ztest.c revision 268855
1168404Spjd/* 2168404Spjd * CDDL HEADER START 3168404Spjd * 4168404Spjd * The contents of this file are subject to the terms of the 5168404Spjd * Common Development and Distribution License (the "License"). 6168404Spjd * You may not use this file except in compliance with the License. 7168404Spjd * 8168404Spjd * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9168404Spjd * or http://www.opensolaris.org/os/licensing. 10168404Spjd * See the License for the specific language governing permissions 11168404Spjd * and limitations under the License. 12168404Spjd * 13168404Spjd * When distributing Covered Code, include this CDDL HEADER in each 14168404Spjd * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15168404Spjd * If applicable, add the following below this CDDL HEADER, with the 16168404Spjd * fields enclosed by brackets "[]" replaced with your own identifying 17168404Spjd * information: Portions Copyright [yyyy] [name of copyright owner] 18168404Spjd * 19168404Spjd * CDDL HEADER END 20168404Spjd */ 21168404Spjd/* 22219089Spjd * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23268086Sdelphij * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 24228103Smm * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25236143Smm * Copyright (c) 2012 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 26251646Sdelphij * Copyright (c) 2013 Steven Hartland. All rights reserved. 27168404Spjd */ 28168404Spjd 29168404Spjd/* 30168404Spjd * The objective of this program is to provide a DMU/ZAP/SPA stress test 31168404Spjd * that runs entirely in userland, is easy to use, and easy to extend. 32168404Spjd * 33168404Spjd * The overall design of the ztest program is as follows: 34168404Spjd * 35168404Spjd * (1) For each major functional area (e.g. adding vdevs to a pool, 36168404Spjd * creating and destroying datasets, reading and writing objects, etc) 37168404Spjd * we have a simple routine to test that functionality. These 38168404Spjd * individual routines do not have to do anything "stressful". 39168404Spjd * 40168404Spjd * (2) We turn these simple functionality tests into a stress test by 41168404Spjd * running them all in parallel, with as many threads as desired, 42168404Spjd * and spread across as many datasets, objects, and vdevs as desired. 43168404Spjd * 44168404Spjd * (3) While all this is happening, we inject faults into the pool to 45168404Spjd * verify that self-healing data really works. 46168404Spjd * 47168404Spjd * (4) Every time we open a dataset, we change its checksum and compression 48168404Spjd * functions. Thus even individual objects vary from block to block 49168404Spjd * in which checksum they use and whether they're compressed. 50168404Spjd * 51168404Spjd * (5) To verify that we never lose on-disk consistency after a crash, 52168404Spjd * we run the entire test in a child of the main process. 53168404Spjd * At random times, the child self-immolates with a SIGKILL. 54168404Spjd * This is the software equivalent of pulling the power cord. 55168404Spjd * The parent then runs the test again, using the existing 56268075Sdelphij * storage pool, as many times as desired. If backwards compatibility 57236143Smm * testing is enabled ztest will sometimes run the "older" version 58236143Smm * of ztest after a SIGKILL. 59168404Spjd * 60168404Spjd * (6) To verify that we don't have future leaks or temporal incursions, 61168404Spjd * many of the functional tests record the transaction group number 62168404Spjd * as part of their data. When reading old data, they verify that 63168404Spjd * the transaction group number is less than the current, open txg. 64168404Spjd * If you add a new test, please do this if applicable. 65168404Spjd * 66168404Spjd * When run with no arguments, ztest runs for about five minutes and 67168404Spjd * produces no output if successful. To get a little bit of information, 68168404Spjd * specify -V. To get more information, specify -VV, and so on. 69168404Spjd * 70168404Spjd * To turn this into an overnight stress test, use -T to specify run time. 71168404Spjd * 72168404Spjd * You can ask more more vdevs [-v], datasets [-d], or threads [-t] 73168404Spjd * to increase the pool capacity, fanout, and overall stress level. 74168404Spjd * 75236143Smm * Use the -k option to set the desired frequency of kills. 76236143Smm * 77236143Smm * When ztest invokes itself it passes all relevant information through a 78236143Smm * temporary file which is mmap-ed in the child process. This allows shared 79236143Smm * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always 80236143Smm * stored at offset 0 of this file and contains information on the size and 81236143Smm * number of shared structures in the file. The information stored in this file 82236143Smm * must remain backwards compatible with older versions of ztest so that 83236143Smm * ztest can invoke them during backwards compatibility testing (-B). 84168404Spjd */ 85168404Spjd 86168404Spjd#include <sys/zfs_context.h> 87168404Spjd#include <sys/spa.h> 88168404Spjd#include <sys/dmu.h> 89168404Spjd#include <sys/txg.h> 90209962Smm#include <sys/dbuf.h> 91168404Spjd#include <sys/zap.h> 92168404Spjd#include <sys/dmu_objset.h> 93168404Spjd#include <sys/poll.h> 94168404Spjd#include <sys/stat.h> 95168404Spjd#include <sys/time.h> 96168404Spjd#include <sys/wait.h> 97168404Spjd#include <sys/mman.h> 98168404Spjd#include <sys/resource.h> 99168404Spjd#include <sys/zio.h> 100168404Spjd#include <sys/zil.h> 101219089Spjd#include <sys/zil_impl.h> 102168404Spjd#include <sys/vdev_impl.h> 103185029Spjd#include <sys/vdev_file.h> 104168404Spjd#include <sys/spa_impl.h> 105219089Spjd#include <sys/metaslab_impl.h> 106168404Spjd#include <sys/dsl_prop.h> 107207910Smm#include <sys/dsl_dataset.h> 108248571Smm#include <sys/dsl_destroy.h> 109219089Spjd#include <sys/dsl_scan.h> 110219089Spjd#include <sys/zio_checksum.h> 111168404Spjd#include <sys/refcount.h> 112236884Smm#include <sys/zfeature.h> 113248571Smm#include <sys/dsl_userhold.h> 114168404Spjd#include <stdio.h> 115168404Spjd#include <stdio_ext.h> 116168404Spjd#include <stdlib.h> 117168404Spjd#include <unistd.h> 118168404Spjd#include <signal.h> 119168404Spjd#include <umem.h> 120168404Spjd#include <dlfcn.h> 121168404Spjd#include <ctype.h> 122168404Spjd#include <math.h> 123168404Spjd#include <errno.h> 124168404Spjd#include <sys/fs/zfs.h> 125219089Spjd#include <libnvpair.h> 126168404Spjd 127242845Sdelphijstatic int ztest_fd_data = -1; 128242845Sdelphijstatic int ztest_fd_rand = -1; 129168404Spjd 130236143Smmtypedef struct ztest_shared_hdr { 131236143Smm uint64_t zh_hdr_size; 132236143Smm uint64_t zh_opts_size; 133236143Smm uint64_t zh_size; 134236143Smm uint64_t zh_stats_size; 135236143Smm uint64_t zh_stats_count; 136236143Smm uint64_t zh_ds_size; 137236143Smm uint64_t zh_ds_count; 138236143Smm} ztest_shared_hdr_t; 139168404Spjd 140236143Smmstatic ztest_shared_hdr_t *ztest_shared_hdr; 141236143Smm 142236143Smmtypedef struct ztest_shared_opts { 143236143Smm char zo_pool[MAXNAMELEN]; 144236143Smm char zo_dir[MAXNAMELEN]; 145236143Smm char zo_alt_ztest[MAXNAMELEN]; 146236143Smm char zo_alt_libpath[MAXNAMELEN]; 147236143Smm uint64_t zo_vdevs; 148236143Smm uint64_t zo_vdevtime; 149236143Smm size_t zo_vdev_size; 150236143Smm int zo_ashift; 151236143Smm int zo_mirrors; 152236143Smm int zo_raidz; 153236143Smm int zo_raidz_parity; 154236143Smm int zo_datasets; 155236143Smm int zo_threads; 156236143Smm uint64_t zo_passtime; 157236143Smm uint64_t zo_killrate; 158236143Smm int zo_verbose; 159236143Smm int zo_init; 160236143Smm uint64_t zo_time; 161236143Smm uint64_t zo_maxloops; 162236143Smm uint64_t zo_metaslab_gang_bang; 163236143Smm} ztest_shared_opts_t; 164236143Smm 165236143Smmstatic const ztest_shared_opts_t ztest_opts_defaults = { 166236143Smm .zo_pool = { 'z', 't', 'e', 's', 't', '\0' }, 167236143Smm .zo_dir = { '/', 't', 'm', 'p', '\0' }, 168236143Smm .zo_alt_ztest = { '\0' }, 169236143Smm .zo_alt_libpath = { '\0' }, 170236143Smm .zo_vdevs = 5, 171236143Smm .zo_ashift = SPA_MINBLOCKSHIFT, 172236143Smm .zo_mirrors = 2, 173236143Smm .zo_raidz = 4, 174236143Smm .zo_raidz_parity = 1, 175236143Smm .zo_vdev_size = SPA_MINDEVSIZE, 176236143Smm .zo_datasets = 7, 177236143Smm .zo_threads = 23, 178236143Smm .zo_passtime = 60, /* 60 seconds */ 179236143Smm .zo_killrate = 70, /* 70% kill rate */ 180236143Smm .zo_verbose = 0, 181236143Smm .zo_init = 1, 182236143Smm .zo_time = 300, /* 5 minutes */ 183236143Smm .zo_maxloops = 50, /* max loops during spa_freeze() */ 184236143Smm .zo_metaslab_gang_bang = 32 << 10 185236143Smm}; 186236143Smm 187236143Smmextern uint64_t metaslab_gang_bang; 188236143Smmextern uint64_t metaslab_df_alloc_threshold; 189258632Savgextern uint64_t zfs_deadman_synctime_ms; 190268086Sdelphijextern int metaslab_preload_limit; 191236143Smm 192236143Smmstatic ztest_shared_opts_t *ztest_shared_opts; 193236143Smmstatic ztest_shared_opts_t ztest_opts; 194236143Smm 195236143Smmtypedef struct ztest_shared_ds { 196236143Smm uint64_t zd_seq; 197236143Smm} ztest_shared_ds_t; 198236143Smm 199236143Smmstatic ztest_shared_ds_t *ztest_shared_ds; 200236143Smm#define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d]) 201236143Smm 202219089Spjd#define BT_MAGIC 0x123456789abcdefULL 203236143Smm#define MAXFAULTS() \ 204236143Smm (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1) 205219089Spjd 206219089Spjdenum ztest_io_type { 207219089Spjd ZTEST_IO_WRITE_TAG, 208219089Spjd ZTEST_IO_WRITE_PATTERN, 209219089Spjd ZTEST_IO_WRITE_ZEROES, 210219089Spjd ZTEST_IO_TRUNCATE, 211219089Spjd ZTEST_IO_SETATTR, 212243524Smm ZTEST_IO_REWRITE, 213219089Spjd ZTEST_IO_TYPES 214219089Spjd}; 215219089Spjd 216185029Spjdtypedef struct ztest_block_tag { 217219089Spjd uint64_t bt_magic; 218185029Spjd uint64_t bt_objset; 219185029Spjd uint64_t bt_object; 220185029Spjd uint64_t bt_offset; 221219089Spjd uint64_t bt_gen; 222185029Spjd uint64_t bt_txg; 223219089Spjd uint64_t bt_crtxg; 224185029Spjd} ztest_block_tag_t; 225185029Spjd 226219089Spjdtypedef struct bufwad { 227219089Spjd uint64_t bw_index; 228219089Spjd uint64_t bw_txg; 229219089Spjd uint64_t bw_data; 230219089Spjd} bufwad_t; 231168404Spjd 232219089Spjd/* 233219089Spjd * XXX -- fix zfs range locks to be generic so we can use them here. 234219089Spjd */ 235219089Spjdtypedef enum { 236219089Spjd RL_READER, 237219089Spjd RL_WRITER, 238219089Spjd RL_APPEND 239219089Spjd} rl_type_t; 240168404Spjd 241219089Spjdtypedef struct rll { 242219089Spjd void *rll_writer; 243219089Spjd int rll_readers; 244219089Spjd mutex_t rll_lock; 245219089Spjd cond_t rll_cv; 246219089Spjd} rll_t; 247219089Spjd 248219089Spjdtypedef struct rl { 249219089Spjd uint64_t rl_object; 250219089Spjd uint64_t rl_offset; 251219089Spjd uint64_t rl_size; 252219089Spjd rll_t *rl_lock; 253219089Spjd} rl_t; 254219089Spjd 255219089Spjd#define ZTEST_RANGE_LOCKS 64 256219089Spjd#define ZTEST_OBJECT_LOCKS 64 257219089Spjd 258168404Spjd/* 259219089Spjd * Object descriptor. Used as a template for object lookup/create/remove. 260219089Spjd */ 261219089Spjdtypedef struct ztest_od { 262219089Spjd uint64_t od_dir; 263219089Spjd uint64_t od_object; 264219089Spjd dmu_object_type_t od_type; 265219089Spjd dmu_object_type_t od_crtype; 266219089Spjd uint64_t od_blocksize; 267219089Spjd uint64_t od_crblocksize; 268219089Spjd uint64_t od_gen; 269219089Spjd uint64_t od_crgen; 270219089Spjd char od_name[MAXNAMELEN]; 271219089Spjd} ztest_od_t; 272219089Spjd 273219089Spjd/* 274219089Spjd * Per-dataset state. 275219089Spjd */ 276219089Spjdtypedef struct ztest_ds { 277236143Smm ztest_shared_ds_t *zd_shared; 278219089Spjd objset_t *zd_os; 279224526Smm rwlock_t zd_zilog_lock; 280219089Spjd zilog_t *zd_zilog; 281219089Spjd ztest_od_t *zd_od; /* debugging aid */ 282219089Spjd char zd_name[MAXNAMELEN]; 283219089Spjd mutex_t zd_dirobj_lock; 284219089Spjd rll_t zd_object_lock[ZTEST_OBJECT_LOCKS]; 285219089Spjd rll_t zd_range_lock[ZTEST_RANGE_LOCKS]; 286219089Spjd} ztest_ds_t; 287219089Spjd 288219089Spjd/* 289219089Spjd * Per-iteration state. 290219089Spjd */ 291219089Spjdtypedef void ztest_func_t(ztest_ds_t *zd, uint64_t id); 292219089Spjd 293219089Spjdtypedef struct ztest_info { 294219089Spjd ztest_func_t *zi_func; /* test function */ 295219089Spjd uint64_t zi_iters; /* iterations per execution */ 296219089Spjd uint64_t *zi_interval; /* execute every <interval> seconds */ 297219089Spjd} ztest_info_t; 298219089Spjd 299236143Smmtypedef struct ztest_shared_callstate { 300236143Smm uint64_t zc_count; /* per-pass count */ 301236143Smm uint64_t zc_time; /* per-pass time */ 302236143Smm uint64_t zc_next; /* next time to call this function */ 303236143Smm} ztest_shared_callstate_t; 304236143Smm 305236143Smmstatic ztest_shared_callstate_t *ztest_shared_callstate; 306236143Smm#define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c]) 307236143Smm 308219089Spjd/* 309168404Spjd * Note: these aren't static because we want dladdr() to work. 310168404Spjd */ 311168404Spjdztest_func_t ztest_dmu_read_write; 312168404Spjdztest_func_t ztest_dmu_write_parallel; 313168404Spjdztest_func_t ztest_dmu_object_alloc_free; 314219089Spjdztest_func_t ztest_dmu_commit_callbacks; 315168404Spjdztest_func_t ztest_zap; 316168404Spjdztest_func_t ztest_zap_parallel; 317219089Spjdztest_func_t ztest_zil_commit; 318224526Smmztest_func_t ztest_zil_remount; 319219089Spjdztest_func_t ztest_dmu_read_write_zcopy; 320168404Spjdztest_func_t ztest_dmu_objset_create_destroy; 321219089Spjdztest_func_t ztest_dmu_prealloc; 322219089Spjdztest_func_t ztest_fzap; 323168404Spjdztest_func_t ztest_dmu_snapshot_create_destroy; 324219089Spjdztest_func_t ztest_dsl_prop_get_set; 325219089Spjdztest_func_t ztest_spa_prop_get_set; 326168404Spjdztest_func_t ztest_spa_create_destroy; 327168404Spjdztest_func_t ztest_fault_inject; 328219089Spjdztest_func_t ztest_ddt_repair; 329219089Spjdztest_func_t ztest_dmu_snapshot_hold; 330185029Spjdztest_func_t ztest_spa_rename; 331219089Spjdztest_func_t ztest_scrub; 332219089Spjdztest_func_t ztest_dsl_dataset_promote_busy; 333168404Spjdztest_func_t ztest_vdev_attach_detach; 334168404Spjdztest_func_t ztest_vdev_LUN_growth; 335168404Spjdztest_func_t ztest_vdev_add_remove; 336185029Spjdztest_func_t ztest_vdev_aux_add_remove; 337219089Spjdztest_func_t ztest_split_pool; 338228103Smmztest_func_t ztest_reguid; 339243505Smmztest_func_t ztest_spa_upgrade; 340168404Spjd 341219089Spjduint64_t zopt_always = 0ULL * NANOSEC; /* all the time */ 342219089Spjduint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */ 343219089Spjduint64_t zopt_often = 1ULL * NANOSEC; /* every second */ 344219089Spjduint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */ 345219089Spjduint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */ 346168404Spjd 347168404Spjdztest_info_t ztest_info[] = { 348185029Spjd { ztest_dmu_read_write, 1, &zopt_always }, 349219089Spjd { ztest_dmu_write_parallel, 10, &zopt_always }, 350185029Spjd { ztest_dmu_object_alloc_free, 1, &zopt_always }, 351219089Spjd { ztest_dmu_commit_callbacks, 1, &zopt_always }, 352185029Spjd { ztest_zap, 30, &zopt_always }, 353185029Spjd { ztest_zap_parallel, 100, &zopt_always }, 354219089Spjd { ztest_split_pool, 1, &zopt_always }, 355219089Spjd { ztest_zil_commit, 1, &zopt_incessant }, 356224526Smm { ztest_zil_remount, 1, &zopt_sometimes }, 357219089Spjd { ztest_dmu_read_write_zcopy, 1, &zopt_often }, 358219089Spjd { ztest_dmu_objset_create_destroy, 1, &zopt_often }, 359219089Spjd { ztest_dsl_prop_get_set, 1, &zopt_often }, 360219089Spjd { ztest_spa_prop_get_set, 1, &zopt_sometimes }, 361219089Spjd#if 0 362219089Spjd { ztest_dmu_prealloc, 1, &zopt_sometimes }, 363219089Spjd#endif 364219089Spjd { ztest_fzap, 1, &zopt_sometimes }, 365219089Spjd { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes }, 366219089Spjd { ztest_spa_create_destroy, 1, &zopt_sometimes }, 367185029Spjd { ztest_fault_inject, 1, &zopt_sometimes }, 368219089Spjd { ztest_ddt_repair, 1, &zopt_sometimes }, 369219089Spjd { ztest_dmu_snapshot_hold, 1, &zopt_sometimes }, 370254074Sdelphij { ztest_reguid, 1, &zopt_rarely }, 371185029Spjd { ztest_spa_rename, 1, &zopt_rarely }, 372219089Spjd { ztest_scrub, 1, &zopt_rarely }, 373243505Smm { ztest_spa_upgrade, 1, &zopt_rarely }, 374219089Spjd { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely }, 375248571Smm { ztest_vdev_attach_detach, 1, &zopt_sometimes }, 376185029Spjd { ztest_vdev_LUN_growth, 1, &zopt_rarely }, 377236143Smm { ztest_vdev_add_remove, 1, 378236143Smm &ztest_opts.zo_vdevtime }, 379236143Smm { ztest_vdev_aux_add_remove, 1, 380236143Smm &ztest_opts.zo_vdevtime }, 381168404Spjd}; 382168404Spjd 383168404Spjd#define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) 384168404Spjd 385219089Spjd/* 386219089Spjd * The following struct is used to hold a list of uncalled commit callbacks. 387219089Spjd * The callbacks are ordered by txg number. 388219089Spjd */ 389219089Spjdtypedef struct ztest_cb_list { 390219089Spjd mutex_t zcl_callbacks_lock; 391219089Spjd list_t zcl_callbacks; 392219089Spjd} ztest_cb_list_t; 393168404Spjd 394168404Spjd/* 395168404Spjd * Stuff we need to share writably between parent and child. 396168404Spjd */ 397168404Spjdtypedef struct ztest_shared { 398236143Smm boolean_t zs_do_init; 399219089Spjd hrtime_t zs_proc_start; 400219089Spjd hrtime_t zs_proc_stop; 401219089Spjd hrtime_t zs_thread_start; 402219089Spjd hrtime_t zs_thread_stop; 403219089Spjd hrtime_t zs_thread_kill; 404219089Spjd uint64_t zs_enospc_count; 405219089Spjd uint64_t zs_vdev_next_leaf; 406185029Spjd uint64_t zs_vdev_aux; 407168404Spjd uint64_t zs_alloc; 408168404Spjd uint64_t zs_space; 409219089Spjd uint64_t zs_splits; 410219089Spjd uint64_t zs_mirrors; 411236143Smm uint64_t zs_metaslab_sz; 412236143Smm uint64_t zs_metaslab_df_alloc_threshold; 413236143Smm uint64_t zs_guid; 414168404Spjd} ztest_shared_t; 415168404Spjd 416219089Spjd#define ID_PARALLEL -1ULL 417219089Spjd 418168404Spjdstatic char ztest_dev_template[] = "%s/%s.%llua"; 419185029Spjdstatic char ztest_aux_template[] = "%s/%s.%s.%llu"; 420219089Spjdztest_shared_t *ztest_shared; 421168404Spjd 422236143Smmstatic spa_t *ztest_spa = NULL; 423236143Smmstatic ztest_ds_t *ztest_ds; 424168404Spjd 425236143Smmstatic mutex_t ztest_vdev_lock; 426239620Smm 427239620Smm/* 428239620Smm * The ztest_name_lock protects the pool and dataset namespace used by 429239620Smm * the individual tests. To modify the namespace, consumers must grab 430239620Smm * this lock as writer. Grabbing the lock as reader will ensure that the 431239620Smm * namespace does not change while the lock is held. 432239620Smm */ 433236143Smmstatic rwlock_t ztest_name_lock; 434236143Smm 435236143Smmstatic boolean_t ztest_dump_core = B_TRUE; 436185029Spjdstatic boolean_t ztest_exiting; 437168404Spjd 438219089Spjd/* Global commit callback list */ 439219089Spjdstatic ztest_cb_list_t zcl; 440219089Spjd 441219089Spjdenum ztest_object { 442219089Spjd ZTEST_META_DNODE = 0, 443219089Spjd ZTEST_DIROBJ, 444219089Spjd ZTEST_OBJECTS 445219089Spjd}; 446168404Spjd 447168676Spjdstatic void usage(boolean_t) __NORETURN; 448168498Spjd 449168404Spjd/* 450168404Spjd * These libumem hooks provide a reasonable set of defaults for the allocator's 451168404Spjd * debugging facilities. 452168404Spjd */ 453168404Spjdconst char * 454168404Spjd_umem_debug_init() 455168404Spjd{ 456168404Spjd return ("default,verbose"); /* $UMEM_DEBUG setting */ 457168404Spjd} 458168404Spjd 459168404Spjdconst char * 460168404Spjd_umem_logging_init(void) 461168404Spjd{ 462168404Spjd return ("fail,contents"); /* $UMEM_LOGGING setting */ 463168404Spjd} 464168404Spjd 465168404Spjd#define FATAL_MSG_SZ 1024 466168404Spjd 467168404Spjdchar *fatal_msg; 468168404Spjd 469168404Spjdstatic void 470168404Spjdfatal(int do_perror, char *message, ...) 471168404Spjd{ 472168404Spjd va_list args; 473168404Spjd int save_errno = errno; 474168404Spjd char buf[FATAL_MSG_SZ]; 475168404Spjd 476168404Spjd (void) fflush(stdout); 477168404Spjd 478168404Spjd va_start(args, message); 479168404Spjd (void) sprintf(buf, "ztest: "); 480168404Spjd /* LINTED */ 481168404Spjd (void) vsprintf(buf + strlen(buf), message, args); 482168404Spjd va_end(args); 483168404Spjd if (do_perror) { 484168404Spjd (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), 485168404Spjd ": %s", strerror(save_errno)); 486168404Spjd } 487168404Spjd (void) fprintf(stderr, "%s\n", buf); 488168404Spjd fatal_msg = buf; /* to ease debugging */ 489168404Spjd if (ztest_dump_core) 490168404Spjd abort(); 491168404Spjd exit(3); 492168404Spjd} 493168404Spjd 494168404Spjdstatic int 495168404Spjdstr2shift(const char *buf) 496168404Spjd{ 497168404Spjd const char *ends = "BKMGTPEZ"; 498168404Spjd int i; 499168404Spjd 500168404Spjd if (buf[0] == '\0') 501168404Spjd return (0); 502168404Spjd for (i = 0; i < strlen(ends); i++) { 503168404Spjd if (toupper(buf[0]) == ends[i]) 504168404Spjd break; 505168404Spjd } 506168498Spjd if (i == strlen(ends)) { 507168498Spjd (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", 508168498Spjd buf); 509168498Spjd usage(B_FALSE); 510168498Spjd } 511168404Spjd if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { 512168404Spjd return (10*i); 513168404Spjd } 514168498Spjd (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); 515168498Spjd usage(B_FALSE); 516168498Spjd /* NOTREACHED */ 517168404Spjd} 518168404Spjd 519168404Spjdstatic uint64_t 520168404Spjdnicenumtoull(const char *buf) 521168404Spjd{ 522168404Spjd char *end; 523168404Spjd uint64_t val; 524168404Spjd 525168404Spjd val = strtoull(buf, &end, 0); 526168404Spjd if (end == buf) { 527168498Spjd (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); 528168498Spjd usage(B_FALSE); 529168404Spjd } else if (end[0] == '.') { 530168404Spjd double fval = strtod(buf, &end); 531168404Spjd fval *= pow(2, str2shift(end)); 532168498Spjd if (fval > UINT64_MAX) { 533168498Spjd (void) fprintf(stderr, "ztest: value too large: %s\n", 534168498Spjd buf); 535168498Spjd usage(B_FALSE); 536168498Spjd } 537168404Spjd val = (uint64_t)fval; 538168404Spjd } else { 539168404Spjd int shift = str2shift(end); 540168498Spjd if (shift >= 64 || (val << shift) >> shift != val) { 541168498Spjd (void) fprintf(stderr, "ztest: value too large: %s\n", 542168498Spjd buf); 543168498Spjd usage(B_FALSE); 544168498Spjd } 545168404Spjd val <<= shift; 546168404Spjd } 547168404Spjd return (val); 548168404Spjd} 549168404Spjd 550168404Spjdstatic void 551168498Spjdusage(boolean_t requested) 552168404Spjd{ 553236143Smm const ztest_shared_opts_t *zo = &ztest_opts_defaults; 554236143Smm 555168404Spjd char nice_vdev_size[10]; 556168404Spjd char nice_gang_bang[10]; 557168498Spjd FILE *fp = requested ? stdout : stderr; 558168404Spjd 559236143Smm nicenum(zo->zo_vdev_size, nice_vdev_size); 560236143Smm nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang); 561168404Spjd 562168498Spjd (void) fprintf(fp, "Usage: %s\n" 563168404Spjd "\t[-v vdevs (default: %llu)]\n" 564168404Spjd "\t[-s size_of_each_vdev (default: %s)]\n" 565219089Spjd "\t[-a alignment_shift (default: %d)] use 0 for random\n" 566168404Spjd "\t[-m mirror_copies (default: %d)]\n" 567168404Spjd "\t[-r raidz_disks (default: %d)]\n" 568168404Spjd "\t[-R raidz_parity (default: %d)]\n" 569168404Spjd "\t[-d datasets (default: %d)]\n" 570168404Spjd "\t[-t threads (default: %d)]\n" 571168404Spjd "\t[-g gang_block_threshold (default: %s)]\n" 572219089Spjd "\t[-i init_count (default: %d)] initialize pool i times\n" 573219089Spjd "\t[-k kill_percentage (default: %llu%%)]\n" 574168404Spjd "\t[-p pool_name (default: %s)]\n" 575219089Spjd "\t[-f dir (default: %s)] file directory for vdev files\n" 576219089Spjd "\t[-V] verbose (use multiple times for ever more blather)\n" 577219089Spjd "\t[-E] use existing pool instead of creating new one\n" 578219089Spjd "\t[-T time (default: %llu sec)] total run time\n" 579219089Spjd "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n" 580219089Spjd "\t[-P passtime (default: %llu sec)] time per pass\n" 581236143Smm "\t[-B alt_ztest (default: <none>)] alternate ztest path\n" 582168498Spjd "\t[-h] (print help)\n" 583168404Spjd "", 584236143Smm zo->zo_pool, 585236143Smm (u_longlong_t)zo->zo_vdevs, /* -v */ 586185029Spjd nice_vdev_size, /* -s */ 587236143Smm zo->zo_ashift, /* -a */ 588236143Smm zo->zo_mirrors, /* -m */ 589236143Smm zo->zo_raidz, /* -r */ 590236143Smm zo->zo_raidz_parity, /* -R */ 591236143Smm zo->zo_datasets, /* -d */ 592236143Smm zo->zo_threads, /* -t */ 593185029Spjd nice_gang_bang, /* -g */ 594236143Smm zo->zo_init, /* -i */ 595236143Smm (u_longlong_t)zo->zo_killrate, /* -k */ 596236143Smm zo->zo_pool, /* -p */ 597236143Smm zo->zo_dir, /* -f */ 598236143Smm (u_longlong_t)zo->zo_time, /* -T */ 599236143Smm (u_longlong_t)zo->zo_maxloops, /* -F */ 600236143Smm (u_longlong_t)zo->zo_passtime); 601168498Spjd exit(requested ? 0 : 1); 602168404Spjd} 603168404Spjd 604168404Spjdstatic void 605168404Spjdprocess_options(int argc, char **argv) 606168404Spjd{ 607236143Smm char *path; 608236143Smm ztest_shared_opts_t *zo = &ztest_opts; 609236143Smm 610168404Spjd int opt; 611168404Spjd uint64_t value; 612236143Smm char altdir[MAXNAMELEN] = { 0 }; 613168404Spjd 614236143Smm bcopy(&ztest_opts_defaults, zo, sizeof (*zo)); 615168404Spjd 616168404Spjd while ((opt = getopt(argc, argv, 617236143Smm "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) { 618168404Spjd value = 0; 619168404Spjd switch (opt) { 620185029Spjd case 'v': 621185029Spjd case 's': 622185029Spjd case 'a': 623185029Spjd case 'm': 624185029Spjd case 'r': 625185029Spjd case 'R': 626185029Spjd case 'd': 627185029Spjd case 't': 628185029Spjd case 'g': 629185029Spjd case 'i': 630185029Spjd case 'k': 631185029Spjd case 'T': 632185029Spjd case 'P': 633219089Spjd case 'F': 634168404Spjd value = nicenumtoull(optarg); 635168404Spjd } 636168404Spjd switch (opt) { 637185029Spjd case 'v': 638236143Smm zo->zo_vdevs = value; 639168404Spjd break; 640185029Spjd case 's': 641236143Smm zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value); 642168404Spjd break; 643185029Spjd case 'a': 644236143Smm zo->zo_ashift = value; 645168404Spjd break; 646185029Spjd case 'm': 647236143Smm zo->zo_mirrors = value; 648168404Spjd break; 649185029Spjd case 'r': 650236143Smm zo->zo_raidz = MAX(1, value); 651168404Spjd break; 652185029Spjd case 'R': 653236143Smm zo->zo_raidz_parity = MIN(MAX(value, 1), 3); 654168404Spjd break; 655185029Spjd case 'd': 656236143Smm zo->zo_datasets = MAX(1, value); 657168404Spjd break; 658185029Spjd case 't': 659236143Smm zo->zo_threads = MAX(1, value); 660168404Spjd break; 661185029Spjd case 'g': 662236143Smm zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, 663236143Smm value); 664168404Spjd break; 665185029Spjd case 'i': 666236143Smm zo->zo_init = value; 667168404Spjd break; 668185029Spjd case 'k': 669236143Smm zo->zo_killrate = value; 670168404Spjd break; 671185029Spjd case 'p': 672236143Smm (void) strlcpy(zo->zo_pool, optarg, 673236143Smm sizeof (zo->zo_pool)); 674168404Spjd break; 675185029Spjd case 'f': 676236143Smm path = realpath(optarg, NULL); 677236143Smm if (path == NULL) { 678236143Smm (void) fprintf(stderr, "error: %s: %s\n", 679236143Smm optarg, strerror(errno)); 680236143Smm usage(B_FALSE); 681236143Smm } else { 682236143Smm (void) strlcpy(zo->zo_dir, path, 683236143Smm sizeof (zo->zo_dir)); 684236143Smm } 685168404Spjd break; 686185029Spjd case 'V': 687236143Smm zo->zo_verbose++; 688168404Spjd break; 689185029Spjd case 'E': 690236143Smm zo->zo_init = 0; 691168404Spjd break; 692185029Spjd case 'T': 693236143Smm zo->zo_time = value; 694168404Spjd break; 695185029Spjd case 'P': 696236143Smm zo->zo_passtime = MAX(1, value); 697168404Spjd break; 698219089Spjd case 'F': 699236143Smm zo->zo_maxloops = MAX(1, value); 700219089Spjd break; 701236143Smm case 'B': 702236143Smm (void) strlcpy(altdir, optarg, sizeof (altdir)); 703236143Smm break; 704185029Spjd case 'h': 705168498Spjd usage(B_TRUE); 706168498Spjd break; 707185029Spjd case '?': 708185029Spjd default: 709168498Spjd usage(B_FALSE); 710168404Spjd break; 711168404Spjd } 712168404Spjd } 713168404Spjd 714236143Smm zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1); 715168404Spjd 716236143Smm zo->zo_vdevtime = 717236143Smm (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs : 718219089Spjd UINT64_MAX >> 2); 719236143Smm 720236143Smm if (strlen(altdir) > 0) { 721242845Sdelphij char *cmd; 722242845Sdelphij char *realaltdir; 723236143Smm char *bin; 724236143Smm char *ztest; 725236143Smm char *isa; 726236143Smm int isalen; 727236143Smm 728242845Sdelphij cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 729242845Sdelphij realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 730242845Sdelphij 731242845Sdelphij VERIFY(NULL != realpath(getexecname(), cmd)); 732236143Smm if (0 != access(altdir, F_OK)) { 733236143Smm ztest_dump_core = B_FALSE; 734236143Smm fatal(B_TRUE, "invalid alternate ztest path: %s", 735236143Smm altdir); 736236143Smm } 737236143Smm VERIFY(NULL != realpath(altdir, realaltdir)); 738236143Smm 739236143Smm /* 740236143Smm * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest". 741236143Smm * We want to extract <isa> to determine if we should use 742236143Smm * 32 or 64 bit binaries. 743236143Smm */ 744236143Smm bin = strstr(cmd, "/usr/bin/"); 745236143Smm ztest = strstr(bin, "/ztest"); 746236143Smm isa = bin + 9; 747236143Smm isalen = ztest - isa; 748236143Smm (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest), 749236143Smm "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa); 750236143Smm (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath), 751236143Smm "%s/usr/lib/%.*s", realaltdir, isalen, isa); 752236143Smm 753236143Smm if (0 != access(zo->zo_alt_ztest, X_OK)) { 754236143Smm ztest_dump_core = B_FALSE; 755236143Smm fatal(B_TRUE, "invalid alternate ztest: %s", 756236143Smm zo->zo_alt_ztest); 757236143Smm } else if (0 != access(zo->zo_alt_libpath, X_OK)) { 758236143Smm ztest_dump_core = B_FALSE; 759236143Smm fatal(B_TRUE, "invalid alternate lib directory %s", 760236143Smm zo->zo_alt_libpath); 761236143Smm } 762242845Sdelphij 763242845Sdelphij umem_free(cmd, MAXPATHLEN); 764242845Sdelphij umem_free(realaltdir, MAXPATHLEN); 765236143Smm } 766168404Spjd} 767168404Spjd 768219089Spjdstatic void 769219089Spjdztest_kill(ztest_shared_t *zs) 770219089Spjd{ 771236143Smm zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa)); 772236143Smm zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa)); 773254112Sdelphij 774254112Sdelphij /* 775254112Sdelphij * Before we kill off ztest, make sure that the config is updated. 