1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2012 by Delphix. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  25  */
  26 
  27 /*
  28  * The objective of this program is to provide a DMU/ZAP/SPA stress test
  29  * that runs entirely in userland, is easy to use, and easy to extend.
  30  *
  31  * The overall design of the ztest program is as follows:
  32  *
  33  * (1) For each major functional area (e.g. adding vdevs to a pool,
  34  *     creating and destroying datasets, reading and writing objects, etc)
  35  *     we have a simple routine to test that functionality.  These
  36  *     individual routines do not have to do anything "stressful".
  37  *
  38  * (2) We turn these simple functionality tests into a stress test by
  39  *     running them all in parallel, with as many threads as desired,
  40  *     and spread across as many datasets, objects, and vdevs as desired.
  41  *
  42  * (3) While all this is happening, we inject faults into the pool to
  43  *     verify that self-healing data really works.
  44  *
  45  * (4) Every time we open a dataset, we change its checksum and compression
  46  *     functions.  Thus even individual objects vary from block to block
  47  *     in which checksum they use and whether they're compressed.
  48  *
  49  * (5) To verify that we never lose on-disk consistency after a crash,
  50  *     we run the entire test in a child of the main process.
  51  *     At random times, the child self-immolates with a SIGKILL.
  52  *     This is the software equivalent of pulling the power cord.
  53  *     The parent then runs the test again, using the existing
  54  *     storage pool, as many times as desired. If backwards compatability
  55  *     testing is enabled ztest will sometimes run the "older" version
  56  *     of ztest after a SIGKILL.
  57  *
  58  * (6) To verify that we don't have future leaks or temporal incursions,
  59  *     many of the functional tests record the transaction group number
  60  *     as part of their data.  When reading old data, they verify that
  61  *     the transaction group number is less than the current, open txg.
  62  *     If you add a new test, please do this if applicable.
  63  *
  64  * When run with no arguments, ztest runs for about five minutes and
  65  * produces no output if successful.  To get a little bit of information,
  66  * specify -V.  To get more information, specify -VV, and so on.
  67  *
  68  * To turn this into an overnight stress test, use -T to specify run time.
  69  *
  70  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
  71  * to increase the pool capacity, fanout, and overall stress level.
  72  *
  73  * Use the -k option to set the desired frequency of kills.
  74  *
  75  * When ztest invokes itself it passes all relevant information through a
  76  * temporary file which is mmap-ed in the child process. This allows shared
  77  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
  78  * stored at offset 0 of this file and contains information on the size and
  79  * number of shared structures in the file. The information stored in this file
  80  * must remain backwards compatible with older versions of ztest so that
  81  * ztest can invoke them during backwards compatibility testing (-B).
  82  */
  83 
  84 #include <sys/zfs_context.h>
  85 #include <sys/spa.h>
  86 #include <sys/dmu.h>
  87 #include <sys/txg.h>
  88 #include <sys/dbuf.h>
  89 #include <sys/zap.h>
  90 #include <sys/dmu_objset.h>
  91 #include <sys/poll.h>
  92 #include <sys/stat.h>
  93 #include <sys/time.h>
  94 #include <sys/wait.h>
  95 #include <sys/mman.h>
  96 #include <sys/resource.h>
  97 #include <sys/zio.h>
  98 #include <sys/zil.h>
  99 #include <sys/zil_impl.h>
 100 #include <sys/vdev_impl.h>
 101 #include <sys/vdev_file.h>
 102 #include <sys/spa_impl.h>
 103 #include <sys/metaslab_impl.h>
 104 #include <sys/dsl_prop.h>
 105 #include <sys/dsl_dataset.h>
