1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2012 by Delphix. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  25  */
  26 
  27 /*
  28  * The objective of this program is to provide a DMU/ZAP/SPA stress test
  29  * that runs entirely in userland, is easy to use, and easy to extend.
  30  *
  31  * The overall design of the ztest program is as follows:
  32  *
  33  * (1) For each major functional area (e.g. adding vdevs to a pool,
  34  *     creating and destroying datasets, reading and writing objects, etc)
  35  *     we have a simple routine to test that functionality.  These
  36  *     individual routines do not have to do anything "stressful".
  37  *
  38  * (2) We turn these simple functionality tests into a stress test by
  39  *     running them all in parallel, with as many threads as desired,
  40  *     and spread across as many datasets, objects, and vdevs as desired.
  41  *
  42  * (3) While all this is happening, we inject faults into the pool to
  43  *     verify that self-healing data really works.
  44  *
  45  * (4) Every time we open a dataset, we change its checksum and compression
  46  *     functions.  Thus even individual objects vary from block to block
  47  *     in which checksum they use and whether they're compressed.
  48  *
  49  * (5) To verify that we never lose on-disk consistency after a crash,
  50  *     we run the entire test in a child of the main process.
  51  *     At random times, the child self-immolates with a SIGKILL.
  52  *     This is the software equivalent of pulling the power cord.
  53  *     The parent then runs the test again, using the existing
  54  *     storage pool, as many times as desired. If backwards compatability
  55  *     testing is enabled ztest will sometimes run the "older" version
  56  *     of ztest after a SIGKILL.
  57  *
  58  * (6) To verify that we don't have future leaks or temporal incursions,
  59  *     many of the functional tests record the transaction group number
  60  *     as part of their data.  When reading old data, they verify that
  61  *     the transaction group number is less than the current, open txg.
  62  *     If you add a new test, please do this if applicable.
  63  *
  64  * When run with no arguments, ztest runs for about five minutes and
  65  * produces no output if successful.  To get a little bit of information,
  66  * specify -V.  To get more information, specify -VV, and so on.
  67  *
  68  * To turn this into an overnight stress test, use -T to specify run time.
  69  *
  70  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
  71  * to increase the pool capacity, fanout, and overall stress level.
  72  *
  73  * Use the -k option to set the desired frequency of kills.
  74  *
  75  * When ztest invokes itself it passes all relevant information through a
  76  * temporary file which is mmap-ed in the child process. This allows shared
  77  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
  78  * stored at offset 0 of this file and contains information on the size and
  79  * number of shared structures in the file. The information stored in this file
  80  * must remain backwards compatible with older versions of ztest so that
  81  * ztest can invoke them during backwards compatibility testing (-B).
  82  */
  83 
  84 #include <sys/zfs_context.h>
  85 #include <sys/spa.h>
  86 #include <sys/dmu.h>
  87 #include <sys/txg.h>
  88 #include <sys/dbuf.h>
  89 #include <sys/zap.h>
  90 #include <sys/dmu_objset.h>
  91 #include <sys/poll.h>
  92 #include <sys/stat.h>
  93 #include <sys/time.h>
  94 #include <sys/wait.h>
  95 #include <sys/mman.h>
  96 #include <sys/resource.h>
  97 #include <sys/zio.h>
  98 #include <sys/zil.h>
  99 #include <sys/zil_impl.h>
 100 #include <sys/vdev_impl.h>
 101 #include <sys/vdev_file.h>
 102 #include <sys/spa_impl.h>
 103 #include <sys/metaslab_impl.h>
 104 #include <sys/dsl_prop.h>
 105 #include <sys/dsl_dataset.h>
 106 #include <sys/dsl_scan.h>
 107 #include <sys/zio_checksum.h>
 108 #include <sys/refcount.h>
 109 #include <sys/zfeature.h>
 110 #include <stdio.h>
 111 #include <stdio_ext.h>
 112 #include <stdlib.h>
 113 #include <unistd.h>
 114 #include <signal.h>
 115 #include <umem.h>
 116 #include <dlfcn.h>
 117 #include <ctype.h>
 118 #include <math.h>
 119 #include <sys/fs/zfs.h>
 120 #include <libnvpair.h>
 121 
 122 #define ZTEST_FD_DATA 3
 123 #define ZTEST_FD_RAND 4
 124 
 125 typedef struct ztest_shared_hdr {
 126         uint64_t        zh_hdr_size;
 127         uint64_t        zh_opts_size;
 128         uint64_t        zh_size;
 129         uint64_t        zh_stats_size;
 130         uint64_t        zh_stats_count;
 131         uint64_t        zh_ds_size;
 132         uint64_t        zh_ds_count;
 133 } ztest_shared_hdr_t;
 134 
 135 static ztest_shared_hdr_t *ztest_shared_hdr;
 136 
 137 typedef struct ztest_shared_opts {
 138         char zo_pool[MAXNAMELEN];
 139         char zo_dir[MAXNAMELEN];
 140         char zo_alt_ztest[MAXNAMELEN];
 141         char zo_alt_libpath[MAXNAMELEN];
 142         uint64_t zo_vdevs;
 143         uint64_t zo_vdevtime;
 144         size_t zo_vdev_size;
 145         int zo_ashift;
 146         int zo_mirrors;
 147         int zo_raidz;
 148         int zo_raidz_parity;
 149         int zo_datasets;
 150         int zo_threads;
 151         uint64_t zo_passtime;
 152         uint64_t zo_killrate;
 153         int zo_verbose;
 154         int zo_init;
 155         uint64_t zo_time;
 156         uint64_t zo_maxloops;
 157         uint64_t zo_metaslab_gang_bang;
 158 } ztest_shared_opts_t;
 159 
 160 static const ztest_shared_opts_t ztest_opts_defaults = {
 161         .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
 162         .zo_dir = { '/', 't', 'm', 'p', '\0' },
 163         .zo_alt_ztest = { '\0' },
 164         .zo_alt_libpath = { '\0' },
 165         .zo_vdevs = 5,
 166         .zo_ashift = SPA_MINBLOCKSHIFT,
 167         .zo_mirrors = 2,
 168         .zo_raidz = 4,
 169         .zo_raidz_parity = 1,
 170         .zo_vdev_size = SPA_MINDEVSIZE,
 171         .zo_datasets = 7,
 172         .zo_threads = 23,
 173         .zo_passtime = 60,              /* 60 seconds */
 174         .zo_killrate = 70,              /* 70% kill rate */
 175         .zo_verbose = 0,
 176         .zo_init = 1,
 177         .zo_time = 300,                 /* 5 minutes */
 178         .zo_maxloops = 50,              /* max loops during spa_freeze() */
 179         .zo_metaslab_gang_bang = 32 << 10
 180 };
 181 
 182 extern uint64_t metaslab_gang_bang;
 183 extern uint64_t metaslab_df_alloc_threshold;
 184 
 185 static ztest_shared_opts_t *ztest_shared_opts;
 186 static ztest_shared_opts_t ztest_opts;
 187 
 188 typedef struct ztest_shared_ds {
 189         uint64_t        zd_seq;
 190 } ztest_shared_ds_t;
 191 
 192 static ztest_shared_ds_t *ztest_shared_ds;
 193 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
 194 
 195 #define BT_MAGIC        0x123456789abcdefULL
 196 #define MAXFAULTS() \
 197         (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
 198 
 199 enum ztest_io_type {
 200         ZTEST_IO_WRITE_TAG,
 201         ZTEST_IO_WRITE_PATTERN,
 202         ZTEST_IO_WRITE_ZEROES,
 203         ZTEST_IO_TRUNCATE,
 204         ZTEST_IO_SETATTR,
 205         ZTEST_IO_TYPES
 206 };
 207 
 208 typedef struct ztest_block_tag {
 209         uint64_t        bt_magic;
 210         uint64_t        bt_objset;
 211         uint64_t        bt_object;
 212         uint64_t        bt_offset;
 213         uint64_t        bt_gen;
 214         uint64_t        bt_txg;
 215         uint64_t        bt_crtxg;
 216 } ztest_block_tag_t;
 217 
 218 typedef struct bufwad {
 219         uint64_t        bw_index;
 220         uint64_t        bw_txg;
 221         uint64_t        bw_data;
 222 } bufwad_t;
 223 
 224 /*
 225  * XXX -- fix zfs range locks to be generic so we can use them here.
 226  */
 227 typedef enum {
 228         RL_READER,
 229         RL_WRITER,
 230         RL_APPEND
 231 } rl_type_t;
 232 
 233 typedef struct rll {
 234         void            *rll_writer;
 235         int             rll_readers;
 236         mutex_t         rll_lock;
 237         cond_t          rll_cv;
 238 } rll_t;
 239 
 240 typedef struct rl {
 241         uint64_t        rl_object;
 242         uint64_t        rl_offset;
 243         uint64_t        rl_size;
 244         rll_t           *rl_lock;
 245 } rl_t;
 246 
 247 #define ZTEST_RANGE_LOCKS       64
 248 #define ZTEST_OBJECT_LOCKS      64
 249 
 250 /*
 251  * Object descriptor.  Used as a template for object lookup/create/remove.
 252  */
 253 typedef struct ztest_od {
 254         uint64_t        od_dir;
 255         uint64_t        od_object;
 256         dmu_object_type_t od_type;
 257         dmu_object_type_t od_crtype;
 258         uint64_t        od_blocksize;
 259         uint64_t        od_crblocksize;
 260         uint64_t        od_gen;
 261         uint64_t        od_crgen;
 262         char            od_name[MAXNAMELEN];
 263 } ztest_od_t;
 264 
 265 /*
 266  * Per-dataset state.
 267  */
 268 typedef struct ztest_ds {
 269         ztest_shared_ds_t *zd_shared;
 270         objset_t        *zd_os;
 271         rwlock_t        zd_zilog_lock;
 272         zilog_t         *zd_zilog;
 273         ztest_od_t      *zd_od;         /* debugging aid */
 274         char            zd_name[MAXNAMELEN];
 275         mutex_t         zd_dirobj_lock;
 276         rll_t           zd_object_lock[ZTEST_OBJECT_LOCKS];
 277         rll_t           zd_range_lock[ZTEST_RANGE_LOCKS];
 278 } ztest_ds_t;
 279 
 280 /*
 281  * Per-iteration state.
 282  */
 283 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
 284 
 285 typedef struct ztest_info {
 286         ztest_func_t    *zi_func;       /* test function */
 287         uint64_t        zi_iters;       /* iterations per execution */
 288         uint64_t        *zi_interval;   /* execute every <interval> seconds */
 289 } ztest_info_t;
 290 
 291 typedef struct ztest_shared_callstate {
 292         uint64_t        zc_count;       /* per-pass count */
 293         uint64_t        zc_time;        /* per-pass time */
 294         uint64_t        zc_next;        /* next time to call this function */
 295 } ztest_shared_callstate_t;
 296 
 297 static ztest_shared_callstate_t *ztest_shared_callstate;
 298 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
 299 
 300 /*
 301  * Note: these aren't static because we want dladdr() to work.
 302  */
 303 ztest_func_t ztest_dmu_read_write;
 304 ztest_func_t ztest_dmu_write_parallel;
 305 ztest_func_t ztest_dmu_object_alloc_free;
 306 ztest_func_t ztest_dmu_commit_callbacks;
 307 ztest_func_t ztest_zap;
 308 ztest_func_t ztest_zap_parallel;
 309 ztest_func_t ztest_zil_commit;
 310 ztest_func_t ztest_zil_remount;
 311 ztest_func_t ztest_dmu_read_write_zcopy;
 312 ztest_func_t ztest_dmu_objset_create_destroy;
 313 ztest_func_t ztest_dmu_prealloc;
 314 ztest_func_t ztest_fzap;
 315 ztest_func_t ztest_dmu_snapshot_create_destroy;
 316 ztest_func_t ztest_dsl_prop_get_set;
 317 ztest_func_t ztest_spa_prop_get_set;
 318 ztest_func_t ztest_spa_create_destroy;
 319 ztest_func_t ztest_fault_inject;
 320 ztest_func_t ztest_ddt_repair;
 321 ztest_func_t ztest_dmu_snapshot_hold;
 322 ztest_func_t ztest_spa_rename;
 323 ztest_func_t ztest_scrub;
 324 ztest_func_t ztest_dsl_dataset_promote_busy;
 325 ztest_func_t ztest_vdev_attach_detach;
 326 ztest_func_t ztest_vdev_LUN_growth;
 327 ztest_func_t ztest_vdev_add_remove;
 328 ztest_func_t ztest_vdev_aux_add_remove;
 329 ztest_func_t ztest_split_pool;
 330 ztest_func_t ztest_reguid;
 331 
 332 uint64_t zopt_always = 0ULL * NANOSEC;          /* all the time */
 333 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;  /* every 1/10 second */
 334 uint64_t zopt_often = 1ULL * NANOSEC;           /* every second */
 335 uint64_t zopt_sometimes = 10ULL * NANOSEC;      /* every 10 seconds */
 336 uint64_t zopt_rarely = 60ULL * NANOSEC;         /* every 60 seconds */
 337 
 338 ztest_info_t ztest_info[] = {
 339         { ztest_dmu_read_write,                 1,      &zopt_always        },
 340         { ztest_dmu_write_parallel,             10,     &zopt_always        },
 341         { ztest_dmu_object_alloc_free,          1,      &zopt_always        },
 342         { ztest_dmu_commit_callbacks,           1,      &zopt_always        },
 343         { ztest_zap,                            30,     &zopt_always        },
 344         { ztest_zap_parallel,                   100,    &zopt_always        },
 345         { ztest_split_pool,                     1,      &zopt_always        },
 346         { ztest_zil_commit,                     1,      &zopt_incessant     },
 347         { ztest_zil_remount,                    1,      &zopt_sometimes     },
 348         { ztest_dmu_read_write_zcopy,           1,      &zopt_often },
 349         { ztest_dmu_objset_create_destroy,      1,      &zopt_often },
 350         { ztest_dsl_prop_get_set,               1,      &zopt_often },
 351         { ztest_spa_prop_get_set,               1,      &zopt_sometimes     },
 352 #if 0
 353         { ztest_dmu_prealloc,                   1,      &zopt_sometimes     },
 354 #endif
 355         { ztest_fzap,                           1,      &zopt_sometimes     },
 356         { ztest_dmu_snapshot_create_destroy,    1,      &zopt_sometimes     },
 357         { ztest_spa_create_destroy,             1,      &zopt_sometimes     },
 358         { ztest_fault_inject,                   1,      &zopt_sometimes     },
 359         { ztest_ddt_repair,                     1,      &zopt_sometimes     },
 360         { ztest_dmu_snapshot_hold,              1,      &zopt_sometimes     },
 361         { ztest_reguid,                         1,      &zopt_sometimes },
 362         { ztest_spa_rename,                     1,      &zopt_rarely        },
 363         { ztest_scrub,                          1,      &zopt_rarely        },
 364         { ztest_dsl_dataset_promote_busy,       1,      &zopt_rarely        },
 365         { ztest_vdev_attach_detach,             1,      &zopt_rarely },
 366         { ztest_vdev_LUN_growth,                1,      &zopt_rarely        },
 367         { ztest_vdev_add_remove,                1,
 368             &ztest_opts.zo_vdevtime                         },
 369         { ztest_vdev_aux_add_remove,            1,
 370             &ztest_opts.zo_vdevtime                         },
 371 };
 372 
 373 #define ZTEST_FUNCS     (sizeof (ztest_info) / sizeof (ztest_info_t))
 374 
 375 /*
 376  * The following struct is used to hold a list of uncalled commit callbacks.
 377  * The callbacks are ordered by txg number.
 378  */
 379 typedef struct ztest_cb_list {
 380         mutex_t zcl_callbacks_lock;
 381         list_t  zcl_callbacks;
 382 } ztest_cb_list_t;
 383 
 384 /*
 385  * Stuff we need to share writably between parent and child.
 386  */
 387 typedef struct ztest_shared {
 388         boolean_t       zs_do_init;
 389         hrtime_t        zs_proc_start;
 390         hrtime_t        zs_proc_stop;
 391         hrtime_t        zs_thread_start;
 392         hrtime_t        zs_thread_stop;
 393         hrtime_t        zs_thread_kill;
 394         uint64_t        zs_enospc_count;
 395         uint64_t        zs_vdev_next_leaf;
 396         uint64_t        zs_vdev_aux;
 397         uint64_t        zs_alloc;
 398         uint64_t        zs_space;
 399         uint64_t        zs_splits;
 400         uint64_t        zs_mirrors;
 401         uint64_t        zs_metaslab_sz;
 402         uint64_t        zs_metaslab_df_alloc_threshold;
 403         uint64_t        zs_guid;
 404 } ztest_shared_t;
 405 
 406 #define ID_PARALLEL     -1ULL
 407 
 408 static char ztest_dev_template[] = "%s/%s.%llua";
 409 static char ztest_aux_template[] = "%s/%s.%s.%llu";
 410 ztest_shared_t *ztest_shared;
 411 
 412 static spa_t *ztest_spa = NULL;
 413 static ztest_ds_t *ztest_ds;
 414 
 415 static mutex_t ztest_vdev_lock;
 416 static rwlock_t ztest_name_lock;
 417 
 418 static boolean_t ztest_dump_core = B_TRUE;
 419 static boolean_t ztest_exiting;
 420 
 421 /* Global commit callback list */
 422 static ztest_cb_list_t zcl;
 423 
 424 enum ztest_object {
 425         ZTEST_META_DNODE = 0,
 426         ZTEST_DIROBJ,
 427         ZTEST_OBJECTS
 428 };
 429 
 430 static void usage(boolean_t) __NORETURN;
 431 
 432 /*
 433  * These libumem hooks provide a reasonable set of defaults for the allocator's
 434  * debugging facilities.
 435  */
 436 const char *
 437 _umem_debug_init()
 438 {
 439         return ("default,verbose"); /* $UMEM_DEBUG setting */
 440 }
 441 
 442 const char *
 443 _umem_logging_init(void)
 444 {
 445         return ("fail,contents"); /* $UMEM_LOGGING setting */
 446 }
 447 
 448 #define FATAL_MSG_SZ    1024
 449 
 450 char *fatal_msg;
 451 
 452 static void
 453 fatal(int do_perror, char *message, ...)
 454 {
 455         va_list args;
 456         int save_errno = errno;
 457         char buf[FATAL_MSG_SZ];
 458 
 459         (void) fflush(stdout);
 460 
 461         va_start(args, message);
 462         (void) sprintf(buf, "ztest: ");
 463         /* LINTED */
 464         (void) vsprintf(buf + strlen(buf), message, args);
 465         va_end(args);
 466         if (do_perror) {
 467                 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
 468                     ": %s", strerror(save_errno));
 469         }
 470         (void) fprintf(stderr, "%s\n", buf);
 471         fatal_msg = buf;                        /* to ease debugging */
 472         if (ztest_dump_core)
 473                 abort();
 474         exit(3);
 475 }
 476 
 477 static int
 478 str2shift(const char *buf)
 479 {
 480         const char *ends = "BKMGTPEZ";
 481         int i;
 482 
 483         if (buf[0] == '\0')
 484                 return (0);
 485         for (i = 0; i < strlen(ends); i++) {
 486                 if (toupper(buf[0]) == ends[i])
 487                         break;
 488         }
 489         if (i == strlen(ends)) {
 490                 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
 491                     buf);
 492                 usage(B_FALSE);
 493         }
 494         if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
 495                 return (10*i);
 496         }
 497         (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
 498         usage(B_FALSE);
 499         /* NOTREACHED */
 500 }
 501 
 502 static uint64_t
 503 nicenumtoull(const char *buf)
 504 {
 505         char *end;
 506         uint64_t val;
 507 
 508         val = strtoull(buf, &end, 0);
 509         if (end == buf) {
 510                 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
 511                 usage(B_FALSE);
 512         } else if (end[0] == '.') {
 513                 double fval = strtod(buf, &end);
 514                 fval *= pow(2, str2shift(end));
 515                 if (fval > UINT64_MAX) {
 516                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 517                             buf);
 518                         usage(B_FALSE);
 519                 }
 520                 val = (uint64_t)fval;
 521         } else {
 522                 int shift = str2shift(end);
 523                 if (shift >= 64 || (val << shift) >> shift != val) {
 524                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 525                             buf);
 526                         usage(B_FALSE);
 527                 }
 528                 val <<= shift;
 529         }
 530         return (val);
 531 }
 532 
 533 static void
 534 usage(boolean_t requested)
 535 {
 536         const ztest_shared_opts_t *zo = &ztest_opts_defaults;
 537 
 538         char nice_vdev_size[10];
 539         char nice_gang_bang[10];
 540         FILE *fp = requested ? stdout : stderr;
 541 
 542         nicenum(zo->zo_vdev_size, nice_vdev_size);
 543         nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
 544 
 545         (void) fprintf(fp, "Usage: %s\n"
 546             "\t[-v vdevs (default: %llu)]\n"
 547             "\t[-s size_of_each_vdev (default: %s)]\n"
 548             "\t[-a alignment_shift (default: %d)] use 0 for random\n"
 549             "\t[-m mirror_copies (default: %d)]\n"
 550             "\t[-r raidz_disks (default: %d)]\n"
 551             "\t[-R raidz_parity (default: %d)]\n"
 552             "\t[-d datasets (default: %d)]\n"
 553             "\t[-t threads (default: %d)]\n"
 554             "\t[-g gang_block_threshold (default: %s)]\n"
 555             "\t[-i init_count (default: %d)] initialize pool i times\n"
 556             "\t[-k kill_percentage (default: %llu%%)]\n"
 557             "\t[-p pool_name (default: %s)]\n"
 558             "\t[-f dir (default: %s)] file directory for vdev files\n"
 559             "\t[-V] verbose (use multiple times for ever more blather)\n"
 560             "\t[-E] use existing pool instead of creating new one\n"
 561             "\t[-T time (default: %llu sec)] total run time\n"
 562             "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
 563             "\t[-P passtime (default: %llu sec)] time per pass\n"
 564             "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
 565             "\t[-h] (print help)\n"
 566             "",
 567             zo->zo_pool,
 568             (u_longlong_t)zo->zo_vdevs,                      /* -v */
 569             nice_vdev_size,                             /* -s */
 570             zo->zo_ashift,                           /* -a */
 571             zo->zo_mirrors,                          /* -m */
 572             zo->zo_raidz,                            /* -r */
 573             zo->zo_raidz_parity,                     /* -R */
 574             zo->zo_datasets,                         /* -d */
 575             zo->zo_threads,                          /* -t */
 576             nice_gang_bang,                             /* -g */
 577             zo->zo_init,                             /* -i */
 578             (u_longlong_t)zo->zo_killrate,           /* -k */
 579             zo->zo_pool,                             /* -p */
 580             zo->zo_dir,                                      /* -f */
 581             (u_longlong_t)zo->zo_time,                       /* -T */
 582             (u_longlong_t)zo->zo_maxloops,           /* -F */
 583             (u_longlong_t)zo->zo_passtime);
 584         exit(requested ? 0 : 1);
 585 }
 586 
 587 static void
 588 process_options(int argc, char **argv)
 589 {
 590         char *path;
 591         ztest_shared_opts_t *zo = &ztest_opts;
 592 
 593         int opt;
 594         uint64_t value;
 595         char altdir[MAXNAMELEN] = { 0 };
 596 
 597         bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
 598 
 599         while ((opt = getopt(argc, argv,
 600             "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
 601                 value = 0;
 602                 switch (opt) {
 603                 case 'v':
 604                 case 's':
 605                 case 'a':
 606                 case 'm':
 607                 case 'r':
 608                 case 'R':
 609                 case 'd':
 610                 case 't':
 611                 case 'g':
 612                 case 'i':
 613                 case 'k':
 614                 case 'T':
 615                 case 'P':
 616                 case 'F':
 617                         value = nicenumtoull(optarg);
 618                 }
 619                 switch (opt) {
 620                 case 'v':
 621                         zo->zo_vdevs = value;
 622                         break;
 623                 case 's':
 624                         zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
 625                         break;
 626                 case 'a':
 627                         zo->zo_ashift = value;
 628                         break;
 629                 case 'm':
 630                         zo->zo_mirrors = value;
 631                         break;
 632                 case 'r':
 633                         zo->zo_raidz = MAX(1, value);
 634                         break;
 635                 case 'R':
 636                         zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
 637                         break;
 638                 case 'd':
 639                         zo->zo_datasets = MAX(1, value);
 640                         break;
 641                 case 't':
 642                         zo->zo_threads = MAX(1, value);
 643                         break;
 644                 case 'g':
 645                         zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
 646                             value);
 647                         break;
 648                 case 'i':
 649                         zo->zo_init = value;
 650                         break;
 651                 case 'k':
 652                         zo->zo_killrate = value;
 653                         break;
 654                 case 'p':
 655                         (void) strlcpy(zo->zo_pool, optarg,
 656                             sizeof (zo->zo_pool));
 657                         break;
 658                 case 'f':
 659                         path = realpath(optarg, NULL);
 660                         if (path == NULL) {
 661                                 (void) fprintf(stderr, "error: %s: %s\n",
 662                                     optarg, strerror(errno));
 663                                 usage(B_FALSE);
 664                         } else {
 665                                 (void) strlcpy(zo->zo_dir, path,
 666                                     sizeof (zo->zo_dir));
 667                         }
 668                         break;
 669                 case 'V':
 670                         zo->zo_verbose++;
 671                         break;
 672                 case 'E':
 673                         zo->zo_init = 0;
 674                         break;
 675                 case 'T':
 676                         zo->zo_time = value;
 677                         break;
 678                 case 'P':
 679                         zo->zo_passtime = MAX(1, value);
 680                         break;
 681                 case 'F':
 682                         zo->zo_maxloops = MAX(1, value);
 683                         break;
 684                 case 'B':
 685                         (void) strlcpy(altdir, optarg, sizeof (altdir));
 686                         break;
 687                 case 'h':
 688                         usage(B_TRUE);
 689                         break;
 690                 case '?':
 691                 default:
 692                         usage(B_FALSE);
 693                         break;
 694                 }
 695         }
 696 
 697         zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
 698 
 699         zo->zo_vdevtime =
 700             (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
 701             UINT64_MAX >> 2);
 702 
 703         if (strlen(altdir) > 0) {
 704                 char cmd[MAXNAMELEN];
 705                 char realaltdir[MAXNAMELEN];
 706                 char *bin;
 707                 char *ztest;
 708                 char *isa;
 709                 int isalen;
 710 
 711                 (void) realpath(getexecname(), cmd);
 712                 if (0 != access(altdir, F_OK)) {
 713                         ztest_dump_core = B_FALSE;
 714                         fatal(B_TRUE, "invalid alternate ztest path: %s",
 715                             altdir);
 716                 }
 717                 VERIFY(NULL != realpath(altdir, realaltdir));
 718 
 719                 /*
 720                  * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
 721                  * We want to extract <isa> to determine if we should use
 722                  * 32 or 64 bit binaries.
