(error); 4998 4999 rw_exit(&ztest_name_lock); 5000 } 5001 5002 /* ARGSUSED */ 5003 void 5004 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id) 5005 { 5006 nvlist_t *props = NULL; 5007 5008 rw_enter(&ztest_name_lock, RW_READER); 5009 5010 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO, 5011 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN)); 5012 5013 VERIFY0(spa_prop_get(ztest_spa, &props)); 5014 5015 if (ztest_opts.zo_verbose >= 6) 5016 dump_nvlist(props, 4); 5017 5018 nvlist_free(props); 5019 5020 rw_exit(&ztest_name_lock); 5021 } 5022 5023 static int 5024 user_release_one(const char *snapname, const char *holdname) 5025 { 5026 nvlist_t *snaps, *holds; 5027 int error; 5028 5029 snaps = fnvlist_alloc(); 5030 holds = fnvlist_alloc(); 5031 fnvlist_add_boolean(holds, holdname); 5032 fnvlist_add_nvlist(snaps, snapname, holds); 5033 fnvlist_free(holds); 5034 error = dsl_dataset_user_release(snaps, NULL); 5035 fnvlist_free(snaps); 5036 return (error); 5037 } 5038 5039 /* 5040 * Test snapshot hold/release and deferred destroy. 5041 */ 5042 void 5043 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) 5044 { 5045 int error; 5046 objset_t *os = zd->zd_os; 5047 objset_t *origin; 5048 char snapname[100]; 5049 char fullname[100]; 5050 char clonename[100]; 5051 char tag[100]; 5052 char osname[ZFS_MAX_DATASET_NAME_LEN]; 5053 nvlist_t *holds; 5054 5055 rw_enter(&ztest_name_lock, RW_READER); 5056 5057 dmu_objset_name(os, osname); 5058 5059 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id); 5060 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname); 5061 (void) snprintf(clonename, sizeof (clonename), 5062 "%s/ch1_%llu", osname, id); 5063 (void) snprintf(tag, sizeof (tag), "tag_%llu", id); 5064 5065 /* 5066 * Clean up from any previous run. 5067 */ 5068 error = dsl_destroy_head(clonename); 5069 if (error != ENOENT) 5070 ASSERT0(error); 5071 error = user_release_one(fullname, tag); 5072 if (error != ESRCH && error != ENOENT) 5073 ASSERT0(error); 5074 error = dsl_destroy_snapshot(fullname, B_FALSE); 5075 if (error != ENOENT) 5076 ASSERT0(error); 5077 5078 /* 5079 * Create snapshot, clone it, mark snap for deferred destroy, 5080 * destroy clone, verify snap was also destroyed. 5081 */ 5082 error = dmu_objset_snapshot_one(osname, snapname); 5083 if (error) { 5084 if (error == ENOSPC) { 5085 ztest_record_enospc("dmu_objset_snapshot"); 5086 goto out; 5087 } 5088 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); 5089 } 5090 5091 error = dmu_objset_clone(clonename, fullname); 5092 if (error) { 5093 if (error == ENOSPC) { 5094 ztest_record_enospc("dmu_objset_clone"); 5095 goto out; 5096 } 5097 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error); 5098 } 5099 5100 error = dsl_destroy_snapshot(fullname, B_TRUE); 5101 if (error) { 5102 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", 5103 fullname, error); 5104 } 5105 5106 error = dsl_destroy_head(clonename); 5107 if (error) 5108 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error); 5109 5110 error = dmu_objset_hold(fullname, FTAG, &origin); 5111 if (error != ENOENT) 5112 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); 5113 5114 /* 5115 * Create snapshot, add temporary hold, verify that we can't 5116 * destroy a held snapshot, mark for deferred destroy, 5117 * release hold, verify snapshot was destroyed. 5118 */ 5119 error = dmu_objset_snapshot_one(osname, snapname); 5120 if (error) { 5121 if (error == ENOSPC) { 5122 ztest_record_enospc("dmu_objset_snapshot"); 5123 goto out; 5124 } 5125 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); 5126 } 5127 5128 holds = fnvlist_alloc(); 5129 fnvlist_add_string(holds, fullname, tag); 5130 error = dsl_dataset_user_hold(holds, 0, NULL); 5131 fnvlist_free(holds); 5132 5133 if (error == ENOSPC) { 5134 ztest_record_enospc("dsl_dataset_user_hold"); 5135 goto out; 5136 } else if (error) { 5137 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u", 5138 fullname, tag, error); 5139 } 5140 5141 error = dsl_destroy_snapshot(fullname, B_FALSE); 5142 if (error != EBUSY) { 5143 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d", 5144 fullname, error); 5145 } 5146 5147 error = dsl_destroy_snapshot(fullname, B_TRUE); 5148 if (error) { 5149 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", 5150 fullname, error); 5151 } 5152 5153 error = user_release_one(fullname, tag); 5154 if (error) 5155 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error); 5156 5157 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT); 5158 5159 out: 5160 rw_exit(&ztest_name_lock); 5161 } 5162 5163 /* 5164 * Inject random faults into the on-disk data. 5165 */ 5166 /* ARGSUSED */ 5167 void 5168 ztest_fault_inject(ztest_ds_t *zd, uint64_t id) 5169 { 5170 ztest_shared_t *zs = ztest_shared; 5171 spa_t *spa = ztest_spa; 5172 int fd; 5173 uint64_t offset; 5174 uint64_t leaves; 5175 uint64_t bad = 0x1990c0ffeedecade; 5176 uint64_t top, leaf; 5177 char path0[MAXPATHLEN]; 5178 char pathrand[MAXPATHLEN]; 5179 size_t fsize; 5180 int bshift = SPA_MAXBLOCKSHIFT + 2; 5181 int iters = 1000; 5182 int maxfaults; 5183 int mirror_save; 5184 vdev_t *vd0 = NULL; 5185 uint64_t guid0 = 0; 5186 boolean_t islog = B_FALSE; 5187 5188 mutex_enter(&ztest_vdev_lock); 5189 5190 /* 5191 * Device removal is in progress, fault injection must be disabled 5192 * until it completes and the pool is scrubbed. The fault injection 5193 * strategy for damaging blocks does not take in to account evacuated 5194 * blocks which may have already been damaged. 5195 */ 5196 if (ztest_device_removal_active) { 5197 mutex_exit(&ztest_vdev_lock); 5198 return; 5199 } 5200 5201 maxfaults = MAXFAULTS(); 5202 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; 5203 mirror_save = zs->zs_mirrors; 5204 mutex_exit(&ztest_vdev_lock); 5205 5206 ASSERT(leaves >= 1); 5207 5208 /* 5209 * Grab the name lock as reader. There are some operations 5210 * which don't like to have their vdevs changed while 5211 * they are in progress (i.e. spa_change_guid). Those 5212 * operations will have grabbed the name lock as writer. 5213 */ 5214 rw_enter(&ztest_name_lock, RW_READER); 5215 5216 /* 5217 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd. 5218 */ 5219 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 5220 5221 if (ztest_random(2) == 0) { 5222 /* 5223 * Inject errors on a normal data device or slog device. 