 106 #include <sys/dsl_scan.h>
 107 #include <sys/zio_checksum.h>
 108 #include <sys/refcount.h>
 109 #include <sys/zfeature.h>
 110 #include <stdio.h>
 111 #include <stdio_ext.h>
 112 #include <stdlib.h>
 113 #include <unistd.h>
 114 #include <signal.h>
 115 #include <umem.h>
 116 #include <dlfcn.h>
 117 #include <ctype.h>
 118 #include <math.h>
 119 #include <sys/fs/zfs.h>
 120 #include <libnvpair.h>
 121 
 122 #define ZTEST_FD_DATA 3
 123 #define ZTEST_FD_RAND 4
 124 
 125 typedef struct ztest_shared_hdr {
 126         uint64_t        zh_hdr_size;
 127         uint64_t        zh_opts_size;
 128         uint64_t        zh_size;
 129         uint64_t        zh_stats_size;
 130         uint64_t        zh_stats_count;
 131         uint64_t        zh_ds_size;
 132         uint64_t        zh_ds_count;
 133 } ztest_shared_hdr_t;
 134 
 135 static ztest_shared_hdr_t *ztest_shared_hdr;
 136 
 137 typedef struct ztest_shared_opts {
 138         char zo_pool[MAXNAMELEN];
 139         char zo_dir[MAXNAMELEN];
 140         char zo_alt_ztest[MAXNAMELEN];
 141         char zo_alt_libpath[MAXNAMELEN];
 142         uint64_t zo_vdevs;
 143         uint64_t zo_vdevtime;
 144         size_t zo_vdev_size;
 145         int zo_ashift;
 146         int zo_mirrors;
 147         int zo_raidz;
 148         int zo_raidz_parity;
 149         int zo_datasets;
 150         int zo_threads;
 151         uint64_t zo_passtime;
 152         uint64_t zo_killrate;
 153         int zo_verbose;
 154         int zo_init;
 155         uint64_t zo_time;
 156         uint64_t zo_maxloops;
 157         uint64_t zo_metaslab_gang_bang;
 158 } ztest_shared_opts_t;
 159 
 160 static const ztest_shared_opts_t ztest_opts_defaults = {
 161         .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
 162         .zo_dir = { '/', 't', 'm', 'p', '\0' },
 163         .zo_alt_ztest = { '\0' },
 164         .zo_alt_libpath = { '\0' },
 165         .zo_vdevs = 5,
 166         .zo_ashift = SPA_MINBLOCKSHIFT,
 167         .zo_mirrors = 2,
 168         .zo_raidz = 4,
 169         .zo_raidz_parity = 1,
 170         .zo_vdev_size = SPA_MINDEVSIZE,
 171         .zo_datasets = 7,
 172         .zo_threads = 23,
 173         .zo_passtime = 60,              /* 60 seconds */
 174         .zo_killrate = 70,              /* 70% kill rate */
 175         .zo_verbose = 0,
 176         .zo_init = 1,
 177         .zo_time = 300,                 /* 5 minutes */
 178         .zo_maxloops = 50,              /* max loops during spa_freeze() */
 179         .zo_metaslab_gang_bang = 32 << 10
 180 };
 181 
 182 extern uint64_t metaslab_gang_bang;
 183 extern uint64_t metaslab_df_alloc_threshold;
 184 
 185 static ztest_shared_opts_t *ztest_shared_opts;
 186 static ztest_shared_opts_t ztest_opts;
 187 
 188 typedef struct ztest_shared_ds {
 189         uint64_t        zd_seq;
 190 } ztest_shared_ds_t;
 191 
 192 static ztest_shared_ds_t *ztest_shared_ds;
 193 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
 194 
 195 #define BT_MAGIC        0x123456789abcdefULL
 196 #define MAXFAULTS() \
 197         (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
 198 
 199 enum ztest_io_type {
 200         ZTEST_IO_WRITE_TAG,
 201         ZTEST_IO_WRITE_PATTERN,
 202         ZTEST_IO_WRITE_ZEROES,
 203         ZTEST_IO_TRUNCATE,
 204         ZTEST_IO_SETATTR,
 205         ZTEST_IO_TYPES
 206 };
 207 
 208 typedef struct ztest_block_tag {
 209         uint64_t        bt_magic;
 210         uint64_t        bt_objset;
 211         uint64_t        bt_object;
 212         uint64_t        bt_offset;
 213         uint64_t        bt_gen;
 214         uint64_t        bt_txg;
 215         uint64_t        bt_crtxg;
 216 } ztest_block_tag_t;
 217 
 218 typedef struct bufwad {
 219         uint64_t        bw_index;
 220         uint64_t        bw_txg;
 221         uint64_t        bw_data;
 222 } bufwad_t;
 223 
 224 /*
 225  * XXX -- fix zfs range locks to be generic so we can use them here.
