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_destroy.h>
 107 #include <sys/dsl_scan.h>
 108 #include <sys/zio_checksum.h>
 109 #include <sys/refcount.h>
 110 #include <sys/zfeature.h>
 111 #include <sys/dsl_userhold.h>
 112 #include <stdio.h>
 113 #include <stdio_ext.h>
 114 #include <stdlib.h>
 115 #include <unistd.h>
 116 #include <signal.h>
 117 #include <umem.h>
 118 #include <dlfcn.h>
 119 #include <ctype.h>
 120 #include <math.h>
 121 #include <sys/fs/zfs.h>
 122 #include <libnvpair.h>
 123 
 124 static int ztest_fd_data = -1;
 125 static int ztest_fd_rand = -1;
 126 
 127 typedef struct ztest_shared_hdr {
 128         uint64_t        zh_hdr_size;
 129         uint64_t        zh_opts_size;
 130         uint64_t        zh_size;
 131         uint64_t        zh_stats_size;
 132         uint64_t        zh_stats_count;
 133         uint64_t        zh_ds_size;
 134         uint64_t        zh_ds_count;
 135 } ztest_shared_hdr_t;
 136 
 137 static ztest_shared_hdr_t *ztest_shared_hdr;
 138 
 139 typedef struct ztest_shared_opts {
 140         char zo_pool[MAXNAMELEN];
 141         char zo_dir[MAXNAMELEN];
 142         char zo_alt_ztest[MAXNAMELEN];
 143         char zo_alt_libpath[MAXNAMELEN];
 144         uint64_t zo_vdevs;
 145         uint64_t zo_vdevtime;
 146         size_t zo_vdev_size;
 147         int zo_ashift;
 148         int zo_mirrors;
 149         int zo_raidz;
 150         int zo_raidz_parity;
 151         int zo_datasets;
 152         int zo_threads;
 153         uint64_t zo_passtime;
 154         uint64_t zo_killrate;
 155         int zo_verbose;
 156         int zo_init;
 157         uint64_t zo_time;
 158         uint64_t zo_maxloops;
 159         uint64_t zo_metaslab_gang_bang;
 160 } ztest_shared_opts_t;
 161 
 162 static const ztest_shared_opts_t ztest_opts_defaults = {
 163         .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
 164         .zo_dir = { '/', 't', 'm', 'p', '\0' },
 165         .zo_alt_ztest = { '\0' },
 166         .zo_alt_libpath = { '\0' },
 167         .zo_vdevs = 5,
 168         .zo_ashift = SPA_MINBLOCKSHIFT,
 169         .zo_mirrors = 2,
 170         .zo_raidz = 4,
 171         .zo_raidz_parity = 1,
 172         .zo_vdev_size = SPA_MINDEVSIZE,
 173         .zo_datasets = 7,
 174         .zo_threads = 23,
 175         .zo_passtime = 60,              /* 60 seconds */
 176         .zo_killrate = 70,              /* 70% kill rate */
 177         .zo_verbose = 0,
 178         .zo_init = 1,
 179         .zo_time = 300,                 /* 5 minutes */
 180         .zo_maxloops = 50,              /* max loops during spa_freeze() */
 181         .zo_metaslab_gang_bang = 32 << 10
 182 };
 183 
 184 extern uint64_t metaslab_gang_bang;
 185 extern uint64_t metaslab_df_alloc_threshold;
 186 
 187 static ztest_shared_opts_t *ztest_shared_opts;
 188 static ztest_shared_opts_t ztest_opts;
 189 
 190 typedef struct ztest_shared_ds {
 191         uint64_t        zd_seq;
 192 } ztest_shared_ds_t;
 193 
 194 static ztest_shared_ds_t *ztest_shared_ds;
 195 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
 196 
 197 #define BT_MAGIC        0x123456789abcdefULL
 198 #define MAXFAULTS() \
 199         (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
 200 
 201 enum ztest_io_type {
 202         ZTEST_IO_WRITE_TAG,
 203         ZTEST_IO_WRITE_PATTERN,
 204         ZTEST_IO_WRITE_ZEROES,
 205         ZTEST_IO_TRUNCATE,
 206         ZTEST_IO_SETATTR,
 207         ZTEST_IO_REWRITE,
 208         ZTEST_IO_TYPES
 209 };
 210 
 211 typedef struct ztest_block_tag {
 212         uint64_t        bt_magic;
 213         uint64_t        bt_objset;
 214         uint64_t        bt_object;
 215         uint64_t        bt_offset;
 216         uint64_t        bt_gen;
 217         uint64_t        bt_txg;
 218         uint64_t        bt_crtxg;
 219 } ztest_block_tag_t;
 220 
 221 typedef struct bufwad {
 222         uint64_t        bw_index;
 223         uint64_t        bw_txg;
 224         uint64_t        bw_data;
 225 } bufwad_t;
 226 
 227 /*
 228  * XXX -- fix zfs range locks to be generic so we can use them here.
 229  */
 230 typedef enum {
 231         RL_READER,
 232         RL_WRITER,
 233         RL_APPEND
 234 } rl_type_t;
 235 
 236 typedef struct rll {
 237         void            *rll_writer;
 238         int             rll_readers;
 239         mutex_t         rll_lock;
 240         cond_t          rll_cv;
 241 } rll_t;
 242 
 243 typedef struct rl {
 244         uint64_t        rl_object;
 245         uint64_t        rl_offset;
 246         uint64_t        rl_size;
 247         rll_t           *rl_lock;
 248 } rl_t;
 249 
 250 #define ZTEST_RANGE_LOCKS       64
 251 #define ZTEST_OBJECT_LOCKS      64
 252 
 253 /*
 254  * Object descriptor.  Used as a template for object lookup/create/remove.
 255  */
 256 typedef struct ztest_od {
 257         uint64_t        od_dir;
 258         uint64_t        od_object;
 259         dmu_object_type_t od_type;
 260         dmu_object_type_t od_crtype;
 261         uint64_t        od_blocksize;
 262         uint64_t        od_crblocksize;
 263         uint64_t        od_gen;
 264         uint64_t        od_crgen;
 265         char            od_name[MAXNAMELEN];
 266 } ztest_od_t;
 267 
 268 /*
 269  * Per-dataset state.
 270  */
 271 typedef struct ztest_ds {
 272         ztest_shared_ds_t *zd_shared;
 273         objset_t        *zd_os;
 274         rwlock_t        zd_zilog_lock;
 275         zilog_t         *zd_zilog;
 276         ztest_od_t      *zd_od;         /* debugging aid */
 277         char            zd_name[MAXNAMELEN];
 278         mutex_t         zd_dirobj_lock;
 279         rll_t           zd_object_lock[ZTEST_OBJECT_LOCKS];
 280         rll_t           zd_range_lock[ZTEST_RANGE_LOCKS];
 281 } ztest_ds_t;
 282 
 283 /*
 284  * Per-iteration state.
 285  */
 286 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
 287 
 288 typedef struct ztest_info {
 289         ztest_func_t    *zi_func;       /* test function */
 290         uint64_t        zi_iters;       /* iterations per execution */
 291         uint64_t        *zi_interval;   /* execute every <interval> seconds */
 292 } ztest_info_t;
 293 
 294 typedef struct ztest_shared_callstate {
 295         uint64_t        zc_count;       /* per-pass count */
 296         uint64_t        zc_time;        /* per-pass time */
 297         uint64_t        zc_next;        /* next time to call this function */
 298 } ztest_shared_callstate_t;
 299 
 300 static ztest_shared_callstate_t *ztest_shared_callstate;
 301 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
 302 
 303 /*
 304  * Note: these aren't static because we want dladdr() to work.
 305  */
 306 ztest_func_t ztest_dmu_read_write;
 307 ztest_func_t ztest_dmu_write_parallel;
 308 ztest_func_t ztest_dmu_object_alloc_free;
 309 ztest_func_t ztest_dmu_commit_callbacks;
 310 ztest_func_t ztest_zap;
 311 ztest_func_t ztest_zap_parallel;
 312 ztest_func_t ztest_zil_commit;
 313 ztest_func_t ztest_zil_remount;
 314 ztest_func_t ztest_dmu_read_write_zcopy;
 315 ztest_func_t ztest_dmu_objset_create_destroy;
 316 ztest_func_t ztest_dmu_prealloc;
 317 ztest_func_t ztest_fzap;
 318 ztest_func_t ztest_dmu_snapshot_create_destroy;
 319 ztest_func_t ztest_dsl_prop_get_set;
 320 ztest_func_t ztest_spa_prop_get_set;
 321 ztest_func_t ztest_spa_create_destroy;
 322 ztest_func_t ztest_fault_inject;
 323 ztest_func_t ztest_ddt_repair;
 324 ztest_func_t ztest_dmu_snapshot_hold;
 325 ztest_func_t ztest_spa_rename;
 326 ztest_func_t ztest_scrub;
 327 ztest_func_t ztest_dsl_dataset_promote_busy;
 328 ztest_func_t ztest_vdev_attach_detach;
 329 ztest_func_t ztest_vdev_LUN_growth;
 330 ztest_func_t ztest_vdev_add_remove;
 331 ztest_func_t ztest_vdev_aux_add_remove;
 332 ztest_func_t ztest_split_pool;
 333 ztest_func_t ztest_reguid;
 334 ztest_func_t ztest_spa_upgrade;
 335 
 336 uint64_t zopt_always = 0ULL * NANOSEC;          /* all the time */
 337 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;  /* every 1/10 second */
 338 uint64_t zopt_often = 1ULL * NANOSEC;           /* every second */
 339 uint64_t zopt_sometimes = 10ULL * NANOSEC;      /* every 10 seconds */
 340 uint64_t zopt_rarely = 60ULL * NANOSEC;         /* every 60 seconds */
 341 
 342 ztest_info_t ztest_info[] = {
 343         { ztest_dmu_read_write,                 1,      &zopt_always        },
 344         { ztest_dmu_write_parallel,             10,     &zopt_always        },
 345         { ztest_dmu_object_alloc_free,          1,      &zopt_always        },
 346         { ztest_dmu_commit_callbacks,           1,      &zopt_always        },
 347         { ztest_zap,                            30,     &zopt_always        },
 348         { ztest_zap_parallel,                   100,    &zopt_always        },
 349         { ztest_split_pool,                     1,      &zopt_always        },
 350         { ztest_zil_commit,                     1,      &zopt_incessant     },
 351         { ztest_zil_remount,                    1,      &zopt_sometimes     },
 352         { ztest_dmu_read_write_zcopy,           1,      &zopt_often },
 353         { ztest_dmu_objset_create_destroy,      1,      &zopt_often },
 354         { ztest_dsl_prop_get_set,               1,      &zopt_often },
 355         { ztest_spa_prop_get_set,               1,      &zopt_sometimes     },
 356 #if 0
 357         { ztest_dmu_prealloc,                   1,      &zopt_sometimes     },
 358 #endif
 359         { ztest_fzap,                           1,      &zopt_sometimes     },
 360         { ztest_dmu_snapshot_create_destroy,    1,      &zopt_sometimes     },
 361         { ztest_spa_create_destroy,             1,      &zopt_sometimes     },
 362         { ztest_fault_inject,                   1,      &zopt_sometimes     },
 363         { ztest_ddt_repair,                     1,      &zopt_sometimes     },
 364         { ztest_dmu_snapshot_hold,              1,      &zopt_sometimes     },
 365         { ztest_reguid,                         1,      &zopt_sometimes },
 366         { ztest_spa_rename,                     1,      &zopt_rarely        },
 367         { ztest_scrub,                          1,      &zopt_rarely        },
 368         { ztest_spa_upgrade,                    1,      &zopt_rarely        },
 369         { ztest_dsl_dataset_promote_busy,       1,      &zopt_rarely        },
 370         { ztest_vdev_attach_detach,             1,      &zopt_sometimes     },
 371         { ztest_vdev_LUN_growth,                1,      &zopt_rarely        },
 372         { ztest_vdev_add_remove,                1,
 373             &ztest_opts.zo_vdevtime                         },
 374         { ztest_vdev_aux_add_remove,            1,
 375             &ztest_opts.zo_vdevtime                         },
 376 };
 377 
 378 #define ZTEST_FUNCS     (sizeof (ztest_info) / sizeof (ztest_info_t))
 379 
 380 /*
 381  * The following struct is used to hold a list of uncalled commit callbacks.
 382  * The callbacks are ordered by txg number.
 383  */
 384 typedef struct ztest_cb_list {
 385         mutex_t zcl_callbacks_lock;
 386         list_t  zcl_callbacks;
 387 } ztest_cb_list_t;
 388 
 389 /*
 390  * Stuff we need to share writably between parent and child.
 391  */
 392 typedef struct ztest_shared {
 393         boolean_t       zs_do_init;
 394         hrtime_t        zs_proc_start;
 395         hrtime_t        zs_proc_stop;
 396         hrtime_t        zs_thread_start;
 397         hrtime_t        zs_thread_stop;
 398         hrtime_t        zs_thread_kill;
 399         uint64_t        zs_enospc_count;
 400         uint64_t        zs_vdev_next_leaf;
 401         uint64_t        zs_vdev_aux;
 402         uint64_t        zs_alloc;
 403         uint64_t        zs_space;
 404         uint64_t        zs_splits;
 405         uint64_t        zs_mirrors;
 406         uint64_t        zs_metaslab_sz;
 407         uint64_t        zs_metaslab_df_alloc_threshold;
 408         uint64_t        zs_guid;
 409 } ztest_shared_t;
 410 
 411 #define ID_PARALLEL     -1ULL
 412 
 413 static char ztest_dev_template[] = "%s/%s.%llua";
 414 static char ztest_aux_template[] = "%s/%s.%s.%llu";
 415 ztest_shared_t *ztest_shared;
 416 
 417 static spa_t *ztest_spa = NULL;
 418 static ztest_ds_t *ztest_ds;
 419 
 420 static mutex_t ztest_vdev_lock;
 421 
 422 /*
 423  * The ztest_name_lock protects the pool and dataset namespace used by
 424  * the individual tests. To modify the namespace, consumers must grab
 425  * this lock as writer. Grabbing the lock as reader will ensure that the
 426  * namespace does not change while the lock is held.
 427  */
 428 static rwlock_t ztest_name_lock;
 429 
 430 static boolean_t ztest_dump_core = B_TRUE;
 431 static boolean_t ztest_exiting;
 432 
 433 /* Global commit callback list */
 434 static ztest_cb_list_t zcl;
 435 
 436 enum ztest_object {
 437         ZTEST_META_DNODE = 0,
 438         ZTEST_DIROBJ,
 439         ZTEST_OBJECTS
 440 };
 441 
 442 static void usage(boolean_t) __NORETURN;
 443 
 444 /*
 445  * These libumem hooks provide a reasonable set of defaults for the allocator's
 446  * debugging facilities.
 447  */
 448 const char *
 449 _umem_debug_init()
 450 {
 451         return ("default,verbose"); /* $UMEM_DEBUG setting */
 452 }
 453 
 454 const char *
 455 _umem_logging_init(void)
 456 {
 457         return ("fail,contents"); /* $UMEM_LOGGING setting */
 458 }
 459 
 460 #define FATAL_MSG_SZ    1024
 461 
 462 char *fatal_msg;
 463 
 464 static void
 465 fatal(int do_perror, char *message, ...)
 466 {
 467         va_list args;
 468         int save_errno = errno;
 469         char buf[FATAL_MSG_SZ];
 470 
 471         (void) fflush(stdout);
 472 
 473         va_start(args, message);
 474         (void) sprintf(buf, "ztest: ");
 475         /* LINTED */
 476         (void) vsprintf(buf + strlen(buf), message, args);
 477         va_end(args);
 478         if (do_perror) {
 479                 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
 480                     ": %s", strerror(save_errno));
 481         }
 482         (void) fprintf(stderr, "%s\n", buf);
 483         fatal_msg = buf;                        /* to ease debugging */
 484         if (ztest_dump_core)
 485                 abort();
 486         exit(3);
 487 }
 488 
 489 static int
 490 str2shift(const char *buf)
 491 {
 492         const char *ends = "BKMGTPEZ";
 493         int i;
 494 
 495         if (buf[0] == '\0')
 496                 return (0);
 497         for (i = 0; i < strlen(ends); i++) {
 498                 if (toupper(buf[0]) == ends[i])
 499                         break;
 500         }
 501         if (i == strlen(ends)) {
 502                 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
 503                     buf);
 504                 usage(B_FALSE);
 505         }
 506         if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
 507                 return (10*i);
 508         }
 509         (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
 510         usage(B_FALSE);
 511         /* NOTREACHED */
 512 }
 513 
 514 static uint64_t
 515 nicenumtoull(const char *buf)
 516 {
 517         char *end;
 518         uint64_t val;
 519 
 520         val = strtoull(buf, &end, 0);
 521         if (end == buf) {
 522                 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
 523                 usage(B_FALSE);
 524         } else if (end[0] == '.') {
 525                 double fval = strtod(buf, &end);
 526                 fval *= pow(2, str2shift(end));
 527                 if (fval > UINT64_MAX) {
 528                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 529                             buf);
 530                         usage(B_FALSE);
 531                 }
 532                 val = (uint64_t)fval;
 533         } else {
 534                 int shift = str2shift(end);
 535                 if (shift >= 64 || (val << shift) >> shift != val) {
 536                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 537                             buf);
 538                         usage(B_FALSE);
 539                 }
 540                 val <<= shift;
 541         }
 542         return (val);
 543 }
 544 
 545 static void
 546 usage(boolean_t requested)
 547 {
 548         const ztest_shared_opts_t *zo = &ztest_opts_defaults;
 549 
 550         char nice_vdev_size[10];
 551         char nice_gang_bang[10];
 552         FILE *fp = requested ? stdout : stderr;
 553 
 554         nicenum(zo->zo_vdev_size, nice_vdev_size);
 555         nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
 556 
 557         (void) fprintf(fp, "Usage: %s\n"
 558             "\t[-v vdevs (default: %llu)]\n"
 559             "\t[-s size_of_each_vdev (default: %s)]\n"
 560             "\t[-a alignment_shift (default: %d)] use 0 for random\n"
 561             "\t[-m mirror_copies (default: %d)]\n"
 562             "\t[-r raidz_disks (default: %d)]\n"
 563             "\t[-R raidz_parity (default: %d)]\n"
 564             "\t[-d datasets (default: %d)]\n"
 565             "\t[-t threads (default: %d)]\n"
 566             "\t[-g gang_block_threshold (default: %s)]\n"
 567             "\t[-i init_count (default: %d)] initialize pool i times\n"
 568             "\t[-k kill_percentage (default: %llu%%)]\n"
 569             "\t[-p pool_name (default: %s)]\n"
 570             "\t[-f dir (default: %s)] file directory for vdev files\n"
 571             "\t[-V] verbose (use multiple times for ever more blather)\n"
 572             "\t[-E] use existing pool instead of creating new one\n"
 573             "\t[-T time (default: %llu sec)] total run time\n"
 574             "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
 575             "\t[-P passtime (default: %llu sec)] time per pass\n"
 576             "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
 577             "\t[-h] (print help)\n"
 578             "",
 579             zo->zo_pool,
 580             (u_longlong_t)zo->zo_vdevs,                      /* -v */
 581             nice_vdev_size,                             /* -s */
 582             zo->zo_ashift,                           /* -a */
 583             zo->zo_mirrors,                          /* -m */
 584             zo->zo_raidz,                            /* -r */
 585             zo->zo_raidz_parity,                     /* -R */
 586             zo->zo_datasets,                         /* -d */
 587             zo->zo_threads,                          /* -t */
 588             nice_gang_bang,                             /* -g */
 589             zo->zo_init,                             /* -i */
 590             (u_longlong_t)zo->zo_killrate,           /* -k */
 591             zo->zo_pool,                             /* -p */
 592             zo->zo_dir,                                      /* -f */
 593             (u_longlong_t)zo->zo_time,                       /* -T */
 594             (u_longlong_t)zo->zo_maxloops,           /* -F */
 595             (u_longlong_t)zo->zo_passtime);
 596         exit(requested ? 0 : 1);
 597 }
 598 
 599 static void
 600 process_options(int argc, char **argv)
 601 {
 602         char *path;
 603         ztest_shared_opts_t *zo = &ztest_opts;
 604 
 605         int opt;
 606         uint64_t value;
 607         char altdir[MAXNAMELEN] = { 0 };
 608 
 609         bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
 610 
 611         while ((opt = getopt(argc, argv,
 612             "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
 613                 value = 0;
 614                 switch (opt) {
 615                 case 'v':
 616                 case 's':
 617                 case 'a':
 618                 case 'm':
 619                 case 'r':
 620                 case 'R':
 621                 case 'd':
 622                 case 't':
 623                 case 'g':
 624                 case 'i':
 625                 case 'k':
 626                 case 'T':
 627                 case 'P':
 628                 case 'F':
 629                         value = nicenumtoull(optarg);
 630                 }
 631                 switch (opt) {
 632                 case 'v':
 633                         zo->zo_vdevs = value;
 634                         break;
 635                 case 's':
 636                         zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
 637                         break;
 638                 case 'a':
 639                         zo->zo_ashift = value;
 640                         break;
 641                 case 'm':
 642                         zo->zo_mirrors = value;
 643                         break;
 644                 case 'r':
 645                         zo->zo_raidz = MAX(1, value);
 646                         break;
 647                 case 'R':
 648                         zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
 649                         break;
 650                 case 'd':
 651                         zo->zo_datasets = MAX(1, value);
 652                         break;
 653                 case 't':
 654                         zo->zo_threads = MAX(1, value);
 655                         break;
 656                 case 'g':
 657                         zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
 658                             value);
 659                         break;
 660                 case 'i':
 661                         zo->zo_init = value;
 662                         break;
 663                 case 'k':
 664                         zo->zo_killrate = value;
 665                         break;
 666                 case 'p':
 667                         (void) strlcpy(zo->zo_pool, optarg,
 668                             sizeof (zo->zo_pool));
 669                         break;
 670                 case 'f':
 671                         path = realpath(optarg, NULL);
 672                         if (path == NULL) {
 673                                 (void) fprintf(stderr, "error: %s: %s\n",
 674                                     optarg, strerror(errno));
 675                                 usage(B_FALSE);
 676                         } else {
 677                                 (void) strlcpy(zo->zo_dir, path,
 678                                     sizeof (zo->zo_dir));
 679                         }
 680                         break;
 681                 case 'V':
 682                         zo->zo_verbose++;
 683                         break;
 684                 case 'E':
 685                         zo->zo_init = 0;
 686                         break;
 687                 case 'T':
 688                         zo->zo_time = value;
 689                         break;
 690                 case 'P':
 691                         zo->zo_passtime = MAX(1, value);
 692                         break;
 693                 case 'F':
 694                         zo->zo_maxloops = MAX(1, value);
 695                         break;
 696                 case 'B':
 697                         (void) strlcpy(altdir, optarg, sizeof (altdir));
 698                         break;
 699                 case 'h':
 700                         usage(B_TRUE);
 701                         break;
 702                 case '?':
 703                 default:
 704                         usage(B_FALSE);
 705                         break;
 706                 }
 707         }
 708 
 709         zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
 710 
 711         zo->zo_vdevtime =
 712             (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
 713             UINT64_MAX >> 2);
 714 
 715         if (strlen(altdir) > 0) {
 716                 char *cmd;
 717                 char *realaltdir;
 718                 char *bin;
 719                 char *ztest;
 720                 char *isa;
 721                 int isalen;
 722 
 723                 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
 724                 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
 725 
 726                 VERIFY(NULL != realpath(getexecname(), cmd));
 727                 if (0 != access(altdir, F_OK)) {
 728                         ztest_dump_core = B_FALSE;
 729                         fatal(B_TRUE, "invalid alternate ztest path: %s",
 730                             altdir);
 731                 }
 732                 VERIFY(NULL != realpath(altdir, realaltdir));
 733 
 734                 /*
 735                  * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
 736                  * We want to extract <isa> to determine if we should use
 737                  * 32 or 64 bit binaries.