776254112Sdelphij * See comment above spa_config_sync(). 777254112Sdelphij */ 778254112Sdelphij mutex_enter(&spa_namespace_lock); 779254112Sdelphij spa_config_sync(ztest_spa, B_FALSE, B_FALSE); 780254112Sdelphij mutex_exit(&spa_namespace_lock); 781254112Sdelphij 782254112Sdelphij zfs_dbgmsg_print(FTAG); 783219089Spjd (void) kill(getpid(), SIGKILL); 784219089Spjd} 785219089Spjd 786168404Spjdstatic uint64_t 787219089Spjdztest_random(uint64_t range) 788219089Spjd{ 789219089Spjd uint64_t r; 790219089Spjd 791242845Sdelphij ASSERT3S(ztest_fd_rand, >=, 0); 792242845Sdelphij 793219089Spjd if (range == 0) 794219089Spjd return (0); 795219089Spjd 796242845Sdelphij if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r)) 797219089Spjd fatal(1, "short read from /dev/urandom"); 798219089Spjd 799219089Spjd return (r % range); 800219089Spjd} 801219089Spjd 802219089Spjd/* ARGSUSED */ 803219089Spjdstatic void 804219089Spjdztest_record_enospc(const char *s) 805219089Spjd{ 806219089Spjd ztest_shared->zs_enospc_count++; 807219089Spjd} 808219089Spjd 809219089Spjdstatic uint64_t 810168404Spjdztest_get_ashift(void) 811168404Spjd{ 812236143Smm if (ztest_opts.zo_ashift == 0) 813268855Sdelphij return (SPA_MINBLOCKSHIFT + ztest_random(5)); 814236143Smm return (ztest_opts.zo_ashift); 815168404Spjd} 816168404Spjd 817168404Spjdstatic nvlist_t * 818243505Smmmake_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift) 819168404Spjd{ 820185029Spjd char pathbuf[MAXPATHLEN]; 821168404Spjd uint64_t vdev; 822168404Spjd nvlist_t *file; 823168404Spjd 824185029Spjd if (ashift == 0) 825185029Spjd ashift = ztest_get_ashift(); 826168404Spjd 827185029Spjd if (path == NULL) { 828185029Spjd path = pathbuf; 829185029Spjd 830185029Spjd if (aux != NULL) { 831185029Spjd vdev = ztest_shared->zs_vdev_aux; 832236143Smm (void) snprintf(path, sizeof (pathbuf), 833236143Smm ztest_aux_template, ztest_opts.zo_dir, 834243505Smm pool == NULL ? ztest_opts.zo_pool : pool, 835243505Smm aux, vdev); 836185029Spjd } else { 837219089Spjd vdev = ztest_shared->zs_vdev_next_leaf++; 838236143Smm (void) snprintf(path, sizeof (pathbuf), 839236143Smm ztest_dev_template, ztest_opts.zo_dir, 840243505Smm pool == NULL ? ztest_opts.zo_pool : pool, vdev); 841185029Spjd } 842185029Spjd } 843185029Spjd 844185029Spjd if (size != 0) { 845185029Spjd int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666); 846168404Spjd if (fd == -1) 847185029Spjd fatal(1, "can't open %s", path); 848168404Spjd if (ftruncate(fd, size) != 0) 849185029Spjd fatal(1, "can't ftruncate %s", path); 850168404Spjd (void) close(fd); 851168404Spjd } 852168404Spjd 853168404Spjd VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 854168404Spjd VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 855185029Spjd VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0); 856168404Spjd VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 857168404Spjd 858168404Spjd return (file); 859168404Spjd} 860168404Spjd 861168404Spjdstatic nvlist_t * 862243505Smmmake_vdev_raidz(char *path, char *aux, char *pool, size_t size, 863243505Smm uint64_t ashift, int r) 864168404Spjd{ 865168404Spjd nvlist_t *raidz, **child; 866168404Spjd int c; 867168404Spjd 868168404Spjd if (r < 2) 869243505Smm return (make_vdev_file(path, aux, pool, size, ashift)); 870168404Spjd child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); 871168404Spjd 872168404Spjd for (c = 0; c < r; c++) 873243505Smm child[c] = make_vdev_file(path, aux, pool, size, ashift); 874168404Spjd 875168404Spjd VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); 876168404Spjd VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, 877168404Spjd VDEV_TYPE_RAIDZ) == 0); 878168404Spjd VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, 879236143Smm ztest_opts.zo_raidz_parity) == 0); 880168404Spjd VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, 881168404Spjd child, r) == 0); 882168404Spjd 883168404Spjd for (c = 0; c < r; c++) 884168404Spjd nvlist_free(child[c]); 885168404Spjd 886168404Spjd umem_free(child, r * sizeof (nvlist_t *)); 887168404Spjd 888168404Spjd return (raidz); 889168404Spjd} 890168404Spjd 891168404Spjdstatic nvlist_t * 892243505Smmmake_vdev_mirror(char *path, char *aux, char *pool, size_t size, 893243505Smm uint64_t ashift, int r, int m) 894168404Spjd{ 895168404Spjd nvlist_t *mirror, **child; 896168404Spjd int c; 897168404Spjd 898168404Spjd if (m < 1) 899243505Smm return (make_vdev_raidz(path, aux, pool, size, ashift, r)); 900168404Spjd 901168404Spjd child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); 902168404Spjd 903168404Spjd for (c = 0; c < m; c++) 904243505Smm child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r); 905168404Spjd 906168404Spjd VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); 907168404Spjd VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, 908168404Spjd VDEV_TYPE_MIRROR) == 0); 909168404Spjd VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, 910168404Spjd child, m) == 0); 911168404Spjd 912168404Spjd for (c = 0; c < m; c++) 913168404Spjd nvlist_free(child[c]); 914168404Spjd 915168404Spjd umem_free(child, m * sizeof (nvlist_t *)); 916168404Spjd 917168404Spjd return (mirror); 918168404Spjd} 919168404Spjd 920168404Spjdstatic nvlist_t * 921243505Smmmake_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift, 922243505Smm int log, int r, int m, int t) 923168404Spjd{ 924168404Spjd nvlist_t *root, **child; 925168404Spjd int c; 926168404Spjd 927168404Spjd ASSERT(t > 0); 928168404Spjd 929168404Spjd child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); 930168404Spjd 931185029Spjd for (c = 0; c < t; c++) { 932243505Smm child[c] = make_vdev_mirror(path, aux, pool, size, ashift, 933243505Smm r, m); 934185029Spjd VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, 935185029Spjd log) == 0); 936185029Spjd } 937168404Spjd 938168404Spjd VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 939168404Spjd VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 940185029Spjd VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN, 941168404Spjd child, t) == 0); 942168404Spjd 943168404Spjd for (c = 0; c < t; c++) 944168404Spjd nvlist_free(child[c]); 945168404Spjd 946168404Spjd umem_free(child, t * sizeof (nvlist_t *)); 947168404Spjd 948168404Spjd return (root); 949168404Spjd} 950168404Spjd 951243505Smm/* 952243505Smm * Find a random spa version. Returns back a random spa version in the 953243505Smm * range [initial_version, SPA_VERSION_FEATURES]. 954243505Smm */ 955243505Smmstatic uint64_t 956243505Smmztest_random_spa_version(uint64_t initial_version) 957243505Smm{ 958243505Smm uint64_t version = initial_version; 959243505Smm 960243505Smm if (version <= SPA_VERSION_BEFORE_FEATURES) { 961243505Smm version = version + 962243505Smm ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1); 963243505Smm } 964243505Smm 965243505Smm if (version > SPA_VERSION_BEFORE_FEATURES) 966243505Smm version = SPA_VERSION_FEATURES; 967243505Smm 968243505Smm ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 969243505Smm return (version); 970243505Smm} 971243505Smm 972268855Sdelphij/* 973268855Sdelphij * Find the largest ashift used 974268855Sdelphij */ 975268855Sdelphijstatic uint64_t 976268855Sdelphijztest_spa_get_ashift() { 977268855Sdelphij uint64_t i; 978268855Sdelphij uint64_t ashift = SPA_MINBLOCKSHIFT; 979268855Sdelphij vdev_t *rvd = ztest_spa->spa_root_vdev; 980268855Sdelphij 981268855Sdelphij for (i = 0; i < rvd->vdev_children; i++) { 982268855Sdelphij ashift = MAX(ashift, rvd->vdev_child[i]->vdev_ashift); 983268855Sdelphij } 984268855Sdelphij return (ashift); 985268855Sdelphij} 986268855Sdelphij 987219089Spjdstatic int 988219089Spjdztest_random_blocksize(void) 989219089Spjd{ 990268855Sdelphij // Choose a block size >= the ashift. 991268855Sdelphij uint64_t block_shift = 992268855Sdelphij ztest_random(SPA_MAXBLOCKSHIFT - ztest_spa_get_ashift() + 1); 993268855Sdelphij return (1 << (SPA_MINBLOCKSHIFT + block_shift)); 994219089Spjd} 995219089Spjd 996219089Spjdstatic int 997219089Spjdztest_random_ibshift(void) 998219089Spjd{ 999219089Spjd return (DN_MIN_INDBLKSHIFT + 1000219089Spjd ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1)); 1001219089Spjd} 1002219089Spjd 1003219089Spjdstatic uint64_t 1004219089Spjdztest_random_vdev_top(spa_t *spa, boolean_t log_ok) 1005219089Spjd{ 1006219089Spjd uint64_t top; 1007219089Spjd vdev_t *rvd = spa->spa_root_vdev; 1008219089Spjd vdev_t *tvd; 1009219089Spjd 1010219089Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1011219089Spjd 1012219089Spjd do { 1013219089Spjd top = ztest_random(rvd->vdev_children); 1014219089Spjd tvd = rvd->vdev_child[top]; 1015219089Spjd } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) || 1016219089Spjd tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL); 1017219089Spjd 1018219089Spjd return (top); 1019219089Spjd} 1020219089Spjd 1021219089Spjdstatic uint64_t 1022219089Spjdztest_random_dsl_prop(zfs_prop_t prop) 1023219089Spjd{ 1024219089Spjd uint64_t value; 1025219089Spjd 1026219089Spjd do { 1027219089Spjd value = zfs_prop_random_value(prop, ztest_random(-1ULL)); 1028219089Spjd } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF); 1029219089Spjd 1030219089Spjd return (value); 1031219089Spjd} 1032219089Spjd 1033219089Spjdstatic int 1034219089Spjdztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value, 1035219089Spjd boolean_t inherit) 1036219089Spjd{ 1037219089Spjd const char *propname = zfs_prop_to_name(prop); 1038219089Spjd const char *valname; 1039219089Spjd char setpoint[MAXPATHLEN]; 1040219089Spjd uint64_t curval; 1041219089Spjd int error; 1042219089Spjd 1043248571Smm error = dsl_prop_set_int(osname, propname, 1044248571Smm (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value); 1045219089Spjd 1046219089Spjd if (error == ENOSPC) { 1047219089Spjd ztest_record_enospc(FTAG); 1048219089Spjd return (error); 1049219089Spjd } 1050240415Smm ASSERT0(error); 1051219089Spjd 1052248571Smm VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint)); 1053219089Spjd 1054236143Smm if (ztest_opts.zo_verbose >= 6) { 1055219089Spjd VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0); 1056219089Spjd (void) printf("%s %s = %s at '%s'\n", 1057219089Spjd osname, propname, valname, setpoint); 1058219089Spjd } 1059219089Spjd 1060219089Spjd return (error); 1061219089Spjd} 1062219089Spjd 1063219089Spjdstatic int 1064236143Smmztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value) 1065219089Spjd{ 1066236143Smm spa_t *spa = ztest_spa; 1067219089Spjd nvlist_t *props = NULL; 1068219089Spjd int error; 1069219089Spjd 1070219089Spjd VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); 1071219089Spjd VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0); 1072219089Spjd 1073219089Spjd error = spa_prop_set(spa, props); 1074219089Spjd 1075219089Spjd nvlist_free(props); 1076219089Spjd 1077219089Spjd if (error == ENOSPC) { 1078219089Spjd ztest_record_enospc(FTAG); 1079219089Spjd return (error); 1080219089Spjd } 1081240415Smm ASSERT0(error); 1082219089Spjd 1083219089Spjd return (error); 1084219089Spjd} 1085219089Spjd 1086168404Spjdstatic void 1087219089Spjdztest_rll_init(rll_t *rll) 1088168404Spjd{ 1089219089Spjd rll->rll_writer = NULL; 1090219089Spjd rll->rll_readers = 0; 1091219089Spjd VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0); 1092219089Spjd VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0); 1093219089Spjd} 1094219089Spjd 1095219089Spjdstatic void 1096219089Spjdztest_rll_destroy(rll_t *rll) 1097219089Spjd{ 1098219089Spjd ASSERT(rll->rll_writer == NULL); 1099219089Spjd ASSERT(rll->rll_readers == 0); 1100219089Spjd VERIFY(_mutex_destroy(&rll->rll_lock) == 0); 1101219089Spjd VERIFY(cond_destroy(&rll->rll_cv) == 0); 1102219089Spjd} 1103219089Spjd 1104219089Spjdstatic void 1105219089Spjdztest_rll_lock(rll_t *rll, rl_type_t type) 1106219089Spjd{ 1107219089Spjd VERIFY(mutex_lock(&rll->rll_lock) == 0); 1108219089Spjd 1109219089Spjd if (type == RL_READER) { 1110219089Spjd while (rll->rll_writer != NULL) 1111219089Spjd (void) cond_wait(&rll->rll_cv, &rll->rll_lock); 1112219089Spjd rll->rll_readers++; 1113219089Spjd } else { 1114219089Spjd while (rll->rll_writer != NULL || rll->rll_readers) 1115219089Spjd (void) cond_wait(&rll->rll_cv, &rll->rll_lock); 1116219089Spjd rll->rll_writer = curthread; 1117219089Spjd } 1118219089Spjd 1119219089Spjd VERIFY(mutex_unlock(&rll->rll_lock) == 0); 1120219089Spjd} 1121219089Spjd 1122219089Spjdstatic void 1123219089Spjdztest_rll_unlock(rll_t *rll) 1124219089Spjd{ 1125219089Spjd VERIFY(mutex_lock(&rll->rll_lock) == 0); 1126219089Spjd 1127219089Spjd if (rll->rll_writer) { 1128219089Spjd ASSERT(rll->rll_readers == 0); 1129219089Spjd rll->rll_writer = NULL; 1130219089Spjd } else { 1131219089Spjd ASSERT(rll->rll_readers != 0); 1132219089Spjd ASSERT(rll->rll_writer == NULL); 1133219089Spjd rll->rll_readers--; 1134219089Spjd } 1135219089Spjd 1136219089Spjd if (rll->rll_writer == NULL && rll->rll_readers == 0) 1137219089Spjd VERIFY(cond_broadcast(&rll->rll_cv) == 0); 1138219089Spjd 1139219089Spjd VERIFY(mutex_unlock(&rll->rll_lock) == 0); 1140219089Spjd} 1141219089Spjd 1142219089Spjdstatic void 1143219089Spjdztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type) 1144219089Spjd{ 1145219089Spjd rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; 1146219089Spjd 1147219089Spjd ztest_rll_lock(rll, type); 1148219089Spjd} 1149219089Spjd 1150219089Spjdstatic void 1151219089Spjdztest_object_unlock(ztest_ds_t *zd, uint64_t object) 1152219089Spjd{ 1153219089Spjd rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)]; 1154219089Spjd 1155219089Spjd ztest_rll_unlock(rll); 1156219089Spjd} 1157219089Spjd 1158219089Spjdstatic rl_t * 1159219089Spjdztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset, 1160219089Spjd uint64_t size, rl_type_t type) 1161219089Spjd{ 1162219089Spjd uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1)); 1163219089Spjd rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)]; 1164219089Spjd rl_t *rl; 1165219089Spjd 1166219089Spjd rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL); 1167219089Spjd rl->rl_object = object; 1168219089Spjd rl->rl_offset = offset; 1169219089Spjd rl->rl_size = size; 1170219089Spjd rl->rl_lock = rll; 1171219089Spjd 1172219089Spjd ztest_rll_lock(rll, type); 1173219089Spjd 1174219089Spjd return (rl); 1175219089Spjd} 1176219089Spjd 1177219089Spjdstatic void 1178219089Spjdztest_range_unlock(rl_t *rl) 1179219089Spjd{ 1180219089Spjd rll_t *rll = rl->rl_lock; 1181219089Spjd 1182219089Spjd ztest_rll_unlock(rll); 1183219089Spjd 1184219089Spjd umem_free(rl, sizeof (*rl)); 1185219089Spjd} 1186219089Spjd 1187219089Spjdstatic void 1188236143Smmztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os) 1189219089Spjd{ 1190219089Spjd zd->zd_os = os; 1191219089Spjd zd->zd_zilog = dmu_objset_zil(os); 1192236143Smm zd->zd_shared = szd; 1193219089Spjd dmu_objset_name(os, zd->zd_name); 1194219089Spjd 1195236143Smm if (zd->zd_shared != NULL) 1196236143Smm zd->zd_shared->zd_seq = 0; 1197236143Smm 1198224526Smm VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0); 1199219089Spjd VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0); 1200219089Spjd 1201219089Spjd for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++) 1202219089Spjd ztest_rll_init(&zd->zd_object_lock[l]); 1203219089Spjd 1204219089Spjd for (int l = 0; l < ZTEST_RANGE_LOCKS; l++) 1205219089Spjd ztest_rll_init(&zd->zd_range_lock[l]); 1206219089Spjd} 1207219089Spjd 1208219089Spjdstatic void 1209219089Spjdztest_zd_fini(ztest_ds_t *zd) 1210219089Spjd{ 1211219089Spjd VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0); 1212219089Spjd 1213219089Spjd for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++) 1214219089Spjd ztest_rll_destroy(&zd->zd_object_lock[l]); 1215219089Spjd 1216219089Spjd for (int l = 0; l < ZTEST_RANGE_LOCKS; l++) 1217219089Spjd ztest_rll_destroy(&zd->zd_range_lock[l]); 1218219089Spjd} 1219219089Spjd 1220219089Spjd#define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT) 1221219089Spjd 1222219089Spjdstatic uint64_t 1223219089Spjdztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag) 1224219089Spjd{ 1225219089Spjd uint64_t txg; 1226168404Spjd int error; 1227168404Spjd 1228219089Spjd /* 1229219089Spjd * Attempt to assign tx to some transaction group. 1230219089Spjd */ 1231219089Spjd error = dmu_tx_assign(tx, txg_how); 1232168404Spjd if (error) { 1233219089Spjd if (error == ERESTART) { 1234219089Spjd ASSERT(txg_how == TXG_NOWAIT); 1235219089Spjd dmu_tx_wait(tx); 1236219089Spjd } else { 1237219089Spjd ASSERT3U(error, ==, ENOSPC); 1238219089Spjd ztest_record_enospc(tag); 1239219089Spjd } 1240219089Spjd dmu_tx_abort(tx); 1241219089Spjd return (0); 1242168404Spjd } 1243219089Spjd txg = dmu_tx_get_txg(tx); 1244219089Spjd ASSERT(txg != 0); 1245219089Spjd return (txg); 1246168404Spjd} 1247168404Spjd 1248219089Spjdstatic void 1249219089Spjdztest_pattern_set(void *buf, uint64_t size, uint64_t value) 1250168404Spjd{ 1251219089Spjd uint64_t *ip = buf; 1252219089Spjd uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); 1253168404Spjd 1254219089Spjd while (ip < ip_end) 1255219089Spjd *ip++ = value; 1256219089Spjd} 1257168404Spjd 1258219089Spjdstatic boolean_t 1259219089Spjdztest_pattern_match(void *buf, uint64_t size, uint64_t value) 1260219089Spjd{ 1261219089Spjd uint64_t *ip = buf; 1262219089Spjd uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size); 1263219089Spjd uint64_t diff = 0; 1264168404Spjd 1265219089Spjd while (ip < ip_end) 1266219089Spjd diff |= (value - *ip++); 1267219089Spjd 1268219089Spjd return (diff == 0); 1269168404Spjd} 1270168404Spjd 1271219089Spjdstatic void 1272219089Spjdztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object, 1273219089Spjd uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) 1274168404Spjd{ 1275219089Spjd bt->bt_magic = BT_MAGIC; 1276219089Spjd bt->bt_objset = dmu_objset_id(os); 1277219089Spjd bt->bt_object = object; 1278219089Spjd bt->bt_offset = offset; 1279219089Spjd bt->bt_gen = gen; 1280219089Spjd bt->bt_txg = txg; 1281219089Spjd bt->bt_crtxg = crtxg; 1282168404Spjd} 1283168404Spjd 1284219089Spjdstatic void 1285219089Spjdztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object, 1286219089Spjd uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg) 1287219089Spjd{ 1288268075Sdelphij ASSERT3U(bt->bt_magic, ==, BT_MAGIC); 1289268075Sdelphij ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os)); 1290268075Sdelphij ASSERT3U(bt->bt_object, ==, object); 1291268075Sdelphij ASSERT3U(bt->bt_offset, ==, offset); 1292268075Sdelphij ASSERT3U(bt->bt_gen, <=, gen); 1293268075Sdelphij ASSERT3U(bt->bt_txg, <=, txg); 1294268075Sdelphij ASSERT3U(bt->bt_crtxg, ==, crtxg); 1295219089Spjd} 1296219089Spjd 1297219089Spjdstatic ztest_block_tag_t * 1298219089Spjdztest_bt_bonus(dmu_buf_t *db) 1299219089Spjd{ 1300219089Spjd dmu_object_info_t doi; 1301219089Spjd ztest_block_tag_t *bt; 1302219089Spjd 1303219089Spjd dmu_object_info_from_db(db, &doi); 1304219089Spjd ASSERT3U(doi.doi_bonus_size, <=, db->db_size); 1305219089Spjd ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt)); 1306219089Spjd bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt)); 1307219089Spjd 1308219089Spjd return (bt); 1309219089Spjd} 1310219089Spjd 1311219089Spjd/* 1312219089Spjd * ZIL logging ops 1313219089Spjd */ 1314219089Spjd 1315219089Spjd#define lrz_type lr_mode 1316219089Spjd#define lrz_blocksize lr_uid 1317219089Spjd#define lrz_ibshift lr_gid 1318219089Spjd#define lrz_bonustype lr_rdev 1319219089Spjd#define lrz_bonuslen lr_crtime[1] 1320219089Spjd 1321219089Spjdstatic void 1322219089Spjdztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr) 1323219089Spjd{ 1324219089Spjd char *name = (void *)(lr + 1); /* name follows lr */ 1325219089Spjd size_t namesize = strlen(name) + 1; 1326219089Spjd itx_t *itx; 1327219089Spjd 1328219089Spjd if (zil_replaying(zd->zd_zilog, tx)) 1329219089Spjd return; 1330219089Spjd 1331219089Spjd itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize); 1332219089Spjd bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1333219089Spjd sizeof (*lr) + namesize - sizeof (lr_t)); 1334219089Spjd 1335219089Spjd zil_itx_assign(zd->zd_zilog, itx, tx); 1336219089Spjd} 1337219089Spjd 1338219089Spjdstatic void 1339219089Spjdztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object) 1340219089Spjd{ 1341219089Spjd char *name = (void *)(lr + 1); /* name follows lr */ 1342219089Spjd size_t namesize = strlen(name) + 1; 1343219089Spjd itx_t *itx; 1344219089Spjd 1345219089Spjd if (zil_replaying(zd->zd_zilog, tx)) 1346219089Spjd return; 1347219089Spjd 1348219089Spjd itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize); 1349219089Spjd bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1350219089Spjd sizeof (*lr) + namesize - sizeof (lr_t)); 1351219089Spjd 1352219089Spjd itx->itx_oid = object; 1353219089Spjd zil_itx_assign(zd->zd_zilog, itx, tx); 1354219089Spjd} 1355219089Spjd 1356219089Spjdstatic void 1357219089Spjdztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr) 1358219089Spjd{ 1359219089Spjd itx_t *itx; 1360219089Spjd itx_wr_state_t write_state = ztest_random(WR_NUM_STATES); 1361219089Spjd 1362219089Spjd if (zil_replaying(zd->zd_zilog, tx)) 1363219089Spjd return; 1364219089Spjd 1365219089Spjd if (lr->lr_length > ZIL_MAX_LOG_DATA) 1366219089Spjd write_state = WR_INDIRECT; 1367219089Spjd 1368219089Spjd itx = zil_itx_create(TX_WRITE, 1369219089Spjd sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0)); 1370219089Spjd 1371219089Spjd if (write_state == WR_COPIED && 1372219089Spjd dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length, 1373219089Spjd ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) { 1374219089Spjd zil_itx_destroy(itx); 1375219089Spjd itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1376219089Spjd write_state = WR_NEED_COPY; 1377219089Spjd } 1378219089Spjd itx->itx_private = zd; 1379219089Spjd itx->itx_wr_state = write_state; 1380219089Spjd itx->itx_sync = (ztest_random(8) == 0); 1381219089Spjd itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0); 1382219089Spjd 1383219089Spjd bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1384219089Spjd sizeof (*lr) - sizeof (lr_t)); 1385219089Spjd 1386219089Spjd zil_itx_assign(zd->zd_zilog, itx, tx); 1387219089Spjd} 1388219089Spjd 1389219089Spjdstatic void 1390219089Spjdztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr) 1391219089Spjd{ 1392219089Spjd itx_t *itx; 1393219089Spjd 1394219089Spjd if (zil_replaying(zd->zd_zilog, tx)) 1395219089Spjd return; 1396219089Spjd 1397219089Spjd itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 1398219089Spjd bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1399219089Spjd sizeof (*lr) - sizeof (lr_t)); 1400219089Spjd 1401219089Spjd itx->itx_sync = B_FALSE; 1402219089Spjd zil_itx_assign(zd->zd_zilog, itx, tx); 1403219089Spjd} 1404219089Spjd 1405219089Spjdstatic void 1406219089Spjdztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr) 1407219089Spjd{ 1408219089Spjd itx_t *itx; 1409219089Spjd 1410219089Spjd if (zil_replaying(zd->zd_zilog, tx)) 1411219089Spjd return; 1412219089Spjd 1413219089Spjd itx = zil_itx_create(TX_SETATTR, sizeof (*lr)); 1414219089Spjd bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, 1415219089Spjd sizeof (*lr) - sizeof (lr_t)); 1416219089Spjd 1417219089Spjd itx->itx_sync = B_FALSE; 1418219089Spjd zil_itx_assign(zd->zd_zilog, itx, tx); 1419219089Spjd} 1420219089Spjd 1421219089Spjd/* 1422219089Spjd * ZIL replay ops 1423219089Spjd */ 1424168404Spjdstatic int 1425219089Spjdztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap) 1426168404Spjd{ 1427219089Spjd char *name = (void *)(lr + 1); /* name follows lr */ 1428219089Spjd objset_t *os = zd->zd_os; 1429219089Spjd ztest_block_tag_t *bbt; 1430219089Spjd dmu_buf_t *db; 1431168404Spjd dmu_tx_t *tx; 1432219089Spjd uint64_t txg; 1433219089Spjd int error = 0; 1434168404Spjd 1435168404Spjd if (byteswap) 1436168404Spjd byteswap_uint64_array(lr, sizeof (*lr)); 1437168404Spjd 1438219089Spjd ASSERT(lr->lr_doid == ZTEST_DIROBJ); 1439219089Spjd ASSERT(name[0] != '\0'); 1440219089Spjd 1441168404Spjd tx = dmu_tx_create(os); 1442219089Spjd 1443219089Spjd dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name); 1444219089Spjd 1445219089Spjd if (lr->lrz_type == DMU_OT_ZAP_OTHER) { 1446219089Spjd dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); 1447219089Spjd } else { 1448219089Spjd dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1449219089Spjd } 1450219089Spjd 1451219089Spjd txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1452219089Spjd if (txg == 0) 1453219089Spjd return (ENOSPC); 1454219089Spjd 1455219089Spjd ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid); 1456219089Spjd 1457219089Spjd if (lr->lrz_type == DMU_OT_ZAP_OTHER) { 1458219089Spjd if (lr->lr_foid == 0) { 1459219089Spjd lr->lr_foid = zap_create(os, 1460219089Spjd lr->lrz_type, lr->lrz_bonustype, 1461219089Spjd lr->lrz_bonuslen, tx); 1462219089Spjd } else { 1463219089Spjd error = zap_create_claim(os, lr->lr_foid, 1464219089Spjd lr->lrz_type, lr->lrz_bonustype, 1465219089Spjd lr->lrz_bonuslen, tx); 1466219089Spjd } 1467219089Spjd } else { 1468219089Spjd if (lr->lr_foid == 0) { 1469219089Spjd lr->lr_foid = dmu_object_alloc(os, 1470219089Spjd lr->lrz_type, 0, lr->lrz_bonustype, 1471219089Spjd lr->lrz_bonuslen, tx); 1472219089Spjd } else { 1473219089Spjd error = dmu_object_claim(os, lr->lr_foid, 1474219089Spjd lr->lrz_type, 0, lr->lrz_bonustype, 1475219089Spjd lr->lrz_bonuslen, tx); 1476219089Spjd } 1477219089Spjd } 1478219089Spjd 1479168404Spjd if (error) { 1480219089Spjd ASSERT3U(error, ==, EEXIST); 1481219089Spjd ASSERT(zd->zd_zilog->zl_replay); 1482219089Spjd dmu_tx_commit(tx); 1483168404Spjd return (error); 1484168404Spjd } 1485168404Spjd 1486219089Spjd ASSERT(lr->lr_foid != 0); 1487219089Spjd 1488219089Spjd if (lr->lrz_type != DMU_OT_ZAP_OTHER) 1489219089Spjd VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid, 1490219089Spjd lr->lrz_blocksize, lr->lrz_ibshift, tx)); 1491219089Spjd 1492219089Spjd VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1493219089Spjd bbt = ztest_bt_bonus(db); 1494219089Spjd dmu_buf_will_dirty(db, tx); 1495219089Spjd ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg); 1496219089Spjd dmu_buf_rele(db, FTAG); 1497219089Spjd 1498219089Spjd VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1, 1499219089Spjd &lr->lr_foid, tx)); 1500219089Spjd 1501219089Spjd (void) ztest_log_create(zd, tx, lr); 1502219089Spjd 1503168404Spjd dmu_tx_commit(tx); 1504168404Spjd 1505219089Spjd return (0); 1506219089Spjd} 1507219089Spjd 1508219089Spjdstatic int 1509219089Spjdztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap) 1510219089Spjd{ 1511219089Spjd char *name = (void *)(lr + 1); /* name follows lr */ 1512219089Spjd objset_t *os = zd->zd_os; 1513219089Spjd dmu_object_info_t doi; 1514219089Spjd dmu_tx_t *tx; 1515219089Spjd uint64_t object, txg; 1516219089Spjd 1517219089Spjd if (byteswap) 1518219089Spjd byteswap_uint64_array(lr, sizeof (*lr)); 1519219089Spjd 1520219089Spjd ASSERT(lr->lr_doid == ZTEST_DIROBJ); 1521219089Spjd ASSERT(name[0] != '\0'); 1522219089Spjd 1523219089Spjd VERIFY3U(0, ==, 1524219089Spjd zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object)); 1525219089Spjd ASSERT(object != 0); 1526219089Spjd 1527219089Spjd ztest_object_lock(zd, object, RL_WRITER); 1528219089Spjd 1529219089Spjd VERIFY3U(0, ==, dmu_object_info(os, object, &doi)); 1530219089Spjd 1531219089Spjd tx = dmu_tx_create(os); 1532219089Spjd 1533219089Spjd dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name); 1534219089Spjd dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 1535219089Spjd 1536219089Spjd txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1537219089Spjd if (txg == 0) { 1538219089Spjd ztest_object_unlock(zd, object); 1539219089Spjd return (ENOSPC); 1540168404Spjd } 1541168404Spjd 1542219089Spjd if (doi.doi_type == DMU_OT_ZAP_OTHER) { 1543219089Spjd VERIFY3U(0, ==, zap_destroy(os, object, tx)); 1544219089Spjd } else { 1545219089Spjd VERIFY3U(0, ==, dmu_object_free(os, object, tx)); 1546219089Spjd } 1547219089Spjd 1548219089Spjd VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx)); 1549219089Spjd 1550219089Spjd (void) ztest_log_remove(zd, tx, lr, object); 1551219089Spjd 1552219089Spjd dmu_tx_commit(tx); 1553219089Spjd 1554219089Spjd ztest_object_unlock(zd, object); 1555219089Spjd 1556219089Spjd return (0); 1557168404Spjd} 1558168404Spjd 1559168404Spjdstatic int 1560219089Spjdztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap) 1561168404Spjd{ 1562219089Spjd objset_t *os = zd->zd_os; 1563219089Spjd void *data = lr + 1; /* data follows lr */ 1564219089Spjd uint64_t offset, length; 1565219089Spjd ztest_block_tag_t *bt = data; 1566219089Spjd ztest_block_tag_t *bbt; 1567219089Spjd uint64_t gen, txg, lrtxg, crtxg; 1568219089Spjd dmu_object_info_t doi; 1569168404Spjd dmu_tx_t *tx; 1570219089Spjd dmu_buf_t *db; 1571219089Spjd arc_buf_t *abuf = NULL; 1572219089Spjd rl_t *rl; 1573168404Spjd 1574168404Spjd if (byteswap) 1575168404Spjd byteswap_uint64_array(lr, sizeof (*lr)); 1576168404Spjd 1577219089Spjd offset = lr->lr_offset; 1578219089Spjd length = lr->lr_length; 1579219089Spjd 1580219089Spjd /* If it's a dmu_sync() block, write the whole block */ 1581219089Spjd if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 1582219089Spjd uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 1583219089Spjd if (length < blocksize) { 1584219089Spjd offset -= offset % blocksize; 1585219089Spjd length = blocksize; 1586219089Spjd } 1587219089Spjd } 1588219089Spjd 1589219089Spjd if (bt->bt_magic == BSWAP_64(BT_MAGIC)) 1590219089Spjd byteswap_uint64_array(bt, sizeof (*bt)); 1591219089Spjd 1592219089Spjd if (bt->bt_magic != BT_MAGIC) 1593219089Spjd bt = NULL; 1594219089Spjd 1595219089Spjd ztest_object_lock(zd, lr->lr_foid, RL_READER); 1596219089Spjd rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER); 1597219089Spjd 1598219089Spjd VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1599219089Spjd 1600219089Spjd dmu_object_info_from_db(db, &doi); 1601219089Spjd 1602219089Spjd bbt = ztest_bt_bonus(db); 1603219089Spjd ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1604219089Spjd gen = bbt->bt_gen; 1605219089Spjd crtxg = bbt->bt_crtxg; 1606219089Spjd lrtxg = lr->lr_common.lrc_txg; 1607219089Spjd 1608168404Spjd tx = dmu_tx_create(os); 1609219089Spjd 1610219089Spjd dmu_tx_hold_write(tx, lr->lr_foid, offset, length); 1611219089Spjd 1612219089Spjd if (ztest_random(8) == 0 && length == doi.doi_data_block_size && 1613219089Spjd P2PHASE(offset, length) == 0) 1614219089Spjd abuf = dmu_request_arcbuf(db, length); 1615219089Spjd 1616219089Spjd txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1617219089Spjd if (txg == 0) { 1618219089Spjd if (abuf != NULL) 1619219089Spjd dmu_return_arcbuf(abuf); 1620219089Spjd dmu_buf_rele(db, FTAG); 1621219089Spjd ztest_range_unlock(rl); 1622219089Spjd ztest_object_unlock(zd, lr->lr_foid); 1623219089Spjd return (ENOSPC); 1624168404Spjd } 1625168404Spjd 1626219089Spjd if (bt != NULL) { 1627219089Spjd /* 1628219089Spjd * Usually, verify the old data before writing new data -- 1629219089Spjd * but not always, because we also want to verify correct 1630219089Spjd * behavior when the data was not recently read into cache. 