 723                  */
 724                 bin = strstr(cmd, "/usr/bin/");
 725                 ztest = strstr(bin, "/ztest");
 726                 isa = bin + 9;
 727                 isalen = ztest - isa;
 728                 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
 729                     "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
 730                 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
 731                     "%s/usr/lib/%.*s", realaltdir, isalen, isa);
 732 
 733                 if (0 != access(zo->zo_alt_ztest, X_OK)) {
 734                         ztest_dump_core = B_FALSE;
 735                         fatal(B_TRUE, "invalid alternate ztest: %s",
 736                             zo->zo_alt_ztest);
 737                 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
 738                         ztest_dump_core = B_FALSE;
 739                         fatal(B_TRUE, "invalid alternate lib directory %s",
 740                             zo->zo_alt_libpath);
 741                 }
 742         }
 743 }
 744 
 745 static void
 746 ztest_kill(ztest_shared_t *zs)
 747 {
 748         zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
 749         zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
 750         (void) kill(getpid(), SIGKILL);
 751 }
 752 
 753 static uint64_t
 754 ztest_random(uint64_t range)
 755 {
 756         uint64_t r;
 757 
 758         if (range == 0)
 759                 return (0);
 760 
 761         if (read(ZTEST_FD_RAND, &r, sizeof (r)) != sizeof (r))
 762                 fatal(1, "short read from /dev/urandom");
 763 
 764         return (r % range);
 765 }
 766 
 767 /* ARGSUSED */
 768 static void
 769 ztest_record_enospc(const char *s)
 770 {
 771         ztest_shared->zs_enospc_count++;
 772 }
 773 
 774 static uint64_t
 775 ztest_get_ashift(void)
 776 {
 777         if (ztest_opts.zo_ashift == 0)
 778                 return (SPA_MINBLOCKSHIFT + ztest_random(3));
 779         return (ztest_opts.zo_ashift);
 780 }
 781 
 782 static nvlist_t *
 783 make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift)
 784 {
 785         char pathbuf[MAXPATHLEN];
 786         uint64_t vdev;
 787         nvlist_t *file;
 788 
 789         if (ashift == 0)
 790                 ashift = ztest_get_ashift();
 791 
 792         if (path == NULL) {
 793                 path = pathbuf;
 794 
 795                 if (aux != NULL) {
 796                         vdev = ztest_shared->zs_vdev_aux;
 797                         (void) snprintf(path, sizeof (pathbuf),
 798                             ztest_aux_template, ztest_opts.zo_dir,
 799                             ztest_opts.zo_pool, aux, vdev);
 800                 } else {
 801                         vdev = ztest_shared->zs_vdev_next_leaf++;
 802                         (void) snprintf(path, sizeof (pathbuf),
 803                             ztest_dev_template, ztest_opts.zo_dir,
 804                             ztest_opts.zo_pool, vdev);
 805                 }
 806         }
 807 
 808         if (size != 0) {
 809                 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
 810                 if (fd == -1)
 811                         fatal(1, "can't open %s", path);
 812                 if (ftruncate(fd, size) != 0)
 813                         fatal(1, "can't ftruncate %s", path);
 814                 (void) close(fd);
 815         }
 816 
 817         VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
 818         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
 819         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
 820         VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
 821 
 822         return (file);
 823 }
 824 
 825 static nvlist_t *
 826 make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r)
 827 {
 828         nvlist_t *raidz, **child;
 829         int c;
 830 
 831         if (r < 2)
 832                 return (make_vdev_file(path, aux, size, ashift));
 833         child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
 834 
 835         for (c = 0; c < r; c++)
 836                 child[c] = make_vdev_file(path, aux, size, ashift);
 837 
 838         VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
 839         VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
 840             VDEV_TYPE_RAIDZ) == 0);
 841         VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
 842             ztest_opts.zo_raidz_parity) == 0);
 843         VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
 844             child, r) == 0);
 845 
 846         for (c = 0; c < r; c++)
 847                 nvlist_free(child[c]);
 848 
 849         umem_free(child, r * sizeof (nvlist_t *));
 850 
 851         return (raidz);
 852 }
 853 
 854 static nvlist_t *
 855 make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift,
 856         int r, int m)
 857 {
 858         nvlist_t *mirror, **child;
 859         int c;
 860 
 861         if (m < 1)
 862                 return (make_vdev_raidz(path, aux, size, ashift, r));
 863 
 864         child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
 865 
 866         for (c = 0; c < m; c++)
 867                 child[c] = make_vdev_raidz(path, aux, size, ashift, r);
 868 
 869         VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
 870         VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
 871             VDEV_TYPE_MIRROR) == 0);
 872         VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
 873             child, m) == 0);
 874 
 875         for (c = 0; c < m; c++)
 876                 nvlist_free(child[c]);
 877 
 878         umem_free(child, m * sizeof (nvlist_t *));
 879 
 880         return (mirror);
 881 }
 882 
 883 static nvlist_t *
 884 make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift,
 885         int log, int r, int m, int t)
 886 {
 887         nvlist_t *root, **child;
 888         int c;
 889 
 890         ASSERT(t > 0);
 891 
 892         child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
 893 
 894         for (c = 0; c < t; c++) {
 895                 child[c] = make_vdev_mirror(path, aux, size, ashift, r, m);
 896                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
 897                     log) == 0);
 898         }
 899 
 900         VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
 901         VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
 902         VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
 903             child, t) == 0);
 904 
 905         for (c = 0; c < t; c++)
 906                 nvlist_free(child[c]);
 907 
 908         umem_free(child, t * sizeof (nvlist_t *));
 909 
 910         return (root);
 911 }
 912 
 913 static int
 914 ztest_random_blocksize(void)
 915 {
 916         return (1 << (SPA_MINBLOCKSHIFT +
 917             ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
 918 }
 919 
 920 static int
 921 ztest_random_ibshift(void)
 922 {
 923         return (DN_MIN_INDBLKSHIFT +
 924             ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
 925 }
 926 
 927 static uint64_t
 928 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
 929 {
 930         uint64_t top;
 931         vdev_t *rvd = spa->spa_root_vdev;
 932         vdev_t *tvd;
 933 
 934         ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
 935 
 936         do {
 937                 top = ztest_random(rvd->vdev_children);
 938                 tvd = rvd->vdev_child[top];
 939         } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
 940             tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
 941 
 942         return (top);
 943 }
 944 
 945 static uint64_t
 946 ztest_random_dsl_prop(zfs_prop_t prop)
 947 {
 948         uint64_t value;
 949 
 950         do {
 951                 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
 952         } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
 953 
 954         return (value);
 955 }
 956 
 957 static int
 958 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
 959     boolean_t inherit)
 960 {
 961         const char *propname = zfs_prop_to_name(prop);
 962         const char *valname;
 963         char setpoint[MAXPATHLEN];
 964         uint64_t curval;
 965         int error;
 966 
 967         error = dsl_prop_set(osname, propname,
 968             (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL),
 969             sizeof (value), 1, &value);
 970 
 971         if (error == ENOSPC) {
 972                 ztest_record_enospc(FTAG);
 973                 return (error);
 974         }
 975         ASSERT3U(error, ==, 0);
 976 
 977         VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval),
 978             1, &curval, setpoint), ==, 0);
 979 
 980         if (ztest_opts.zo_verbose >= 6) {
 981                 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
 982                 (void) printf("%s %s = %s at '%s'\n",
 983                     osname, propname, valname, setpoint);
 984         }
 985 
 986         return (error);
 987 }
 988 
 989 static int
 990 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
 991 {
 992         spa_t *spa = ztest_spa;
 993         nvlist_t *props = NULL;
 994         int error;
 995 
 996         VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
 997         VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
 998 
 999         error = spa_prop_set(spa, props);
1000 
1001         nvlist_free(props);
1002 
1003         if (error == ENOSPC) {
1004                 ztest_record_enospc(FTAG);
1005                 return (error);
1006         }
1007         ASSERT3U(error, ==, 0);
1008 
1009         return (error);
1010 }
1011 
1012 static void
1013 ztest_rll_init(rll_t *rll)
1014 {
1015         rll->rll_writer = NULL;
1016         rll->rll_readers = 0;
1017         VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1018         VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1019 }
1020 
1021 static void
1022 ztest_rll_destroy(rll_t *rll)
1023 {
1024         ASSERT(rll->rll_writer == NULL);
1025         ASSERT(rll->rll_readers == 0);
1026         VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1027         VERIFY(cond_destroy(&rll->rll_cv) == 0);
1028 }
1029 
1030 static void
1031 ztest_rll_lock(rll_t *rll, rl_type_t type)
1032 {
1033         VERIFY(mutex_lock(&rll->rll_lock) == 0);
1034 
1035         if (type == RL_READER) {
1036                 while (rll->rll_writer != NULL)
1037                         (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1038                 rll->rll_readers++;
1039         } else {
1040                 while (rll->rll_writer != NULL || rll->rll_readers)
1041                         (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1042                 rll->rll_writer = curthread;
1043         }
1044 
1045         VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1046 }
1047 
1048 static void
1049 ztest_rll_unlock(rll_t *rll)
1050 {
1051         VERIFY(mutex_lock(&rll->rll_lock) == 0);
1052 
1053         if (rll->rll_writer) {
1054                 ASSERT(rll->rll_readers == 0);
1055                 rll->rll_writer = NULL;
1056         } else {
1057                 ASSERT(rll->rll_readers != 0);
1058                 ASSERT(rll->rll_writer == NULL);
1059                 rll->rll_readers--;
1060         }
1061 
1062         if (rll->rll_writer == NULL && rll->rll_readers == 0)
1063                 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1064 
1065         VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1066 }
1067 
1068 static void
1069 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1070 {
1071         rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1072 
1073         ztest_rll_lock(rll, type);
1074 }
1075 
1076 static void
1077 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1078 {
1079         rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1080 
1081         ztest_rll_unlock(rll);
1082 }
1083 
1084 static rl_t *
1085 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1086     uint64_t size, rl_type_t type)
1087 {
1088         uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1089         rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1090         rl_t *rl;
1091 
1092         rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1093         rl->rl_object = object;
1094         rl->rl_offset = offset;
1095         rl->rl_size = size;
1096         rl->rl_lock = rll;
1097 
1098         ztest_rll_lock(rll, type);
1099 
1100         return (rl);
1101 }
1102 
1103 static void
1104 ztest_range_unlock(rl_t *rl)
1105 {
1106         rll_t *rll = rl->rl_lock;
1107 
1108         ztest_rll_unlock(rll);
1109 
1110         umem_free(rl, sizeof (*rl));
1111 }
1112 
1113 static void
1114 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1115 {
1116         zd->zd_os = os;
1117         zd->zd_zilog = dmu_objset_zil(os);
1118         zd->zd_shared = szd;
1119         dmu_objset_name(os, zd->zd_name);
1120 
1121         if (zd->zd_shared != NULL)
1122                 zd->zd_shared->zd_seq = 0;
1123 
1124         VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1125         VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1126 
1127         for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1128                 ztest_rll_init(&zd->zd_object_lock[l]);
1129 
1130         for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1131                 ztest_rll_init(&zd->zd_range_lock[l]);
1132 }
1133 
1134 static void
1135 ztest_zd_fini(ztest_ds_t *zd)
1136 {
1137         VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1138 
1139         for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1140                 ztest_rll_destroy(&zd->zd_object_lock[l]);
1141 
1142         for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1143                 ztest_rll_destroy(&zd->zd_range_lock[l]);
1144 }
1145 
1146 #define TXG_MIGHTWAIT   (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1147 
1148 static uint64_t
1149 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1150 {
1151         uint64_t txg;
1152         int error;
1153 
1154         /*
1155          * Attempt to assign tx to some transaction group.
1156          */
1157         error = dmu_tx_assign(tx, txg_how);
1158         if (error) {
1159                 if (error == ERESTART) {
1160                         ASSERT(txg_how == TXG_NOWAIT);
1161                         dmu_tx_wait(tx);
1162                 } else {
1163                         ASSERT3U(error, ==, ENOSPC);
1164                         ztest_record_enospc(tag);
1165                 }
1166                 dmu_tx_abort(tx);
1167                 return (0);
1168         }
1169         txg = dmu_tx_get_txg(tx);
1170         ASSERT(txg != 0);
1171         return (txg);
1172 }
1173 
1174 static void
1175 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1176 {
1177         uint64_t *ip = buf;
1178         uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1179 
1180         while (ip < ip_end)
1181                 *ip++ = value;
1182 }
1183 
1184 static boolean_t
1185 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1186 {
1187         uint64_t *ip = buf;
1188         uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1189         uint64_t diff = 0;
1190 
1191         while (ip < ip_end)
1192                 diff |= (value - *ip++);
1193 
1194         return (diff == 0);
1195 }
1196 
1197 static void
1198 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1199     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1200 {
1201         bt->bt_magic = BT_MAGIC;
1202         bt->bt_objset = dmu_objset_id(os);
1203         bt->bt_object = object;
1204         bt->bt_offset = offset;
1205         bt->bt_gen = gen;
1206         bt->bt_txg = txg;
1207         bt->bt_crtxg = crtxg;
1208 }
1209 
1210 static void
1211 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1212     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1213 {
1214         ASSERT(bt->bt_magic == BT_MAGIC);
1215         ASSERT(bt->bt_objset == dmu_objset_id(os));
1216         ASSERT(bt->bt_object == object);
1217         ASSERT(bt->bt_offset == offset);
1218         ASSERT(bt->bt_gen <= gen);
1219         ASSERT(bt->bt_txg <= txg);
1220         ASSERT(bt->bt_crtxg == crtxg);
1221 }
1222 
1223 static ztest_block_tag_t *
1224 ztest_bt_bonus(dmu_buf_t *db)
1225 {
1226         dmu_object_info_t doi;
1227         ztest_block_tag_t *bt;
1228 
1229         dmu_object_info_from_db(db, &doi);
1230         ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1231         ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1232         bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1233 
1234         return (bt);
1235 }
1236 
1237 /*
1238  * ZIL logging ops
1239  */
1240 
1241 #define lrz_type        lr_mode
1242 #define lrz_blocksize   lr_uid
1243 #define lrz_ibshift     lr_gid
1244 #define lrz_bonustype   lr_rdev
1245 #define lrz_bonuslen    lr_crtime[1]
1246 
1247 static void
1248 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1249 {
1250         char *name = (void *)(lr + 1);          /* name follows lr */
1251         size_t namesize = strlen(name) + 1;
1252         itx_t *itx;
1253 
1254         if (zil_replaying(zd->zd_zilog, tx))
1255                 return;
1256 
1257         itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1258         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1259             sizeof (*lr) + namesize - sizeof (lr_t));
1260 
1261         zil_itx_assign(zd->zd_zilog, itx, tx);
1262 }
1263 
1264 static void
1265 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1266 {
1267         char *name = (void *)(lr + 1);          /* name follows lr */
1268         size_t namesize = strlen(name) + 1;
1269         itx_t *itx;
1270 
1271         if (zil_replaying(zd->zd_zilog, tx))
1272                 return;
1273 
1274         itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1275         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1276             sizeof (*lr) + namesize - sizeof (lr_t));
1277 
1278         itx->itx_oid = object;
1279         zil_itx_assign(zd->zd_zilog, itx, tx);
1280 }
1281 
1282 static void
1283 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1284 {
1285         itx_t *itx;
1286         itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1287 
1288         if (zil_replaying(zd->zd_zilog, tx))
1289                 return;
1290 
1291         if (lr->lr_length > ZIL_MAX_LOG_DATA)
1292                 write_state = WR_INDIRECT;
1293 
1294         itx = zil_itx_create(TX_WRITE,
1295             sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1296 
1297         if (write_state == WR_COPIED &&
1298             dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1299             ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1300                 zil_itx_destroy(itx);
1301                 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1302                 write_state = WR_NEED_COPY;
1303         }
1304         itx->itx_private = zd;
1305         itx->itx_wr_state = write_state;
1306         itx->itx_sync = (ztest_random(8) == 0);
1307         itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1308 
1309         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1310             sizeof (*lr) - sizeof (lr_t));
1311 
1312         zil_itx_assign(zd->zd_zilog, itx, tx);
1313 }
1314 
1315 static void
1316 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1317 {
1318         itx_t *itx;
1319 
1320         if (zil_replaying(zd->zd_zilog, tx))
1321                 return;
1322 
1323         itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1324         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1325             sizeof (*lr) - sizeof (lr_t));
1326 
1327         itx->itx_sync = B_FALSE;
1328         zil_itx_assign(zd->zd_zilog, itx, tx);
1329 }
1330 
1331 static void
1332 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1333 {
1334         itx_t *itx;
1335 
1336         if (zil_replaying(zd->zd_zilog, tx))
1337                 return;
1338 
1339         itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1340         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1341             sizeof (*lr) - sizeof (lr_t));
1342 
1343         itx->itx_sync = B_FALSE;
1344         zil_itx_assign(zd->zd_zilog, itx, tx);
1345 }
1346 
1347 /*
1348  * ZIL replay ops
1349  */
1350 static int
1351 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1352 {
1353         char *name = (void *)(lr + 1);          /* name follows lr */
1354         objset_t *os = zd->zd_os;
1355         ztest_block_tag_t *bbt;
1356         dmu_buf_t *db;
1357         dmu_tx_t *tx;
1358         uint64_t txg;
1359         int error = 0;
1360 
1361         if (byteswap)
1362                 byteswap_uint64_array(lr, sizeof (*lr));
1363 
1364         ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1365         ASSERT(name[0] != '\0');
1366 
1367         tx = dmu_tx_create(os);
1368 
1369         dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1370 
1371         if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1372                 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1373         } else {
1374                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1375         }
1376 
1377         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1378         if (txg == 0)
1379                 return (ENOSPC);
1380 
1381         ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1382 
1383         if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1384                 if (lr->lr_foid == 0) {
1385                         lr->lr_foid = zap_create(os,
1386                             lr->lrz_type, lr->lrz_bonustype,
1387                             lr->lrz_bonuslen, tx);
1388                 } else {
1389                         error = zap_create_claim(os, lr->lr_foid,
1390                             lr->lrz_type, lr->lrz_bonustype,
1391                             lr->lrz_bonuslen, tx);
1392                 }
1393         } else {
1394                 if (lr->lr_foid == 0) {
1395                         lr->lr_foid = dmu_object_alloc(os,
1396                             lr->lrz_type, 0, lr->lrz_bonustype,
1397                             lr->lrz_bonuslen, tx);
1398                 } else {
1399                         error = dmu_object_claim(os, lr->lr_foid,
1400                             lr->lrz_type, 0, lr->lrz_bonustype,
1401                             lr->lrz_bonuslen, tx);
1402                 }
1403         }
1404 
1405         if (error) {
1406                 ASSERT3U(error, ==, EEXIST);
1407                 ASSERT(zd->zd_zilog->zl_replay);
1408                 dmu_tx_commit(tx);
1409                 return (error);
1410         }
1411 
1412         ASSERT(lr->lr_foid != 0);
1413 
1414         if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1415                 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1416                     lr->lrz_blocksize, lr->lrz_ibshift, tx));
1417 
1418         VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1419         bbt = ztest_bt_bonus(db);
1420         dmu_buf_will_dirty(db, tx);
1421         ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1422         dmu_buf_rele(db, FTAG);
1423 
1424         VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1425             &lr->lr_foid, tx));
1426 
1427         (void) ztest_log_create(zd, tx, lr);
1428 
1429         dmu_tx_commit(tx);
1430 
1431         return (0);
1432 }
1433 
1434 static int
1435 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1436 {
1437         char *name = (void *)(lr + 1);          /* name follows lr */
1438         objset_t *os = zd->zd_os;
1439         dmu_object_info_t doi;
1440         dmu_tx_t *tx;
1441         uint64_t object, txg;
1442 
1443         if (byteswap)
1444                 byteswap_uint64_array(lr, sizeof (*lr));
1445 
1446         ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1447         ASSERT(name[0] != '\0');
1448 
1449         VERIFY3U(0, ==,
1450             zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1451         ASSERT(object != 0);
1452 
1453         ztest_object_lock(zd, object, RL_WRITER);
1454 
1455         VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1456 
1457         tx = dmu_tx_create(os);
1458 
1459         dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1460         dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1461 
1462         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1463         if (txg == 0) {
1464                 ztest_object_unlock(zd, object);
1465                 return (ENOSPC);
1466         }
1467 
1468         if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1469                 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1470         } else {
1471                 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1472         }
1473 
1474         VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1475 
1476         (void) ztest_log_remove(zd, tx, lr, object);
1477 
1478         dmu_tx_commit(tx);
1479 
1480         ztest_object_unlock(zd, object);
1481 
1482         return (0);
1483 }
1484 
1485 static int
1486 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1487 {
1488         objset_t *os = zd->zd_os;
1489         void *data = lr + 1;                    /* data follows lr */
1490         uint64_t offset, length;
1491         ztest_block_tag_t *bt = data;
1492         ztest_block_tag_t *bbt;
1493         uint64_t gen, txg, lrtxg, crtxg;
1494         dmu_object_info_t doi;
1495         dmu_tx_t *tx;
1496         dmu_buf_t *db;
1497         arc_buf_t *abuf = NULL;
1498         rl_t *rl;
1499 
1500         if (byteswap)
1501                 byteswap_uint64_array(lr, sizeof (*lr));
1502 
1503         offset = lr->lr_offset;
1504         length = lr->lr_length;
1505 
1506         /* If it's a dmu_sync() block, write the whole block */
1507         if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1508                 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1509                 if (length < blocksize) {
1510                         offset -= offset % blocksize;
1511                         length = blocksize;
1512                 }
1513         }
1514 
1515         if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1516                 byteswap_uint64_array(bt, sizeof (*bt));
1517 
1518         if (bt->bt_magic != BT_MAGIC)
1519                 bt = NULL;
1520 
1521         ztest_object_lock(zd, lr->lr_foid, RL_READER);
1522         rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1523 
1524         VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1525 
1526         dmu_object_info_from_db(db, &doi);
1527 
1528         bbt = ztest_bt_bonus(db);
1529         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1530         gen = bbt->bt_gen;
1531         crtxg = bbt->bt_crtxg;
1532         lrtxg = lr->lr_common.lrc_txg;
1533 
1534         tx = dmu_tx_create(os);
1535 
1536         dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1537 
1538         if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1539             P2PHASE(offset, length) == 0)
1540                 abuf = dmu_request_arcbuf(db, length);
1541 
1542         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1543         if (txg == 0) {
1544                 if (abuf != NULL)
1545                         dmu_return_arcbuf(abuf);
1546                 dmu_buf_rele(db, FTAG);
1547                 ztest_range_unlock(rl);
1548                 ztest_object_unlock(zd, lr->lr_foid);
1549                 return (ENOSPC);
1550         }
1551 
1552         if (bt != NULL) {
1553                 /*
1554                  * Usually, verify the old data before writing new data --
1555                  * but not always, because we also want to verify correct
1556                  * behavior when the data was not recently read into cache.