 738                  */
 739                 bin = strstr(cmd, "/usr/bin/");
 740                 ztest = strstr(bin, "/ztest");
 741                 isa = bin + 9;
 742                 isalen = ztest - isa;
 743                 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
 744                     "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
 745                 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
 746                     "%s/usr/lib/%.*s", realaltdir, isalen, isa);
 747 
 748                 if (0 != access(zo->zo_alt_ztest, X_OK)) {
 749                         ztest_dump_core = B_FALSE;
 750                         fatal(B_TRUE, "invalid alternate ztest: %s",
 751                             zo->zo_alt_ztest);
 752                 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
 753                         ztest_dump_core = B_FALSE;
 754                         fatal(B_TRUE, "invalid alternate lib directory %s",
 755                             zo->zo_alt_libpath);
 756                 }
 757 
 758                 umem_free(cmd, MAXPATHLEN);
 759                 umem_free(realaltdir, MAXPATHLEN);
 760         }
 761 }
 762 
 763 static void
 764 ztest_kill(ztest_shared_t *zs)
 765 {
 766         zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
 767         zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
 768         (void) kill(getpid(), SIGKILL);
 769 }
 770 
 771 static uint64_t
 772 ztest_random(uint64_t range)
 773 {
 774         uint64_t r;
 775 
 776         ASSERT3S(ztest_fd_rand, >=, 0);
 777 
 778         if (range == 0)
 779                 return (0);
 780 
 781         if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
 782                 fatal(1, "short read from /dev/urandom");
 783 
 784         return (r % range);
 785 }
 786 
 787 /* ARGSUSED */
 788 static void
 789 ztest_record_enospc(const char *s)
 790 {
 791         ztest_shared->zs_enospc_count++;
 792 }
 793 
 794 static uint64_t
 795 ztest_get_ashift(void)
 796 {
 797         if (ztest_opts.zo_ashift == 0)
 798                 return (SPA_MINBLOCKSHIFT + ztest_random(3));
 799         return (ztest_opts.zo_ashift);
 800 }
 801 
 802 static nvlist_t *
 803 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
 804 {
 805         char pathbuf[MAXPATHLEN];
 806         uint64_t vdev;
 807         nvlist_t *file;
 808 
 809         if (ashift == 0)
 810                 ashift = ztest_get_ashift();
 811 
 812         if (path == NULL) {
 813                 path = pathbuf;
 814 
 815                 if (aux != NULL) {
 816                         vdev = ztest_shared->zs_vdev_aux;
 817                         (void) snprintf(path, sizeof (pathbuf),
 818                             ztest_aux_template, ztest_opts.zo_dir,
 819                             pool == NULL ? ztest_opts.zo_pool : pool,
 820                             aux, vdev);
 821                 } else {
 822                         vdev = ztest_shared->zs_vdev_next_leaf++;
 823                         (void) snprintf(path, sizeof (pathbuf),
 824                             ztest_dev_template, ztest_opts.zo_dir,
 825                             pool == NULL ? ztest_opts.zo_pool : pool, vdev);
 826                 }
 827         }
 828 
 829         if (size != 0) {
 830                 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
 831                 if (fd == -1)
 832                         fatal(1, "can't open %s", path);
 833                 if (ftruncate(fd, size) != 0)
 834                         fatal(1, "can't ftruncate %s", path);
 835                 (void) close(fd);
 836         }
 837 
 838         VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
 839         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
 840         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
 841         VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
 842 
 843         return (file);
 844 }
 845 
 846 static nvlist_t *
 847 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
 848     uint64_t ashift, int r)
 849 {
 850         nvlist_t *raidz, **child;
 851         int c;
 852 
 853         if (r < 2)
 854                 return (make_vdev_file(path, aux, pool, size, ashift));
 855         child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
 856 
 857         for (c = 0; c < r; c++)
 858                 child[c] = make_vdev_file(path, aux, pool, size, ashift);
 859 
 860         VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
 861         VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
 862             VDEV_TYPE_RAIDZ) == 0);
 863         VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
 864             ztest_opts.zo_raidz_parity) == 0);
 865         VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
 866             child, r) == 0);
 867 
 868         for (c = 0; c < r; c++)
 869                 nvlist_free(child[c]);
 870 
 871         umem_free(child, r * sizeof (nvlist_t *));
 872 
 873         return (raidz);
 874 }
 875 
 876 static nvlist_t *
 877 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
 878     uint64_t ashift, int r, int m)
 879 {
 880         nvlist_t *mirror, **child;
 881         int c;
 882 
 883         if (m < 1)
 884                 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
 885 
 886         child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
 887 
 888         for (c = 0; c < m; c++)
 889                 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
 890 
 891         VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
 892         VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
 893             VDEV_TYPE_MIRROR) == 0);
 894         VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
 895             child, m) == 0);
 896 
 897         for (c = 0; c < m; c++)
 898                 nvlist_free(child[c]);
 899 
 900         umem_free(child, m * sizeof (nvlist_t *));
 901 
 902         return (mirror);
 903 }
 904 
 905 static nvlist_t *
 906 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
 907     int log, int r, int m, int t)
 908 {
 909         nvlist_t *root, **child;
 910         int c;
 911 
 912         ASSERT(t > 0);
 913 
 914         child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
 915 
 916         for (c = 0; c < t; c++) {
 917                 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
 918                     r, m);
 919                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
 920                     log) == 0);
 921         }
 922 
 923         VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
 924         VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
 925         VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
 926             child, t) == 0);
 927 
 928         for (c = 0; c < t; c++)
 929                 nvlist_free(child[c]);
 930 
 931         umem_free(child, t * sizeof (nvlist_t *));
 932 
 933         return (root);
 934 }
 935 
 936 /*
 937  * Find a random spa version. Returns back a random spa version in the
 938  * range [initial_version, SPA_VERSION_FEATURES].
 939  */
 940 static uint64_t
 941 ztest_random_spa_version(uint64_t initial_version)
 942 {
 943         uint64_t version = initial_version;
 944 
 945         if (version <= SPA_VERSION_BEFORE_FEATURES) {
 946                 version = version +
 947                     ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
 948         }
 949 
 950         if (version > SPA_VERSION_BEFORE_FEATURES)
 951                 version = SPA_VERSION_FEATURES;
 952 
 953         ASSERT(SPA_VERSION_IS_SUPPORTED(version));
 954         return (version);
 955 }
 956 
 957 static int
 958 ztest_random_blocksize(void)
 959 {
 960         return (1 << (SPA_MINBLOCKSHIFT +
 961             ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
 962 }
 963 
 964 static int
 965 ztest_random_ibshift(void)
 966 {
 967         return (DN_MIN_INDBLKSHIFT +
 968             ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
 969 }
 970 
 971 static uint64_t
 972 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
 973 {
 974         uint64_t top;
 975         vdev_t *rvd = spa->spa_root_vdev;
 976         vdev_t *tvd;
 977 
 978         ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
 979 
 980         do {
 981                 top = ztest_random(rvd->vdev_children);
 982                 tvd = rvd->vdev_child[top];
 983         } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
 984             tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
 985 
 986         return (top);
 987 }
 988 
 989 static uint64_t
 990 ztest_random_dsl_prop(zfs_prop_t prop)
 991 {
 992         uint64_t value;
 993 
 994         do {
 995                 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
 996         } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
 997 
 998         return (value);
 999 }
1000 
1001 static int
1002 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1003     boolean_t inherit)
1004 {
1005         const char *propname = zfs_prop_to_name(prop);
1006         const char *valname;
1007         char setpoint[MAXPATHLEN];
1008         uint64_t curval;
1009         int error;
1010 
1011         error = dsl_prop_set_int(osname, propname,
1012             (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1013 
1014         if (error == ENOSPC) {
1015                 ztest_record_enospc(FTAG);
1016                 return (error);
1017         }
1018         ASSERT0(error);
1019 
1020         VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1021 
1022         if (ztest_opts.zo_verbose >= 6) {
1023                 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1024                 (void) printf("%s %s = %s at '%s'\n",
1025                     osname, propname, valname, setpoint);
1026         }
1027 
1028         return (error);
1029 }
1030 
1031 static int
1032 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1033 {
1034         spa_t *spa = ztest_spa;
1035         nvlist_t *props = NULL;
1036         int error;
1037 
1038         VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1039         VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1040 
1041         error = spa_prop_set(spa, props);
1042 
1043         nvlist_free(props);
1044 
1045         if (error == ENOSPC) {
1046                 ztest_record_enospc(FTAG);
1047                 return (error);
1048         }
1049         ASSERT0(error);
1050 
1051         return (error);
1052 }
1053 
1054 static void
1055 ztest_rll_init(rll_t *rll)
1056 {
1057         rll->rll_writer = NULL;
1058         rll->rll_readers = 0;
1059         VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1060         VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1061 }
1062 
1063 static void
1064 ztest_rll_destroy(rll_t *rll)
1065 {
1066         ASSERT(rll->rll_writer == NULL);
1067         ASSERT(rll->rll_readers == 0);
1068         VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1069         VERIFY(cond_destroy(&rll->rll_cv) == 0);
1070 }
1071 
1072 static void
1073 ztest_rll_lock(rll_t *rll, rl_type_t type)
1074 {
1075         VERIFY(mutex_lock(&rll->rll_lock) == 0);
1076 
1077         if (type == RL_READER) {
1078                 while (rll->rll_writer != NULL)
1079                         (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1080                 rll->rll_readers++;
1081         } else {
1082                 while (rll->rll_writer != NULL || rll->rll_readers)
1083                         (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1084                 rll->rll_writer = curthread;
1085         }
1086 
1087         VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1088 }
1089 
1090 static void
1091 ztest_rll_unlock(rll_t *rll)
1092 {
1093         VERIFY(mutex_lock(&rll->rll_lock) == 0);
1094 
1095         if (rll->rll_writer) {
1096                 ASSERT(rll->rll_readers == 0);
1097                 rll->rll_writer = NULL;
1098         } else {
1099                 ASSERT(rll->rll_readers != 0);
1100                 ASSERT(rll->rll_writer == NULL);
1101                 rll->rll_readers--;
1102         }
1103 
1104         if (rll->rll_writer == NULL && rll->rll_readers == 0)
1105                 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1106 
1107         VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1108 }
1109 
1110 static void
1111 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1112 {
1113         rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1114 
1115         ztest_rll_lock(rll, type);
1116 }
1117 
1118 static void
1119 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1120 {
1121         rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1122 
1123         ztest_rll_unlock(rll);
1124 }
1125 
1126 static rl_t *
1127 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1128     uint64_t size, rl_type_t type)
1129 {
1130         uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1131         rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1132         rl_t *rl;
1133 
1134         rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1135         rl->rl_object = object;
1136         rl->rl_offset = offset;
1137         rl->rl_size = size;
1138         rl->rl_lock = rll;
1139 
1140         ztest_rll_lock(rll, type);
1141 
1142         return (rl);
1143 }
1144 
1145 static void
1146 ztest_range_unlock(rl_t *rl)
1147 {
1148         rll_t *rll = rl->rl_lock;
1149 
1150         ztest_rll_unlock(rll);
1151 
1152         umem_free(rl, sizeof (*rl));
1153 }
1154 
1155 static void
1156 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1157 {
1158         zd->zd_os = os;
1159         zd->zd_zilog = dmu_objset_zil(os);
1160         zd->zd_shared = szd;
1161         dmu_objset_name(os, zd->zd_name);
1162 
1163         if (zd->zd_shared != NULL)
1164                 zd->zd_shared->zd_seq = 0;
1165 
1166         VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1167         VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1168 
1169         for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1170                 ztest_rll_init(&zd->zd_object_lock[l]);
1171 
1172         for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1173                 ztest_rll_init(&zd->zd_range_lock[l]);
1174 }
1175 
1176 static void
1177 ztest_zd_fini(ztest_ds_t *zd)
1178 {
1179         VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1180 
1181         for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1182                 ztest_rll_destroy(&zd->zd_object_lock[l]);
1183 
1184         for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1185                 ztest_rll_destroy(&zd->zd_range_lock[l]);
1186 }
1187 
1188 #define TXG_MIGHTWAIT   (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1189 
1190 static uint64_t
1191 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1192 {
1193         uint64_t txg;
1194         int error;
1195 
1196         /*
1197          * Attempt to assign tx to some transaction group.
1198          */
1199         error = dmu_tx_assign(tx, txg_how);
1200         if (error) {
1201                 if (error == ERESTART) {
1202                         ASSERT(txg_how == TXG_NOWAIT);
1203                         dmu_tx_wait(tx);
1204                 } else {
1205                         ASSERT3U(error, ==, ENOSPC);
1206                         ztest_record_enospc(tag);
1207                 }
1208                 dmu_tx_abort(tx);
1209                 return (0);
1210         }
1211         txg = dmu_tx_get_txg(tx);
1212         ASSERT(txg != 0);
1213         return (txg);
1214 }
1215 
1216 static void
1217 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1218 {
1219         uint64_t *ip = buf;
1220         uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1221 
1222         while (ip < ip_end)
1223                 *ip++ = value;
1224 }
1225 
1226 static boolean_t
1227 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1228 {
1229         uint64_t *ip = buf;
1230         uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1231         uint64_t diff = 0;
1232 
1233         while (ip < ip_end)
1234                 diff |= (value - *ip++);
1235 
1236         return (diff == 0);
1237 }
1238 
1239 static void
1240 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1241     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1242 {
1243         bt->bt_magic = BT_MAGIC;
1244         bt->bt_objset = dmu_objset_id(os);
1245         bt->bt_object = object;
1246         bt->bt_offset = offset;
1247         bt->bt_gen = gen;
1248         bt->bt_txg = txg;
1249         bt->bt_crtxg = crtxg;
1250 }
1251 
1252 static void
1253 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1254     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1255 {
1256         ASSERT(bt->bt_magic == BT_MAGIC);
1257         ASSERT(bt->bt_objset == dmu_objset_id(os));
1258         ASSERT(bt->bt_object == object);
1259         ASSERT(bt->bt_offset == offset);
1260         ASSERT(bt->bt_gen <= gen);
1261         ASSERT(bt->bt_txg <= txg);
1262         ASSERT(bt->bt_crtxg == crtxg);
1263 }
1264 
1265 static ztest_block_tag_t *
1266 ztest_bt_bonus(dmu_buf_t *db)
1267 {
1268         dmu_object_info_t doi;
1269         ztest_block_tag_t *bt;
1270 
1271         dmu_object_info_from_db(db, &doi);
1272         ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1273         ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1274         bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1275 
1276         return (bt);
1277 }
1278 
1279 /*
1280  * ZIL logging ops
1281  */
1282 
1283 #define lrz_type        lr_mode
1284 #define lrz_blocksize   lr_uid
1285 #define lrz_ibshift     lr_gid
1286 #define lrz_bonustype   lr_rdev
1287 #define lrz_bonuslen    lr_crtime[1]
1288 
1289 static void
1290 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1291 {
1292         char *name = (void *)(lr + 1);          /* name follows lr */
1293         size_t namesize = strlen(name) + 1;
1294         itx_t *itx;
1295 
1296         if (zil_replaying(zd->zd_zilog, tx))
1297                 return;
1298 
1299         itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1300         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1301             sizeof (*lr) + namesize - sizeof (lr_t));
1302 
1303         zil_itx_assign(zd->zd_zilog, itx, tx);
1304 }
1305 
1306 static void
1307 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1308 {
1309         char *name = (void *)(lr + 1);          /* name follows lr */
1310         size_t namesize = strlen(name) + 1;
1311         itx_t *itx;
1312 
1313         if (zil_replaying(zd->zd_zilog, tx))
1314                 return;
1315 
1316         itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1317         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1318             sizeof (*lr) + namesize - sizeof (lr_t));
1319 
1320         itx->itx_oid = object;
1321         zil_itx_assign(zd->zd_zilog, itx, tx);
1322 }
1323 
1324 static void
1325 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1326 {
1327         itx_t *itx;
1328         itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1329 
1330         if (zil_replaying(zd->zd_zilog, tx))
1331                 return;
1332 
1333         if (lr->lr_length > ZIL_MAX_LOG_DATA)
1334                 write_state = WR_INDIRECT;
1335 
1336         itx = zil_itx_create(TX_WRITE,
1337             sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1338 
1339         if (write_state == WR_COPIED &&
1340             dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1341             ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1342                 zil_itx_destroy(itx);
1343                 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1344                 write_state = WR_NEED_COPY;
1345         }
1346         itx->itx_private = zd;
1347         itx->itx_wr_state = write_state;
1348         itx->itx_sync = (ztest_random(8) == 0);
1349         itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1350 
1351         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1352             sizeof (*lr) - sizeof (lr_t));
1353 
1354         zil_itx_assign(zd->zd_zilog, itx, tx);
1355 }
1356 
1357 static void
1358 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1359 {
1360         itx_t *itx;
1361 
1362         if (zil_replaying(zd->zd_zilog, tx))
1363                 return;
1364 
1365         itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1366         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1367             sizeof (*lr) - sizeof (lr_t));
1368 
1369         itx->itx_sync = B_FALSE;
1370         zil_itx_assign(zd->zd_zilog, itx, tx);
1371 }
1372 
1373 static void
1374 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1375 {
1376         itx_t *itx;
1377 
1378         if (zil_replaying(zd->zd_zilog, tx))
1379                 return;
1380 
1381         itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1382         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1383             sizeof (*lr) - sizeof (lr_t));
1384 
1385         itx->itx_sync = B_FALSE;
1386         zil_itx_assign(zd->zd_zilog, itx, tx);
1387 }
1388 
1389 /*
1390  * ZIL replay ops
1391  */
1392 static int
1393 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1394 {
1395         char *name = (void *)(lr + 1);          /* name follows lr */
1396         objset_t *os = zd->zd_os;
1397         ztest_block_tag_t *bbt;
1398         dmu_buf_t *db;
1399         dmu_tx_t *tx;
1400         uint64_t txg;
1401         int error = 0;
1402 
1403         if (byteswap)
1404                 byteswap_uint64_array(lr, sizeof (*lr));
1405 
1406         ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1407         ASSERT(name[0] != '\0');
1408 
1409         tx = dmu_tx_create(os);
1410 
1411         dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1412 
1413         if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1414                 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1415         } else {
1416                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1417         }
1418 
1419         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1420         if (txg == 0)
1421                 return (ENOSPC);
1422 
1423         ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1424 
1425         if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1426                 if (lr->lr_foid == 0) {
1427                         lr->lr_foid = zap_create(os,
1428                             lr->lrz_type, lr->lrz_bonustype,
1429                             lr->lrz_bonuslen, tx);
1430                 } else {
1431                         error = zap_create_claim(os, lr->lr_foid,
1432                             lr->lrz_type, lr->lrz_bonustype,
1433                             lr->lrz_bonuslen, tx);
1434                 }
1435         } else {
1436                 if (lr->lr_foid == 0) {
1437                         lr->lr_foid = dmu_object_alloc(os,
1438                             lr->lrz_type, 0, lr->lrz_bonustype,
1439                             lr->lrz_bonuslen, tx);
1440                 } else {
1441                         error = dmu_object_claim(os, lr->lr_foid,
1442                             lr->lrz_type, 0, lr->lrz_bonustype,
1443                             lr->lrz_bonuslen, tx);
1444                 }
1445         }
1446 
1447         if (error) {
1448                 ASSERT3U(error, ==, EEXIST);
1449                 ASSERT(zd->zd_zilog->zl_replay);
1450                 dmu_tx_commit(tx);
1451                 return (error);
1452         }
1453 
1454         ASSERT(lr->lr_foid != 0);
1455 
1456         if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1457                 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1458                     lr->lrz_blocksize, lr->lrz_ibshift, tx));
1459 
1460         VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1461         bbt = ztest_bt_bonus(db);
1462         dmu_buf_will_dirty(db, tx);
1463         ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1464         dmu_buf_rele(db, FTAG);
1465 
1466         VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1467             &lr->lr_foid, tx));
1468 
1469         (void) ztest_log_create(zd, tx, lr);
1470 
1471         dmu_tx_commit(tx);
1472 
1473         return (0);
1474 }
1475 
1476 static int
1477 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1478 {
1479         char *name = (void *)(lr + 1);          /* name follows lr */
1480         objset_t *os = zd->zd_os;
1481         dmu_object_info_t doi;
1482         dmu_tx_t *tx;
1483         uint64_t object, txg;
1484 
1485         if (byteswap)
1486                 byteswap_uint64_array(lr, sizeof (*lr));
1487 
1488         ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1489         ASSERT(name[0] != '\0');
1490 
1491         VERIFY3U(0, ==,
1492             zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1493         ASSERT(object != 0);
1494 
1495         ztest_object_lock(zd, object, RL_WRITER);
1496 
1497         VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1498 
1499         tx = dmu_tx_create(os);
1500 
1501         dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1502         dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1503 
1504         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1505         if (txg == 0) {
1506                 ztest_object_unlock(zd, object);
1507                 return (ENOSPC);
1508         }
1509 
1510         if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1511                 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1512         } else {
1513                 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1514         }
1515 
1516         VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1517 
1518         (void) ztest_log_remove(zd, tx, lr, object);
1519 
1520         dmu_tx_commit(tx);
1521 
1522         ztest_object_unlock(zd, object);
1523 
1524         return (0);
1525 }
1526 
1527 static int
1528 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1529 {
1530         objset_t *os = zd->zd_os;
1531         void *data = lr + 1;                    /* data follows lr */
1532         uint64_t offset, length;
1533         ztest_block_tag_t *bt = data;
1534         ztest_block_tag_t *bbt;
1535         uint64_t gen, txg, lrtxg, crtxg;
1536         dmu_object_info_t doi;
1537         dmu_tx_t *tx;
1538         dmu_buf_t *db;
1539         arc_buf_t *abuf = NULL;
1540         rl_t *rl;
1541 
1542         if (byteswap)
1543                 byteswap_uint64_array(lr, sizeof (*lr));
1544 
1545         offset = lr->lr_offset;
1546         length = lr->lr_length;
1547 
1548         /* If it's a dmu_sync() block, write the whole block */
1549         if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1550                 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1551                 if (length < blocksize) {
1552                         offset -= offset % blocksize;
1553                         length = blocksize;
1554                 }
1555         }
1556 
1557         if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1558                 byteswap_uint64_array(bt, sizeof (*bt));
1559 
1560         if (bt->bt_magic != BT_MAGIC)
1561                 bt = NULL;
1562 
1563         ztest_object_lock(zd, lr->lr_foid, RL_READER);
1564         rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1565 
1566         VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1567 
1568         dmu_object_info_from_db(db, &doi);
1569 
1570         bbt = ztest_bt_bonus(db);
1571         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1572         gen = bbt->bt_gen;
1573         crtxg = bbt->bt_crtxg;
1574         lrtxg = lr->lr_common.lrc_txg;
1575 
1576         tx = dmu_tx_create(os);
1577 
1578         dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1579 
1580         if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1581             P2PHASE(offset, length) == 0)
1582                 abuf = dmu_request_arcbuf(db, length);
1583 
1584         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1585         if (txg == 0) {
1586                 if (abuf != NULL)
1587                         dmu_return_arcbuf(abuf);
1588                 dmu_buf_rele(db, FTAG);
1589                 ztest_range_unlock(rl);
1590                 ztest_object_unlock(zd, lr->lr_foid);
1591                 return (ENOSPC);
1592         }
1593 
1594         if (bt != NULL) {
1595                 /*
1596                  * Usually, verify the old data before writing new data --
1597                  * but not always, because we also want to verify correct
1598                  * behavior when the data was not recently read into cache.