1631219089Spjd */ 1632219089Spjd ASSERT(offset % doi.doi_data_block_size == 0); 1633219089Spjd if (ztest_random(4) != 0) { 1634219089Spjd int prefetch = ztest_random(2) ? 1635219089Spjd DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH; 1636219089Spjd ztest_block_tag_t rbt; 1637219089Spjd 1638219089Spjd VERIFY(dmu_read(os, lr->lr_foid, offset, 1639219089Spjd sizeof (rbt), &rbt, prefetch) == 0); 1640219089Spjd if (rbt.bt_magic == BT_MAGIC) { 1641219089Spjd ztest_bt_verify(&rbt, os, lr->lr_foid, 1642219089Spjd offset, gen, txg, crtxg); 1643219089Spjd } 1644219089Spjd } 1645219089Spjd 1646219089Spjd /* 1647219089Spjd * Writes can appear to be newer than the bonus buffer because 1648219089Spjd * the ztest_get_data() callback does a dmu_read() of the 1649219089Spjd * open-context data, which may be different than the data 1650219089Spjd * as it was when the write was generated. 1651219089Spjd */ 1652219089Spjd if (zd->zd_zilog->zl_replay) { 1653219089Spjd ztest_bt_verify(bt, os, lr->lr_foid, offset, 1654219089Spjd MAX(gen, bt->bt_gen), MAX(txg, lrtxg), 1655219089Spjd bt->bt_crtxg); 1656219089Spjd } 1657219089Spjd 1658219089Spjd /* 1659219089Spjd * Set the bt's gen/txg to the bonus buffer's gen/txg 1660219089Spjd * so that all of the usual ASSERTs will work. 1661219089Spjd */ 1662219089Spjd ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg); 1663219089Spjd } 1664219089Spjd 1665219089Spjd if (abuf == NULL) { 1666219089Spjd dmu_write(os, lr->lr_foid, offset, length, data, tx); 1667219089Spjd } else { 1668219089Spjd bcopy(data, abuf->b_data, length); 1669219089Spjd dmu_assign_arcbuf(db, offset, abuf, tx); 1670219089Spjd } 1671219089Spjd 1672219089Spjd (void) ztest_log_write(zd, tx, lr); 1673219089Spjd 1674219089Spjd dmu_buf_rele(db, FTAG); 1675219089Spjd 1676168404Spjd dmu_tx_commit(tx); 1677168404Spjd 1678219089Spjd ztest_range_unlock(rl); 1679219089Spjd ztest_object_unlock(zd, lr->lr_foid); 1680219089Spjd 1681219089Spjd return (0); 1682168404Spjd} 1683168404Spjd 1684219089Spjdstatic int 1685219089Spjdztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap) 1686219089Spjd{ 1687219089Spjd objset_t *os = zd->zd_os; 1688219089Spjd dmu_tx_t *tx; 1689219089Spjd uint64_t txg; 1690219089Spjd rl_t *rl; 1691219089Spjd 1692219089Spjd if (byteswap) 1693219089Spjd byteswap_uint64_array(lr, sizeof (*lr)); 1694219089Spjd 1695219089Spjd ztest_object_lock(zd, lr->lr_foid, RL_READER); 1696219089Spjd rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length, 1697219089Spjd RL_WRITER); 1698219089Spjd 1699219089Spjd tx = dmu_tx_create(os); 1700219089Spjd 1701219089Spjd dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length); 1702219089Spjd 1703219089Spjd txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1704219089Spjd if (txg == 0) { 1705219089Spjd ztest_range_unlock(rl); 1706219089Spjd ztest_object_unlock(zd, lr->lr_foid); 1707219089Spjd return (ENOSPC); 1708219089Spjd } 1709219089Spjd 1710219089Spjd VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset, 1711219089Spjd lr->lr_length, tx) == 0); 1712219089Spjd 1713219089Spjd (void) ztest_log_truncate(zd, tx, lr); 1714219089Spjd 1715219089Spjd dmu_tx_commit(tx); 1716219089Spjd 1717219089Spjd ztest_range_unlock(rl); 1718219089Spjd ztest_object_unlock(zd, lr->lr_foid); 1719219089Spjd 1720219089Spjd return (0); 1721219089Spjd} 1722219089Spjd 1723219089Spjdstatic int 1724219089Spjdztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap) 1725219089Spjd{ 1726219089Spjd objset_t *os = zd->zd_os; 1727219089Spjd dmu_tx_t *tx; 1728219089Spjd dmu_buf_t *db; 1729219089Spjd ztest_block_tag_t *bbt; 1730219089Spjd uint64_t txg, lrtxg, crtxg; 1731219089Spjd 1732219089Spjd if (byteswap) 1733219089Spjd byteswap_uint64_array(lr, sizeof (*lr)); 1734219089Spjd 1735219089Spjd ztest_object_lock(zd, lr->lr_foid, RL_WRITER); 1736219089Spjd 1737219089Spjd VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db)); 1738219089Spjd 1739219089Spjd tx = dmu_tx_create(os); 1740219089Spjd dmu_tx_hold_bonus(tx, lr->lr_foid); 1741219089Spjd 1742219089Spjd txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 1743219089Spjd if (txg == 0) { 1744219089Spjd dmu_buf_rele(db, FTAG); 1745219089Spjd ztest_object_unlock(zd, lr->lr_foid); 1746219089Spjd return (ENOSPC); 1747219089Spjd } 1748219089Spjd 1749219089Spjd bbt = ztest_bt_bonus(db); 1750219089Spjd ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1751219089Spjd crtxg = bbt->bt_crtxg; 1752219089Spjd lrtxg = lr->lr_common.lrc_txg; 1753219089Spjd 1754219089Spjd if (zd->zd_zilog->zl_replay) { 1755219089Spjd ASSERT(lr->lr_size != 0); 1756219089Spjd ASSERT(lr->lr_mode != 0); 1757219089Spjd ASSERT(lrtxg != 0); 1758219089Spjd } else { 1759219089Spjd /* 1760219089Spjd * Randomly change the size and increment the generation. 1761219089Spjd */ 1762219089Spjd lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) * 1763219089Spjd sizeof (*bbt); 1764219089Spjd lr->lr_mode = bbt->bt_gen + 1; 1765219089Spjd ASSERT(lrtxg == 0); 1766219089Spjd } 1767219089Spjd 1768219089Spjd /* 1769219089Spjd * Verify that the current bonus buffer is not newer than our txg. 1770219089Spjd */ 1771219089Spjd ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, 1772219089Spjd MAX(txg, lrtxg), crtxg); 1773219089Spjd 1774219089Spjd dmu_buf_will_dirty(db, tx); 1775219089Spjd 1776219089Spjd ASSERT3U(lr->lr_size, >=, sizeof (*bbt)); 1777219089Spjd ASSERT3U(lr->lr_size, <=, db->db_size); 1778240415Smm VERIFY0(dmu_set_bonus(db, lr->lr_size, tx)); 1779219089Spjd bbt = ztest_bt_bonus(db); 1780219089Spjd 1781219089Spjd ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg); 1782219089Spjd 1783219089Spjd dmu_buf_rele(db, FTAG); 1784219089Spjd 1785219089Spjd (void) ztest_log_setattr(zd, tx, lr); 1786219089Spjd 1787219089Spjd dmu_tx_commit(tx); 1788219089Spjd 1789219089Spjd ztest_object_unlock(zd, lr->lr_foid); 1790219089Spjd 1791219089Spjd return (0); 1792219089Spjd} 1793219089Spjd 1794168404Spjdzil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { 1795168404Spjd NULL, /* 0 no such transaction type */ 1796168404Spjd ztest_replay_create, /* TX_CREATE */ 1797168404Spjd NULL, /* TX_MKDIR */ 1798168404Spjd NULL, /* TX_MKXATTR */ 1799168404Spjd NULL, /* TX_SYMLINK */ 1800168404Spjd ztest_replay_remove, /* TX_REMOVE */ 1801168404Spjd NULL, /* TX_RMDIR */ 1802168404Spjd NULL, /* TX_LINK */ 1803168404Spjd NULL, /* TX_RENAME */ 1804219089Spjd ztest_replay_write, /* TX_WRITE */ 1805219089Spjd ztest_replay_truncate, /* TX_TRUNCATE */ 1806219089Spjd ztest_replay_setattr, /* TX_SETATTR */ 1807168404Spjd NULL, /* TX_ACL */ 1808209962Smm NULL, /* TX_CREATE_ACL */ 1809209962Smm NULL, /* TX_CREATE_ATTR */ 1810209962Smm NULL, /* TX_CREATE_ACL_ATTR */ 1811209962Smm NULL, /* TX_MKDIR_ACL */ 1812209962Smm NULL, /* TX_MKDIR_ATTR */ 1813209962Smm NULL, /* TX_MKDIR_ACL_ATTR */ 1814209962Smm NULL, /* TX_WRITE2 */ 1815168404Spjd}; 1816168404Spjd 1817168404Spjd/* 1818219089Spjd * ZIL get_data callbacks 1819219089Spjd */ 1820219089Spjd 1821219089Spjdstatic void 1822219089Spjdztest_get_done(zgd_t *zgd, int error) 1823219089Spjd{ 1824219089Spjd ztest_ds_t *zd = zgd->zgd_private; 1825219089Spjd uint64_t object = zgd->zgd_rl->rl_object; 1826219089Spjd 1827219089Spjd if (zgd->zgd_db) 1828219089Spjd dmu_buf_rele(zgd->zgd_db, zgd); 1829219089Spjd 1830219089Spjd ztest_range_unlock(zgd->zgd_rl); 1831219089Spjd ztest_object_unlock(zd, object); 1832219089Spjd 1833219089Spjd if (error == 0 && zgd->zgd_bp) 1834219089Spjd zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1835219089Spjd 1836219089Spjd umem_free(zgd, sizeof (*zgd)); 1837219089Spjd} 1838219089Spjd 1839219089Spjdstatic int 1840219089Spjdztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1841219089Spjd{ 1842219089Spjd ztest_ds_t *zd = arg; 1843219089Spjd objset_t *os = zd->zd_os; 1844219089Spjd uint64_t object = lr->lr_foid; 1845219089Spjd uint64_t offset = lr->lr_offset; 1846219089Spjd uint64_t size = lr->lr_length; 1847219089Spjd blkptr_t *bp = &lr->lr_blkptr; 1848219089Spjd uint64_t txg = lr->lr_common.lrc_txg; 1849219089Spjd uint64_t crtxg; 1850219089Spjd dmu_object_info_t doi; 1851219089Spjd dmu_buf_t *db; 1852219089Spjd zgd_t *zgd; 1853219089Spjd int error; 1854219089Spjd 1855219089Spjd ztest_object_lock(zd, object, RL_READER); 1856219089Spjd error = dmu_bonus_hold(os, object, FTAG, &db); 1857219089Spjd if (error) { 1858219089Spjd ztest_object_unlock(zd, object); 1859219089Spjd return (error); 1860219089Spjd } 1861219089Spjd 1862219089Spjd crtxg = ztest_bt_bonus(db)->bt_crtxg; 1863219089Spjd 1864219089Spjd if (crtxg == 0 || crtxg > txg) { 1865219089Spjd dmu_buf_rele(db, FTAG); 1866219089Spjd ztest_object_unlock(zd, object); 1867219089Spjd return (ENOENT); 1868219089Spjd } 1869219089Spjd 1870219089Spjd dmu_object_info_from_db(db, &doi); 1871219089Spjd dmu_buf_rele(db, FTAG); 1872219089Spjd db = NULL; 1873219089Spjd 1874219089Spjd zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL); 1875219089Spjd zgd->zgd_zilog = zd->zd_zilog; 1876219089Spjd zgd->zgd_private = zd; 1877219089Spjd 1878219089Spjd if (buf != NULL) { /* immediate write */ 1879219089Spjd zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, 1880219089Spjd RL_READER); 1881219089Spjd 1882219089Spjd error = dmu_read(os, object, offset, size, buf, 1883219089Spjd DMU_READ_NO_PREFETCH); 1884219089Spjd ASSERT(error == 0); 1885219089Spjd } else { 1886219089Spjd size = doi.doi_data_block_size; 1887219089Spjd if (ISP2(size)) { 1888219089Spjd offset = P2ALIGN(offset, size); 1889219089Spjd } else { 1890219089Spjd ASSERT(offset < size); 1891219089Spjd offset = 0; 1892219089Spjd } 1893219089Spjd 1894219089Spjd zgd->zgd_rl = ztest_range_lock(zd, object, offset, size, 1895219089Spjd RL_READER); 1896219089Spjd 1897219089Spjd error = dmu_buf_hold(os, object, offset, zgd, &db, 1898219089Spjd DMU_READ_NO_PREFETCH); 1899219089Spjd 1900219089Spjd if (error == 0) { 1901243524Smm blkptr_t *obp = dmu_buf_get_blkptr(db); 1902243524Smm if (obp) { 1903243524Smm ASSERT(BP_IS_HOLE(bp)); 1904243524Smm *bp = *obp; 1905243524Smm } 1906243524Smm 1907219089Spjd zgd->zgd_db = db; 1908219089Spjd zgd->zgd_bp = bp; 1909219089Spjd 1910219089Spjd ASSERT(db->db_offset == offset); 1911219089Spjd ASSERT(db->db_size == size); 1912219089Spjd 1913219089Spjd error = dmu_sync(zio, lr->lr_common.lrc_txg, 1914219089Spjd ztest_get_done, zgd); 1915219089Spjd 1916219089Spjd if (error == 0) 1917219089Spjd return (0); 1918219089Spjd } 1919219089Spjd } 1920219089Spjd 1921219089Spjd ztest_get_done(zgd, error); 1922219089Spjd 1923219089Spjd return (error); 1924219089Spjd} 1925219089Spjd 1926219089Spjdstatic void * 1927219089Spjdztest_lr_alloc(size_t lrsize, char *name) 1928219089Spjd{ 1929219089Spjd char *lr; 1930219089Spjd size_t namesize = name ? strlen(name) + 1 : 0; 1931219089Spjd 1932219089Spjd lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL); 1933219089Spjd 1934219089Spjd if (name) 1935219089Spjd bcopy(name, lr + lrsize, namesize); 1936219089Spjd 1937219089Spjd return (lr); 1938219089Spjd} 1939219089Spjd 1940219089Spjdvoid 1941219089Spjdztest_lr_free(void *lr, size_t lrsize, char *name) 1942219089Spjd{ 1943219089Spjd size_t namesize = name ? strlen(name) + 1 : 0; 1944219089Spjd 1945219089Spjd umem_free(lr, lrsize + namesize); 1946219089Spjd} 1947219089Spjd 1948219089Spjd/* 1949219089Spjd * Lookup a bunch of objects. Returns the number of objects not found. 1950219089Spjd */ 1951219089Spjdstatic int 1952219089Spjdztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count) 1953219089Spjd{ 1954219089Spjd int missing = 0; 1955219089Spjd int error; 1956219089Spjd 1957219089Spjd ASSERT(_mutex_held(&zd->zd_dirobj_lock)); 1958219089Spjd 1959219089Spjd for (int i = 0; i < count; i++, od++) { 1960219089Spjd od->od_object = 0; 1961219089Spjd error = zap_lookup(zd->zd_os, od->od_dir, od->od_name, 1962219089Spjd sizeof (uint64_t), 1, &od->od_object); 1963219089Spjd if (error) { 1964219089Spjd ASSERT(error == ENOENT); 1965219089Spjd ASSERT(od->od_object == 0); 1966219089Spjd missing++; 1967219089Spjd } else { 1968219089Spjd dmu_buf_t *db; 1969219089Spjd ztest_block_tag_t *bbt; 1970219089Spjd dmu_object_info_t doi; 1971219089Spjd 1972219089Spjd ASSERT(od->od_object != 0); 1973219089Spjd ASSERT(missing == 0); /* there should be no gaps */ 1974219089Spjd 1975219089Spjd ztest_object_lock(zd, od->od_object, RL_READER); 1976219089Spjd VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os, 1977219089Spjd od->od_object, FTAG, &db)); 1978219089Spjd dmu_object_info_from_db(db, &doi); 1979219089Spjd bbt = ztest_bt_bonus(db); 1980219089Spjd ASSERT3U(bbt->bt_magic, ==, BT_MAGIC); 1981219089Spjd od->od_type = doi.doi_type; 1982219089Spjd od->od_blocksize = doi.doi_data_block_size; 1983219089Spjd od->od_gen = bbt->bt_gen; 1984219089Spjd dmu_buf_rele(db, FTAG); 1985219089Spjd ztest_object_unlock(zd, od->od_object); 1986219089Spjd } 1987219089Spjd } 1988219089Spjd 1989219089Spjd return (missing); 1990219089Spjd} 1991219089Spjd 1992219089Spjdstatic int 1993219089Spjdztest_create(ztest_ds_t *zd, ztest_od_t *od, int count) 1994219089Spjd{ 1995219089Spjd int missing = 0; 1996219089Spjd 1997219089Spjd ASSERT(_mutex_held(&zd->zd_dirobj_lock)); 1998219089Spjd 1999219089Spjd for (int i = 0; i < count; i++, od++) { 2000219089Spjd if (missing) { 2001219089Spjd od->od_object = 0; 2002219089Spjd missing++; 2003219089Spjd continue; 2004219089Spjd } 2005219089Spjd 2006219089Spjd lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); 2007219089Spjd 2008219089Spjd lr->lr_doid = od->od_dir; 2009219089Spjd lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */ 2010219089Spjd lr->lrz_type = od->od_crtype; 2011219089Spjd lr->lrz_blocksize = od->od_crblocksize; 2012219089Spjd lr->lrz_ibshift = ztest_random_ibshift(); 2013219089Spjd lr->lrz_bonustype = DMU_OT_UINT64_OTHER; 2014219089Spjd lr->lrz_bonuslen = dmu_bonus_max(); 2015219089Spjd lr->lr_gen = od->od_crgen; 2016219089Spjd lr->lr_crtime[0] = time(NULL); 2017219089Spjd 2018219089Spjd if (ztest_replay_create(zd, lr, B_FALSE) != 0) { 2019219089Spjd ASSERT(missing == 0); 2020219089Spjd od->od_object = 0; 2021219089Spjd missing++; 2022219089Spjd } else { 2023219089Spjd od->od_object = lr->lr_foid; 2024219089Spjd od->od_type = od->od_crtype; 2025219089Spjd od->od_blocksize = od->od_crblocksize; 2026219089Spjd od->od_gen = od->od_crgen; 2027219089Spjd ASSERT(od->od_object != 0); 2028219089Spjd } 2029219089Spjd 2030219089Spjd ztest_lr_free(lr, sizeof (*lr), od->od_name); 2031219089Spjd } 2032219089Spjd 2033219089Spjd return (missing); 2034219089Spjd} 2035219089Spjd 2036219089Spjdstatic int 2037219089Spjdztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count) 2038219089Spjd{ 2039219089Spjd int missing = 0; 2040219089Spjd int error; 2041219089Spjd 2042219089Spjd ASSERT(_mutex_held(&zd->zd_dirobj_lock)); 2043219089Spjd 2044219089Spjd od += count - 1; 2045219089Spjd 2046219089Spjd for (int i = count - 1; i >= 0; i--, od--) { 2047219089Spjd if (missing) { 2048219089Spjd missing++; 2049219089Spjd continue; 2050219089Spjd } 2051219089Spjd 2052243524Smm /* 2053243524Smm * No object was found. 2054243524Smm */ 2055219089Spjd if (od->od_object == 0) 2056219089Spjd continue; 2057219089Spjd 2058219089Spjd lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name); 2059219089Spjd 2060219089Spjd lr->lr_doid = od->od_dir; 2061219089Spjd 2062219089Spjd if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) { 2063219089Spjd ASSERT3U(error, ==, ENOSPC); 2064219089Spjd missing++; 2065219089Spjd } else { 2066219089Spjd od->od_object = 0; 2067219089Spjd } 2068219089Spjd ztest_lr_free(lr, sizeof (*lr), od->od_name); 2069219089Spjd } 2070219089Spjd 2071219089Spjd return (missing); 2072219089Spjd} 2073219089Spjd 2074219089Spjdstatic int 2075219089Spjdztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size, 2076219089Spjd void *data) 2077219089Spjd{ 2078219089Spjd lr_write_t *lr; 2079219089Spjd int error; 2080219089Spjd 2081219089Spjd lr = ztest_lr_alloc(sizeof (*lr) + size, NULL); 2082219089Spjd 2083219089Spjd lr->lr_foid = object; 2084219089Spjd lr->lr_offset = offset; 2085219089Spjd lr->lr_length = size; 2086219089Spjd lr->lr_blkoff = 0; 2087219089Spjd BP_ZERO(&lr->lr_blkptr); 2088219089Spjd 2089219089Spjd bcopy(data, lr + 1, size); 2090219089Spjd 2091219089Spjd error = ztest_replay_write(zd, lr, B_FALSE); 2092219089Spjd 2093219089Spjd ztest_lr_free(lr, sizeof (*lr) + size, NULL); 2094219089Spjd 2095219089Spjd return (error); 2096219089Spjd} 2097219089Spjd 2098219089Spjdstatic int 2099219089Spjdztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) 2100219089Spjd{ 2101219089Spjd lr_truncate_t *lr; 2102219089Spjd int error; 2103219089Spjd 2104219089Spjd lr = ztest_lr_alloc(sizeof (*lr), NULL); 2105219089Spjd 2106219089Spjd lr->lr_foid = object; 2107219089Spjd lr->lr_offset = offset; 2108219089Spjd lr->lr_length = size; 2109219089Spjd 2110219089Spjd error = ztest_replay_truncate(zd, lr, B_FALSE); 2111219089Spjd 2112219089Spjd ztest_lr_free(lr, sizeof (*lr), NULL); 2113219089Spjd 2114219089Spjd return (error); 2115219089Spjd} 2116219089Spjd 2117219089Spjdstatic int 2118219089Spjdztest_setattr(ztest_ds_t *zd, uint64_t object) 2119219089Spjd{ 2120219089Spjd lr_setattr_t *lr; 2121219089Spjd int error; 2122219089Spjd 2123219089Spjd lr = ztest_lr_alloc(sizeof (*lr), NULL); 2124219089Spjd 2125219089Spjd lr->lr_foid = object; 2126219089Spjd lr->lr_size = 0; 2127219089Spjd lr->lr_mode = 0; 2128219089Spjd 2129219089Spjd error = ztest_replay_setattr(zd, lr, B_FALSE); 2130219089Spjd 2131219089Spjd ztest_lr_free(lr, sizeof (*lr), NULL); 2132219089Spjd 2133219089Spjd return (error); 2134219089Spjd} 2135219089Spjd 2136219089Spjdstatic void 2137219089Spjdztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size) 2138219089Spjd{ 2139219089Spjd objset_t *os = zd->zd_os; 2140219089Spjd dmu_tx_t *tx; 2141219089Spjd uint64_t txg; 2142219089Spjd rl_t *rl; 2143219089Spjd 2144219089Spjd txg_wait_synced(dmu_objset_pool(os), 0); 2145219089Spjd 2146219089Spjd ztest_object_lock(zd, object, RL_READER); 2147219089Spjd rl = ztest_range_lock(zd, object, offset, size, RL_WRITER); 2148219089Spjd 2149219089Spjd tx = dmu_tx_create(os); 2150219089Spjd 2151219089Spjd dmu_tx_hold_write(tx, object, offset, size); 2152219089Spjd 2153219089Spjd txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 2154219089Spjd 2155219089Spjd if (txg != 0) { 2156219089Spjd dmu_prealloc(os, object, offset, size, tx); 2157219089Spjd dmu_tx_commit(tx); 2158219089Spjd txg_wait_synced(dmu_objset_pool(os), txg); 2159219089Spjd } else { 2160219089Spjd (void) dmu_free_long_range(os, object, offset, size); 2161219089Spjd } 2162219089Spjd 2163219089Spjd ztest_range_unlock(rl); 2164219089Spjd ztest_object_unlock(zd, object); 2165219089Spjd} 2166219089Spjd 2167219089Spjdstatic void 2168219089Spjdztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset) 2169219089Spjd{ 2170243524Smm int err; 2171219089Spjd ztest_block_tag_t wbt; 2172219089Spjd dmu_object_info_t doi; 2173219089Spjd enum ztest_io_type io_type; 2174219089Spjd uint64_t blocksize; 2175219089Spjd void *data; 2176219089Spjd 2177219089Spjd VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0); 2178219089Spjd blocksize = doi.doi_data_block_size; 2179219089Spjd data = umem_alloc(blocksize, UMEM_NOFAIL); 2180219089Spjd 2181219089Spjd /* 2182219089Spjd * Pick an i/o type at random, biased toward writing block tags. 2183219089Spjd */ 2184219089Spjd io_type = ztest_random(ZTEST_IO_TYPES); 2185219089Spjd if (ztest_random(2) == 0) 2186219089Spjd io_type = ZTEST_IO_WRITE_TAG; 2187219089Spjd 2188224526Smm (void) rw_rdlock(&zd->zd_zilog_lock); 2189224526Smm 2190219089Spjd switch (io_type) { 2191219089Spjd 2192219089Spjd case ZTEST_IO_WRITE_TAG: 2193219089Spjd ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0); 2194219089Spjd (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt); 2195219089Spjd break; 2196219089Spjd 2197219089Spjd case ZTEST_IO_WRITE_PATTERN: 2198219089Spjd (void) memset(data, 'a' + (object + offset) % 5, blocksize); 2199219089Spjd if (ztest_random(2) == 0) { 2200219089Spjd /* 2201219089Spjd * Induce fletcher2 collisions to ensure that 2202219089Spjd * zio_ddt_collision() detects and resolves them 2203219089Spjd * when using fletcher2-verify for deduplication. 2204219089Spjd */ 2205219089Spjd ((uint64_t *)data)[0] ^= 1ULL << 63; 2206219089Spjd ((uint64_t *)data)[4] ^= 1ULL << 63; 2207219089Spjd } 2208219089Spjd (void) ztest_write(zd, object, offset, blocksize, data); 2209219089Spjd break; 2210219089Spjd 2211219089Spjd case ZTEST_IO_WRITE_ZEROES: 2212219089Spjd bzero(data, blocksize); 2213219089Spjd (void) ztest_write(zd, object, offset, blocksize, data); 2214219089Spjd break; 2215219089Spjd 2216219089Spjd case ZTEST_IO_TRUNCATE: 2217219089Spjd (void) ztest_truncate(zd, object, offset, blocksize); 2218219089Spjd break; 2219219089Spjd 2220219089Spjd case ZTEST_IO_SETATTR: 2221219089Spjd (void) ztest_setattr(zd, object); 2222219089Spjd break; 2223243524Smm 2224243524Smm case ZTEST_IO_REWRITE: 2225243524Smm (void) rw_rdlock(&ztest_name_lock); 2226243524Smm err = ztest_dsl_prop_set_uint64(zd->zd_name, 2227243524Smm ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa), 2228243524Smm B_FALSE); 2229243524Smm VERIFY(err == 0 || err == ENOSPC); 2230243524Smm err = ztest_dsl_prop_set_uint64(zd->zd_name, 2231243524Smm ZFS_PROP_COMPRESSION, 2232243524Smm ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), 2233243524Smm B_FALSE); 2234243524Smm VERIFY(err == 0 || err == ENOSPC); 2235243524Smm (void) rw_unlock(&ztest_name_lock); 2236243524Smm 2237243524Smm VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data, 2238243524Smm DMU_READ_NO_PREFETCH)); 2239243524Smm 2240243524Smm (void) ztest_write(zd, object, offset, blocksize, data); 2241243524Smm break; 2242219089Spjd } 2243219089Spjd 2244224526Smm (void) rw_unlock(&zd->zd_zilog_lock); 2245224526Smm 2246219089Spjd umem_free(data, blocksize); 2247219089Spjd} 2248219089Spjd 2249219089Spjd/* 2250219089Spjd * Initialize an object description template. 2251219089Spjd */ 2252219089Spjdstatic void 2253219089Spjdztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index, 2254219089Spjd dmu_object_type_t type, uint64_t blocksize, uint64_t gen) 2255219089Spjd{ 2256219089Spjd od->od_dir = ZTEST_DIROBJ; 2257219089Spjd od->od_object = 0; 2258219089Spjd 2259219089Spjd od->od_crtype = type; 2260219089Spjd od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize(); 2261219089Spjd od->od_crgen = gen; 2262219089Spjd 2263219089Spjd od->od_type = DMU_OT_NONE; 2264219089Spjd od->od_blocksize = 0; 2265219089Spjd od->od_gen = 0; 2266219089Spjd 2267219089Spjd (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]", 2268219089Spjd tag, (int64_t)id, index); 2269219089Spjd} 2270219089Spjd 2271219089Spjd/* 2272219089Spjd * Lookup or create the objects for a test using the od template. 2273219089Spjd * If the objects do not all exist, or if 'remove' is specified, 2274219089Spjd * remove any existing objects and create new ones. Otherwise, 2275219089Spjd * use the existing objects. 2276219089Spjd */ 2277219089Spjdstatic int 2278219089Spjdztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove) 2279219089Spjd{ 2280219089Spjd int count = size / sizeof (*od); 2281219089Spjd int rv = 0; 2282219089Spjd 2283219089Spjd VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0); 2284219089Spjd if ((ztest_lookup(zd, od, count) != 0 || remove) && 2285219089Spjd (ztest_remove(zd, od, count) != 0 || 2286219089Spjd ztest_create(zd, od, count) != 0)) 2287219089Spjd rv = -1; 2288219089Spjd zd->zd_od = od; 2289219089Spjd VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0); 2290219089Spjd 2291219089Spjd return (rv); 2292219089Spjd} 2293219089Spjd 2294219089Spjd/* ARGSUSED */ 2295219089Spjdvoid 2296219089Spjdztest_zil_commit(ztest_ds_t *zd, uint64_t id) 2297219089Spjd{ 2298219089Spjd zilog_t *zilog = zd->zd_zilog; 2299219089Spjd 2300224526Smm (void) rw_rdlock(&zd->zd_zilog_lock); 2301224526Smm 2302219089Spjd zil_commit(zilog, ztest_random(ZTEST_OBJECTS)); 2303219089Spjd 2304219089Spjd /* 2305219089Spjd * Remember the committed values in zd, which is in parent/child 2306219089Spjd * shared memory. If we die, the next iteration of ztest_run() 2307219089Spjd * will verify that the log really does contain this record. 2308219089Spjd */ 2309219089Spjd mutex_enter(&zilog->zl_lock); 2310236143Smm ASSERT(zd->zd_shared != NULL); 2311236143Smm ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq); 2312236143Smm zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq; 2313219089Spjd mutex_exit(&zilog->zl_lock); 2314224526Smm 2315224526Smm (void) rw_unlock(&zd->zd_zilog_lock); 2316219089Spjd} 2317219089Spjd 2318219089Spjd/* 2319224526Smm * This function is designed to simulate the operations that occur during a 2320224526Smm * mount/unmount operation. We hold the dataset across these operations in an 2321224526Smm * attempt to expose any implicit assumptions about ZIL management. 2322224526Smm */ 2323224526Smm/* ARGSUSED */ 2324224526Smmvoid 2325224526Smmztest_zil_remount(ztest_ds_t *zd, uint64_t id) 2326224526Smm{ 2327224526Smm objset_t *os = zd->zd_os; 2328224526Smm 2329243524Smm /* 2330243524Smm * We grab the zd_dirobj_lock to ensure that no other thread is 2331243524Smm * updating the zil (i.e. adding in-memory log records) and the 2332243524Smm * zd_zilog_lock to block any I/O. 2333243524Smm */ 2334243524Smm VERIFY0(mutex_lock(&zd->zd_dirobj_lock)); 2335224526Smm (void) rw_wrlock(&zd->zd_zilog_lock); 2336224526Smm 2337224526Smm /* zfsvfs_teardown() */ 2338224526Smm zil_close(zd->zd_zilog); 2339224526Smm 2340224526Smm /* zfsvfs_setup() */ 2341224526Smm VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog); 2342224526Smm zil_replay(os, zd, ztest_replay_vector); 2343224526Smm 2344224526Smm (void) rw_unlock(&zd->zd_zilog_lock); 2345239620Smm VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0); 2346224526Smm} 2347224526Smm 2348224526Smm/* 2349168404Spjd * Verify that we can't destroy an active pool, create an existing pool, 2350168404Spjd * or create a pool with a bad vdev spec. 2351168404Spjd */ 2352219089Spjd/* ARGSUSED */ 2353168404Spjdvoid 2354219089Spjdztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id) 2355168404Spjd{ 2356236143Smm ztest_shared_opts_t *zo = &ztest_opts; 2357168404Spjd spa_t *spa; 2358168404Spjd nvlist_t *nvroot; 2359168404Spjd 2360168404Spjd /* 2361168404Spjd * Attempt to create using a bad file. 2362168404Spjd */ 2363243505Smm nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); 2364219089Spjd VERIFY3U(ENOENT, ==, 2365248571Smm spa_create("ztest_bad_file", nvroot, NULL, NULL)); 2366168404Spjd nvlist_free(nvroot); 2367168404Spjd 2368168404Spjd /* 2369168404Spjd * Attempt to create using a bad mirror. 2370168404Spjd */ 2371243505Smm nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1); 2372219089Spjd VERIFY3U(ENOENT, ==, 2373248571Smm spa_create("ztest_bad_mirror", nvroot, NULL, NULL)); 2374168404Spjd nvlist_free(nvroot); 2375168404Spjd 2376168404Spjd /* 2377168404Spjd * Attempt to create an existing pool. It shouldn't matter 2378168404Spjd * what's in the nvroot; we should fail with EEXIST. 2379168404Spjd */ 2380236143Smm (void) rw_rdlock(&ztest_name_lock); 2381243505Smm nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1); 2382248571Smm VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL)); 2383168404Spjd nvlist_free(nvroot); 2384236143Smm VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG)); 2385236143Smm VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool)); 2386219089Spjd spa_close(spa, FTAG); 2387168404Spjd 2388236143Smm (void) rw_unlock(&ztest_name_lock); 2389168404Spjd} 2390168404Spjd 2391243505Smm/* ARGSUSED */ 2392243505Smmvoid 2393243505Smmztest_spa_upgrade(ztest_ds_t *zd, uint64_t id) 2394243505Smm{ 2395243505Smm spa_t *spa; 2396243505Smm uint64_t initial_version = SPA_VERSION_INITIAL; 2397243505Smm uint64_t version, newversion; 2398243505Smm nvlist_t *nvroot, *props; 2399243505Smm char *name; 2400243505Smm 2401243505Smm VERIFY0(mutex_lock(&ztest_vdev_lock)); 2402243505Smm name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool); 2403243505Smm 2404243505Smm /* 2405243505Smm * Clean up from previous runs. 2406243505Smm */ 2407243505Smm (void) spa_destroy(name); 2408243505Smm 2409243505Smm nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0, 2410243505Smm 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1); 2411243505Smm 2412243505Smm /* 2413243505Smm * If we're configuring a RAIDZ device then make sure that the 2414243505Smm * the initial version is capable of supporting that feature. 2415243505Smm */ 2416243505Smm switch (ztest_opts.zo_raidz_parity) { 2417243505Smm case 0: 2418243505Smm case 1: 2419243505Smm initial_version = SPA_VERSION_INITIAL; 2420243505Smm break; 2421243505Smm case 2: 2422243505Smm initial_version = SPA_VERSION_RAIDZ2; 2423243505Smm break; 2424243505Smm case 3: 2425243505Smm initial_version = SPA_VERSION_RAIDZ3; 2426243505Smm break; 2427243505Smm } 2428243505Smm 2429243505Smm /* 2430243505Smm * Create a pool with a spa version that can be upgraded. Pick 2431243505Smm * a value between initial_version and SPA_VERSION_BEFORE_FEATURES. 