1557                  */
1558                 ASSERT(offset % doi.doi_data_block_size == 0);
1559                 if (ztest_random(4) != 0) {
1560                         int prefetch = ztest_random(2) ?
1561                             DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1562                         ztest_block_tag_t rbt;
1563 
1564                         VERIFY(dmu_read(os, lr->lr_foid, offset,
1565                             sizeof (rbt), &rbt, prefetch) == 0);
1566                         if (rbt.bt_magic == BT_MAGIC) {
1567                                 ztest_bt_verify(&rbt, os, lr->lr_foid,
1568                                     offset, gen, txg, crtxg);
1569                         }
1570                 }
1571 
1572                 /*
1573                  * Writes can appear to be newer than the bonus buffer because
1574                  * the ztest_get_data() callback does a dmu_read() of the
1575                  * open-context data, which may be different than the data
1576                  * as it was when the write was generated.
1577                  */
1578                 if (zd->zd_zilog->zl_replay) {
1579                         ztest_bt_verify(bt, os, lr->lr_foid, offset,
1580                             MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1581                             bt->bt_crtxg);
1582                 }
1583 
1584                 /*
1585                  * Set the bt's gen/txg to the bonus buffer's gen/txg
1586                  * so that all of the usual ASSERTs will work.
1587                  */
1588                 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1589         }
1590 
1591         if (abuf == NULL) {
1592                 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1593         } else {
1594                 bcopy(data, abuf->b_data, length);
1595                 dmu_assign_arcbuf(db, offset, abuf, tx);
1596         }
1597 
1598         (void) ztest_log_write(zd, tx, lr);
1599 
1600         dmu_buf_rele(db, FTAG);
1601 
1602         dmu_tx_commit(tx);
1603 
1604         ztest_range_unlock(rl);
1605         ztest_object_unlock(zd, lr->lr_foid);
1606 
1607         return (0);
1608 }
1609 
1610 static int
1611 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1612 {
1613         objset_t *os = zd->zd_os;
1614         dmu_tx_t *tx;
1615         uint64_t txg;
1616         rl_t *rl;
1617 
1618         if (byteswap)
1619                 byteswap_uint64_array(lr, sizeof (*lr));
1620 
1621         ztest_object_lock(zd, lr->lr_foid, RL_READER);
1622         rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1623             RL_WRITER);
1624 
1625         tx = dmu_tx_create(os);
1626 
1627         dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1628 
1629         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1630         if (txg == 0) {
1631                 ztest_range_unlock(rl);
1632                 ztest_object_unlock(zd, lr->lr_foid);
1633                 return (ENOSPC);
1634         }
1635 
1636         VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1637             lr->lr_length, tx) == 0);
1638 
1639         (void) ztest_log_truncate(zd, tx, lr);
1640 
1641         dmu_tx_commit(tx);
1642 
1643         ztest_range_unlock(rl);
1644         ztest_object_unlock(zd, lr->lr_foid);
1645 
1646         return (0);
1647 }
1648 
1649 static int
1650 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1651 {
1652         objset_t *os = zd->zd_os;
1653         dmu_tx_t *tx;
1654         dmu_buf_t *db;
1655         ztest_block_tag_t *bbt;
1656         uint64_t txg, lrtxg, crtxg;
1657 
1658         if (byteswap)
1659                 byteswap_uint64_array(lr, sizeof (*lr));
1660 
1661         ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1662 
1663         VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1664 
1665         tx = dmu_tx_create(os);
1666         dmu_tx_hold_bonus(tx, lr->lr_foid);
1667 
1668         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1669         if (txg == 0) {
1670                 dmu_buf_rele(db, FTAG);
1671                 ztest_object_unlock(zd, lr->lr_foid);
1672                 return (ENOSPC);
1673         }
1674 
1675         bbt = ztest_bt_bonus(db);
1676         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1677         crtxg = bbt->bt_crtxg;
1678         lrtxg = lr->lr_common.lrc_txg;
1679 
1680         if (zd->zd_zilog->zl_replay) {
1681                 ASSERT(lr->lr_size != 0);
1682                 ASSERT(lr->lr_mode != 0);
1683                 ASSERT(lrtxg != 0);
1684         } else {
1685                 /*
1686                  * Randomly change the size and increment the generation.
1687                  */
1688                 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1689                     sizeof (*bbt);
1690                 lr->lr_mode = bbt->bt_gen + 1;
1691                 ASSERT(lrtxg == 0);
1692         }
1693 
1694         /*
1695          * Verify that the current bonus buffer is not newer than our txg.
1696          */
1697         ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1698             MAX(txg, lrtxg), crtxg);
1699 
1700         dmu_buf_will_dirty(db, tx);
1701 
1702         ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1703         ASSERT3U(lr->lr_size, <=, db->db_size);
1704         VERIFY3U(dmu_set_bonus(db, lr->lr_size, tx), ==, 0);
1705         bbt = ztest_bt_bonus(db);
1706 
1707         ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1708 
1709         dmu_buf_rele(db, FTAG);
1710 
1711         (void) ztest_log_setattr(zd, tx, lr);
1712 
1713         dmu_tx_commit(tx);
1714 
1715         ztest_object_unlock(zd, lr->lr_foid);
1716 
1717         return (0);
1718 }
1719 
1720 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1721         NULL,                   /* 0 no such transaction type */
1722         ztest_replay_create,    /* TX_CREATE */
1723         NULL,                   /* TX_MKDIR */
1724         NULL,                   /* TX_MKXATTR */
1725         NULL,                   /* TX_SYMLINK */
1726         ztest_replay_remove,    /* TX_REMOVE */
1727         NULL,                   /* TX_RMDIR */
1728         NULL,                   /* TX_LINK */
1729         NULL,                   /* TX_RENAME */
1730         ztest_replay_write,     /* TX_WRITE */
1731         ztest_replay_truncate,  /* TX_TRUNCATE */
1732         ztest_replay_setattr,   /* TX_SETATTR */
1733         NULL,                   /* TX_ACL */
1734         NULL,                   /* TX_CREATE_ACL */
1735         NULL,                   /* TX_CREATE_ATTR */
1736         NULL,                   /* TX_CREATE_ACL_ATTR */
1737         NULL,                   /* TX_MKDIR_ACL */
1738         NULL,                   /* TX_MKDIR_ATTR */
1739         NULL,                   /* TX_MKDIR_ACL_ATTR */
1740         NULL,                   /* TX_WRITE2 */
1741 };
1742 
1743 /*
1744  * ZIL get_data callbacks
1745  */
1746 
1747 static void
1748 ztest_get_done(zgd_t *zgd, int error)
1749 {
1750         ztest_ds_t *zd = zgd->zgd_private;
1751         uint64_t object = zgd->zgd_rl->rl_object;
1752 
1753         if (zgd->zgd_db)
1754                 dmu_buf_rele(zgd->zgd_db, zgd);
1755 
1756         ztest_range_unlock(zgd->zgd_rl);
1757         ztest_object_unlock(zd, object);
1758 
1759         if (error == 0 && zgd->zgd_bp)
1760                 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1761 
1762         umem_free(zgd, sizeof (*zgd));
1763 }
1764 
1765 static int
1766 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1767 {
1768         ztest_ds_t *zd = arg;
1769         objset_t *os = zd->zd_os;
1770         uint64_t object = lr->lr_foid;
1771         uint64_t offset = lr->lr_offset;
1772         uint64_t size = lr->lr_length;
1773         blkptr_t *bp = &lr->lr_blkptr;
1774         uint64_t txg = lr->lr_common.lrc_txg;
1775         uint64_t crtxg;
1776         dmu_object_info_t doi;
1777         dmu_buf_t *db;
1778         zgd_t *zgd;
1779         int error;
1780 
1781         ztest_object_lock(zd, object, RL_READER);
1782         error = dmu_bonus_hold(os, object, FTAG, &db);
1783         if (error) {
1784                 ztest_object_unlock(zd, object);
1785                 return (error);
1786         }
1787 
1788         crtxg = ztest_bt_bonus(db)->bt_crtxg;
1789 
1790         if (crtxg == 0 || crtxg > txg) {
1791                 dmu_buf_rele(db, FTAG);
1792                 ztest_object_unlock(zd, object);
1793                 return (ENOENT);
1794         }
1795 
1796         dmu_object_info_from_db(db, &doi);
1797         dmu_buf_rele(db, FTAG);
1798         db = NULL;
1799 
1800         zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1801         zgd->zgd_zilog = zd->zd_zilog;
1802         zgd->zgd_private = zd;
1803 
1804         if (buf != NULL) {      /* immediate write */
1805                 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1806                     RL_READER);
1807 
1808                 error = dmu_read(os, object, offset, size, buf,
1809                     DMU_READ_NO_PREFETCH);
1810                 ASSERT(error == 0);
1811         } else {
1812                 size = doi.doi_data_block_size;
1813                 if (ISP2(size)) {
1814                         offset = P2ALIGN(offset, size);
1815                 } else {
1816                         ASSERT(offset < size);
1817                         offset = 0;
1818                 }
1819 
1820                 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1821                     RL_READER);
1822 
1823                 error = dmu_buf_hold(os, object, offset, zgd, &db,
1824                     DMU_READ_NO_PREFETCH);
1825 
1826                 if (error == 0) {
1827                         zgd->zgd_db = db;
1828                         zgd->zgd_bp = bp;
1829 
1830                         ASSERT(db->db_offset == offset);
1831                         ASSERT(db->db_size == size);
1832 
1833                         error = dmu_sync(zio, lr->lr_common.lrc_txg,
1834                             ztest_get_done, zgd);
1835 
1836                         if (error == 0)
1837                                 return (0);
1838                 }
1839         }
1840 
1841         ztest_get_done(zgd, error);
1842 
1843         return (error);
1844 }
1845 
1846 static void *
1847 ztest_lr_alloc(size_t lrsize, char *name)
1848 {
1849         char *lr;
1850         size_t namesize = name ? strlen(name) + 1 : 0;
1851 
1852         lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1853 
1854         if (name)
1855                 bcopy(name, lr + lrsize, namesize);
1856 
1857         return (lr);
1858 }
1859 
1860 void
1861 ztest_lr_free(void *lr, size_t lrsize, char *name)
1862 {
1863         size_t namesize = name ? strlen(name) + 1 : 0;
1864 
1865         umem_free(lr, lrsize + namesize);
1866 }
1867 
1868 /*
1869  * Lookup a bunch of objects.  Returns the number of objects not found.
1870  */
1871 static int
1872 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1873 {
1874         int missing = 0;
1875         int error;
1876 
1877         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1878 
1879         for (int i = 0; i < count; i++, od++) {
1880                 od->od_object = 0;
1881                 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1882                     sizeof (uint64_t), 1, &od->od_object);
1883                 if (error) {
1884                         ASSERT(error == ENOENT);
1885                         ASSERT(od->od_object == 0);
1886                         missing++;
1887                 } else {
1888                         dmu_buf_t *db;
1889                         ztest_block_tag_t *bbt;
1890                         dmu_object_info_t doi;
1891 
1892                         ASSERT(od->od_object != 0);
1893                         ASSERT(missing == 0);   /* there should be no gaps */
1894 
1895                         ztest_object_lock(zd, od->od_object, RL_READER);
1896                         VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1897                             od->od_object, FTAG, &db));
1898                         dmu_object_info_from_db(db, &doi);
1899                         bbt = ztest_bt_bonus(db);
1900                         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1901                         od->od_type = doi.doi_type;
1902                         od->od_blocksize = doi.doi_data_block_size;
1903                         od->od_gen = bbt->bt_gen;
1904                         dmu_buf_rele(db, FTAG);
1905                         ztest_object_unlock(zd, od->od_object);
1906                 }
1907         }
1908 
1909         return (missing);
1910 }
1911 
1912 static int
1913 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1914 {
1915         int missing = 0;
1916 
1917         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1918 
1919         for (int i = 0; i < count; i++, od++) {
1920                 if (missing) {
1921                         od->od_object = 0;
1922                         missing++;
1923                         continue;
1924                 }
1925 
1926                 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1927 
1928                 lr->lr_doid = od->od_dir;
1929                 lr->lr_foid = 0;     /* 0 to allocate, > 0 to claim */
1930                 lr->lrz_type = od->od_crtype;
1931                 lr->lrz_blocksize = od->od_crblocksize;
1932                 lr->lrz_ibshift = ztest_random_ibshift();
1933                 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1934                 lr->lrz_bonuslen = dmu_bonus_max();
1935                 lr->lr_gen = od->od_crgen;
1936                 lr->lr_crtime[0] = time(NULL);
1937 
1938                 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1939                         ASSERT(missing == 0);
1940                         od->od_object = 0;
1941                         missing++;
1942                 } else {
1943                         od->od_object = lr->lr_foid;
1944                         od->od_type = od->od_crtype;
1945                         od->od_blocksize = od->od_crblocksize;
1946                         od->od_gen = od->od_crgen;
1947                         ASSERT(od->od_object != 0);
1948                 }
1949 
1950                 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1951         }
1952 
1953         return (missing);
1954 }
1955 
1956 static int
1957 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
1958 {
1959         int missing = 0;
1960         int error;
1961 
1962         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1963 
1964         od += count - 1;
1965 
1966         for (int i = count - 1; i >= 0; i--, od--) {
1967                 if (missing) {
1968                         missing++;
1969                         continue;
1970                 }
1971 
1972                 if (od->od_object == 0)
1973                         continue;
1974 
1975                 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1976 
1977                 lr->lr_doid = od->od_dir;
1978 
1979                 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
1980                         ASSERT3U(error, ==, ENOSPC);
1981                         missing++;
1982                 } else {
1983                         od->od_object = 0;
1984                 }
1985                 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1986         }
1987 
1988         return (missing);
1989 }
1990 
1991 static int
1992 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
1993     void *data)
1994 {
1995         lr_write_t *lr;
1996         int error;
1997 
1998         lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
1999 
2000         lr->lr_foid = object;
2001         lr->lr_offset = offset;
2002         lr->lr_length = size;
2003         lr->lr_blkoff = 0;
2004         BP_ZERO(&lr->lr_blkptr);
2005 
2006         bcopy(data, lr + 1, size);
2007 
2008         error = ztest_replay_write(zd, lr, B_FALSE);
2009 
2010         ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2011 
2012         return (error);
2013 }
2014 
2015 static int
2016 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2017 {
2018         lr_truncate_t *lr;
2019         int error;
2020 
2021         lr = ztest_lr_alloc(sizeof (*lr), NULL);
2022 
2023         lr->lr_foid = object;
2024         lr->lr_offset = offset;
2025         lr->lr_length = size;
2026 
2027         error = ztest_replay_truncate(zd, lr, B_FALSE);
2028 
2029         ztest_lr_free(lr, sizeof (*lr), NULL);
2030 
2031         return (error);
2032 }
2033 
2034 static int
2035 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2036 {
2037         lr_setattr_t *lr;
2038         int error;
2039 
2040         lr = ztest_lr_alloc(sizeof (*lr), NULL);
2041 
2042         lr->lr_foid = object;
2043         lr->lr_size = 0;
2044         lr->lr_mode = 0;
2045 
2046         error = ztest_replay_setattr(zd, lr, B_FALSE);
2047 
2048         ztest_lr_free(lr, sizeof (*lr), NULL);
2049 
2050         return (error);
2051 }
2052 
2053 static void
2054 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2055 {
2056         objset_t *os = zd->zd_os;
2057         dmu_tx_t *tx;
2058         uint64_t txg;
2059         rl_t *rl;
2060 
2061         txg_wait_synced(dmu_objset_pool(os), 0);
2062 
2063         ztest_object_lock(zd, object, RL_READER);
2064         rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2065 
2066         tx = dmu_tx_create(os);
2067 
2068         dmu_tx_hold_write(tx, object, offset, size);
2069 
2070         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2071 
2072         if (txg != 0) {
2073                 dmu_prealloc(os, object, offset, size, tx);
2074                 dmu_tx_commit(tx);
2075                 txg_wait_synced(dmu_objset_pool(os), txg);
2076         } else {
2077                 (void) dmu_free_long_range(os, object, offset, size);
2078         }
2079 
2080         ztest_range_unlock(rl);
2081         ztest_object_unlock(zd, object);
2082 }
2083 
2084 static void
2085 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2086 {
2087         ztest_block_tag_t wbt;
2088         dmu_object_info_t doi;
2089         enum ztest_io_type io_type;
2090         uint64_t blocksize;
2091         void *data;
2092 
2093         VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2094         blocksize = doi.doi_data_block_size;
2095         data = umem_alloc(blocksize, UMEM_NOFAIL);
2096 
2097         /*
2098          * Pick an i/o type at random, biased toward writing block tags.
2099          */
2100         io_type = ztest_random(ZTEST_IO_TYPES);
2101         if (ztest_random(2) == 0)
2102                 io_type = ZTEST_IO_WRITE_TAG;
2103 
2104         (void) rw_rdlock(&zd->zd_zilog_lock);
2105 
2106         switch (io_type) {
2107 
2108         case ZTEST_IO_WRITE_TAG:
2109                 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2110                 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2111                 break;
2112 
2113         case ZTEST_IO_WRITE_PATTERN:
2114                 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2115                 if (ztest_random(2) == 0) {
2116                         /*
2117                          * Induce fletcher2 collisions to ensure that
2118                          * zio_ddt_collision() detects and resolves them
2119                          * when using fletcher2-verify for deduplication.
2120                          */
2121                         ((uint64_t *)data)[0] ^= 1ULL << 63;
2122                         ((uint64_t *)data)[4] ^= 1ULL << 63;
2123                 }
2124                 (void) ztest_write(zd, object, offset, blocksize, data);
2125                 break;
2126 
2127         case ZTEST_IO_WRITE_ZEROES:
2128                 bzero(data, blocksize);
2129                 (void) ztest_write(zd, object, offset, blocksize, data);
2130                 break;
2131 
2132         case ZTEST_IO_TRUNCATE:
2133                 (void) ztest_truncate(zd, object, offset, blocksize);
2134                 break;
2135 
2136         case ZTEST_IO_SETATTR:
2137                 (void) ztest_setattr(zd, object);
2138                 break;
2139         }
2140 
2141         (void) rw_unlock(&zd->zd_zilog_lock);
2142 
2143         umem_free(data, blocksize);
2144 }
2145 
2146 /*
2147  * Initialize an object description template.
2148  */
2149 static void
2150 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2151     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2152 {
2153         od->od_dir = ZTEST_DIROBJ;
2154         od->od_object = 0;
2155 
2156         od->od_crtype = type;
2157         od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2158         od->od_crgen = gen;
2159 
2160         od->od_type = DMU_OT_NONE;
2161         od->od_blocksize = 0;
2162         od->od_gen = 0;
2163 
2164         (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2165             tag, (int64_t)id, index);
2166 }
2167 
2168 /*
2169  * Lookup or create the objects for a test using the od template.
2170  * If the objects do not all exist, or if 'remove' is specified,
2171  * remove any existing objects and create new ones.  Otherwise,
2172  * use the existing objects.
2173  */
2174 static int
2175 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2176 {
2177         int count = size / sizeof (*od);
2178         int rv = 0;
2179 
2180         VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2181         if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2182             (ztest_remove(zd, od, count) != 0 ||
2183             ztest_create(zd, od, count) != 0))
2184                 rv = -1;
2185         zd->zd_od = od;
2186         VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2187 
2188         return (rv);
2189 }
2190 
2191 /* ARGSUSED */
2192 void
2193 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2194 {
2195         zilog_t *zilog = zd->zd_zilog;
2196 
2197         (void) rw_rdlock(&zd->zd_zilog_lock);
2198 
2199         zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2200 
2201         /*
2202          * Remember the committed values in zd, which is in parent/child
2203          * shared memory.  If we die, the next iteration of ztest_run()
2204          * will verify that the log really does contain this record.
2205          */
2206         mutex_enter(&zilog->zl_lock);
2207         ASSERT(zd->zd_shared != NULL);
2208         ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2209         zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2210         mutex_exit(&zilog->zl_lock);
2211 
2212         (void) rw_unlock(&zd->zd_zilog_lock);
2213 }
2214 
2215 /*
2216  * This function is designed to simulate the operations that occur during a
2217  * mount/unmount operation.  We hold the dataset across these operations in an
2218  * attempt to expose any implicit assumptions about ZIL management.
2219  */
2220 /* ARGSUSED */
2221 void
2222 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2223 {
2224         objset_t *os = zd->zd_os;
2225 
2226         (void) rw_wrlock(&zd->zd_zilog_lock);
2227 
2228         /* zfsvfs_teardown() */
2229         zil_close(zd->zd_zilog);
2230 
2231         /* zfsvfs_setup() */
2232         VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2233         zil_replay(os, zd, ztest_replay_vector);
2234 
2235         (void) rw_unlock(&zd->zd_zilog_lock);
2236 }
2237 
2238 /*
2239  * Verify that we can't destroy an active pool, create an existing pool,
2240  * or create a pool with a bad vdev spec.