5224 */ 5225 top = ztest_random_vdev_top(spa, B_TRUE); 5226 leaf = ztest_random(leaves) + zs->zs_splits; 5227 5228 /* 5229 * Generate paths to the first leaf in this top-level vdev, 5230 * and to the random leaf we selected. We'll induce transient 5231 * write failures and random online/offline activity on leaf 0, 5232 * and we'll write random garbage to the randomly chosen leaf. 5233 */ 5234 (void) snprintf(path0, sizeof (path0), ztest_dev_template, 5235 ztest_opts.zo_dir, ztest_opts.zo_pool, 5236 top * leaves + zs->zs_splits); 5237 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template, 5238 ztest_opts.zo_dir, ztest_opts.zo_pool, 5239 top * leaves + leaf); 5240 5241 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); 5242 if (vd0 != NULL && vd0->vdev_top->vdev_islog) 5243 islog = B_TRUE; 5244 5245 /* 5246 * If the top-level vdev needs to be resilvered 5247 * then we only allow faults on the device that is 5248 * resilvering. 5249 */ 5250 if (vd0 != NULL && maxfaults != 1 && 5251 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) || 5252 vd0->vdev_resilver_txg != 0)) { 5253 /* 5254 * Make vd0 explicitly claim to be unreadable, 5255 * or unwriteable, or reach behind its back 5256 * and close the underlying fd. We can do this if 5257 * maxfaults == 0 because we'll fail and reexecute, 5258 * and we can do it if maxfaults >= 2 because we'll 5259 * have enough redundancy. If maxfaults == 1, the 5260 * combination of this with injection of random data 5261 * corruption below exceeds the pool's fault tolerance. 5262 */ 5263 vdev_file_t *vf = vd0->vdev_tsd; 5264 5265 zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d", 5266 (long long)vd0->vdev_id, (int)maxfaults); 5267 5268 if (vf != NULL && ztest_random(3) == 0) { 5269 (void) close(vf->vf_vnode->v_fd); 5270 vf->vf_vnode->v_fd = -1; 5271 } else if (ztest_random(2) == 0) { 5272 vd0->vdev_cant_read = B_TRUE; 5273 } else { 5274 vd0->vdev_cant_write = B_TRUE; 5275 } 5276 guid0 = vd0->vdev_guid; 5277 } 5278 } else { 5279 /* 5280 * Inject errors on an l2cache device. 5281 */ 5282 spa_aux_vdev_t *sav = &spa->spa_l2cache; 5283 5284 if (sav->sav_count == 0) { 5285 spa_config_exit(spa, SCL_STATE, FTAG); 5286 rw_exit(&ztest_name_lock); 5287 return; 5288 } 5289 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)]; 5290 guid0 = vd0->vdev_guid; 5291 (void) strcpy(path0, vd0->vdev_path); 5292 (void) strcpy(pathrand, vd0->vdev_path); 5293 5294 leaf = 0; 5295 leaves = 1; 5296 maxfaults = INT_MAX; /* no limit on cache devices */ 5297 } 5298 5299 spa_config_exit(spa, SCL_STATE, FTAG); 5300 rw_exit(&ztest_name_lock); 5301 5302 /* 5303 * If we can tolerate two or more faults, or we're dealing 5304 * with a slog, randomly online/offline vd0. 5305 */ 5306 if ((maxfaults >= 2 || islog) && guid0 != 0) { 5307 if (ztest_random(10) < 6) { 5308 int flags = (ztest_random(2) == 0 ? 5309 ZFS_OFFLINE_TEMPORARY : 0); 5310 5311 /* 5312 * We have to grab the zs_name_lock as writer to 5313 * prevent a race between offlining a slog and 5314 * destroying a dataset. Offlining the slog will 5315 * grab a reference on the dataset which may cause 5316 * dmu_objset_destroy() to fail with EBUSY thus 5317 * leaving the dataset in an inconsistent state. 5318 */ 5319 if (islog) 5320 rw_enter(&ztest_name_lock, RW_WRITER); 5321 5322 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY); 5323 5324 if (islog) 5325 rw_exit(&ztest_name_lock); 5326 } else { 5327 /* 5328 * Ideally we would like to be able to randomly 5329 * call vdev_[on|off]line without holding locks 5330 * to force unpredictable failures but the side 5331 * effects of vdev_[on|off]line prevent us from 5332 * doing so. We grab the ztest_vdev_lock here to 5333 * prevent a race between injection testing and 5334 * aux_vdev removal. 5335 */ 5336 mutex_enter(&ztest_vdev_lock); 5337 (void) vdev_online(spa, guid0, 0, NULL); 5338 mutex_exit(&ztest_vdev_lock); 5339 } 5340 } 5341 5342 if (maxfaults == 0) 5343 return; 5344 5345 /* 5346 * We have at least single-fault tolerance, so inject data corruption. 5347 */ 5348 fd = open(pathrand, O_RDWR); 5349 5350 if (fd == -1) /* we hit a gap in the device namespace */ 5351 return; 5352 5353 fsize = lseek(fd, 0, SEEK_END); 5354 5355 while (--iters != 0) { 5356 /* 5357 * The offset must be chosen carefully to ensure that 5358 * we do not inject a given logical block with errors 5359 * on two different leaf devices, because ZFS can not 5360 * tolerate that (if maxfaults==1). 5361 * 5362 * We divide each leaf into chunks of size 5363 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk 5364 * there is a series of ranges to which we can inject errors. 5365 * Each range can accept errors on only a single leaf vdev. 5366 * The error injection ranges are separated by ranges 5367 * which we will not inject errors on any device (DMZs). 5368 * Each DMZ must be large enough such that a single block 5369 * can not straddle it, so that a single block can not be 5370 * a target in two different injection ranges (on different 5371 * leaf vdevs). 5372 * 5373 * For example, with 3 leaves, each chunk looks like: 5374 * 0 to 32M: injection range for leaf 0 5375 * 32M to 64M: DMZ - no injection allowed 5376 * 64M to 96M: injection range for leaf 1 5377 * 96M to 128M: DMZ - no injection allowed 5378 * 128M to 160M: injection range for leaf 2 5379 * 160M to 192M: DMZ - no injection allowed 5380 */ 5381 offset = ztest_random(fsize / (leaves << bshift)) * 5382 (leaves << bshift) + (leaf << bshift) + 5383 (ztest_random(1ULL << (bshift - 1)) & -8ULL); 5384 5385 /* 5386 * Only allow damage to the labels at one end of the vdev. 5387 * 5388 * If all labels are damaged, the device will be totally 5389 * inaccessible, which will result in loss of data, 5390 * because we also damage (parts of) the other side of 5391 * the mirror/raidz. 5392 * 5393 * Additionally, we will always have both an even and an 5394 * odd label, so that we can handle crashes in the 5395 * middle of vdev_config_sync(). 5396 */ 5397 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE) 5398 continue; 5399 5400 /* 5401 * The two end labels are stored at the "end" of the disk, but 5402 * the end of the disk (vdev_psize) is aligned to 5403 * sizeof (vdev_label_t). 5404 */ 5405 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t)); 5406 if ((leaf & 1) == 1 && 5407 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE) 5408 continue; 5409 5410 mutex_enter(&ztest_vdev_lock); 5411 if (mirror_save != zs->zs_mirrors) { 5412 mutex_exit(&ztest_vdev_lock); 5413 (void) close(fd); 5414 return; 5415 } 5416 5417 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) 5418 fatal(1, "can't inject bad word at 0x%llx in %s", 5419 offset, pathrand); 5420 5421 mutex_exit(&ztest_vdev_lock); 5422 5423 if (ztest_opts.zo_verbose >= 7) 5424 (void) printf("injected bad word into %s," 5425 " offset 0x%llx\n", pathrand, (u_longlong_t)offset); 5426 } 5427 5428 (void) close(fd); 5429 } 5430 5431 /* 5432 * Verify that DDT repair works as expected. 