2432243505Smm */ 2433243505Smm do { 2434243505Smm version = ztest_random_spa_version(initial_version); 2435243505Smm } while (version > SPA_VERSION_BEFORE_FEATURES); 2436243505Smm 2437243505Smm props = fnvlist_alloc(); 2438243505Smm fnvlist_add_uint64(props, 2439243505Smm zpool_prop_to_name(ZPOOL_PROP_VERSION), version); 2440248571Smm VERIFY0(spa_create(name, nvroot, props, NULL)); 2441243505Smm fnvlist_free(nvroot); 2442243505Smm fnvlist_free(props); 2443243505Smm 2444243505Smm VERIFY0(spa_open(name, &spa, FTAG)); 2445243505Smm VERIFY3U(spa_version(spa), ==, version); 2446243505Smm newversion = ztest_random_spa_version(version + 1); 2447243505Smm 2448243505Smm if (ztest_opts.zo_verbose >= 4) { 2449243505Smm (void) printf("upgrading spa version from %llu to %llu\n", 2450243505Smm (u_longlong_t)version, (u_longlong_t)newversion); 2451243505Smm } 2452243505Smm 2453243505Smm spa_upgrade(spa, newversion); 2454243505Smm VERIFY3U(spa_version(spa), >, version); 2455243505Smm VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config, 2456243505Smm zpool_prop_to_name(ZPOOL_PROP_VERSION))); 2457243505Smm spa_close(spa, FTAG); 2458243505Smm 2459243505Smm strfree(name); 2460243505Smm VERIFY0(mutex_unlock(&ztest_vdev_lock)); 2461243505Smm} 2462243505Smm 2463185029Spjdstatic vdev_t * 2464185029Spjdvdev_lookup_by_path(vdev_t *vd, const char *path) 2465185029Spjd{ 2466185029Spjd vdev_t *mvd; 2467185029Spjd 2468185029Spjd if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0) 2469185029Spjd return (vd); 2470185029Spjd 2471185029Spjd for (int c = 0; c < vd->vdev_children; c++) 2472185029Spjd if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != 2473185029Spjd NULL) 2474185029Spjd return (mvd); 2475185029Spjd 2476185029Spjd return (NULL); 2477185029Spjd} 2478185029Spjd 2479168404Spjd/* 2480219089Spjd * Find the first available hole which can be used as a top-level. 2481219089Spjd */ 2482219089Spjdint 2483219089Spjdfind_vdev_hole(spa_t *spa) 2484219089Spjd{ 2485219089Spjd vdev_t *rvd = spa->spa_root_vdev; 2486219089Spjd int c; 2487219089Spjd 2488219089Spjd ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV); 2489219089Spjd 2490219089Spjd for (c = 0; c < rvd->vdev_children; c++) { 2491219089Spjd vdev_t *cvd = rvd->vdev_child[c]; 2492219089Spjd 2493219089Spjd if (cvd->vdev_ishole) 2494219089Spjd break; 2495219089Spjd } 2496219089Spjd return (c); 2497219089Spjd} 2498219089Spjd 2499219089Spjd/* 2500168404Spjd * Verify that vdev_add() works as expected. 2501168404Spjd */ 2502219089Spjd/* ARGSUSED */ 2503168404Spjdvoid 2504219089Spjdztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id) 2505168404Spjd{ 2506219089Spjd ztest_shared_t *zs = ztest_shared; 2507236143Smm spa_t *spa = ztest_spa; 2508219089Spjd uint64_t leaves; 2509219089Spjd uint64_t guid; 2510168404Spjd nvlist_t *nvroot; 2511168404Spjd int error; 2512168404Spjd 2513236143Smm VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2514248571Smm leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz; 2515168404Spjd 2516185029Spjd spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2517168404Spjd 2518219089Spjd ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves; 2519168404Spjd 2520185029Spjd /* 2521219089Spjd * If we have slogs then remove them 1/4 of the time. 2522185029Spjd */ 2523219089Spjd if (spa_has_slogs(spa) && ztest_random(4) == 0) { 2524219089Spjd /* 2525219089Spjd * Grab the guid from the head of the log class rotor. 2526219089Spjd */ 2527219089Spjd guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid; 2528185029Spjd 2529219089Spjd spa_config_exit(spa, SCL_VDEV, FTAG); 2530168404Spjd 2531219089Spjd /* 2532219089Spjd * We have to grab the zs_name_lock as writer to 2533219089Spjd * prevent a race between removing a slog (dmu_objset_find) 2534219089Spjd * and destroying a dataset. Removing the slog will 2535219089Spjd * grab a reference on the dataset which may cause 2536219089Spjd * dmu_objset_destroy() to fail with EBUSY thus 2537219089Spjd * leaving the dataset in an inconsistent state. 2538219089Spjd */ 2539236143Smm VERIFY(rw_wrlock(&ztest_name_lock) == 0); 2540219089Spjd error = spa_vdev_remove(spa, guid, B_FALSE); 2541236143Smm VERIFY(rw_unlock(&ztest_name_lock) == 0); 2542168404Spjd 2543219089Spjd if (error && error != EEXIST) 2544219089Spjd fatal(0, "spa_vdev_remove() = %d", error); 2545219089Spjd } else { 2546219089Spjd spa_config_exit(spa, SCL_VDEV, FTAG); 2547219089Spjd 2548219089Spjd /* 2549219089Spjd * Make 1/4 of the devices be log devices. 2550219089Spjd */ 2551243505Smm nvroot = make_vdev_root(NULL, NULL, NULL, 2552236143Smm ztest_opts.zo_vdev_size, 0, 2553236143Smm ztest_random(4) == 0, ztest_opts.zo_raidz, 2554236143Smm zs->zs_mirrors, 1); 2555219089Spjd 2556219089Spjd error = spa_vdev_add(spa, nvroot); 2557219089Spjd nvlist_free(nvroot); 2558219089Spjd 2559219089Spjd if (error == ENOSPC) 2560219089Spjd ztest_record_enospc("spa_vdev_add"); 2561219089Spjd else if (error != 0) 2562219089Spjd fatal(0, "spa_vdev_add() = %d", error); 2563219089Spjd } 2564219089Spjd 2565236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2566168404Spjd} 2567168404Spjd 2568185029Spjd/* 2569185029Spjd * Verify that adding/removing aux devices (l2arc, hot spare) works as expected. 2570185029Spjd */ 2571219089Spjd/* ARGSUSED */ 2572185029Spjdvoid 2573219089Spjdztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id) 2574168404Spjd{ 2575219089Spjd ztest_shared_t *zs = ztest_shared; 2576236143Smm spa_t *spa = ztest_spa; 2577185029Spjd vdev_t *rvd = spa->spa_root_vdev; 2578185029Spjd spa_aux_vdev_t *sav; 2579185029Spjd char *aux; 2580185029Spjd uint64_t guid = 0; 2581185029Spjd int error; 2582168404Spjd 2583185029Spjd if (ztest_random(2) == 0) { 2584185029Spjd sav = &spa->spa_spares; 2585185029Spjd aux = ZPOOL_CONFIG_SPARES; 2586185029Spjd } else { 2587185029Spjd sav = &spa->spa_l2cache; 2588185029Spjd aux = ZPOOL_CONFIG_L2CACHE; 2589185029Spjd } 2590185029Spjd 2591236143Smm VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2592185029Spjd 2593185029Spjd spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2594185029Spjd 2595185029Spjd if (sav->sav_count != 0 && ztest_random(4) == 0) { 2596185029Spjd /* 2597185029Spjd * Pick a random device to remove. 2598185029Spjd */ 2599185029Spjd guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid; 2600185029Spjd } else { 2601185029Spjd /* 2602185029Spjd * Find an unused device we can add. 2603185029Spjd */ 2604219089Spjd zs->zs_vdev_aux = 0; 2605185029Spjd for (;;) { 2606185029Spjd char path[MAXPATHLEN]; 2607185029Spjd int c; 2608236143Smm (void) snprintf(path, sizeof (path), ztest_aux_template, 2609236143Smm ztest_opts.zo_dir, ztest_opts.zo_pool, aux, 2610236143Smm zs->zs_vdev_aux); 2611185029Spjd for (c = 0; c < sav->sav_count; c++) 2612185029Spjd if (strcmp(sav->sav_vdevs[c]->vdev_path, 2613185029Spjd path) == 0) 2614185029Spjd break; 2615185029Spjd if (c == sav->sav_count && 2616185029Spjd vdev_lookup_by_path(rvd, path) == NULL) 2617185029Spjd break; 2618219089Spjd zs->zs_vdev_aux++; 2619168404Spjd } 2620168404Spjd } 2621168404Spjd 2622185029Spjd spa_config_exit(spa, SCL_VDEV, FTAG); 2623168404Spjd 2624185029Spjd if (guid == 0) { 2625185029Spjd /* 2626185029Spjd * Add a new device. 2627185029Spjd */ 2628243505Smm nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL, 2629236143Smm (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1); 2630185029Spjd error = spa_vdev_add(spa, nvroot); 2631185029Spjd if (error != 0) 2632185029Spjd fatal(0, "spa_vdev_add(%p) = %d", nvroot, error); 2633185029Spjd nvlist_free(nvroot); 2634185029Spjd } else { 2635185029Spjd /* 2636185029Spjd * Remove an existing device. Sometimes, dirty its 2637185029Spjd * vdev state first to make sure we handle removal 2638185029Spjd * of devices that have pending state changes. 2639185029Spjd */ 2640185029Spjd if (ztest_random(2) == 0) 2641219089Spjd (void) vdev_online(spa, guid, 0, NULL); 2642185029Spjd 2643185029Spjd error = spa_vdev_remove(spa, guid, B_FALSE); 2644185029Spjd if (error != 0 && error != EBUSY) 2645185029Spjd fatal(0, "spa_vdev_remove(%llu) = %d", guid, error); 2646185029Spjd } 2647185029Spjd 2648236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2649168404Spjd} 2650168404Spjd 2651168404Spjd/* 2652219089Spjd * split a pool if it has mirror tlvdevs 2653219089Spjd */ 2654219089Spjd/* ARGSUSED */ 2655219089Spjdvoid 2656219089Spjdztest_split_pool(ztest_ds_t *zd, uint64_t id) 2657219089Spjd{ 2658219089Spjd ztest_shared_t *zs = ztest_shared; 2659236143Smm spa_t *spa = ztest_spa; 2660219089Spjd vdev_t *rvd = spa->spa_root_vdev; 2661219089Spjd nvlist_t *tree, **child, *config, *split, **schild; 2662219089Spjd uint_t c, children, schildren = 0, lastlogid = 0; 2663219089Spjd int error = 0; 2664219089Spjd 2665236143Smm VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2666219089Spjd 2667219089Spjd /* ensure we have a useable config; mirrors of raidz aren't supported */ 2668236143Smm if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) { 2669236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2670219089Spjd return; 2671219089Spjd } 2672219089Spjd 2673219089Spjd /* clean up the old pool, if any */ 2674219089Spjd (void) spa_destroy("splitp"); 2675219089Spjd 2676219089Spjd spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2677219089Spjd 2678219089Spjd /* generate a config from the existing config */ 2679219089Spjd mutex_enter(&spa->spa_props_lock); 2680219089Spjd VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE, 2681219089Spjd &tree) == 0); 2682219089Spjd mutex_exit(&spa->spa_props_lock); 2683219089Spjd 2684219089Spjd VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child, 2685219089Spjd &children) == 0); 2686219089Spjd 2687219089Spjd schild = malloc(rvd->vdev_children * sizeof (nvlist_t *)); 2688219089Spjd for (c = 0; c < children; c++) { 2689219089Spjd vdev_t *tvd = rvd->vdev_child[c]; 2690219089Spjd nvlist_t **mchild; 2691219089Spjd uint_t mchildren; 2692219089Spjd 2693219089Spjd if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) { 2694219089Spjd VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME, 2695219089Spjd 0) == 0); 2696219089Spjd VERIFY(nvlist_add_string(schild[schildren], 2697219089Spjd ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); 2698219089Spjd VERIFY(nvlist_add_uint64(schild[schildren], 2699219089Spjd ZPOOL_CONFIG_IS_HOLE, 1) == 0); 2700219089Spjd if (lastlogid == 0) 2701219089Spjd lastlogid = schildren; 2702219089Spjd ++schildren; 2703219089Spjd continue; 2704219089Spjd } 2705219089Spjd lastlogid = 0; 2706219089Spjd VERIFY(nvlist_lookup_nvlist_array(child[c], 2707219089Spjd ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0); 2708219089Spjd VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0); 2709219089Spjd } 2710219089Spjd 2711219089Spjd /* OK, create a config that can be used to split */ 2712219089Spjd VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0); 2713219089Spjd VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE, 2714219089Spjd VDEV_TYPE_ROOT) == 0); 2715219089Spjd VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild, 2716219089Spjd lastlogid != 0 ? lastlogid : schildren) == 0); 2717219089Spjd 2718219089Spjd VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0); 2719219089Spjd VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0); 2720219089Spjd 2721219089Spjd for (c = 0; c < schildren; c++) 2722219089Spjd nvlist_free(schild[c]); 2723219089Spjd free(schild); 2724219089Spjd nvlist_free(split); 2725219089Spjd 2726219089Spjd spa_config_exit(spa, SCL_VDEV, FTAG); 2727219089Spjd 2728236143Smm (void) rw_wrlock(&ztest_name_lock); 2729219089Spjd error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE); 2730236143Smm (void) rw_unlock(&ztest_name_lock); 2731219089Spjd 2732219089Spjd nvlist_free(config); 2733219089Spjd 2734219089Spjd if (error == 0) { 2735219089Spjd (void) printf("successful split - results:\n"); 2736219089Spjd mutex_enter(&spa_namespace_lock); 2737219089Spjd show_pool_stats(spa); 2738219089Spjd show_pool_stats(spa_lookup("splitp")); 2739219089Spjd mutex_exit(&spa_namespace_lock); 2740219089Spjd ++zs->zs_splits; 2741219089Spjd --zs->zs_mirrors; 2742219089Spjd } 2743236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2744219089Spjd 2745219089Spjd} 2746219089Spjd 2747219089Spjd/* 2748168404Spjd * Verify that we can attach and detach devices. 2749168404Spjd */ 2750219089Spjd/* ARGSUSED */ 2751168404Spjdvoid 2752219089Spjdztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id) 2753168404Spjd{ 2754219089Spjd ztest_shared_t *zs = ztest_shared; 2755236143Smm spa_t *spa = ztest_spa; 2756185029Spjd spa_aux_vdev_t *sav = &spa->spa_spares; 2757168404Spjd vdev_t *rvd = spa->spa_root_vdev; 2758168404Spjd vdev_t *oldvd, *newvd, *pvd; 2759185029Spjd nvlist_t *root; 2760219089Spjd uint64_t leaves; 2761168404Spjd uint64_t leaf, top; 2762168404Spjd uint64_t ashift = ztest_get_ashift(); 2763209962Smm uint64_t oldguid, pguid; 2764254112Sdelphij uint64_t oldsize, newsize; 2765168404Spjd char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; 2766168404Spjd int replacing; 2767185029Spjd int oldvd_has_siblings = B_FALSE; 2768185029Spjd int newvd_is_spare = B_FALSE; 2769185029Spjd int oldvd_is_log; 2770168404Spjd int error, expected_error; 2771168404Spjd 2772236143Smm VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 2773236143Smm leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; 2774168404Spjd 2775185029Spjd spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2776168404Spjd 2777168404Spjd /* 2778168404Spjd * Decide whether to do an attach or a replace. 2779168404Spjd */ 2780168404Spjd replacing = ztest_random(2); 2781168404Spjd 2782168404Spjd /* 2783168404Spjd * Pick a random top-level vdev. 2784168404Spjd */ 2785219089Spjd top = ztest_random_vdev_top(spa, B_TRUE); 2786168404Spjd 2787168404Spjd /* 2788168404Spjd * Pick a random leaf within it. 2789168404Spjd */ 2790168404Spjd leaf = ztest_random(leaves); 2791168404Spjd 2792168404Spjd /* 2793185029Spjd * Locate this vdev. 2794168404Spjd */ 2795185029Spjd oldvd = rvd->vdev_child[top]; 2796219089Spjd if (zs->zs_mirrors >= 1) { 2797209962Smm ASSERT(oldvd->vdev_ops == &vdev_mirror_ops); 2798219089Spjd ASSERT(oldvd->vdev_children >= zs->zs_mirrors); 2799236143Smm oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz]; 2800209962Smm } 2801236143Smm if (ztest_opts.zo_raidz > 1) { 2802209962Smm ASSERT(oldvd->vdev_ops == &vdev_raidz_ops); 2803236143Smm ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz); 2804236143Smm oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz]; 2805209962Smm } 2806168404Spjd 2807168404Spjd /* 2808185029Spjd * If we're already doing an attach or replace, oldvd may be a 2809185029Spjd * mirror vdev -- in which case, pick a random child. 2810168404Spjd */ 2811185029Spjd while (oldvd->vdev_children != 0) { 2812185029Spjd oldvd_has_siblings = B_TRUE; 2813209962Smm ASSERT(oldvd->vdev_children >= 2); 2814209962Smm oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)]; 2815185029Spjd } 2816168404Spjd 2817185029Spjd oldguid = oldvd->vdev_guid; 2818219089Spjd oldsize = vdev_get_min_asize(oldvd); 2819185029Spjd oldvd_is_log = oldvd->vdev_top->vdev_islog; 2820185029Spjd (void) strcpy(oldpath, oldvd->vdev_path); 2821185029Spjd pvd = oldvd->vdev_parent; 2822209962Smm pguid = pvd->vdev_guid; 2823185029Spjd 2824168404Spjd /* 2825185029Spjd * If oldvd has siblings, then half of the time, detach it. 2826168404Spjd */ 2827185029Spjd if (oldvd_has_siblings && ztest_random(2) == 0) { 2828185029Spjd spa_config_exit(spa, SCL_VDEV, FTAG); 2829209962Smm error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE); 2830209962Smm if (error != 0 && error != ENODEV && error != EBUSY && 2831209962Smm error != ENOTSUP) 2832209962Smm fatal(0, "detach (%s) returned %d", oldpath, error); 2833236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2834185029Spjd return; 2835185029Spjd } 2836168404Spjd 2837168404Spjd /* 2838185029Spjd * For the new vdev, choose with equal probability between the two 2839185029Spjd * standard paths (ending in either 'a' or 'b') or a random hot spare. 2840168404Spjd */ 2841185029Spjd if (sav->sav_count != 0 && ztest_random(3) == 0) { 2842185029Spjd newvd = sav->sav_vdevs[ztest_random(sav->sav_count)]; 2843185029Spjd newvd_is_spare = B_TRUE; 2844185029Spjd (void) strcpy(newpath, newvd->vdev_path); 2845185029Spjd } else { 2846185029Spjd (void) snprintf(newpath, sizeof (newpath), ztest_dev_template, 2847236143Smm ztest_opts.zo_dir, ztest_opts.zo_pool, 2848236143Smm top * leaves + leaf); 2849185029Spjd if (ztest_random(2) == 0) 2850185029Spjd newpath[strlen(newpath) - 1] = 'b'; 2851185029Spjd newvd = vdev_lookup_by_path(rvd, newpath); 2852185029Spjd } 2853168404Spjd 2854185029Spjd if (newvd) { 2855219089Spjd newsize = vdev_get_min_asize(newvd); 2856185029Spjd } else { 2857185029Spjd /* 2858185029Spjd * Make newsize a little bigger or smaller than oldsize. 2859185029Spjd * If it's smaller, the attach should fail. 2860185029Spjd * If it's larger, and we're doing a replace, 2861185029Spjd * we should get dynamic LUN growth when we're done. 2862185029Spjd */ 2863185029Spjd newsize = 10 * oldsize / (9 + ztest_random(3)); 2864185029Spjd } 2865185029Spjd 2866168404Spjd /* 2867168404Spjd * If pvd is not a mirror or root, the attach should fail with ENOTSUP, 2868168404Spjd * unless it's a replace; in that case any non-replacing parent is OK. 2869168404Spjd * 2870168404Spjd * If newvd is already part of the pool, it should fail with EBUSY. 2871168404Spjd * 2872168404Spjd * If newvd is too small, it should fail with EOVERFLOW. 2873168404Spjd */ 2874185029Spjd if (pvd->vdev_ops != &vdev_mirror_ops && 2875185029Spjd pvd->vdev_ops != &vdev_root_ops && (!replacing || 2876185029Spjd pvd->vdev_ops == &vdev_replacing_ops || 2877185029Spjd pvd->vdev_ops == &vdev_spare_ops)) 2878185029Spjd expected_error = ENOTSUP; 2879185029Spjd else if (newvd_is_spare && (!replacing || oldvd_is_log)) 2880185029Spjd expected_error = ENOTSUP; 2881185029Spjd else if (newvd == oldvd) 2882185029Spjd expected_error = replacing ? 0 : EBUSY; 2883185029Spjd else if (vdev_lookup_by_path(rvd, newpath) != NULL) 2884168404Spjd expected_error = EBUSY; 2885168404Spjd else if (newsize < oldsize) 2886168404Spjd expected_error = EOVERFLOW; 2887168404Spjd else if (ashift > oldvd->vdev_top->vdev_ashift) 2888168404Spjd expected_error = EDOM; 2889168404Spjd else 2890168404Spjd expected_error = 0; 2891168404Spjd 2892185029Spjd spa_config_exit(spa, SCL_VDEV, FTAG); 2893168404Spjd 2894168404Spjd /* 2895168404Spjd * Build the nvlist describing newpath. 2896168404Spjd */ 2897243505Smm root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0, 2898185029Spjd ashift, 0, 0, 0, 1); 2899168404Spjd 2900185029Spjd error = spa_vdev_attach(spa, oldguid, root, replacing); 2901168404Spjd 2902168404Spjd nvlist_free(root); 2903168404Spjd 2904168404Spjd /* 2905168404Spjd * If our parent was the replacing vdev, but the replace completed, 2906168404Spjd * then instead of failing with ENOTSUP we may either succeed, 2907168404Spjd * fail with ENODEV, or fail with EOVERFLOW. 2908168404Spjd */ 2909168404Spjd if (expected_error == ENOTSUP && 2910168404Spjd (error == 0 || error == ENODEV || error == EOVERFLOW)) 2911168404Spjd expected_error = error; 2912168404Spjd 2913168404Spjd /* 2914168404Spjd * If someone grew the LUN, the replacement may be too small. 2915168404Spjd */ 2916185029Spjd if (error == EOVERFLOW || error == EBUSY) 2917168404Spjd expected_error = error; 2918168404Spjd 2919185029Spjd /* XXX workaround 6690467 */ 2920185029Spjd if (error != expected_error && expected_error != EBUSY) { 2921185029Spjd fatal(0, "attach (%s %llu, %s %llu, %d) " 2922185029Spjd "returned %d, expected %d", 2923254112Sdelphij oldpath, oldsize, newpath, 2924254112Sdelphij newsize, replacing, error, expected_error); 2925168404Spjd } 2926168404Spjd 2927236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 2928168404Spjd} 2929168404Spjd 2930168404Spjd/* 2931219089Spjd * Callback function which expands the physical size of the vdev. 2932168404Spjd */ 2933219089Spjdvdev_t * 2934219089Spjdgrow_vdev(vdev_t *vd, void *arg) 2935168404Spjd{ 2936219089Spjd spa_t *spa = vd->vdev_spa; 2937219089Spjd size_t *newsize = arg; 2938168404Spjd size_t fsize; 2939168404Spjd int fd; 2940168404Spjd 2941219089Spjd ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); 2942219089Spjd ASSERT(vd->vdev_ops->vdev_op_leaf); 2943168404Spjd 2944219089Spjd if ((fd = open(vd->vdev_path, O_RDWR)) == -1) 2945219089Spjd return (vd); 2946219089Spjd 2947219089Spjd fsize = lseek(fd, 0, SEEK_END); 2948219089Spjd (void) ftruncate(fd, *newsize); 2949219089Spjd 2950236143Smm if (ztest_opts.zo_verbose >= 6) { 2951219089Spjd (void) printf("%s grew from %lu to %lu bytes\n", 2952219089Spjd vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize); 2953219089Spjd } 2954219089Spjd (void) close(fd); 2955219089Spjd return (NULL); 2956219089Spjd} 2957219089Spjd 2958219089Spjd/* 2959219089Spjd * Callback function which expands a given vdev by calling vdev_online(). 2960219089Spjd */ 2961219089Spjd/* ARGSUSED */ 2962219089Spjdvdev_t * 2963219089Spjdonline_vdev(vdev_t *vd, void *arg) 2964219089Spjd{ 2965219089Spjd spa_t *spa = vd->vdev_spa; 2966219089Spjd vdev_t *tvd = vd->vdev_top; 2967219089Spjd uint64_t guid = vd->vdev_guid; 2968219089Spjd uint64_t generation = spa->spa_config_generation + 1; 2969219089Spjd vdev_state_t newstate = VDEV_STATE_UNKNOWN; 2970219089Spjd int error; 2971219089Spjd 2972219089Spjd ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE); 2973219089Spjd ASSERT(vd->vdev_ops->vdev_op_leaf); 2974219089Spjd 2975219089Spjd /* Calling vdev_online will initialize the new metaslabs */ 2976219089Spjd spa_config_exit(spa, SCL_STATE, spa); 2977219089Spjd error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate); 2978219089Spjd spa_config_enter(spa, SCL_STATE, spa, RW_READER); 2979219089Spjd 2980168404Spjd /* 2981219089Spjd * If vdev_online returned an error or the underlying vdev_open 2982219089Spjd * failed then we abort the expand. The only way to know that 2983219089Spjd * vdev_open fails is by checking the returned newstate. 2984168404Spjd */ 2985219089Spjd if (error || newstate != VDEV_STATE_HEALTHY) { 2986236143Smm if (ztest_opts.zo_verbose >= 5) { 2987219089Spjd (void) printf("Unable to expand vdev, state %llu, " 2988219089Spjd "error %d\n", (u_longlong_t)newstate, error); 2989219089Spjd } 2990219089Spjd return (vd); 2991219089Spjd } 2992219089Spjd ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY); 2993168404Spjd 2994219089Spjd /* 2995219089Spjd * Since we dropped the lock we need to ensure that we're 2996219089Spjd * still talking to the original vdev. It's possible this 2997219089Spjd * vdev may have been detached/replaced while we were 2998219089Spjd * trying to online it. 2999219089Spjd */ 3000219089Spjd if (generation != spa->spa_config_generation) { 3001236143Smm if (ztest_opts.zo_verbose >= 5) { 3002219089Spjd (void) printf("vdev configuration has changed, " 3003219089Spjd "guid %llu, state %llu, expected gen %llu, " 3004219089Spjd "got gen %llu\n", 3005219089Spjd (u_longlong_t)guid, 3006219089Spjd (u_longlong_t)tvd->vdev_state, 3007219089Spjd (u_longlong_t)generation, 3008219089Spjd (u_longlong_t)spa->spa_config_generation); 3009219089Spjd } 3010219089Spjd return (vd); 3011219089Spjd } 3012219089Spjd return (NULL); 3013219089Spjd} 3014168404Spjd 3015219089Spjd/* 3016219089Spjd * Traverse the vdev tree calling the supplied function. 3017219089Spjd * We continue to walk the tree until we either have walked all 3018219089Spjd * children or we receive a non-NULL return from the callback. 3019219089Spjd * If a NULL callback is passed, then we just return back the first 3020219089Spjd * leaf vdev we encounter. 3021219089Spjd */ 3022219089Spjdvdev_t * 3023219089Spjdvdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg) 3024219089Spjd{ 3025219089Spjd if (vd->vdev_ops->vdev_op_leaf) { 3026219089Spjd if (func == NULL) 3027219089Spjd return (vd); 3028219089Spjd else 3029219089Spjd return (func(vd, arg)); 3030219089Spjd } 3031168404Spjd 3032219089Spjd for (uint_t c = 0; c < vd->vdev_children; c++) { 3033219089Spjd vdev_t *cvd = vd->vdev_child[c]; 3034219089Spjd if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL) 3035219089Spjd return (cvd); 3036219089Spjd } 3037219089Spjd return (NULL); 3038219089Spjd} 3039219089Spjd 3040219089Spjd/* 3041219089Spjd * Verify that dynamic LUN growth works as expected. 3042219089Spjd */ 3043219089Spjd/* ARGSUSED */ 3044219089Spjdvoid 3045219089Spjdztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id) 3046219089Spjd{ 3047236143Smm spa_t *spa = ztest_spa; 3048219089Spjd vdev_t *vd, *tvd; 3049219089Spjd metaslab_class_t *mc; 3050219089Spjd metaslab_group_t *mg; 3051219089Spjd size_t psize, newsize; 3052219089Spjd uint64_t top; 3053219089Spjd uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count; 3054219089Spjd 3055236143Smm VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 3056219089Spjd spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3057219089Spjd 3058219089Spjd top = ztest_random_vdev_top(spa, B_TRUE); 3059219089Spjd 3060219089Spjd tvd = spa->spa_root_vdev->vdev_child[top]; 3061219089Spjd mg = tvd->vdev_mg; 3062219089Spjd mc = mg->mg_class; 3063219089Spjd old_ms_count = tvd->vdev_ms_count; 3064219089Spjd old_class_space = metaslab_class_get_space(mc); 3065219089Spjd 3066219089Spjd /* 3067219089Spjd * Determine the size of the first leaf vdev associated with 3068219089Spjd * our top-level device. 3069219089Spjd */ 3070219089Spjd vd = vdev_walk_tree(tvd, NULL, NULL); 3071219089Spjd ASSERT3P(vd, !=, NULL); 3072219089Spjd ASSERT(vd->vdev_ops->vdev_op_leaf); 3073219089Spjd 3074219089Spjd psize = vd->vdev_psize; 3075219089Spjd 3076219089Spjd /* 3077219089Spjd * We only try to expand the vdev if it's healthy, less than 4x its 3078219089Spjd * original size, and it has a valid psize. 3079219089Spjd */ 3080219089Spjd if (tvd->vdev_state != VDEV_STATE_HEALTHY || 3081236143Smm psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) { 3082219089Spjd spa_config_exit(spa, SCL_STATE, spa); 3083236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3084219089Spjd return; 3085219089Spjd } 3086219089Spjd ASSERT(psize > 0); 3087219089Spjd newsize = psize + psize / 8; 3088219089Spjd ASSERT3U(newsize, >, psize); 3089219089Spjd 3090236143Smm if (ztest_opts.zo_verbose >= 6) { 3091219089Spjd (void) printf("Expanding LUN %s from %lu to %lu\n", 3092219089Spjd vd->vdev_path, (ulong_t)psize, (ulong_t)newsize); 3093219089Spjd } 3094219089Spjd 3095219089Spjd /* 3096219089Spjd * Growing the vdev is a two step process: 3097219089Spjd * 1). expand the physical size (i.e. relabel) 3098219089Spjd * 2). online the vdev to create the new metaslabs 3099219089Spjd */ 3100219089Spjd if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL || 3101219089Spjd vdev_walk_tree(tvd, online_vdev, NULL) != NULL || 3102219089Spjd tvd->vdev_state != VDEV_STATE_HEALTHY) { 3103236143Smm if (ztest_opts.zo_verbose >= 5) { 3104219089Spjd (void) printf("Could not expand LUN because " 3105219089Spjd "the vdev configuration changed.\n"); 3106168404Spjd } 3107219089Spjd spa_config_exit(spa, SCL_STATE, spa); 3108236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3109219089Spjd return; 3110168404Spjd } 3111168404Spjd 3112219089Spjd spa_config_exit(spa, SCL_STATE, spa); 3113219089Spjd 3114219089Spjd /* 3115219089Spjd * Expanding the LUN will update the config asynchronously, 3116219089Spjd * thus we must wait for the async thread to complete any 3117219089Spjd * pending tasks before proceeding. 3118219089Spjd */ 3119219089Spjd for (;;) { 3120219089Spjd boolean_t done; 3121219089Spjd mutex_enter(&spa->spa_async_lock); 3122219089Spjd done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks); 3123219089Spjd mutex_exit(&spa->spa_async_lock); 3124219089Spjd if (done) 3125219089Spjd break; 3126219089Spjd txg_wait_synced(spa_get_dsl(spa), 0); 3127219089Spjd (void) poll(NULL, 0, 100); 3128219089Spjd } 3129219089Spjd 3130219089Spjd spa_config_enter(spa, SCL_STATE, spa, RW_READER); 3131219089Spjd 3132219089Spjd tvd = spa->spa_root_vdev->vdev_child[top]; 3133219089Spjd new_ms_count = tvd->vdev_ms_count; 3134219089Spjd new_class_space = metaslab_class_get_space(mc); 3135219089Spjd 3136219089Spjd if (tvd->vdev_mg != mg || mg->mg_class != mc) { 3137236143Smm if (ztest_opts.zo_verbose >= 5) { 3138219089Spjd (void) printf("Could not verify LUN expansion due to " 3139219089Spjd "intervening vdev offline or remove.\n"); 3140219089Spjd } 3141219089Spjd spa_config_exit(spa, SCL_STATE, spa); 3142236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3143219089Spjd return; 3144219089Spjd } 3145219089Spjd 3146219089Spjd /* 3147219089Spjd * Make sure we were able to grow the vdev. 3148219089Spjd */ 3149219089Spjd if (new_ms_count <= old_ms_count) 3150219089Spjd fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n", 3151219089Spjd old_ms_count, new_ms_count); 3152219089Spjd 3153219089Spjd /* 3154219089Spjd * Make sure we were able to grow the pool. 3155219089Spjd */ 3156219089Spjd if (new_class_space <= old_class_space) 3157219089Spjd fatal(0, "LUN expansion failed: class_space %llu <= %llu\n", 3158219089Spjd old_class_space, new_class_space); 3159219089Spjd 3160236143Smm if (ztest_opts.zo_verbose >= 5) { 3161219089Spjd char oldnumbuf[6], newnumbuf[6]; 3162219089Spjd 3163219089Spjd nicenum(old_class_space, oldnumbuf); 3164219089Spjd nicenum(new_class_space, newnumbuf); 3165219089Spjd (void) printf("%s grew from %s to %s\n", 3166219089Spjd spa->spa_name, oldnumbuf, newnumbuf); 3167219089Spjd } 3168219089Spjd 3169219089Spjd spa_config_exit(spa, SCL_STATE, spa); 3170236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 3171168404Spjd} 3172168404Spjd 3173219089Spjd/* 3174219089Spjd * Verify that dmu_objset_{create,destroy,open,close} work as expected. 3175219089Spjd */ 3176168404Spjd/* ARGSUSED */ 3177168404Spjdstatic void 3178219089Spjdztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 3179168404Spjd{ 3180168404Spjd /* 3181219089Spjd * Create the objects common to all ztest datasets. 3182168404Spjd */ 3183219089Spjd VERIFY(zap_create_claim(os, ZTEST_DIROBJ, 3184168404Spjd DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 3185219089Spjd} 3186168404Spjd 3187219089Spjdstatic int 3188219089Spjdztest_dataset_create(char *dsname) 3189219089Spjd{ 3190219089Spjd uint64_t zilset = ztest_random(100); 3191219089Spjd int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, 3192219089Spjd ztest_objset_create_cb, NULL); 3193219089Spjd 3194219089Spjd if (err || zilset < 80) 3195219089Spjd return (err); 3196219089Spjd 3197236143Smm if (ztest_opts.zo_verbose >= 6) 3198236143Smm (void) printf("Setting dataset %s to sync always\n", dsname); 3199219089Spjd return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC, 3200219089Spjd ZFS_SYNC_ALWAYS, B_FALSE)); 3201168404Spjd} 3202168404Spjd 3203219089Spjd/* ARGSUSED */ 3204168404Spjdstatic int 3205219089Spjdztest_objset_destroy_cb(const char *name, void *arg) 3206168404Spjd{ 3207168404Spjd objset_t *os; 3208219089Spjd dmu_object_info_t doi; 3209168404Spjd int error; 3210168404Spjd 3211168404Spjd /* 3212168404Spjd * Verify that the dataset contains a directory object. 