2241  */
2242 /* ARGSUSED */
2243 void
2244 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2245 {
2246         ztest_shared_opts_t *zo = &ztest_opts;
2247         spa_t *spa;
2248         nvlist_t *nvroot;
2249 
2250         /*
2251          * Attempt to create using a bad file.
2252          */
2253         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2254         VERIFY3U(ENOENT, ==,
2255             spa_create("ztest_bad_file", nvroot, NULL, NULL));
2256         nvlist_free(nvroot);
2257 
2258         /*
2259          * Attempt to create using a bad mirror.
2260          */
2261         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1);
2262         VERIFY3U(ENOENT, ==,
2263             spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2264         nvlist_free(nvroot);
2265 
2266         /*
2267          * Attempt to create an existing pool.  It shouldn't matter
2268          * what's in the nvroot; we should fail with EEXIST.
2269          */
2270         (void) rw_rdlock(&ztest_name_lock);
2271         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2272         VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2273         nvlist_free(nvroot);
2274         VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2275         VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2276         spa_close(spa, FTAG);
2277 
2278         (void) rw_unlock(&ztest_name_lock);
2279 }
2280 
2281 static vdev_t *
2282 vdev_lookup_by_path(vdev_t *vd, const char *path)
2283 {
2284         vdev_t *mvd;
2285 
2286         if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2287                 return (vd);
2288 
2289         for (int c = 0; c < vd->vdev_children; c++)
2290                 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2291                     NULL)
2292                         return (mvd);
2293 
2294         return (NULL);
2295 }
2296 
2297 /*
2298  * Find the first available hole which can be used as a top-level.
2299  */
2300 int
2301 find_vdev_hole(spa_t *spa)
2302 {
2303         vdev_t *rvd = spa->spa_root_vdev;
2304         int c;
2305 
2306         ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2307 
2308         for (c = 0; c < rvd->vdev_children; c++) {
2309                 vdev_t *cvd = rvd->vdev_child[c];
2310 
2311                 if (cvd->vdev_ishole)
2312                         break;
2313         }
2314         return (c);
2315 }
2316 
2317 /*
2318  * Verify that vdev_add() works as expected.
2319  */
2320 /* ARGSUSED */
2321 void
2322 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2323 {
2324         ztest_shared_t *zs = ztest_shared;
2325         spa_t *spa = ztest_spa;
2326         uint64_t leaves;
2327         uint64_t guid;
2328         nvlist_t *nvroot;
2329         int error;
2330 
2331         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2332         leaves =
2333             MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2334 
2335         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2336 
2337         ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2338 
2339         /*
2340          * If we have slogs then remove them 1/4 of the time.
2341          */
2342         if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2343                 /*
2344                  * Grab the guid from the head of the log class rotor.
2345                  */
2346                 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2347 
2348                 spa_config_exit(spa, SCL_VDEV, FTAG);
2349 
2350                 /*
2351                  * We have to grab the zs_name_lock as writer to
2352                  * prevent a race between removing a slog (dmu_objset_find)
2353                  * and destroying a dataset. Removing the slog will
2354                  * grab a reference on the dataset which may cause
2355                  * dmu_objset_destroy() to fail with EBUSY thus
2356                  * leaving the dataset in an inconsistent state.
2357                  */
2358                 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2359                 error = spa_vdev_remove(spa, guid, B_FALSE);
2360                 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2361 
2362                 if (error && error != EEXIST)
2363                         fatal(0, "spa_vdev_remove() = %d", error);
2364         } else {
2365                 spa_config_exit(spa, SCL_VDEV, FTAG);
2366 
2367                 /*
2368                  * Make 1/4 of the devices be log devices.
2369                  */
2370                 nvroot = make_vdev_root(NULL, NULL,
2371                     ztest_opts.zo_vdev_size, 0,
2372                     ztest_random(4) == 0, ztest_opts.zo_raidz,
2373                     zs->zs_mirrors, 1);
2374 
2375                 error = spa_vdev_add(spa, nvroot);
2376                 nvlist_free(nvroot);
2377 
2378                 if (error == ENOSPC)
2379                         ztest_record_enospc("spa_vdev_add");
2380                 else if (error != 0)
2381                         fatal(0, "spa_vdev_add() = %d", error);
2382         }
2383 
2384         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2385 }
2386 
2387 /*
2388  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2389  */
2390 /* ARGSUSED */
2391 void
2392 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2393 {
2394         ztest_shared_t *zs = ztest_shared;
2395         spa_t *spa = ztest_spa;
2396         vdev_t *rvd = spa->spa_root_vdev;
2397         spa_aux_vdev_t *sav;
2398         char *aux;
2399         uint64_t guid = 0;
2400         int error;
2401 
2402         if (ztest_random(2) == 0) {
2403                 sav = &spa->spa_spares;
2404                 aux = ZPOOL_CONFIG_SPARES;
2405         } else {
2406                 sav = &spa->spa_l2cache;
2407                 aux = ZPOOL_CONFIG_L2CACHE;
2408         }
2409 
2410         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2411 
2412         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2413 
2414         if (sav->sav_count != 0 && ztest_random(4) == 0) {
2415                 /*
2416                  * Pick a random device to remove.
2417                  */
2418                 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2419         } else {
2420                 /*
2421                  * Find an unused device we can add.
2422                  */
2423                 zs->zs_vdev_aux = 0;
2424                 for (;;) {
2425                         char path[MAXPATHLEN];
2426                         int c;
2427                         (void) snprintf(path, sizeof (path), ztest_aux_template,
2428                             ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2429                             zs->zs_vdev_aux);
2430                         for (c = 0; c < sav->sav_count; c++)
2431                                 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2432                                     path) == 0)
2433                                         break;
2434                         if (c == sav->sav_count &&
2435                             vdev_lookup_by_path(rvd, path) == NULL)
2436                                 break;
2437                         zs->zs_vdev_aux++;
2438                 }
2439         }
2440 
2441         spa_config_exit(spa, SCL_VDEV, FTAG);
2442 
2443         if (guid == 0) {
2444                 /*
2445                  * Add a new device.
2446                  */
2447                 nvlist_t *nvroot = make_vdev_root(NULL, aux,
2448                     (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2449                 error = spa_vdev_add(spa, nvroot);
2450                 if (error != 0)
2451                         fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2452                 nvlist_free(nvroot);
2453         } else {
2454                 /*
2455                  * Remove an existing device.  Sometimes, dirty its
2456                  * vdev state first to make sure we handle removal
2457                  * of devices that have pending state changes.
2458                  */
2459                 if (ztest_random(2) == 0)
2460                         (void) vdev_online(spa, guid, 0, NULL);
2461 
2462                 error = spa_vdev_remove(spa, guid, B_FALSE);
2463                 if (error != 0 && error != EBUSY)
2464                         fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2465         }
2466 
2467         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2468 }
2469 
2470 /*
2471  * split a pool if it has mirror tlvdevs
2472  */
2473 /* ARGSUSED */
2474 void
2475 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2476 {
2477         ztest_shared_t *zs = ztest_shared;
2478         spa_t *spa = ztest_spa;
2479         vdev_t *rvd = spa->spa_root_vdev;
2480         nvlist_t *tree, **child, *config, *split, **schild;
2481         uint_t c, children, schildren = 0, lastlogid = 0;
2482         int error = 0;
2483 
2484         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2485 
2486         /* ensure we have a useable config; mirrors of raidz aren't supported */
2487         if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2488                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2489                 return;
2490         }
2491 
2492         /* clean up the old pool, if any */
2493         (void) spa_destroy("splitp");
2494 
2495         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2496 
2497         /* generate a config from the existing config */
2498         mutex_enter(&spa->spa_props_lock);
2499         VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2500             &tree) == 0);
2501         mutex_exit(&spa->spa_props_lock);
2502 
2503         VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2504             &children) == 0);
2505 
2506         schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2507         for (c = 0; c < children; c++) {
2508                 vdev_t *tvd = rvd->vdev_child[c];
2509                 nvlist_t **mchild;
2510                 uint_t mchildren;
2511 
2512                 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2513                         VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2514                             0) == 0);
2515                         VERIFY(nvlist_add_string(schild[schildren],
2516                             ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2517                         VERIFY(nvlist_add_uint64(schild[schildren],
2518                             ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2519                         if (lastlogid == 0)
2520                                 lastlogid = schildren;
2521                         ++schildren;
2522                         continue;
2523                 }
2524                 lastlogid = 0;
2525                 VERIFY(nvlist_lookup_nvlist_array(child[c],
2526                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2527                 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2528         }
2529 
2530         /* OK, create a config that can be used to split */
2531         VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2532         VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2533             VDEV_TYPE_ROOT) == 0);
2534         VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2535             lastlogid != 0 ? lastlogid : schildren) == 0);
2536 
2537         VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2538         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2539 
2540         for (c = 0; c < schildren; c++)
2541                 nvlist_free(schild[c]);
2542         free(schild);
2543         nvlist_free(split);
2544 
2545         spa_config_exit(spa, SCL_VDEV, FTAG);
2546 
2547         (void) rw_wrlock(&ztest_name_lock);
2548         error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2549         (void) rw_unlock(&ztest_name_lock);
2550 
2551         nvlist_free(config);
2552 
2553         if (error == 0) {
2554                 (void) printf("successful split - results:\n");
2555                 mutex_enter(&spa_namespace_lock);
2556                 show_pool_stats(spa);
2557                 show_pool_stats(spa_lookup("splitp"));
2558                 mutex_exit(&spa_namespace_lock);
2559                 ++zs->zs_splits;
2560                 --zs->zs_mirrors;
2561         }
2562         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2563 
2564 }
2565 
2566 /*
2567  * Verify that we can attach and detach devices.
2568  */
2569 /* ARGSUSED */
2570 void
2571 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2572 {
2573         ztest_shared_t *zs = ztest_shared;
2574         spa_t *spa = ztest_spa;
2575         spa_aux_vdev_t *sav = &spa->spa_spares;
2576         vdev_t *rvd = spa->spa_root_vdev;
2577         vdev_t *oldvd, *newvd, *pvd;
2578         nvlist_t *root;
2579         uint64_t leaves;
2580         uint64_t leaf, top;
2581         uint64_t ashift = ztest_get_ashift();
2582         uint64_t oldguid, pguid;
2583         size_t oldsize, newsize;
2584         char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2585         int replacing;
2586         int oldvd_has_siblings = B_FALSE;
2587         int newvd_is_spare = B_FALSE;
2588         int oldvd_is_log;
2589         int error, expected_error;
2590 
2591         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2592         leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2593 
2594         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2595 
2596         /*
2597          * Decide whether to do an attach or a replace.
2598          */
2599         replacing = ztest_random(2);
2600 
2601         /*
2602          * Pick a random top-level vdev.
2603          */
2604         top = ztest_random_vdev_top(spa, B_TRUE);
2605 
2606         /*
2607          * Pick a random leaf within it.
2608          */
2609         leaf = ztest_random(leaves);
2610 
2611         /*
2612          * Locate this vdev.
2613          */
2614         oldvd = rvd->vdev_child[top];
2615         if (zs->zs_mirrors >= 1) {
2616                 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2617                 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2618                 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2619         }
2620         if (ztest_opts.zo_raidz > 1) {
2621                 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2622                 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2623                 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2624         }
2625 
2626         /*
2627          * If we're already doing an attach or replace, oldvd may be a
2628          * mirror vdev -- in which case, pick a random child.
2629          */
2630         while (oldvd->vdev_children != 0) {
2631                 oldvd_has_siblings = B_TRUE;
2632                 ASSERT(oldvd->vdev_children >= 2);
2633                 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2634         }
2635 
2636         oldguid = oldvd->vdev_guid;
2637         oldsize = vdev_get_min_asize(oldvd);
2638         oldvd_is_log = oldvd->vdev_top->vdev_islog;
2639         (void) strcpy(oldpath, oldvd->vdev_path);
2640         pvd = oldvd->vdev_parent;
2641         pguid = pvd->vdev_guid;
2642 
2643         /*
2644          * If oldvd has siblings, then half of the time, detach it.
2645          */
2646         if (oldvd_has_siblings && ztest_random(2) == 0) {
2647                 spa_config_exit(spa, SCL_VDEV, FTAG);
2648                 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2649                 if (error != 0 && error != ENODEV && error != EBUSY &&
2650                     error != ENOTSUP)
2651                         fatal(0, "detach (%s) returned %d", oldpath, error);
2652                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2653                 return;
2654         }
2655 
2656         /*
2657          * For the new vdev, choose with equal probability between the two
2658          * standard paths (ending in either 'a' or 'b') or a random hot spare.
2659          */
2660         if (sav->sav_count != 0 && ztest_random(3) == 0) {
2661                 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2662                 newvd_is_spare = B_TRUE;
2663                 (void) strcpy(newpath, newvd->vdev_path);
2664         } else {
2665                 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2666                     ztest_opts.zo_dir, ztest_opts.zo_pool,
2667                     top * leaves + leaf);
2668                 if (ztest_random(2) == 0)
2669                         newpath[strlen(newpath) - 1] = 'b';
2670                 newvd = vdev_lookup_by_path(rvd, newpath);
2671         }
2672 
2673         if (newvd) {
2674                 newsize = vdev_get_min_asize(newvd);
2675         } else {
2676                 /*
2677                  * Make newsize a little bigger or smaller than oldsize.
2678                  * If it's smaller, the attach should fail.
2679                  * If it's larger, and we're doing a replace,
2680                  * we should get dynamic LUN growth when we're done.
2681                  */
2682                 newsize = 10 * oldsize / (9 + ztest_random(3));
2683         }
2684 
2685         /*
2686          * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2687          * unless it's a replace; in that case any non-replacing parent is OK.
2688          *
2689          * If newvd is already part of the pool, it should fail with EBUSY.
2690          *
2691          * If newvd is too small, it should fail with EOVERFLOW.
2692          */
2693         if (pvd->vdev_ops != &vdev_mirror_ops &&
2694             pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2695             pvd->vdev_ops == &vdev_replacing_ops ||
2696             pvd->vdev_ops == &vdev_spare_ops))
2697                 expected_error = ENOTSUP;
2698         else if (newvd_is_spare && (!replacing || oldvd_is_log))
2699                 expected_error = ENOTSUP;
2700         else if (newvd == oldvd)
2701                 expected_error = replacing ? 0 : EBUSY;
2702         else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2703                 expected_error = EBUSY;
2704         else if (newsize < oldsize)
2705                 expected_error = EOVERFLOW;
2706         else if (ashift > oldvd->vdev_top->vdev_ashift)
2707                 expected_error = EDOM;
2708         else
2709                 expected_error = 0;
2710 
2711         spa_config_exit(spa, SCL_VDEV, FTAG);
2712 
2713         /*
2714          * Build the nvlist describing newpath.
2715          */
2716         root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0,
2717             ashift, 0, 0, 0, 1);
2718 
2719         error = spa_vdev_attach(spa, oldguid, root, replacing);
2720 
2721         nvlist_free(root);
2722 
2723         /*
2724          * If our parent was the replacing vdev, but the replace completed,
2725          * then instead of failing with ENOTSUP we may either succeed,
2726          * fail with ENODEV, or fail with EOVERFLOW.
2727          */
2728         if (expected_error == ENOTSUP &&
2729             (error == 0 || error == ENODEV || error == EOVERFLOW))
2730                 expected_error = error;
2731 
2732         /*
2733          * If someone grew the LUN, the replacement may be too small.
2734          */
2735         if (error == EOVERFLOW || error == EBUSY)
2736                 expected_error = error;
2737 
2738         /* XXX workaround 6690467 */
2739         if (error != expected_error && expected_error != EBUSY) {
2740                 fatal(0, "attach (%s %llu, %s %llu, %d) "
2741                     "returned %d, expected %d",
2742                     oldpath, (longlong_t)oldsize, newpath,
2743                     (longlong_t)newsize, replacing, error, expected_error);
2744         }
2745 
2746         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2747 }
2748 
2749 /*
2750  * Callback function which expands the physical size of the vdev.
2751  */
2752 vdev_t *
2753 grow_vdev(vdev_t *vd, void *arg)
2754 {
2755         spa_t *spa = vd->vdev_spa;
2756         size_t *newsize = arg;
2757         size_t fsize;
2758         int fd;
2759 
2760         ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2761         ASSERT(vd->vdev_ops->vdev_op_leaf);
2762 
2763         if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2764                 return (vd);
2765 
2766         fsize = lseek(fd, 0, SEEK_END);
2767         (void) ftruncate(fd, *newsize);
2768 
2769         if (ztest_opts.zo_verbose >= 6) {
2770                 (void) printf("%s grew from %lu to %lu bytes\n",
2771                     vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2772         }
2773         (void) close(fd);
2774         return (NULL);
2775 }
2776 
2777 /*
2778  * Callback function which expands a given vdev by calling vdev_online().
2779  */
2780 /* ARGSUSED */
2781 vdev_t *
2782 online_vdev(vdev_t *vd, void *arg)
2783 {
2784         spa_t *spa = vd->vdev_spa;
2785         vdev_t *tvd = vd->vdev_top;
2786         uint64_t guid = vd->vdev_guid;
2787         uint64_t generation = spa->spa_config_generation + 1;
2788         vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2789         int error;
2790 
2791         ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2792         ASSERT(vd->vdev_ops->vdev_op_leaf);
2793 
2794         /* Calling vdev_online will initialize the new metaslabs */
2795         spa_config_exit(spa, SCL_STATE, spa);
2796         error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2797         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2798 
2799         /*
2800          * If vdev_online returned an error or the underlying vdev_open
2801          * failed then we abort the expand. The only way to know that
2802          * vdev_open fails is by checking the returned newstate.
2803          */
2804         if (error || newstate != VDEV_STATE_HEALTHY) {
2805                 if (ztest_opts.zo_verbose >= 5) {
2806                         (void) printf("Unable to expand vdev, state %llu, "
2807                             "error %d\n", (u_longlong_t)newstate, error);
2808                 }
2809                 return (vd);
2810         }
2811         ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2812 
2813         /*
2814          * Since we dropped the lock we need to ensure that we're
2815          * still talking to the original vdev. It's possible this
2816          * vdev may have been detached/replaced while we were
2817          * trying to online it.
2818          */
2819         if (generation != spa->spa_config_generation) {
2820                 if (ztest_opts.zo_verbose >= 5) {
2821                         (void) printf("vdev configuration has changed, "
2822                             "guid %llu, state %llu, expected gen %llu, "
2823                             "got gen %llu\n",
2824                             (u_longlong_t)guid,
2825                             (u_longlong_t)tvd->vdev_state,
2826                             (u_longlong_t)generation,
2827                             (u_longlong_t)spa->spa_config_generation);
2828                 }
2829                 return (vd);
2830         }
2831         return (NULL);
2832 }
2833 
2834 /*
2835  * Traverse the vdev tree calling the supplied function.
2836  * We continue to walk the tree until we either have walked all
2837  * children or we receive a non-NULL return from the callback.
2838  * If a NULL callback is passed, then we just return back the first
2839  * leaf vdev we encounter.
2840  */
2841 vdev_t *
2842 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2843 {
2844         if (vd->vdev_ops->vdev_op_leaf) {
2845                 if (func == NULL)
2846                         return (vd);
2847                 else
2848                         return (func(vd, arg));
2849         }
2850 
2851         for (uint_t c = 0; c < vd->vdev_children; c++) {
2852                 vdev_t *cvd = vd->vdev_child[c];
2853                 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
2854                         return (cvd);
2855         }
2856         return (NULL);
2857 }
2858 
2859 /*
2860  * Verify that dynamic LUN growth works as expected.
2861  */
2862 /* ARGSUSED */
2863 void
2864 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
2865 {
2866         spa_t *spa = ztest_spa;
2867         vdev_t *vd, *tvd;
2868         metaslab_class_t *mc;
2869         metaslab_group_t *mg;
2870         size_t psize, newsize;
2871         uint64_t top;
2872         uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
2873 
2874         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2875         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2876 
2877         top = ztest_random_vdev_top(spa, B_TRUE);
2878 
2879         tvd = spa->spa_root_vdev->vdev_child[top];
2880         mg = tvd->vdev_mg;
2881         mc = mg->mg_class;
2882         old_ms_count = tvd->vdev_ms_count;
2883         old_class_space = metaslab_class_get_space(mc);
2884 
2885         /*
2886          * Determine the size of the first leaf vdev associated with
2887          * our top-level device.
2888          */
2889         vd = vdev_walk_tree(tvd, NULL, NULL);
2890         ASSERT3P(vd, !=, NULL);
2891         ASSERT(vd->vdev_ops->vdev_op_leaf);
2892 
2893         psize = vd->vdev_psize;
2894 
2895         /*
2896          * We only try to expand the vdev if it's healthy, less than 4x its
2897          * original size, and it has a valid psize.
2898          */
2899         if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
2900             psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
2901                 spa_config_exit(spa, SCL_STATE, spa);
2902                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2903                 return;
2904         }
2905         ASSERT(psize > 0);
2906         newsize = psize + psize / 8;
2907         ASSERT3U(newsize, >, psize);
2908 
2909         if (ztest_opts.zo_verbose >= 6) {
2910                 (void) printf("Expanding LUN %s from %lu to %lu\n",
2911                     vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
2912         }
2913 
2914         /*
2915          * Growing the vdev is a two step process:
2916          *      1). expand the physical size (i.e. relabel)
2917          *      2). online the vdev to create the new metaslabs
2918          */
2919         if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
2920             vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
2921             tvd->vdev_state != VDEV_STATE_HEALTHY) {
2922                 if (ztest_opts.zo_verbose >= 5) {
2923                         (void) printf("Could not expand LUN because "
2924                             "the vdev configuration changed.\n");
2925                 }
2926                 spa_config_exit(spa, SCL_STATE, spa);
2927                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2928                 return;
2929         }
2930 
2931         spa_config_exit(spa, SCL_STATE, spa);
2932 
2933         /*
2934          * Expanding the LUN will update the config asynchronously,
2935          * thus we must wait for the async thread to complete any
2936          * pending tasks before proceeding.
2937          */
2938         for (;;) {
2939                 boolean_t done;
2940                 mutex_enter(&spa->spa_async_lock);
2941                 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
2942                 mutex_exit(&spa->spa_async_lock);
2943                 if (done)
2944                         break;
2945                 txg_wait_synced(spa_get_dsl(spa), 0);
2946                 (void) poll(NULL, 0, 100);
2947         }
2948 
2949         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2950 
2951         tvd = spa->spa_root_vdev->vdev_child[top];
2952         new_ms_count = tvd->vdev_ms_count;
2953         new_class_space = metaslab_class_get_space(mc);
2954 
2955         if (tvd->vdev_mg != mg || mg->mg_class != mc) {
2956                 if (ztest_opts.zo_verbose >= 5) {
2957                         (void) printf("Could not verify LUN expansion due to "
2958                             "intervening vdev offline or remove.\n");
2959                 }
2960                 spa_config_exit(spa, SCL_STATE, spa);
2961                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2962                 return;
2963         }
2964 
2965         /*
2966          * Make sure we were able to grow the vdev.
2967          */
2968         if (new_ms_count <= old_ms_count)
2969                 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
2970                     old_ms_count, new_ms_count);
2971 
2972         /*
2973          * Make sure we were able to grow the pool.