1599                  */
1600                 ASSERT(offset % doi.doi_data_block_size == 0);
1601                 if (ztest_random(4) != 0) {
1602                         int prefetch = ztest_random(2) ?
1603                             DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1604                         ztest_block_tag_t rbt;
1605 
1606                         VERIFY(dmu_read(os, lr->lr_foid, offset,
1607                             sizeof (rbt), &rbt, prefetch) == 0);
1608                         if (rbt.bt_magic == BT_MAGIC) {
1609                                 ztest_bt_verify(&rbt, os, lr->lr_foid,
1610                                     offset, gen, txg, crtxg);
1611                         }
1612                 }
1613 
1614                 /*
1615                  * Writes can appear to be newer than the bonus buffer because
1616                  * the ztest_get_data() callback does a dmu_read() of the
1617                  * open-context data, which may be different than the data
1618                  * as it was when the write was generated.
1619                  */
1620                 if (zd->zd_zilog->zl_replay) {
1621                         ztest_bt_verify(bt, os, lr->lr_foid, offset,
1622                             MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1623                             bt->bt_crtxg);
1624                 }
1625 
1626                 /*
1627                  * Set the bt's gen/txg to the bonus buffer's gen/txg
1628                  * so that all of the usual ASSERTs will work.
1629                  */
1630                 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1631         }
1632 
1633         if (abuf == NULL) {
1634                 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1635         } else {
1636                 bcopy(data, abuf->b_data, length);
1637                 dmu_assign_arcbuf(db, offset, abuf, tx);
1638         }
1639 
1640         (void) ztest_log_write(zd, tx, lr);
1641 
1642         dmu_buf_rele(db, FTAG);
1643 
1644         dmu_tx_commit(tx);
1645 
1646         ztest_range_unlock(rl);
1647         ztest_object_unlock(zd, lr->lr_foid);
1648 
1649         return (0);
1650 }
1651 
1652 static int
1653 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1654 {
1655         objset_t *os = zd->zd_os;
1656         dmu_tx_t *tx;
1657         uint64_t txg;
1658         rl_t *rl;
1659 
1660         if (byteswap)
1661                 byteswap_uint64_array(lr, sizeof (*lr));
1662 
1663         ztest_object_lock(zd, lr->lr_foid, RL_READER);
1664         rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1665             RL_WRITER);
1666 
1667         tx = dmu_tx_create(os);
1668 
1669         dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1670 
1671         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1672         if (txg == 0) {
1673                 ztest_range_unlock(rl);
1674                 ztest_object_unlock(zd, lr->lr_foid);
1675                 return (ENOSPC);
1676         }
1677 
1678         VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1679             lr->lr_length, tx) == 0);
1680 
1681         (void) ztest_log_truncate(zd, tx, lr);
1682 
1683         dmu_tx_commit(tx);
1684 
1685         ztest_range_unlock(rl);
1686         ztest_object_unlock(zd, lr->lr_foid);
1687 
1688         return (0);
1689 }
1690 
1691 static int
1692 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1693 {
1694         objset_t *os = zd->zd_os;
1695         dmu_tx_t *tx;
1696         dmu_buf_t *db;
1697         ztest_block_tag_t *bbt;
1698         uint64_t txg, lrtxg, crtxg;
1699 
1700         if (byteswap)
1701                 byteswap_uint64_array(lr, sizeof (*lr));
1702 
1703         ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1704 
1705         VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1706 
1707         tx = dmu_tx_create(os);
1708         dmu_tx_hold_bonus(tx, lr->lr_foid);
1709 
1710         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1711         if (txg == 0) {
1712                 dmu_buf_rele(db, FTAG);
1713                 ztest_object_unlock(zd, lr->lr_foid);
1714                 return (ENOSPC);
1715         }
1716 
1717         bbt = ztest_bt_bonus(db);
1718         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1719         crtxg = bbt->bt_crtxg;
1720         lrtxg = lr->lr_common.lrc_txg;
1721 
1722         if (zd->zd_zilog->zl_replay) {
1723                 ASSERT(lr->lr_size != 0);
1724                 ASSERT(lr->lr_mode != 0);
1725                 ASSERT(lrtxg != 0);
1726         } else {
1727                 /*
1728                  * Randomly change the size and increment the generation.
1729                  */
1730                 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1731                     sizeof (*bbt);
1732                 lr->lr_mode = bbt->bt_gen + 1;
1733                 ASSERT(lrtxg == 0);
1734         }
1735 
1736         /*
1737          * Verify that the current bonus buffer is not newer than our txg.
1738          */
1739         ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1740             MAX(txg, lrtxg), crtxg);
1741 
1742         dmu_buf_will_dirty(db, tx);
1743 
1744         ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1745         ASSERT3U(lr->lr_size, <=, db->db_size);
1746         VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1747         bbt = ztest_bt_bonus(db);
1748 
1749         ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1750 
1751         dmu_buf_rele(db, FTAG);
1752 
1753         (void) ztest_log_setattr(zd, tx, lr);
1754 
1755         dmu_tx_commit(tx);
1756 
1757         ztest_object_unlock(zd, lr->lr_foid);
1758 
1759         return (0);
1760 }
1761 
1762 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1763         NULL,                   /* 0 no such transaction type */
1764         ztest_replay_create,    /* TX_CREATE */
1765         NULL,                   /* TX_MKDIR */
1766         NULL,                   /* TX_MKXATTR */
1767         NULL,                   /* TX_SYMLINK */
1768         ztest_replay_remove,    /* TX_REMOVE */
1769         NULL,                   /* TX_RMDIR */
1770         NULL,                   /* TX_LINK */
1771         NULL,                   /* TX_RENAME */
1772         ztest_replay_write,     /* TX_WRITE */
1773         ztest_replay_truncate,  /* TX_TRUNCATE */
1774         ztest_replay_setattr,   /* TX_SETATTR */
1775         NULL,                   /* TX_ACL */
1776         NULL,                   /* TX_CREATE_ACL */
1777         NULL,                   /* TX_CREATE_ATTR */
1778         NULL,                   /* TX_CREATE_ACL_ATTR */
1779         NULL,                   /* TX_MKDIR_ACL */
1780         NULL,                   /* TX_MKDIR_ATTR */
1781         NULL,                   /* TX_MKDIR_ACL_ATTR */
1782         NULL,                   /* TX_WRITE2 */
1783 };
1784 
1785 /*
1786  * ZIL get_data callbacks
1787  */
1788 
1789 static void
1790 ztest_get_done(zgd_t *zgd, int error)
1791 {
1792         ztest_ds_t *zd = zgd->zgd_private;
1793         uint64_t object = zgd->zgd_rl->rl_object;
1794 
1795         if (zgd->zgd_db)
1796                 dmu_buf_rele(zgd->zgd_db, zgd);
1797 
1798         ztest_range_unlock(zgd->zgd_rl);
1799         ztest_object_unlock(zd, object);
1800 
1801         if (error == 0 && zgd->zgd_bp)
1802                 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1803 
1804         umem_free(zgd, sizeof (*zgd));
1805 }
1806 
1807 static int
1808 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1809 {
1810         ztest_ds_t *zd = arg;
1811         objset_t *os = zd->zd_os;
1812         uint64_t object = lr->lr_foid;
1813         uint64_t offset = lr->lr_offset;
1814         uint64_t size = lr->lr_length;
1815         blkptr_t *bp = &lr->lr_blkptr;
1816         uint64_t txg = lr->lr_common.lrc_txg;
1817         uint64_t crtxg;
1818         dmu_object_info_t doi;
1819         dmu_buf_t *db;
1820         zgd_t *zgd;
1821         int error;
1822 
1823         ztest_object_lock(zd, object, RL_READER);
1824         error = dmu_bonus_hold(os, object, FTAG, &db);
1825         if (error) {
1826                 ztest_object_unlock(zd, object);
1827                 return (error);
1828         }
1829 
1830         crtxg = ztest_bt_bonus(db)->bt_crtxg;
1831 
1832         if (crtxg == 0 || crtxg > txg) {
1833                 dmu_buf_rele(db, FTAG);
1834                 ztest_object_unlock(zd, object);
1835                 return (ENOENT);
1836         }
1837 
1838         dmu_object_info_from_db(db, &doi);
1839         dmu_buf_rele(db, FTAG);
1840         db = NULL;
1841 
1842         zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1843         zgd->zgd_zilog = zd->zd_zilog;
1844         zgd->zgd_private = zd;
1845 
1846         if (buf != NULL) {      /* immediate write */
1847                 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1848                     RL_READER);
1849 
1850                 error = dmu_read(os, object, offset, size, buf,
1851                     DMU_READ_NO_PREFETCH);
1852                 ASSERT(error == 0);
1853         } else {
1854                 size = doi.doi_data_block_size;
1855                 if (ISP2(size)) {
1856                         offset = P2ALIGN(offset, size);
1857                 } else {
1858                         ASSERT(offset < size);
1859                         offset = 0;
1860                 }
1861 
1862                 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1863                     RL_READER);
1864 
1865                 error = dmu_buf_hold(os, object, offset, zgd, &db,
1866                     DMU_READ_NO_PREFETCH);
1867 
1868                 if (error == 0) {
1869                         blkptr_t *obp = dmu_buf_get_blkptr(db);
1870                         if (obp) {
1871                                 ASSERT(BP_IS_HOLE(bp));
1872                                 *bp = *obp;
1873                         }
1874 
1875                         zgd->zgd_db = db;
1876                         zgd->zgd_bp = bp;
1877 
1878                         ASSERT(db->db_offset == offset);
1879                         ASSERT(db->db_size == size);
1880 
1881                         error = dmu_sync(zio, lr->lr_common.lrc_txg,
1882                             ztest_get_done, zgd);
1883 
1884                         if (error == 0)
1885                                 return (0);
1886                 }
1887         }
1888 
1889         ztest_get_done(zgd, error);
1890 
1891         return (error);
1892 }
1893 
1894 static void *
1895 ztest_lr_alloc(size_t lrsize, char *name)
1896 {
1897         char *lr;
1898         size_t namesize = name ? strlen(name) + 1 : 0;
1899 
1900         lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1901 
1902         if (name)
1903                 bcopy(name, lr + lrsize, namesize);
1904 
1905         return (lr);
1906 }
1907 
1908 void
1909 ztest_lr_free(void *lr, size_t lrsize, char *name)
1910 {
1911         size_t namesize = name ? strlen(name) + 1 : 0;
1912 
1913         umem_free(lr, lrsize + namesize);
1914 }
1915 
1916 /*
1917  * Lookup a bunch of objects.  Returns the number of objects not found.
1918  */
1919 static int
1920 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1921 {
1922         int missing = 0;
1923         int error;
1924 
1925         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1926 
1927         for (int i = 0; i < count; i++, od++) {
1928                 od->od_object = 0;
1929                 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1930                     sizeof (uint64_t), 1, &od->od_object);
1931                 if (error) {
1932                         ASSERT(error == ENOENT);
1933                         ASSERT(od->od_object == 0);
1934                         missing++;
1935                 } else {
1936                         dmu_buf_t *db;
1937                         ztest_block_tag_t *bbt;
1938                         dmu_object_info_t doi;
1939 
1940                         ASSERT(od->od_object != 0);
1941                         ASSERT(missing == 0);   /* there should be no gaps */
1942 
1943                         ztest_object_lock(zd, od->od_object, RL_READER);
1944                         VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1945                             od->od_object, FTAG, &db));
1946                         dmu_object_info_from_db(db, &doi);
1947                         bbt = ztest_bt_bonus(db);
1948                         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1949                         od->od_type = doi.doi_type;
1950                         od->od_blocksize = doi.doi_data_block_size;
1951                         od->od_gen = bbt->bt_gen;
1952                         dmu_buf_rele(db, FTAG);
1953                         ztest_object_unlock(zd, od->od_object);
1954                 }
1955         }
1956 
1957         return (missing);
1958 }
1959 
1960 static int
1961 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1962 {
1963         int missing = 0;
1964 
1965         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1966 
1967         for (int i = 0; i < count; i++, od++) {
1968                 if (missing) {
1969                         od->od_object = 0;
1970                         missing++;
1971                         continue;
1972                 }
1973 
1974                 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1975 
1976                 lr->lr_doid = od->od_dir;
1977                 lr->lr_foid = 0;     /* 0 to allocate, > 0 to claim */
1978                 lr->lrz_type = od->od_crtype;
1979                 lr->lrz_blocksize = od->od_crblocksize;
1980                 lr->lrz_ibshift = ztest_random_ibshift();
1981                 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1982                 lr->lrz_bonuslen = dmu_bonus_max();
1983                 lr->lr_gen = od->od_crgen;
1984                 lr->lr_crtime[0] = time(NULL);
1985 
1986                 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1987                         ASSERT(missing == 0);
1988                         od->od_object = 0;
1989                         missing++;
1990                 } else {
1991                         od->od_object = lr->lr_foid;
1992                         od->od_type = od->od_crtype;
1993                         od->od_blocksize = od->od_crblocksize;
1994                         od->od_gen = od->od_crgen;
1995                         ASSERT(od->od_object != 0);
1996                 }
1997 
1998                 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1999         }
2000 
2001         return (missing);
2002 }
2003 
2004 static int
2005 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2006 {
2007         int missing = 0;
2008         int error;
2009 
2010         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2011 
2012         od += count - 1;
2013 
2014         for (int i = count - 1; i >= 0; i--, od--) {
2015                 if (missing) {
2016                         missing++;
2017                         continue;
2018                 }
2019 
2020                 /*
2021                  * No object was found.
2022                  */
2023                 if (od->od_object == 0)
2024                         continue;
2025 
2026                 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2027 
2028                 lr->lr_doid = od->od_dir;
2029 
2030                 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2031                         ASSERT3U(error, ==, ENOSPC);
2032                         missing++;
2033                 } else {
2034                         od->od_object = 0;
2035                 }
2036                 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2037         }
2038 
2039         return (missing);
2040 }
2041 
2042 static int
2043 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2044     void *data)
2045 {
2046         lr_write_t *lr;
2047         int error;
2048 
2049         lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2050 
2051         lr->lr_foid = object;
2052         lr->lr_offset = offset;
2053         lr->lr_length = size;
2054         lr->lr_blkoff = 0;
2055         BP_ZERO(&lr->lr_blkptr);
2056 
2057         bcopy(data, lr + 1, size);
2058 
2059         error = ztest_replay_write(zd, lr, B_FALSE);
2060 
2061         ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2062 
2063         return (error);
2064 }
2065 
2066 static int
2067 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2068 {
2069         lr_truncate_t *lr;
2070         int error;
2071 
2072         lr = ztest_lr_alloc(sizeof (*lr), NULL);
2073 
2074         lr->lr_foid = object;
2075         lr->lr_offset = offset;
2076         lr->lr_length = size;
2077 
2078         error = ztest_replay_truncate(zd, lr, B_FALSE);
2079 
2080         ztest_lr_free(lr, sizeof (*lr), NULL);
2081 
2082         return (error);
2083 }
2084 
2085 static int
2086 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2087 {
2088         lr_setattr_t *lr;
2089         int error;
2090 
2091         lr = ztest_lr_alloc(sizeof (*lr), NULL);
2092 
2093         lr->lr_foid = object;
2094         lr->lr_size = 0;
2095         lr->lr_mode = 0;
2096 
2097         error = ztest_replay_setattr(zd, lr, B_FALSE);
2098 
2099         ztest_lr_free(lr, sizeof (*lr), NULL);
2100 
2101         return (error);
2102 }
2103 
2104 static void
2105 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2106 {
2107         objset_t *os = zd->zd_os;
2108         dmu_tx_t *tx;
2109         uint64_t txg;
2110         rl_t *rl;
2111 
2112         txg_wait_synced(dmu_objset_pool(os), 0);
2113 
2114         ztest_object_lock(zd, object, RL_READER);
2115         rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2116 
2117         tx = dmu_tx_create(os);
2118 
2119         dmu_tx_hold_write(tx, object, offset, size);
2120 
2121         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2122 
2123         if (txg != 0) {
2124                 dmu_prealloc(os, object, offset, size, tx);
2125                 dmu_tx_commit(tx);
2126                 txg_wait_synced(dmu_objset_pool(os), txg);
2127         } else {
2128                 (void) dmu_free_long_range(os, object, offset, size);
2129         }
2130 
2131         ztest_range_unlock(rl);
2132         ztest_object_unlock(zd, object);
2133 }
2134 
2135 static void
2136 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2137 {
2138         int err;
2139         ztest_block_tag_t wbt;
2140         dmu_object_info_t doi;
2141         enum ztest_io_type io_type;
2142         uint64_t blocksize;
2143         void *data;
2144 
2145         VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2146         blocksize = doi.doi_data_block_size;
2147         data = umem_alloc(blocksize, UMEM_NOFAIL);
2148 
2149         /*
2150          * Pick an i/o type at random, biased toward writing block tags.
2151          */
2152         io_type = ztest_random(ZTEST_IO_TYPES);
2153         if (ztest_random(2) == 0)
2154                 io_type = ZTEST_IO_WRITE_TAG;
2155 
2156         (void) rw_rdlock(&zd->zd_zilog_lock);
2157 
2158         switch (io_type) {
2159 
2160         case ZTEST_IO_WRITE_TAG:
2161                 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2162                 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2163                 break;
2164 
2165         case ZTEST_IO_WRITE_PATTERN:
2166                 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2167                 if (ztest_random(2) == 0) {
2168                         /*
2169                          * Induce fletcher2 collisions to ensure that
2170                          * zio_ddt_collision() detects and resolves them
2171                          * when using fletcher2-verify for deduplication.
2172                          */
2173                         ((uint64_t *)data)[0] ^= 1ULL << 63;
2174                         ((uint64_t *)data)[4] ^= 1ULL << 63;
2175                 }
2176                 (void) ztest_write(zd, object, offset, blocksize, data);
2177                 break;
2178 
2179         case ZTEST_IO_WRITE_ZEROES:
2180                 bzero(data, blocksize);
2181                 (void) ztest_write(zd, object, offset, blocksize, data);
2182                 break;
2183 
2184         case ZTEST_IO_TRUNCATE:
2185                 (void) ztest_truncate(zd, object, offset, blocksize);
2186                 break;
2187 
2188         case ZTEST_IO_SETATTR:
2189                 (void) ztest_setattr(zd, object);
2190                 break;
2191 
2192         case ZTEST_IO_REWRITE:
2193                 (void) rw_rdlock(&ztest_name_lock);
2194                 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2195                     ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2196                     B_FALSE);
2197                 VERIFY(err == 0 || err == ENOSPC);
2198                 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2199                     ZFS_PROP_COMPRESSION,
2200                     ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2201                     B_FALSE);
2202                 VERIFY(err == 0 || err == ENOSPC);
2203                 (void) rw_unlock(&ztest_name_lock);
2204 
2205                 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2206                     DMU_READ_NO_PREFETCH));
2207 
2208                 (void) ztest_write(zd, object, offset, blocksize, data);
2209                 break;
2210         }
2211 
2212         (void) rw_unlock(&zd->zd_zilog_lock);
2213 
2214         umem_free(data, blocksize);
2215 }
2216 
2217 /*
2218  * Initialize an object description template.
2219  */
2220 static void
2221 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2222     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2223 {
2224         od->od_dir = ZTEST_DIROBJ;
2225         od->od_object = 0;
2226 
2227         od->od_crtype = type;
2228         od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2229         od->od_crgen = gen;
2230 
2231         od->od_type = DMU_OT_NONE;
2232         od->od_blocksize = 0;
2233         od->od_gen = 0;
2234 
2235         (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2236             tag, (int64_t)id, index);
2237 }
2238 
2239 /*
2240  * Lookup or create the objects for a test using the od template.
2241  * If the objects do not all exist, or if 'remove' is specified,
2242  * remove any existing objects and create new ones.  Otherwise,
2243  * use the existing objects.
2244  */
2245 static int
2246 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2247 {
2248         int count = size / sizeof (*od);
2249         int rv = 0;
2250 
2251         VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2252         if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2253             (ztest_remove(zd, od, count) != 0 ||
2254             ztest_create(zd, od, count) != 0))
2255                 rv = -1;
2256         zd->zd_od = od;
2257         VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2258 
2259         return (rv);
2260 }
2261 
2262 /* ARGSUSED */
2263 void
2264 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2265 {
2266         zilog_t *zilog = zd->zd_zilog;
2267 
2268         (void) rw_rdlock(&zd->zd_zilog_lock);
2269 
2270         zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2271 
2272         /*
2273          * Remember the committed values in zd, which is in parent/child
2274          * shared memory.  If we die, the next iteration of ztest_run()
2275          * will verify that the log really does contain this record.
2276          */
2277         mutex_enter(&zilog->zl_lock);
2278         ASSERT(zd->zd_shared != NULL);
2279         ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2280         zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2281         mutex_exit(&zilog->zl_lock);
2282 
2283         (void) rw_unlock(&zd->zd_zilog_lock);
2284 }
2285 
2286 /*
2287  * This function is designed to simulate the operations that occur during a
2288  * mount/unmount operation.  We hold the dataset across these operations in an
2289  * attempt to expose any implicit assumptions about ZIL management.
2290  */
2291 /* ARGSUSED */
2292 void
2293 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2294 {
2295         objset_t *os = zd->zd_os;
2296 
2297         /*
2298          * We grab the zd_dirobj_lock to ensure that no other thread is
2299          * updating the zil (i.e. adding in-memory log records) and the
2300          * zd_zilog_lock to block any I/O.
2301          */
2302         VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2303         (void) rw_wrlock(&zd->zd_zilog_lock);
2304 
2305         /* zfsvfs_teardown() */
2306         zil_close(zd->zd_zilog);
2307 
2308         /* zfsvfs_setup() */
2309         VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2310         zil_replay(os, zd, ztest_replay_vector);
2311 
2312         (void) rw_unlock(&zd->zd_zilog_lock);
2313         VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2314 }
2315 
2316 /*
2317  * Verify that we can't destroy an active pool, create an existing pool,
2318  * or create a pool with a bad vdev spec.
2319  */
2320 /* ARGSUSED */
2321 void
2322 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2323 {
2324         ztest_shared_opts_t *zo = &ztest_opts;
2325         spa_t *spa;
2326         nvlist_t *nvroot;
2327 
2328         /*
2329          * Attempt to create using a bad file.
2330          */
2331         nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2332         VERIFY3U(ENOENT, ==,
2333             spa_create("ztest_bad_file", nvroot, NULL, NULL));
2334         nvlist_free(nvroot);
2335 
2336         /*
2337          * Attempt to create using a bad mirror.
2338          */
2339         nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2340         VERIFY3U(ENOENT, ==,
2341             spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2342         nvlist_free(nvroot);
2343 
2344         /*
2345          * Attempt to create an existing pool.  It shouldn't matter
2346          * what's in the nvroot; we should fail with EEXIST.
2347          */
2348         (void) rw_rdlock(&ztest_name_lock);
2349         nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2350         VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2351         nvlist_free(nvroot);
2352         VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2353         VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2354         spa_close(spa, FTAG);
2355 
2356         (void) rw_unlock(&ztest_name_lock);
2357 }
2358 
2359 /* ARGSUSED */
2360 void
2361 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2362 {
2363         spa_t *spa;
2364         uint64_t initial_version = SPA_VERSION_INITIAL;
2365         uint64_t version, newversion;
2366         nvlist_t *nvroot, *props;
2367         char *name;
2368 
2369         VERIFY0(mutex_lock(&ztest_vdev_lock));
2370         name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2371 
2372         /*
2373          * Clean up from previous runs.