5433 */ 5434 void 5435 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id) 5436 { 5437 ztest_shared_t *zs = ztest_shared; 5438 spa_t *spa = ztest_spa; 5439 objset_t *os = zd->zd_os; 5440 ztest_od_t od[1]; 5441 uint64_t object, blocksize, txg, pattern, psize; 5442 enum zio_checksum checksum = spa_dedup_checksum(spa); 5443 dmu_buf_t *db; 5444 dmu_tx_t *tx; 5445 abd_t *abd; 5446 blkptr_t blk; 5447 int copies = 2 * ZIO_DEDUPDITTO_MIN; 5448 5449 blocksize = ztest_random_blocksize(); 5450 blocksize = MIN(blocksize, 2048); /* because we write so many */ 5451 5452 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 5453 0, 0); 5454 5455 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 5456 return; 5457 5458 /* 5459 * Take the name lock as writer to prevent anyone else from changing 5460 * the pool and dataset properies we need to maintain during this test. 5461 */ 5462 rw_enter(&ztest_name_lock, RW_WRITER); 5463 5464 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum, 5465 B_FALSE) != 0 || 5466 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1, 5467 B_FALSE) != 0) { 5468 rw_exit(&ztest_name_lock); 5469 return; 5470 } 5471 5472 dmu_objset_stats_t dds; 5473 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 5474 dmu_objset_fast_stat(os, &dds); 5475 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 5476 5477 object = od[0].od_object; 5478 blocksize = od[0].od_blocksize; 5479 pattern = zs->zs_guid ^ dds.dds_guid; 5480 5481 ASSERT(object != 0); 5482 5483 tx = dmu_tx_create(os); 5484 dmu_tx_hold_write(tx, object, 0, copies * blocksize); 5485 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 5486 if (txg == 0) { 5487 rw_exit(&ztest_name_lock); 5488 return; 5489 } 5490 5491 /* 5492 * Write all the copies of our block. 5493 */ 5494 for (int i = 0; i < copies; i++) { 5495 uint64_t offset = i * blocksize; 5496 int error = dmu_buf_hold(os, object, offset, FTAG, &db, 5497 DMU_READ_NO_PREFETCH); 5498 if (error != 0) { 5499 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u", 5500 os, (long long)object, (long long) offset, error); 5501 } 5502 ASSERT(db->db_offset == offset); 5503 ASSERT(db->db_size == blocksize); 5504 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) || 5505 ztest_pattern_match(db->db_data, db->db_size, 0ULL)); 5506 dmu_buf_will_fill(db, tx); 5507 ztest_pattern_set(db->db_data, db->db_size, pattern); 5508 dmu_buf_rele(db, FTAG); 5509 } 5510 5511 dmu_tx_commit(tx); 5512 txg_wait_synced(spa_get_dsl(spa), txg); 5513 5514 /* 5515 * Find out what block we got. 5516 */ 5517 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db, 5518 DMU_READ_NO_PREFETCH)); 5519 blk = *((dmu_buf_impl_t *)db)->db_blkptr; 5520 dmu_buf_rele(db, FTAG); 5521 5522 /* 5523 * Damage the block. Dedup-ditto will save us when we read it later. 5524 */ 5525 psize = BP_GET_PSIZE(&blk); 5526 abd = abd_alloc_linear(psize, B_TRUE); 5527 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern); 5528 5529 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk, 5530 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE, 5531 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL)); 5532 5533 abd_free(abd); 5534 5535 rw_exit(&ztest_name_lock); 5536 } 5537 5538 /* 5539 * Scrub the pool. 5540 */ 5541 /* ARGSUSED */ 5542 void 5543 ztest_scrub(ztest_ds_t *zd, uint64_t id) 5544 { 5545 spa_t *spa = ztest_spa; 5546 5547 /* 5548 * Scrub in progress by device removal. 5549 */ 5550 if (ztest_device_removal_active) 5551 return; 5552 5553 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5554 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */ 5555 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5556 } 5557 5558 /* 5559 * Change the guid for the pool. 5560 */ 5561 /* ARGSUSED */ 5562 void 5563 ztest_reguid(ztest_ds_t *zd, uint64_t id) 5564 { 5565 spa_t *spa = ztest_spa; 5566 uint64_t orig, load; 5567 int error; 5568 5569 orig = spa_guid(spa); 5570 load = spa_load_guid(spa); 5571 5572 rw_enter(&ztest_name_lock, RW_WRITER); 5573 error = spa_change_guid(spa); 5574 rw_exit(&ztest_name_lock); 5575 5576 if (error != 0) 5577 return; 5578 5579 if (ztest_opts.zo_verbose >= 4) { 5580 (void) printf("Changed guid old %llu -> %llu\n", 5581 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa)); 5582 } 5583 5584 VERIFY3U(orig, !=, spa_guid(spa)); 5585 VERIFY3U(load, ==, spa_load_guid(spa)); 5586 } 5587 5588 static vdev_t * 5589 ztest_random_concrete_vdev_leaf(vdev_t *vd) 5590 { 5591 if (vd == NULL) 5592 return (NULL); 5593 5594 if (vd->vdev_children == 0) 5595 return (vd); 5596 5597 vdev_t *eligible[vd->vdev_children]; 5598 int eligible_idx = 0, i; 5599 for (i = 0; i < vd->vdev_children; i++) { 5600 vdev_t *cvd = vd->vdev_child[i]; 5601 if (cvd->vdev_top->vdev_removing) 5602 continue; 5603 if (cvd->vdev_children > 0 || 5604 (vdev_is_concrete(cvd) && !cvd->vdev_detached)) { 5605 eligible[eligible_idx++] = cvd; 5606 } 5607 } 5608 VERIFY(eligible_idx > 0); 5609 5610 uint64_t child_no = ztest_random(eligible_idx); 5611 return (ztest_random_concrete_vdev_leaf(eligible[child_no])); 5612 } 5613 5614 /* ARGSUSED */ 5615 void 5616 ztest_initialize(ztest_ds_t *zd, uint64_t id) 5617 { 5618 spa_t *spa = ztest_spa; 5619 int error = 0; 5620 5621 mutex_enter(&ztest_vdev_lock); 5622 5623 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 5624 5625 /* Random leaf vdev */ 5626 vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev); 5627 if (rand_vd == NULL) { 5628 spa_config_exit(spa, SCL_VDEV, FTAG); 5629 mutex_exit(&ztest_vdev_lock); 5630 return; 5631 } 5632 5633 /* 5634 * The random vdev we've selected may change as soon as we 5635 * drop the spa_config_lock. We create local copies of things 5636 * we're interested in. 5637 */ 5638 uint64_t guid = rand_vd->vdev_guid; 5639 char *path = strdup(rand_vd->vdev_path); 5640 boolean_t active = rand_vd->vdev_initialize_thread != NULL; 5641 5642 zfs_dbgmsg("vd %p, guid %llu", rand_vd, guid); 5643 spa_config_exit(spa, SCL_VDEV, FTAG); 5644 5645 uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS); 5646 error = spa_vdev_initialize(spa, guid, cmd); 5647 switch (cmd) { 5648 case POOL_INITIALIZE_CANCEL: 5649 if (ztest_opts.zo_verbose >= 4) { 5650 (void) printf("Cancel initialize %s", path); 5651 if (!active) 5652 (void) printf(" failed (no initialize active)"); 5653 (void) printf("\n"); 5654 } 5655 break; 5656 case POOL_INITIALIZE_DO: 5657 if (ztest_opts.zo_verbose >= 4) { 5658 (void) printf("Start initialize %s", path); 5659 if (active && error == 0) 5660 (void) printf(" failed (already active)"); 5661 else if (error != 0) 5662 (void) printf(" failed (error %d)", error); 5663 (void) printf("\n"); 5664 } 5665 break; 5666 