2974          */
2975         if (new_class_space <= old_class_space)
2976                 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
2977                     old_class_space, new_class_space);
2978 
2979         if (ztest_opts.zo_verbose >= 5) {
2980                 char oldnumbuf[6], newnumbuf[6];
2981 
2982                 nicenum(old_class_space, oldnumbuf);
2983                 nicenum(new_class_space, newnumbuf);
2984                 (void) printf("%s grew from %s to %s\n",
2985                     spa->spa_name, oldnumbuf, newnumbuf);
2986         }
2987 
2988         spa_config_exit(spa, SCL_STATE, spa);
2989         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2990 }
2991 
2992 /*
2993  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
2994  */
2995 /* ARGSUSED */
2996 static void
2997 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
2998 {
2999         /*
3000          * Create the objects common to all ztest datasets.
3001          */
3002         VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3003             DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3004 }
3005 
3006 static int
3007 ztest_dataset_create(char *dsname)
3008 {
3009         uint64_t zilset = ztest_random(100);
3010         int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3011             ztest_objset_create_cb, NULL);
3012 
3013         if (err || zilset < 80)
3014                 return (err);
3015 
3016         if (ztest_opts.zo_verbose >= 6)
3017                 (void) printf("Setting dataset %s to sync always\n", dsname);
3018         return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3019             ZFS_SYNC_ALWAYS, B_FALSE));
3020 }
3021 
3022 /* ARGSUSED */
3023 static int
3024 ztest_objset_destroy_cb(const char *name, void *arg)
3025 {
3026         objset_t *os;
3027         dmu_object_info_t doi;
3028         int error;
3029 
3030         /*
3031          * Verify that the dataset contains a directory object.
3032          */
3033         VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os));
3034         error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3035         if (error != ENOENT) {
3036                 /* We could have crashed in the middle of destroying it */
3037                 ASSERT3U(error, ==, 0);
3038                 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3039                 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3040         }
3041         dmu_objset_rele(os, FTAG);
3042 
3043         /*
3044          * Destroy the dataset.
3045          */
3046         VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE));
3047         return (0);
3048 }
3049 
3050 static boolean_t
3051 ztest_snapshot_create(char *osname, uint64_t id)
3052 {
3053         char snapname[MAXNAMELEN];
3054         int error;
3055 
3056         (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3057             (u_longlong_t)id);
3058 
3059         error = dmu_objset_snapshot_one(osname, strchr(snapname, '@') + 1);
3060         if (error == ENOSPC) {
3061                 ztest_record_enospc(FTAG);
3062                 return (B_FALSE);
3063         }
3064         if (error != 0 && error != EEXIST)
3065                 fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error);
3066         return (B_TRUE);
3067 }
3068 
3069 static boolean_t
3070 ztest_snapshot_destroy(char *osname, uint64_t id)
3071 {
3072         char snapname[MAXNAMELEN];
3073         int error;
3074 
3075         (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3076             (u_longlong_t)id);
3077 
3078         error = dmu_objset_destroy(snapname, B_FALSE);
3079         if (error != 0 && error != ENOENT)
3080                 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3081         return (B_TRUE);
3082 }
3083 
3084 /* ARGSUSED */
3085 void
3086 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3087 {
3088         ztest_ds_t zdtmp;
3089         int iters;
3090         int error;
3091         objset_t *os, *os2;
3092         char name[MAXNAMELEN];
3093         zilog_t *zilog;
3094 
3095         (void) rw_rdlock(&ztest_name_lock);
3096 
3097         (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3098             ztest_opts.zo_pool, (u_longlong_t)id);
3099 
3100         /*
3101          * If this dataset exists from a previous run, process its replay log
3102          * half of the time.  If we don't replay it, then dmu_objset_destroy()
3103          * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3104          */
3105         if (ztest_random(2) == 0 &&
3106             dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3107                 ztest_zd_init(&zdtmp, NULL, os);
3108                 zil_replay(os, &zdtmp, ztest_replay_vector);
3109                 ztest_zd_fini(&zdtmp);
3110                 dmu_objset_disown(os, FTAG);
3111         }
3112 
3113         /*
3114          * There may be an old instance of the dataset we're about to
3115          * create lying around from a previous run.  If so, destroy it
3116          * and all of its snapshots.
3117          */
3118         (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3119             DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3120 
3121         /*
3122          * Verify that the destroyed dataset is no longer in the namespace.
3123          */
3124         VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os));
3125 
3126         /*
3127          * Verify that we can create a new dataset.
3128          */
3129         error = ztest_dataset_create(name);
3130         if (error) {
3131                 if (error == ENOSPC) {
3132                         ztest_record_enospc(FTAG);
3133                         (void) rw_unlock(&ztest_name_lock);
3134                         return;
3135                 }
3136                 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3137         }
3138 
3139         VERIFY3U(0, ==,
3140             dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3141 
3142         ztest_zd_init(&zdtmp, NULL, os);
3143 
3144         /*
3145          * Open the intent log for it.
3146          */
3147         zilog = zil_open(os, ztest_get_data);
3148 
3149         /*
3150          * Put some objects in there, do a little I/O to them,
3151          * and randomly take a couple of snapshots along the way.
3152          */
3153         iters = ztest_random(5);
3154         for (int i = 0; i < iters; i++) {
3155                 ztest_dmu_object_alloc_free(&zdtmp, id);
3156                 if (ztest_random(iters) == 0)
3157                         (void) ztest_snapshot_create(name, i);
3158         }
3159 
3160         /*
3161          * Verify that we cannot create an existing dataset.
3162          */
3163         VERIFY3U(EEXIST, ==,
3164             dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3165 
3166         /*
3167          * Verify that we can hold an objset that is also owned.
3168          */
3169         VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3170         dmu_objset_rele(os2, FTAG);
3171 
3172         /*
3173          * Verify that we cannot own an objset that is already owned.
3174          */
3175         VERIFY3U(EBUSY, ==,
3176             dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3177 
3178         zil_close(zilog);
3179         dmu_objset_disown(os, FTAG);
3180         ztest_zd_fini(&zdtmp);
3181 
3182         (void) rw_unlock(&ztest_name_lock);
3183 }
3184 
3185 /*
3186  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3187  */
3188 void
3189 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3190 {
3191         (void) rw_rdlock(&ztest_name_lock);
3192         (void) ztest_snapshot_destroy(zd->zd_name, id);
3193         (void) ztest_snapshot_create(zd->zd_name, id);
3194         (void) rw_unlock(&ztest_name_lock);
3195 }
3196 
3197 /*
3198  * Cleanup non-standard snapshots and clones.
3199  */
3200 void
3201 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3202 {
3203         char snap1name[MAXNAMELEN];
3204         char clone1name[MAXNAMELEN];
3205         char snap2name[MAXNAMELEN];
3206         char clone2name[MAXNAMELEN];
3207         char snap3name[MAXNAMELEN];
3208         int error;
3209 
3210         (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3211         (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3212         (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3213         (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3214         (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3215 
3216         error = dmu_objset_destroy(clone2name, B_FALSE);
3217         if (error && error != ENOENT)
3218                 fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error);
3219         error = dmu_objset_destroy(snap3name, B_FALSE);
3220         if (error && error != ENOENT)
3221                 fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error);
3222         error = dmu_objset_destroy(snap2name, B_FALSE);
3223         if (error && error != ENOENT)
3224                 fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error);
3225         error = dmu_objset_destroy(clone1name, B_FALSE);
3226         if (error && error != ENOENT)
3227                 fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error);
3228         error = dmu_objset_destroy(snap1name, B_FALSE);
3229         if (error && error != ENOENT)
3230                 fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error);
3231 }
3232 
3233 /*
3234  * Verify dsl_dataset_promote handles EBUSY
3235  */
3236 void
3237 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3238 {
3239         objset_t *clone;
3240         dsl_dataset_t *ds;
3241         char snap1name[MAXNAMELEN];
3242         char clone1name[MAXNAMELEN];
3243         char snap2name[MAXNAMELEN];
3244         char clone2name[MAXNAMELEN];
3245         char snap3name[MAXNAMELEN];
3246         char *osname = zd->zd_name;
3247         int error;
3248 
3249         (void) rw_rdlock(&ztest_name_lock);
3250 
3251         ztest_dsl_dataset_cleanup(osname, id);
3252 
3253         (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3254         (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3255         (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3256         (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3257         (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3258 
3259         error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3260         if (error && error != EEXIST) {
3261                 if (error == ENOSPC) {
3262                         ztest_record_enospc(FTAG);
3263                         goto out;
3264                 }
3265                 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3266         }
3267 
3268         error = dmu_objset_hold(snap1name, FTAG, &clone);
3269         if (error)
3270                 fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error);
3271 
3272         error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0);
3273         dmu_objset_rele(clone, FTAG);
3274         if (error) {
3275                 if (error == ENOSPC) {
3276                         ztest_record_enospc(FTAG);
3277                         goto out;
3278                 }
3279                 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3280         }
3281 
3282         error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3283         if (error && error != EEXIST) {
3284                 if (error == ENOSPC) {
3285                         ztest_record_enospc(FTAG);
3286                         goto out;
3287                 }
3288                 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3289         }
3290 
3291         error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3292         if (error && error != EEXIST) {
3293                 if (error == ENOSPC) {
3294                         ztest_record_enospc(FTAG);
3295                         goto out;
3296                 }
3297                 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3298         }
3299 
3300         error = dmu_objset_hold(snap3name, FTAG, &clone);
3301         if (error)
3302                 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3303 
3304         error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0);
3305         dmu_objset_rele(clone, FTAG);
3306         if (error) {
3307                 if (error == ENOSPC) {
3308                         ztest_record_enospc(FTAG);
3309                         goto out;
3310                 }
3311                 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3312         }
3313 
3314         error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds);
3315         if (error)
3316                 fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error);
3317         error = dsl_dataset_promote(clone2name, NULL);
3318         if (error != EBUSY)
3319                 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3320                     error);
3321         dsl_dataset_disown(ds, FTAG);
3322 
3323 out:
3324         ztest_dsl_dataset_cleanup(osname, id);
3325 
3326         (void) rw_unlock(&ztest_name_lock);
3327 }
3328 
3329 /*
3330  * Verify that dmu_object_{alloc,free} work as expected.
3331  */
3332 void
3333 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3334 {
3335         ztest_od_t od[4];
3336         int batchsize = sizeof (od) / sizeof (od[0]);
3337 
3338         for (int b = 0; b < batchsize; b++)
3339                 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3340 
3341         /*
3342          * Destroy the previous batch of objects, create a new batch,
3343          * and do some I/O on the new objects.
3344          */
3345         if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3346                 return;
3347 
3348         while (ztest_random(4 * batchsize) != 0)
3349                 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3350                     ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3351 }
3352 
3353 /*
3354  * Verify that dmu_{read,write} work as expected.
3355  */
3356 void
3357 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3358 {
3359         objset_t *os = zd->zd_os;
3360         ztest_od_t od[2];
3361         dmu_tx_t *tx;
3362         int i, freeit, error;
3363         uint64_t n, s, txg;
3364         bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3365         uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3366         uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3367         uint64_t regions = 997;
3368         uint64_t stride = 123456789ULL;
3369         uint64_t width = 40;
3370         int free_percent = 5;
3371 
3372         /*
3373          * This test uses two objects, packobj and bigobj, that are always
3374          * updated together (i.e. in the same tx) so that their contents are
3375          * in sync and can be compared.  Their contents relate to each other
3376          * in a simple way: packobj is a dense array of 'bufwad' structures,
3377          * while bigobj is a sparse array of the same bufwads.  Specifically,
3378          * for any index n, there are three bufwads that should be identical:
3379          *
3380          *      packobj, at offset n * sizeof (bufwad_t)
3381          *      bigobj, at the head of the nth chunk
3382          *      bigobj, at the tail of the nth chunk
3383          *
3384          * The chunk size is arbitrary. It doesn't have to be a power of two,
3385          * and it doesn't have any relation to the object blocksize.
3386          * The only requirement is that it can hold at least two bufwads.
3387          *
3388          * Normally, we write the bufwad to each of these locations.
3389          * However, free_percent of the time we instead write zeroes to
3390          * packobj and perform a dmu_free_range() on bigobj.  By comparing
3391          * bigobj to packobj, we can verify that the DMU is correctly
3392          * tracking which parts of an object are allocated and free,
3393          * and that the contents of the allocated blocks are correct.
3394          */
3395 
3396         /*
3397          * Read the directory info.  If it's the first time, set things up.
3398          */
3399         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3400         ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3401 
3402         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3403                 return;
3404 
3405         bigobj = od[0].od_object;
3406         packobj = od[1].od_object;
3407         chunksize = od[0].od_gen;
3408         ASSERT(chunksize == od[1].od_gen);
3409 
3410         /*
3411          * Prefetch a random chunk of the big object.
3412          * Our aim here is to get some async reads in flight
3413          * for blocks that we may free below; the DMU should
3414          * handle this race correctly.
3415          */
3416         n = ztest_random(regions) * stride + ztest_random(width);
3417         s = 1 + ztest_random(2 * width - 1);
3418         dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3419 
3420         /*
3421          * Pick a random index and compute the offsets into packobj and bigobj.
3422          */
3423         n = ztest_random(regions) * stride + ztest_random(width);
3424         s = 1 + ztest_random(width - 1);
3425 
3426         packoff = n * sizeof (bufwad_t);
3427         packsize = s * sizeof (bufwad_t);
3428 
3429         bigoff = n * chunksize;
3430         bigsize = s * chunksize;
3431 
3432         packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3433         bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3434 
3435         /*
3436          * free_percent of the time, free a range of bigobj rather than
3437          * overwriting it.
3438          */
3439         freeit = (ztest_random(100) < free_percent);
3440 
3441         /*
3442          * Read the current contents of our objects.
3443          */
3444         error = dmu_read(os, packobj, packoff, packsize, packbuf,
3445             DMU_READ_PREFETCH);
3446         ASSERT3U(error, ==, 0);
3447         error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3448             DMU_READ_PREFETCH);
3449         ASSERT3U(error, ==, 0);
3450 
3451         /*
3452          * Get a tx for the mods to both packobj and bigobj.
3453          */
3454         tx = dmu_tx_create(os);
3455 
3456         dmu_tx_hold_write(tx, packobj, packoff, packsize);
3457 
3458         if (freeit)
3459                 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3460         else
3461                 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3462 
3463         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3464         if (txg == 0) {
3465                 umem_free(packbuf, packsize);
3466                 umem_free(bigbuf, bigsize);
3467                 return;
3468         }
3469 
3470         dmu_object_set_checksum(os, bigobj,
3471             (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3472 
3473         dmu_object_set_compress(os, bigobj,
3474             (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3475 
3476         /*
3477          * For each index from n to n + s, verify that the existing bufwad
3478          * in packobj matches the bufwads at the head and tail of the
3479          * corresponding chunk in bigobj.  Then update all three bufwads
3480          * with the new values we want to write out.
3481          */
3482         for (i = 0; i < s; i++) {
3483                 /* LINTED */
3484                 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3485                 /* LINTED */
3486                 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3487                 /* LINTED */
3488                 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3489 
3490                 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3491                 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3492 
3493                 if (pack->bw_txg > txg)
3494                         fatal(0, "future leak: got %llx, open txg is %llx",
3495                             pack->bw_txg, txg);
3496 
3497                 if (pack->bw_data != 0 && pack->bw_index != n + i)
3498                         fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3499                             pack->bw_index, n, i);
3500 
3501                 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3502                         fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3503 
3504                 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3505                         fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3506 
3507                 if (freeit) {
3508                         bzero(pack, sizeof (bufwad_t));
3509                 } else {
3510                         pack->bw_index = n + i;
3511                         pack->bw_txg = txg;
3512                         pack->bw_data = 1 + ztest_random(-2ULL);
3513                 }
3514                 *bigH = *pack;
3515                 *bigT = *pack;
3516         }
3517 
3518         /*
3519          * We've verified all the old bufwads, and made new ones.
3520          * Now write them out.
3521          */
3522         dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3523 
3524         if (freeit) {
3525                 if (ztest_opts.zo_verbose >= 7) {
3526                         (void) printf("freeing offset %llx size %llx"
3527                             " txg %llx\n",
3528                             (u_longlong_t)bigoff,
3529                             (u_longlong_t)bigsize,
3530                             (u_longlong_t)txg);
3531                 }
3532                 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3533         } else {
3534                 if (ztest_opts.zo_verbose >= 7) {
3535                         (void) printf("writing offset %llx size %llx"
3536                             " txg %llx\n",
3537                             (u_longlong_t)bigoff,
3538                             (u_longlong_t)bigsize,
3539                             (u_longlong_t)txg);
3540                 }
3541                 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3542         }
3543 
3544         dmu_tx_commit(tx);
3545 
3546         /*
3547          * Sanity check the stuff we just wrote.
3548          */
3549         {
3550                 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3551                 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3552 
3553                 VERIFY(0 == dmu_read(os, packobj, packoff,
3554                     packsize, packcheck, DMU_READ_PREFETCH));
3555                 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3556                     bigsize, bigcheck, DMU_READ_PREFETCH));
3557 
3558                 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3559                 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3560 
3561                 umem_free(packcheck, packsize);
3562                 umem_free(bigcheck, bigsize);
3563         }
3564 
3565         umem_free(packbuf, packsize);
3566         umem_free(bigbuf, bigsize);
3567 }
3568 
3569 void
3570 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3571     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3572 {
3573         uint64_t i;
3574         bufwad_t *pack;
3575         bufwad_t *bigH;
3576         bufwad_t *bigT;
3577 
3578         /*
3579          * For each index from n to n + s, verify that the existing bufwad
3580          * in packobj matches the bufwads at the head and tail of the
3581          * corresponding chunk in bigobj.  Then update all three bufwads
3582          * with the new values we want to write out.
3583          */
3584         for (i = 0; i < s; i++) {
3585                 /* LINTED */
3586                 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3587                 /* LINTED */
3588                 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3589                 /* LINTED */
3590                 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3591 
3592                 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3593                 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3594 
3595                 if (pack->bw_txg > txg)
3596                         fatal(0, "future leak: got %llx, open txg is %llx",
3597                             pack->bw_txg, txg);
3598 
3599                 if (pack->bw_data != 0 && pack->bw_index != n + i)
3600                         fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3601                             pack->bw_index, n, i);
3602 
3603                 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3604                         fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3605 
3606                 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3607                         fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3608 
3609                 pack->bw_index = n + i;
3610                 pack->bw_txg = txg;
3611                 pack->bw_data = 1 + ztest_random(-2ULL);
3612 
3613                 *bigH = *pack;
3614                 *bigT = *pack;
3615         }
3616 }
3617 
3618 void
3619 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3620 {
3621         objset_t *os = zd->zd_os;
3622         ztest_od_t od[2];
3623         dmu_tx_t *tx;
3624         uint64_t i;
3625         int error;
3626         uint64_t n, s, txg;
3627         bufwad_t *packbuf, *bigbuf;
3628         uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3629         uint64_t blocksize = ztest_random_blocksize();
3630         uint64_t chunksize = blocksize;
3631         uint64_t regions = 997;
3632         uint64_t stride = 123456789ULL;
3633         uint64_t width = 9;
3634         dmu_buf_t *bonus_db;
3635         arc_buf_t **bigbuf_arcbufs;
3636         dmu_object_info_t doi;
3637 
3638         /*
3639          * This test uses two objects, packobj and bigobj, that are always
3640          * updated together (i.e. in the same tx) so that their contents are
3641          * in sync and can be compared.  Their contents relate to each other
3642          * in a simple way: packobj is a dense array of 'bufwad' structures,
3643          * while bigobj is a sparse array of the same bufwads.  Specifically,
3644          * for any index n, there are three bufwads that should be identical:
3645          *
3646          *      packobj, at offset n * sizeof (bufwad_t)
3647          *      bigobj, at the head of the nth chunk
3648          *      bigobj, at the tail of the nth chunk
3649          *
3650          * The chunk size is set equal to bigobj block size so that
3651          * dmu_assign_arcbuf() can be tested for object updates.
3652          */
3653 
3654         /*
3655          * Read the directory info.  If it's the first time, set things up.
3656          */
3657         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3658         ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3659 
3660         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3661                 return;
3662 
3663         bigobj = od[0].od_object;
3664         packobj = od[1].od_object;
3665         blocksize = od[0].od_blocksize;
3666         chunksize = blocksize;
3667         ASSERT(chunksize == od[1].od_gen);
3668 
3669         VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3670         VERIFY(ISP2(doi.doi_data_block_size));
3671         VERIFY(chunksize == doi.doi_data_block_size);
3672         VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3673 
3674         /*
3675          * Pick a random index and compute the offsets into packobj and bigobj.
3676          */
3677         n = ztest_random(regions) * stride + ztest_random(width);
3678         s = 1 + ztest_random(width - 1);
3679 
3680         packoff = n * sizeof (bufwad_t);
3681         packsize = s * sizeof (bufwad_t);
3682 
3683         bigoff = n * chunksize;
3684         bigsize = s * chunksize;
3685 
3686         packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3687         bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3688 
3689         VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3690 
3691         bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3692 
3693         /*
3694          * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3695          * Iteration 1 test zcopy to already referenced dbufs.
3696          * Iteration 2 test zcopy to dirty dbuf in the same txg.
3697          * Iteration 3 test zcopy to dbuf dirty in previous txg.
3698          * Iteration 4 test zcopy when dbuf is no longer dirty.
3699          * Iteration 5 test zcopy when it can't be done.
3700          * Iteration 6 one more zcopy write.
3701          */
3702         for (i = 0; i < 7; i++) {
3703                 uint64_t j;
3704                 uint64_t off;
3705 
3706                 /*
3707                  * In iteration 5 (i == 5) use arcbufs
3708                  * that don't match bigobj blksz to test
3709                  * dmu_assign_arcbuf() when it can't directly
3710                  * assign an arcbuf to a dbuf.
3711                  */
3712                 for (j = 0; j < s; j++) {
3713                         if (i != 5) {
3714                                 bigbuf_arcbufs[j] =
3715                                     dmu_request_arcbuf(bonus_db, chunksize);
3716                         } else {
3717                                 bigbuf_arcbufs[2 * j] =
3718                                     dmu_request_arcbuf(bonus_db, chunksize / 2);
3719                                 bigbuf_arcbufs[2 * j + 1] =
3720                                     dmu_request_arcbuf(bonus_db, chunksize / 2);
3721                         }
3722                 }
3723 
3724                 /*
3725                  * Get a tx for the mods to both packobj and bigobj.
3726                  */
3727                 tx = dmu_tx_create(os);
3728 
3729                 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3730                 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3731 
3732                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3733                 if (txg == 0) {
3734                         umem_free(packbuf, packsize);
3735                         umem_free(bigbuf, bigsize);
3736                         for (j = 0; j < s; j++) {
3737                                 if (i != 5) {
3738                                         dmu_return_arcbuf(bigbuf_arcbufs[j]);
3739                                 } else {
3740                                         dmu_return_arcbuf(
3741                                             bigbuf_arcbufs[2 * j]);
3742                                         dmu_return_arcbuf(
3743                                             bigbuf_arcbufs[2 * j + 1]);
3744                                 }
3745                         }
3746                         umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3747                         dmu_buf_rele(bonus_db, FTAG);
3748                         return;
3749                 }
3750 
3751                 /*
3752                  * 50% of the time don't read objects in the 1st iteration to
3753                  * test dmu_assign_arcbuf() for the case when there're no
3754                  * existing dbufs for the specified offsets.