2374          */
2375         (void) spa_destroy(name);
2376 
2377         nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2378             0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2379 
2380         /*
2381          * If we're configuring a RAIDZ device then make sure that the
2382          * the initial version is capable of supporting that feature.
2383          */
2384         switch (ztest_opts.zo_raidz_parity) {
2385         case 0:
2386         case 1:
2387                 initial_version = SPA_VERSION_INITIAL;
2388                 break;
2389         case 2:
2390                 initial_version = SPA_VERSION_RAIDZ2;
2391                 break;
2392         case 3:
2393                 initial_version = SPA_VERSION_RAIDZ3;
2394                 break;
2395         }
2396 
2397         /*
2398          * Create a pool with a spa version that can be upgraded. Pick
2399          * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2400          */
2401         do {
2402                 version = ztest_random_spa_version(initial_version);
2403         } while (version > SPA_VERSION_BEFORE_FEATURES);
2404 
2405         props = fnvlist_alloc();
2406         fnvlist_add_uint64(props,
2407             zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2408         VERIFY0(spa_create(name, nvroot, props, NULL));
2409         fnvlist_free(nvroot);
2410         fnvlist_free(props);
2411 
2412         VERIFY0(spa_open(name, &spa, FTAG));
2413         VERIFY3U(spa_version(spa), ==, version);
2414         newversion = ztest_random_spa_version(version + 1);
2415 
2416         if (ztest_opts.zo_verbose >= 4) {
2417                 (void) printf("upgrading spa version from %llu to %llu\n",
2418                     (u_longlong_t)version, (u_longlong_t)newversion);
2419         }
2420 
2421         spa_upgrade(spa, newversion);
2422         VERIFY3U(spa_version(spa), >, version);
2423         VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2424             zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2425         spa_close(spa, FTAG);
2426 
2427         strfree(name);
2428         VERIFY0(mutex_unlock(&ztest_vdev_lock));
2429 }
2430 
2431 static vdev_t *
2432 vdev_lookup_by_path(vdev_t *vd, const char *path)
2433 {
2434         vdev_t *mvd;
2435 
2436         if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2437                 return (vd);
2438 
2439         for (int c = 0; c < vd->vdev_children; c++)
2440                 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2441                     NULL)
2442                         return (mvd);
2443 
2444         return (NULL);
2445 }
2446 
2447 /*
2448  * Find the first available hole which can be used as a top-level.
2449  */
2450 int
2451 find_vdev_hole(spa_t *spa)
2452 {
2453         vdev_t *rvd = spa->spa_root_vdev;
2454         int c;
2455 
2456         ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2457 
2458         for (c = 0; c < rvd->vdev_children; c++) {
2459                 vdev_t *cvd = rvd->vdev_child[c];
2460 
2461                 if (cvd->vdev_ishole)
2462                         break;
2463         }
2464         return (c);
2465 }
2466 
2467 /*
2468  * Verify that vdev_add() works as expected.
2469  */
2470 /* ARGSUSED */
2471 void
2472 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2473 {
2474         ztest_shared_t *zs = ztest_shared;
2475         spa_t *spa = ztest_spa;
2476         uint64_t leaves;
2477         uint64_t guid;
2478         nvlist_t *nvroot;
2479         int error;
2480 
2481         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2482         leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2483 
2484         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2485 
2486         ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2487 
2488         /*
2489          * If we have slogs then remove them 1/4 of the time.
2490          */
2491         if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2492                 /*
2493                  * Grab the guid from the head of the log class rotor.
2494                  */
2495                 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2496 
2497                 spa_config_exit(spa, SCL_VDEV, FTAG);
2498 
2499                 /*
2500                  * We have to grab the zs_name_lock as writer to
2501                  * prevent a race between removing a slog (dmu_objset_find)
2502                  * and destroying a dataset. Removing the slog will
2503                  * grab a reference on the dataset which may cause
2504                  * dmu_objset_destroy() to fail with EBUSY thus
2505                  * leaving the dataset in an inconsistent state.
2506                  */
2507                 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2508                 error = spa_vdev_remove(spa, guid, B_FALSE);
2509                 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2510 
2511                 if (error && error != EEXIST)
2512                         fatal(0, "spa_vdev_remove() = %d", error);
2513         } else {
2514                 spa_config_exit(spa, SCL_VDEV, FTAG);
2515 
2516                 /*
2517                  * Make 1/4 of the devices be log devices.
2518                  */
2519                 nvroot = make_vdev_root(NULL, NULL, NULL,
2520                     ztest_opts.zo_vdev_size, 0,
2521                     ztest_random(4) == 0, ztest_opts.zo_raidz,
2522                     zs->zs_mirrors, 1);
2523 
2524                 error = spa_vdev_add(spa, nvroot);
2525                 nvlist_free(nvroot);
2526 
2527                 if (error == ENOSPC)
2528                         ztest_record_enospc("spa_vdev_add");
2529                 else if (error != 0)
2530                         fatal(0, "spa_vdev_add() = %d", error);
2531         }
2532 
2533         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2534 }
2535 
2536 /*
2537  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2538  */
2539 /* ARGSUSED */
2540 void
2541 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2542 {
2543         ztest_shared_t *zs = ztest_shared;
2544         spa_t *spa = ztest_spa;
2545         vdev_t *rvd = spa->spa_root_vdev;
2546         spa_aux_vdev_t *sav;
2547         char *aux;
2548         uint64_t guid = 0;
2549         int error;
2550 
2551         if (ztest_random(2) == 0) {
2552                 sav = &spa->spa_spares;
2553                 aux = ZPOOL_CONFIG_SPARES;
2554         } else {
2555                 sav = &spa->spa_l2cache;
2556                 aux = ZPOOL_CONFIG_L2CACHE;
2557         }
2558 
2559         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2560 
2561         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2562 
2563         if (sav->sav_count != 0 && ztest_random(4) == 0) {
2564                 /*
2565                  * Pick a random device to remove.
2566                  */
2567                 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2568         } else {
2569                 /*
2570                  * Find an unused device we can add.
2571                  */
2572                 zs->zs_vdev_aux = 0;
2573                 for (;;) {
2574                         char path[MAXPATHLEN];
2575                         int c;
2576                         (void) snprintf(path, sizeof (path), ztest_aux_template,
2577                             ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2578                             zs->zs_vdev_aux);
2579                         for (c = 0; c < sav->sav_count; c++)
2580                                 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2581                                     path) == 0)
2582                                         break;
2583                         if (c == sav->sav_count &&
2584                             vdev_lookup_by_path(rvd, path) == NULL)
2585                                 break;
2586                         zs->zs_vdev_aux++;
2587                 }
2588         }
2589 
2590         spa_config_exit(spa, SCL_VDEV, FTAG);
2591 
2592         if (guid == 0) {
2593                 /*
2594                  * Add a new device.
2595                  */
2596                 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2597                     (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2598                 error = spa_vdev_add(spa, nvroot);
2599                 if (error != 0)
2600                         fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2601                 nvlist_free(nvroot);
2602         } else {
2603                 /*
2604                  * Remove an existing device.  Sometimes, dirty its
2605                  * vdev state first to make sure we handle removal
2606                  * of devices that have pending state changes.
2607                  */
2608                 if (ztest_random(2) == 0)
2609                         (void) vdev_online(spa, guid, 0, NULL);
2610 
2611                 error = spa_vdev_remove(spa, guid, B_FALSE);
2612                 if (error != 0 && error != EBUSY)
2613                         fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2614         }
2615 
2616         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2617 }
2618 
2619 /*
2620  * split a pool if it has mirror tlvdevs
2621  */
2622 /* ARGSUSED */
2623 void
2624 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2625 {
2626         ztest_shared_t *zs = ztest_shared;
2627         spa_t *spa = ztest_spa;
2628         vdev_t *rvd = spa->spa_root_vdev;
2629         nvlist_t *tree, **child, *config, *split, **schild;
2630         uint_t c, children, schildren = 0, lastlogid = 0;
2631         int error = 0;
2632 
2633         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2634 
2635         /* ensure we have a useable config; mirrors of raidz aren't supported */
2636         if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2637                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2638                 return;
2639         }
2640 
2641         /* clean up the old pool, if any */
2642         (void) spa_destroy("splitp");
2643 
2644         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2645 
2646         /* generate a config from the existing config */
2647         mutex_enter(&spa->spa_props_lock);
2648         VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2649             &tree) == 0);
2650         mutex_exit(&spa->spa_props_lock);
2651 
2652         VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2653             &children) == 0);
2654 
2655         schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2656         for (c = 0; c < children; c++) {
2657                 vdev_t *tvd = rvd->vdev_child[c];
2658                 nvlist_t **mchild;
2659                 uint_t mchildren;
2660 
2661                 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2662                         VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2663                             0) == 0);
2664                         VERIFY(nvlist_add_string(schild[schildren],
2665                             ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2666                         VERIFY(nvlist_add_uint64(schild[schildren],
2667                             ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2668                         if (lastlogid == 0)
2669                                 lastlogid = schildren;
2670                         ++schildren;
2671                         continue;
2672                 }
2673                 lastlogid = 0;
2674                 VERIFY(nvlist_lookup_nvlist_array(child[c],
2675                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2676                 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2677         }
2678 
2679         /* OK, create a config that can be used to split */
2680         VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2681         VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2682             VDEV_TYPE_ROOT) == 0);
2683         VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2684             lastlogid != 0 ? lastlogid : schildren) == 0);
2685 
2686         VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2687         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2688 
2689         for (c = 0; c < schildren; c++)
2690                 nvlist_free(schild[c]);
2691         free(schild);
2692         nvlist_free(split);
2693 
2694         spa_config_exit(spa, SCL_VDEV, FTAG);
2695 
2696         (void) rw_wrlock(&ztest_name_lock);
2697         error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2698         (void) rw_unlock(&ztest_name_lock);
2699 
2700         nvlist_free(config);
2701 
2702         if (error == 0) {
2703                 (void) printf("successful split - results:\n");
2704                 mutex_enter(&spa_namespace_lock);
2705                 show_pool_stats(spa);
2706                 show_pool_stats(spa_lookup("splitp"));
2707                 mutex_exit(&spa_namespace_lock);
2708                 ++zs->zs_splits;
2709                 --zs->zs_mirrors;
2710         }
2711         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2712 
2713 }
2714 
2715 /*
2716  * Verify that we can attach and detach devices.
2717  */
2718 /* ARGSUSED */
2719 void
2720 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2721 {
2722         ztest_shared_t *zs = ztest_shared;
2723         spa_t *spa = ztest_spa;
2724         spa_aux_vdev_t *sav = &spa->spa_spares;
2725         vdev_t *rvd = spa->spa_root_vdev;
2726         vdev_t *oldvd, *newvd, *pvd;
2727         nvlist_t *root;
2728         uint64_t leaves;
2729         uint64_t leaf, top;
2730         uint64_t ashift = ztest_get_ashift();
2731         uint64_t oldguid, pguid;
2732         size_t oldsize, newsize;
2733         char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2734         int replacing;
2735         int oldvd_has_siblings = B_FALSE;
2736         int newvd_is_spare = B_FALSE;
2737         int oldvd_is_log;
2738         int error, expected_error;
2739 
2740         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2741         leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2742 
2743         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2744 
2745         /*
2746          * Decide whether to do an attach or a replace.
2747          */
2748         replacing = ztest_random(2);
2749 
2750         /*
2751          * Pick a random top-level vdev.
2752          */
2753         top = ztest_random_vdev_top(spa, B_TRUE);
2754 
2755         /*
2756          * Pick a random leaf within it.
2757          */
2758         leaf = ztest_random(leaves);
2759 
2760         /*
2761          * Locate this vdev.
2762          */
2763         oldvd = rvd->vdev_child[top];
2764         if (zs->zs_mirrors >= 1) {
2765                 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2766                 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2767                 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2768         }
2769         if (ztest_opts.zo_raidz > 1) {
2770                 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2771                 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2772                 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2773         }
2774 
2775         /*
2776          * If we're already doing an attach or replace, oldvd may be a
2777          * mirror vdev -- in which case, pick a random child.
2778          */
2779         while (oldvd->vdev_children != 0) {
2780                 oldvd_has_siblings = B_TRUE;
2781                 ASSERT(oldvd->vdev_children >= 2);
2782                 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2783         }
2784 
2785         oldguid = oldvd->vdev_guid;
2786         oldsize = vdev_get_min_asize(oldvd);
2787         oldvd_is_log = oldvd->vdev_top->vdev_islog;
2788         (void) strcpy(oldpath, oldvd->vdev_path);
2789         pvd = oldvd->vdev_parent;
2790         pguid = pvd->vdev_guid;
2791 
2792         /*
2793          * If oldvd has siblings, then half of the time, detach it.
2794          */
2795         if (oldvd_has_siblings && ztest_random(2) == 0) {
2796                 spa_config_exit(spa, SCL_VDEV, FTAG);
2797                 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2798                 if (error != 0 && error != ENODEV && error != EBUSY &&
2799                     error != ENOTSUP)
2800                         fatal(0, "detach (%s) returned %d", oldpath, error);
2801                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2802                 return;
2803         }
2804 
2805         /*
2806          * For the new vdev, choose with equal probability between the two
2807          * standard paths (ending in either 'a' or 'b') or a random hot spare.
2808          */
2809         if (sav->sav_count != 0 && ztest_random(3) == 0) {
2810                 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2811                 newvd_is_spare = B_TRUE;
2812                 (void) strcpy(newpath, newvd->vdev_path);
2813         } else {
2814                 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2815                     ztest_opts.zo_dir, ztest_opts.zo_pool,
2816                     top * leaves + leaf);
2817                 if (ztest_random(2) == 0)
2818                         newpath[strlen(newpath) - 1] = 'b';
2819                 newvd = vdev_lookup_by_path(rvd, newpath);
2820         }
2821 
2822         if (newvd) {
2823                 newsize = vdev_get_min_asize(newvd);
2824         } else {
2825                 /*
2826                  * Make newsize a little bigger or smaller than oldsize.
2827                  * If it's smaller, the attach should fail.
2828                  * If it's larger, and we're doing a replace,
2829                  * we should get dynamic LUN growth when we're done.
2830                  */
2831                 newsize = 10 * oldsize / (9 + ztest_random(3));
2832         }
2833 
2834         /*
2835          * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2836          * unless it's a replace; in that case any non-replacing parent is OK.
2837          *
2838          * If newvd is already part of the pool, it should fail with EBUSY.
2839          *
2840          * If newvd is too small, it should fail with EOVERFLOW.
2841          */
2842         if (pvd->vdev_ops != &vdev_mirror_ops &&
2843             pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2844             pvd->vdev_ops == &vdev_replacing_ops ||
2845             pvd->vdev_ops == &vdev_spare_ops))
2846                 expected_error = ENOTSUP;
2847         else if (newvd_is_spare && (!replacing || oldvd_is_log))
2848                 expected_error = ENOTSUP;
2849         else if (newvd == oldvd)
2850                 expected_error = replacing ? 0 : EBUSY;
2851         else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2852                 expected_error = EBUSY;
2853         else if (newsize < oldsize)
2854                 expected_error = EOVERFLOW;
2855         else if (ashift > oldvd->vdev_top->vdev_ashift)
2856                 expected_error = EDOM;
2857         else
2858                 expected_error = 0;
2859 
2860         spa_config_exit(spa, SCL_VDEV, FTAG);
2861 
2862         /*
2863          * Build the nvlist describing newpath.
2864          */
2865         root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2866             ashift, 0, 0, 0, 1);
2867 
2868         error = spa_vdev_attach(spa, oldguid, root, replacing);
2869 
2870         nvlist_free(root);
2871 
2872         /*
2873          * If our parent was the replacing vdev, but the replace completed,
2874          * then instead of failing with ENOTSUP we may either succeed,
2875          * fail with ENODEV, or fail with EOVERFLOW.
2876          */
2877         if (expected_error == ENOTSUP &&
2878             (error == 0 || error == ENODEV || error == EOVERFLOW))
2879                 expected_error = error;
2880 
2881         /*
2882          * If someone grew the LUN, the replacement may be too small.
2883          */
2884         if (error == EOVERFLOW || error == EBUSY)
2885                 expected_error = error;
2886 
2887         /* XXX workaround 6690467 */
2888         if (error != expected_error && expected_error != EBUSY) {
2889                 fatal(0, "attach (%s %llu, %s %llu, %d) "
2890                     "returned %d, expected %d",
2891                     oldpath, (longlong_t)oldsize, newpath,
2892                     (longlong_t)newsize, replacing, error, expected_error);
2893         }
2894 
2895         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2896 }
2897 
2898 /*
2899  * Callback function which expands the physical size of the vdev.
2900  */
2901 vdev_t *
2902 grow_vdev(vdev_t *vd, void *arg)
2903 {
2904         spa_t *spa = vd->vdev_spa;
2905         size_t *newsize = arg;
2906         size_t fsize;
2907         int fd;
2908 
2909         ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2910         ASSERT(vd->vdev_ops->vdev_op_leaf);
2911 
2912         if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2913                 return (vd);
2914 
2915         fsize = lseek(fd, 0, SEEK_END);
2916         (void) ftruncate(fd, *newsize);
2917 
2918         if (ztest_opts.zo_verbose >= 6) {
2919                 (void) printf("%s grew from %lu to %lu bytes\n",
2920                     vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2921         }
2922         (void) close(fd);
2923         return (NULL);
2924 }
2925 
2926 /*
2927  * Callback function which expands a given vdev by calling vdev_online().
2928  */
2929 /* ARGSUSED */
2930 vdev_t *
2931 online_vdev(vdev_t *vd, void *arg)
2932 {
2933         spa_t *spa = vd->vdev_spa;
2934         vdev_t *tvd = vd->vdev_top;
2935         uint64_t guid = vd->vdev_guid;
2936         uint64_t generation = spa->spa_config_generation + 1;
2937         vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2938         int error;
2939 
2940         ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2941         ASSERT(vd->vdev_ops->vdev_op_leaf);
2942 
2943         /* Calling vdev_online will initialize the new metaslabs */
2944         spa_config_exit(spa, SCL_STATE, spa);
2945         error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2946         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2947 
2948         /*
2949          * If vdev_online returned an error or the underlying vdev_open
2950          * failed then we abort the expand. The only way to know that
2951          * vdev_open fails is by checking the returned newstate.
2952          */
2953         if (error || newstate != VDEV_STATE_HEALTHY) {
2954                 if (ztest_opts.zo_verbose >= 5) {
2955                         (void) printf("Unable to expand vdev, state %llu, "
2956                             "error %d\n", (u_longlong_t)newstate, error);
2957                 }
2958                 return (vd);
2959         }
2960         ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2961 
2962         /*
2963          * Since we dropped the lock we need to ensure that we're
2964          * still talking to the original vdev. It's possible this
2965          * vdev may have been detached/replaced while we were
2966          * trying to online it.
2967          */
2968         if (generation != spa->spa_config_generation) {
2969                 if (ztest_opts.zo_verbose >= 5) {
2970                         (void) printf("vdev configuration has changed, "
2971                             "guid %llu, state %llu, expected gen %llu, "
2972                             "got gen %llu\n",
2973                             (u_longlong_t)guid,
2974                             (u_longlong_t)tvd->vdev_state,
2975                             (u_longlong_t)generation,
2976                             (u_longlong_t)spa->spa_config_generation);
2977                 }
2978                 return (vd);
2979         }
2980         return (NULL);
2981 }
2982 
2983 /*
2984  * Traverse the vdev tree calling the supplied function.
2985  * We continue to walk the tree until we either have walked all
2986  * children or we receive a non-NULL return from the callback.
2987  * If a NULL callback is passed, then we just return back the first
2988  * leaf vdev we encounter.
2989  */
2990 vdev_t *
2991 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2992 {
2993         if (vd->vdev_ops->vdev_op_leaf) {
2994                 if (func == NULL)
2995                         return (vd);
2996                 else
2997                         return (func(vd, arg));
2998         }
2999 
3000         for (uint_t c = 0; c < vd->vdev_children; c++) {
3001                 vdev_t *cvd = vd->vdev_child[c];
3002                 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3003                         return (cvd);
3004         }
3005         return (NULL);
3006 }
3007 
3008 /*
3009  * Verify that dynamic LUN growth works as expected.
3010  */
3011 /* ARGSUSED */
3012 void
3013 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3014 {
3015         spa_t *spa = ztest_spa;
3016         vdev_t *vd, *tvd;
3017         metaslab_class_t *mc;
3018         metaslab_group_t *mg;
3019         size_t psize, newsize;
3020         uint64_t top;
3021         uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3022 
3023         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3024         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3025 
3026         top = ztest_random_vdev_top(spa, B_TRUE);
3027 
3028         tvd = spa->spa_root_vdev->vdev_child[top];
3029         mg = tvd->vdev_mg;
3030         mc = mg->mg_class;
3031         old_ms_count = tvd->vdev_ms_count;
3032         old_class_space = metaslab_class_get_space(mc);
3033 
3034         /*
3035          * Determine the size of the first leaf vdev associated with
3036          * our top-level device.
3037          */
3038         vd = vdev_walk_tree(tvd, NULL, NULL);
3039         ASSERT3P(vd, !=, NULL);
3040         ASSERT(vd->vdev_ops->vdev_op_leaf);
3041 
3042         psize = vd->vdev_psize;
3043 
3044         /*
3045          * We only try to expand the vdev if it's healthy, less than 4x its
3046          * original size, and it has a valid psize.
3047          */
3048         if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3049             psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3050                 spa_config_exit(spa, SCL_STATE, spa);
3051                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3052                 return;
3053         }
3054         ASSERT(psize > 0);
3055         newsize = psize + psize / 8;
3056         ASSERT3U(newsize, >, psize);
3057 
3058         if (ztest_opts.zo_verbose >= 6) {
3059                 (void) printf("Expanding LUN %s from %lu to %lu\n",
3060                     vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3061         }
3062 
3063         /*
3064          * Growing the vdev is a two step process:
3065          *      1). expand the physical size (i.e. relabel)
3066          *      2). online the vdev to create the new metaslabs
3067          */
3068         if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3069             vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3070             tvd->vdev_state != VDEV_STATE_HEALTHY) {
3071                 if (ztest_opts.zo_verbose >= 5) {
3072                         (void) printf("Could not expand LUN because "
3073                             "the vdev configuration changed.\n");
3074                 }
3075                 spa_config_exit(spa, SCL_STATE, spa);
3076                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3077                 return;
3078         }
3079 
3080         spa_config_exit(spa, SCL_STATE, spa);
3081 
3082         /*
3083          * Expanding the LUN will update the config asynchronously,
3084          * thus we must wait for the async thread to complete any
3085          * pending tasks before proceeding.
3086          */
3087         for (;;) {
3088                 boolean_t done;
3089                 mutex_enter(&spa->spa_async_lock);
3090                 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3091                 mutex_exit(&spa->spa_async_lock);
3092                 if (done)
3093                         break;
3094                 txg_wait_synced(spa_get_dsl(spa), 0);
3095                 (void) poll(NULL, 0, 100);
3096         }
3097 
3098         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3099 
3100         tvd = spa->spa_root_vdev->vdev_child[top];
3101         new_ms_count = tvd->vdev_ms_count;
3102         new_class_space = metaslab_class_get_space(mc);
3103 
3104         if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3105                 if (ztest_opts.zo_verbose >= 5) {
3106                         (void) printf("Could not verify LUN expansion due to "
3107                             "intervening vdev offline or remove.\n");
3108                 }
3109                 spa_config_exit(spa, SCL_STATE, spa);
3110                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3111                 return;
3112         }
3113 
3114         /*
3115          * Make sure we were able to grow the vdev.