case POOL_INITIALIZE_SUSPEND: 5667 if (ztest_opts.zo_verbose >= 4) { 5668 (void) printf("Suspend initialize %s", path); 5669 if (!active) 5670 (void) printf(" failed (no initialize active)"); 5671 (void) printf("\n"); 5672 } 5673 break; 5674 } 5675 free(path); 5676 mutex_exit(&ztest_vdev_lock); 5677 } 5678 5679 /* 5680 * Verify pool integrity by running zdb. 5681 */ 5682 static void 5683 ztest_run_zdb(char *pool) 5684 { 5685 int status; 5686 char zdb[MAXPATHLEN + MAXNAMELEN + 20]; 5687 char zbuf[1024]; 5688 char *bin; 5689 char *ztest; 5690 char *isa; 5691 int isalen; 5692 FILE *fp; 5693 5694 (void) realpath(getexecname(), zdb); 5695 5696 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ 5697 bin = strstr(zdb, "/usr/bin/"); 5698 ztest = strstr(bin, "/ztest"); 5699 isa = bin + 8; 5700 isalen = ztest - isa; 5701 isa = strdup(isa); 5702 /* LINTED */ 5703 (void) sprintf(bin, 5704 "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s " 5705 "-o zfs_reconstruct_indirect_combinations_max=65536 %s", 5706 isalen, 5707 isa, 5708 ztest_opts.zo_verbose >= 3 ? "s" : "", 5709 ztest_opts.zo_verbose >= 4 ? "v" : "", 5710 spa_config_path, 5711 pool); 5712 free(isa); 5713 5714 if (ztest_opts.zo_verbose >= 5) 5715 (void) printf("Executing %s\n", strstr(zdb, "zdb ")); 5716 5717 fp = popen(zdb, "r"); 5718 5719 while (fgets(zbuf, sizeof (zbuf), fp) != NULL) 5720 if (ztest_opts.zo_verbose >= 3) 5721 (void) printf("%s", zbuf); 5722 5723 status = pclose(fp); 5724 5725 if (status == 0) 5726 return; 5727 5728 ztest_dump_core = 0; 5729 if (WIFEXITED(status)) 5730 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); 5731 else 5732 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); 5733 } 5734 5735 static void 5736 ztest_walk_pool_directory(char *header) 5737 { 5738 spa_t *spa = NULL; 5739 5740 if (ztest_opts.zo_verbose >= 6) 5741 (void) printf("%s\n", header); 5742 5743 mutex_enter(&spa_namespace_lock); 5744 while ((spa = spa_next(spa)) != NULL) 5745 if (ztest_opts.zo_verbose >= 6) 5746 (void) printf("\t%s\n", spa_name(spa)); 5747 mutex_exit(&spa_namespace_lock); 5748 } 5749 5750 static void 5751 ztest_spa_import_export(char *oldname, char *newname) 5752 { 5753 nvlist_t *config, *newconfig; 5754 uint64_t pool_guid; 5755 spa_t *spa; 5756 int error; 5757 5758 if (ztest_opts.zo_verbose >= 4) { 5759 (void) printf("import/export: old = %s, new = %s\n", 5760 oldname, newname); 5761 } 5762 5763 /* 5764 * Clean up from previous runs. 5765 */ 5766 (void) spa_destroy(newname); 5767 5768 /* 5769 * Get the pool's configuration and guid. 5770 */ 5771 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); 5772 5773 /* 5774 * Kick off a scrub to tickle scrub/export races. 5775 */ 5776 if (ztest_random(2) == 0) 5777 (void) spa_scan(spa, POOL_SCAN_SCRUB); 5778 5779 pool_guid = spa_guid(spa); 5780 spa_close(spa, FTAG); 5781 5782 ztest_walk_pool_directory("pools before export"); 5783 5784 /* 5785 * Export it. 5786 */ 5787 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE)); 5788 5789 ztest_walk_pool_directory("pools after export"); 5790 5791 /* 5792 * Try to import it. 5793 */ 5794 newconfig = spa_tryimport(config); 5795 ASSERT(newconfig != NULL); 5796 nvlist_free(newconfig); 5797 5798 /* 5799 * Import it under the new name. 5800 */ 5801 error = spa_import(newname, config, NULL, 0); 5802 if (error != 0) { 5803 dump_nvlist(config, 0); 5804 fatal(B_FALSE, "couldn't import pool %s as %s: error %u", 5805 oldname, newname, error); 5806 } 5807 5808 ztest_walk_pool_directory("pools after import"); 5809 5810 /* 5811 * Try to import it again -- should fail with EEXIST. 5812 */ 5813 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0)); 5814 5815 /* 5816 * Try to import it under a different name -- should fail with EEXIST. 5817 */ 5818 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0)); 5819 5820 /* 5821 * Verify that the pool is no longer visible under the old name. 5822 */ 5823 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); 5824 5825 /* 5826 * Verify that we can open and close the pool using the new name. 5827 */ 5828 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); 5829 ASSERT(pool_guid == spa_guid(spa)); 5830 spa_close(spa, FTAG); 5831 5832 nvlist_free(config); 5833 } 5834 5835 static void 5836 ztest_resume(spa_t *spa) 5837 { 5838 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6) 5839 (void) printf("resuming from suspended state\n"); 5840 spa_vdev_state_enter(spa, SCL_NONE); 5841 vdev_clear(spa, NULL); 5842 (void) spa_vdev_state_exit(spa, NULL, 0); 5843 (void) zio_resume(spa); 5844 } 5845 5846 static void * 5847 ztest_resume_thread(void *arg) 5848 { 5849 spa_t *spa = arg; 5850 5851 while (!ztest_exiting) { 5852 if (spa_suspended(spa)) 5853 ztest_resume(spa); 5854 (void) poll(NULL, 0, 100); 5855 5856 /* 5857 * Periodically change the zfs_compressed_arc_enabled setting. 5858 */ 5859 if (ztest_random(10) == 0) 5860 zfs_compressed_arc_enabled = ztest_random(2); 5861 5862 /* 5863 * Periodically change the zfs_abd_scatter_enabled setting. 5864 */ 5865 if (ztest_random(10) == 0) 5866 zfs_abd_scatter_enabled = ztest_random(2); 5867 } 5868 return (NULL); 5869 } 5870 5871 static void * 5872 ztest_deadman_thread(void *arg) 5873 { 5874 ztest_shared_t *zs = arg; 5875 spa_t *spa = ztest_spa; 5876 hrtime_t delta, total = 0; 5877 5878 for (;;) { 5879 delta = zs->zs_thread_stop - zs->zs_thread_start + 5880 MSEC2NSEC(zfs_deadman_synctime_ms); 5881 5882 (void) poll(NULL, 0, (int)NSEC2MSEC(delta)); 5883 5884 /* 5885 * If the pool is suspended then fail immediately. Otherwise, 5886 * check to see if the pool is making any progress. If 5887 * vdev_deadman() discovers that there hasn't been any recent 5888 * I/Os then it will end up aborting the tests. 5889 */ 5890 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) { 5891 fatal(0, "aborting test after %llu seconds because " 5892 "pool has transitioned to a suspended state.", 5893 zfs_deadman_synctime_ms / 1000); 5894 return (NULL); 5895 } 5896 vdev_deadman(spa->spa_root_vdev); 5897 5898 total += zfs_deadman_synctime_ms/1000; 5899 (void) printf("ztest has been running for %lld seconds\n", 5900 total); 5901 } 5902 } 5903 5904 static void 5905 ztest_execute(int test, ztest_info_t *zi, uint64_t id) 5906 { 5907 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets]; 5908 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test); 5909 hrtime_t functime = gethrtime(); 5910 5911 for (int i = 0; i < zi->zi_iters; i++) 5912 zi->zi_func(zd, id); 5913 5914 functime = gethrtime() - functime; 5915 5916 atomic_add_64(&zc->zc_count, 1); 5917 atomic_add_64(&zc->zc_time, functime); 5918 5919 if (ztest_opts.zo_verbose >= 4) { 5920 Dl_info dli; 5921 (void) dladdr((void *)zi->zi_func, &dli); 5922 (void) printf("%6.2f sec in %s\n", 5923 (double)functime / NANOSEC, dli.dli_sname); 5924 } 5925 } 5926 5927 static void * 5928 ztest_thread(void *arg) 5929 { 5930 int rand; 5931 uint64_t id = (uintptr_t)arg; 5932 ztest_shared_t *zs = ztest_shared; 5933 uint64_t call_next; 5934 hrtime_t now; 5935 ztest_info_t *zi; 5936 ztest_shared_callstate_t *zc; 5937 5938 while ((now = gethrtime()) < zs->zs_thread_stop) { 5939 /* 5940 * See if it's time to force a crash. 