3755                  */
3756                 if (i != 0 || ztest_random(2) != 0) {
3757                         error = dmu_read(os, packobj, packoff,
3758                             packsize, packbuf, DMU_READ_PREFETCH);
3759                         ASSERT3U(error, ==, 0);
3760                         error = dmu_read(os, bigobj, bigoff, bigsize,
3761                             bigbuf, DMU_READ_PREFETCH);
3762                         ASSERT3U(error, ==, 0);
3763                 }
3764                 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3765                     n, chunksize, txg);
3766 
3767                 /*
3768                  * We've verified all the old bufwads, and made new ones.
3769                  * Now write them out.
3770                  */
3771                 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3772                 if (ztest_opts.zo_verbose >= 7) {
3773                         (void) printf("writing offset %llx size %llx"
3774                             " txg %llx\n",
3775                             (u_longlong_t)bigoff,
3776                             (u_longlong_t)bigsize,
3777                             (u_longlong_t)txg);
3778                 }
3779                 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3780                         dmu_buf_t *dbt;
3781                         if (i != 5) {
3782                                 bcopy((caddr_t)bigbuf + (off - bigoff),
3783                                     bigbuf_arcbufs[j]->b_data, chunksize);
3784                         } else {
3785                                 bcopy((caddr_t)bigbuf + (off - bigoff),
3786                                     bigbuf_arcbufs[2 * j]->b_data,
3787                                     chunksize / 2);
3788                                 bcopy((caddr_t)bigbuf + (off - bigoff) +
3789                                     chunksize / 2,
3790                                     bigbuf_arcbufs[2 * j + 1]->b_data,
3791                                     chunksize / 2);
3792                         }
3793 
3794                         if (i == 1) {
3795                                 VERIFY(dmu_buf_hold(os, bigobj, off,
3796                                     FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3797                         }
3798                         if (i != 5) {
3799                                 dmu_assign_arcbuf(bonus_db, off,
3800                                     bigbuf_arcbufs[j], tx);
3801                         } else {
3802                                 dmu_assign_arcbuf(bonus_db, off,
3803                                     bigbuf_arcbufs[2 * j], tx);
3804                                 dmu_assign_arcbuf(bonus_db,
3805                                     off + chunksize / 2,
3806                                     bigbuf_arcbufs[2 * j + 1], tx);
3807                         }
3808                         if (i == 1) {
3809                                 dmu_buf_rele(dbt, FTAG);
3810                         }
3811                 }
3812                 dmu_tx_commit(tx);
3813 
3814                 /*
3815                  * Sanity check the stuff we just wrote.
3816                  */
3817                 {
3818                         void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3819                         void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3820 
3821                         VERIFY(0 == dmu_read(os, packobj, packoff,
3822                             packsize, packcheck, DMU_READ_PREFETCH));
3823                         VERIFY(0 == dmu_read(os, bigobj, bigoff,
3824                             bigsize, bigcheck, DMU_READ_PREFETCH));
3825 
3826                         ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3827                         ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3828 
3829                         umem_free(packcheck, packsize);
3830                         umem_free(bigcheck, bigsize);
3831                 }
3832                 if (i == 2) {
3833                         txg_wait_open(dmu_objset_pool(os), 0);
3834                 } else if (i == 3) {
3835                         txg_wait_synced(dmu_objset_pool(os), 0);
3836                 }
3837         }
3838 
3839         dmu_buf_rele(bonus_db, FTAG);
3840         umem_free(packbuf, packsize);
3841         umem_free(bigbuf, bigsize);
3842         umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3843 }
3844 
3845 /* ARGSUSED */
3846 void
3847 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3848 {
3849         ztest_od_t od[1];
3850         uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3851             (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3852 
3853         /*
3854          * Have multiple threads write to large offsets in an object
3855          * to verify that parallel writes to an object -- even to the
3856          * same blocks within the object -- doesn't cause any trouble.
3857          */
3858         ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
3859 
3860         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3861                 return;
3862 
3863         while (ztest_random(10) != 0)
3864                 ztest_io(zd, od[0].od_object, offset);
3865 }
3866 
3867 void
3868 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
3869 {
3870         ztest_od_t od[1];
3871         uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
3872             (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3873         uint64_t count = ztest_random(20) + 1;
3874         uint64_t blocksize = ztest_random_blocksize();
3875         void *data;
3876 
3877         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3878 
3879         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3880                 return;
3881 
3882         if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
3883                 return;
3884 
3885         ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
3886 
3887         data = umem_zalloc(blocksize, UMEM_NOFAIL);
3888 
3889         while (ztest_random(count) != 0) {
3890                 uint64_t randoff = offset + (ztest_random(count) * blocksize);
3891                 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
3892                     data) != 0)
3893                         break;
3894                 while (ztest_random(4) != 0)
3895                         ztest_io(zd, od[0].od_object, randoff);
3896         }
3897 
3898         umem_free(data, blocksize);
3899 }
3900 
3901 /*
3902  * Verify that zap_{create,destroy,add,remove,update} work as expected.
3903  */
3904 #define ZTEST_ZAP_MIN_INTS      1
3905 #define ZTEST_ZAP_MAX_INTS      4
3906 #define ZTEST_ZAP_MAX_PROPS     1000
3907 
3908 void
3909 ztest_zap(ztest_ds_t *zd, uint64_t id)
3910 {
3911         objset_t *os = zd->zd_os;
3912         ztest_od_t od[1];
3913         uint64_t object;
3914         uint64_t txg, last_txg;
3915         uint64_t value[ZTEST_ZAP_MAX_INTS];
3916         uint64_t zl_ints, zl_intsize, prop;
3917         int i, ints;
3918         dmu_tx_t *tx;
3919         char propname[100], txgname[100];
3920         int error;
3921         char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
3922 
3923         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
3924 
3925         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3926                 return;
3927 
3928         object = od[0].od_object;
3929 
3930         /*
3931          * Generate a known hash collision, and verify that
3932          * we can lookup and remove both entries.
3933          */
3934         tx = dmu_tx_create(os);
3935         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3936         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3937         if (txg == 0)
3938                 return;
3939         for (i = 0; i < 2; i++) {
3940                 value[i] = i;
3941                 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
3942                     1, &value[i], tx));
3943         }
3944         for (i = 0; i < 2; i++) {
3945                 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
3946                     sizeof (uint64_t), 1, &value[i], tx));
3947                 VERIFY3U(0, ==,
3948                     zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
3949                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3950                 ASSERT3U(zl_ints, ==, 1);
3951         }
3952         for (i = 0; i < 2; i++) {
3953                 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
3954         }
3955         dmu_tx_commit(tx);
3956 
3957         /*
3958          * Generate a buch of random entries.
3959          */
3960         ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
3961 
3962         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
3963         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
3964         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
3965         bzero(value, sizeof (value));
3966         last_txg = 0;
3967 
3968         /*
3969          * If these zap entries already exist, validate their contents.
3970          */
3971         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
3972         if (error == 0) {
3973                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3974                 ASSERT3U(zl_ints, ==, 1);
3975 
3976                 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
3977                     zl_ints, &last_txg) == 0);
3978 
3979                 VERIFY(zap_length(os, object, propname, &zl_intsize,
3980                     &zl_ints) == 0);
3981 
3982                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3983                 ASSERT3U(zl_ints, ==, ints);
3984 
3985                 VERIFY(zap_lookup(os, object, propname, zl_intsize,
3986                     zl_ints, value) == 0);
3987 
3988                 for (i = 0; i < ints; i++) {
3989                         ASSERT3U(value[i], ==, last_txg + object + i);
3990                 }
3991         } else {
3992                 ASSERT3U(error, ==, ENOENT);
3993         }
3994 
3995         /*
3996          * Atomically update two entries in our zap object.
3997          * The first is named txg_%llu, and contains the txg
3998          * in which the property was last updated.  The second
3999          * is named prop_%llu, and the nth element of its value
4000          * should be txg + object + n.
4001          */
4002         tx = dmu_tx_create(os);
4003         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4004         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4005         if (txg == 0)
4006                 return;
4007 
4008         if (last_txg > txg)
4009                 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4010 
4011         for (i = 0; i < ints; i++)
4012                 value[i] = txg + object + i;
4013 
4014         VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4015             1, &txg, tx));
4016         VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4017             ints, value, tx));
4018 
4019         dmu_tx_commit(tx);
4020 
4021         /*
4022          * Remove a random pair of entries.
4023          */
4024         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4025         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4026         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4027 
4028         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4029 
4030         if (error == ENOENT)
4031                 return;
4032 
4033         ASSERT3U(error, ==, 0);
4034 
4035         tx = dmu_tx_create(os);
4036         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4037         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4038         if (txg == 0)
4039                 return;
4040         VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4041         VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4042         dmu_tx_commit(tx);
4043 }
4044 
4045 /*
4046  * Testcase to test the upgrading of a microzap to fatzap.
4047  */
4048 void
4049 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4050 {
4051         objset_t *os = zd->zd_os;
4052         ztest_od_t od[1];
4053         uint64_t object, txg;
4054 
4055         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4056 
4057         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4058                 return;
4059 
4060         object = od[0].od_object;
4061 
4062         /*
4063          * Add entries to this ZAP and make sure it spills over
4064          * and gets upgraded to a fatzap. Also, since we are adding
4065          * 2050 entries we should see ptrtbl growth and leaf-block split.
4066          */
4067         for (int i = 0; i < 2050; i++) {
4068                 char name[MAXNAMELEN];
4069                 uint64_t value = i;
4070                 dmu_tx_t *tx;
4071                 int error;
4072 
4073                 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4074                     id, value);
4075 
4076                 tx = dmu_tx_create(os);
4077                 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4078                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4079                 if (txg == 0)
4080                         return;
4081                 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4082                     &value, tx);
4083                 ASSERT(error == 0 || error == EEXIST);
4084                 dmu_tx_commit(tx);
4085         }
4086 }
4087 
4088 /* ARGSUSED */
4089 void
4090 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4091 {
4092         objset_t *os = zd->zd_os;
4093         ztest_od_t od[1];
4094         uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4095         dmu_tx_t *tx;
4096         int i, namelen, error;
4097         int micro = ztest_random(2);
4098         char name[20], string_value[20];
4099         void *data;
4100 
4101         ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4102 
4103         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4104                 return;
4105 
4106         object = od[0].od_object;
4107 
4108         /*
4109          * Generate a random name of the form 'xxx.....' where each
4110          * x is a random printable character and the dots are dots.
4111          * There are 94 such characters, and the name length goes from
4112          * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4113          */
4114         namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4115 
4116         for (i = 0; i < 3; i++)
4117                 name[i] = '!' + ztest_random('~' - '!' + 1);
4118         for (; i < namelen - 1; i++)
4119                 name[i] = '.';
4120         name[i] = '\0';
4121 
4122         if ((namelen & 1) || micro) {
4123                 wsize = sizeof (txg);
4124                 wc = 1;
4125                 data = &txg;
4126         } else {
4127                 wsize = 1;
4128                 wc = namelen;
4129                 data = string_value;
4130         }
4131 
4132         count = -1ULL;
4133         VERIFY(zap_count(os, object, &count) == 0);
4134         ASSERT(count != -1ULL);
4135 
4136         /*
4137          * Select an operation: length, lookup, add, update, remove.
4138          */
4139         i = ztest_random(5);
4140 
4141         if (i >= 2) {
4142                 tx = dmu_tx_create(os);
4143                 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4144                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4145                 if (txg == 0)
4146                         return;
4147                 bcopy(name, string_value, namelen);
4148         } else {
4149                 tx = NULL;
4150                 txg = 0;
4151                 bzero(string_value, namelen);
4152         }
4153 
4154         switch (i) {
4155 
4156         case 0:
4157                 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4158                 if (error == 0) {
4159                         ASSERT3U(wsize, ==, zl_wsize);
4160                         ASSERT3U(wc, ==, zl_wc);
4161                 } else {
4162                         ASSERT3U(error, ==, ENOENT);
4163                 }
4164                 break;
4165 
4166         case 1:
4167                 error = zap_lookup(os, object, name, wsize, wc, data);
4168                 if (error == 0) {
4169                         if (data == string_value &&
4170                             bcmp(name, data, namelen) != 0)
4171                                 fatal(0, "name '%s' != val '%s' len %d",
4172                                     name, data, namelen);
4173                 } else {
4174                         ASSERT3U(error, ==, ENOENT);
4175                 }
4176                 break;
4177 
4178         case 2:
4179                 error = zap_add(os, object, name, wsize, wc, data, tx);
4180                 ASSERT(error == 0 || error == EEXIST);
4181                 break;
4182 
4183         case 3:
4184                 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4185                 break;
4186 
4187         case 4:
4188                 error = zap_remove(os, object, name, tx);
4189                 ASSERT(error == 0 || error == ENOENT);
4190                 break;
4191         }
4192 
4193         if (tx != NULL)
4194                 dmu_tx_commit(tx);
4195 }
4196 
4197 /*
4198  * Commit callback data.
4199  */
4200 typedef struct ztest_cb_data {
4201         list_node_t             zcd_node;
4202         uint64_t                zcd_txg;
4203         int                     zcd_expected_err;
4204         boolean_t               zcd_added;
4205         boolean_t               zcd_called;
4206         spa_t                   *zcd_spa;
4207 } ztest_cb_data_t;
4208 
4209 /* This is the actual commit callback function */
4210 static void
4211 ztest_commit_callback(void *arg, int error)
4212 {
4213         ztest_cb_data_t *data = arg;
4214         uint64_t synced_txg;
4215 
4216         VERIFY(data != NULL);
4217         VERIFY3S(data->zcd_expected_err, ==, error);
4218         VERIFY(!data->zcd_called);
4219 
4220         synced_txg = spa_last_synced_txg(data->zcd_spa);
4221         if (data->zcd_txg > synced_txg)
4222                 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4223                     ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4224                     synced_txg);
4225 
4226         data->zcd_called = B_TRUE;
4227 
4228         if (error == ECANCELED) {
4229                 ASSERT3U(data->zcd_txg, ==, 0);
4230                 ASSERT(!data->zcd_added);
4231 
4232                 /*
4233                  * The private callback data should be destroyed here, but
4234                  * since we are going to check the zcd_called field after
4235                  * dmu_tx_abort(), we will destroy it there.
4236                  */
4237                 return;
4238         }
4239 
4240         /* Was this callback added to the global callback list? */
4241         if (!data->zcd_added)
4242                 goto out;
4243 
4244         ASSERT3U(data->zcd_txg, !=, 0);
4245 
4246         /* Remove our callback from the list */
4247         (void) mutex_lock(&zcl.zcl_callbacks_lock);
4248         list_remove(&zcl.zcl_callbacks, data);
4249         (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4250 
4251 out:
4252         umem_free(data, sizeof (ztest_cb_data_t));
4253 }
4254 
4255 /* Allocate and initialize callback data structure */
4256 static ztest_cb_data_t *
4257 ztest_create_cb_data(objset_t *os, uint64_t txg)
4258 {
4259         ztest_cb_data_t *cb_data;
4260 
4261         cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4262 
4263         cb_data->zcd_txg = txg;
4264         cb_data->zcd_spa = dmu_objset_spa(os);
4265 
4266         return (cb_data);
4267 }
4268 
4269 /*
4270  * If a number of txgs equal to this threshold have been created after a commit
4271  * callback has been registered but not called, then we assume there is an
4272  * implementation bug.
4273  */
4274 #define ZTEST_COMMIT_CALLBACK_THRESH    (TXG_CONCURRENT_STATES + 2)
4275 
4276 /*
4277  * Commit callback test.
4278  */
4279 void
4280 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4281 {
4282         objset_t *os = zd->zd_os;
4283         ztest_od_t od[1];
4284         dmu_tx_t *tx;
4285         ztest_cb_data_t *cb_data[3], *tmp_cb;
4286         uint64_t old_txg, txg;
4287         int i, error;
4288 
4289         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4290 
4291         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4292                 return;
4293 
4294         tx = dmu_tx_create(os);
4295 
4296         cb_data[0] = ztest_create_cb_data(os, 0);
4297         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4298 
4299         dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4300 
4301         /* Every once in a while, abort the transaction on purpose */
4302         if (ztest_random(100) == 0)
4303                 error = -1;
4304 
4305         if (!error)
4306                 error = dmu_tx_assign(tx, TXG_NOWAIT);
4307 
4308         txg = error ? 0 : dmu_tx_get_txg(tx);
4309 
4310         cb_data[0]->zcd_txg = txg;
4311         cb_data[1] = ztest_create_cb_data(os, txg);
4312         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4313 
4314         if (error) {
4315                 /*
4316                  * It's not a strict requirement to call the registered
4317                  * callbacks from inside dmu_tx_abort(), but that's what
4318                  * it's supposed to happen in the current implementation
4319                  * so we will check for that.
4320                  */
4321                 for (i = 0; i < 2; i++) {
4322                         cb_data[i]->zcd_expected_err = ECANCELED;
4323                         VERIFY(!cb_data[i]->zcd_called);
4324                 }
4325 
4326                 dmu_tx_abort(tx);
4327 
4328                 for (i = 0; i < 2; i++) {
4329                         VERIFY(cb_data[i]->zcd_called);
4330                         umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4331                 }
4332 
4333                 return;
4334         }
4335 
4336         cb_data[2] = ztest_create_cb_data(os, txg);
4337         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4338 
4339         /*
4340          * Read existing data to make sure there isn't a future leak.
4341          */
4342         VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4343             &old_txg, DMU_READ_PREFETCH));
4344 
4345         if (old_txg > txg)
4346                 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4347                     old_txg, txg);
4348 
4349         dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4350 
4351         (void) mutex_lock(&zcl.zcl_callbacks_lock);
4352 
4353         /*
4354          * Since commit callbacks don't have any ordering requirement and since
4355          * it is theoretically possible for a commit callback to be called
4356          * after an arbitrary amount of time has elapsed since its txg has been
4357          * synced, it is difficult to reliably determine whether a commit
4358          * callback hasn't been called due to high load or due to a flawed
4359          * implementation.
4360          *
4361          * In practice, we will assume that if after a certain number of txgs a
4362          * commit callback hasn't been called, then most likely there's an
4363          * implementation bug..
4364          */
4365         tmp_cb = list_head(&zcl.zcl_callbacks);
4366         if (tmp_cb != NULL &&
4367             tmp_cb->zcd_txg > txg - ZTEST_COMMIT_CALLBACK_THRESH) {
4368                 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4369                     PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4370         }
4371 
4372         /*
4373          * Let's find the place to insert our callbacks.
4374          *
4375          * Even though the list is ordered by txg, it is possible for the
4376          * insertion point to not be the end because our txg may already be
4377          * quiescing at this point and other callbacks in the open txg
4378          * (from other objsets) may have sneaked in.
4379          */
4380         tmp_cb = list_tail(&zcl.zcl_callbacks);
4381         while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4382                 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4383 
4384         /* Add the 3 callbacks to the list */
4385         for (i = 0; i < 3; i++) {
4386                 if (tmp_cb == NULL)
4387                         list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4388                 else
4389                         list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4390                             cb_data[i]);
4391 
4392                 cb_data[i]->zcd_added = B_TRUE;
4393                 VERIFY(!cb_data[i]->zcd_called);
4394 
4395                 tmp_cb = cb_data[i];
4396         }
4397 
4398         (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4399 
4400         dmu_tx_commit(tx);
4401 }
4402 
4403 /* ARGSUSED */
4404 void
4405 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4406 {
4407         zfs_prop_t proplist[] = {
4408                 ZFS_PROP_CHECKSUM,
4409                 ZFS_PROP_COMPRESSION,
4410                 ZFS_PROP_COPIES,
4411                 ZFS_PROP_DEDUP
4412         };
4413 
4414         (void) rw_rdlock(&ztest_name_lock);
4415 
4416         for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4417                 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4418                     ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4419 
4420         (void) rw_unlock(&ztest_name_lock);
4421 }
4422 
4423 /* ARGSUSED */
4424 void
4425 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4426 {
4427         nvlist_t *props = NULL;
4428 
4429         (void) rw_rdlock(&ztest_name_lock);
4430 
4431         (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4432             ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4433 
4434         VERIFY3U(spa_prop_get(ztest_spa, &props), ==, 0);
4435 
4436         if (ztest_opts.zo_verbose >= 6)
4437                 dump_nvlist(props, 4);
4438 
4439         nvlist_free(props);
4440 
4441         (void) rw_unlock(&ztest_name_lock);
4442 }
4443 
4444 /*
4445  * Test snapshot hold/release and deferred destroy.
4446  */
4447 void
4448 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4449 {
4450         int error;
4451         objset_t *os = zd->zd_os;
4452         objset_t *origin;
4453         char snapname[100];
4454         char fullname[100];
4455         char clonename[100];
4456         char tag[100];
4457         char osname[MAXNAMELEN];
4458 
4459         (void) rw_rdlock(&ztest_name_lock);
4460 
4461         dmu_objset_name(os, osname);
4462 
4463         (void) snprintf(snapname, 100, "sh1_%llu", id);
4464         (void) snprintf(fullname, 100, "%s@%s", osname, snapname);
4465         (void) snprintf(clonename, 100, "%s/ch1_%llu", osname, id);
4466         (void) snprintf(tag, 100, "%tag_%llu", id);
4467 
4468         /*
4469          * Clean up from any previous run.
4470          */
4471         (void) dmu_objset_destroy(clonename, B_FALSE);
4472         (void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4473         (void) dmu_objset_destroy(fullname, B_FALSE);
4474 
4475         /*
4476          * Create snapshot, clone it, mark snap for deferred destroy,
4477          * destroy clone, verify snap was also destroyed.
4478          */
4479         error = dmu_objset_snapshot_one(osname, snapname);
4480         if (error) {
4481                 if (error == ENOSPC) {
4482                         ztest_record_enospc("dmu_objset_snapshot");
4483                         goto out;
4484                 }
4485                 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4486         }
4487 
4488         error = dmu_objset_hold(fullname, FTAG, &origin);
4489         if (error)
4490                 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4491 
4492         error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0);
4493         dmu_objset_rele(origin, FTAG);
4494         if (error) {
4495                 if (error == ENOSPC) {
4496                         ztest_record_enospc("dmu_objset_clone");
4497                         goto out;
4498                 }
4499                 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4500         }
4501 
4502         error = dmu_objset_destroy(fullname, B_TRUE);
4503         if (error) {
4504                 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4505                     fullname, error);
4506         }
4507 
4508         error = dmu_objset_destroy(clonename, B_FALSE);
4509         if (error)
4510                 fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error);
4511 
4512         error = dmu_objset_hold(fullname, FTAG, &origin);
4513         if (error != ENOENT)
4514                 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4515 
4516         /*
4517          * Create snapshot, add temporary hold, verify that we can't
4518          * destroy a held snapshot, mark for deferred destroy,
4519          * release hold, verify snapshot was destroyed.
4520          */
4521         error = dmu_objset_snapshot_one(osname, snapname);
4522         if (error) {
4523                 if (error == ENOSPC) {
4524                         ztest_record_enospc("dmu_objset_snapshot");
4525                         goto out;
4526                 }
4527                 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4528         }
4529 
4530         error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE,
4531             B_TRUE, -1);
4532         if (error)
4533                 fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4534 
4535         error = dmu_objset_destroy(fullname, B_FALSE);
4536         if (error != EBUSY) {
4537                 fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4538                     fullname, error);
4539         }
4540 
4541         error = dmu_objset_destroy(fullname, B_TRUE);
4542         if (error) {
4543                 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4544                     fullname, error);
4545         }
4546 
4547         error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4548         if (error)
4549                 fatal(0, "dsl_dataset_user_release(%s)", fullname, tag);
4550 
4551         VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT);
4552 
4553 out:
4554         (void) rw_unlock(&ztest_name_lock);
4555 }
4556 
4557 /*
4558  * Inject random faults into the on-disk data.