3213168404Spjd */ 3214248571Smm VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os)); 3215219089Spjd error = dmu_object_info(os, ZTEST_DIROBJ, &doi); 3216168404Spjd if (error != ENOENT) { 3217168404Spjd /* We could have crashed in the middle of destroying it */ 3218240415Smm ASSERT0(error); 3219219089Spjd ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER); 3220219089Spjd ASSERT3S(doi.doi_physical_blocks_512, >=, 0); 3221168404Spjd } 3222248571Smm dmu_objset_disown(os, FTAG); 3223168404Spjd 3224168404Spjd /* 3225168404Spjd * Destroy the dataset. 3226168404Spjd */ 3227248571Smm if (strchr(name, '@') != NULL) { 3228248571Smm VERIFY0(dsl_destroy_snapshot(name, B_FALSE)); 3229248571Smm } else { 3230248571Smm VERIFY0(dsl_destroy_head(name)); 3231248571Smm } 3232168404Spjd return (0); 3233168404Spjd} 3234168404Spjd 3235219089Spjdstatic boolean_t 3236219089Spjdztest_snapshot_create(char *osname, uint64_t id) 3237168404Spjd{ 3238219089Spjd char snapname[MAXNAMELEN]; 3239219089Spjd int error; 3240168404Spjd 3241248571Smm (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id); 3242168404Spjd 3243248571Smm error = dmu_objset_snapshot_one(osname, snapname); 3244219089Spjd if (error == ENOSPC) { 3245219089Spjd ztest_record_enospc(FTAG); 3246219089Spjd return (B_FALSE); 3247219089Spjd } 3248248571Smm if (error != 0 && error != EEXIST) { 3249248571Smm fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname, 3250248571Smm snapname, error); 3251248571Smm } 3252219089Spjd return (B_TRUE); 3253219089Spjd} 3254168404Spjd 3255219089Spjdstatic boolean_t 3256219089Spjdztest_snapshot_destroy(char *osname, uint64_t id) 3257219089Spjd{ 3258219089Spjd char snapname[MAXNAMELEN]; 3259219089Spjd int error; 3260219089Spjd 3261219089Spjd (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname, 3262219089Spjd (u_longlong_t)id); 3263219089Spjd 3264248571Smm error = dsl_destroy_snapshot(snapname, B_FALSE); 3265219089Spjd if (error != 0 && error != ENOENT) 3266219089Spjd fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error); 3267219089Spjd return (B_TRUE); 3268168404Spjd} 3269168404Spjd 3270219089Spjd/* ARGSUSED */ 3271168404Spjdvoid 3272219089Spjdztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id) 3273168404Spjd{ 3274219089Spjd ztest_ds_t zdtmp; 3275219089Spjd int iters; 3276168404Spjd int error; 3277185029Spjd objset_t *os, *os2; 3278219089Spjd char name[MAXNAMELEN]; 3279168404Spjd zilog_t *zilog; 3280168404Spjd 3281236143Smm (void) rw_rdlock(&ztest_name_lock); 3282168404Spjd 3283219089Spjd (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu", 3284236143Smm ztest_opts.zo_pool, (u_longlong_t)id); 3285168404Spjd 3286168404Spjd /* 3287168404Spjd * If this dataset exists from a previous run, process its replay log 3288168404Spjd * half of the time. If we don't replay it, then dmu_objset_destroy() 3289219089Spjd * (invoked from ztest_objset_destroy_cb()) should just throw it away. 3290168404Spjd */ 3291168404Spjd if (ztest_random(2) == 0 && 3292219089Spjd dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) { 3293236143Smm ztest_zd_init(&zdtmp, NULL, os); 3294219089Spjd zil_replay(os, &zdtmp, ztest_replay_vector); 3295219089Spjd ztest_zd_fini(&zdtmp); 3296219089Spjd dmu_objset_disown(os, FTAG); 3297168404Spjd } 3298168404Spjd 3299168404Spjd /* 3300168404Spjd * There may be an old instance of the dataset we're about to 3301168404Spjd * create lying around from a previous run. If so, destroy it 3302168404Spjd * and all of its snapshots. 3303168404Spjd */ 3304219089Spjd (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, 3305168404Spjd DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 3306168404Spjd 3307168404Spjd /* 3308168404Spjd * Verify that the destroyed dataset is no longer in the namespace. 3309168404Spjd */ 3310248571Smm VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, 3311248571Smm FTAG, &os)); 3312168404Spjd 3313168404Spjd /* 3314168404Spjd * Verify that we can create a new dataset. 3315168404Spjd */ 3316219089Spjd error = ztest_dataset_create(name); 3317168404Spjd if (error) { 3318168404Spjd if (error == ENOSPC) { 3319219089Spjd ztest_record_enospc(FTAG); 3320236143Smm (void) rw_unlock(&ztest_name_lock); 3321168404Spjd return; 3322168404Spjd } 3323168404Spjd fatal(0, "dmu_objset_create(%s) = %d", name, error); 3324168404Spjd } 3325168404Spjd 3326248571Smm VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os)); 3327168404Spjd 3328236143Smm ztest_zd_init(&zdtmp, NULL, os); 3329219089Spjd 3330168404Spjd /* 3331168404Spjd * Open the intent log for it. 3332168404Spjd */ 3333219089Spjd zilog = zil_open(os, ztest_get_data); 3334168404Spjd 3335168404Spjd /* 3336219089Spjd * Put some objects in there, do a little I/O to them, 3337219089Spjd * and randomly take a couple of snapshots along the way. 3338168404Spjd */ 3339219089Spjd iters = ztest_random(5); 3340219089Spjd for (int i = 0; i < iters; i++) { 3341219089Spjd ztest_dmu_object_alloc_free(&zdtmp, id); 3342219089Spjd if (ztest_random(iters) == 0) 3343219089Spjd (void) ztest_snapshot_create(name, i); 3344168404Spjd } 3345168404Spjd 3346168404Spjd /* 3347168404Spjd * Verify that we cannot create an existing dataset. 3348168404Spjd */ 3349219089Spjd VERIFY3U(EEXIST, ==, 3350219089Spjd dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL)); 3351168404Spjd 3352168404Spjd /* 3353219089Spjd * Verify that we can hold an objset that is also owned. 3354168404Spjd */ 3355219089Spjd VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2)); 3356219089Spjd dmu_objset_rele(os2, FTAG); 3357168404Spjd 3358219089Spjd /* 3359219089Spjd * Verify that we cannot own an objset that is already owned. 3360219089Spjd */ 3361219089Spjd VERIFY3U(EBUSY, ==, 3362219089Spjd dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2)); 3363219089Spjd 3364168404Spjd zil_close(zilog); 3365219089Spjd dmu_objset_disown(os, FTAG); 3366219089Spjd ztest_zd_fini(&zdtmp); 3367168404Spjd 3368236143Smm (void) rw_unlock(&ztest_name_lock); 3369168404Spjd} 3370168404Spjd 3371168404Spjd/* 3372168404Spjd * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. 3373168404Spjd */ 3374168404Spjdvoid 3375219089Spjdztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id) 3376168404Spjd{ 3377236143Smm (void) rw_rdlock(&ztest_name_lock); 3378219089Spjd (void) ztest_snapshot_destroy(zd->zd_name, id); 3379219089Spjd (void) ztest_snapshot_create(zd->zd_name, id); 3380236143Smm (void) rw_unlock(&ztest_name_lock); 3381219089Spjd} 3382219089Spjd 3383219089Spjd/* 3384219089Spjd * Cleanup non-standard snapshots and clones. 3385219089Spjd */ 3386219089Spjdvoid 3387219089Spjdztest_dsl_dataset_cleanup(char *osname, uint64_t id) 3388219089Spjd{ 3389219089Spjd char snap1name[MAXNAMELEN]; 3390219089Spjd char clone1name[MAXNAMELEN]; 3391219089Spjd char snap2name[MAXNAMELEN]; 3392219089Spjd char clone2name[MAXNAMELEN]; 3393219089Spjd char snap3name[MAXNAMELEN]; 3394168404Spjd int error; 3395168404Spjd 3396219089Spjd (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id); 3397219089Spjd (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id); 3398219089Spjd (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id); 3399219089Spjd (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id); 3400219089Spjd (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id); 3401168404Spjd 3402248571Smm error = dsl_destroy_head(clone2name); 3403219089Spjd if (error && error != ENOENT) 3404248571Smm fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error); 3405248571Smm error = dsl_destroy_snapshot(snap3name, B_FALSE); 3406219089Spjd if (error && error != ENOENT) 3407248571Smm fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error); 3408248571Smm error = dsl_destroy_snapshot(snap2name, B_FALSE); 3409219089Spjd if (error && error != ENOENT) 3410248571Smm fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error); 3411248571Smm error = dsl_destroy_head(clone1name); 3412219089Spjd if (error && error != ENOENT) 3413248571Smm fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error); 3414248571Smm error = dsl_destroy_snapshot(snap1name, B_FALSE); 3415219089Spjd if (error && error != ENOENT) 3416248571Smm fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error); 3417168404Spjd} 3418168404Spjd 3419168404Spjd/* 3420207910Smm * Verify dsl_dataset_promote handles EBUSY 3421207910Smm */ 3422207910Smmvoid 3423219089Spjdztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id) 3424207910Smm{ 3425248571Smm objset_t *os; 3426219089Spjd char snap1name[MAXNAMELEN]; 3427219089Spjd char clone1name[MAXNAMELEN]; 3428219089Spjd char snap2name[MAXNAMELEN]; 3429219089Spjd char clone2name[MAXNAMELEN]; 3430219089Spjd char snap3name[MAXNAMELEN]; 3431219089Spjd char *osname = zd->zd_name; 3432219089Spjd int error; 3433207910Smm 3434236143Smm (void) rw_rdlock(&ztest_name_lock); 3435207910Smm 3436219089Spjd ztest_dsl_dataset_cleanup(osname, id); 3437207910Smm 3438219089Spjd (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id); 3439219089Spjd (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id); 3440219089Spjd (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id); 3441219089Spjd (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id); 3442219089Spjd (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id); 3443207910Smm 3444248571Smm error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1); 3445209962Smm if (error && error != EEXIST) { 3446209962Smm if (error == ENOSPC) { 3447209962Smm ztest_record_enospc(FTAG); 3448209962Smm goto out; 3449209962Smm } 3450209962Smm fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error); 3451209962Smm } 3452207910Smm 3453248571Smm error = dmu_objset_clone(clone1name, snap1name); 3454209962Smm if (error) { 3455209962Smm if (error == ENOSPC) { 3456209962Smm ztest_record_enospc(FTAG); 3457209962Smm goto out; 3458209962Smm } 3459207910Smm fatal(0, "dmu_objset_create(%s) = %d", clone1name, error); 3460209962Smm } 3461207910Smm 3462248571Smm error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1); 3463209962Smm if (error && error != EEXIST) { 3464209962Smm if (error == ENOSPC) { 3465209962Smm ztest_record_enospc(FTAG); 3466209962Smm goto out; 3467209962Smm } 3468209962Smm fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error); 3469209962Smm } 3470207910Smm 3471248571Smm error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1); 3472209962Smm if (error && error != EEXIST) { 3473209962Smm if (error == ENOSPC) { 3474209962Smm ztest_record_enospc(FTAG); 3475209962Smm goto out; 3476209962Smm } 3477209962Smm fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error); 3478209962Smm } 3479207910Smm 3480248571Smm error = dmu_objset_clone(clone2name, snap3name); 3481209962Smm if (error) { 3482209962Smm if (error == ENOSPC) { 3483219089Spjd ztest_record_enospc(FTAG); 3484209962Smm goto out; 3485209962Smm } 3486207910Smm fatal(0, "dmu_objset_create(%s) = %d", clone2name, error); 3487209962Smm } 3488207910Smm 3489248571Smm error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os); 3490207910Smm if (error) 3491248571Smm fatal(0, "dmu_objset_own(%s) = %d", snap2name, error); 3492219089Spjd error = dsl_dataset_promote(clone2name, NULL); 3493268075Sdelphij if (error == ENOSPC) { 3494268075Sdelphij dmu_objset_disown(os, FTAG); 3495268075Sdelphij ztest_record_enospc(FTAG); 3496268075Sdelphij goto out; 3497268075Sdelphij } 3498207910Smm if (error != EBUSY) 3499207910Smm fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name, 3500207910Smm error); 3501248571Smm dmu_objset_disown(os, FTAG); 3502207910Smm 3503209962Smmout: 3504219089Spjd ztest_dsl_dataset_cleanup(osname, id); 3505207910Smm 3506236143Smm (void) rw_unlock(&ztest_name_lock); 3507207910Smm} 3508207910Smm 3509207910Smm/* 3510168404Spjd * Verify that dmu_object_{alloc,free} work as expected. 3511168404Spjd */ 3512168404Spjdvoid 3513219089Spjdztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id) 3514168404Spjd{ 3515219089Spjd ztest_od_t od[4]; 3516219089Spjd int batchsize = sizeof (od) / sizeof (od[0]); 3517168404Spjd 3518219089Spjd for (int b = 0; b < batchsize; b++) 3519219089Spjd ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0); 3520168404Spjd 3521168404Spjd /* 3522219089Spjd * Destroy the previous batch of objects, create a new batch, 3523219089Spjd * and do some I/O on the new objects. 3524168404Spjd */ 3525219089Spjd if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0) 3526219089Spjd return; 3527168404Spjd 3528219089Spjd while (ztest_random(4 * batchsize) != 0) 3529219089Spjd ztest_io(zd, od[ztest_random(batchsize)].od_object, 3530219089Spjd ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 3531168404Spjd} 3532168404Spjd 3533168404Spjd/* 3534168404Spjd * Verify that dmu_{read,write} work as expected. 3535168404Spjd */ 3536168404Spjdvoid 3537219089Spjdztest_dmu_read_write(ztest_ds_t *zd, uint64_t id) 3538168404Spjd{ 3539219089Spjd objset_t *os = zd->zd_os; 3540219089Spjd ztest_od_t od[2]; 3541168404Spjd dmu_tx_t *tx; 3542168404Spjd int i, freeit, error; 3543168404Spjd uint64_t n, s, txg; 3544168404Spjd bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; 3545219089Spjd uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; 3546219089Spjd uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t); 3547168404Spjd uint64_t regions = 997; 3548168404Spjd uint64_t stride = 123456789ULL; 3549168404Spjd uint64_t width = 40; 3550168404Spjd int free_percent = 5; 3551168404Spjd 3552168404Spjd /* 3553168404Spjd * This test uses two objects, packobj and bigobj, that are always 3554168404Spjd * updated together (i.e. in the same tx) so that their contents are 3555168404Spjd * in sync and can be compared. Their contents relate to each other 3556168404Spjd * in a simple way: packobj is a dense array of 'bufwad' structures, 3557168404Spjd * while bigobj is a sparse array of the same bufwads. Specifically, 3558168404Spjd * for any index n, there are three bufwads that should be identical: 3559168404Spjd * 3560168404Spjd * packobj, at offset n * sizeof (bufwad_t) 3561168404Spjd * bigobj, at the head of the nth chunk 3562168404Spjd * bigobj, at the tail of the nth chunk 3563168404Spjd * 3564168404Spjd * The chunk size is arbitrary. It doesn't have to be a power of two, 3565168404Spjd * and it doesn't have any relation to the object blocksize. 3566168404Spjd * The only requirement is that it can hold at least two bufwads. 3567168404Spjd * 3568168404Spjd * Normally, we write the bufwad to each of these locations. 3569168404Spjd * However, free_percent of the time we instead write zeroes to 3570168404Spjd * packobj and perform a dmu_free_range() on bigobj. By comparing 3571168404Spjd * bigobj to packobj, we can verify that the DMU is correctly 3572168404Spjd * tracking which parts of an object are allocated and free, 3573168404Spjd * and that the contents of the allocated blocks are correct. 3574168404Spjd */ 3575168404Spjd 3576168404Spjd /* 3577168404Spjd * Read the directory info. If it's the first time, set things up. 3578168404Spjd */ 3579219089Spjd ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize); 3580219089Spjd ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); 3581168404Spjd 3582219089Spjd if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 3583219089Spjd return; 3584168404Spjd 3585219089Spjd bigobj = od[0].od_object; 3586219089Spjd packobj = od[1].od_object; 3587219089Spjd chunksize = od[0].od_gen; 3588219089Spjd ASSERT(chunksize == od[1].od_gen); 3589168404Spjd 3590168404Spjd /* 3591168404Spjd * Prefetch a random chunk of the big object. 3592168404Spjd * Our aim here is to get some async reads in flight 3593168404Spjd * for blocks that we may free below; the DMU should 3594168404Spjd * handle this race correctly. 3595168404Spjd */ 3596168404Spjd n = ztest_random(regions) * stride + ztest_random(width); 3597168404Spjd s = 1 + ztest_random(2 * width - 1); 3598219089Spjd dmu_prefetch(os, bigobj, n * chunksize, s * chunksize); 3599168404Spjd 3600168404Spjd /* 3601168404Spjd * Pick a random index and compute the offsets into packobj and bigobj. 3602168404Spjd */ 3603168404Spjd n = ztest_random(regions) * stride + ztest_random(width); 3604168404Spjd s = 1 + ztest_random(width - 1); 3605168404Spjd 3606168404Spjd packoff = n * sizeof (bufwad_t); 3607168404Spjd packsize = s * sizeof (bufwad_t); 3608168404Spjd 3609219089Spjd bigoff = n * chunksize; 3610219089Spjd bigsize = s * chunksize; 3611168404Spjd 3612168404Spjd packbuf = umem_alloc(packsize, UMEM_NOFAIL); 3613168404Spjd bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); 3614168404Spjd 3615168404Spjd /* 3616168404Spjd * free_percent of the time, free a range of bigobj rather than 3617168404Spjd * overwriting it. 3618168404Spjd */ 3619168404Spjd freeit = (ztest_random(100) < free_percent); 3620168404Spjd 3621168404Spjd /* 3622168404Spjd * Read the current contents of our objects. 3623168404Spjd */ 3624219089Spjd error = dmu_read(os, packobj, packoff, packsize, packbuf, 3625209962Smm DMU_READ_PREFETCH); 3626240415Smm ASSERT0(error); 3627219089Spjd error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf, 3628209962Smm DMU_READ_PREFETCH); 3629240415Smm ASSERT0(error); 3630168404Spjd 3631168404Spjd /* 3632168404Spjd * Get a tx for the mods to both packobj and bigobj. 3633168404Spjd */ 3634168404Spjd tx = dmu_tx_create(os); 3635168404Spjd 3636219089Spjd dmu_tx_hold_write(tx, packobj, packoff, packsize); 3637168404Spjd 3638168404Spjd if (freeit) 3639219089Spjd dmu_tx_hold_free(tx, bigobj, bigoff, bigsize); 3640168404Spjd else 3641219089Spjd dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); 3642168404Spjd 3643254077Sdelphij /* This accounts for setting the checksum/compression. */ 3644254077Sdelphij dmu_tx_hold_bonus(tx, bigobj); 3645254077Sdelphij 3646219089Spjd txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 3647219089Spjd if (txg == 0) { 3648168404Spjd umem_free(packbuf, packsize); 3649168404Spjd umem_free(bigbuf, bigsize); 3650168404Spjd return; 3651168404Spjd } 3652168404Spjd 3653268075Sdelphij enum zio_checksum cksum; 3654268075Sdelphij do { 3655268075Sdelphij cksum = (enum zio_checksum) 3656268075Sdelphij ztest_random_dsl_prop(ZFS_PROP_CHECKSUM); 3657268075Sdelphij } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS); 3658268075Sdelphij dmu_object_set_checksum(os, bigobj, cksum, tx); 3659168404Spjd 3660268075Sdelphij enum zio_compress comp; 3661268075Sdelphij do { 3662268075Sdelphij comp = (enum zio_compress) 3663268075Sdelphij ztest_random_dsl_prop(ZFS_PROP_COMPRESSION); 3664268075Sdelphij } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS); 3665268075Sdelphij dmu_object_set_compress(os, bigobj, comp, tx); 3666219089Spjd 3667168404Spjd /* 3668168404Spjd * For each index from n to n + s, verify that the existing bufwad 3669168404Spjd * in packobj matches the bufwads at the head and tail of the 3670168404Spjd * corresponding chunk in bigobj. Then update all three bufwads 3671168404Spjd * with the new values we want to write out. 3672168404Spjd */ 3673168404Spjd for (i = 0; i < s; i++) { 3674168404Spjd /* LINTED */ 3675168404Spjd pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 3676168404Spjd /* LINTED */ 3677219089Spjd bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); 3678168404Spjd /* LINTED */ 3679219089Spjd bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; 3680168404Spjd 3681168404Spjd ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 3682168404Spjd ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 3683168404Spjd 3684168404Spjd if (pack->bw_txg > txg) 3685168404Spjd fatal(0, "future leak: got %llx, open txg is %llx", 3686168404Spjd pack->bw_txg, txg); 3687168404Spjd 3688168404Spjd if (pack->bw_data != 0 && pack->bw_index != n + i) 3689168404Spjd fatal(0, "wrong index: got %llx, wanted %llx+%llx", 3690168404Spjd pack->bw_index, n, i); 3691168404Spjd 3692168404Spjd if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 3693168404Spjd fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 3694168404Spjd 3695168404Spjd if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 3696168404Spjd fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 3697168404Spjd 3698168404Spjd if (freeit) { 3699168404Spjd bzero(pack, sizeof (bufwad_t)); 3700168404Spjd } else { 3701168404Spjd pack->bw_index = n + i; 3702168404Spjd pack->bw_txg = txg; 3703168404Spjd pack->bw_data = 1 + ztest_random(-2ULL); 3704168404Spjd } 3705168404Spjd *bigH = *pack; 3706168404Spjd *bigT = *pack; 3707168404Spjd } 3708168404Spjd 3709168404Spjd /* 3710168404Spjd * We've verified all the old bufwads, and made new ones. 3711168404Spjd * Now write them out. 3712168404Spjd */ 3713219089Spjd dmu_write(os, packobj, packoff, packsize, packbuf, tx); 3714168404Spjd 3715168404Spjd if (freeit) { 3716236143Smm if (ztest_opts.zo_verbose >= 7) { 3717168404Spjd (void) printf("freeing offset %llx size %llx" 3718168404Spjd " txg %llx\n", 3719168404Spjd (u_longlong_t)bigoff, 3720168404Spjd (u_longlong_t)bigsize, 3721168404Spjd (u_longlong_t)txg); 3722168404Spjd } 3723219089Spjd VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx)); 3724168404Spjd } else { 3725236143Smm if (ztest_opts.zo_verbose >= 7) { 3726168404Spjd (void) printf("writing offset %llx size %llx" 3727168404Spjd " txg %llx\n", 3728168404Spjd (u_longlong_t)bigoff, 3729168404Spjd (u_longlong_t)bigsize, 3730168404Spjd (u_longlong_t)txg); 3731168404Spjd } 3732219089Spjd dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx); 3733168404Spjd } 3734168404Spjd 3735168404Spjd dmu_tx_commit(tx); 3736168404Spjd 3737168404Spjd /* 3738168404Spjd * Sanity check the stuff we just wrote. 3739168404Spjd */ 3740168404Spjd { 3741168404Spjd void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 3742168404Spjd void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 3743168404Spjd 3744219089Spjd VERIFY(0 == dmu_read(os, packobj, packoff, 3745209962Smm packsize, packcheck, DMU_READ_PREFETCH)); 3746219089Spjd VERIFY(0 == dmu_read(os, bigobj, bigoff, 3747209962Smm bigsize, bigcheck, DMU_READ_PREFETCH)); 3748168404Spjd 3749168404Spjd ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 3750168404Spjd ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 3751168404Spjd 3752168404Spjd umem_free(packcheck, packsize); 3753168404Spjd umem_free(bigcheck, bigsize); 3754168404Spjd } 3755168404Spjd 3756168404Spjd umem_free(packbuf, packsize); 3757168404Spjd umem_free(bigbuf, bigsize); 3758168404Spjd} 3759168404Spjd 3760168404Spjdvoid 3761209962Smmcompare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf, 3762219089Spjd uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg) 3763209962Smm{ 3764209962Smm uint64_t i; 3765209962Smm bufwad_t *pack; 3766209962Smm bufwad_t *bigH; 3767209962Smm bufwad_t *bigT; 3768209962Smm 3769209962Smm /* 3770209962Smm * For each index from n to n + s, verify that the existing bufwad 3771209962Smm * in packobj matches the bufwads at the head and tail of the 3772209962Smm * corresponding chunk in bigobj. Then update all three bufwads 3773209962Smm * with the new values we want to write out. 3774209962Smm */ 3775209962Smm for (i = 0; i < s; i++) { 3776209962Smm /* LINTED */ 3777209962Smm pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 3778209962Smm /* LINTED */ 3779219089Spjd bigH = (bufwad_t *)((char *)bigbuf + i * chunksize); 3780209962Smm /* LINTED */ 3781219089Spjd bigT = (bufwad_t *)((char *)bigH + chunksize) - 1; 3782209962Smm 3783209962Smm ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 3784209962Smm ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 3785209962Smm 3786209962Smm if (pack->bw_txg > txg) 3787209962Smm fatal(0, "future leak: got %llx, open txg is %llx", 3788209962Smm pack->bw_txg, txg); 3789209962Smm 3790209962Smm if (pack->bw_data != 0 && pack->bw_index != n + i) 3791209962Smm fatal(0, "wrong index: got %llx, wanted %llx+%llx", 3792209962Smm pack->bw_index, n, i); 3793209962Smm 3794209962Smm if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 3795209962Smm fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 3796209962Smm 3797209962Smm if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 3798209962Smm fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 3799209962Smm 3800209962Smm pack->bw_index = n + i; 3801209962Smm pack->bw_txg = txg; 3802209962Smm pack->bw_data = 1 + ztest_random(-2ULL); 3803209962Smm 3804209962Smm *bigH = *pack; 3805209962Smm *bigT = *pack; 3806209962Smm } 3807209962Smm} 3808209962Smm 3809209962Smmvoid 3810219089Spjdztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id) 3811209962Smm{ 3812219089Spjd objset_t *os = zd->zd_os; 3813219089Spjd ztest_od_t od[2]; 3814209962Smm dmu_tx_t *tx; 3815209962Smm uint64_t i; 3816209962Smm int error; 3817209962Smm uint64_t n, s, txg; 3818209962Smm bufwad_t *packbuf, *bigbuf; 3819219089Spjd uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize; 3820219089Spjd uint64_t blocksize = ztest_random_blocksize(); 3821219089Spjd uint64_t chunksize = blocksize; 3822209962Smm uint64_t regions = 997; 3823209962Smm uint64_t stride = 123456789ULL; 3824209962Smm uint64_t width = 9; 3825209962Smm dmu_buf_t *bonus_db; 3826209962Smm arc_buf_t **bigbuf_arcbufs; 3827219089Spjd dmu_object_info_t doi; 3828209962Smm 3829209962Smm /* 3830209962Smm * This test uses two objects, packobj and bigobj, that are always 3831209962Smm * updated together (i.e. in the same tx) so that their contents are 3832209962Smm * in sync and can be compared. Their contents relate to each other 3833209962Smm * in a simple way: packobj is a dense array of 'bufwad' structures, 3834209962Smm * while bigobj is a sparse array of the same bufwads. Specifically, 3835209962Smm * for any index n, there are three bufwads that should be identical: 3836209962Smm * 3837209962Smm * packobj, at offset n * sizeof (bufwad_t) 3838209962Smm * bigobj, at the head of the nth chunk 3839209962Smm * bigobj, at the tail of the nth chunk 3840209962Smm * 3841209962Smm * The chunk size is set equal to bigobj block size so that 3842209962Smm * dmu_assign_arcbuf() can be tested for object updates. 3843209962Smm */ 3844209962Smm 3845209962Smm /* 3846209962Smm * Read the directory info. If it's the first time, set things up. 3847209962Smm */ 3848219089Spjd ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 3849219089Spjd ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize); 3850209962Smm 3851219089Spjd if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 3852219089Spjd return; 3853209962Smm 3854219089Spjd bigobj = od[0].od_object; 3855219089Spjd packobj = od[1].od_object; 3856219089Spjd blocksize = od[0].od_blocksize; 3857219089Spjd chunksize = blocksize; 3858219089Spjd ASSERT(chunksize == od[1].od_gen); 3859209962Smm 3860219089Spjd VERIFY(dmu_object_info(os, bigobj, &doi) == 0); 3861219089Spjd VERIFY(ISP2(doi.doi_data_block_size)); 3862219089Spjd VERIFY(chunksize == doi.doi_data_block_size); 3863219089Spjd VERIFY(chunksize >= 2 * sizeof (bufwad_t)); 3864209962Smm 3865209962Smm /* 3866209962Smm * Pick a random index and compute the offsets into packobj and bigobj. 3867209962Smm */ 3868209962Smm n = ztest_random(regions) * stride + ztest_random(width); 3869209962Smm s = 1 + ztest_random(width - 1); 3870209962Smm 3871209962Smm packoff = n * sizeof (bufwad_t); 3872209962Smm packsize = s * sizeof (bufwad_t); 3873209962Smm 3874219089Spjd bigoff = n * chunksize; 3875219089Spjd bigsize = s * chunksize; 3876209962Smm 3877209962Smm packbuf = umem_zalloc(packsize, UMEM_NOFAIL); 3878209962Smm bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL); 3879209962Smm 3880219089Spjd VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db)); 3881209962Smm 3882209962Smm bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL); 3883209962Smm 3884209962Smm /* 3885209962Smm * Iteration 0 test zcopy for DB_UNCACHED dbufs. 3886209962Smm * Iteration 1 test zcopy to already referenced dbufs. 3887209962Smm * Iteration 2 test zcopy to dirty dbuf in the same txg. 3888209962Smm * Iteration 3 test zcopy to dbuf dirty in previous txg. 3889209962Smm * Iteration 4 test zcopy when dbuf is no longer dirty. 3890209962Smm * Iteration 5 test zcopy when it can't be done. 3891209962Smm * Iteration 6 one more zcopy write. 3892209962Smm */ 3893209962Smm for (i = 0; i < 7; i++) { 3894209962Smm uint64_t j; 3895209962Smm uint64_t off; 3896209962Smm 3897209962Smm /* 3898209962Smm * In iteration 5 (i == 5) use arcbufs 3899209962Smm * that don't match bigobj blksz to test 3900209962Smm * dmu_assign_arcbuf() when it can't directly 3901209962Smm * assign an arcbuf to a dbuf. 3902209962Smm */ 3903209962Smm for (j = 0; j < s; j++) { 3904209962Smm if (i != 5) { 3905209962Smm bigbuf_arcbufs[j] = 3906219089Spjd dmu_request_arcbuf(bonus_db, chunksize); 3907209962Smm } else { 3908209962Smm bigbuf_arcbufs[2 * j] = 3909219089Spjd dmu_request_arcbuf(bonus_db, chunksize / 2); 3910209962Smm bigbuf_arcbufs[2 * j + 1] = 3911219089Spjd dmu_request_arcbuf(bonus_db, chunksize / 2); 3912209962Smm } 3913209962Smm } 3914209962Smm 3915209962Smm /* 3916209962Smm * Get a tx for the mods to both packobj and bigobj. 3917209962Smm */ 3918209962Smm tx = dmu_tx_create(os); 3919209962Smm 3920219089Spjd dmu_tx_hold_write(tx, packobj, packoff, packsize); 3921219089Spjd dmu_tx_hold_write(tx, bigobj, bigoff, bigsize); 3922209962Smm 3923219089Spjd txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 3924219089Spjd if (txg == 0) { 3925209962Smm umem_free(packbuf, packsize); 3926209962Smm umem_free(bigbuf, bigsize); 3927209962Smm for (j = 0; j < s; j++) { 3928209962Smm if (i != 5) { 3929209962Smm dmu_return_arcbuf(bigbuf_arcbufs[j]); 3930209962Smm } else { 3931209962Smm dmu_return_arcbuf( 3932209962Smm bigbuf_arcbufs[2 * j]); 3933209962Smm dmu_return_arcbuf( 3934209962Smm bigbuf_arcbufs[2 * j + 1]); 3935209962Smm } 3936209962Smm } 3937209962Smm umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); 3938209962Smm dmu_buf_rele(bonus_db, FTAG); 3939209962Smm return; 3940209962Smm } 3941209962Smm 3942209962Smm /* 3943209962Smm * 50% of the time don't read objects in the 1st iteration to 3944209962Smm * test dmu_assign_arcbuf() for the case when there're no 3945209962Smm * existing dbufs for the specified offsets. 