5941 */ 5942 if (now > zs->zs_thread_kill) 5943 ztest_kill(zs); 5944 5945 /* 5946 * If we're getting ENOSPC with some regularity, stop. 5947 */ 5948 if (zs->zs_enospc_count > 10) 5949 break; 5950 5951 /* 5952 * Pick a random function to execute. 5953 */ 5954 rand = ztest_random(ZTEST_FUNCS); 5955 zi = &ztest_info[rand]; 5956 zc = ZTEST_GET_SHARED_CALLSTATE(rand); 5957 call_next = zc->zc_next; 5958 5959 if (now >= call_next && 5960 atomic_cas_64(&zc->zc_next, call_next, call_next + 5961 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) { 5962 ztest_execute(rand, zi, id); 5963 } 5964 } 5965 5966 return (NULL); 5967 } 5968 5969 static void 5970 ztest_dataset_name(char *dsname, char *pool, int d) 5971 { 5972 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d); 5973 } 5974 5975 static void 5976 ztest_dataset_destroy(int d) 5977 { 5978 char name[ZFS_MAX_DATASET_NAME_LEN]; 5979 5980 ztest_dataset_name(name, ztest_opts.zo_pool, d); 5981 5982 if (ztest_opts.zo_verbose >= 3) 5983 (void) printf("Destroying %s to free up space\n", name); 5984 5985 /* 5986 * Cleanup any non-standard clones and snapshots. In general, 5987 * ztest thread t operates on dataset (t % zopt_datasets), 5988 * so there may be more than one thing to clean up. 5989 */ 5990 for (int t = d; t < ztest_opts.zo_threads; 5991 t += ztest_opts.zo_datasets) { 5992 ztest_dsl_dataset_cleanup(name, t); 5993 } 5994 5995 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, 5996 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); 5997 } 5998 5999 static void 6000 ztest_dataset_dirobj_verify(ztest_ds_t *zd) 6001 { 6002 uint64_t usedobjs, dirobjs, scratch; 6003 6004 /* 6005 * ZTEST_DIROBJ is the object directory for the entire dataset. 6006 * Therefore, the number of objects in use should equal the 6007 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself. 6008 * If not, we have an object leak. 6009 * 6010 * Note that we can only check this in ztest_dataset_open(), 6011 * when the open-context and syncing-context values agree. 6012 * That's because zap_count() returns the open-context value, 6013 * while dmu_objset_space() returns the rootbp fill count. 6014 */ 6015 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs)); 6016 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch); 6017 ASSERT3U(dirobjs + 1, ==, usedobjs); 6018 } 6019 6020 static int 6021 ztest_dataset_open(int d) 6022 { 6023 ztest_ds_t *zd = &ztest_ds[d]; 6024 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq; 6025 objset_t *os; 6026 zilog_t *zilog; 6027 char name[ZFS_MAX_DATASET_NAME_LEN]; 6028 int error; 6029 6030 ztest_dataset_name(name, ztest_opts.zo_pool, d); 6031 6032 rw_enter(&ztest_name_lock, RW_READER); 6033 6034 error = ztest_dataset_create(name); 6035 if (error == ENOSPC) { 6036 rw_exit(&ztest_name_lock); 6037 ztest_record_enospc(FTAG); 6038 return (error); 6039 } 6040 ASSERT(error == 0 || error == EEXIST); 6041 6042 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os)); 6043 rw_exit(&ztest_name_lock); 6044 6045 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os); 6046 6047 zilog = zd->zd_zilog; 6048 6049 if (zilog->zl_header->zh_claim_lr_seq != 0 && 6050 zilog->zl_header->zh_claim_lr_seq < committed_seq) 6051 fatal(0, "missing log records: claimed %llu < committed %llu", 6052 zilog->zl_header->zh_claim_lr_seq, committed_seq); 6053 6054 ztest_dataset_dirobj_verify(zd); 6055 6056 zil_replay(os, zd, ztest_replay_vector); 6057 6058 ztest_dataset_dirobj_verify(zd); 6059 6060 if (ztest_opts.zo_verbose >= 6) 6061 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n", 6062 zd->zd_name, 6063 (u_longlong_t)zilog->zl_parse_blk_count, 6064 (u_longlong_t)zilog->zl_parse_lr_count, 6065 (u_longlong_t)zilog->zl_replaying_seq); 6066 6067 zilog = zil_open(os, ztest_get_data); 6068 6069 if (zilog->zl_replaying_seq != 0 && 6070 zilog->zl_replaying_seq < committed_seq) 6071 fatal(0, "missing log records: replayed %llu < committed %llu", 6072 zilog->zl_replaying_seq, committed_seq); 6073 6074 return (0); 6075 } 6076 6077 static void 6078 ztest_dataset_close(int d) 6079 { 6080 ztest_ds_t *zd = &ztest_ds[d]; 6081 6082 zil_close(zd->zd_zilog); 6083 dmu_objset_disown(zd->zd_os, zd); 6084 6085 ztest_zd_fini(zd); 6086 } 6087 6088 /* 6089 * Kick off threads to run tests on all datasets in parallel. 6090 */ 6091 static void 6092 ztest_run(ztest_shared_t *zs) 6093 { 6094 thread_t *tid; 6095 spa_t *spa; 6096 objset_t *os; 6097 thread_t resume_tid; 6098 int error; 6099 6100 ztest_exiting = B_FALSE; 6101 6102 /* 6103 * Initialize parent/child shared state. 6104 */ 6105 mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL); 6106 mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL); 6107 rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL); 6108 6109 zs->zs_thread_start = gethrtime(); 6110 zs->zs_thread_stop = 6111 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC; 6112 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop); 6113 zs->zs_thread_kill = zs->zs_thread_stop; 6114 if (ztest_random(100) < ztest_opts.zo_killrate) { 6115 zs->zs_thread_kill -= 6116 ztest_random(ztest_opts.zo_passtime * NANOSEC); 6117 } 6118 6119 mutex_init(&zcl.zcl_callbacks_lock, NULL, USYNC_THREAD, NULL); 6120 6121 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t), 6122 offsetof(ztest_cb_data_t, zcd_node)); 6123 6124 /* 6125 * Open our pool. 6126 */ 6127 kernel_init(FREAD | FWRITE); 6128 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG)); 6129 metaslab_preload_limit = ztest_random(20) + 1; 6130 ztest_spa = spa; 6131 6132 dmu_objset_stats_t dds; 6133 VERIFY0(dmu_objset_own(ztest_opts.zo_pool, 6134 DMU_OST_ANY, B_TRUE, FTAG, &os)); 6135 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 6136 dmu_objset_fast_stat(os, &dds); 6137 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 6138 zs->zs_guid = dds.dds_guid; 6139 dmu_objset_disown(os, FTAG); 6140 6141 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN; 6142 6143 /* 6144 * We don't expect the pool to suspend unless maxfaults == 0, 6145 * in which case ztest_fault_inject() temporarily takes away 6146 * the only valid replica. 