4559  */
4560 /* ARGSUSED */
4561 void
4562 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4563 {
4564         ztest_shared_t *zs = ztest_shared;
4565         spa_t *spa = ztest_spa;
4566         int fd;
4567         uint64_t offset;
4568         uint64_t leaves;
4569         uint64_t bad = 0x1990c0ffeedecade;
4570         uint64_t top, leaf;
4571         char path0[MAXPATHLEN];
4572         char pathrand[MAXPATHLEN];
4573         size_t fsize;
4574         int bshift = SPA_MAXBLOCKSHIFT + 2;     /* don't scrog all labels */
4575         int iters = 1000;
4576         int maxfaults;
4577         int mirror_save;
4578         vdev_t *vd0 = NULL;
4579         uint64_t guid0 = 0;
4580         boolean_t islog = B_FALSE;
4581 
4582         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4583         maxfaults = MAXFAULTS();
4584         leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4585         mirror_save = zs->zs_mirrors;
4586         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4587 
4588         ASSERT(leaves >= 1);
4589 
4590         /*
4591          * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4592          */
4593         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4594 
4595         if (ztest_random(2) == 0) {
4596                 /*
4597                  * Inject errors on a normal data device or slog device.
4598                  */
4599                 top = ztest_random_vdev_top(spa, B_TRUE);
4600                 leaf = ztest_random(leaves) + zs->zs_splits;
4601 
4602                 /*
4603                  * Generate paths to the first leaf in this top-level vdev,
4604                  * and to the random leaf we selected.  We'll induce transient
4605                  * write failures and random online/offline activity on leaf 0,
4606                  * and we'll write random garbage to the randomly chosen leaf.
4607                  */
4608                 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4609                     ztest_opts.zo_dir, ztest_opts.zo_pool,
4610                     top * leaves + zs->zs_splits);
4611                 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4612                     ztest_opts.zo_dir, ztest_opts.zo_pool,
4613                     top * leaves + leaf);
4614 
4615                 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4616                 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4617                         islog = B_TRUE;
4618 
4619                 if (vd0 != NULL && maxfaults != 1) {
4620                         /*
4621                          * Make vd0 explicitly claim to be unreadable,
4622                          * or unwriteable, or reach behind its back
4623                          * and close the underlying fd.  We can do this if
4624                          * maxfaults == 0 because we'll fail and reexecute,
4625                          * and we can do it if maxfaults >= 2 because we'll
4626                          * have enough redundancy.  If maxfaults == 1, the
4627                          * combination of this with injection of random data
4628                          * corruption below exceeds the pool's fault tolerance.
4629                          */
4630                         vdev_file_t *vf = vd0->vdev_tsd;
4631 
4632                         if (vf != NULL && ztest_random(3) == 0) {
4633                                 (void) close(vf->vf_vnode->v_fd);
4634                                 vf->vf_vnode->v_fd = -1;
4635                         } else if (ztest_random(2) == 0) {
4636                                 vd0->vdev_cant_read = B_TRUE;
4637                         } else {
4638                                 vd0->vdev_cant_write = B_TRUE;
4639                         }
4640                         guid0 = vd0->vdev_guid;
4641                 }
4642         } else {
4643                 /*
4644                  * Inject errors on an l2cache device.
4645                  */
4646                 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4647 
4648                 if (sav->sav_count == 0) {
4649                         spa_config_exit(spa, SCL_STATE, FTAG);
4650                         return;
4651                 }
4652                 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4653                 guid0 = vd0->vdev_guid;
4654                 (void) strcpy(path0, vd0->vdev_path);
4655                 (void) strcpy(pathrand, vd0->vdev_path);
4656 
4657                 leaf = 0;
4658                 leaves = 1;
4659                 maxfaults = INT_MAX;    /* no limit on cache devices */
4660         }
4661 
4662         spa_config_exit(spa, SCL_STATE, FTAG);
4663 
4664         /*
4665          * If we can tolerate two or more faults, or we're dealing
4666          * with a slog, randomly online/offline vd0.
4667          */
4668         if ((maxfaults >= 2 || islog) && guid0 != 0) {
4669                 if (ztest_random(10) < 6) {
4670                         int flags = (ztest_random(2) == 0 ?
4671                             ZFS_OFFLINE_TEMPORARY : 0);
4672 
4673                         /*
4674                          * We have to grab the zs_name_lock as writer to
4675                          * prevent a race between offlining a slog and
4676                          * destroying a dataset. Offlining the slog will
4677                          * grab a reference on the dataset which may cause
4678                          * dmu_objset_destroy() to fail with EBUSY thus
4679                          * leaving the dataset in an inconsistent state.
4680                          */
4681                         if (islog)
4682                                 (void) rw_wrlock(&ztest_name_lock);
4683 
4684                         VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4685 
4686                         if (islog)
4687                                 (void) rw_unlock(&ztest_name_lock);
4688                 } else {
4689                         (void) vdev_online(spa, guid0, 0, NULL);
4690                 }
4691         }
4692 
4693         if (maxfaults == 0)
4694                 return;
4695 
4696         /*
4697          * We have at least single-fault tolerance, so inject data corruption.
4698          */
4699         fd = open(pathrand, O_RDWR);
4700 
4701         if (fd == -1)   /* we hit a gap in the device namespace */
4702                 return;
4703 
4704         fsize = lseek(fd, 0, SEEK_END);
4705 
4706         while (--iters != 0) {
4707                 offset = ztest_random(fsize / (leaves << bshift)) *
4708                     (leaves << bshift) + (leaf << bshift) +
4709                     (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4710 
4711                 if (offset >= fsize)
4712                         continue;
4713 
4714                 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4715                 if (mirror_save != zs->zs_mirrors) {
4716                         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4717                         (void) close(fd);
4718                         return;
4719                 }
4720 
4721                 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4722                         fatal(1, "can't inject bad word at 0x%llx in %s",
4723                             offset, pathrand);
4724 
4725                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4726 
4727                 if (ztest_opts.zo_verbose >= 7)
4728                         (void) printf("injected bad word into %s,"
4729                             " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4730         }
4731 
4732         (void) close(fd);
4733 }
4734 
4735 /*
4736  * Verify that DDT repair works as expected.
4737  */
4738 void
4739 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4740 {
4741         ztest_shared_t *zs = ztest_shared;
4742         spa_t *spa = ztest_spa;
4743         objset_t *os = zd->zd_os;
4744         ztest_od_t od[1];
4745         uint64_t object, blocksize, txg, pattern, psize;
4746         enum zio_checksum checksum = spa_dedup_checksum(spa);
4747         dmu_buf_t *db;
4748         dmu_tx_t *tx;
4749         void *buf;
4750         blkptr_t blk;
4751         int copies = 2 * ZIO_DEDUPDITTO_MIN;
4752 
4753         blocksize = ztest_random_blocksize();
4754         blocksize = MIN(blocksize, 2048);       /* because we write so many */
4755 
4756         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4757 
4758         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4759                 return;
4760 
4761         /*
4762          * Take the name lock as writer to prevent anyone else from changing
4763          * the pool and dataset properies we need to maintain during this test.
4764          */
4765         (void) rw_wrlock(&ztest_name_lock);
4766 
4767         if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4768             B_FALSE) != 0 ||
4769             ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4770             B_FALSE) != 0) {
4771                 (void) rw_unlock(&ztest_name_lock);
4772                 return;
4773         }
4774 
4775         object = od[0].od_object;
4776         blocksize = od[0].od_blocksize;
4777         pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4778 
4779         ASSERT(object != 0);
4780 
4781         tx = dmu_tx_create(os);
4782         dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4783         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4784         if (txg == 0) {
4785                 (void) rw_unlock(&ztest_name_lock);
4786                 return;
4787         }
4788 
4789         /*
4790          * Write all the copies of our block.
4791          */
4792         for (int i = 0; i < copies; i++) {
4793                 uint64_t offset = i * blocksize;
4794                 VERIFY(dmu_buf_hold(os, object, offset, FTAG, &db,
4795                     DMU_READ_NO_PREFETCH) == 0);
4796                 ASSERT(db->db_offset == offset);
4797                 ASSERT(db->db_size == blocksize);
4798                 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4799                     ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4800                 dmu_buf_will_fill(db, tx);
4801                 ztest_pattern_set(db->db_data, db->db_size, pattern);
4802                 dmu_buf_rele(db, FTAG);
4803         }
4804 
4805         dmu_tx_commit(tx);
4806         txg_wait_synced(spa_get_dsl(spa), txg);
4807 
4808         /*
4809          * Find out what block we got.
4810          */
4811         VERIFY(dmu_buf_hold(os, object, 0, FTAG, &db,
4812             DMU_READ_NO_PREFETCH) == 0);
4813         blk = *((dmu_buf_impl_t *)db)->db_blkptr;
4814         dmu_buf_rele(db, FTAG);
4815 
4816         /*
4817          * Damage the block.  Dedup-ditto will save us when we read it later.
4818          */
4819         psize = BP_GET_PSIZE(&blk);
4820         buf = zio_buf_alloc(psize);
4821         ztest_pattern_set(buf, psize, ~pattern);
4822 
4823         (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
4824             buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
4825             ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
4826 
4827         zio_buf_free(buf, psize);
4828 
4829         (void) rw_unlock(&ztest_name_lock);
4830 }
4831 
4832 /*
4833  * Scrub the pool.
4834  */
4835 /* ARGSUSED */
4836 void
4837 ztest_scrub(ztest_ds_t *zd, uint64_t id)
4838 {
4839         spa_t *spa = ztest_spa;
4840 
4841         (void) spa_scan(spa, POOL_SCAN_SCRUB);
4842         (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
4843         (void) spa_scan(spa, POOL_SCAN_SCRUB);
4844 }
4845 
4846 /*
4847  * Change the guid for the pool.
4848  */
4849 /* ARGSUSED */
4850 void
4851 ztest_reguid(ztest_ds_t *zd, uint64_t id)
4852 {
4853         spa_t *spa = ztest_spa;
4854         uint64_t orig, load;
4855 
4856         orig = spa_guid(spa);
4857         load = spa_load_guid(spa);
4858         if (spa_change_guid(spa) != 0)
4859                 return;
4860 
4861         if (ztest_opts.zo_verbose >= 3) {
4862                 (void) printf("Changed guid old %llu -> %llu\n",
4863                     (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
4864         }
4865 
4866         VERIFY3U(orig, !=, spa_guid(spa));
4867         VERIFY3U(load, ==, spa_load_guid(spa));
4868 }
4869 
4870 /*
4871  * Rename the pool to a different name and then rename it back.
4872  */
4873 /* ARGSUSED */
4874 void
4875 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
4876 {
4877         char *oldname, *newname;
4878         spa_t *spa;
4879 
4880         (void) rw_wrlock(&ztest_name_lock);
4881 
4882         oldname = ztest_opts.zo_pool;
4883         newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
4884         (void) strcpy(newname, oldname);
4885         (void) strcat(newname, "_tmp");
4886 
4887         /*
4888          * Do the rename
4889          */
4890         VERIFY3U(0, ==, spa_rename(oldname, newname));
4891 
4892         /*
4893          * Try to open it under the old name, which shouldn't exist
4894          */
4895         VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
4896 
4897         /*
4898          * Open it under the new name and make sure it's still the same spa_t.
4899          */
4900         VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
4901 
4902         ASSERT(spa == ztest_spa);
4903         spa_close(spa, FTAG);
4904 
4905         /*
4906          * Rename it back to the original
4907          */
4908         VERIFY3U(0, ==, spa_rename(newname, oldname));
4909 
4910         /*
4911          * Make sure it can still be opened
4912          */
4913         VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
4914 
4915         ASSERT(spa == ztest_spa);
4916         spa_close(spa, FTAG);
4917 
4918         umem_free(newname, strlen(newname) + 1);
4919 
4920         (void) rw_unlock(&ztest_name_lock);
4921 }
4922 
4923 /*
4924  * Verify pool integrity by running zdb.
4925  */
4926 static void
4927 ztest_run_zdb(char *pool)
4928 {
4929         int status;
4930         char zdb[MAXPATHLEN + MAXNAMELEN + 20];
4931         char zbuf[1024];
4932         char *bin;
4933         char *ztest;
4934         char *isa;
4935         int isalen;
4936         FILE *fp;
4937 
4938         (void) realpath(getexecname(), zdb);
4939 
4940         /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
4941         bin = strstr(zdb, "/usr/bin/");
4942         ztest = strstr(bin, "/ztest");
4943         isa = bin + 8;
4944         isalen = ztest - isa;
4945         isa = strdup(isa);
4946         /* LINTED */
4947         (void) sprintf(bin,
4948             "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
4949             isalen,
4950             isa,
4951             ztest_opts.zo_verbose >= 3 ? "s" : "",
4952             ztest_opts.zo_verbose >= 4 ? "v" : "",
4953             spa_config_path,
4954             pool);
4955         free(isa);
4956 
4957         if (ztest_opts.zo_verbose >= 5)
4958                 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
4959 
4960         fp = popen(zdb, "r");
4961 
4962         while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
4963                 if (ztest_opts.zo_verbose >= 3)
4964                         (void) printf("%s", zbuf);
4965 
4966         status = pclose(fp);
4967 
4968         if (status == 0)
4969                 return;
4970 
4971         ztest_dump_core = 0;
4972         if (WIFEXITED(status))
4973                 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
4974         else
4975                 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
4976 }
4977 
4978 static void
4979 ztest_walk_pool_directory(char *header)
4980 {
4981         spa_t *spa = NULL;
4982 
4983         if (ztest_opts.zo_verbose >= 6)
4984                 (void) printf("%s\n", header);
4985 
4986         mutex_enter(&spa_namespace_lock);
4987         while ((spa = spa_next(spa)) != NULL)
4988                 if (ztest_opts.zo_verbose >= 6)
4989                         (void) printf("\t%s\n", spa_name(spa));
4990         mutex_exit(&spa_namespace_lock);
4991 }
4992 
4993 static void
4994 ztest_spa_import_export(char *oldname, char *newname)
4995 {
4996         nvlist_t *config, *newconfig;
4997         uint64_t pool_guid;
4998         spa_t *spa;
4999 
5000         if (ztest_opts.zo_verbose >= 4) {
5001                 (void) printf("import/export: old = %s, new = %s\n",
5002                     oldname, newname);
5003         }
5004 
5005         /*
5006          * Clean up from previous runs.
5007          */
5008         (void) spa_destroy(newname);
5009 
5010         /*
5011          * Get the pool's configuration and guid.
5012          */
5013         VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5014 
5015         /*
5016          * Kick off a scrub to tickle scrub/export races.
5017          */
5018         if (ztest_random(2) == 0)
5019                 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5020 
5021         pool_guid = spa_guid(spa);
5022         spa_close(spa, FTAG);
5023 
5024         ztest_walk_pool_directory("pools before export");
5025 
5026         /*
5027          * Export it.
5028          */
5029         VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5030 
5031         ztest_walk_pool_directory("pools after export");
5032 
5033         /*
5034          * Try to import it.
5035          */
5036         newconfig = spa_tryimport(config);
5037         ASSERT(newconfig != NULL);
5038         nvlist_free(newconfig);
5039 
5040         /*
5041          * Import it under the new name.
5042          */
5043         VERIFY3U(0, ==, spa_import(newname, config, NULL, 0));
5044 
5045         ztest_walk_pool_directory("pools after import");
5046 
5047         /*
5048          * Try to import it again -- should fail with EEXIST.
5049          */
5050         VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5051 
5052         /*
5053          * Try to import it under a different name -- should fail with EEXIST.
5054          */
5055         VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5056 
5057         /*
5058          * Verify that the pool is no longer visible under the old name.
5059          */
5060         VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5061 
5062         /*
5063          * Verify that we can open and close the pool using the new name.
5064          */
5065         VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5066         ASSERT(pool_guid == spa_guid(spa));
5067         spa_close(spa, FTAG);
5068 
5069         nvlist_free(config);
5070 }
5071 
5072 static void
5073 ztest_resume(spa_t *spa)
5074 {
5075         if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5076                 (void) printf("resuming from suspended state\n");
5077         spa_vdev_state_enter(spa, SCL_NONE);
5078         vdev_clear(spa, NULL);
5079         (void) spa_vdev_state_exit(spa, NULL, 0);
5080         (void) zio_resume(spa);
5081 }
5082 
5083 static void *
5084 ztest_resume_thread(void *arg)
5085 {
5086         spa_t *spa = arg;
5087 
5088         while (!ztest_exiting) {
5089                 if (spa_suspended(spa))
5090                         ztest_resume(spa);
5091                 (void) poll(NULL, 0, 100);
5092         }
5093         return (NULL);
5094 }
5095 
5096 static void *
5097 ztest_deadman_thread(void *arg)
5098 {
5099         ztest_shared_t *zs = arg;
5100         int grace = 300;
5101         hrtime_t delta;
5102 
5103         delta = (zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + grace;
5104 
5105         (void) poll(NULL, 0, (int)(1000 * delta));
5106 
5107         fatal(0, "failed to complete within %d seconds of deadline", grace);
5108 
5109         return (NULL);
5110 }
5111 
5112 static void
5113 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5114 {
5115         ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5116         ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5117         hrtime_t functime = gethrtime();
5118 
5119         for (int i = 0; i < zi->zi_iters; i++)
5120                 zi->zi_func(zd, id);
5121 
5122         functime = gethrtime() - functime;
5123 
5124         atomic_add_64(&zc->zc_count, 1);
5125         atomic_add_64(&zc->zc_time, functime);
5126 
5127         if (ztest_opts.zo_verbose >= 4) {
5128                 Dl_info dli;
5129                 (void) dladdr((void *)zi->zi_func, &dli);
5130                 (void) printf("%6.2f sec in %s\n",
5131                     (double)functime / NANOSEC, dli.dli_sname);
5132         }
5133 }
5134 
5135 static void *
5136 ztest_thread(void *arg)
5137 {
5138         int rand;
5139         uint64_t id = (uintptr_t)arg;
5140         ztest_shared_t *zs = ztest_shared;
5141         uint64_t call_next;
5142         hrtime_t now;
5143         ztest_info_t *zi;
5144         ztest_shared_callstate_t *zc;
5145 
5146         while ((now = gethrtime()) < zs->zs_thread_stop) {
5147                 /*
5148                  * See if it's time to force a crash.
5149                  */
5150                 if (now > zs->zs_thread_kill)
5151                         ztest_kill(zs);
5152 
5153                 /*
5154                  * If we're getting ENOSPC with some regularity, stop.
5155                  */
5156                 if (zs->zs_enospc_count > 10)
5157                         break;
5158 
5159                 /*
5160                  * Pick a random function to execute.
5161                  */
5162                 rand = ztest_random(ZTEST_FUNCS);
5163                 zi = &ztest_info[rand];
5164                 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5165                 call_next = zc->zc_next;
5166 
5167                 if (now >= call_next &&
5168                     atomic_cas_64(&zc->zc_next, call_next, call_next +
5169                     ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5170                         ztest_execute(rand, zi, id);
5171                 }
5172         }
5173 
5174         return (NULL);
5175 }
5176 
5177 static void
5178 ztest_dataset_name(char *dsname, char *pool, int d)
5179 {
5180         (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5181 }
5182 
5183 static void
5184 ztest_dataset_destroy(int d)
5185 {
5186         char name[MAXNAMELEN];
5187 
5188         ztest_dataset_name(name, ztest_opts.zo_pool, d);
5189 
5190         if (ztest_opts.zo_verbose >= 3)
5191                 (void) printf("Destroying %s to free up space\n", name);
5192 
5193         /*
5194          * Cleanup any non-standard clones and snapshots.  In general,
5195          * ztest thread t operates on dataset (t % zopt_datasets),
5196          * so there may be more than one thing to clean up.
5197          */
5198         for (int t = d; t < ztest_opts.zo_threads;
5199             t += ztest_opts.zo_datasets) {
5200                 ztest_dsl_dataset_cleanup(name, t);
5201         }
5202 
5203         (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5204             DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5205 }
5206 
5207 static void
5208 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5209 {
5210         uint64_t usedobjs, dirobjs, scratch;
5211 
5212         /*
5213          * ZTEST_DIROBJ is the object directory for the entire dataset.
5214          * Therefore, the number of objects in use should equal the
5215          * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5216          * If not, we have an object leak.
5217          *
5218          * Note that we can only check this in ztest_dataset_open(),
5219          * when the open-context and syncing-context values agree.
5220          * That's because zap_count() returns the open-context value,
5221          * while dmu_objset_space() returns the rootbp fill count.
5222          */
5223         VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5224         dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5225         ASSERT3U(dirobjs + 1, ==, usedobjs);
5226 }
5227 
5228 static int
5229 ztest_dataset_open(int d)
5230 {
5231         ztest_ds_t *zd = &ztest_ds[d];
5232         uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5233         objset_t *os;
5234         zilog_t *zilog;
5235         char name[MAXNAMELEN];
5236         int error;
5237 
5238         ztest_dataset_name(name, ztest_opts.zo_pool, d);
5239 
5240         (void) rw_rdlock(&ztest_name_lock);
5241 
5242         error = ztest_dataset_create(name);
5243         if (error == ENOSPC) {
5244                 (void) rw_unlock(&ztest_name_lock);
5245                 ztest_record_enospc(FTAG);
5246                 return (error);
5247         }
5248         ASSERT(error == 0 || error == EEXIST);
5249 
5250         VERIFY3U(dmu_objset_hold(name, zd, &os), ==, 0);
5251         (void) rw_unlock(&ztest_name_lock);
5252 
5253         ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5254 
5255         zilog = zd->zd_zilog;
5256 
5257         if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5258             zilog->zl_header->zh_claim_lr_seq < committed_seq)
5259                 fatal(0, "missing log records: claimed %llu < committed %llu",
5260                     zilog->zl_header->zh_claim_lr_seq, committed_seq);
5261 
5262         ztest_dataset_dirobj_verify(zd);
5263 
5264         zil_replay(os, zd, ztest_replay_vector);
5265 
5266         ztest_dataset_dirobj_verify(zd);
5267 
5268         if (ztest_opts.zo_verbose >= 6)
5269                 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5270                     zd->zd_name,
5271                     (u_longlong_t)zilog->zl_parse_blk_count,
5272                     (u_longlong_t)zilog->zl_parse_lr_count,
5273                     (u_longlong_t)zilog->zl_replaying_seq);
5274 
5275         zilog = zil_open(os, ztest_get_data);
5276 
5277         if (zilog->zl_replaying_seq != 0 &&
5278             zilog->zl_replaying_seq < committed_seq)
5279                 fatal(0, "missing log records: replayed %llu < committed %llu",
5280                     zilog->zl_replaying_seq, committed_seq);
5281 
5282         return (0);
5283 }
5284 
5285 static void
5286 ztest_dataset_close(int d)
5287 {
5288         ztest_ds_t *zd = &ztest_ds[d];
5289 
5290         zil_close(zd->zd_zilog);
5291         dmu_objset_rele(zd->zd_os, zd);
5292 
5293         ztest_zd_fini(zd);
5294 }
5295 
5296 /*
5297  * Kick off threads to run tests on all datasets in parallel.
5298  */
5299 static void
5300 ztest_run(ztest_shared_t *zs)
5301 {
5302         thread_t *tid;
5303         spa_t *spa;
5304         objset_t *os;
5305         thread_t resume_tid;
5306         int error;
5307 
5308         ztest_exiting = B_FALSE;
5309 
5310         /*
5311          * Initialize parent/child shared state.