3116          */
3117         if (new_ms_count <= old_ms_count)
3118                 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3119                     old_ms_count, new_ms_count);
3120 
3121         /*
3122          * Make sure we were able to grow the pool.
3123          */
3124         if (new_class_space <= old_class_space)
3125                 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3126                     old_class_space, new_class_space);
3127 
3128         if (ztest_opts.zo_verbose >= 5) {
3129                 char oldnumbuf[6], newnumbuf[6];
3130 
3131                 nicenum(old_class_space, oldnumbuf);
3132                 nicenum(new_class_space, newnumbuf);
3133                 (void) printf("%s grew from %s to %s\n",
3134                     spa->spa_name, oldnumbuf, newnumbuf);
3135         }
3136 
3137         spa_config_exit(spa, SCL_STATE, spa);
3138         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3139 }
3140 
3141 /*
3142  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3143  */
3144 /* ARGSUSED */
3145 static void
3146 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3147 {
3148         /*
3149          * Create the objects common to all ztest datasets.
3150          */
3151         VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3152             DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3153 }
3154 
3155 static int
3156 ztest_dataset_create(char *dsname)
3157 {
3158         uint64_t zilset = ztest_random(100);
3159         int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3160             ztest_objset_create_cb, NULL);
3161 
3162         if (err || zilset < 80)
3163                 return (err);
3164 
3165         if (ztest_opts.zo_verbose >= 6)
3166                 (void) printf("Setting dataset %s to sync always\n", dsname);
3167         return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3168             ZFS_SYNC_ALWAYS, B_FALSE));
3169 }
3170 
3171 /* ARGSUSED */
3172 static int
3173 ztest_objset_destroy_cb(const char *name, void *arg)
3174 {
3175         objset_t *os;
3176         dmu_object_info_t doi;
3177         int error;
3178 
3179         /*
3180          * Verify that the dataset contains a directory object.
3181          */
3182         VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3183         error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3184         if (error != ENOENT) {
3185                 /* We could have crashed in the middle of destroying it */
3186                 ASSERT0(error);
3187                 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3188                 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3189         }
3190         dmu_objset_disown(os, FTAG);
3191 
3192         /*
3193          * Destroy the dataset.
3194          */
3195         if (strchr(name, '@') != NULL) {
3196                 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3197         } else {
3198                 VERIFY0(dsl_destroy_head(name));
3199         }
3200         return (0);
3201 }
3202 
3203 static boolean_t
3204 ztest_snapshot_create(char *osname, uint64_t id)
3205 {
3206         char snapname[MAXNAMELEN];
3207         int error;
3208 
3209         (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3210 
3211         error = dmu_objset_snapshot_one(osname, snapname);
3212         if (error == ENOSPC) {
3213                 ztest_record_enospc(FTAG);
3214                 return (B_FALSE);
3215         }
3216         if (error != 0 && error != EEXIST) {
3217                 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3218                     snapname, error);
3219         }
3220         return (B_TRUE);
3221 }
3222 
3223 static boolean_t
3224 ztest_snapshot_destroy(char *osname, uint64_t id)
3225 {
3226         char snapname[MAXNAMELEN];
3227         int error;
3228 
3229         (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3230             (u_longlong_t)id);
3231 
3232         error = dsl_destroy_snapshot(snapname, B_FALSE);
3233         if (error != 0 && error != ENOENT)
3234                 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3235         return (B_TRUE);
3236 }
3237 
3238 /* ARGSUSED */
3239 void
3240 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3241 {
3242         ztest_ds_t zdtmp;
3243         int iters;
3244         int error;
3245         objset_t *os, *os2;
3246         char name[MAXNAMELEN];
3247         zilog_t *zilog;
3248 
3249         (void) rw_rdlock(&ztest_name_lock);
3250 
3251         (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3252             ztest_opts.zo_pool, (u_longlong_t)id);
3253 
3254         /*
3255          * If this dataset exists from a previous run, process its replay log
3256          * half of the time.  If we don't replay it, then dmu_objset_destroy()
3257          * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3258          */
3259         if (ztest_random(2) == 0 &&
3260             dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3261                 ztest_zd_init(&zdtmp, NULL, os);
3262                 zil_replay(os, &zdtmp, ztest_replay_vector);
3263                 ztest_zd_fini(&zdtmp);
3264                 dmu_objset_disown(os, FTAG);
3265         }
3266 
3267         /*
3268          * There may be an old instance of the dataset we're about to
3269          * create lying around from a previous run.  If so, destroy it
3270          * and all of its snapshots.
3271          */
3272         (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3273             DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3274 
3275         /*
3276          * Verify that the destroyed dataset is no longer in the namespace.
3277          */
3278         VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3279             FTAG, &os));
3280 
3281         /*
3282          * Verify that we can create a new dataset.
3283          */
3284         error = ztest_dataset_create(name);
3285         if (error) {
3286                 if (error == ENOSPC) {
3287                         ztest_record_enospc(FTAG);
3288                         (void) rw_unlock(&ztest_name_lock);
3289                         return;
3290                 }
3291                 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3292         }
3293 
3294         VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3295 
3296         ztest_zd_init(&zdtmp, NULL, os);
3297 
3298         /*
3299          * Open the intent log for it.
3300          */
3301         zilog = zil_open(os, ztest_get_data);
3302 
3303         /*
3304          * Put some objects in there, do a little I/O to them,
3305          * and randomly take a couple of snapshots along the way.
3306          */
3307         iters = ztest_random(5);
3308         for (int i = 0; i < iters; i++) {
3309                 ztest_dmu_object_alloc_free(&zdtmp, id);
3310                 if (ztest_random(iters) == 0)
3311                         (void) ztest_snapshot_create(name, i);
3312         }
3313 
3314         /*
3315          * Verify that we cannot create an existing dataset.
3316          */
3317         VERIFY3U(EEXIST, ==,
3318             dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3319 
3320         /*
3321          * Verify that we can hold an objset that is also owned.
3322          */
3323         VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3324         dmu_objset_rele(os2, FTAG);
3325 
3326         /*
3327          * Verify that we cannot own an objset that is already owned.
3328          */
3329         VERIFY3U(EBUSY, ==,
3330             dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3331 
3332         zil_close(zilog);
3333         dmu_objset_disown(os, FTAG);
3334         ztest_zd_fini(&zdtmp);
3335 
3336         (void) rw_unlock(&ztest_name_lock);
3337 }
3338 
3339 /*
3340  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3341  */
3342 void
3343 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3344 {
3345         (void) rw_rdlock(&ztest_name_lock);
3346         (void) ztest_snapshot_destroy(zd->zd_name, id);
3347         (void) ztest_snapshot_create(zd->zd_name, id);
3348         (void) rw_unlock(&ztest_name_lock);
3349 }
3350 
3351 /*
3352  * Cleanup non-standard snapshots and clones.
3353  */
3354 void
3355 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3356 {
3357         char snap1name[MAXNAMELEN];
3358         char clone1name[MAXNAMELEN];
3359         char snap2name[MAXNAMELEN];
3360         char clone2name[MAXNAMELEN];
3361         char snap3name[MAXNAMELEN];
3362         int error;
3363 
3364         (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3365         (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3366         (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3367         (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3368         (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3369 
3370         error = dsl_destroy_head(clone2name);
3371         if (error && error != ENOENT)
3372                 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3373         error = dsl_destroy_snapshot(snap3name, B_FALSE);
3374         if (error && error != ENOENT)
3375                 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3376         error = dsl_destroy_snapshot(snap2name, B_FALSE);
3377         if (error && error != ENOENT)
3378                 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3379         error = dsl_destroy_head(clone1name);
3380         if (error && error != ENOENT)
3381                 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3382         error = dsl_destroy_snapshot(snap1name, B_FALSE);
3383         if (error && error != ENOENT)
3384                 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3385 }
3386 
3387 /*
3388  * Verify dsl_dataset_promote handles EBUSY
3389  */
3390 void
3391 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3392 {
3393         objset_t *os;
3394         char snap1name[MAXNAMELEN];
3395         char clone1name[MAXNAMELEN];
3396         char snap2name[MAXNAMELEN];
3397         char clone2name[MAXNAMELEN];
3398         char snap3name[MAXNAMELEN];
3399         char *osname = zd->zd_name;
3400         int error;
3401 
3402         (void) rw_rdlock(&ztest_name_lock);
3403 
3404         ztest_dsl_dataset_cleanup(osname, id);
3405 
3406         (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3407         (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3408         (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3409         (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3410         (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3411 
3412         error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3413         if (error && error != EEXIST) {
3414                 if (error == ENOSPC) {
3415                         ztest_record_enospc(FTAG);
3416                         goto out;
3417                 }
3418                 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3419         }
3420 
3421         error = dmu_objset_clone(clone1name, snap1name);
3422         if (error) {
3423                 if (error == ENOSPC) {
3424                         ztest_record_enospc(FTAG);
3425                         goto out;
3426                 }
3427                 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3428         }
3429 
3430         error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3431         if (error && error != EEXIST) {
3432                 if (error == ENOSPC) {
3433                         ztest_record_enospc(FTAG);
3434                         goto out;
3435                 }
3436                 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3437         }
3438 
3439         error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3440         if (error && error != EEXIST) {
3441                 if (error == ENOSPC) {
3442                         ztest_record_enospc(FTAG);
3443                         goto out;
3444                 }
3445                 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3446         }
3447 
3448         error = dmu_objset_clone(clone2name, snap3name);
3449         if (error) {
3450                 if (error == ENOSPC) {
3451                         ztest_record_enospc(FTAG);
3452                         goto out;
3453                 }
3454                 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3455         }
3456 
3457         error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3458         if (error)
3459                 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3460         error = dsl_dataset_promote(clone2name, NULL);
3461         if (error != EBUSY)
3462                 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3463                     error);
3464         dmu_objset_disown(os, FTAG);
3465 
3466 out:
3467         ztest_dsl_dataset_cleanup(osname, id);
3468 
3469         (void) rw_unlock(&ztest_name_lock);
3470 }
3471 
3472 /*
3473  * Verify that dmu_object_{alloc,free} work as expected.
3474  */
3475 void
3476 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3477 {
3478         ztest_od_t od[4];
3479         int batchsize = sizeof (od) / sizeof (od[0]);
3480 
3481         for (int b = 0; b < batchsize; b++)
3482                 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3483 
3484         /*
3485          * Destroy the previous batch of objects, create a new batch,
3486          * and do some I/O on the new objects.
3487          */
3488         if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3489                 return;
3490 
3491         while (ztest_random(4 * batchsize) != 0)
3492                 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3493                     ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3494 }
3495 
3496 /*
3497  * Verify that dmu_{read,write} work as expected.
3498  */
3499 void
3500 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3501 {
3502         objset_t *os = zd->zd_os;
3503         ztest_od_t od[2];
3504         dmu_tx_t *tx;
3505         int i, freeit, error;
3506         uint64_t n, s, txg;
3507         bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3508         uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3509         uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3510         uint64_t regions = 997;
3511         uint64_t stride = 123456789ULL;
3512         uint64_t width = 40;
3513         int free_percent = 5;
3514 
3515         /*
3516          * This test uses two objects, packobj and bigobj, that are always
3517          * updated together (i.e. in the same tx) so that their contents are
3518          * in sync and can be compared.  Their contents relate to each other
3519          * in a simple way: packobj is a dense array of 'bufwad' structures,
3520          * while bigobj is a sparse array of the same bufwads.  Specifically,
3521          * for any index n, there are three bufwads that should be identical:
3522          *
3523          *      packobj, at offset n * sizeof (bufwad_t)
3524          *      bigobj, at the head of the nth chunk
3525          *      bigobj, at the tail of the nth chunk
3526          *
3527          * The chunk size is arbitrary. It doesn't have to be a power of two,
3528          * and it doesn't have any relation to the object blocksize.
3529          * The only requirement is that it can hold at least two bufwads.
3530          *
3531          * Normally, we write the bufwad to each of these locations.
3532          * However, free_percent of the time we instead write zeroes to
3533          * packobj and perform a dmu_free_range() on bigobj.  By comparing
3534          * bigobj to packobj, we can verify that the DMU is correctly
3535          * tracking which parts of an object are allocated and free,
3536          * and that the contents of the allocated blocks are correct.
3537          */
3538 
3539         /*
3540          * Read the directory info.  If it's the first time, set things up.
3541          */
3542         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3543         ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3544 
3545         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3546                 return;
3547 
3548         bigobj = od[0].od_object;
3549         packobj = od[1].od_object;
3550         chunksize = od[0].od_gen;
3551         ASSERT(chunksize == od[1].od_gen);
3552 
3553         /*
3554          * Prefetch a random chunk of the big object.
3555          * Our aim here is to get some async reads in flight
3556          * for blocks that we may free below; the DMU should
3557          * handle this race correctly.
3558          */
3559         n = ztest_random(regions) * stride + ztest_random(width);
3560         s = 1 + ztest_random(2 * width - 1);
3561         dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3562 
3563         /*
3564          * Pick a random index and compute the offsets into packobj and bigobj.
3565          */
3566         n = ztest_random(regions) * stride + ztest_random(width);
3567         s = 1 + ztest_random(width - 1);
3568 
3569         packoff = n * sizeof (bufwad_t);
3570         packsize = s * sizeof (bufwad_t);
3571 
3572         bigoff = n * chunksize;
3573         bigsize = s * chunksize;
3574 
3575         packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3576         bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3577 
3578         /*
3579          * free_percent of the time, free a range of bigobj rather than
3580          * overwriting it.
3581          */
3582         freeit = (ztest_random(100) < free_percent);
3583 
3584         /*
3585          * Read the current contents of our objects.
3586          */
3587         error = dmu_read(os, packobj, packoff, packsize, packbuf,
3588             DMU_READ_PREFETCH);
3589         ASSERT0(error);
3590         error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3591             DMU_READ_PREFETCH);
3592         ASSERT0(error);
3593 
3594         /*
3595          * Get a tx for the mods to both packobj and bigobj.
3596          */
3597         tx = dmu_tx_create(os);
3598 
3599         dmu_tx_hold_write(tx, packobj, packoff, packsize);
3600 
3601         if (freeit)
3602                 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3603         else
3604                 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3605 
3606         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3607         if (txg == 0) {
3608                 umem_free(packbuf, packsize);
3609                 umem_free(bigbuf, bigsize);
3610                 return;
3611         }
3612 
3613         dmu_object_set_checksum(os, bigobj,
3614             (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3615 
3616         dmu_object_set_compress(os, bigobj,
3617             (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3618 
3619         /*
3620          * For each index from n to n + s, verify that the existing bufwad
3621          * in packobj matches the bufwads at the head and tail of the
3622          * corresponding chunk in bigobj.  Then update all three bufwads
3623          * with the new values we want to write out.
3624          */
3625         for (i = 0; i < s; i++) {
3626                 /* LINTED */
3627                 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3628                 /* LINTED */
3629                 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3630                 /* LINTED */
3631                 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3632 
3633                 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3634                 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3635 
3636                 if (pack->bw_txg > txg)
3637                         fatal(0, "future leak: got %llx, open txg is %llx",
3638                             pack->bw_txg, txg);
3639 
3640                 if (pack->bw_data != 0 && pack->bw_index != n + i)
3641                         fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3642                             pack->bw_index, n, i);
3643 
3644                 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3645                         fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3646 
3647                 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3648                         fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3649 
3650                 if (freeit) {
3651                         bzero(pack, sizeof (bufwad_t));
3652                 } else {
3653                         pack->bw_index = n + i;
3654                         pack->bw_txg = txg;
3655                         pack->bw_data = 1 + ztest_random(-2ULL);
3656                 }
3657                 *bigH = *pack;
3658                 *bigT = *pack;
3659         }
3660 
3661         /*
3662          * We've verified all the old bufwads, and made new ones.
3663          * Now write them out.
3664          */
3665         dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3666 
3667         if (freeit) {
3668                 if (ztest_opts.zo_verbose >= 7) {
3669                         (void) printf("freeing offset %llx size %llx"
3670                             " txg %llx\n",
3671                             (u_longlong_t)bigoff,
3672                             (u_longlong_t)bigsize,
3673                             (u_longlong_t)txg);
3674                 }
3675                 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3676         } else {
3677                 if (ztest_opts.zo_verbose >= 7) {
3678                         (void) printf("writing offset %llx size %llx"
3679                             " txg %llx\n",
3680                             (u_longlong_t)bigoff,
3681                             (u_longlong_t)bigsize,
3682                             (u_longlong_t)txg);
3683                 }
3684                 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3685         }
3686 
3687         dmu_tx_commit(tx);
3688 
3689         /*
3690          * Sanity check the stuff we just wrote.
3691          */
3692         {
3693                 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3694                 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3695 
3696                 VERIFY(0 == dmu_read(os, packobj, packoff,
3697                     packsize, packcheck, DMU_READ_PREFETCH));
3698                 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3699                     bigsize, bigcheck, DMU_READ_PREFETCH));
3700 
3701                 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3702                 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3703 
3704                 umem_free(packcheck, packsize);
3705                 umem_free(bigcheck, bigsize);
3706         }
3707 
3708         umem_free(packbuf, packsize);
3709         umem_free(bigbuf, bigsize);
3710 }
3711 
3712 void
3713 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3714     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3715 {
3716         uint64_t i;
3717         bufwad_t *pack;
3718         bufwad_t *bigH;
3719         bufwad_t *bigT;
3720 
3721         /*
3722          * For each index from n to n + s, verify that the existing bufwad
3723          * in packobj matches the bufwads at the head and tail of the
3724          * corresponding chunk in bigobj.  Then update all three bufwads
3725          * with the new values we want to write out.
3726          */
3727         for (i = 0; i < s; i++) {
3728                 /* LINTED */
3729                 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3730                 /* LINTED */
3731                 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3732                 /* LINTED */
3733                 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3734 
3735                 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3736                 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3737 
3738                 if (pack->bw_txg > txg)
3739                         fatal(0, "future leak: got %llx, open txg is %llx",
3740                             pack->bw_txg, txg);
3741 
3742                 if (pack->bw_data != 0 && pack->bw_index != n + i)
3743                         fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3744                             pack->bw_index, n, i);
3745 
3746                 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3747                         fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3748 
3749                 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3750                         fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3751 
3752                 pack->bw_index = n + i;
3753                 pack->bw_txg = txg;
3754                 pack->bw_data = 1 + ztest_random(-2ULL);
3755 
3756                 *bigH = *pack;
3757                 *bigT = *pack;
3758         }
3759 }
3760 
3761 void
3762 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3763 {
3764         objset_t *os = zd->zd_os;
3765         ztest_od_t od[2];
3766         dmu_tx_t *tx;
3767         uint64_t i;
3768         int error;
3769         uint64_t n, s, txg;
3770         bufwad_t *packbuf, *bigbuf;
3771         uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3772         uint64_t blocksize = ztest_random_blocksize();
3773         uint64_t chunksize = blocksize;
3774         uint64_t regions = 997;
3775         uint64_t stride = 123456789ULL;
3776         uint64_t width = 9;
3777         dmu_buf_t *bonus_db;
3778         arc_buf_t **bigbuf_arcbufs;
3779         dmu_object_info_t doi;
3780 
3781         /*
3782          * This test uses two objects, packobj and bigobj, that are always
3783          * updated together (i.e. in the same tx) so that their contents are
3784          * in sync and can be compared.  Their contents relate to each other
3785          * in a simple way: packobj is a dense array of 'bufwad' structures,
3786          * while bigobj is a sparse array of the same bufwads.  Specifically,
3787          * for any index n, there are three bufwads that should be identical:
3788          *
3789          *      packobj, at offset n * sizeof (bufwad_t)
3790          *      bigobj, at the head of the nth chunk
3791          *      bigobj, at the tail of the nth chunk
3792          *
3793          * The chunk size is set equal to bigobj block size so that
3794          * dmu_assign_arcbuf() can be tested for object updates.
3795          */
3796 
3797         /*
3798          * Read the directory info.  If it's the first time, set things up.
3799          */
3800         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3801         ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3802 
3803         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3804                 return;
3805 
3806         bigobj = od[0].od_object;
3807         packobj = od[1].od_object;
3808         blocksize = od[0].od_blocksize;
3809         chunksize = blocksize;
3810         ASSERT(chunksize == od[1].od_gen);
3811 
3812         VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3813         VERIFY(ISP2(doi.doi_data_block_size));
3814         VERIFY(chunksize == doi.doi_data_block_size);
3815         VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3816 
3817         /*
3818          * Pick a random index and compute the offsets into packobj and bigobj.
3819          */
3820         n = ztest_random(regions) * stride + ztest_random(width);
3821         s = 1 + ztest_random(width - 1);
3822 
3823         packoff = n * sizeof (bufwad_t);
3824         packsize = s * sizeof (bufwad_t);
3825 
3826         bigoff = n * chunksize;
3827         bigsize = s * chunksize;
3828 
3829         packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3830         bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3831 
3832         VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3833 
3834         bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3835 
3836         /*
3837          * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3838          * Iteration 1 test zcopy to already referenced dbufs.
3839          * Iteration 2 test zcopy to dirty dbuf in the same txg.
3840          * Iteration 3 test zcopy to dbuf dirty in previous txg.
3841          * Iteration 4 test zcopy when dbuf is no longer dirty.
3842          * Iteration 5 test zcopy when it can't be done.
3843          * Iteration 6 one more zcopy write.
3844          */
3845         for (i = 0; i < 7; i++) {
3846                 uint64_t j;
3847                 uint64_t off;
3848 
3849                 /*
3850                  * In iteration 5 (i == 5) use arcbufs
3851                  * that don't match bigobj blksz to test
3852                  * dmu_assign_arcbuf() when it can't directly
3853                  * assign an arcbuf to a dbuf.
3854                  */
3855                 for (j = 0; j < s; j++) {
3856                         if (i != 5) {
3857                                 bigbuf_arcbufs[j] =
3858                                     dmu_request_arcbuf(bonus_db, chunksize);
3859                         } else {
3860                                 bigbuf_arcbufs[2 * j] =
3861                                     dmu_request_arcbuf(bonus_db, chunksize / 2);
3862                                 bigbuf_arcbufs[2 * j + 1] =
3863                                     dmu_request_arcbuf(bonus_db, chunksize / 2);
3864                         }
3865                 }
3866 
3867                 /*
3868                  * Get a tx for the mods to both packobj and bigobj.
3869                  */
3870                 tx = dmu_tx_create(os);
3871 
3872                 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3873                 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3874 
3875                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3876                 if (txg == 0) {
3877                         umem_free(packbuf, packsize);
3878                         umem_free(bigbuf, bigsize);
3879                         for (j = 0; j < s; j++) {
3880                                 if (i != 5) {
3881                                         dmu_return_arcbuf(bigbuf_arcbufs[j]);
3882                                 } else {
3883                                         dmu_return_arcbuf(
3884                                             bigbuf_arcbufs[2 * j]);
3885                                         dmu_return_arcbuf(
3886                                             bigbuf_arcbufs[2 * j + 1]);
3887                                 }
3888                         }
3889                         umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3890                         dmu_buf_rele(bonus_db, FTAG);
3891                         return;
3892                 }
3893 
3894                 /*
3895                  * 50% of the time don't read objects in the 1st iteration to
3896                  * test dmu_assign_arcbuf() for the case when there're no
3897                  * existing dbufs for the specified offsets.