6147 */ 6148 if (MAXFAULTS() == 0) 6149 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT; 6150 else 6151 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC; 6152 6153 /* 6154 * Create a thread to periodically resume suspended I/O. 6155 */ 6156 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND, 6157 &resume_tid) == 0); 6158 6159 /* 6160 * Create a deadman thread to abort() if we hang. 6161 */ 6162 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND, 6163 NULL) == 0); 6164 6165 /* 6166 * Verify that we can safely inquire about any object, 6167 * whether it's allocated or not. To make it interesting, 6168 * we probe a 5-wide window around each power of two. 6169 * This hits all edge cases, including zero and the max. 6170 */ 6171 for (int t = 0; t < 64; t++) { 6172 for (int d = -5; d <= 5; d++) { 6173 error = dmu_object_info(spa->spa_meta_objset, 6174 (1ULL << t) + d, NULL); 6175 ASSERT(error == 0 || error == ENOENT || 6176 error == EINVAL); 6177 } 6178 } 6179 6180 /* 6181 * If we got any ENOSPC errors on the previous run, destroy something. 6182 */ 6183 if (zs->zs_enospc_count != 0) { 6184 int d = ztest_random(ztest_opts.zo_datasets); 6185 ztest_dataset_destroy(d); 6186 } 6187 zs->zs_enospc_count = 0; 6188 6189 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t), 6190 UMEM_NOFAIL); 6191 6192 if (ztest_opts.zo_verbose >= 4) 6193 (void) printf("starting main threads...\n"); 6194 6195 /* 6196 * Kick off all the tests that run in parallel. 6197 */ 6198 for (int t = 0; t < ztest_opts.zo_threads; t++) { 6199 if (t < ztest_opts.zo_datasets && 6200 ztest_dataset_open(t) != 0) 6201 return; 6202 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t, 6203 THR_BOUND, &tid[t]) == 0); 6204 } 6205 6206 /* 6207 * Wait for all of the tests to complete. We go in reverse order 6208 * so we don't close datasets while threads are still using them. 6209 */ 6210 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) { 6211 VERIFY(thr_join(tid[t], NULL, NULL) == 0); 6212 if (t < ztest_opts.zo_datasets) 6213 ztest_dataset_close(t); 6214 } 6215 6216 txg_wait_synced(spa_get_dsl(spa), 0); 6217 6218 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 6219 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa)); 6220 zfs_dbgmsg_print(FTAG); 6221 6222 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t)); 6223 6224 /* Kill the resume thread */ 6225 ztest_exiting = B_TRUE; 6226 VERIFY(thr_join(resume_tid, NULL, NULL) == 0); 6227 ztest_resume(spa); 6228 6229 /* 6230 * Right before closing the pool, kick off a bunch of async I/O; 6231 * spa_close() should wait for it to complete. 6232 */ 6233 for (uint64_t object = 1; object < 50; object++) { 6234 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20, 6235 ZIO_PRIORITY_SYNC_READ); 6236 } 6237 6238 spa_close(spa, FTAG); 6239 6240 /* 6241 * Verify that we can loop over all pools. 6242 */ 6243 mutex_enter(&spa_namespace_lock); 6244 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) 6245 if (ztest_opts.zo_verbose > 3) 6246 (void) printf("spa_next: found %s\n", spa_name(spa)); 6247 mutex_exit(&spa_namespace_lock); 6248 6249 /* 6250 * Verify that we can export the pool and reimport it under a 6251 * different name. 6252 */ 6253 if (ztest_random(2) == 0) { 6254 char name[ZFS_MAX_DATASET_NAME_LEN]; 6255 (void) snprintf(name, sizeof (name), "%s_import", 6256 ztest_opts.zo_pool); 6257 ztest_spa_import_export(ztest_opts.zo_pool, name); 6258 ztest_spa_import_export(name, ztest_opts.zo_pool); 6259 } 6260 6261 kernel_fini(); 6262 6263 list_destroy(&zcl.zcl_callbacks); 6264 6265 mutex_destroy(&zcl.zcl_callbacks_lock); 6266 6267 rw_destroy(&ztest_name_lock); 6268 mutex_destroy(&ztest_vdev_lock); 6269 mutex_destroy(&ztest_checkpoint_lock); 6270 } 6271 6272 static void 6273 ztest_freeze(void) 6274 { 6275 ztest_ds_t *zd = &ztest_ds[0]; 6276 spa_t *spa; 6277 int numloops = 0; 6278 6279 if (ztest_opts.zo_verbose >= 3) 6280 (void) printf("testing spa_freeze()...\n"); 6281 6282 kernel_init(FREAD | FWRITE); 6283 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 6284 VERIFY3U(0, ==, ztest_dataset_open(0)); 6285 ztest_spa = spa; 6286 6287 /* 6288 * Force the first log block to be transactionally allocated. 6289 * We have to do this before we freeze the pool -- otherwise 6290 * the log chain won't be anchored. 6291 */ 6292 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) { 6293 ztest_dmu_object_alloc_free(zd, 0); 6294 zil_commit(zd->zd_zilog, 0); 6295 } 6296 6297 txg_wait_synced(spa_get_dsl(spa), 0); 6298 6299 /* 6300 * Freeze the pool. This stops spa_sync() from doing anything, 6301 * so that the only way to record changes from now on is the ZIL. 6302 */ 6303 spa_freeze(spa); 6304 6305 /* 6306 * Because it is hard to predict how much space a write will actually 6307 * require beforehand, we leave ourselves some fudge space to write over 6308 * capacity. 6309 */ 6310 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2; 6311 6312 /* 6313 * Run tests that generate log records but don't alter the pool config 6314 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc). 6315 * We do a txg_wait_synced() after each iteration to force the txg 6316 * to increase well beyond the last synced value in the uberblock. 6317 * The ZIL should be OK with that. 6318 * 6319 * Run a random number of times less than zo_maxloops and ensure we do 6320 * not run out of space on the pool. 6321 */ 6322 while (ztest_random(10) != 0 && 6323 numloops++ < ztest_opts.zo_maxloops && 6324 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) { 6325 ztest_od_t od; 6326 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0); 6327 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE)); 6328 ztest_io(zd, od.od_object, 6329 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 6330 txg_wait_synced(spa_get_dsl(spa), 0); 6331 } 6332 6333 /* 6334 * Commit all of the changes we just generated. 6335 */ 6336 zil_commit(zd->zd_zilog, 0); 6337 txg_wait_synced(spa_get_dsl(spa), 0); 6338 6339 /* 6340 * Close our dataset and close the pool. 6341 */ 6342 ztest_dataset_close(0); 6343 spa_close(spa, FTAG); 6344 kernel_fini(); 6345 6346 /* 6347 * Open and close the pool and dataset to induce log replay. 