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