3898                  */
3899                 if (i != 0 || ztest_random(2) != 0) {
3900                         error = dmu_read(os, packobj, packoff,
3901                             packsize, packbuf, DMU_READ_PREFETCH);
3902                         ASSERT0(error);
3903                         error = dmu_read(os, bigobj, bigoff, bigsize,
3904                             bigbuf, DMU_READ_PREFETCH);
3905                         ASSERT0(error);
3906                 }
3907                 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3908                     n, chunksize, txg);
3909 
3910                 /*
3911                  * We've verified all the old bufwads, and made new ones.
3912                  * Now write them out.
3913                  */
3914                 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3915                 if (ztest_opts.zo_verbose >= 7) {
3916                         (void) printf("writing offset %llx size %llx"
3917                             " txg %llx\n",
3918                             (u_longlong_t)bigoff,
3919                             (u_longlong_t)bigsize,
3920                             (u_longlong_t)txg);
3921                 }
3922                 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3923                         dmu_buf_t *dbt;
3924                         if (i != 5) {
3925                                 bcopy((caddr_t)bigbuf + (off - bigoff),
3926                                     bigbuf_arcbufs[j]->b_data, chunksize);
3927                         } else {
3928                                 bcopy((caddr_t)bigbuf + (off - bigoff),
3929                                     bigbuf_arcbufs[2 * j]->b_data,
3930                                     chunksize / 2);
3931                                 bcopy((caddr_t)bigbuf + (off - bigoff) +
3932                                     chunksize / 2,
3933                                     bigbuf_arcbufs[2 * j + 1]->b_data,
3934                                     chunksize / 2);
3935                         }
3936 
3937                         if (i == 1) {
3938                                 VERIFY(dmu_buf_hold(os, bigobj, off,
3939                                     FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3940                         }
3941                         if (i != 5) {
3942                                 dmu_assign_arcbuf(bonus_db, off,
3943                                     bigbuf_arcbufs[j], tx);
3944                         } else {
3945                                 dmu_assign_arcbuf(bonus_db, off,
3946                                     bigbuf_arcbufs[2 * j], tx);
3947                                 dmu_assign_arcbuf(bonus_db,
3948                                     off + chunksize / 2,
3949                                     bigbuf_arcbufs[2 * j + 1], tx);
3950                         }
3951                         if (i == 1) {
3952                                 dmu_buf_rele(dbt, FTAG);
3953                         }
3954                 }
3955                 dmu_tx_commit(tx);
3956 
3957                 /*
3958                  * Sanity check the stuff we just wrote.
3959                  */
3960                 {
3961                         void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3962                         void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3963 
3964                         VERIFY(0 == dmu_read(os, packobj, packoff,
3965                             packsize, packcheck, DMU_READ_PREFETCH));
3966                         VERIFY(0 == dmu_read(os, bigobj, bigoff,
3967                             bigsize, bigcheck, DMU_READ_PREFETCH));
3968 
3969                         ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3970                         ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3971 
3972                         umem_free(packcheck, packsize);
3973                         umem_free(bigcheck, bigsize);
3974                 }
3975                 if (i == 2) {
3976                         txg_wait_open(dmu_objset_pool(os), 0);
3977                 } else if (i == 3) {
3978                         txg_wait_synced(dmu_objset_pool(os), 0);
3979                 }
3980         }
3981 
3982         dmu_buf_rele(bonus_db, FTAG);
3983         umem_free(packbuf, packsize);
3984         umem_free(bigbuf, bigsize);
3985         umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3986 }
3987 
3988 /* ARGSUSED */
3989 void
3990 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3991 {
3992         ztest_od_t od[1];
3993         uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3994             (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3995 
3996         /*
3997          * Have multiple threads write to large offsets in an object
3998          * to verify that parallel writes to an object -- even to the
3999          * same blocks within the object -- doesn't cause any trouble.
4000          */
4001         ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4002 
4003         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4004                 return;
4005 
4006         while (ztest_random(10) != 0)
4007                 ztest_io(zd, od[0].od_object, offset);
4008 }
4009 
4010 void
4011 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4012 {
4013         ztest_od_t od[1];
4014         uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4015             (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4016         uint64_t count = ztest_random(20) + 1;
4017         uint64_t blocksize = ztest_random_blocksize();
4018         void *data;
4019 
4020         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4021 
4022         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4023                 return;
4024 
4025         if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4026                 return;
4027 
4028         ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4029 
4030         data = umem_zalloc(blocksize, UMEM_NOFAIL);
4031 
4032         while (ztest_random(count) != 0) {
4033                 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4034                 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4035                     data) != 0)
4036                         break;
4037                 while (ztest_random(4) != 0)
4038                         ztest_io(zd, od[0].od_object, randoff);
4039         }
4040 
4041         umem_free(data, blocksize);
4042 }
4043 
4044 /*
4045  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4046  */
4047 #define ZTEST_ZAP_MIN_INTS      1
4048 #define ZTEST_ZAP_MAX_INTS      4
4049 #define ZTEST_ZAP_MAX_PROPS     1000
4050 
4051 void
4052 ztest_zap(ztest_ds_t *zd, uint64_t id)
4053 {
4054         objset_t *os = zd->zd_os;
4055         ztest_od_t od[1];
4056         uint64_t object;
4057         uint64_t txg, last_txg;
4058         uint64_t value[ZTEST_ZAP_MAX_INTS];
4059         uint64_t zl_ints, zl_intsize, prop;
4060         int i, ints;
4061         dmu_tx_t *tx;
4062         char propname[100], txgname[100];
4063         int error;
4064         char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4065 
4066         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4067 
4068         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4069                 return;
4070 
4071         object = od[0].od_object;
4072 
4073         /*
4074          * Generate a known hash collision, and verify that
4075          * we can lookup and remove both entries.
4076          */
4077         tx = dmu_tx_create(os);
4078         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4079         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4080         if (txg == 0)
4081                 return;
4082         for (i = 0; i < 2; i++) {
4083                 value[i] = i;
4084                 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4085                     1, &value[i], tx));
4086         }
4087         for (i = 0; i < 2; i++) {
4088                 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4089                     sizeof (uint64_t), 1, &value[i], tx));
4090                 VERIFY3U(0, ==,
4091                     zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4092                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4093                 ASSERT3U(zl_ints, ==, 1);
4094         }
4095         for (i = 0; i < 2; i++) {
4096                 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4097         }
4098         dmu_tx_commit(tx);
4099 
4100         /*
4101          * Generate a buch of random entries.
4102          */
4103         ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4104 
4105         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4106         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4107         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4108         bzero(value, sizeof (value));
4109         last_txg = 0;
4110 
4111         /*
4112          * If these zap entries already exist, validate their contents.
4113          */
4114         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4115         if (error == 0) {
4116                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4117                 ASSERT3U(zl_ints, ==, 1);
4118 
4119                 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4120                     zl_ints, &last_txg) == 0);
4121 
4122                 VERIFY(zap_length(os, object, propname, &zl_intsize,
4123                     &zl_ints) == 0);
4124 
4125                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4126                 ASSERT3U(zl_ints, ==, ints);
4127 
4128                 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4129                     zl_ints, value) == 0);
4130 
4131                 for (i = 0; i < ints; i++) {
4132                         ASSERT3U(value[i], ==, last_txg + object + i);
4133                 }
4134         } else {
4135                 ASSERT3U(error, ==, ENOENT);
4136         }
4137 
4138         /*
4139          * Atomically update two entries in our zap object.
4140          * The first is named txg_%llu, and contains the txg
4141          * in which the property was last updated.  The second
4142          * is named prop_%llu, and the nth element of its value
4143          * should be txg + object + n.
4144          */
4145         tx = dmu_tx_create(os);
4146         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4147         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4148         if (txg == 0)
4149                 return;
4150 
4151         if (last_txg > txg)
4152                 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4153 
4154         for (i = 0; i < ints; i++)
4155                 value[i] = txg + object + i;
4156 
4157         VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4158             1, &txg, tx));
4159         VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4160             ints, value, tx));
4161 
4162         dmu_tx_commit(tx);
4163 
4164         /*
4165          * Remove a random pair of entries.
4166          */
4167         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4168         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4169         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4170 
4171         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4172 
4173         if (error == ENOENT)
4174                 return;
4175 
4176         ASSERT0(error);
4177 
4178         tx = dmu_tx_create(os);
4179         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4180         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4181         if (txg == 0)
4182                 return;
4183         VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4184         VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4185         dmu_tx_commit(tx);
4186 }
4187 
4188 /*
4189  * Testcase to test the upgrading of a microzap to fatzap.
4190  */
4191 void
4192 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4193 {
4194         objset_t *os = zd->zd_os;
4195         ztest_od_t od[1];
4196         uint64_t object, txg;
4197 
4198         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4199 
4200         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4201                 return;
4202 
4203         object = od[0].od_object;
4204 
4205         /*
4206          * Add entries to this ZAP and make sure it spills over
4207          * and gets upgraded to a fatzap. Also, since we are adding
4208          * 2050 entries we should see ptrtbl growth and leaf-block split.
4209          */
4210         for (int i = 0; i < 2050; i++) {
4211                 char name[MAXNAMELEN];
4212                 uint64_t value = i;
4213                 dmu_tx_t *tx;
4214                 int error;
4215 
4216                 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4217                     id, value);
4218 
4219                 tx = dmu_tx_create(os);
4220                 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4221                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4222                 if (txg == 0)
4223                         return;
4224                 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4225                     &value, tx);
4226                 ASSERT(error == 0 || error == EEXIST);
4227                 dmu_tx_commit(tx);
4228         }
4229 }
4230 
4231 /* ARGSUSED */
4232 void
4233 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4234 {
4235         objset_t *os = zd->zd_os;
4236         ztest_od_t od[1];
4237         uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4238         dmu_tx_t *tx;
4239         int i, namelen, error;
4240         int micro = ztest_random(2);
4241         char name[20], string_value[20];
4242         void *data;
4243 
4244         ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4245 
4246         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4247                 return;
4248 
4249         object = od[0].od_object;
4250 
4251         /*
4252          * Generate a random name of the form 'xxx.....' where each
4253          * x is a random printable character and the dots are dots.
4254          * There are 94 such characters, and the name length goes from
4255          * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4256          */
4257         namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4258 
4259         for (i = 0; i < 3; i++)
4260                 name[i] = '!' + ztest_random('~' - '!' + 1);
4261         for (; i < namelen - 1; i++)
4262                 name[i] = '.';
4263         name[i] = '\0';
4264 
4265         if ((namelen & 1) || micro) {
4266                 wsize = sizeof (txg);
4267                 wc = 1;
4268                 data = &txg;
4269         } else {
4270                 wsize = 1;
4271                 wc = namelen;
4272                 data = string_value;
4273         }
4274 
4275         count = -1ULL;
4276         VERIFY0(zap_count(os, object, &count));
4277         ASSERT(count != -1ULL);
4278 
4279         /*
4280          * Select an operation: length, lookup, add, update, remove.
4281          */
4282         i = ztest_random(5);
4283 
4284         if (i >= 2) {
4285                 tx = dmu_tx_create(os);
4286                 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4287                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4288                 if (txg == 0)
4289                         return;
4290                 bcopy(name, string_value, namelen);
4291         } else {
4292                 tx = NULL;
4293                 txg = 0;
4294                 bzero(string_value, namelen);
4295         }
4296 
4297         switch (i) {
4298 
4299         case 0:
4300                 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4301                 if (error == 0) {
4302                         ASSERT3U(wsize, ==, zl_wsize);
4303                         ASSERT3U(wc, ==, zl_wc);
4304                 } else {
4305                         ASSERT3U(error, ==, ENOENT);
4306                 }
4307                 break;
4308 
4309         case 1:
4310                 error = zap_lookup(os, object, name, wsize, wc, data);
4311                 if (error == 0) {
4312                         if (data == string_value &&
4313                             bcmp(name, data, namelen) != 0)
4314                                 fatal(0, "name '%s' != val '%s' len %d",
4315                                     name, data, namelen);
4316                 } else {
4317                         ASSERT3U(error, ==, ENOENT);
4318                 }
4319                 break;
4320 
4321         case 2:
4322                 error = zap_add(os, object, name, wsize, wc, data, tx);
4323                 ASSERT(error == 0 || error == EEXIST);
4324                 break;
4325 
4326         case 3:
4327                 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4328                 break;
4329 
4330         case 4:
4331                 error = zap_remove(os, object, name, tx);
4332                 ASSERT(error == 0 || error == ENOENT);
4333                 break;
4334         }
4335 
4336         if (tx != NULL)
4337                 dmu_tx_commit(tx);
4338 }
4339 
4340 /*
4341  * Commit callback data.
4342  */
4343 typedef struct ztest_cb_data {
4344         list_node_t             zcd_node;
4345         uint64_t                zcd_txg;
4346         int                     zcd_expected_err;
4347         boolean_t               zcd_added;
4348         boolean_t               zcd_called;
4349         spa_t                   *zcd_spa;
4350 } ztest_cb_data_t;
4351 
4352 /* This is the actual commit callback function */
4353 static void
4354 ztest_commit_callback(void *arg, int error)
4355 {
4356         ztest_cb_data_t *data = arg;
4357         uint64_t synced_txg;
4358 
4359         VERIFY(data != NULL);
4360         VERIFY3S(data->zcd_expected_err, ==, error);
4361         VERIFY(!data->zcd_called);
4362 
4363         synced_txg = spa_last_synced_txg(data->zcd_spa);
4364         if (data->zcd_txg > synced_txg)
4365                 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4366                     ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4367                     synced_txg);
4368 
4369         data->zcd_called = B_TRUE;
4370 
4371         if (error == ECANCELED) {
4372                 ASSERT0(data->zcd_txg);
4373                 ASSERT(!data->zcd_added);
4374 
4375                 /*
4376                  * The private callback data should be destroyed here, but
4377                  * since we are going to check the zcd_called field after
4378                  * dmu_tx_abort(), we will destroy it there.
4379                  */
4380                 return;
4381         }
4382 
4383         /* Was this callback added to the global callback list? */
4384         if (!data->zcd_added)
4385                 goto out;
4386 
4387         ASSERT3U(data->zcd_txg, !=, 0);
4388 
4389         /* Remove our callback from the list */
4390         (void) mutex_lock(&zcl.zcl_callbacks_lock);
4391         list_remove(&zcl.zcl_callbacks, data);
4392         (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4393 
4394 out:
4395         umem_free(data, sizeof (ztest_cb_data_t));
4396 }
4397 
4398 /* Allocate and initialize callback data structure */
4399 static ztest_cb_data_t *
4400 ztest_create_cb_data(objset_t *os, uint64_t txg)
4401 {
4402         ztest_cb_data_t *cb_data;
4403 
4404         cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4405 
4406         cb_data->zcd_txg = txg;
4407         cb_data->zcd_spa = dmu_objset_spa(os);
4408 
4409         return (cb_data);
4410 }
4411 
4412 /*
4413  * If a number of txgs equal to this threshold have been created after a commit
4414  * callback has been registered but not called, then we assume there is an
4415  * implementation bug.
4416  */
4417 #define ZTEST_COMMIT_CALLBACK_THRESH    (TXG_CONCURRENT_STATES + 2)
4418 
4419 /*
4420  * Commit callback test.
4421  */
4422 void
4423 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4424 {
4425         objset_t *os = zd->zd_os;
4426         ztest_od_t od[1];
4427         dmu_tx_t *tx;
4428         ztest_cb_data_t *cb_data[3], *tmp_cb;
4429         uint64_t old_txg, txg;
4430         int i, error;
4431 
4432         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4433 
4434         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4435                 return;
4436 
4437         tx = dmu_tx_create(os);
4438 
4439         cb_data[0] = ztest_create_cb_data(os, 0);
4440         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4441 
4442         dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4443 
4444         /* Every once in a while, abort the transaction on purpose */
4445         if (ztest_random(100) == 0)
4446                 error = -1;
4447 
4448         if (!error)
4449                 error = dmu_tx_assign(tx, TXG_NOWAIT);
4450 
4451         txg = error ? 0 : dmu_tx_get_txg(tx);
4452 
4453         cb_data[0]->zcd_txg = txg;
4454         cb_data[1] = ztest_create_cb_data(os, txg);
4455         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4456 
4457         if (error) {
4458                 /*
4459                  * It's not a strict requirement to call the registered
4460                  * callbacks from inside dmu_tx_abort(), but that's what
4461                  * it's supposed to happen in the current implementation
4462                  * so we will check for that.
4463                  */
4464                 for (i = 0; i < 2; i++) {
4465                         cb_data[i]->zcd_expected_err = ECANCELED;
4466                         VERIFY(!cb_data[i]->zcd_called);
4467                 }
4468 
4469                 dmu_tx_abort(tx);
4470 
4471                 for (i = 0; i < 2; i++) {
4472                         VERIFY(cb_data[i]->zcd_called);
4473                         umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4474                 }
4475 
4476                 return;
4477         }
4478 
4479         cb_data[2] = ztest_create_cb_data(os, txg);
4480         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4481 
4482         /*
4483          * Read existing data to make sure there isn't a future leak.
4484          */
4485         VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4486             &old_txg, DMU_READ_PREFETCH));
4487 
4488         if (old_txg > txg)
4489                 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4490                     old_txg, txg);
4491 
4492         dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4493 
4494         (void) mutex_lock(&zcl.zcl_callbacks_lock);
4495 
4496         /*
4497          * Since commit callbacks don't have any ordering requirement and since
4498          * it is theoretically possible for a commit callback to be called
4499          * after an arbitrary amount of time has elapsed since its txg has been
4500          * synced, it is difficult to reliably determine whether a commit
4501          * callback hasn't been called due to high load or due to a flawed
4502          * implementation.
4503          *
4504          * In practice, we will assume that if after a certain number of txgs a
4505          * commit callback hasn't been called, then most likely there's an
4506          * implementation bug..
4507          */
4508         tmp_cb = list_head(&zcl.zcl_callbacks);
4509         if (tmp_cb != NULL &&
4510             tmp_cb->zcd_txg > txg - ZTEST_COMMIT_CALLBACK_THRESH) {
4511                 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4512                     PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4513         }
4514 
4515         /*
4516          * Let's find the place to insert our callbacks.
4517          *
4518          * Even though the list is ordered by txg, it is possible for the
4519          * insertion point to not be the end because our txg may already be
4520          * quiescing at this point and other callbacks in the open txg
4521          * (from other objsets) may have sneaked in.
4522          */
4523         tmp_cb = list_tail(&zcl.zcl_callbacks);
4524         while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4525                 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4526 
4527         /* Add the 3 callbacks to the list */
4528         for (i = 0; i < 3; i++) {
4529                 if (tmp_cb == NULL)
4530                         list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4531                 else
4532                         list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4533                             cb_data[i]);
4534 
4535                 cb_data[i]->zcd_added = B_TRUE;
4536                 VERIFY(!cb_data[i]->zcd_called);
4537 
4538                 tmp_cb = cb_data[i];
4539         }
4540 
4541         (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4542 
4543         dmu_tx_commit(tx);
4544 }
4545 
4546 /* ARGSUSED */
4547 void
4548 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4549 {
4550         zfs_prop_t proplist[] = {
4551                 ZFS_PROP_CHECKSUM,
4552                 ZFS_PROP_COMPRESSION,
4553                 ZFS_PROP_COPIES,
4554                 ZFS_PROP_DEDUP
4555         };
4556 
4557         (void) rw_rdlock(&ztest_name_lock);
4558 
4559         for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4560                 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4561                     ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4562 
4563         (void) rw_unlock(&ztest_name_lock);
4564 }
4565 
4566 /* ARGSUSED */
4567 void
4568 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4569 {
4570         nvlist_t *props = NULL;
4571 
4572         (void) rw_rdlock(&ztest_name_lock);
4573 
4574         (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4575             ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4576 
4577         VERIFY0(spa_prop_get(ztest_spa, &props));
4578 
4579         if (ztest_opts.zo_verbose >= 6)
4580                 dump_nvlist(props, 4);
4581 
4582         nvlist_free(props);
4583 
4584         (void) rw_unlock(&ztest_name_lock);
4585 }
4586 
4587 static int
4588 user_release_one(const char *snapname, const char *holdname)
4589 {
4590         nvlist_t *snaps, *holds;
4591         int error;
4592 
4593         snaps = fnvlist_alloc();
4594         holds = fnvlist_alloc();
4595         fnvlist_add_boolean(holds, holdname);
4596         fnvlist_add_nvlist(snaps, snapname, holds);
4597         fnvlist_free(holds);
4598         error = dsl_dataset_user_release(snaps, NULL);
4599         fnvlist_free(snaps);
4600         return (error);
4601 }
4602 
4603 /*
4604  * Test snapshot hold/release and deferred destroy.
4605  */
4606 void
4607 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4608 {
4609         int error;
4610         objset_t *os = zd->zd_os;
4611         objset_t *origin;
4612         char snapname[100];
4613         char fullname[100];
4614         char clonename[100];
4615         char tag[100];
4616         char osname[MAXNAMELEN];
4617         nvlist_t *holds;
4618 
4619         (void) rw_rdlock(&ztest_name_lock);
4620 
4621         dmu_objset_name(os, osname);
4622 
4623         (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4624         (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4625         (void) snprintf(clonename, sizeof (clonename),
4626             "%s/ch1_%llu", osname, id);
4627         (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4628 
4629         /*
4630          * Clean up from any previous run.
4631          */
4632         error = dsl_destroy_head(clonename);
4633         if (error != ENOENT)
4634                 ASSERT0(error);
4635         error = user_release_one(fullname, tag);
4636         if (error != ESRCH && error != ENOENT)
4637                 ASSERT0(error);
4638         error = dsl_destroy_snapshot(fullname, B_FALSE);
4639         if (error != ENOENT)
4640                 ASSERT0(error);
4641 
4642         /*
4643          * Create snapshot, clone it, mark snap for deferred destroy,
4644          * destroy clone, verify snap was also destroyed.
4645          */
4646         error = dmu_objset_snapshot_one(osname, snapname);
4647         if (error) {
4648                 if (error == ENOSPC) {
4649                         ztest_record_enospc("dmu_objset_snapshot");
4650                         goto out;
4651                 }
4652                 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4653         }
4654 
4655         error = dmu_objset_clone(clonename, fullname);
4656         if (error) {
4657                 if (error == ENOSPC) {
4658                         ztest_record_enospc("dmu_objset_clone");
4659                         goto out;
4660                 }
4661                 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4662         }
4663 
4664         error = dsl_destroy_snapshot(fullname, B_TRUE);
4665         if (error) {
4666                 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4667                     fullname, error);
4668         }
4669 
4670         error = dsl_destroy_head(clonename);
4671         if (error)
4672                 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4673 
4674         error = dmu_objset_hold(fullname, FTAG, &origin);
4675         if (error != ENOENT)
4676                 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4677 
4678         /*
4679          * Create snapshot, add temporary hold, verify that we can't
4680          * destroy a held snapshot, mark for deferred destroy,
4681          * release hold, verify snapshot was destroyed.
4682          */
4683         error = dmu_objset_snapshot_one(osname, snapname);
4684         if (error) {
4685                 if (error == ENOSPC) {
4686                         ztest_record_enospc("dmu_objset_snapshot");
4687                         goto out;
4688                 }
4689                 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4690         }
4691 
4692         holds = fnvlist_alloc();
4693         fnvlist_add_string(holds, fullname, tag);
4694         error = dsl_dataset_user_hold(holds, 0, NULL);
4695         fnvlist_free(holds);
4696 
4697         if (error)
4698                 fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4699 
4700         error = dsl_destroy_snapshot(fullname, B_FALSE);
4701         if (error != EBUSY) {
4702                 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4703                     fullname, error);
4704         }
4705 
4706         error = dsl_destroy_snapshot(fullname, B_TRUE);
4707         if (error) {
4708                 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4709                     fullname, error);
4710         }
4711 
4712         error = user_release_one(fullname, tag);
4713         if (error)
4714                 fatal(0, "user_release_one(%s)", fullname, tag);
4715 
4716         VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4717 
4718 out:
4719         (void) rw_unlock(&ztest_name_lock);
4720 }
4721 
4722 /*
4723  * Inject random faults into the on-disk data.