3946209962Smm */ 3947209962Smm if (i != 0 || ztest_random(2) != 0) { 3948219089Spjd error = dmu_read(os, packobj, packoff, 3949209962Smm packsize, packbuf, DMU_READ_PREFETCH); 3950240415Smm ASSERT0(error); 3951219089Spjd error = dmu_read(os, bigobj, bigoff, bigsize, 3952209962Smm bigbuf, DMU_READ_PREFETCH); 3953240415Smm ASSERT0(error); 3954209962Smm } 3955209962Smm compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize, 3956219089Spjd n, chunksize, txg); 3957209962Smm 3958209962Smm /* 3959209962Smm * We've verified all the old bufwads, and made new ones. 3960209962Smm * Now write them out. 3961209962Smm */ 3962219089Spjd dmu_write(os, packobj, packoff, packsize, packbuf, tx); 3963236143Smm if (ztest_opts.zo_verbose >= 7) { 3964209962Smm (void) printf("writing offset %llx size %llx" 3965209962Smm " txg %llx\n", 3966209962Smm (u_longlong_t)bigoff, 3967209962Smm (u_longlong_t)bigsize, 3968209962Smm (u_longlong_t)txg); 3969209962Smm } 3970219089Spjd for (off = bigoff, j = 0; j < s; j++, off += chunksize) { 3971209962Smm dmu_buf_t *dbt; 3972209962Smm if (i != 5) { 3973209962Smm bcopy((caddr_t)bigbuf + (off - bigoff), 3974219089Spjd bigbuf_arcbufs[j]->b_data, chunksize); 3975209962Smm } else { 3976209962Smm bcopy((caddr_t)bigbuf + (off - bigoff), 3977209962Smm bigbuf_arcbufs[2 * j]->b_data, 3978219089Spjd chunksize / 2); 3979209962Smm bcopy((caddr_t)bigbuf + (off - bigoff) + 3980219089Spjd chunksize / 2, 3981209962Smm bigbuf_arcbufs[2 * j + 1]->b_data, 3982219089Spjd chunksize / 2); 3983209962Smm } 3984209962Smm 3985209962Smm if (i == 1) { 3986219089Spjd VERIFY(dmu_buf_hold(os, bigobj, off, 3987219089Spjd FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0); 3988209962Smm } 3989209962Smm if (i != 5) { 3990209962Smm dmu_assign_arcbuf(bonus_db, off, 3991209962Smm bigbuf_arcbufs[j], tx); 3992209962Smm } else { 3993209962Smm dmu_assign_arcbuf(bonus_db, off, 3994209962Smm bigbuf_arcbufs[2 * j], tx); 3995209962Smm dmu_assign_arcbuf(bonus_db, 3996219089Spjd off + chunksize / 2, 3997209962Smm bigbuf_arcbufs[2 * j + 1], tx); 3998209962Smm } 3999209962Smm if (i == 1) { 4000209962Smm dmu_buf_rele(dbt, FTAG); 4001209962Smm } 4002209962Smm } 4003209962Smm dmu_tx_commit(tx); 4004209962Smm 4005209962Smm /* 4006209962Smm * Sanity check the stuff we just wrote. 4007209962Smm */ 4008209962Smm { 4009209962Smm void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 4010209962Smm void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 4011209962Smm 4012219089Spjd VERIFY(0 == dmu_read(os, packobj, packoff, 4013209962Smm packsize, packcheck, DMU_READ_PREFETCH)); 4014219089Spjd VERIFY(0 == dmu_read(os, bigobj, bigoff, 4015209962Smm bigsize, bigcheck, DMU_READ_PREFETCH)); 4016209962Smm 4017209962Smm ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 4018209962Smm ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 4019209962Smm 4020209962Smm umem_free(packcheck, packsize); 4021209962Smm umem_free(bigcheck, bigsize); 4022209962Smm } 4023209962Smm if (i == 2) { 4024209962Smm txg_wait_open(dmu_objset_pool(os), 0); 4025209962Smm } else if (i == 3) { 4026209962Smm txg_wait_synced(dmu_objset_pool(os), 0); 4027209962Smm } 4028209962Smm } 4029209962Smm 4030209962Smm dmu_buf_rele(bonus_db, FTAG); 4031209962Smm umem_free(packbuf, packsize); 4032209962Smm umem_free(bigbuf, bigsize); 4033209962Smm umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *)); 4034209962Smm} 4035209962Smm 4036219089Spjd/* ARGSUSED */ 4037209962Smmvoid 4038219089Spjdztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id) 4039168404Spjd{ 4040219089Spjd ztest_od_t od[1]; 4041219089Spjd uint64_t offset = (1ULL << (ztest_random(20) + 43)) + 4042219089Spjd (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 4043168404Spjd 4044168404Spjd /* 4045219089Spjd * Have multiple threads write to large offsets in an object 4046219089Spjd * to verify that parallel writes to an object -- even to the 4047219089Spjd * same blocks within the object -- doesn't cause any trouble. 4048168404Spjd */ 4049219089Spjd ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 4050219089Spjd 4051219089Spjd if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4052219089Spjd return; 4053219089Spjd 4054219089Spjd while (ztest_random(10) != 0) 4055219089Spjd ztest_io(zd, od[0].od_object, offset); 4056168404Spjd} 4057168404Spjd 4058168404Spjdvoid 4059219089Spjdztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id) 4060168404Spjd{ 4061219089Spjd ztest_od_t od[1]; 4062219089Spjd uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) + 4063219089Spjd (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 4064219089Spjd uint64_t count = ztest_random(20) + 1; 4065219089Spjd uint64_t blocksize = ztest_random_blocksize(); 4066219089Spjd void *data; 4067168404Spjd 4068219089Spjd ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 4069168404Spjd 4070219089Spjd if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4071185029Spjd return; 4072168404Spjd 4073219089Spjd if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0) 4074185029Spjd return; 4075168404Spjd 4076219089Spjd ztest_prealloc(zd, od[0].od_object, offset, count * blocksize); 4077185029Spjd 4078219089Spjd data = umem_zalloc(blocksize, UMEM_NOFAIL); 4079185029Spjd 4080219089Spjd while (ztest_random(count) != 0) { 4081219089Spjd uint64_t randoff = offset + (ztest_random(count) * blocksize); 4082219089Spjd if (ztest_write(zd, od[0].od_object, randoff, blocksize, 4083219089Spjd data) != 0) 4084219089Spjd break; 4085219089Spjd while (ztest_random(4) != 0) 4086219089Spjd ztest_io(zd, od[0].od_object, randoff); 4087185029Spjd } 4088168404Spjd 4089219089Spjd umem_free(data, blocksize); 4090168404Spjd} 4091168404Spjd 4092168404Spjd/* 4093168404Spjd * Verify that zap_{create,destroy,add,remove,update} work as expected. 4094168404Spjd */ 4095168404Spjd#define ZTEST_ZAP_MIN_INTS 1 4096168404Spjd#define ZTEST_ZAP_MAX_INTS 4 4097168404Spjd#define ZTEST_ZAP_MAX_PROPS 1000 4098168404Spjd 4099168404Spjdvoid 4100219089Spjdztest_zap(ztest_ds_t *zd, uint64_t id) 4101168404Spjd{ 4102219089Spjd objset_t *os = zd->zd_os; 4103219089Spjd ztest_od_t od[1]; 4104168404Spjd uint64_t object; 4105168404Spjd uint64_t txg, last_txg; 4106168404Spjd uint64_t value[ZTEST_ZAP_MAX_INTS]; 4107168404Spjd uint64_t zl_ints, zl_intsize, prop; 4108168404Spjd int i, ints; 4109168404Spjd dmu_tx_t *tx; 4110168404Spjd char propname[100], txgname[100]; 4111168404Spjd int error; 4112168404Spjd char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; 4113168404Spjd 4114219089Spjd ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); 4115168404Spjd 4116219089Spjd if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4117219089Spjd return; 4118219089Spjd 4119219089Spjd object = od[0].od_object; 4120219089Spjd 4121168404Spjd /* 4122219089Spjd * Generate a known hash collision, and verify that 4123219089Spjd * we can lookup and remove both entries. 4124168404Spjd */ 4125219089Spjd tx = dmu_tx_create(os); 4126219089Spjd dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4127219089Spjd txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4128219089Spjd if (txg == 0) 4129219089Spjd return; 4130219089Spjd for (i = 0; i < 2; i++) { 4131219089Spjd value[i] = i; 4132219089Spjd VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t), 4133219089Spjd 1, &value[i], tx)); 4134168404Spjd } 4135219089Spjd for (i = 0; i < 2; i++) { 4136219089Spjd VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i], 4137219089Spjd sizeof (uint64_t), 1, &value[i], tx)); 4138219089Spjd VERIFY3U(0, ==, 4139219089Spjd zap_length(os, object, hc[i], &zl_intsize, &zl_ints)); 4140219089Spjd ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4141219089Spjd ASSERT3U(zl_ints, ==, 1); 4142219089Spjd } 4143219089Spjd for (i = 0; i < 2; i++) { 4144219089Spjd VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx)); 4145219089Spjd } 4146219089Spjd dmu_tx_commit(tx); 4147168404Spjd 4148219089Spjd /* 4149219089Spjd * Generate a buch of random entries. 4150219089Spjd */ 4151168404Spjd ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); 4152168404Spjd 4153185029Spjd prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 4154185029Spjd (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 4155185029Spjd (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 4156185029Spjd bzero(value, sizeof (value)); 4157185029Spjd last_txg = 0; 4158168404Spjd 4159185029Spjd /* 4160185029Spjd * If these zap entries already exist, validate their contents. 4161185029Spjd */ 4162185029Spjd error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 4163185029Spjd if (error == 0) { 4164185029Spjd ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4165185029Spjd ASSERT3U(zl_ints, ==, 1); 4166168404Spjd 4167185029Spjd VERIFY(zap_lookup(os, object, txgname, zl_intsize, 4168185029Spjd zl_ints, &last_txg) == 0); 4169168404Spjd 4170185029Spjd VERIFY(zap_length(os, object, propname, &zl_intsize, 4171185029Spjd &zl_ints) == 0); 4172168404Spjd 4173185029Spjd ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 4174185029Spjd ASSERT3U(zl_ints, ==, ints); 4175168404Spjd 4176185029Spjd VERIFY(zap_lookup(os, object, propname, zl_intsize, 4177185029Spjd zl_ints, value) == 0); 4178168404Spjd 4179185029Spjd for (i = 0; i < ints; i++) { 4180185029Spjd ASSERT3U(value[i], ==, last_txg + object + i); 4181168404Spjd } 4182185029Spjd } else { 4183185029Spjd ASSERT3U(error, ==, ENOENT); 4184185029Spjd } 4185168404Spjd 4186185029Spjd /* 4187185029Spjd * Atomically update two entries in our zap object. 4188185029Spjd * The first is named txg_%llu, and contains the txg 4189185029Spjd * in which the property was last updated. The second 4190185029Spjd * is named prop_%llu, and the nth element of its value 4191185029Spjd * should be txg + object + n. 4192185029Spjd */ 4193185029Spjd tx = dmu_tx_create(os); 4194219089Spjd dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4195219089Spjd txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4196219089Spjd if (txg == 0) 4197185029Spjd return; 4198168404Spjd 4199185029Spjd if (last_txg > txg) 4200185029Spjd fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); 4201168404Spjd 4202185029Spjd for (i = 0; i < ints; i++) 4203185029Spjd value[i] = txg + object + i; 4204168404Spjd 4205219089Spjd VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t), 4206219089Spjd 1, &txg, tx)); 4207219089Spjd VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t), 4208219089Spjd ints, value, tx)); 4209168404Spjd 4210185029Spjd dmu_tx_commit(tx); 4211168404Spjd 4212185029Spjd /* 4213185029Spjd * Remove a random pair of entries. 4214185029Spjd */ 4215185029Spjd prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 4216185029Spjd (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 4217185029Spjd (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 4218168404Spjd 4219185029Spjd error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 4220168404Spjd 4221185029Spjd if (error == ENOENT) 4222185029Spjd return; 4223168404Spjd 4224240415Smm ASSERT0(error); 4225168404Spjd 4226185029Spjd tx = dmu_tx_create(os); 4227219089Spjd dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4228219089Spjd txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4229219089Spjd if (txg == 0) 4230185029Spjd return; 4231219089Spjd VERIFY3U(0, ==, zap_remove(os, object, txgname, tx)); 4232219089Spjd VERIFY3U(0, ==, zap_remove(os, object, propname, tx)); 4233185029Spjd dmu_tx_commit(tx); 4234168404Spjd} 4235168404Spjd 4236209962Smm/* 4237209962Smm * Testcase to test the upgrading of a microzap to fatzap. 4238209962Smm */ 4239168404Spjdvoid 4240219089Spjdztest_fzap(ztest_ds_t *zd, uint64_t id) 4241209962Smm{ 4242219089Spjd objset_t *os = zd->zd_os; 4243219089Spjd ztest_od_t od[1]; 4244219089Spjd uint64_t object, txg; 4245209962Smm 4246219089Spjd ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0); 4247209962Smm 4248219089Spjd if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0) 4249219089Spjd return; 4250209962Smm 4251219089Spjd object = od[0].od_object; 4252209962Smm 4253209962Smm /* 4254219089Spjd * Add entries to this ZAP and make sure it spills over 4255209962Smm * and gets upgraded to a fatzap. Also, since we are adding 4256219089Spjd * 2050 entries we should see ptrtbl growth and leaf-block split. 4257209962Smm */ 4258219089Spjd for (int i = 0; i < 2050; i++) { 4259219089Spjd char name[MAXNAMELEN]; 4260219089Spjd uint64_t value = i; 4261219089Spjd dmu_tx_t *tx; 4262219089Spjd int error; 4263209962Smm 4264219089Spjd (void) snprintf(name, sizeof (name), "fzap-%llu-%llu", 4265219089Spjd id, value); 4266219089Spjd 4267209962Smm tx = dmu_tx_create(os); 4268219089Spjd dmu_tx_hold_zap(tx, object, B_TRUE, name); 4269219089Spjd txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4270219089Spjd if (txg == 0) 4271209962Smm return; 4272219089Spjd error = zap_add(os, object, name, sizeof (uint64_t), 1, 4273219089Spjd &value, tx); 4274209962Smm ASSERT(error == 0 || error == EEXIST); 4275209962Smm dmu_tx_commit(tx); 4276209962Smm } 4277209962Smm} 4278209962Smm 4279219089Spjd/* ARGSUSED */ 4280209962Smmvoid 4281219089Spjdztest_zap_parallel(ztest_ds_t *zd, uint64_t id) 4282168404Spjd{ 4283219089Spjd objset_t *os = zd->zd_os; 4284219089Spjd ztest_od_t od[1]; 4285168404Spjd uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; 4286168404Spjd dmu_tx_t *tx; 4287168404Spjd int i, namelen, error; 4288219089Spjd int micro = ztest_random(2); 4289168404Spjd char name[20], string_value[20]; 4290168404Spjd void *data; 4291168404Spjd 4292219089Spjd ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0); 4293219089Spjd 4294219089Spjd if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4295219089Spjd return; 4296219089Spjd 4297219089Spjd object = od[0].od_object; 4298219089Spjd 4299185029Spjd /* 4300185029Spjd * Generate a random name of the form 'xxx.....' where each 4301185029Spjd * x is a random printable character and the dots are dots. 4302185029Spjd * There are 94 such characters, and the name length goes from 4303185029Spjd * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. 4304185029Spjd */ 4305185029Spjd namelen = ztest_random(sizeof (name) - 5) + 5 + 1; 4306168404Spjd 4307185029Spjd for (i = 0; i < 3; i++) 4308185029Spjd name[i] = '!' + ztest_random('~' - '!' + 1); 4309185029Spjd for (; i < namelen - 1; i++) 4310185029Spjd name[i] = '.'; 4311185029Spjd name[i] = '\0'; 4312168404Spjd 4313219089Spjd if ((namelen & 1) || micro) { 4314185029Spjd wsize = sizeof (txg); 4315185029Spjd wc = 1; 4316185029Spjd data = &txg; 4317185029Spjd } else { 4318185029Spjd wsize = 1; 4319185029Spjd wc = namelen; 4320185029Spjd data = string_value; 4321185029Spjd } 4322168404Spjd 4323185029Spjd count = -1ULL; 4324248571Smm VERIFY0(zap_count(os, object, &count)); 4325185029Spjd ASSERT(count != -1ULL); 4326168404Spjd 4327185029Spjd /* 4328185029Spjd * Select an operation: length, lookup, add, update, remove. 4329185029Spjd */ 4330185029Spjd i = ztest_random(5); 4331168404Spjd 4332185029Spjd if (i >= 2) { 4333185029Spjd tx = dmu_tx_create(os); 4334219089Spjd dmu_tx_hold_zap(tx, object, B_TRUE, NULL); 4335219089Spjd txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG); 4336219089Spjd if (txg == 0) 4337185029Spjd return; 4338185029Spjd bcopy(name, string_value, namelen); 4339185029Spjd } else { 4340185029Spjd tx = NULL; 4341185029Spjd txg = 0; 4342185029Spjd bzero(string_value, namelen); 4343185029Spjd } 4344168404Spjd 4345185029Spjd switch (i) { 4346168404Spjd 4347185029Spjd case 0: 4348185029Spjd error = zap_length(os, object, name, &zl_wsize, &zl_wc); 4349185029Spjd if (error == 0) { 4350185029Spjd ASSERT3U(wsize, ==, zl_wsize); 4351185029Spjd ASSERT3U(wc, ==, zl_wc); 4352185029Spjd } else { 4353185029Spjd ASSERT3U(error, ==, ENOENT); 4354185029Spjd } 4355185029Spjd break; 4356168404Spjd 4357185029Spjd case 1: 4358185029Spjd error = zap_lookup(os, object, name, wsize, wc, data); 4359185029Spjd if (error == 0) { 4360185029Spjd if (data == string_value && 4361185029Spjd bcmp(name, data, namelen) != 0) 4362185029Spjd fatal(0, "name '%s' != val '%s' len %d", 4363185029Spjd name, data, namelen); 4364185029Spjd } else { 4365185029Spjd ASSERT3U(error, ==, ENOENT); 4366185029Spjd } 4367185029Spjd break; 4368168404Spjd 4369185029Spjd case 2: 4370185029Spjd error = zap_add(os, object, name, wsize, wc, data, tx); 4371185029Spjd ASSERT(error == 0 || error == EEXIST); 4372185029Spjd break; 4373168404Spjd 4374185029Spjd case 3: 4375185029Spjd VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); 4376185029Spjd break; 4377168404Spjd 4378185029Spjd case 4: 4379185029Spjd error = zap_remove(os, object, name, tx); 4380185029Spjd ASSERT(error == 0 || error == ENOENT); 4381185029Spjd break; 4382185029Spjd } 4383168404Spjd 4384185029Spjd if (tx != NULL) 4385185029Spjd dmu_tx_commit(tx); 4386168404Spjd} 4387168404Spjd 4388219089Spjd/* 4389219089Spjd * Commit callback data. 4390219089Spjd */ 4391219089Spjdtypedef struct ztest_cb_data { 4392219089Spjd list_node_t zcd_node; 4393219089Spjd uint64_t zcd_txg; 4394219089Spjd int zcd_expected_err; 4395219089Spjd boolean_t zcd_added; 4396219089Spjd boolean_t zcd_called; 4397219089Spjd spa_t *zcd_spa; 4398219089Spjd} ztest_cb_data_t; 4399219089Spjd 4400219089Spjd/* This is the actual commit callback function */ 4401219089Spjdstatic void 4402219089Spjdztest_commit_callback(void *arg, int error) 4403219089Spjd{ 4404219089Spjd ztest_cb_data_t *data = arg; 4405219089Spjd uint64_t synced_txg; 4406219089Spjd 4407219089Spjd VERIFY(data != NULL); 4408219089Spjd VERIFY3S(data->zcd_expected_err, ==, error); 4409219089Spjd VERIFY(!data->zcd_called); 4410219089Spjd 4411219089Spjd synced_txg = spa_last_synced_txg(data->zcd_spa); 4412219089Spjd if (data->zcd_txg > synced_txg) 4413219089Spjd fatal(0, "commit callback of txg %" PRIu64 " called prematurely" 4414219089Spjd ", last synced txg = %" PRIu64 "\n", data->zcd_txg, 4415219089Spjd synced_txg); 4416219089Spjd 4417219089Spjd data->zcd_called = B_TRUE; 4418219089Spjd 4419219089Spjd if (error == ECANCELED) { 4420240415Smm ASSERT0(data->zcd_txg); 4421219089Spjd ASSERT(!data->zcd_added); 4422219089Spjd 4423219089Spjd /* 4424219089Spjd * The private callback data should be destroyed here, but 4425219089Spjd * since we are going to check the zcd_called field after 4426219089Spjd * dmu_tx_abort(), we will destroy it there. 4427219089Spjd */ 4428219089Spjd return; 4429219089Spjd } 4430219089Spjd 4431219089Spjd /* Was this callback added to the global callback list? */ 4432219089Spjd if (!data->zcd_added) 4433219089Spjd goto out; 4434219089Spjd 4435219089Spjd ASSERT3U(data->zcd_txg, !=, 0); 4436219089Spjd 4437219089Spjd /* Remove our callback from the list */ 4438219089Spjd (void) mutex_lock(&zcl.zcl_callbacks_lock); 4439219089Spjd list_remove(&zcl.zcl_callbacks, data); 4440219089Spjd (void) mutex_unlock(&zcl.zcl_callbacks_lock); 4441219089Spjd 4442219089Spjdout: 4443219089Spjd umem_free(data, sizeof (ztest_cb_data_t)); 4444219089Spjd} 4445219089Spjd 4446219089Spjd/* Allocate and initialize callback data structure */ 4447219089Spjdstatic ztest_cb_data_t * 4448219089Spjdztest_create_cb_data(objset_t *os, uint64_t txg) 4449219089Spjd{ 4450219089Spjd ztest_cb_data_t *cb_data; 4451219089Spjd 4452219089Spjd cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL); 4453219089Spjd 4454219089Spjd cb_data->zcd_txg = txg; 4455219089Spjd cb_data->zcd_spa = dmu_objset_spa(os); 4456219089Spjd 4457219089Spjd return (cb_data); 4458219089Spjd} 4459219089Spjd 4460219089Spjd/* 4461219089Spjd * If a number of txgs equal to this threshold have been created after a commit 4462219089Spjd * callback has been registered but not called, then we assume there is an 4463219089Spjd * implementation bug. 4464219089Spjd */ 4465219089Spjd#define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2) 4466219089Spjd 4467219089Spjd/* 4468219089Spjd * Commit callback test. 4469219089Spjd */ 4470168404Spjdvoid 4471219089Spjdztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id) 4472168404Spjd{ 4473219089Spjd objset_t *os = zd->zd_os; 4474219089Spjd ztest_od_t od[1]; 4475219089Spjd dmu_tx_t *tx; 4476219089Spjd ztest_cb_data_t *cb_data[3], *tmp_cb; 4477219089Spjd uint64_t old_txg, txg; 4478219089Spjd int i, error; 4479219089Spjd 4480219089Spjd ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 4481219089Spjd 4482219089Spjd if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 4483219089Spjd return; 4484219089Spjd 4485219089Spjd tx = dmu_tx_create(os); 4486219089Spjd 4487219089Spjd cb_data[0] = ztest_create_cb_data(os, 0); 4488219089Spjd dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]); 4489219089Spjd 4490219089Spjd dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t)); 4491219089Spjd 4492219089Spjd /* Every once in a while, abort the transaction on purpose */ 4493219089Spjd if (ztest_random(100) == 0) 4494219089Spjd error = -1; 4495219089Spjd 4496219089Spjd if (!error) 4497219089Spjd error = dmu_tx_assign(tx, TXG_NOWAIT); 4498219089Spjd 4499219089Spjd txg = error ? 0 : dmu_tx_get_txg(tx); 4500219089Spjd 4501219089Spjd cb_data[0]->zcd_txg = txg; 4502219089Spjd cb_data[1] = ztest_create_cb_data(os, txg); 4503219089Spjd dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]); 4504219089Spjd 4505219089Spjd if (error) { 4506219089Spjd /* 4507219089Spjd * It's not a strict requirement to call the registered 4508219089Spjd * callbacks from inside dmu_tx_abort(), but that's what 4509219089Spjd * it's supposed to happen in the current implementation 4510219089Spjd * so we will check for that. 4511219089Spjd */ 4512219089Spjd for (i = 0; i < 2; i++) { 4513219089Spjd cb_data[i]->zcd_expected_err = ECANCELED; 4514219089Spjd VERIFY(!cb_data[i]->zcd_called); 4515219089Spjd } 4516219089Spjd 4517219089Spjd dmu_tx_abort(tx); 4518219089Spjd 4519219089Spjd for (i = 0; i < 2; i++) { 4520219089Spjd VERIFY(cb_data[i]->zcd_called); 4521219089Spjd umem_free(cb_data[i], sizeof (ztest_cb_data_t)); 4522219089Spjd } 4523219089Spjd 4524219089Spjd return; 4525219089Spjd } 4526219089Spjd 4527219089Spjd cb_data[2] = ztest_create_cb_data(os, txg); 4528219089Spjd dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]); 4529219089Spjd 4530219089Spjd /* 4531219089Spjd * Read existing data to make sure there isn't a future leak. 4532219089Spjd */ 4533219089Spjd VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t), 4534219089Spjd &old_txg, DMU_READ_PREFETCH)); 4535219089Spjd 4536219089Spjd if (old_txg > txg) 4537219089Spjd fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64, 4538219089Spjd old_txg, txg); 4539219089Spjd 4540219089Spjd dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx); 4541219089Spjd 4542219089Spjd (void) mutex_lock(&zcl.zcl_callbacks_lock); 4543219089Spjd 4544219089Spjd /* 4545219089Spjd * Since commit callbacks don't have any ordering requirement and since 4546219089Spjd * it is theoretically possible for a commit callback to be called 4547219089Spjd * after an arbitrary amount of time has elapsed since its txg has been 4548219089Spjd * synced, it is difficult to reliably determine whether a commit 4549219089Spjd * callback hasn't been called due to high load or due to a flawed 4550219089Spjd * implementation. 4551219089Spjd * 4552219089Spjd * In practice, we will assume that if after a certain number of txgs a 4553219089Spjd * commit callback hasn't been called, then most likely there's an 4554219089Spjd * implementation bug.. 4555219089Spjd */ 4556219089Spjd tmp_cb = list_head(&zcl.zcl_callbacks); 4557219089Spjd if (tmp_cb != NULL && 4558251635Sdelphij (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) { 4559219089Spjd fatal(0, "Commit callback threshold exceeded, oldest txg: %" 4560219089Spjd PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg); 4561219089Spjd } 4562219089Spjd 4563219089Spjd /* 4564219089Spjd * Let's find the place to insert our callbacks. 4565219089Spjd * 4566219089Spjd * Even though the list is ordered by txg, it is possible for the 4567219089Spjd * insertion point to not be the end because our txg may already be 4568219089Spjd * quiescing at this point and other callbacks in the open txg 4569219089Spjd * (from other objsets) may have sneaked in. 4570219089Spjd */ 4571219089Spjd tmp_cb = list_tail(&zcl.zcl_callbacks); 4572219089Spjd while (tmp_cb != NULL && tmp_cb->zcd_txg > txg) 4573219089Spjd tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb); 4574219089Spjd 4575219089Spjd /* Add the 3 callbacks to the list */ 4576219089Spjd for (i = 0; i < 3; i++) { 4577219089Spjd if (tmp_cb == NULL) 4578219089Spjd list_insert_head(&zcl.zcl_callbacks, cb_data[i]); 4579219089Spjd else 4580219089Spjd list_insert_after(&zcl.zcl_callbacks, tmp_cb, 4581219089Spjd cb_data[i]); 4582219089Spjd 4583219089Spjd cb_data[i]->zcd_added = B_TRUE; 4584219089Spjd VERIFY(!cb_data[i]->zcd_called); 4585219089Spjd 4586219089Spjd tmp_cb = cb_data[i]; 4587219089Spjd } 4588219089Spjd 4589219089Spjd (void) mutex_unlock(&zcl.zcl_callbacks_lock); 4590219089Spjd 4591219089Spjd dmu_tx_commit(tx); 4592219089Spjd} 4593219089Spjd 4594219089Spjd/* ARGSUSED */ 4595219089Spjdvoid 4596219089Spjdztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id) 4597219089Spjd{ 4598219089Spjd zfs_prop_t proplist[] = { 4599219089Spjd ZFS_PROP_CHECKSUM, 4600219089Spjd ZFS_PROP_COMPRESSION, 4601219089Spjd ZFS_PROP_COPIES, 4602219089Spjd ZFS_PROP_DEDUP 4603219089Spjd }; 4604219089Spjd 4605236143Smm (void) rw_rdlock(&ztest_name_lock); 4606219089Spjd 4607219089Spjd for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) 4608219089Spjd (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p], 4609219089Spjd ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2)); 4610219089Spjd 4611236143Smm (void) rw_unlock(&ztest_name_lock); 4612219089Spjd} 4613219089Spjd 4614219089Spjd/* ARGSUSED */ 4615219089Spjdvoid 4616219089Spjdztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id) 4617219089Spjd{ 4618219089Spjd nvlist_t *props = NULL; 4619219089Spjd 4620236143Smm (void) rw_rdlock(&ztest_name_lock); 4621219089Spjd 4622236143Smm (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO, 4623219089Spjd ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN)); 4624219089Spjd 4625240415Smm VERIFY0(spa_prop_get(ztest_spa, &props)); 4626219089Spjd 4627236143Smm if (ztest_opts.zo_verbose >= 6) 4628219089Spjd dump_nvlist(props, 4); 4629219089Spjd 4630219089Spjd nvlist_free(props); 4631219089Spjd 4632236143Smm (void) rw_unlock(&ztest_name_lock); 4633219089Spjd} 4634219089Spjd 4635248571Smmstatic int 4636248571Smmuser_release_one(const char *snapname, const char *holdname) 4637248571Smm{ 4638248571Smm nvlist_t *snaps, *holds; 4639248571Smm int error; 4640248571Smm 4641248571Smm snaps = fnvlist_alloc(); 4642248571Smm holds = fnvlist_alloc(); 4643248571Smm fnvlist_add_boolean(holds, holdname); 4644248571Smm fnvlist_add_nvlist(snaps, snapname, holds); 4645248571Smm fnvlist_free(holds); 4646248571Smm error = dsl_dataset_user_release(snaps, NULL); 4647248571Smm fnvlist_free(snaps); 4648248571Smm return (error); 4649248571Smm} 4650248571Smm 4651219089Spjd/* 4652219089Spjd * Test snapshot hold/release and deferred destroy. 4653219089Spjd */ 4654219089Spjdvoid 4655219089Spjdztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id) 4656219089Spjd{ 4657219089Spjd int error; 4658219089Spjd objset_t *os = zd->zd_os; 4659219089Spjd objset_t *origin; 4660219089Spjd char snapname[100]; 4661219089Spjd char fullname[100]; 4662219089Spjd char clonename[100]; 4663219089Spjd char tag[100]; 4664168404Spjd char osname[MAXNAMELEN]; 4665248571Smm nvlist_t *holds; 4666168404Spjd 4667236143Smm (void) rw_rdlock(&ztest_name_lock); 4668168404Spjd 4669168404Spjd dmu_objset_name(os, osname); 4670168404Spjd 4671248571Smm (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id); 4672248571Smm (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname); 4673248571Smm (void) snprintf(clonename, sizeof (clonename), 4674248571Smm "%s/ch1_%llu", osname, id); 4675248571Smm (void) snprintf(tag, sizeof (tag), "tag_%llu", id); 4676219089Spjd 4677219089Spjd /* 4678219089Spjd * Clean up from any previous run. 4679219089Spjd */ 4680248571Smm error = dsl_destroy_head(clonename); 4681248571Smm if (error != ENOENT) 4682248571Smm ASSERT0(error); 4683248571Smm error = user_release_one(fullname, tag); 4684248571Smm if (error != ESRCH && error != ENOENT) 4685248571Smm ASSERT0(error); 4686248571Smm error = dsl_destroy_snapshot(fullname, B_FALSE); 4687248571Smm if (error != ENOENT) 4688248571Smm ASSERT0(error); 4689219089Spjd 4690219089Spjd /* 4691219089Spjd * Create snapshot, clone it, mark snap for deferred destroy, 4692219089Spjd * destroy clone, verify snap was also destroyed. 4693219089Spjd */ 4694248571Smm error = dmu_objset_snapshot_one(osname, snapname); 4695219089Spjd if (error) { 4696219089Spjd if (error == ENOSPC) { 4697219089Spjd ztest_record_enospc("dmu_objset_snapshot"); 4698219089Spjd goto out; 4699168404Spjd } 4700219089Spjd fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); 4701219089Spjd } 4702168404Spjd 4703248571Smm error = dmu_objset_clone(clonename, fullname); 4704219089Spjd if (error) { 4705168404Spjd if (error == ENOSPC) { 4706219089Spjd ztest_record_enospc("dmu_objset_clone"); 4707219089Spjd goto out; 4708168404Spjd } 4709219089Spjd fatal(0, "dmu_objset_clone(%s) = %d", clonename, error); 4710219089Spjd } 4711168404Spjd 4712248571Smm error = dsl_destroy_snapshot(fullname, B_TRUE); 4713219089Spjd if (error) { 4714248571Smm fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", 4715219089Spjd fullname, error); 4716219089Spjd } 4717168404Spjd 4718248571Smm error = dsl_destroy_head(clonename); 4719219089Spjd if (error) 4720248571Smm fatal(0, "dsl_destroy_head(%s) = %d", clonename, error); 4721168404Spjd 4722219089Spjd error = dmu_objset_hold(fullname, FTAG, &origin); 4723219089Spjd if (error != ENOENT) 4724219089Spjd fatal(0, "dmu_objset_hold(%s) = %d", fullname, error); 4725168404Spjd 4726219089Spjd /* 4727219089Spjd * Create snapshot, add temporary hold, verify that we can't 4728219089Spjd * destroy a held snapshot, mark for deferred destroy, 4729219089Spjd * release hold, verify snapshot was destroyed. 