5312          */
5313         VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5314         VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5315 
5316         zs->zs_thread_start = gethrtime();
5317         zs->zs_thread_stop =
5318             zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5319         zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5320         zs->zs_thread_kill = zs->zs_thread_stop;
5321         if (ztest_random(100) < ztest_opts.zo_killrate) {
5322                 zs->zs_thread_kill -=
5323                     ztest_random(ztest_opts.zo_passtime * NANOSEC);
5324         }
5325 
5326         (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5327 
5328         list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5329             offsetof(ztest_cb_data_t, zcd_node));
5330 
5331         /*
5332          * Open our pool.
5333          */
5334         kernel_init(FREAD | FWRITE);
5335         VERIFY(spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0);
5336         spa->spa_debug = B_TRUE;
5337         ztest_spa = spa;
5338 
5339         VERIFY3U(0, ==, dmu_objset_hold(ztest_opts.zo_pool, FTAG, &os));
5340         zs->zs_guid = dmu_objset_fsid_guid(os);
5341         dmu_objset_rele(os, FTAG);
5342 
5343         spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5344 
5345         /*
5346          * We don't expect the pool to suspend unless maxfaults == 0,
5347          * in which case ztest_fault_inject() temporarily takes away
5348          * the only valid replica.
5349          */
5350         if (MAXFAULTS() == 0)
5351                 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5352         else
5353                 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5354 
5355         /*
5356          * Create a thread to periodically resume suspended I/O.
5357          */
5358         VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5359             &resume_tid) == 0);
5360 
5361         /*
5362          * Create a deadman thread to abort() if we hang.
5363          */
5364         VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5365             NULL) == 0);
5366 
5367         /*
5368          * Verify that we can safely inquire about about any object,
5369          * whether it's allocated or not.  To make it interesting,
5370          * we probe a 5-wide window around each power of two.
5371          * This hits all edge cases, including zero and the max.
5372          */
5373         for (int t = 0; t < 64; t++) {
5374                 for (int d = -5; d <= 5; d++) {
5375                         error = dmu_object_info(spa->spa_meta_objset,
5376                             (1ULL << t) + d, NULL);
5377                         ASSERT(error == 0 || error == ENOENT ||
5378                             error == EINVAL);
5379                 }
5380         }
5381 
5382         /*
5383          * If we got any ENOSPC errors on the previous run, destroy something.
5384          */
5385         if (zs->zs_enospc_count != 0) {
5386                 int d = ztest_random(ztest_opts.zo_datasets);
5387                 ztest_dataset_destroy(d);
5388         }
5389         zs->zs_enospc_count = 0;
5390 
5391         tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5392             UMEM_NOFAIL);
5393 
5394         if (ztest_opts.zo_verbose >= 4)
5395                 (void) printf("starting main threads...\n");
5396 
5397         /*
5398          * Kick off all the tests that run in parallel.
5399          */
5400         for (int t = 0; t < ztest_opts.zo_threads; t++) {
5401                 if (t < ztest_opts.zo_datasets &&
5402                     ztest_dataset_open(t) != 0)
5403                         return;
5404                 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5405                     THR_BOUND, &tid[t]) == 0);
5406         }
5407 
5408         /*
5409          * Wait for all of the tests to complete.  We go in reverse order
5410          * so we don't close datasets while threads are still using them.
5411          */
5412         for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5413                 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5414                 if (t < ztest_opts.zo_datasets)
5415                         ztest_dataset_close(t);
5416         }
5417 
5418         txg_wait_synced(spa_get_dsl(spa), 0);
5419 
5420         zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5421         zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5422 
5423         umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5424 
5425         /* Kill the resume thread */
5426         ztest_exiting = B_TRUE;
5427         VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5428         ztest_resume(spa);
5429 
5430         /*
5431          * Right before closing the pool, kick off a bunch of async I/O;
5432          * spa_close() should wait for it to complete.
5433          */
5434         for (uint64_t object = 1; object < 50; object++)
5435                 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5436 
5437         spa_close(spa, FTAG);
5438 
5439         /*
5440          * Verify that we can loop over all pools.
5441          */
5442         mutex_enter(&spa_namespace_lock);
5443         for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5444                 if (ztest_opts.zo_verbose > 3)
5445                         (void) printf("spa_next: found %s\n", spa_name(spa));
5446         mutex_exit(&spa_namespace_lock);
5447 
5448         /*
5449          * Verify that we can export the pool and reimport it under a
5450          * different name.
5451          */
5452         if (ztest_random(2) == 0) {
5453                 char name[MAXNAMELEN];
5454                 (void) snprintf(name, MAXNAMELEN, "%s_import",
5455                     ztest_opts.zo_pool);
5456                 ztest_spa_import_export(ztest_opts.zo_pool, name);
5457                 ztest_spa_import_export(name, ztest_opts.zo_pool);
5458         }
5459 
5460         kernel_fini();
5461 
5462         list_destroy(&zcl.zcl_callbacks);
5463 
5464         (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5465 
5466         (void) rwlock_destroy(&ztest_name_lock);
5467         (void) _mutex_destroy(&ztest_vdev_lock);
5468 }
5469 
5470 static void
5471 ztest_freeze(void)
5472 {
5473         ztest_ds_t *zd = &ztest_ds[0];
5474         spa_t *spa;
5475         int numloops = 0;
5476 
5477         if (ztest_opts.zo_verbose >= 3)
5478                 (void) printf("testing spa_freeze()...\n");
5479 
5480         kernel_init(FREAD | FWRITE);
5481         VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5482         VERIFY3U(0, ==, ztest_dataset_open(0));
5483 
5484         /*
5485          * Force the first log block to be transactionally allocated.
5486          * We have to do this before we freeze the pool -- otherwise
5487          * the log chain won't be anchored.
5488          */
5489         while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5490                 ztest_dmu_object_alloc_free(zd, 0);
5491                 zil_commit(zd->zd_zilog, 0);
5492         }
5493 
5494         txg_wait_synced(spa_get_dsl(spa), 0);
5495 
5496         /*
5497          * Freeze the pool.  This stops spa_sync() from doing anything,
5498          * so that the only way to record changes from now on is the ZIL.
5499          */
5500         spa_freeze(spa);
5501 
5502         /*
5503          * Run tests that generate log records but don't alter the pool config
5504          * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5505          * We do a txg_wait_synced() after each iteration to force the txg
5506          * to increase well beyond the last synced value in the uberblock.
5507          * The ZIL should be OK with that.
5508          */
5509         while (ztest_random(10) != 0 &&
5510             numloops++ < ztest_opts.zo_maxloops) {
5511                 ztest_dmu_write_parallel(zd, 0);
5512                 ztest_dmu_object_alloc_free(zd, 0);
5513                 txg_wait_synced(spa_get_dsl(spa), 0);
5514         }
5515 
5516         /*
5517          * Commit all of the changes we just generated.
5518          */
5519         zil_commit(zd->zd_zilog, 0);
5520         txg_wait_synced(spa_get_dsl(spa), 0);
5521 
5522         /*
5523          * Close our dataset and close the pool.
5524          */
5525         ztest_dataset_close(0);
5526         spa_close(spa, FTAG);
5527         kernel_fini();
5528 
5529         /*
5530          * Open and close the pool and dataset to induce log replay.
5531          */
5532         kernel_init(FREAD | FWRITE);
5533         VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5534         VERIFY3U(0, ==, ztest_dataset_open(0));
5535         ztest_dataset_close(0);
5536         spa_close(spa, FTAG);
5537         kernel_fini();
5538 }
5539 
5540 void
5541 print_time(hrtime_t t, char *timebuf)
5542 {
5543         hrtime_t s = t / NANOSEC;
5544         hrtime_t m = s / 60;
5545         hrtime_t h = m / 60;
5546         hrtime_t d = h / 24;
5547 
5548         s -= m * 60;
5549         m -= h * 60;
5550         h -= d * 24;
5551 
5552         timebuf[0] = '\0';
5553 
5554         if (d)
5555                 (void) sprintf(timebuf,
5556                     "%llud%02lluh%02llum%02llus", d, h, m, s);
5557         else if (h)
5558                 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5559         else if (m)
5560                 (void) sprintf(timebuf, "%llum%02llus", m, s);
5561         else
5562                 (void) sprintf(timebuf, "%llus", s);
5563 }
5564 
5565 static nvlist_t *
5566 make_random_props()
5567 {
5568         nvlist_t *props;
5569 
5570         VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5571         if (ztest_random(2) == 0)
5572                 return (props);
5573         VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5574 
5575         return (props);
5576 }
5577 
5578 /*
5579  * Create a storage pool with the given name and initial vdev size.
5580  * Then test spa_freeze() functionality.
5581  */
5582 static void
5583 ztest_init(ztest_shared_t *zs)
5584 {
5585         spa_t *spa;
5586         nvlist_t *nvroot, *props;
5587 
5588         VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5589         VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5590 
5591         kernel_init(FREAD | FWRITE);
5592 
5593         /*
5594          * Create the storage pool.
5595          */
5596         (void) spa_destroy(ztest_opts.zo_pool);
5597         ztest_shared->zs_vdev_next_leaf = 0;
5598         zs->zs_splits = 0;
5599         zs->zs_mirrors = ztest_opts.zo_mirrors;
5600         nvroot = make_vdev_root(NULL, NULL, ztest_opts.zo_vdev_size, 0,
5601             0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5602         props = make_random_props();
5603         for (int i = 0; i < SPA_FEATURES; i++) {
5604                 char buf[1024];
5605                 (void) snprintf(buf, sizeof (buf), "feature@%s",
5606                     spa_feature_table[i].fi_uname);
5607                 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5608         }
5609         VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5610         nvlist_free(nvroot);
5611 
5612         VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5613         zs->zs_metaslab_sz =
5614             1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5615 
5616         spa_close(spa, FTAG);
5617 
5618         kernel_fini();
5619 
5620         ztest_run_zdb(ztest_opts.zo_pool);
5621 
5622         ztest_freeze();
5623 
5624         ztest_run_zdb(ztest_opts.zo_pool);
5625 
5626         (void) rwlock_destroy(&ztest_name_lock);
5627         (void) _mutex_destroy(&ztest_vdev_lock);
5628 }
5629 
5630 static void
5631 setup_fds(void)
5632 {
5633         int fd;
5634 
5635         char *tmp = tempnam(NULL, NULL);
5636         fd = open(tmp, O_RDWR | O_CREAT, 0700);
5637         ASSERT3U(fd, ==, ZTEST_FD_DATA);
5638         (void) unlink(tmp);
5639         free(tmp);
5640 
5641         fd = open("/dev/urandom", O_RDONLY);
5642         ASSERT3U(fd, ==, ZTEST_FD_RAND);
5643 }
5644 
5645 static int
5646 shared_data_size(ztest_shared_hdr_t *hdr)
5647 {
5648         int size;
5649 
5650         size = hdr->zh_hdr_size;
5651         size += hdr->zh_opts_size;
5652         size += hdr->zh_size;
5653         size += hdr->zh_stats_size * hdr->zh_stats_count;
5654         size += hdr->zh_ds_size * hdr->zh_ds_count;
5655 
5656         return (size);
5657 }
5658 
5659 static void
5660 setup_hdr(void)
5661 {
5662         int size;
5663         ztest_shared_hdr_t *hdr;
5664 
5665         hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5666             PROT_READ | PROT_WRITE, MAP_SHARED, ZTEST_FD_DATA, 0);
5667         ASSERT(hdr != MAP_FAILED);
5668 
5669         VERIFY3U(0, ==, ftruncate(ZTEST_FD_DATA, sizeof (ztest_shared_hdr_t)));
5670 
5671         hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5672         hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5673         hdr->zh_size = sizeof (ztest_shared_t);
5674         hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5675         hdr->zh_stats_count = ZTEST_FUNCS;
5676         hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5677         hdr->zh_ds_count = ztest_opts.zo_datasets;
5678 
5679         size = shared_data_size(hdr);
5680         VERIFY3U(0, ==, ftruncate(ZTEST_FD_DATA, size));
5681 
5682         (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5683 }
5684 
5685 static void
5686 setup_data(void)
5687 {
5688         int size, offset;
5689         ztest_shared_hdr_t *hdr;
5690         uint8_t *buf;
5691 
5692         hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5693             PROT_READ, MAP_SHARED, ZTEST_FD_DATA, 0);
5694         ASSERT(hdr != MAP_FAILED);
5695 
5696         size = shared_data_size(hdr);
5697 
5698         (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5699         hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5700             PROT_READ | PROT_WRITE, MAP_SHARED, ZTEST_FD_DATA, 0);
5701         ASSERT(hdr != MAP_FAILED);
5702         buf = (uint8_t *)hdr;
5703 
5704         offset = hdr->zh_hdr_size;
5705         ztest_shared_opts = (void *)&buf[offset];
5706         offset += hdr->zh_opts_size;
5707         ztest_shared = (void *)&buf[offset];
5708         offset += hdr->zh_size;
5709         ztest_shared_callstate = (void *)&buf[offset];
5710         offset += hdr->zh_stats_size * hdr->zh_stats_count;
5711         ztest_shared_ds = (void *)&buf[offset];
5712 }
5713 
5714 static boolean_t
5715 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5716 {
5717         pid_t pid;
5718         int status;
5719         char cmdbuf[MAXPATHLEN];
5720 
5721         pid = fork();
5722 
5723         if (cmd == NULL) {
5724                 (void) strlcpy(cmdbuf, getexecname(), sizeof (cmdbuf));
5725                 cmd = cmdbuf;
5726         }
5727 
5728         if (pid == -1)
5729                 fatal(1, "fork failed");
5730 
5731         if (pid == 0) { /* child */
5732                 char *emptyargv[2] = { cmd, NULL };
5733 
5734                 struct rlimit rl = { 1024, 1024 };
5735                 (void) setrlimit(RLIMIT_NOFILE, &rl);
5736                 (void) enable_extended_FILE_stdio(-1, -1);
5737                 if (libpath != NULL)
5738                         VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5739                 (void) execv(cmd, emptyargv);
5740                 ztest_dump_core = B_FALSE;
5741                 fatal(B_TRUE, "exec failed: %s", cmd);
5742         }
5743 
5744         while (waitpid(pid, &status, 0) != pid)
5745                 continue;
5746         if (statusp != NULL)
5747                 *statusp = status;
5748 
5749         if (WIFEXITED(status)) {
5750                 if (WEXITSTATUS(status) != 0) {
5751                         (void) fprintf(stderr, "child exited with code %d\n",
5752                             WEXITSTATUS(status));
5753                         exit(2);
5754                 }
5755                 return (B_FALSE);
5756         } else if (WIFSIGNALED(status)) {
5757                 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
5758                         (void) fprintf(stderr, "child died with signal %d\n",
5759                             WTERMSIG(status));
5760                         exit(3);
5761                 }
5762                 return (B_TRUE);
5763         } else {
5764                 (void) fprintf(stderr, "something strange happened to child\n");
5765                 exit(4);
5766                 /* NOTREACHED */
5767         }
5768 }
5769 
5770 static void
5771 ztest_run_init(void)
5772 {
5773         ztest_shared_t *zs = ztest_shared;
5774 
5775         ASSERT(ztest_opts.zo_init != 0);
5776 
5777         /*
5778          * Blow away any existing copy of zpool.cache
5779          */
5780         (void) remove(spa_config_path);
5781 
5782         /*
5783          * Create and initialize our storage pool.
5784          */
5785         for (int i = 1; i <= ztest_opts.zo_init; i++) {
5786                 bzero(zs, sizeof (ztest_shared_t));
5787                 if (ztest_opts.zo_verbose >= 3 &&
5788                     ztest_opts.zo_init != 1) {
5789                         (void) printf("ztest_init(), pass %d\n", i);
5790                 }
5791                 ztest_init(zs);
5792         }
5793 }
5794 
5795 int
5796 main(int argc, char **argv)
5797 {
5798         int kills = 0;
5799         int iters = 0;
5800         int older = 0;
5801         int newer = 0;
5802         ztest_shared_t *zs;
5803         ztest_info_t *zi;
5804         ztest_shared_callstate_t *zc;
5805         char timebuf[100];
5806         char numbuf[6];
5807         spa_t *spa;
5808         char cmd[MAXNAMELEN];
5809         boolean_t hasalt;
5810 
5811         boolean_t ischild = (0 == lseek(ZTEST_FD_DATA, 0, SEEK_CUR));
5812         ASSERT(ischild || errno == EBADF);
5813 
5814         (void) setvbuf(stdout, NULL, _IOLBF, 0);
5815 
5816         if (!ischild) {
5817                 process_options(argc, argv);
5818 
5819                 setup_fds();
5820                 setup_hdr();
5821                 setup_data();
5822                 bcopy(&ztest_opts, ztest_shared_opts,
5823                     sizeof (*ztest_shared_opts));
5824         } else {
5825                 setup_data();
5826                 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
5827         }
5828         ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
5829 
5830         /* Override location of zpool.cache */
5831         (void) asprintf((char **)&spa_config_path, "%s/zpool.cache",
5832             ztest_opts.zo_dir);
5833 
5834         ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
5835             UMEM_NOFAIL);
5836         zs = ztest_shared;
5837 
5838         if (ischild) {
5839                 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
5840                 metaslab_df_alloc_threshold =
5841                     zs->zs_metaslab_df_alloc_threshold;
5842 
5843                 if (zs->zs_do_init)
5844                         ztest_run_init();
5845                 else
5846                         ztest_run(zs);
5847                 exit(0);
5848         }
5849 
5850         hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
5851 
5852         if (ztest_opts.zo_verbose >= 1) {
5853                 (void) printf("%llu vdevs, %d datasets, %d threads,"
5854                     " %llu seconds...\n",
5855                     (u_longlong_t)ztest_opts.zo_vdevs,
5856                     ztest_opts.zo_datasets,
5857                     ztest_opts.zo_threads,
5858                     (u_longlong_t)ztest_opts.zo_time);
5859         }
5860 
5861         (void) strlcpy(cmd, getexecname(), sizeof (cmd));
5862 
5863         zs->zs_do_init = B_TRUE;
5864         if (strlen(ztest_opts.zo_alt_ztest) != 0) {
5865                 if (ztest_opts.zo_verbose >= 1) {
5866                         (void) printf("Executing older ztest for "
5867                             "initialization: %s\n", ztest_opts.zo_alt_ztest);
5868                 }
5869                 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
5870                     ztest_opts.zo_alt_libpath, B_FALSE, NULL));
5871         } else {
5872                 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
5873         }
5874         zs->zs_do_init = B_FALSE;
5875 
5876         zs->zs_proc_start = gethrtime();
5877         zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
5878 
5879         for (int f = 0; f < ZTEST_FUNCS; f++) {
5880                 zi = &ztest_info[f];
5881                 zc = ZTEST_GET_SHARED_CALLSTATE(f);
5882                 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
5883                         zc->zc_next = UINT64_MAX;
5884                 else
5885                         zc->zc_next = zs->zs_proc_start +
5886                             ztest_random(2 * zi->zi_interval[0] + 1);
5887         }
5888 
5889         /*
5890          * Run the tests in a loop.  These tests include fault injection
5891          * to verify that self-healing data works, and forced crashes
5892          * to verify that we never lose on-disk consistency.
5893          */
5894         while (gethrtime() < zs->zs_proc_stop) {
5895                 int status;
5896                 boolean_t killed;
5897 
5898                 /*
5899                  * Initialize the workload counters for each function.
5900                  */
5901                 for (int f = 0; f < ZTEST_FUNCS; f++) {
5902                         zc = ZTEST_GET_SHARED_CALLSTATE(f);
5903                         zc->zc_count = 0;
5904                         zc->zc_time = 0;
5905                 }
5906 
5907                 /* Set the allocation switch size */
5908                 zs->zs_metaslab_df_alloc_threshold =
5909                     ztest_random(zs->zs_metaslab_sz / 4) + 1;
5910 
5911                 if (!hasalt || ztest_random(2) == 0) {
5912                         if (hasalt && ztest_opts.zo_verbose >= 1) {
5913                                 (void) printf("Executing newer ztest: %s\n",
5914                                     cmd);
5915                         }
5916                         newer++;
5917                         killed = exec_child(cmd, NULL, B_TRUE, &status);
5918                 } else {
5919                         if (hasalt && ztest_opts.zo_verbose >= 1) {
5920                                 (void) printf("Executing older ztest: %s\n",
5921                                     ztest_opts.zo_alt_ztest);
5922                         }
5923                         older++;
5924                         killed = exec_child(ztest_opts.zo_alt_ztest,
5925                             ztest_opts.zo_alt_libpath, B_TRUE, &status);
5926                 }
5927 
5928                 if (killed)
5929                         kills++;
5930                 iters++;
5931 
5932                 if (ztest_opts.zo_verbose >= 1) {
5933                         hrtime_t now = gethrtime();
5934 
5935                         now = MIN(now, zs->zs_proc_stop);
5936                         print_time(zs->zs_proc_stop - now, timebuf);
5937                         nicenum(zs->zs_space, numbuf);
5938 
5939                         (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
5940                             "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
5941                             iters,
5942                             WIFEXITED(status) ? "Complete" : "SIGKILL",
5943                             (u_longlong_t)zs->zs_enospc_count,
5944                             100.0 * zs->zs_alloc / zs->zs_space,
5945                             numbuf,
5946                             100.0 * (now - zs->zs_proc_start) /
5947                             (ztest_opts.zo_time * NANOSEC), timebuf);
5948                 }
5949 
5950                 if (ztest_opts.zo_verbose >= 2) {
5951                         (void) printf("\nWorkload summary:\n\n");
5952                         (void) printf("%7s %9s   %s\n",
5953                             "Calls", "Time", "Function");
5954                         (void) printf("%7s %9s   %s\n",
5955                             "-----", "----", "--------");
5956                         for (int f = 0; f < ZTEST_FUNCS; f++) {
5957                                 Dl_info dli;
5958 
5959                                 zi = &ztest_info[f];
5960                                 zc = ZTEST_GET_SHARED_CALLSTATE(f);
5961                                 print_time(zc->zc_time, timebuf);
5962                                 (void) dladdr((void *)zi->zi_func, &dli);
5963                                 (void) printf("%7llu %9s   %s\n",
5964                                     (u_longlong_t)zc->zc_count, timebuf,
5965                                     dli.dli_sname);
5966                         }
5967                         (void) printf("\n");
5968                 }
5969 
5970                 /*
5971                  * It's possible that we killed a child during a rename test,
5972                  * in which case we'll have a 'ztest_tmp' pool lying around
5973                  * instead of 'ztest'.  Do a blind rename in case this happened.
5974                  */
5975                 kernel_init(FREAD);
5976                 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
5977                         spa_close(spa, FTAG);
5978                 } else {
5979                         char tmpname[MAXNAMELEN];
5980                         kernel_fini();
5981                         kernel_init(FREAD | FWRITE);
5982                         (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
5983                             ztest_opts.zo_pool);
5984                         (void) spa_rename(tmpname, ztest_opts.zo_pool);
5985                 }
5986                 kernel_fini();
5987 
5988                 ztest_run_zdb(ztest_opts.zo_pool);
5989         }
5990 
5991         if (ztest_opts.zo_verbose >= 1) {
5992                 if (hasalt) {
5993                         (void) printf("%d runs of older ztest: %s\n", older,
5994                             ztest_opts.zo_alt_ztest);
5995                         (void) printf("%d runs of newer ztest: %s\n", newer,
5996                             cmd);
5997                 }
5998                 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
5999                     kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6000         }
6001 
6002         return (0);
6003 }