6348 */ 6349 kernel_init(FREAD | FWRITE); 6350 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 6351 ASSERT(spa_freeze_txg(spa) == UINT64_MAX); 6352 VERIFY3U(0, ==, ztest_dataset_open(0)); 6353 ztest_dataset_close(0); 6354 6355 ztest_spa = spa; 6356 txg_wait_synced(spa_get_dsl(spa), 0); 6357 ztest_reguid(NULL, 0); 6358 6359 spa_close(spa, FTAG); 6360 kernel_fini(); 6361 } 6362 6363 void 6364 print_time(hrtime_t t, char *timebuf) 6365 { 6366 hrtime_t s = t / NANOSEC; 6367 hrtime_t m = s / 60; 6368 hrtime_t h = m / 60; 6369 hrtime_t d = h / 24; 6370 6371 s -= m * 60; 6372 m -= h * 60; 6373 h -= d * 24; 6374 6375 timebuf[0] = '\0'; 6376 6377 if (d) 6378 (void) sprintf(timebuf, 6379 "%llud%02lluh%02llum%02llus", d, h, m, s); 6380 else if (h) 6381 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); 6382 else if (m) 6383 (void) sprintf(timebuf, "%llum%02llus", m, s); 6384 else 6385 (void) sprintf(timebuf, "%llus", s); 6386 } 6387 6388 static nvlist_t * 6389 make_random_props() 6390 { 6391 nvlist_t *props; 6392 6393 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); 6394 if (ztest_random(2) == 0) 6395 return (props); 6396 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0); 6397 6398 return (props); 6399 } 6400 6401 /* 6402 * Create a storage pool with the given name and initial vdev size. 6403 * Then test spa_freeze() functionality. 6404 */ 6405 static void 6406 ztest_init(ztest_shared_t *zs) 6407 { 6408 spa_t *spa; 6409 nvlist_t *nvroot, *props; 6410 6411 mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL); 6412 mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL); 6413 rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL); 6414 6415 kernel_init(FREAD | FWRITE); 6416 6417 /* 6418 * Create the storage pool. 6419 */ 6420 (void) spa_destroy(ztest_opts.zo_pool); 6421 ztest_shared->zs_vdev_next_leaf = 0; 6422 zs->zs_splits = 0; 6423 zs->zs_mirrors = ztest_opts.zo_mirrors; 6424 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0, 6425 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1); 6426 props = make_random_props(); 6427 for (int i = 0; i < SPA_FEATURES; i++) { 6428 char buf[1024]; 6429 (void) snprintf(buf, sizeof (buf), "feature@%s", 6430 spa_feature_table[i].fi_uname); 6431 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0)); 6432 } 6433 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL)); 6434 nvlist_free(nvroot); 6435 nvlist_free(props); 6436 6437 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 6438 zs->zs_metaslab_sz = 6439 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift; 6440 6441 spa_close(spa, FTAG); 6442 6443 kernel_fini(); 6444 6445 ztest_run_zdb(ztest_opts.zo_pool); 6446 6447 ztest_freeze(); 6448 6449 ztest_run_zdb(ztest_opts.zo_pool); 6450 6451 rw_destroy(&ztest_name_lock); 6452 mutex_destroy(&ztest_vdev_lock); 6453 mutex_destroy(&ztest_checkpoint_lock); 6454 } 6455 6456 static void 6457 setup_data_fd(void) 6458 { 6459 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX"; 6460 6461 ztest_fd_data = mkstemp(ztest_name_data); 6462 ASSERT3S(ztest_fd_data, >=, 0); 6463 (void) unlink(ztest_name_data); 6464 } 6465 6466 6467 static int 6468 shared_data_size(ztest_shared_hdr_t *hdr) 6469 { 6470 int size; 6471 6472 size = hdr->zh_hdr_size; 6473 size += hdr->zh_opts_size; 6474 size += hdr->zh_size; 6475 size += hdr->zh_stats_size * hdr->zh_stats_count; 6476 size += hdr->zh_ds_size * hdr->zh_ds_count; 6477 6478 return (size); 6479 } 6480 6481 static void 6482 setup_hdr(void) 6483 { 6484 int size; 6485 ztest_shared_hdr_t *hdr; 6486 6487 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), 6488 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); 6489 ASSERT(hdr != MAP_FAILED); 6490 6491 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t))); 6492 6493 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t); 6494 hdr->zh_opts_size = sizeof (ztest_shared_opts_t); 6495 hdr->zh_size = sizeof (ztest_shared_t); 6496 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t); 6497 hdr->zh_stats_count = ZTEST_FUNCS; 6498 hdr->zh_ds_size = sizeof (ztest_shared_ds_t); 6499 hdr->zh_ds_count = ztest_opts.zo_datasets; 6500 6501 size = shared_data_size(hdr); 6502 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size)); 6503 6504 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); 6505 } 6506 6507 static void 6508 setup_data(void) 6509 { 6510 int size, offset; 6511 ztest_shared_hdr_t *hdr; 6512 uint8_t *buf; 6513 6514 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), 6515 PROT_READ, MAP_SHARED, ztest_fd_data, 0); 6516 ASSERT(hdr != MAP_FAILED); 6517 6518 size = shared_data_size(hdr); 6519 6520 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); 6521 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()), 6522 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); 6523 ASSERT(hdr != MAP_FAILED); 6524 buf = (uint8_t *)hdr; 6525 6526 offset = hdr->zh_hdr_size; 6527 ztest_shared_opts = (void *)&buf[offset]; 6528 offset += hdr->zh_opts_size; 6529 ztest_shared = (void *)&buf[offset]; 6530 offset += hdr->zh_size; 6531 ztest_shared_callstate = (void *)&buf[offset]; 6532 offset += hdr->zh_stats_size * hdr->zh_stats_count; 6533 ztest_shared_ds = (void *)&buf[offset]; 6534 } 6535 6536 static boolean_t 6537 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp) 6538 { 6539 pid_t pid; 6540 int status; 6541 char *cmdbuf = NULL; 6542 6543 pid = fork(); 6544 6545 if (cmd == NULL) { 6546 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 6547 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN); 6548 cmd = cmdbuf; 6549 } 6550 6551 if (pid == -1) 6552 fatal(1, "fork failed"); 6553 6554 if (pid == 0) { /* child */ 6555 char *emptyargv[2] = { cmd, NULL }; 6556 char fd_data_str[12]; 6557 6558 struct rlimit rl = { 1024, 1024 }; 6559 (void) setrlimit(RLIMIT_NOFILE, &rl); 6560 6561 (void) close(ztest_fd_rand); 6562 VERIFY3U(11, >=, 6563 snprintf(fd_data_str, 12, "%d", ztest_fd_data)); 6564 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1)); 6565 6566 (void) enable_extended_FILE_stdio(-1, -1); 6567 if (libpath != NULL) 6568 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1)); 6569 (void) execv(cmd, emptyargv); 6570 ztest_dump_core = B_FALSE; 6571 fatal(B_TRUE, "exec failed: %s", cmd); 6572 } 6573 6574 if (cmdbuf != NULL) { 6575 umem_free(cmdbuf, MAXPATHLEN); 6576 cmd = NULL; 6577 } 6578 6579 while (waitpid(pid, &status, 0) != pid) 6580 continue; 6581 if (statusp != NULL) 6582 *statusp = status; 6583 6584 if (WIFEXITED(status)) { 6585 if (WEXITSTATUS(status) != 0) { 6586 (void) fprintf(stderr, "child exited with code %d\n", 6587 WEXITSTATUS(status)); 6588 exit(2); 6589 } 6590 return (B_FALSE); 6591 } else if (WIFSIGNALED(status)) { 6592 if (!ignorekill || WTERMSIG(status) != SIGKILL) { 6593 (void) fprintf(stderr, "child died with signal %d\n", 6594 WTERMSIG(status)); 6595 exit(3); 6596 } 6597 return (B_TRUE); 6598 } else { 6599 (void) fprintf(stderr, "something strange happened to child\n"); 6600 exit(4); 6601 /* NOTREACHED */ 6602 } 6603 } 6604 6605 static void 6606 ztest_run_init(void) 6607 { 6608 ztest_shared_t *zs = ztest_shared; 6609 6610 ASSERT(ztest_opts.zo_init != 0); 6611 6612 /* 6613 * Blow away