4730219089Spjd */ 4731248571Smm error = dmu_objset_snapshot_one(osname, snapname); 4732219089Spjd if (error) { 4733219089Spjd if (error == ENOSPC) { 4734219089Spjd ztest_record_enospc("dmu_objset_snapshot"); 4735219089Spjd goto out; 4736168404Spjd } 4737219089Spjd fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error); 4738168404Spjd } 4739168404Spjd 4740248571Smm holds = fnvlist_alloc(); 4741248571Smm fnvlist_add_string(holds, fullname, tag); 4742248571Smm error = dsl_dataset_user_hold(holds, 0, NULL); 4743248571Smm fnvlist_free(holds); 4744248571Smm 4745268075Sdelphij if (error == ENOSPC) { 4746268075Sdelphij ztest_record_enospc("dsl_dataset_user_hold"); 4747268075Sdelphij goto out; 4748268075Sdelphij } else if (error) { 4749268075Sdelphij fatal(0, "dsl_dataset_user_hold(%s, %s) = %u", 4750268075Sdelphij fullname, tag, error); 4751268075Sdelphij } 4752219089Spjd 4753248571Smm error = dsl_destroy_snapshot(fullname, B_FALSE); 4754219089Spjd if (error != EBUSY) { 4755248571Smm fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d", 4756219089Spjd fullname, error); 4757219089Spjd } 4758219089Spjd 4759248571Smm error = dsl_destroy_snapshot(fullname, B_TRUE); 4760219089Spjd if (error) { 4761248571Smm fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d", 4762219089Spjd fullname, error); 4763219089Spjd } 4764219089Spjd 4765248571Smm error = user_release_one(fullname, tag); 4766219089Spjd if (error) 4767251646Sdelphij fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error); 4768219089Spjd 4769248571Smm VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT); 4770219089Spjd 4771219089Spjdout: 4772236143Smm (void) rw_unlock(&ztest_name_lock); 4773168404Spjd} 4774168404Spjd 4775168404Spjd/* 4776168404Spjd * Inject random faults into the on-disk data. 4777168404Spjd */ 4778219089Spjd/* ARGSUSED */ 4779168404Spjdvoid 4780219089Spjdztest_fault_inject(ztest_ds_t *zd, uint64_t id) 4781168404Spjd{ 4782219089Spjd ztest_shared_t *zs = ztest_shared; 4783236143Smm spa_t *spa = ztest_spa; 4784168404Spjd int fd; 4785168404Spjd uint64_t offset; 4786219089Spjd uint64_t leaves; 4787168404Spjd uint64_t bad = 0x1990c0ffeedecadeULL; 4788168404Spjd uint64_t top, leaf; 4789168404Spjd char path0[MAXPATHLEN]; 4790168404Spjd char pathrand[MAXPATHLEN]; 4791168404Spjd size_t fsize; 4792168404Spjd int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ 4793168404Spjd int iters = 1000; 4794219089Spjd int maxfaults; 4795219089Spjd int mirror_save; 4796185029Spjd vdev_t *vd0 = NULL; 4797168404Spjd uint64_t guid0 = 0; 4798219089Spjd boolean_t islog = B_FALSE; 4799168404Spjd 4800236143Smm VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 4801219089Spjd maxfaults = MAXFAULTS(); 4802236143Smm leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz; 4803219089Spjd mirror_save = zs->zs_mirrors; 4804236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4805219089Spjd 4806185029Spjd ASSERT(leaves >= 1); 4807168404Spjd 4808168404Spjd /* 4809254074Sdelphij * Grab the name lock as reader. There are some operations 4810254074Sdelphij * which don't like to have their vdevs changed while 4811254074Sdelphij * they are in progress (i.e. spa_change_guid). Those 4812254074Sdelphij * operations will have grabbed the name lock as writer. 4813254074Sdelphij */ 4814254074Sdelphij (void) rw_rdlock(&ztest_name_lock); 4815254074Sdelphij 4816254074Sdelphij /* 4817185029Spjd * We need SCL_STATE here because we're going to look at vd0->vdev_tsd. 4818168404Spjd */ 4819185029Spjd spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 4820168404Spjd 4821185029Spjd if (ztest_random(2) == 0) { 4822185029Spjd /* 4823219089Spjd * Inject errors on a normal data device or slog device. 4824185029Spjd */ 4825219089Spjd top = ztest_random_vdev_top(spa, B_TRUE); 4826219089Spjd leaf = ztest_random(leaves) + zs->zs_splits; 4827168404Spjd 4828185029Spjd /* 4829185029Spjd * Generate paths to the first leaf in this top-level vdev, 4830185029Spjd * and to the random leaf we selected. We'll induce transient 4831185029Spjd * write failures and random online/offline activity on leaf 0, 4832185029Spjd * and we'll write random garbage to the randomly chosen leaf. 4833185029Spjd */ 4834185029Spjd (void) snprintf(path0, sizeof (path0), ztest_dev_template, 4835236143Smm ztest_opts.zo_dir, ztest_opts.zo_pool, 4836236143Smm top * leaves + zs->zs_splits); 4837185029Spjd (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template, 4838236143Smm ztest_opts.zo_dir, ztest_opts.zo_pool, 4839236143Smm top * leaves + leaf); 4840168404Spjd 4841185029Spjd vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); 4842219089Spjd if (vd0 != NULL && vd0->vdev_top->vdev_islog) 4843219089Spjd islog = B_TRUE; 4844219089Spjd 4845254074Sdelphij /* 4846254074Sdelphij * If the top-level vdev needs to be resilvered 4847254074Sdelphij * then we only allow faults on the device that is 4848254074Sdelphij * resilvering. 4849254074Sdelphij */ 4850254074Sdelphij if (vd0 != NULL && maxfaults != 1 && 4851254074Sdelphij (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) || 4852254112Sdelphij vd0->vdev_resilver_txg != 0)) { 4853185029Spjd /* 4854185029Spjd * Make vd0 explicitly claim to be unreadable, 4855185029Spjd * or unwriteable, or reach behind its back 4856185029Spjd * and close the underlying fd. We can do this if 4857185029Spjd * maxfaults == 0 because we'll fail and reexecute, 4858185029Spjd * and we can do it if maxfaults >= 2 because we'll 4859185029Spjd * have enough redundancy. If maxfaults == 1, the 4860185029Spjd * combination of this with injection of random data 4861185029Spjd * corruption below exceeds the pool's fault tolerance. 4862185029Spjd */ 4863185029Spjd vdev_file_t *vf = vd0->vdev_tsd; 4864168404Spjd 4865185029Spjd if (vf != NULL && ztest_random(3) == 0) { 4866185029Spjd (void) close(vf->vf_vnode->v_fd); 4867185029Spjd vf->vf_vnode->v_fd = -1; 4868185029Spjd } else if (ztest_random(2) == 0) { 4869185029Spjd vd0->vdev_cant_read = B_TRUE; 4870185029Spjd } else { 4871185029Spjd vd0->vdev_cant_write = B_TRUE; 4872185029Spjd } 4873185029Spjd guid0 = vd0->vdev_guid; 4874185029Spjd } 4875185029Spjd } else { 4876185029Spjd /* 4877185029Spjd * Inject errors on an l2cache device. 4878185029Spjd */ 4879185029Spjd spa_aux_vdev_t *sav = &spa->spa_l2cache; 4880168404Spjd 4881185029Spjd if (sav->sav_count == 0) { 4882185029Spjd spa_config_exit(spa, SCL_STATE, FTAG); 4883254074Sdelphij (void) rw_unlock(&ztest_name_lock); 4884185029Spjd return; 4885185029Spjd } 4886185029Spjd vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)]; 4887168404Spjd guid0 = vd0->vdev_guid; 4888185029Spjd (void) strcpy(path0, vd0->vdev_path); 4889185029Spjd (void) strcpy(pathrand, vd0->vdev_path); 4890185029Spjd 4891185029Spjd leaf = 0; 4892185029Spjd leaves = 1; 4893185029Spjd maxfaults = INT_MAX; /* no limit on cache devices */ 4894168404Spjd } 4895168404Spjd 4896185029Spjd spa_config_exit(spa, SCL_STATE, FTAG); 4897254074Sdelphij (void) rw_unlock(&ztest_name_lock); 4898185029Spjd 4899168404Spjd /* 4900219089Spjd * If we can tolerate two or more faults, or we're dealing 4901219089Spjd * with a slog, randomly online/offline vd0. 4902168404Spjd */ 4903219089Spjd if ((maxfaults >= 2 || islog) && guid0 != 0) { 4904209962Smm if (ztest_random(10) < 6) { 4905209962Smm int flags = (ztest_random(2) == 0 ? 4906209962Smm ZFS_OFFLINE_TEMPORARY : 0); 4907219089Spjd 4908219089Spjd /* 4909219089Spjd * We have to grab the zs_name_lock as writer to 4910219089Spjd * prevent a race between offlining a slog and 4911219089Spjd * destroying a dataset. Offlining the slog will 4912219089Spjd * grab a reference on the dataset which may cause 4913219089Spjd * dmu_objset_destroy() to fail with EBUSY thus 4914219089Spjd * leaving the dataset in an inconsistent state. 4915219089Spjd */ 4916219089Spjd if (islog) 4917236143Smm (void) rw_wrlock(&ztest_name_lock); 4918219089Spjd 4919209962Smm VERIFY(vdev_offline(spa, guid0, flags) != EBUSY); 4920219089Spjd 4921219089Spjd if (islog) 4922236143Smm (void) rw_unlock(&ztest_name_lock); 4923209962Smm } else { 4924242845Sdelphij /* 4925242845Sdelphij * Ideally we would like to be able to randomly 4926242845Sdelphij * call vdev_[on|off]line without holding locks 4927242845Sdelphij * to force unpredictable failures but the side 4928242845Sdelphij * effects of vdev_[on|off]line prevent us from 4929242845Sdelphij * doing so. We grab the ztest_vdev_lock here to 4930242845Sdelphij * prevent a race between injection testing and 4931242845Sdelphij * aux_vdev removal. 4932242845Sdelphij */ 4933242845Sdelphij VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 4934209962Smm (void) vdev_online(spa, guid0, 0, NULL); 4935242845Sdelphij VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4936209962Smm } 4937168404Spjd } 4938168404Spjd 4939219089Spjd if (maxfaults == 0) 4940219089Spjd return; 4941219089Spjd 4942168404Spjd /* 4943168404Spjd * We have at least single-fault tolerance, so inject data corruption. 4944168404Spjd */ 4945168404Spjd fd = open(pathrand, O_RDWR); 4946168404Spjd 4947168404Spjd if (fd == -1) /* we hit a gap in the device namespace */ 4948168404Spjd return; 4949168404Spjd 4950168404Spjd fsize = lseek(fd, 0, SEEK_END); 4951168404Spjd 4952168404Spjd while (--iters != 0) { 4953168404Spjd offset = ztest_random(fsize / (leaves << bshift)) * 4954168404Spjd (leaves << bshift) + (leaf << bshift) + 4955168404Spjd (ztest_random(1ULL << (bshift - 1)) & -8ULL); 4956168404Spjd 4957168404Spjd if (offset >= fsize) 4958168404Spjd continue; 4959168404Spjd 4960236143Smm VERIFY(mutex_lock(&ztest_vdev_lock) == 0); 4961219089Spjd if (mirror_save != zs->zs_mirrors) { 4962236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4963219089Spjd (void) close(fd); 4964219089Spjd return; 4965219089Spjd } 4966168404Spjd 4967168404Spjd if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) 4968168404Spjd fatal(1, "can't inject bad word at 0x%llx in %s", 4969168404Spjd offset, pathrand); 4970219089Spjd 4971236143Smm VERIFY(mutex_unlock(&ztest_vdev_lock) == 0); 4972219089Spjd 4973236143Smm if (ztest_opts.zo_verbose >= 7) 4974219089Spjd (void) printf("injected bad word into %s," 4975219089Spjd " offset 0x%llx\n", pathrand, (u_longlong_t)offset); 4976168404Spjd } 4977168404Spjd 4978168404Spjd (void) close(fd); 4979168404Spjd} 4980168404Spjd 4981168404Spjd/* 4982219089Spjd * Verify that DDT repair works as expected. 4983219089Spjd */ 4984219089Spjdvoid 4985219089Spjdztest_ddt_repair(ztest_ds_t *zd, uint64_t id) 4986219089Spjd{ 4987219089Spjd ztest_shared_t *zs = ztest_shared; 4988236143Smm spa_t *spa = ztest_spa; 4989219089Spjd objset_t *os = zd->zd_os; 4990219089Spjd ztest_od_t od[1]; 4991219089Spjd uint64_t object, blocksize, txg, pattern, psize; 4992219089Spjd enum zio_checksum checksum = spa_dedup_checksum(spa); 4993219089Spjd dmu_buf_t *db; 4994219089Spjd dmu_tx_t *tx; 4995219089Spjd void *buf; 4996219089Spjd blkptr_t blk; 4997219089Spjd int copies = 2 * ZIO_DEDUPDITTO_MIN; 4998219089Spjd 4999219089Spjd blocksize = ztest_random_blocksize(); 5000219089Spjd blocksize = MIN(blocksize, 2048); /* because we write so many */ 5001219089Spjd 5002219089Spjd ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0); 5003219089Spjd 5004219089Spjd if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0) 5005219089Spjd return; 5006219089Spjd 5007219089Spjd /* 5008219089Spjd * Take the name lock as writer to prevent anyone else from changing 5009219089Spjd * the pool and dataset properies we need to maintain during this test. 5010219089Spjd */ 5011236143Smm (void) rw_wrlock(&ztest_name_lock); 5012219089Spjd 5013219089Spjd if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum, 5014219089Spjd B_FALSE) != 0 || 5015219089Spjd ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1, 5016219089Spjd B_FALSE) != 0) { 5017236143Smm (void) rw_unlock(&ztest_name_lock); 5018219089Spjd return; 5019219089Spjd } 5020219089Spjd 5021219089Spjd object = od[0].od_object; 5022219089Spjd blocksize = od[0].od_blocksize; 5023228103Smm pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os); 5024219089Spjd 5025219089Spjd ASSERT(object != 0); 5026219089Spjd 5027219089Spjd tx = dmu_tx_create(os); 5028219089Spjd dmu_tx_hold_write(tx, object, 0, copies * blocksize); 5029219089Spjd txg = ztest_tx_assign(tx, TXG_WAIT, FTAG); 5030219089Spjd if (txg == 0) { 5031236143Smm (void) rw_unlock(&ztest_name_lock); 5032219089Spjd return; 5033219089Spjd } 5034219089Spjd 5035219089Spjd /* 5036219089Spjd * Write all the copies of our block. 5037219089Spjd */ 5038219089Spjd for (int i = 0; i < copies; i++) { 5039219089Spjd uint64_t offset = i * blocksize; 5040248571Smm int error = dmu_buf_hold(os, object, offset, FTAG, &db, 5041248571Smm DMU_READ_NO_PREFETCH); 5042248571Smm if (error != 0) { 5043248571Smm fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u", 5044248571Smm os, (long long)object, (long long) offset, error); 5045248571Smm } 5046219089Spjd ASSERT(db->db_offset == offset); 5047219089Spjd ASSERT(db->db_size == blocksize); 5048219089Spjd ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) || 5049219089Spjd ztest_pattern_match(db->db_data, db->db_size, 0ULL)); 5050219089Spjd dmu_buf_will_fill(db, tx); 5051219089Spjd ztest_pattern_set(db->db_data, db->db_size, pattern); 5052219089Spjd dmu_buf_rele(db, FTAG); 5053219089Spjd } 5054219089Spjd 5055219089Spjd dmu_tx_commit(tx); 5056219089Spjd txg_wait_synced(spa_get_dsl(spa), txg); 5057219089Spjd 5058219089Spjd /* 5059219089Spjd * Find out what block we got. 5060219089Spjd */ 5061243524Smm VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db, 5062243524Smm DMU_READ_NO_PREFETCH)); 5063219089Spjd blk = *((dmu_buf_impl_t *)db)->db_blkptr; 5064219089Spjd dmu_buf_rele(db, FTAG); 5065219089Spjd 5066219089Spjd /* 5067219089Spjd * Damage the block. Dedup-ditto will save us when we read it later. 5068219089Spjd */ 5069219089Spjd psize = BP_GET_PSIZE(&blk); 5070219089Spjd buf = zio_buf_alloc(psize); 5071219089Spjd ztest_pattern_set(buf, psize, ~pattern); 5072219089Spjd 5073219089Spjd (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk, 5074219089Spjd buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE, 5075219089Spjd ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL)); 5076219089Spjd 5077219089Spjd zio_buf_free(buf, psize); 5078219089Spjd 5079236143Smm (void) rw_unlock(&ztest_name_lock); 5080219089Spjd} 5081219089Spjd 5082219089Spjd/* 5083168404Spjd * Scrub the pool. 5084168404Spjd */ 5085219089Spjd/* ARGSUSED */ 5086168404Spjdvoid 5087219089Spjdztest_scrub(ztest_ds_t *zd, uint64_t id) 5088168404Spjd{ 5089236143Smm spa_t *spa = ztest_spa; 5090168404Spjd 5091219089Spjd (void) spa_scan(spa, POOL_SCAN_SCRUB); 5092219089Spjd (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */ 5093219089Spjd (void) spa_scan(spa, POOL_SCAN_SCRUB); 5094168404Spjd} 5095168404Spjd 5096168404Spjd/* 5097228103Smm * Change the guid for the pool. 5098228103Smm */ 5099228103Smm/* ARGSUSED */ 5100228103Smmvoid 5101228103Smmztest_reguid(ztest_ds_t *zd, uint64_t id) 5102228103Smm{ 5103236143Smm spa_t *spa = ztest_spa; 5104228103Smm uint64_t orig, load; 5105239620Smm int error; 5106228103Smm 5107228103Smm orig = spa_guid(spa); 5108228103Smm load = spa_load_guid(spa); 5109239620Smm 5110239620Smm (void) rw_wrlock(&ztest_name_lock); 5111239620Smm error = spa_change_guid(spa); 5112239620Smm (void) rw_unlock(&ztest_name_lock); 5113239620Smm 5114239620Smm if (error != 0) 5115228103Smm return; 5116228103Smm 5117243505Smm if (ztest_opts.zo_verbose >= 4) { 5118228103Smm (void) printf("Changed guid old %llu -> %llu\n", 5119228103Smm (u_longlong_t)orig, (u_longlong_t)spa_guid(spa)); 5120228103Smm } 5121228103Smm 5122228103Smm VERIFY3U(orig, !=, spa_guid(spa)); 5123228103Smm VERIFY3U(load, ==, spa_load_guid(spa)); 5124228103Smm} 5125228103Smm 5126228103Smm/* 5127168404Spjd * Rename the pool to a different name and then rename it back. 5128168404Spjd */ 5129219089Spjd/* ARGSUSED */ 5130168404Spjdvoid 5131219089Spjdztest_spa_rename(ztest_ds_t *zd, uint64_t id) 5132168404Spjd{ 5133168404Spjd char *oldname, *newname; 5134168404Spjd spa_t *spa; 5135168404Spjd 5136236143Smm (void) rw_wrlock(&ztest_name_lock); 5137168404Spjd 5138236143Smm oldname = ztest_opts.zo_pool; 5139168404Spjd newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); 5140168404Spjd (void) strcpy(newname, oldname); 5141168404Spjd (void) strcat(newname, "_tmp"); 5142168404Spjd 5143168404Spjd /* 5144168404Spjd * Do the rename 5145168404Spjd */ 5146219089Spjd VERIFY3U(0, ==, spa_rename(oldname, newname)); 5147168404Spjd 5148168404Spjd /* 5149168404Spjd * Try to open it under the old name, which shouldn't exist 5150168404Spjd */ 5151219089Spjd VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); 5152168404Spjd 5153168404Spjd /* 5154168404Spjd * Open it under the new name and make sure it's still the same spa_t. 5155168404Spjd */ 5156219089Spjd VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); 5157168404Spjd 5158236143Smm ASSERT(spa == ztest_spa); 5159168404Spjd spa_close(spa, FTAG); 5160168404Spjd 5161168404Spjd /* 5162168404Spjd * Rename it back to the original 5163168404Spjd */ 5164219089Spjd VERIFY3U(0, ==, spa_rename(newname, oldname)); 5165168404Spjd 5166168404Spjd /* 5167168404Spjd * Make sure it can still be opened 5168168404Spjd */ 5169219089Spjd VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); 5170168404Spjd 5171236143Smm ASSERT(spa == ztest_spa); 5172168404Spjd spa_close(spa, FTAG); 5173168404Spjd 5174168404Spjd umem_free(newname, strlen(newname) + 1); 5175168404Spjd 5176236143Smm (void) rw_unlock(&ztest_name_lock); 5177168404Spjd} 5178168404Spjd 5179168404Spjd/* 5180219089Spjd * Verify pool integrity by running zdb. 5181168404Spjd */ 5182168404Spjdstatic void 5183219089Spjdztest_run_zdb(char *pool) 5184168404Spjd{ 5185168404Spjd int status; 5186168404Spjd char zdb[MAXPATHLEN + MAXNAMELEN + 20]; 5187168404Spjd char zbuf[1024]; 5188168404Spjd char *bin; 5189185029Spjd char *ztest; 5190185029Spjd char *isa; 5191185029Spjd int isalen; 5192168404Spjd FILE *fp; 5193168404Spjd 5194214623Spjd strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb)); 5195168404Spjd 5196168404Spjd /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ 5197168404Spjd bin = strstr(zdb, "/usr/bin/"); 5198185029Spjd ztest = strstr(bin, "/ztest"); 5199185029Spjd isa = bin + 8; 5200185029Spjd isalen = ztest - isa; 5201185029Spjd isa = strdup(isa); 5202168404Spjd /* LINTED */ 5203185029Spjd (void) sprintf(bin, 5204268075Sdelphij "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s", 5205185029Spjd isalen, 5206185029Spjd isa, 5207236143Smm ztest_opts.zo_verbose >= 3 ? "s" : "", 5208236143Smm ztest_opts.zo_verbose >= 4 ? "v" : "", 5209219089Spjd spa_config_path, 5210208047Smm pool); 5211185029Spjd free(isa); 5212168404Spjd 5213236143Smm if (ztest_opts.zo_verbose >= 5) 5214168404Spjd (void) printf("Executing %s\n", strstr(zdb, "zdb ")); 5215168404Spjd 5216168404Spjd fp = popen(zdb, "r"); 5217168404Spjd assert(fp != NULL); 5218168404Spjd 5219168404Spjd while (fgets(zbuf, sizeof (zbuf), fp) != NULL) 5220236143Smm if (ztest_opts.zo_verbose >= 3) 5221168404Spjd (void) printf("%s", zbuf); 5222168404Spjd 5223168404Spjd status = pclose(fp); 5224168404Spjd 5225168404Spjd if (status == 0) 5226168404Spjd return; 5227168404Spjd 5228168404Spjd ztest_dump_core = 0; 5229168404Spjd if (WIFEXITED(status)) 5230168404Spjd fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); 5231168404Spjd else 5232168404Spjd fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); 5233168404Spjd} 5234168404Spjd 5235168404Spjdstatic void 5236168404Spjdztest_walk_pool_directory(char *header) 5237168404Spjd{ 5238168404Spjd spa_t *spa = NULL; 5239168404Spjd 5240236143Smm if (ztest_opts.zo_verbose >= 6) 5241168404Spjd (void) printf("%s\n", header); 5242168404Spjd 5243168404Spjd mutex_enter(&spa_namespace_lock); 5244168404Spjd while ((spa = spa_next(spa)) != NULL) 5245236143Smm if (ztest_opts.zo_verbose >= 6) 5246168404Spjd (void) printf("\t%s\n", spa_name(spa)); 5247168404Spjd mutex_exit(&spa_namespace_lock); 5248168404Spjd} 5249168404Spjd 5250168404Spjdstatic void 5251168404Spjdztest_spa_import_export(char *oldname, char *newname) 5252168404Spjd{ 5253209962Smm nvlist_t *config, *newconfig; 5254168404Spjd uint64_t pool_guid; 5255168404Spjd spa_t *spa; 5256248571Smm int error; 5257168404Spjd 5258236143Smm if (ztest_opts.zo_verbose >= 4) { 5259168404Spjd (void) printf("import/export: old = %s, new = %s\n", 5260168404Spjd oldname, newname); 5261168404Spjd } 5262168404Spjd 5263168404Spjd /* 5264168404Spjd * Clean up from previous runs. 5265168404Spjd */ 5266168404Spjd (void) spa_destroy(newname); 5267168404Spjd 5268168404Spjd /* 5269168404Spjd * Get the pool's configuration and guid. 5270168404Spjd */ 5271219089Spjd VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG)); 5272168404Spjd 5273209962Smm /* 5274209962Smm * Kick off a scrub to tickle scrub/export races. 5275209962Smm */ 5276209962Smm if (ztest_random(2) == 0) 5277219089Spjd (void) spa_scan(spa, POOL_SCAN_SCRUB); 5278209962Smm 5279168404Spjd pool_guid = spa_guid(spa); 5280168404Spjd spa_close(spa, FTAG); 5281168404Spjd 5282168404Spjd ztest_walk_pool_directory("pools before export"); 5283168404Spjd 5284168404Spjd /* 5285168404Spjd * Export it. 5286168404Spjd */ 5287219089Spjd VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE)); 5288168404Spjd 5289168404Spjd ztest_walk_pool_directory("pools after export"); 5290168404Spjd 5291168404Spjd /* 5292209962Smm * Try to import it. 5293209962Smm */ 5294209962Smm newconfig = spa_tryimport(config); 5295209962Smm ASSERT(newconfig != NULL); 5296209962Smm nvlist_free(newconfig); 5297209962Smm 5298209962Smm /* 5299168404Spjd * Import it under the new name. 5300168404Spjd */ 5301248571Smm error = spa_import(newname, config, NULL, 0); 5302248571Smm if (error != 0) { 5303248571Smm dump_nvlist(config, 0); 5304248571Smm fatal(B_FALSE, "couldn't import pool %s as %s: error %u", 5305248571Smm oldname, newname, error); 5306248571Smm } 5307168404Spjd 5308168404Spjd ztest_walk_pool_directory("pools after import"); 5309168404Spjd 5310168404Spjd /* 5311168404Spjd * Try to import it again -- should fail with EEXIST. 5312168404Spjd */ 5313219089Spjd VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0)); 5314168404Spjd 5315168404Spjd /* 5316168404Spjd * Try to import it under a different name -- should fail with EEXIST. 5317168404Spjd */ 5318219089Spjd VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0)); 5319168404Spjd 5320168404Spjd /* 5321168404Spjd * Verify that the pool is no longer visible under the old name. 5322168404Spjd */ 5323219089Spjd VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG)); 5324168404Spjd 5325168404Spjd /* 5326168404Spjd * Verify that we can open and close the pool using the new name. 5327168404Spjd */ 5328219089Spjd VERIFY3U(0, ==, spa_open(newname, &spa, FTAG)); 5329168404Spjd ASSERT(pool_guid == spa_guid(spa)); 5330168404Spjd spa_close(spa, FTAG); 5331168404Spjd 5332168404Spjd nvlist_free(config); 5333168404Spjd} 5334168404Spjd 5335209962Smmstatic void 5336209962Smmztest_resume(spa_t *spa) 5337209962Smm{ 5338236143Smm if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6) 5339219089Spjd (void) printf("resuming from suspended state\n"); 5340219089Spjd spa_vdev_state_enter(spa, SCL_NONE); 5341219089Spjd vdev_clear(spa, NULL); 5342219089Spjd (void) spa_vdev_state_exit(spa, NULL, 0); 5343219089Spjd (void) zio_resume(spa); 5344209962Smm} 5345209962Smm 5346168404Spjdstatic void * 5347209962Smmztest_resume_thread(void *arg) 5348185029Spjd{ 5349185029Spjd spa_t *spa = arg; 5350185029Spjd 5351185029Spjd while (!ztest_exiting) { 5352219089Spjd if (spa_suspended(spa)) 5353219089Spjd ztest_resume(spa); 5354219089Spjd (void) poll(NULL, 0, 100); 5355185029Spjd } 5356185029Spjd return (NULL); 5357185029Spjd} 5358185029Spjd 5359185029Spjdstatic void * 5360219089Spjdztest_deadman_thread(void *arg) 5361219089Spjd{ 5362219089Spjd ztest_shared_t *zs = arg; 5363254074Sdelphij spa_t *spa = ztest_spa; 5364254074Sdelphij hrtime_t delta, total = 0; 5365219089Spjd 5366254074Sdelphij for (;;) { 5367258632Savg delta = zs->zs_thread_stop - zs->zs_thread_start + 5368258632Savg MSEC2NSEC(zfs_deadman_synctime_ms); 5369219089Spjd 5370258632Savg (void) poll(NULL, 0, (int)NSEC2MSEC(delta)); 5371219089Spjd 5372254074Sdelphij /* 5373254074Sdelphij * If the pool is suspended then fail immediately. Otherwise, 5374254074Sdelphij * check to see if the pool is making any progress. If 5375254074Sdelphij * vdev_deadman() discovers that there hasn't been any recent 5376254074Sdelphij * I/Os then it will end up aborting the tests. 5377254074Sdelphij */ 5378258717Savg if (spa_suspended(spa) || spa->spa_root_vdev == NULL) { 5379254074Sdelphij fatal(0, "aborting test after %llu seconds because " 5380254074Sdelphij "pool has transitioned to a suspended state.", 5381258632Savg zfs_deadman_synctime_ms / 1000); 5382254074Sdelphij return (NULL); 5383254074Sdelphij } 5384254074Sdelphij vdev_deadman(spa->spa_root_vdev); 5385219089Spjd 5386258632Savg total += zfs_deadman_synctime_ms/1000; 5387254074Sdelphij (void) printf("ztest has been running for %lld seconds\n", 5388254074Sdelphij total); 5389254074Sdelphij } 5390219089Spjd} 5391219089Spjd 5392219089Spjdstatic void 5393236143Smmztest_execute(int test, ztest_info_t *zi, uint64_t id) 5394219089Spjd{ 5395236143Smm ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets]; 5396236143Smm ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test); 5397219089Spjd hrtime_t functime = gethrtime(); 5398219089Spjd 5399219089Spjd for (int i = 0; i < zi->zi_iters; i++) 5400219089Spjd zi->zi_func(zd, id); 5401219089Spjd 5402219089Spjd functime = gethrtime() - functime; 5403219089Spjd 5404236143Smm atomic_add_64(&zc->zc_count, 1); 5405236143Smm atomic_add_64(&zc->zc_time, functime); 5406219089Spjd 5407236143Smm if (ztest_opts.zo_verbose >= 4) { 5408219089Spjd Dl_info dli; 5409219089Spjd (void) dladdr((void *)zi->zi_func, &dli); 5410219089Spjd (void) printf("%6.2f sec in %s\n", 5411219089Spjd (double)functime / NANOSEC, dli.dli_sname); 5412219089Spjd } 5413219089Spjd} 5414219089Spjd 5415219089Spjdstatic void * 5416168404Spjdztest_thread(void *arg) 5417168404Spjd{ 5418236143Smm int rand; 5419219089Spjd uint64_t id = (uintptr_t)arg; 5420168404Spjd ztest_shared_t *zs = ztest_shared; 5421219089Spjd uint64_t call_next; 5422219089Spjd hrtime_t now; 5423168404Spjd ztest_info_t *zi; 5424236143Smm ztest_shared_callstate_t *zc; 5425168404Spjd 5426219089Spjd while ((now = gethrtime()) < zs->zs_thread_stop) { 5427168404Spjd /* 5428168404Spjd * See if it's time to force a crash. 5429168404Spjd */ 5430219089Spjd if (now > zs->zs_thread_kill) 5431219089Spjd ztest_kill(zs); 5432168404Spjd 5433168404Spjd /* 5434219089Spjd * If we're getting ENOSPC with some regularity, stop. 5435168404Spjd */ 5436219089Spjd if (zs->zs_enospc_count > 10) 5437219089Spjd break; 5438168404Spjd 5439168404Spjd /* 5440219089Spjd * Pick a random function to execute. 5441168404Spjd */ 5442236143Smm rand = ztest_random(ZTEST_FUNCS); 5443236143Smm zi = &ztest_info[rand]; 5444236143Smm zc = ZTEST_GET_SHARED_CALLSTATE(rand); 5445236143Smm call_next = zc->zc_next; 5446168404Spjd 5447219089Spjd if (now >= call_next && 5448236143Smm atomic_cas_64(&zc->zc_next, call_next, call_next + 5449236143Smm ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) { 5450236143Smm ztest_execute(rand, zi, id); 5451236143Smm } 5452219089Spjd } 5453168404Spjd 5454219089Spjd return (NULL); 5455219089Spjd} 5456168404Spjd 5457219089Spjdstatic void 5458219089Spjdztest_dataset_name(char *dsname, char *pool, int d) 5459219089Spjd{ 5460219089Spjd (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d); 5461219089Spjd} 5462168404Spjd 5463219089Spjdstatic void 5464236143Smmztest_dataset_destroy(int d) 5465219089Spjd{ 5466219089Spjd char name[MAXNAMELEN]; 5467168404Spjd 5468236143Smm ztest_dataset_name(name, ztest_opts.zo_pool, d); 5469168404Spjd 5470236143Smm if (ztest_opts.zo_verbose >= 3) 5471219089Spjd (void) printf("Destroying %s to free up space\n", name); 5472168404Spjd 5473219089Spjd /* 5474219089Spjd * Cleanup any non-standard clones and snapshots. In general, 5475219089Spjd * ztest thread t operates on dataset (t % zopt_datasets), 5476219089Spjd * so there may be more than one thing to clean up. 5477219089Spjd */ 5478236143Smm for (int t = d; t < ztest_opts.zo_threads; 5479236143Smm t += ztest_opts.zo_datasets) { 5480219089Spjd ztest_dsl_dataset_cleanup(name, t); 5481236143Smm } 5482219089Spjd 5483219089Spjd (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL, 5484219089Spjd DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); 5485219089Spjd} 5486219089Spjd 5487219089Spjdstatic void 5488219089Spjdztest_dataset_dirobj_verify(ztest_ds_t *zd) 5489219089Spjd{ 5490219089Spjd uint64_t usedobjs, dirobjs, scratch; 5491219089Spjd 5492219089Spjd /* 5493219089Spjd * ZTEST_DIROBJ is the object directory for the entire dataset. 5494219089Spjd * Therefore, the number of objects in use should equal the 5495219089Spjd * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself. 5496219089Spjd * If not, we have an object leak. 5497219089Spjd * 5498219089Spjd * Note that we can only check this in ztest_dataset_open(), 5499219089Spjd * when the open-context and syncing-context values agree. 5500219089Spjd * That's because zap_count() returns the open-context value, 5501219089Spjd * while dmu_objset_space() returns the rootbp fill count. 5502219089Spjd */ 5503219089Spjd VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs)); 5504219089Spjd dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch); 5505219089Spjd ASSERT3U(dirobjs + 1, ==, usedobjs); 5506219089Spjd} 5507219089Spjd 5508219089Spjdstatic int 5509236143Smmztest_dataset_open(int d) 5510219089Spjd{ 5511236143Smm ztest_ds_t *zd = &ztest_ds[d]; 5512236143Smm uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq; 5513219089Spjd objset_t *os; 5514219089Spjd zilog_t *zilog; 5515219089Spjd char name[MAXNAMELEN]; 5516219089Spjd int error; 5517219089Spjd 5518236143Smm ztest_dataset_name(name, ztest_opts.zo_pool, d); 5519219089Spjd 5520236143Smm (void) rw_rdlock(&ztest_name_lock); 5521219089Spjd 5522219089Spjd error = ztest_dataset_create(name); 5523219089Spjd if (error == ENOSPC) { 5524236143Smm (void) rw_unlock(&ztest_name_lock); 5525219089Spjd ztest_record_enospc(FTAG); 5526219089Spjd return (error); 5527168404Spjd } 5528219089Spjd ASSERT(error == 0 || error == EEXIST); 5529168404Spjd 5530248571Smm VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os)); 5531236143Smm (void) rw_unlock(&ztest_name_lock); 5532219089Spjd 5533236143Smm ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os); 5534219089Spjd 5535219089Spjd zilog = zd->zd_zilog; 5536219089Spjd 5537219089Spjd if (zilog->zl_header->zh_claim_lr_seq != 0 && 5538219089Spjd zilog->zl_header->zh_claim_lr_seq < committed_seq) 5539219089Spjd fatal(0, "missing log records: claimed %llu < committed %llu", 5540219089Spjd zilog->zl_header->zh_claim_lr_seq, committed_seq); 5541219089Spjd 5542219089Spjd ztest_dataset_dirobj_verify(zd); 5543219089Spjd 5544219089Spjd zil_replay(os, zd, ztest_replay_vector); 5545219089Spjd 5546219089Spjd ztest_dataset_dirobj_verify(zd); 5547219089Spjd 5548236143Smm if (ztest_opts.zo_verbose >= 6) 5549219089Spjd (void) printf("%s replay %llu blocks, %llu records, seq %llu\n", 5550219089Spjd zd->zd_name, 5551219089Spjd (u_longlong_t)zilog->zl_parse_blk_count, 5552219089Spjd (u_longlong_t)zilog->zl_parse_lr_count, 5553219089Spjd (u_longlong_t)zilog->zl_replaying_seq); 5554219089Spjd 5555219089Spjd zilog = zil_open(os, ztest_get_data); 5556219089Spjd 5557219089Spjd if (zilog->zl_replaying_seq != 0 && 5558219089Spjd zilog->zl_replaying_seq < committed_seq) 5559219089Spjd fatal(0, "missing log records: replayed %llu < committed %llu", 5560219089Spjd zilog->zl_replaying_seq, committed_seq); 5561219089Spjd 5562219089Spjd return (0); 5563168404Spjd} 5564168404Spjd 5565219089Spjdstatic void 5566236143Smmztest_dataset_close(int d) 5567219089Spjd{ 5568236143Smm ztest_ds_t *zd = &ztest_ds[d]; 5569219089Spjd 5570219089Spjd zil_close(zd->zd_zilog); 5571248571Smm dmu_objset_disown(zd->zd_os, zd); 5572219089Spjd 5573219089Spjd ztest_zd_fini(zd); 5574219089Spjd} 5575219089Spjd 5576168404Spjd/* 5577168404Spjd * Kick off threads to run tests on all datasets in parallel. 