4724  */
4725 /* ARGSUSED */
4726 void
4727 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4728 {
4729         ztest_shared_t *zs = ztest_shared;
4730         spa_t *spa = ztest_spa;
4731         int fd;
4732         uint64_t offset;
4733         uint64_t leaves;
4734         uint64_t bad = 0x1990c0ffeedecade;
4735         uint64_t top, leaf;
4736         char path0[MAXPATHLEN];
4737         char pathrand[MAXPATHLEN];
4738         size_t fsize;
4739         int bshift = SPA_MAXBLOCKSHIFT + 2;     /* don't scrog all labels */
4740         int iters = 1000;
4741         int maxfaults;
4742         int mirror_save;
4743         vdev_t *vd0 = NULL;
4744         uint64_t guid0 = 0;
4745         boolean_t islog = B_FALSE;
4746 
4747         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4748         maxfaults = MAXFAULTS();
4749         leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4750         mirror_save = zs->zs_mirrors;
4751         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4752 
4753         ASSERT(leaves >= 1);
4754 
4755         /*
4756          * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4757          */
4758         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4759 
4760         if (ztest_random(2) == 0) {
4761                 /*
4762                  * Inject errors on a normal data device or slog device.
4763                  */
4764                 top = ztest_random_vdev_top(spa, B_TRUE);
4765                 leaf = ztest_random(leaves) + zs->zs_splits;
4766 
4767                 /*
4768                  * Generate paths to the first leaf in this top-level vdev,
4769                  * and to the random leaf we selected.  We'll induce transient
4770                  * write failures and random online/offline activity on leaf 0,
4771                  * and we'll write random garbage to the randomly chosen leaf.
4772                  */
4773                 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4774                     ztest_opts.zo_dir, ztest_opts.zo_pool,
4775                     top * leaves + zs->zs_splits);
4776                 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4777                     ztest_opts.zo_dir, ztest_opts.zo_pool,
4778                     top * leaves + leaf);
4779 
4780                 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4781                 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4782                         islog = B_TRUE;
4783 
4784                 if (vd0 != NULL && maxfaults != 1) {
4785                         /*
4786                          * Make vd0 explicitly claim to be unreadable,
4787                          * or unwriteable, or reach behind its back
4788                          * and close the underlying fd.  We can do this if
4789                          * maxfaults == 0 because we'll fail and reexecute,
4790                          * and we can do it if maxfaults >= 2 because we'll
4791                          * have enough redundancy.  If maxfaults == 1, the
4792                          * combination of this with injection of random data
4793                          * corruption below exceeds the pool's fault tolerance.
4794                          */
4795                         vdev_file_t *vf = vd0->vdev_tsd;
4796 
4797                         if (vf != NULL && ztest_random(3) == 0) {
4798                                 (void) close(vf->vf_vnode->v_fd);
4799                                 vf->vf_vnode->v_fd = -1;
4800                         } else if (ztest_random(2) == 0) {
4801                                 vd0->vdev_cant_read = B_TRUE;
4802                         } else {
4803                                 vd0->vdev_cant_write = B_TRUE;
4804                         }
4805                         guid0 = vd0->vdev_guid;
4806                 }
4807         } else {
4808                 /*
4809                  * Inject errors on an l2cache device.
4810                  */
4811                 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4812 
4813                 if (sav->sav_count == 0) {
4814                         spa_config_exit(spa, SCL_STATE, FTAG);
4815                         return;
4816                 }
4817                 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4818                 guid0 = vd0->vdev_guid;
4819                 (void) strcpy(path0, vd0->vdev_path);
4820                 (void) strcpy(pathrand, vd0->vdev_path);
4821 
4822                 leaf = 0;
4823                 leaves = 1;
4824                 maxfaults = INT_MAX;    /* no limit on cache devices */
4825         }
4826 
4827         spa_config_exit(spa, SCL_STATE, FTAG);
4828 
4829         /*
4830          * If we can tolerate two or more faults, or we're dealing
4831          * with a slog, randomly online/offline vd0.
4832          */
4833         if ((maxfaults >= 2 || islog) && guid0 != 0) {
4834                 if (ztest_random(10) < 6) {
4835                         int flags = (ztest_random(2) == 0 ?
4836                             ZFS_OFFLINE_TEMPORARY : 0);
4837 
4838                         /*
4839                          * We have to grab the zs_name_lock as writer to
4840                          * prevent a race between offlining a slog and
4841                          * destroying a dataset. Offlining the slog will
4842                          * grab a reference on the dataset which may cause
4843                          * dmu_objset_destroy() to fail with EBUSY thus
4844                          * leaving the dataset in an inconsistent state.
4845                          */
4846                         if (islog)
4847                                 (void) rw_wrlock(&ztest_name_lock);
4848 
4849                         VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4850 
4851                         if (islog)
4852                                 (void) rw_unlock(&ztest_name_lock);
4853                 } else {
4854                         /*
4855                          * Ideally we would like to be able to randomly
4856                          * call vdev_[on|off]line without holding locks
4857                          * to force unpredictable failures but the side
4858                          * effects of vdev_[on|off]line prevent us from
4859                          * doing so. We grab the ztest_vdev_lock here to
4860                          * prevent a race between injection testing and
4861                          * aux_vdev removal.
4862                          */
4863                         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4864                         (void) vdev_online(spa, guid0, 0, NULL);
4865                         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4866                 }
4867         }
4868 
4869         if (maxfaults == 0)
4870                 return;
4871 
4872         /*
4873          * We have at least single-fault tolerance, so inject data corruption.
4874          */
4875         fd = open(pathrand, O_RDWR);
4876 
4877         if (fd == -1)   /* we hit a gap in the device namespace */
4878                 return;
4879 
4880         fsize = lseek(fd, 0, SEEK_END);
4881 
4882         while (--iters != 0) {
4883                 offset = ztest_random(fsize / (leaves << bshift)) *
4884                     (leaves << bshift) + (leaf << bshift) +
4885                     (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4886 
4887                 if (offset >= fsize)
4888                         continue;
4889 
4890                 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4891                 if (mirror_save != zs->zs_mirrors) {
4892                         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4893                         (void) close(fd);
4894                         return;
4895                 }
4896 
4897                 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4898                         fatal(1, "can't inject bad word at 0x%llx in %s",
4899                             offset, pathrand);
4900 
4901                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4902 
4903                 if (ztest_opts.zo_verbose >= 7)
4904                         (void) printf("injected bad word into %s,"
4905                             " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4906         }
4907 
4908         (void) close(fd);
4909 }
4910 
4911 /*
4912  * Verify that DDT repair works as expected.
4913  */
4914 void
4915 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4916 {
4917         ztest_shared_t *zs = ztest_shared;
4918         spa_t *spa = ztest_spa;
4919         objset_t *os = zd->zd_os;
4920         ztest_od_t od[1];
4921         uint64_t object, blocksize, txg, pattern, psize;
4922         enum zio_checksum checksum = spa_dedup_checksum(spa);
4923         dmu_buf_t *db;
4924         dmu_tx_t *tx;
4925         void *buf;
4926         blkptr_t blk;
4927         int copies = 2 * ZIO_DEDUPDITTO_MIN;
4928 
4929         blocksize = ztest_random_blocksize();
4930         blocksize = MIN(blocksize, 2048);       /* because we write so many */
4931 
4932         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4933 
4934         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4935                 return;
4936 
4937         /*
4938          * Take the name lock as writer to prevent anyone else from changing
4939          * the pool and dataset properies we need to maintain during this test.
4940          */
4941         (void) rw_wrlock(&ztest_name_lock);
4942 
4943         if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4944             B_FALSE) != 0 ||
4945             ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4946             B_FALSE) != 0) {
4947                 (void) rw_unlock(&ztest_name_lock);
4948                 return;
4949         }
4950 
4951         object = od[0].od_object;
4952         blocksize = od[0].od_blocksize;
4953         pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4954 
4955         ASSERT(object != 0);
4956 
4957         tx = dmu_tx_create(os);
4958         dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4959         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4960         if (txg == 0) {
4961                 (void) rw_unlock(&ztest_name_lock);
4962                 return;
4963         }
4964 
4965         /*
4966          * Write all the copies of our block.
4967          */
4968         for (int i = 0; i < copies; i++) {
4969                 uint64_t offset = i * blocksize;
4970                 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
4971                     DMU_READ_NO_PREFETCH);
4972                 if (error != 0) {
4973                         fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
4974                             os, (long long)object, (long long) offset, error);
4975                 }
4976                 ASSERT(db->db_offset == offset);
4977                 ASSERT(db->db_size == blocksize);
4978                 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4979                     ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4980                 dmu_buf_will_fill(db, tx);
4981                 ztest_pattern_set(db->db_data, db->db_size, pattern);
4982                 dmu_buf_rele(db, FTAG);
4983         }
4984 
4985         dmu_tx_commit(tx);
4986         txg_wait_synced(spa_get_dsl(spa), txg);
4987 
4988         /*
4989          * Find out what block we got.
4990          */
4991         VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
4992             DMU_READ_NO_PREFETCH));
4993         blk = *((dmu_buf_impl_t *)db)->db_blkptr;
4994         dmu_buf_rele(db, FTAG);
4995 
4996         /*
4997          * Damage the block.  Dedup-ditto will save us when we read it later.
4998          */
4999         psize = BP_GET_PSIZE(&blk);
5000         buf = zio_buf_alloc(psize);
5001         ztest_pattern_set(buf, psize, ~pattern);
5002 
5003         (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5004             buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5005             ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5006 
5007         zio_buf_free(buf, psize);
5008 
5009         (void) rw_unlock(&ztest_name_lock);
5010 }
5011 
5012 /*
5013  * Scrub the pool.
5014  */
5015 /* ARGSUSED */
5016 void
5017 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5018 {
5019         spa_t *spa = ztest_spa;
5020 
5021         (void) spa_scan(spa, POOL_SCAN_SCRUB);
5022         (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5023         (void) spa_scan(spa, POOL_SCAN_SCRUB);
5024 }
5025 
5026 /*
5027  * Change the guid for the pool.
5028  */
5029 /* ARGSUSED */
5030 void
5031 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5032 {
5033         spa_t *spa = ztest_spa;
5034         uint64_t orig, load;
5035         int error;
5036 
5037         orig = spa_guid(spa);
5038         load = spa_load_guid(spa);
5039 
5040         (void) rw_wrlock(&ztest_name_lock);
5041         error = spa_change_guid(spa);
5042         (void) rw_unlock(&ztest_name_lock);
5043 
5044         if (error != 0)
5045                 return;
5046 
5047         if (ztest_opts.zo_verbose >= 4) {
5048                 (void) printf("Changed guid old %llu -> %llu\n",
5049                     (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5050         }
5051 
5052         VERIFY3U(orig, !=, spa_guid(spa));
5053         VERIFY3U(load, ==, spa_load_guid(spa));
5054 }
5055 
5056 /*
5057  * Rename the pool to a different name and then rename it back.
5058  */
5059 /* ARGSUSED */
5060 void
5061 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5062 {
5063         char *oldname, *newname;
5064         spa_t *spa;
5065 
5066         (void) rw_wrlock(&ztest_name_lock);
5067 
5068         oldname = ztest_opts.zo_pool;
5069         newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5070         (void) strcpy(newname, oldname);
5071         (void) strcat(newname, "_tmp");
5072 
5073         /*
5074          * Do the rename
5075          */
5076         VERIFY3U(0, ==, spa_rename(oldname, newname));
5077 
5078         /*
5079          * Try to open it under the old name, which shouldn't exist
5080          */
5081         VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5082 
5083         /*
5084          * Open it under the new name and make sure it's still the same spa_t.
5085          */
5086         VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5087 
5088         ASSERT(spa == ztest_spa);
5089         spa_close(spa, FTAG);
5090 
5091         /*
5092          * Rename it back to the original
5093          */
5094         VERIFY3U(0, ==, spa_rename(newname, oldname));
5095 
5096         /*
5097          * Make sure it can still be opened
5098          */
5099         VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5100 
5101         ASSERT(spa == ztest_spa);
5102         spa_close(spa, FTAG);
5103 
5104         umem_free(newname, strlen(newname) + 1);
5105 
5106         (void) rw_unlock(&ztest_name_lock);
5107 }
5108 
5109 /*
5110  * Verify pool integrity by running zdb.
5111  */
5112 static void
5113 ztest_run_zdb(char *pool)
5114 {
5115         int status;
5116         char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5117         char zbuf[1024];
5118         char *bin;
5119         char *ztest;
5120         char *isa;
5121         int isalen;
5122         FILE *fp;
5123 
5124         (void) realpath(getexecname(), zdb);
5125 
5126         /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5127         bin = strstr(zdb, "/usr/bin/");
5128         ztest = strstr(bin, "/ztest");
5129         isa = bin + 8;
5130         isalen = ztest - isa;
5131         isa = strdup(isa);
5132         /* LINTED */
5133         (void) sprintf(bin,
5134             "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
5135             isalen,
5136             isa,
5137             ztest_opts.zo_verbose >= 3 ? "s" : "",
5138             ztest_opts.zo_verbose >= 4 ? "v" : "",
5139             spa_config_path,
5140             pool);
5141         free(isa);
5142 
5143         if (ztest_opts.zo_verbose >= 5)
5144                 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5145 
5146         fp = popen(zdb, "r");
5147 
5148         while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5149                 if (ztest_opts.zo_verbose >= 3)
5150                         (void) printf("%s", zbuf);
5151 
5152         status = pclose(fp);
5153 
5154         if (status == 0)
5155                 return;
5156 
5157         ztest_dump_core = 0;
5158         if (WIFEXITED(status))
5159                 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5160         else
5161                 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5162 }
5163 
5164 static void
5165 ztest_walk_pool_directory(char *header)
5166 {
5167         spa_t *spa = NULL;
5168 
5169         if (ztest_opts.zo_verbose >= 6)
5170                 (void) printf("%s\n", header);
5171 
5172         mutex_enter(&spa_namespace_lock);
5173         while ((spa = spa_next(spa)) != NULL)
5174                 if (ztest_opts.zo_verbose >= 6)
5175                         (void) printf("\t%s\n", spa_name(spa));
5176         mutex_exit(&spa_namespace_lock);
5177 }
5178 
5179 static void
5180 ztest_spa_import_export(char *oldname, char *newname)
5181 {
5182         nvlist_t *config, *newconfig;
5183         uint64_t pool_guid;
5184         spa_t *spa;
5185         int error;
5186 
5187         if (ztest_opts.zo_verbose >= 4) {
5188                 (void) printf("import/export: old = %s, new = %s\n",
5189                     oldname, newname);
5190         }
5191 
5192         /*
5193          * Clean up from previous runs.
5194          */
5195         (void) spa_destroy(newname);
5196 
5197         /*
5198          * Get the pool's configuration and guid.
5199          */
5200         VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5201 
5202         /*
5203          * Kick off a scrub to tickle scrub/export races.
5204          */
5205         if (ztest_random(2) == 0)
5206                 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5207 
5208         pool_guid = spa_guid(spa);
5209         spa_close(spa, FTAG);
5210 
5211         ztest_walk_pool_directory("pools before export");
5212 
5213         /*
5214          * Export it.
5215          */
5216         VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5217 
5218         ztest_walk_pool_directory("pools after export");
5219 
5220         /*
5221          * Try to import it.
5222          */
5223         newconfig = spa_tryimport(config);
5224         ASSERT(newconfig != NULL);
5225         nvlist_free(newconfig);
5226 
5227         /*
5228          * Import it under the new name.
5229          */
5230         error = spa_import(newname, config, NULL, 0);
5231         if (error != 0) {
5232                 dump_nvlist(config, 0);
5233                 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5234                     oldname, newname, error);
5235         }
5236 
5237         ztest_walk_pool_directory("pools after import");
5238 
5239         /*
5240          * Try to import it again -- should fail with EEXIST.
5241          */
5242         VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5243 
5244         /*
5245          * Try to import it under a different name -- should fail with EEXIST.
5246          */
5247         VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5248 
5249         /*
5250          * Verify that the pool is no longer visible under the old name.
5251          */
5252         VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5253 
5254         /*
5255          * Verify that we can open and close the pool using the new name.
5256          */
5257         VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5258         ASSERT(pool_guid == spa_guid(spa));
5259         spa_close(spa, FTAG);
5260 
5261         nvlist_free(config);
5262 }
5263 
5264 static void
5265 ztest_resume(spa_t *spa)
5266 {
5267         if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5268                 (void) printf("resuming from suspended state\n");
5269         spa_vdev_state_enter(spa, SCL_NONE);
5270         vdev_clear(spa, NULL);
5271         (void) spa_vdev_state_exit(spa, NULL, 0);
5272         (void) zio_resume(spa);
5273 }
5274 
5275 static void *
5276 ztest_resume_thread(void *arg)
5277 {
5278         spa_t *spa = arg;
5279 
5280         while (!ztest_exiting) {
5281                 if (spa_suspended(spa))
5282                         ztest_resume(spa);
5283                 (void) poll(NULL, 0, 100);
5284         }
5285         return (NULL);
5286 }
5287 
5288 static void *
5289 ztest_deadman_thread(void *arg)
5290 {
5291         ztest_shared_t *zs = arg;
5292         int grace = 300;
5293         hrtime_t delta;
5294 
5295         delta = (zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + grace;
5296 
5297         (void) poll(NULL, 0, (int)(1000 * delta));
5298 
5299         fatal(0, "failed to complete within %d seconds of deadline", grace);
5300 
5301         return (NULL);
5302 }
5303 
5304 static void
5305 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5306 {
5307         ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5308         ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5309         hrtime_t functime = gethrtime();
5310 
5311         for (int i = 0; i < zi->zi_iters; i++)
5312                 zi->zi_func(zd, id);
5313 
5314         functime = gethrtime() - functime;
5315 
5316         atomic_add_64(&zc->zc_count, 1);
5317         atomic_add_64(&zc->zc_time, functime);
5318 
5319         if (ztest_opts.zo_verbose >= 4) {
5320                 Dl_info dli;
5321                 (void) dladdr((void *)zi->zi_func, &dli);
5322                 (void) printf("%6.2f sec in %s\n",
5323                     (double)functime / NANOSEC, dli.dli_sname);
5324         }
5325 }
5326 
5327 static void *
5328 ztest_thread(void *arg)
5329 {
5330         int rand;
5331         uint64_t id = (uintptr_t)arg;
5332         ztest_shared_t *zs = ztest_shared;
5333         uint64_t call_next;
5334         hrtime_t now;
5335         ztest_info_t *zi;
5336         ztest_shared_callstate_t *zc;
5337 
5338         while ((now = gethrtime()) < zs->zs_thread_stop) {
5339                 /*
5340                  * See if it's time to force a crash.
5341                  */
5342                 if (now > zs->zs_thread_kill)
5343                         ztest_kill(zs);
5344 
5345                 /*
5346                  * If we're getting ENOSPC with some regularity, stop.
5347                  */
5348                 if (zs->zs_enospc_count > 10)
5349                         break;
5350 
5351                 /*
5352                  * Pick a random function to execute.
5353                  */
5354                 rand = ztest_random(ZTEST_FUNCS);
5355                 zi = &ztest_info[rand];
5356                 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5357                 call_next = zc->zc_next;
5358 
5359                 if (now >= call_next &&
5360                     atomic_cas_64(&zc->zc_next, call_next, call_next +
5361                     ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5362                         ztest_execute(rand, zi, id);
5363                 }
5364         }
5365 
5366         return (NULL);
5367 }
5368 
5369 static void
5370 ztest_dataset_name(char *dsname, char *pool, int d)
5371 {
5372         (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5373 }
5374 
5375 static void
5376 ztest_dataset_destroy(int d)
5377 {
5378         char name[MAXNAMELEN];
5379 
5380         ztest_dataset_name(name, ztest_opts.zo_pool, d);
5381 
5382         if (ztest_opts.zo_verbose >= 3)
5383                 (void) printf("Destroying %s to free up space\n", name);
5384 
5385         /*
5386          * Cleanup any non-standard clones and snapshots.  In general,
5387          * ztest thread t operates on dataset (t % zopt_datasets),
5388          * so there may be more than one thing to clean up.
5389          */
5390         for (int t = d; t < ztest_opts.zo_threads;
5391             t += ztest_opts.zo_datasets) {
5392                 ztest_dsl_dataset_cleanup(name, t);
5393         }
5394 
5395         (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5396             DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5397 }
5398 
5399 static void
5400 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5401 {
5402         uint64_t usedobjs, dirobjs, scratch;
5403 
5404         /*
5405          * ZTEST_DIROBJ is the object directory for the entire dataset.
5406          * Therefore, the number of objects in use should equal the
5407          * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5408          * If not, we have an object leak.
5409          *
5410          * Note that we can only check this in ztest_dataset_open(),
5411          * when the open-context and syncing-context values agree.
5412          * That's because zap_count() returns the open-context value,
5413          * while dmu_objset_space() returns the rootbp fill count.
5414          */
5415         VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5416         dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5417         ASSERT3U(dirobjs + 1, ==, usedobjs);
5418 }
5419 
5420 static int
5421 ztest_dataset_open(int d)
5422 {
5423         ztest_ds_t *zd = &ztest_ds[d];
5424         uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5425         objset_t *os;
5426         zilog_t *zilog;
5427         char name[MAXNAMELEN];
5428         int error;
5429 
5430         ztest_dataset_name(name, ztest_opts.zo_pool, d);
5431 
5432         (void) rw_rdlock(&ztest_name_lock);
5433 
5434         error = ztest_dataset_create(name);
5435         if (error == ENOSPC) {
5436                 (void) rw_unlock(&ztest_name_lock);
5437                 ztest_record_enospc(FTAG);
5438                 return (error);
5439         }
5440         ASSERT(error == 0 || error == EEXIST);
5441 
5442         VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5443         (void) rw_unlock(&ztest_name_lock);
5444 
5445         ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5446 
5447         zilog = zd->zd_zilog;
5448 
5449         if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5450             zilog->zl_header->zh_claim_lr_seq < committed_seq)
5451                 fatal(0, "missing log records: claimed %llu < committed %llu",
5452                     zilog->zl_header->zh_claim_lr_seq, committed_seq);
5453 
5454         ztest_dataset_dirobj_verify(zd);
5455 
5456         zil_replay(os, zd, ztest_replay_vector);
5457 
5458         ztest_dataset_dirobj_verify(zd);
5459 
5460         if (ztest_opts.zo_verbose >= 6)
5461                 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5462                     zd->zd_name,
5463                     (u_longlong_t)zilog->zl_parse_blk_count,
5464                     (u_longlong_t)zilog->zl_parse_lr_count,
5465                     (u_longlong_t)zilog->zl_replaying_seq);
5466 
5467         zilog = zil_open(os, ztest_get_data);
5468 
5469         if (zilog->zl_replaying_seq != 0 &&
5470             zilog->zl_replaying_seq < committed_seq)
5471                 fatal(0, "missing log records: replayed %llu < committed %llu",
5472                     zilog->zl_replaying_seq, committed_seq);
5473 
5474         return (0);
5475 }
5476 
5477 static void
5478 ztest_dataset_close(int d)
5479 {
5480         ztest_ds_t *zd = &ztest_ds[d];
5481 
5482         zil_close(zd->zd_zilog);
5483         dmu_objset_disown(zd->zd_os, zd);
5484 
5485         ztest_zd_fini(zd);
5486 }
5487 
5488 /*
5489  * Kick off threads to run tests on all datasets in parallel.