any existing copy of zpool.cache 6614 */ 6615 (void) remove(spa_config_path); 6616 6617 /* 6618 * Create and initialize our storage pool. 6619 */ 6620 for (int i = 1; i <= ztest_opts.zo_init; i++) { 6621 bzero(zs, sizeof (ztest_shared_t)); 6622 if (ztest_opts.zo_verbose >= 3 && 6623 ztest_opts.zo_init != 1) { 6624 (void) printf("ztest_init(), pass %d\n", i); 6625 } 6626 ztest_init(zs); 6627 } 6628 } 6629 6630 int 6631 main(int argc, char **argv) 6632 { 6633 int kills = 0; 6634 int iters = 0; 6635 int older = 0; 6636 int newer = 0; 6637 ztest_shared_t *zs; 6638 ztest_info_t *zi; 6639 ztest_shared_callstate_t *zc; 6640 char timebuf[100]; 6641 char numbuf[NN_NUMBUF_SZ]; 6642 char *cmd; 6643 boolean_t hasalt; 6644 char *fd_data_str = getenv("ZTEST_FD_DATA"); 6645 6646 (void) setvbuf(stdout, NULL, _IOLBF, 0); 6647 6648 dprintf_setup(&argc, argv); 6649 zfs_deadman_synctime_ms = 300000; 6650 /* 6651 * As two-word space map entries may not come up often (especially 6652 * if pool and vdev sizes are small) we want to force at least some 6653 * of them so the feature get tested. 6654 */ 6655 zfs_force_some_double_word_sm_entries = B_TRUE; 6656 6657 /* 6658 * Verify that even extensively damaged split blocks with many 6659 * segments can be reconstructed in a reasonable amount of time 6660 * when reconstruction is known to be possible. 6661 */ 6662 zfs_reconstruct_indirect_damage_fraction = 4; 6663 6664 ztest_fd_rand = open("/dev/urandom", O_RDONLY); 6665 ASSERT3S(ztest_fd_rand, >=, 0); 6666 6667 if (!fd_data_str) { 6668 process_options(argc, argv); 6669 6670 setup_data_fd(); 6671 setup_hdr(); 6672 setup_data(); 6673 bcopy(&ztest_opts, ztest_shared_opts, 6674 sizeof (*ztest_shared_opts)); 6675 } else { 6676 ztest_fd_data = atoi(fd_data_str); 6677 setup_data(); 6678 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts)); 6679 } 6680 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count); 6681 6682 /* Override location of zpool.cache */ 6683 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache", 6684 ztest_opts.zo_dir), !=, -1); 6685 6686 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t), 6687 UMEM_NOFAIL); 6688 zs = ztest_shared; 6689 6690 if (fd_data_str) { 6691 metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging; 6692 metaslab_df_alloc_threshold = 6693 zs->zs_metaslab_df_alloc_threshold; 6694 6695 if (zs->zs_do_init) 6696 ztest_run_init(); 6697 else 6698 ztest_run(zs); 6699 exit(0); 6700 } 6701 6702 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0); 6703 6704 if (ztest_opts.zo_verbose >= 1) { 6705 (void) printf("%llu vdevs, %d datasets, %d threads," 6706 " %llu seconds...\n", 6707 (u_longlong_t)ztest_opts.zo_vdevs, 6708 ztest_opts.zo_datasets, 6709 ztest_opts.zo_threads, 6710 (u_longlong_t)ztest_opts.zo_time); 6711 } 6712 6713 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); 6714 (void) strlcpy(cmd, getexecname(), MAXNAMELEN); 6715 6716 zs->zs_do_init = B_TRUE; 6717 if (strlen(ztest_opts.zo_alt_ztest) != 0) { 6718 if (ztest_opts.zo_verbose >= 1) { 6719 (void) printf("Executing older ztest for " 6720 "initialization: %s\n", ztest_opts.zo_alt_ztest); 6721 } 6722 VERIFY(!exec_child(ztest_opts.zo_alt_ztest, 6723 ztest_opts.zo_alt_libpath, B_FALSE, NULL)); 6724 } else { 6725 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL)); 6726 } 6727 zs->zs_do_init = B_FALSE; 6728 6729 zs->zs_proc_start = gethrtime(); 6730 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC; 6731 6732 for (int f = 0; f < ZTEST_FUNCS; f++) { 6733 zi = &ztest_info[f]; 6734 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6735 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop) 6736 zc->zc_next = UINT64_MAX; 6737 else 6738 zc->zc_next = zs->zs_proc_start + 6739 ztest_random(2 * zi->zi_interval[0] + 1); 6740 } 6741 6742 /* 6743 * Run the tests in a loop. These tests include fault injection 6744 * to verify that self-healing data works, and forced crashes 6745 * to verify that we never lose on-disk consistency. 6746 */ 6747 while (gethrtime() < zs->zs_proc_stop) { 6748 int status; 6749 boolean_t killed; 6750 6751 /* 6752 * Initialize the workload counters for each function. 6753 */ 6754 for (int f = 0; f < ZTEST_FUNCS; f++) { 6755 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6756 zc->zc_count = 0; 6757 zc->zc_time = 0; 6758 } 6759 6760 /* Set the allocation switch size */ 6761 zs->zs_metaslab_df_alloc_threshold = 6762 ztest_random(zs->zs_metaslab_sz / 4) + 1; 6763 6764 if (!hasalt || ztest_random(2) == 0) { 6765 if (hasalt && ztest_opts.zo_verbose >= 1) { 6766 (void) printf("Executing newer ztest: %s\n", 6767 cmd); 6768 } 6769 newer++; 6770 killed = exec_child(cmd, NULL, B_TRUE, &status); 6771 } else { 6772 if (hasalt && ztest_opts.zo_verbose >= 1) { 6773 (void) printf("Executing older ztest: %s\n", 6774 ztest_opts.zo_alt_ztest); 6775 } 6776 older++; 6777 killed = exec_child(ztest_opts.zo_alt_ztest, 6778 ztest_opts.zo_alt_libpath, B_TRUE, &status); 6779 } 6780 6781 if (killed) 6782 kills++; 6783 iters++; 6784 6785 if (ztest_opts.zo_verbose >= 1) { 6786 hrtime_t now = gethrtime(); 6787 6788 now = MIN(now, zs->zs_proc_stop); 6789 print_time(zs->zs_proc_stop - now, timebuf); 6790 nicenum(zs->zs_space, numbuf, sizeof (numbuf)); 6791 6792 (void) printf("Pass %3d, %8s, %3llu ENOSPC, " 6793 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", 6794 iters, 6795 WIFEXITED(status) ? "Complete" : "SIGKILL", 6796 (u_longlong_t)zs->zs_enospc_count, 6797 100.0 * zs->zs_alloc / zs->zs_space, 6798 numbuf, 6799 100.0 * (now - zs->zs_proc_start) / 6800 (ztest_opts.zo_time * NANOSEC), timebuf); 6801 } 6802 6803 if (ztest_opts.zo_verbose >= 2) { 6804 (void) printf("\nWorkload summary:\n\n"); 6805 (void) printf("%7s %9s %s\n", 6806 "Calls", "Time", "Function"); 6807 (void) printf("%7s %9s %s\n", 6808 "-----", "----", "--------"); 6809 for (int f = 0; f < ZTEST_FUNCS; f++) { 6810 Dl_info dli; 6811 6812 zi = &ztest_info[f]; 6813 zc = ZTEST_GET_SHARED_CALLSTATE(f); 6814 print_time(zc->zc_time, timebuf); 6815 (void) dladdr((void *)zi->zi_func, &dli); 6816 (void) printf("%7llu %9s %s\n", 6817 (u_longlong_t)zc->zc_count, timebuf, 6818 dli.dli_sname); 6819 } 6820 (void) printf("\n"); 6821 } 6822 6823 ztest_run_zdb(ztest_opts.zo_pool); 6824 } 6825 6826 if (ztest_opts.zo_verbose >= 1) { 6827 if (hasalt) { 6828 (void) printf("%d runs of older ztest: %s\n", older, 6829 ztest_opts.zo_alt_ztest); 6830 (void) printf("%d runs of newer ztest: %s\n", newer, 6831 cmd); 6832 } 6833 (void) printf("%d killed, %d completed, %.0f%% kill rate\n", 6834 kills, iters - kills, (100.0 * kills) / MAX(1, iters)); 6835 } 6836 6837 umem_free(cmd, MAXNAMELEN); 6838 6839 return (0); 6840 } 6841