5578168404Spjd */ 5579168404Spjdstatic void 5580219089Spjdztest_run(ztest_shared_t *zs) 5581168404Spjd{ 5582219089Spjd thread_t *tid; 5583168404Spjd spa_t *spa; 5584228103Smm objset_t *os; 5585185029Spjd thread_t resume_tid; 5586219089Spjd int error; 5587168404Spjd 5588185029Spjd ztest_exiting = B_FALSE; 5589185029Spjd 5590168404Spjd /* 5591219089Spjd * Initialize parent/child shared state. 5592168404Spjd */ 5593236143Smm VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0); 5594236143Smm VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0); 5595168404Spjd 5596219089Spjd zs->zs_thread_start = gethrtime(); 5597236143Smm zs->zs_thread_stop = 5598236143Smm zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC; 5599219089Spjd zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop); 5600219089Spjd zs->zs_thread_kill = zs->zs_thread_stop; 5601236143Smm if (ztest_random(100) < ztest_opts.zo_killrate) { 5602236143Smm zs->zs_thread_kill -= 5603236143Smm ztest_random(ztest_opts.zo_passtime * NANOSEC); 5604236143Smm } 5605168404Spjd 5606219089Spjd (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL); 5607168404Spjd 5608219089Spjd list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t), 5609219089Spjd offsetof(ztest_cb_data_t, zcd_node)); 5610168404Spjd 5611168404Spjd /* 5612219089Spjd * Open our pool. 5613168404Spjd */ 5614219089Spjd kernel_init(FREAD | FWRITE); 5615248571Smm VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5616224177Smm spa->spa_debug = B_TRUE; 5617268086Sdelphij metaslab_preload_limit = ztest_random(20) + 1; 5618236143Smm ztest_spa = spa; 5619168404Spjd 5620248571Smm VERIFY0(dmu_objset_own(ztest_opts.zo_pool, 5621248571Smm DMU_OST_ANY, B_TRUE, FTAG, &os)); 5622228103Smm zs->zs_guid = dmu_objset_fsid_guid(os); 5623248571Smm dmu_objset_disown(os, FTAG); 5624228103Smm 5625219089Spjd spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN; 5626168404Spjd 5627168404Spjd /* 5628209962Smm * We don't expect the pool to suspend unless maxfaults == 0, 5629209962Smm * in which case ztest_fault_inject() temporarily takes away 5630209962Smm * the only valid replica. 5631209962Smm */ 5632219089Spjd if (MAXFAULTS() == 0) 5633209962Smm spa->spa_failmode = ZIO_FAILURE_MODE_WAIT; 5634209962Smm else 5635209962Smm spa->spa_failmode = ZIO_FAILURE_MODE_PANIC; 5636209962Smm 5637209962Smm /* 5638185029Spjd * Create a thread to periodically resume suspended I/O. 5639185029Spjd */ 5640209962Smm VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND, 5641185029Spjd &resume_tid) == 0); 5642185029Spjd 5643185029Spjd /* 5644219089Spjd * Create a deadman thread to abort() if we hang. 5645219089Spjd */ 5646219089Spjd VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND, 5647219089Spjd NULL) == 0); 5648219089Spjd 5649219089Spjd /* 5650168404Spjd * Verify that we can safely inquire about about any object, 5651168404Spjd * whether it's allocated or not. To make it interesting, 5652168404Spjd * we probe a 5-wide window around each power of two. 5653168404Spjd * This hits all edge cases, including zero and the max. 5654168404Spjd */ 5655219089Spjd for (int t = 0; t < 64; t++) { 5656219089Spjd for (int d = -5; d <= 5; d++) { 5657168404Spjd error = dmu_object_info(spa->spa_meta_objset, 5658168404Spjd (1ULL << t) + d, NULL); 5659168404Spjd ASSERT(error == 0 || error == ENOENT || 5660168404Spjd error == EINVAL); 5661168404Spjd } 5662168404Spjd } 5663168404Spjd 5664168404Spjd /* 5665219089Spjd * If we got any ENOSPC errors on the previous run, destroy something. 5666168404Spjd */ 5667219089Spjd if (zs->zs_enospc_count != 0) { 5668236143Smm int d = ztest_random(ztest_opts.zo_datasets); 5669236143Smm ztest_dataset_destroy(d); 5670219089Spjd } 5671168404Spjd zs->zs_enospc_count = 0; 5672168404Spjd 5673236143Smm tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t), 5674236143Smm UMEM_NOFAIL); 5675168404Spjd 5676236143Smm if (ztest_opts.zo_verbose >= 4) 5677168404Spjd (void) printf("starting main threads...\n"); 5678168404Spjd 5679219089Spjd /* 5680219089Spjd * Kick off all the tests that run in parallel. 5681219089Spjd */ 5682236143Smm for (int t = 0; t < ztest_opts.zo_threads; t++) { 5683236143Smm if (t < ztest_opts.zo_datasets && 5684236143Smm ztest_dataset_open(t) != 0) 5685219089Spjd return; 5686219089Spjd VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t, 5687219089Spjd THR_BOUND, &tid[t]) == 0); 5688219089Spjd } 5689168404Spjd 5690219089Spjd /* 5691219089Spjd * Wait for all of the tests to complete. We go in reverse order 5692219089Spjd * so we don't close datasets while threads are still using them. 5693219089Spjd */ 5694236143Smm for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) { 5695219089Spjd VERIFY(thr_join(tid[t], NULL, NULL) == 0); 5696236143Smm if (t < ztest_opts.zo_datasets) 5697236143Smm ztest_dataset_close(t); 5698219089Spjd } 5699185029Spjd 5700219089Spjd txg_wait_synced(spa_get_dsl(spa), 0); 5701185029Spjd 5702219089Spjd zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 5703219089Spjd zs->zs_space = metaslab_class_get_space(spa_normal_class(spa)); 5704254112Sdelphij zfs_dbgmsg_print(FTAG); 5705168404Spjd 5706236143Smm umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t)); 5707168404Spjd 5708219089Spjd /* Kill the resume thread */ 5709219089Spjd ztest_exiting = B_TRUE; 5710219089Spjd VERIFY(thr_join(resume_tid, NULL, NULL) == 0); 5711219089Spjd ztest_resume(spa); 5712219089Spjd 5713219089Spjd /* 5714219089Spjd * Right before closing the pool, kick off a bunch of async I/O; 5715219089Spjd * spa_close() should wait for it to complete. 5716219089Spjd */ 5717219089Spjd for (uint64_t object = 1; object < 50; object++) 5718219089Spjd dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20); 5719219089Spjd 5720219089Spjd spa_close(spa, FTAG); 5721219089Spjd 5722219089Spjd /* 5723219089Spjd * Verify that we can loop over all pools. 5724219089Spjd */ 5725219089Spjd mutex_enter(&spa_namespace_lock); 5726219089Spjd for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) 5727236143Smm if (ztest_opts.zo_verbose > 3) 5728219089Spjd (void) printf("spa_next: found %s\n", spa_name(spa)); 5729219089Spjd mutex_exit(&spa_namespace_lock); 5730219089Spjd 5731219089Spjd /* 5732219089Spjd * Verify that we can export the pool and reimport it under a 5733219089Spjd * different name. 5734219089Spjd */ 5735219089Spjd if (ztest_random(2) == 0) { 5736219089Spjd char name[MAXNAMELEN]; 5737236143Smm (void) snprintf(name, MAXNAMELEN, "%s_import", 5738236143Smm ztest_opts.zo_pool); 5739236143Smm ztest_spa_import_export(ztest_opts.zo_pool, name); 5740236143Smm ztest_spa_import_export(name, ztest_opts.zo_pool); 5741168404Spjd } 5742168404Spjd 5743219089Spjd kernel_fini(); 5744219089Spjd 5745219089Spjd list_destroy(&zcl.zcl_callbacks); 5746219089Spjd 5747219089Spjd (void) _mutex_destroy(&zcl.zcl_callbacks_lock); 5748219089Spjd 5749236143Smm (void) rwlock_destroy(&ztest_name_lock); 5750236143Smm (void) _mutex_destroy(&ztest_vdev_lock); 5751219089Spjd} 5752219089Spjd 5753219089Spjdstatic void 5754236143Smmztest_freeze(void) 5755219089Spjd{ 5756236143Smm ztest_ds_t *zd = &ztest_ds[0]; 5757219089Spjd spa_t *spa; 5758219089Spjd int numloops = 0; 5759219089Spjd 5760236143Smm if (ztest_opts.zo_verbose >= 3) 5761219089Spjd (void) printf("testing spa_freeze()...\n"); 5762168404Spjd 5763219089Spjd kernel_init(FREAD | FWRITE); 5764236143Smm VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5765236143Smm VERIFY3U(0, ==, ztest_dataset_open(0)); 5766243524Smm spa->spa_debug = B_TRUE; 5767243524Smm ztest_spa = spa; 5768168404Spjd 5769168404Spjd /* 5770219089Spjd * Force the first log block to be transactionally allocated. 5771219089Spjd * We have to do this before we freeze the pool -- otherwise 5772219089Spjd * the log chain won't be anchored. 5773168404Spjd */ 5774219089Spjd while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) { 5775219089Spjd ztest_dmu_object_alloc_free(zd, 0); 5776219089Spjd zil_commit(zd->zd_zilog, 0); 5777168404Spjd } 5778168404Spjd 5779168404Spjd txg_wait_synced(spa_get_dsl(spa), 0); 5780168404Spjd 5781219089Spjd /* 5782219089Spjd * Freeze the pool. This stops spa_sync() from doing anything, 5783219089Spjd * so that the only way to record changes from now on is the ZIL. 5784219089Spjd */ 5785219089Spjd spa_freeze(spa); 5786185029Spjd 5787219089Spjd /* 5788268855Sdelphij * Because it is hard to predict how much space a write will actually 5789268855Sdelphij * require beforehand, we leave ourselves some fudge space to write over 5790268855Sdelphij * capacity. 5791268855Sdelphij */ 5792268855Sdelphij uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2; 5793268855Sdelphij 5794268855Sdelphij /* 5795219089Spjd * Run tests that generate log records but don't alter the pool config 5796219089Spjd * or depend on DSL sync tasks (snapshots, objset create/destroy, etc). 5797219089Spjd * We do a txg_wait_synced() after each iteration to force the txg 5798219089Spjd * to increase well beyond the last synced value in the uberblock. 5799219089Spjd * The ZIL should be OK with that. 5800268855Sdelphij * 5801268855Sdelphij * Run a random number of times less than zo_maxloops and ensure we do 5802268855Sdelphij * not run out of space on the pool. 5803219089Spjd */ 5804236143Smm while (ztest_random(10) != 0 && 5805268855Sdelphij numloops++ < ztest_opts.zo_maxloops && 5806268855Sdelphij metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) { 5807268855Sdelphij ztest_od_t od; 5808268855Sdelphij ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0); 5809268855Sdelphij VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE)); 5810268855Sdelphij ztest_io(zd, od.od_object, 5811268855Sdelphij ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT); 5812219089Spjd txg_wait_synced(spa_get_dsl(spa), 0); 5813219089Spjd } 5814185029Spjd 5815168404Spjd /* 5816219089Spjd * Commit all of the changes we just generated. 5817168404Spjd */ 5818219089Spjd zil_commit(zd->zd_zilog, 0); 5819219089Spjd txg_wait_synced(spa_get_dsl(spa), 0); 5820168404Spjd 5821219089Spjd /* 5822219089Spjd * Close our dataset and close the pool. 5823219089Spjd */ 5824236143Smm ztest_dataset_close(0); 5825168404Spjd spa_close(spa, FTAG); 5826219089Spjd kernel_fini(); 5827168404Spjd 5828219089Spjd /* 5829219089Spjd * Open and close the pool and dataset to induce log replay. 5830219089Spjd */ 5831219089Spjd kernel_init(FREAD | FWRITE); 5832236143Smm VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5833239620Smm ASSERT(spa_freeze_txg(spa) == UINT64_MAX); 5834236143Smm VERIFY3U(0, ==, ztest_dataset_open(0)); 5835236143Smm ztest_dataset_close(0); 5836239620Smm 5837239620Smm spa->spa_debug = B_TRUE; 5838239620Smm ztest_spa = spa; 5839239620Smm txg_wait_synced(spa_get_dsl(spa), 0); 5840239620Smm ztest_reguid(NULL, 0); 5841239620Smm 5842219089Spjd spa_close(spa, FTAG); 5843168404Spjd kernel_fini(); 5844168404Spjd} 5845168404Spjd 5846168404Spjdvoid 5847168404Spjdprint_time(hrtime_t t, char *timebuf) 5848168404Spjd{ 5849168404Spjd hrtime_t s = t / NANOSEC; 5850168404Spjd hrtime_t m = s / 60; 5851168404Spjd hrtime_t h = m / 60; 5852168404Spjd hrtime_t d = h / 24; 5853168404Spjd 5854168404Spjd s -= m * 60; 5855168404Spjd m -= h * 60; 5856168404Spjd h -= d * 24; 5857168404Spjd 5858168404Spjd timebuf[0] = '\0'; 5859168404Spjd 5860168404Spjd if (d) 5861168404Spjd (void) sprintf(timebuf, 5862168404Spjd "%llud%02lluh%02llum%02llus", d, h, m, s); 5863168404Spjd else if (h) 5864168404Spjd (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); 5865168404Spjd else if (m) 5866168404Spjd (void) sprintf(timebuf, "%llum%02llus", m, s); 5867168404Spjd else 5868168404Spjd (void) sprintf(timebuf, "%llus", s); 5869168404Spjd} 5870168404Spjd 5871219089Spjdstatic nvlist_t * 5872219089Spjdmake_random_props() 5873219089Spjd{ 5874219089Spjd nvlist_t *props; 5875219089Spjd 5876236884Smm VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0); 5877219089Spjd if (ztest_random(2) == 0) 5878236884Smm return (props); 5879219089Spjd VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0); 5880219089Spjd 5881219089Spjd return (props); 5882219089Spjd} 5883219089Spjd 5884168404Spjd/* 5885168404Spjd * Create a storage pool with the given name and initial vdev size. 5886219089Spjd * Then test spa_freeze() functionality. 5887168404Spjd */ 5888168404Spjdstatic void 5889219089Spjdztest_init(ztest_shared_t *zs) 5890168404Spjd{ 5891168404Spjd spa_t *spa; 5892219089Spjd nvlist_t *nvroot, *props; 5893168404Spjd 5894236143Smm VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0); 5895236143Smm VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0); 5896219089Spjd 5897168404Spjd kernel_init(FREAD | FWRITE); 5898168404Spjd 5899168404Spjd /* 5900168404Spjd * Create the storage pool. 5901168404Spjd */ 5902236143Smm (void) spa_destroy(ztest_opts.zo_pool); 5903219089Spjd ztest_shared->zs_vdev_next_leaf = 0; 5904219089Spjd zs->zs_splits = 0; 5905236143Smm zs->zs_mirrors = ztest_opts.zo_mirrors; 5906243505Smm nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0, 5907236143Smm 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1); 5908219089Spjd props = make_random_props(); 5909236884Smm for (int i = 0; i < SPA_FEATURES; i++) { 5910236884Smm char buf[1024]; 5911236884Smm (void) snprintf(buf, sizeof (buf), "feature@%s", 5912236884Smm spa_feature_table[i].fi_uname); 5913236884Smm VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0)); 5914236884Smm } 5915248571Smm VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL)); 5916168404Spjd nvlist_free(nvroot); 5917168404Spjd 5918236143Smm VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG)); 5919236143Smm zs->zs_metaslab_sz = 5920236143Smm 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift; 5921236884Smm 5922219089Spjd spa_close(spa, FTAG); 5923209962Smm 5924219089Spjd kernel_fini(); 5925168404Spjd 5926236143Smm ztest_run_zdb(ztest_opts.zo_pool); 5927168404Spjd 5928236143Smm ztest_freeze(); 5929219089Spjd 5930236143Smm ztest_run_zdb(ztest_opts.zo_pool); 5931219089Spjd 5932236143Smm (void) rwlock_destroy(&ztest_name_lock); 5933236143Smm (void) _mutex_destroy(&ztest_vdev_lock); 5934168404Spjd} 5935168404Spjd 5936236143Smmstatic void 5937242845Sdelphijsetup_data_fd(void) 5938236143Smm{ 5939242845Sdelphij static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX"; 5940236143Smm 5941242845Sdelphij ztest_fd_data = mkstemp(ztest_name_data); 5942242845Sdelphij ASSERT3S(ztest_fd_data, >=, 0); 5943242845Sdelphij (void) unlink(ztest_name_data); 5944242845Sdelphij} 5945236143Smm 5946236143Smm 5947236884Smmstatic int 5948236884Smmshared_data_size(ztest_shared_hdr_t *hdr) 5949236884Smm{ 5950236884Smm int size; 5951236884Smm 5952236884Smm size = hdr->zh_hdr_size; 5953236884Smm size += hdr->zh_opts_size; 5954236884Smm size += hdr->zh_size; 5955236884Smm size += hdr->zh_stats_size * hdr->zh_stats_count; 5956236884Smm size += hdr->zh_ds_size * hdr->zh_ds_count; 5957236884Smm 5958236884Smm return (size); 5959236884Smm} 5960236884Smm 5961236143Smmstatic void 5962236143Smmsetup_hdr(void) 5963236143Smm{ 5964236884Smm int size; 5965236143Smm ztest_shared_hdr_t *hdr; 5966236143Smm 5967236143Smm hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), 5968242845Sdelphij PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); 5969236143Smm ASSERT(hdr != MAP_FAILED); 5970236143Smm 5971242845Sdelphij VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t))); 5972236884Smm 5973236143Smm hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t); 5974236143Smm hdr->zh_opts_size = sizeof (ztest_shared_opts_t); 5975236143Smm hdr->zh_size = sizeof (ztest_shared_t); 5976236143Smm hdr->zh_stats_size = sizeof (ztest_shared_callstate_t); 5977236143Smm hdr->zh_stats_count = ZTEST_FUNCS; 5978236143Smm hdr->zh_ds_size = sizeof (ztest_shared_ds_t); 5979236143Smm hdr->zh_ds_count = ztest_opts.zo_datasets; 5980236143Smm 5981236884Smm size = shared_data_size(hdr); 5982242845Sdelphij VERIFY3U(0, ==, ftruncate(ztest_fd_data, size)); 5983236884Smm 5984236143Smm (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); 5985236143Smm} 5986236143Smm 5987236143Smmstatic void 5988236143Smmsetup_data(void) 5989236143Smm{ 5990236143Smm int size, offset; 5991236143Smm ztest_shared_hdr_t *hdr; 5992236143Smm uint8_t *buf; 5993236143Smm 5994236143Smm hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()), 5995242845Sdelphij PROT_READ, MAP_SHARED, ztest_fd_data, 0); 5996236143Smm ASSERT(hdr != MAP_FAILED); 5997236143Smm 5998236884Smm size = shared_data_size(hdr); 5999236143Smm 6000236143Smm (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize())); 6001236143Smm hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()), 6002242845Sdelphij PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0); 6003236143Smm ASSERT(hdr != MAP_FAILED); 6004236143Smm buf = (uint8_t *)hdr; 6005236143Smm 6006236143Smm offset = hdr->zh_hdr_size; 6007236143Smm ztest_shared_opts = (void *)&buf[offset]; 6008236143Smm offset += hdr->zh_opts_size; 6009236143Smm ztest_shared = (void *)&buf[offset]; 6010236143Smm offset += hdr->zh_size; 6011236143Smm ztest_shared_callstate = (void *)&buf[offset]; 6012236143Smm offset += hdr->zh_stats_size * hdr->zh_stats_count; 6013236143Smm ztest_shared_ds = (void *)&buf[offset]; 6014236143Smm} 6015236143Smm 6016236143Smmstatic boolean_t 6017236143Smmexec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp) 6018236143Smm{ 6019236143Smm pid_t pid; 6020236143Smm int status; 6021242845Sdelphij char *cmdbuf = NULL; 6022236143Smm 6023236143Smm pid = fork(); 6024236143Smm 6025236143Smm if (cmd == NULL) { 6026242845Sdelphij cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL); 6027242845Sdelphij (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN); 6028236143Smm cmd = cmdbuf; 6029236143Smm } 6030236143Smm 6031236143Smm if (pid == -1) 6032236143Smm fatal(1, "fork failed"); 6033236143Smm 6034236143Smm if (pid == 0) { /* child */ 6035236143Smm char *emptyargv[2] = { cmd, NULL }; 6036242845Sdelphij char fd_data_str[12]; 6037236143Smm 6038236143Smm struct rlimit rl = { 1024, 1024 }; 6039236143Smm (void) setrlimit(RLIMIT_NOFILE, &rl); 6040242845Sdelphij 6041242845Sdelphij (void) close(ztest_fd_rand); 6042242845Sdelphij VERIFY3U(11, >=, 6043242845Sdelphij snprintf(fd_data_str, 12, "%d", ztest_fd_data)); 6044242845Sdelphij VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1)); 6045242845Sdelphij 6046236143Smm (void) enable_extended_FILE_stdio(-1, -1); 6047236143Smm if (libpath != NULL) 6048236143Smm VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1)); 6049236143Smm#ifdef illumos 6050236143Smm (void) execv(cmd, emptyargv); 6051236143Smm#else 6052236143Smm (void) execvp(cmd, emptyargv); 6053236143Smm#endif 6054236143Smm ztest_dump_core = B_FALSE; 6055236143Smm fatal(B_TRUE, "exec failed: %s", cmd); 6056236143Smm } 6057236143Smm 6058242845Sdelphij if (cmdbuf != NULL) { 6059242845Sdelphij umem_free(cmdbuf, MAXPATHLEN); 6060242845Sdelphij cmd = NULL; 6061242845Sdelphij } 6062242845Sdelphij 6063236143Smm while (waitpid(pid, &status, 0) != pid) 6064236143Smm continue; 6065236143Smm if (statusp != NULL) 6066236143Smm *statusp = status; 6067236143Smm 6068236143Smm if (WIFEXITED(status)) { 6069236143Smm if (WEXITSTATUS(status) != 0) { 6070236143Smm (void) fprintf(stderr, "child exited with code %d\n", 6071236143Smm WEXITSTATUS(status)); 6072236143Smm exit(2); 6073236143Smm } 6074236143Smm return (B_FALSE); 6075236143Smm } else if (WIFSIGNALED(status)) { 6076236143Smm if (!ignorekill || WTERMSIG(status) != SIGKILL) { 6077236143Smm (void) fprintf(stderr, "child died with signal %d\n", 6078236143Smm WTERMSIG(status)); 6079236143Smm exit(3); 6080236143Smm } 6081236143Smm return (B_TRUE); 6082236143Smm } else { 6083236143Smm (void) fprintf(stderr, "something strange happened to child\n"); 6084236143Smm exit(4); 6085236143Smm /* NOTREACHED */ 6086236143Smm } 6087236143Smm} 6088236143Smm 6089236143Smmstatic void 6090236143Smmztest_run_init(void) 6091236143Smm{ 6092236143Smm ztest_shared_t *zs = ztest_shared; 6093236143Smm 6094236143Smm ASSERT(ztest_opts.zo_init != 0); 6095236143Smm 6096236143Smm /* 6097236143Smm * Blow away any existing copy of zpool.cache 6098236143Smm */ 6099236143Smm (void) remove(spa_config_path); 6100236143Smm 6101236143Smm /* 6102236143Smm * Create and initialize our storage pool. 6103236143Smm */ 6104236143Smm for (int i = 1; i <= ztest_opts.zo_init; i++) { 6105236143Smm bzero(zs, sizeof (ztest_shared_t)); 6106236143Smm if (ztest_opts.zo_verbose >= 3 && 6107236143Smm ztest_opts.zo_init != 1) { 6108236143Smm (void) printf("ztest_init(), pass %d\n", i); 6109236143Smm } 6110236143Smm ztest_init(zs); 6111236143Smm } 6112236143Smm} 6113236143Smm 6114168404Spjdint 6115168404Spjdmain(int argc, char **argv) 6116168404Spjd{ 6117168404Spjd int kills = 0; 6118168404Spjd int iters = 0; 6119236143Smm int older = 0; 6120236143Smm int newer = 0; 6121168404Spjd ztest_shared_t *zs; 6122168404Spjd ztest_info_t *zi; 6123236143Smm ztest_shared_callstate_t *zc; 6124168404Spjd char timebuf[100]; 6125168404Spjd char numbuf[6]; 6126219089Spjd spa_t *spa; 6127242845Sdelphij char *cmd; 6128236143Smm boolean_t hasalt; 6129242845Sdelphij char *fd_data_str = getenv("ZTEST_FD_DATA"); 6130168404Spjd 6131168404Spjd (void) setvbuf(stdout, NULL, _IOLBF, 0); 6132168404Spjd 6133240133Smm dprintf_setup(&argc, argv); 6134258632Savg zfs_deadman_synctime_ms = 300000; 6135240133Smm 6136242845Sdelphij ztest_fd_rand = open("/dev/urandom", O_RDONLY); 6137242845Sdelphij ASSERT3S(ztest_fd_rand, >=, 0); 6138242845Sdelphij 6139242845Sdelphij if (!fd_data_str) { 6140236143Smm process_options(argc, argv); 6141168404Spjd 6142242845Sdelphij setup_data_fd(); 6143236143Smm setup_hdr(); 6144236143Smm setup_data(); 6145236143Smm bcopy(&ztest_opts, ztest_shared_opts, 6146236143Smm sizeof (*ztest_shared_opts)); 6147236143Smm } else { 6148242845Sdelphij ztest_fd_data = atoi(fd_data_str); 6149236143Smm setup_data(); 6150236143Smm bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts)); 6151236143Smm } 6152236143Smm ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count); 6153168404Spjd 6154219089Spjd /* Override location of zpool.cache */ 6155242845Sdelphij VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache", 6156242845Sdelphij ztest_opts.zo_dir), !=, -1); 6157219089Spjd 6158236143Smm ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t), 6159236143Smm UMEM_NOFAIL); 6160236143Smm zs = ztest_shared; 6161168404Spjd 6162242845Sdelphij if (fd_data_str) { 6163236143Smm metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang; 6164236143Smm metaslab_df_alloc_threshold = 6165236143Smm zs->zs_metaslab_df_alloc_threshold; 6166219089Spjd 6167236143Smm if (zs->zs_do_init) 6168236143Smm ztest_run_init(); 6169236143Smm else 6170236143Smm ztest_run(zs); 6171236143Smm exit(0); 6172236143Smm } 6173168404Spjd 6174236143Smm hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0); 6175236143Smm 6176236143Smm if (ztest_opts.zo_verbose >= 1) { 6177168404Spjd (void) printf("%llu vdevs, %d datasets, %d threads," 6178168404Spjd " %llu seconds...\n", 6179236143Smm (u_longlong_t)ztest_opts.zo_vdevs, 6180236143Smm ztest_opts.zo_datasets, 6181236143Smm ztest_opts.zo_threads, 6182236143Smm (u_longlong_t)ztest_opts.zo_time); 6183168404Spjd } 6184168404Spjd 6185242845Sdelphij cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL); 6186242845Sdelphij (void) strlcpy(cmd, getexecname(), MAXNAMELEN); 6187236143Smm 6188236143Smm zs->zs_do_init = B_TRUE; 6189236143Smm if (strlen(ztest_opts.zo_alt_ztest) != 0) { 6190236143Smm if (ztest_opts.zo_verbose >= 1) { 6191236143Smm (void) printf("Executing older ztest for " 6192236143Smm "initialization: %s\n", ztest_opts.zo_alt_ztest); 6193236143Smm } 6194236143Smm VERIFY(!exec_child(ztest_opts.zo_alt_ztest, 6195236143Smm ztest_opts.zo_alt_libpath, B_FALSE, NULL)); 6196236143Smm } else { 6197236143Smm VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL)); 6198168404Spjd } 6199236143Smm zs->zs_do_init = B_FALSE; 6200168404Spjd 6201219089Spjd zs->zs_proc_start = gethrtime(); 6202236143Smm zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC; 6203219089Spjd 6204219089Spjd for (int f = 0; f < ZTEST_FUNCS; f++) { 6205236143Smm zi = &ztest_info[f]; 6206236143Smm zc = ZTEST_GET_SHARED_CALLSTATE(f); 6207219089Spjd if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop) 6208236143Smm zc->zc_next = UINT64_MAX; 6209168404Spjd else 6210236143Smm zc->zc_next = zs->zs_proc_start + 6211219089Spjd ztest_random(2 * zi->zi_interval[0] + 1); 6212168404Spjd } 6213168404Spjd 6214168404Spjd /* 6215168404Spjd * Run the tests in a loop. These tests include fault injection 6216168404Spjd * to verify that self-healing data works, and forced crashes 6217168404Spjd * to verify that we never lose on-disk consistency. 6218168404Spjd */ 6219219089Spjd while (gethrtime() < zs->zs_proc_stop) { 6220168404Spjd int status; 6221236143Smm boolean_t killed; 6222168404Spjd 6223168404Spjd /* 6224168404Spjd * Initialize the workload counters for each function. 6225168404Spjd */ 6226219089Spjd for (int f = 0; f < ZTEST_FUNCS; f++) { 6227236143Smm zc = ZTEST_GET_SHARED_CALLSTATE(f); 6228236143Smm zc->zc_count = 0; 6229236143Smm zc->zc_time = 0; 6230168404Spjd } 6231168404Spjd 6232209962Smm /* Set the allocation switch size */ 6233236143Smm zs->zs_metaslab_df_alloc_threshold = 6234236143Smm ztest_random(zs->zs_metaslab_sz / 4) + 1; 6235209962Smm 6236236143Smm if (!hasalt || ztest_random(2) == 0) { 6237236143Smm if (hasalt && ztest_opts.zo_verbose >= 1) { 6238236143Smm (void) printf("Executing newer ztest: %s\n", 6239236143Smm cmd); 6240168404Spjd } 6241236143Smm newer++; 6242236143Smm killed = exec_child(cmd, NULL, B_TRUE, &status); 6243236143Smm } else { 6244236143Smm if (hasalt && ztest_opts.zo_verbose >= 1) { 6245236143Smm (void) printf("Executing older ztest: %s\n", 6246236143Smm ztest_opts.zo_alt_ztest); 6247168404Spjd } 6248236143Smm older++; 6249236143Smm killed = exec_child(ztest_opts.zo_alt_ztest, 6250236143Smm ztest_opts.zo_alt_libpath, B_TRUE, &status); 6251168404Spjd } 6252168404Spjd 6253236143Smm if (killed) 6254236143Smm kills++; 6255168404Spjd iters++; 6256168404Spjd 6257236143Smm if (ztest_opts.zo_verbose >= 1) { 6258168404Spjd hrtime_t now = gethrtime(); 6259168404Spjd 6260219089Spjd now = MIN(now, zs->zs_proc_stop); 6261219089Spjd print_time(zs->zs_proc_stop - now, timebuf); 6262168404Spjd nicenum(zs->zs_space, numbuf); 6263168404Spjd 6264168404Spjd (void) printf("Pass %3d, %8s, %3llu ENOSPC, " 6265168404Spjd "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", 6266168404Spjd iters, 6267168404Spjd WIFEXITED(status) ? "Complete" : "SIGKILL", 6268168404Spjd (u_longlong_t)zs->zs_enospc_count, 6269168404Spjd 100.0 * zs->zs_alloc / zs->zs_space, 6270168404Spjd numbuf, 6271219089Spjd 100.0 * (now - zs->zs_proc_start) / 6272236143Smm (ztest_opts.zo_time * NANOSEC), timebuf); 6273168404Spjd } 6274168404Spjd 6275236143Smm if (ztest_opts.zo_verbose >= 2) { 6276168404Spjd (void) printf("\nWorkload summary:\n\n"); 6277168404Spjd (void) printf("%7s %9s %s\n", 6278168404Spjd "Calls", "Time", "Function"); 6279168404Spjd (void) printf("%7s %9s %s\n", 6280168404Spjd "-----", "----", "--------"); 6281219089Spjd for (int f = 0; f < ZTEST_FUNCS; f++) { 6282168404Spjd Dl_info dli; 6283168404Spjd 6284236143Smm zi = &ztest_info[f]; 6285236143Smm zc = ZTEST_GET_SHARED_CALLSTATE(f); 6286236143Smm print_time(zc->zc_time, timebuf); 6287168404Spjd (void) dladdr((void *)zi->zi_func, &dli); 6288168404Spjd (void) printf("%7llu %9s %s\n", 6289236143Smm (u_longlong_t)zc->zc_count, timebuf, 6290168404Spjd dli.dli_sname); 6291168404Spjd } 6292168404Spjd (void) printf("\n"); 6293168404Spjd } 6294168404Spjd 6295168404Spjd /* 6296219089Spjd * It's possible that we killed a child during a rename test, 6297219089Spjd * in which case we'll have a 'ztest_tmp' pool lying around 6298219089Spjd * instead of 'ztest'. Do a blind rename in case this happened. 6299168404Spjd */ 6300219089Spjd kernel_init(FREAD); 6301236143Smm if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) { 6302219089Spjd spa_close(spa, FTAG); 6303219089Spjd } else { 6304219089Spjd char tmpname[MAXNAMELEN]; 6305219089Spjd kernel_fini(); 6306219089Spjd kernel_init(FREAD | FWRITE); 6307219089Spjd (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp", 6308236143Smm ztest_opts.zo_pool); 6309236143Smm (void) spa_rename(tmpname, ztest_opts.zo_pool); 6310219089Spjd } 6311168404Spjd kernel_fini(); 6312219089Spjd 6313236143Smm ztest_run_zdb(ztest_opts.zo_pool); 6314168404Spjd } 6315168404Spjd 6316236143Smm if (ztest_opts.zo_verbose >= 1) { 6317236143Smm if (hasalt) { 6318236143Smm (void) printf("%d runs of older ztest: %s\n", older, 6319236143Smm ztest_opts.zo_alt_ztest); 6320236143Smm (void) printf("%d runs of newer ztest: %s\n", newer, 6321236143Smm cmd); 6322236143Smm } 6323168404Spjd (void) printf("%d killed, %d completed, %.0f%% kill rate\n", 6324168404Spjd kills, iters - kills, (100.0 * kills) / MAX(1, iters)); 6325168404Spjd } 6326168404Spjd 6327242845Sdelphij umem_free(cmd, MAXNAMELEN); 6328242845Sdelphij 6329168404Spjd return (0); 6330168404Spjd} 6331