5490  */
5491 static void
5492 ztest_run(ztest_shared_t *zs)
5493 {
5494         thread_t *tid;
5495         spa_t *spa;
5496         objset_t *os;
5497         thread_t resume_tid;
5498         int error;
5499 
5500         ztest_exiting = B_FALSE;
5501 
5502         /*
5503          * Initialize parent/child shared state.
5504          */
5505         VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5506         VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5507 
5508         zs->zs_thread_start = gethrtime();
5509         zs->zs_thread_stop =
5510             zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5511         zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5512         zs->zs_thread_kill = zs->zs_thread_stop;
5513         if (ztest_random(100) < ztest_opts.zo_killrate) {
5514                 zs->zs_thread_kill -=
5515                     ztest_random(ztest_opts.zo_passtime * NANOSEC);
5516         }
5517 
5518         (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5519 
5520         list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5521             offsetof(ztest_cb_data_t, zcd_node));
5522 
5523         /*
5524          * Open our pool.
5525          */
5526         kernel_init(FREAD | FWRITE);
5527         VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5528         spa->spa_debug = B_TRUE;
5529         ztest_spa = spa;
5530 
5531         VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5532             DMU_OST_ANY, B_TRUE, FTAG, &os));
5533         zs->zs_guid = dmu_objset_fsid_guid(os);
5534         dmu_objset_disown(os, FTAG);
5535 
5536         spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5537 
5538         /*
5539          * We don't expect the pool to suspend unless maxfaults == 0,
5540          * in which case ztest_fault_inject() temporarily takes away
5541          * the only valid replica.
5542          */
5543         if (MAXFAULTS() == 0)
5544                 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5545         else
5546                 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5547 
5548         /*
5549          * Create a thread to periodically resume suspended I/O.
5550          */
5551         VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5552             &resume_tid) == 0);
5553 
5554         /*
5555          * Create a deadman thread to abort() if we hang.
5556          */
5557         VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5558             NULL) == 0);
5559 
5560         /*
5561          * Verify that we can safely inquire about about any object,
5562          * whether it's allocated or not.  To make it interesting,
5563          * we probe a 5-wide window around each power of two.
5564          * This hits all edge cases, including zero and the max.
5565          */
5566         for (int t = 0; t < 64; t++) {
5567                 for (int d = -5; d <= 5; d++) {
5568                         error = dmu_object_info(spa->spa_meta_objset,
5569                             (1ULL << t) + d, NULL);
5570                         ASSERT(error == 0 || error == ENOENT ||
5571                             error == EINVAL);
5572                 }
5573         }
5574 
5575         /*
5576          * If we got any ENOSPC errors on the previous run, destroy something.
5577          */
5578         if (zs->zs_enospc_count != 0) {
5579                 int d = ztest_random(ztest_opts.zo_datasets);
5580                 ztest_dataset_destroy(d);
5581         }
5582         zs->zs_enospc_count = 0;
5583 
5584         tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5585             UMEM_NOFAIL);
5586 
5587         if (ztest_opts.zo_verbose >= 4)
5588                 (void) printf("starting main threads...\n");
5589 
5590         /*
5591          * Kick off all the tests that run in parallel.
5592          */
5593         for (int t = 0; t < ztest_opts.zo_threads; t++) {
5594                 if (t < ztest_opts.zo_datasets &&
5595                     ztest_dataset_open(t) != 0)
5596                         return;
5597                 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5598                     THR_BOUND, &tid[t]) == 0);
5599         }
5600 
5601         /*
5602          * Wait for all of the tests to complete.  We go in reverse order
5603          * so we don't close datasets while threads are still using them.
5604          */
5605         for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5606                 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5607                 if (t < ztest_opts.zo_datasets)
5608                         ztest_dataset_close(t);
5609         }
5610 
5611         txg_wait_synced(spa_get_dsl(spa), 0);
5612 
5613         zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5614         zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5615 
5616         umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5617 
5618         /* Kill the resume thread */
5619         ztest_exiting = B_TRUE;
5620         VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5621         ztest_resume(spa);
5622 
5623         /*
5624          * Right before closing the pool, kick off a bunch of async I/O;
5625          * spa_close() should wait for it to complete.
5626          */
5627         for (uint64_t object = 1; object < 50; object++)
5628                 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5629 
5630         spa_close(spa, FTAG);
5631 
5632         /*
5633          * Verify that we can loop over all pools.
5634          */
5635         mutex_enter(&spa_namespace_lock);
5636         for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5637                 if (ztest_opts.zo_verbose > 3)
5638                         (void) printf("spa_next: found %s\n", spa_name(spa));
5639         mutex_exit(&spa_namespace_lock);
5640 
5641         /*
5642          * Verify that we can export the pool and reimport it under a
5643          * different name.
5644          */
5645         if (ztest_random(2) == 0) {
5646                 char name[MAXNAMELEN];
5647                 (void) snprintf(name, MAXNAMELEN, "%s_import",
5648                     ztest_opts.zo_pool);
5649                 ztest_spa_import_export(ztest_opts.zo_pool, name);
5650                 ztest_spa_import_export(name, ztest_opts.zo_pool);
5651         }
5652 
5653         kernel_fini();
5654 
5655         list_destroy(&zcl.zcl_callbacks);
5656 
5657         (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5658 
5659         (void) rwlock_destroy(&ztest_name_lock);
5660         (void) _mutex_destroy(&ztest_vdev_lock);
5661 }
5662 
5663 static void
5664 ztest_freeze(void)
5665 {
5666         ztest_ds_t *zd = &ztest_ds[0];
5667         spa_t *spa;
5668         int numloops = 0;
5669 
5670         if (ztest_opts.zo_verbose >= 3)
5671                 (void) printf("testing spa_freeze()...\n");
5672 
5673         kernel_init(FREAD | FWRITE);
5674         VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5675         VERIFY3U(0, ==, ztest_dataset_open(0));
5676         spa->spa_debug = B_TRUE;
5677         ztest_spa = spa;
5678 
5679         /*
5680          * Force the first log block to be transactionally allocated.
5681          * We have to do this before we freeze the pool -- otherwise
5682          * the log chain won't be anchored.
5683          */
5684         while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5685                 ztest_dmu_object_alloc_free(zd, 0);
5686                 zil_commit(zd->zd_zilog, 0);
5687         }
5688 
5689         txg_wait_synced(spa_get_dsl(spa), 0);
5690 
5691         /*
5692          * Freeze the pool.  This stops spa_sync() from doing anything,
5693          * so that the only way to record changes from now on is the ZIL.
5694          */
5695         spa_freeze(spa);
5696 
5697         /*
5698          * Run tests that generate log records but don't alter the pool config
5699          * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5700          * We do a txg_wait_synced() after each iteration to force the txg
5701          * to increase well beyond the last synced value in the uberblock.
5702          * The ZIL should be OK with that.
5703          */
5704         while (ztest_random(10) != 0 &&
5705             numloops++ < ztest_opts.zo_maxloops) {
5706                 ztest_dmu_write_parallel(zd, 0);
5707                 ztest_dmu_object_alloc_free(zd, 0);
5708                 txg_wait_synced(spa_get_dsl(spa), 0);
5709         }
5710 
5711         /*
5712          * Commit all of the changes we just generated.
5713          */
5714         zil_commit(zd->zd_zilog, 0);
5715         txg_wait_synced(spa_get_dsl(spa), 0);
5716 
5717         /*
5718          * Close our dataset and close the pool.
5719          */
5720         ztest_dataset_close(0);
5721         spa_close(spa, FTAG);
5722         kernel_fini();
5723 
5724         /*
5725          * Open and close the pool and dataset to induce log replay.
5726          */
5727         kernel_init(FREAD | FWRITE);
5728         VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5729         ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5730         VERIFY3U(0, ==, ztest_dataset_open(0));
5731         ztest_dataset_close(0);
5732 
5733         spa->spa_debug = B_TRUE;
5734         ztest_spa = spa;
5735         txg_wait_synced(spa_get_dsl(spa), 0);
5736         ztest_reguid(NULL, 0);
5737 
5738         spa_close(spa, FTAG);
5739         kernel_fini();
5740 }
5741 
5742 void
5743 print_time(hrtime_t t, char *timebuf)
5744 {
5745         hrtime_t s = t / NANOSEC;
5746         hrtime_t m = s / 60;
5747         hrtime_t h = m / 60;
5748         hrtime_t d = h / 24;
5749 
5750         s -= m * 60;
5751         m -= h * 60;
5752         h -= d * 24;
5753 
5754         timebuf[0] = '\0';
5755 
5756         if (d)
5757                 (void) sprintf(timebuf,
5758                     "%llud%02lluh%02llum%02llus", d, h, m, s);
5759         else if (h)
5760                 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5761         else if (m)
5762                 (void) sprintf(timebuf, "%llum%02llus", m, s);
5763         else
5764                 (void) sprintf(timebuf, "%llus", s);
5765 }
5766 
5767 static nvlist_t *
5768 make_random_props()
5769 {
5770         nvlist_t *props;
5771 
5772         VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5773         if (ztest_random(2) == 0)
5774                 return (props);
5775         VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5776 
5777         return (props);
5778 }
5779 
5780 /*
5781  * Create a storage pool with the given name and initial vdev size.
5782  * Then test spa_freeze() functionality.
5783  */
5784 static void
5785 ztest_init(ztest_shared_t *zs)
5786 {
5787         spa_t *spa;
5788         nvlist_t *nvroot, *props;
5789 
5790         VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5791         VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5792 
5793         kernel_init(FREAD | FWRITE);
5794 
5795         /*
5796          * Create the storage pool.
5797          */
5798         (void) spa_destroy(ztest_opts.zo_pool);
5799         ztest_shared->zs_vdev_next_leaf = 0;
5800         zs->zs_splits = 0;
5801         zs->zs_mirrors = ztest_opts.zo_mirrors;
5802         nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5803             0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5804         props = make_random_props();
5805         for (int i = 0; i < SPA_FEATURES; i++) {
5806                 char buf[1024];
5807                 (void) snprintf(buf, sizeof (buf), "feature@%s",
5808                     spa_feature_table[i].fi_uname);
5809                 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5810         }
5811         VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5812         nvlist_free(nvroot);
5813 
5814         VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5815         zs->zs_metaslab_sz =
5816             1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5817 
5818         spa_close(spa, FTAG);
5819 
5820         kernel_fini();
5821 
5822         ztest_run_zdb(ztest_opts.zo_pool);
5823 
5824         ztest_freeze();
5825 
5826         ztest_run_zdb(ztest_opts.zo_pool);
5827 
5828         (void) rwlock_destroy(&ztest_name_lock);
5829         (void) _mutex_destroy(&ztest_vdev_lock);
5830 }
5831 
5832 static void
5833 setup_data_fd(void)
5834 {
5835         static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5836 
5837         ztest_fd_data = mkstemp(ztest_name_data);
5838         ASSERT3S(ztest_fd_data, >=, 0);
5839         (void) unlink(ztest_name_data);
5840 }
5841 
5842 
5843 static int
5844 shared_data_size(ztest_shared_hdr_t *hdr)
5845 {
5846         int size;
5847 
5848         size = hdr->zh_hdr_size;
5849         size += hdr->zh_opts_size;
5850         size += hdr->zh_size;
5851         size += hdr->zh_stats_size * hdr->zh_stats_count;
5852         size += hdr->zh_ds_size * hdr->zh_ds_count;
5853 
5854         return (size);
5855 }
5856 
5857 static void
5858 setup_hdr(void)
5859 {
5860         int size;
5861         ztest_shared_hdr_t *hdr;
5862 
5863         hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5864             PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5865         ASSERT(hdr != MAP_FAILED);
5866 
5867         VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5868 
5869         hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5870         hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5871         hdr->zh_size = sizeof (ztest_shared_t);
5872         hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5873         hdr->zh_stats_count = ZTEST_FUNCS;
5874         hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5875         hdr->zh_ds_count = ztest_opts.zo_datasets;
5876 
5877         size = shared_data_size(hdr);
5878         VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5879 
5880         (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5881 }
5882 
5883 static void
5884 setup_data(void)
5885 {
5886         int size, offset;
5887         ztest_shared_hdr_t *hdr;
5888         uint8_t *buf;
5889 
5890         hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5891             PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5892         ASSERT(hdr != MAP_FAILED);
5893 
5894         size = shared_data_size(hdr);
5895 
5896         (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5897         hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5898             PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5899         ASSERT(hdr != MAP_FAILED);
5900         buf = (uint8_t *)hdr;
5901 
5902         offset = hdr->zh_hdr_size;
5903         ztest_shared_opts = (void *)&buf[offset];
5904         offset += hdr->zh_opts_size;
5905         ztest_shared = (void *)&buf[offset];
5906         offset += hdr->zh_size;
5907         ztest_shared_callstate = (void *)&buf[offset];
5908         offset += hdr->zh_stats_size * hdr->zh_stats_count;
5909         ztest_shared_ds = (void *)&buf[offset];
5910 }
5911 
5912 static boolean_t
5913 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5914 {
5915         pid_t pid;
5916         int status;
5917         char *cmdbuf = NULL;
5918 
5919         pid = fork();
5920 
5921         if (cmd == NULL) {
5922                 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5923                 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
5924                 cmd = cmdbuf;
5925         }
5926 
5927         if (pid == -1)
5928                 fatal(1, "fork failed");
5929 
5930         if (pid == 0) { /* child */
5931                 char *emptyargv[2] = { cmd, NULL };
5932                 char fd_data_str[12];
5933 
5934                 struct rlimit rl = { 1024, 1024 };
5935                 (void) setrlimit(RLIMIT_NOFILE, &rl);
5936 
5937                 (void) close(ztest_fd_rand);
5938                 VERIFY3U(11, >=,
5939                     snprintf(fd_data_str, 12, "%d", ztest_fd_data));
5940                 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
5941 
5942                 (void) enable_extended_FILE_stdio(-1, -1);
5943                 if (libpath != NULL)
5944                         VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5945                 (void) execv(cmd, emptyargv);
5946                 ztest_dump_core = B_FALSE;
5947                 fatal(B_TRUE, "exec failed: %s", cmd);
5948         }
5949 
5950         if (cmdbuf != NULL) {
5951                 umem_free(cmdbuf, MAXPATHLEN);
5952                 cmd = NULL;
5953         }
5954 
5955         while (waitpid(pid, &status, 0) != pid)
5956                 continue;
5957         if (statusp != NULL)
5958                 *statusp = status;
5959 
5960         if (WIFEXITED(status)) {
5961                 if (WEXITSTATUS(status) != 0) {
5962                         (void) fprintf(stderr, "child exited with code %d\n",
5963                             WEXITSTATUS(status));
5964                         exit(2);
5965                 }
5966                 return (B_FALSE);
5967         } else if (WIFSIGNALED(status)) {
5968                 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
5969                         (void) fprintf(stderr, "child died with signal %d\n",
5970                             WTERMSIG(status));
5971                         exit(3);
5972                 }
5973                 return (B_TRUE);
5974         } else {
5975                 (void) fprintf(stderr, "something strange happened to child\n");
5976                 exit(4);
5977                 /* NOTREACHED */
5978         }
5979 }
5980 
5981 static void
5982 ztest_run_init(void)
5983 {
5984         ztest_shared_t *zs = ztest_shared;
5985 
5986         ASSERT(ztest_opts.zo_init != 0);
5987 
5988         /*
5989          * Blow away any existing copy of zpool.cache
5990          */
5991         (void) remove(spa_config_path);
5992 
5993         /*
5994          * Create and initialize our storage pool.
5995          */
5996         for (int i = 1; i <= ztest_opts.zo_init; i++) {
5997                 bzero(zs, sizeof (ztest_shared_t));
5998                 if (ztest_opts.zo_verbose >= 3 &&
5999                     ztest_opts.zo_init != 1) {
6000                         (void) printf("ztest_init(), pass %d\n", i);
6001                 }
6002                 ztest_init(zs);
6003         }
6004 }
6005 
6006 int
6007 main(int argc, char **argv)
6008 {
6009         int kills = 0;
6010         int iters = 0;
6011         int older = 0;
6012         int newer = 0;
6013         ztest_shared_t *zs;
6014         ztest_info_t *zi;
6015         ztest_shared_callstate_t *zc;
6016         char timebuf[100];
6017         char numbuf[6];
6018         spa_t *spa;
6019         char *cmd;
6020         boolean_t hasalt;
6021         char *fd_data_str = getenv("ZTEST_FD_DATA");
6022 
6023         (void) setvbuf(stdout, NULL, _IOLBF, 0);
6024 
6025         dprintf_setup(&argc, argv);
6026 
6027         ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6028         ASSERT3S(ztest_fd_rand, >=, 0);
6029 
6030         if (!fd_data_str) {
6031                 process_options(argc, argv);
6032 
6033                 setup_data_fd();
6034                 setup_hdr();
6035                 setup_data();
6036                 bcopy(&ztest_opts, ztest_shared_opts,
6037                     sizeof (*ztest_shared_opts));
6038         } else {
6039                 ztest_fd_data = atoi(fd_data_str);
6040                 setup_data();
6041                 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6042         }
6043         ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6044 
6045         /* Override location of zpool.cache */
6046         VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6047             ztest_opts.zo_dir), !=, -1);
6048 
6049         ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6050             UMEM_NOFAIL);
6051         zs = ztest_shared;
6052 
6053         if (fd_data_str) {
6054                 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6055                 metaslab_df_alloc_threshold =
6056                     zs->zs_metaslab_df_alloc_threshold;
6057 
6058                 if (zs->zs_do_init)
6059                         ztest_run_init();
6060                 else
6061                         ztest_run(zs);
6062                 exit(0);
6063         }
6064 
6065         hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6066 
6067         if (ztest_opts.zo_verbose >= 1) {
6068                 (void) printf("%llu vdevs, %d datasets, %d threads,"
6069                     " %llu seconds...\n",
6070                     (u_longlong_t)ztest_opts.zo_vdevs,
6071                     ztest_opts.zo_datasets,
6072                     ztest_opts.zo_threads,
6073                     (u_longlong_t)ztest_opts.zo_time);
6074         }
6075 
6076         cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6077         (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6078 
6079         zs->zs_do_init = B_TRUE;
6080         if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6081                 if (ztest_opts.zo_verbose >= 1) {
6082                         (void) printf("Executing older ztest for "
6083                             "initialization: %s\n", ztest_opts.zo_alt_ztest);
6084                 }
6085                 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6086                     ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6087         } else {
6088                 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6089         }
6090         zs->zs_do_init = B_FALSE;
6091 
6092         zs->zs_proc_start = gethrtime();
6093         zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6094 
6095         for (int f = 0; f < ZTEST_FUNCS; f++) {
6096                 zi = &ztest_info[f];
6097                 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6098                 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6099                         zc->zc_next = UINT64_MAX;
6100                 else
6101                         zc->zc_next = zs->zs_proc_start +
6102                             ztest_random(2 * zi->zi_interval[0] + 1);
6103         }
6104 
6105         /*
6106          * Run the tests in a loop.  These tests include fault injection
6107          * to verify that self-healing data works, and forced crashes
6108          * to verify that we never lose on-disk consistency.
6109          */
6110         while (gethrtime() < zs->zs_proc_stop) {
6111                 int status;
6112                 boolean_t killed;
6113 
6114                 /*
6115                  * Initialize the workload counters for each function.
6116                  */
6117                 for (int f = 0; f < ZTEST_FUNCS; f++) {
6118                         zc = ZTEST_GET_SHARED_CALLSTATE(f);
6119                         zc->zc_count = 0;
6120                         zc->zc_time = 0;
6121                 }
6122 
6123                 /* Set the allocation switch size */
6124                 zs->zs_metaslab_df_alloc_threshold =
6125                     ztest_random(zs->zs_metaslab_sz / 4) + 1;
6126 
6127                 if (!hasalt || ztest_random(2) == 0) {
6128                         if (hasalt && ztest_opts.zo_verbose >= 1) {
6129                                 (void) printf("Executing newer ztest: %s\n",
6130                                     cmd);
6131                         }
6132                         newer++;
6133                         killed = exec_child(cmd, NULL, B_TRUE, &status);
6134                 } else {
6135                         if (hasalt && ztest_opts.zo_verbose >= 1) {
6136                                 (void) printf("Executing older ztest: %s\n",
6137                                     ztest_opts.zo_alt_ztest);
6138                         }
6139                         older++;
6140                         killed = exec_child(ztest_opts.zo_alt_ztest,
6141                             ztest_opts.zo_alt_libpath, B_TRUE, &status);
6142                 }
6143 
6144                 if (killed)
6145                         kills++;
6146                 iters++;
6147 
6148                 if (ztest_opts.zo_verbose >= 1) {
6149                         hrtime_t now = gethrtime();
6150 
6151                         now = MIN(now, zs->zs_proc_stop);
6152                         print_time(zs->zs_proc_stop - now, timebuf);
6153                         nicenum(zs->zs_space, numbuf);
6154 
6155                         (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6156                             "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6157                             iters,
6158                             WIFEXITED(status) ? "Complete" : "SIGKILL",
6159                             (u_longlong_t)zs->zs_enospc_count,
6160                             100.0 * zs->zs_alloc / zs->zs_space,
6161                             numbuf,
6162                             100.0 * (now - zs->zs_proc_start) /
6163                             (ztest_opts.zo_time * NANOSEC), timebuf);
6164                 }
6165 
6166                 if (ztest_opts.zo_verbose >= 2) {
6167                         (void) printf("\nWorkload summary:\n\n");
6168                         (void) printf("%7s %9s   %s\n",
6169                             "Calls", "Time", "Function");
6170                         (void) printf("%7s %9s   %s\n",
6171                             "-----", "----", "--------");
6172                         for (int f = 0; f < ZTEST_FUNCS; f++) {
6173                                 Dl_info dli;
6174 
6175                                 zi = &ztest_info[f];
6176                                 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6177                                 print_time(zc->zc_time, timebuf);
6178                                 (void) dladdr((void *)zi->zi_func, &dli);
6179                                 (void) printf("%7llu %9s   %s\n",
6180                                     (u_longlong_t)zc->zc_count, timebuf,
6181                                     dli.dli_sname);
6182                         }
6183                         (void) printf("\n");
6184                 }
6185 
6186                 /*
6187                  * It's possible that we killed a child during a rename test,
6188                  * in which case we'll have a 'ztest_tmp' pool lying around
6189                  * instead of 'ztest'.  Do a blind rename in case this happened.
6190                  */
6191                 kernel_init(FREAD);
6192                 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6193                         spa_close(spa, FTAG);
6194                 } else {
6195                         char tmpname[MAXNAMELEN];
6196                         kernel_fini();
6197                         kernel_init(FREAD | FWRITE);
6198                         (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6199                             ztest_opts.zo_pool);
6200                         (void) spa_rename(tmpname, ztest_opts.zo_pool);
6201                 }
6202                 kernel_fini();
6203 
6204                 ztest_run_zdb(ztest_opts.zo_pool);
6205         }
6206 
6207         if (ztest_opts.zo_verbose >= 1) {
6208                 if (hasalt) {
6209                         (void) printf("%d runs of older ztest: %s\n", older,
6210                             ztest_opts.zo_alt_ztest);
6211                         (void) printf("%d runs of newer ztest: %s\n", newer,
6212                             cmd);
6213                 }
6214                 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6215                     kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6216         }
6217 
6218         umem_free(cmd, MAXNAMELEN);
6219 
6220         return (0);
6221 }