 226  */
 227 typedef enum {
 228         RL_READER,
 229         RL_WRITER,
 230         RL_APPEND
 231 } rl_type_t;
 232 
 233 typedef struct rll {
 234         void            *rll_writer;
 235         int             rll_readers;
 236         mutex_t         rll_lock;
 237         cond_t          rll_cv;
 238 } rll_t;
 239 
 240 typedef struct rl {
 241         uint64_t        rl_object;
 242         uint64_t        rl_offset;
 243         uint64_t        rl_size;
 244         rll_t           *rl_lock;
 245 } rl_t;
 246 
 247 #define ZTEST_RANGE_LOCKS       64
 248 #define ZTEST_OBJECT_LOCKS      64
 249 
 250 /*
 251  * Object descriptor.  Used as a template for object lookup/create/remove.
 252  */
 253 typedef struct ztest_od {
 254         uint64_t        od_dir;
 255         uint64_t        od_object;
 256         dmu_object_type_t od_type;
 257         dmu_object_type_t od_crtype;
 258         uint64_t        od_blocksize;
 259         uint64_t        od_crblocksize;
 260         uint64_t        od_gen;
 261         uint64_t        od_crgen;
 262         char            od_name[MAXNAMELEN];
 263 } ztest_od_t;
 264 
 265 /*
 266  * Per-dataset state.
 267  */
 268 typedef struct ztest_ds {
 269         ztest_shared_ds_t *zd_shared;
 270         objset_t        *zd_os;
 271         rwlock_t        zd_zilog_lock;
 272         zilog_t         *zd_zilog;
 273         ztest_od_t      *zd_od;         /* debugging aid */
 274         char            zd_name[MAXNAMELEN];
 275         mutex_t         zd_dirobj_lock;
 276         rll_t           zd_object_lock[ZTEST_OBJECT_LOCKS];
 277         rll_t           zd_range_lock[ZTEST_RANGE_LOCKS];
 278 } ztest_ds_t;
 279 
 280 /*
 281  * Per-iteration state.
 282  */
 283 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
 284 
 285 typedef struct ztest_info {
 286         ztest_func_t    *zi_func;       /* test function */
 287         uint64_t        zi_iters;       /* iterations per execution */
 288         uint64_t        *zi_interval;   /* execute every <interval> seconds */
 289 } ztest_info_t;
 290 
 291 typedef struct ztest_shared_callstate {
 292         uint64_t        zc_count;       /* per-pass count */
 293         uint64_t        zc_time;        /* per-pass time */
 294         uint64_t        zc_next;        /* next time to call this function */
 295 } ztest_shared_callstate_t;
 296 
 297 static ztest_shared_callstate_t *ztest_shared_callstate;
 298 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
 299 
 300 /*
 301  * Note: these aren't static because we want dladdr() to work.
 302  */
 303 ztest_func_t ztest_dmu_read_write;
 304 ztest_func_t ztest_dmu_write_parallel;
 305 ztest_func_t ztest_dmu_object_alloc_free;
 306 ztest_func_t ztest_dmu_commit_callbacks;
 307 ztest_func_t ztest_zap;
 308 ztest_func_t ztest_zap_parallel;
 309 ztest_func_t ztest_zil_commit;
 310 ztest_func_t ztest_zil_remount;
 311 ztest_func_t ztest_dmu_read_write_zcopy;
 312 ztest_func_t ztest_dmu_objset_create_destroy;
 313 ztest_func_t ztest_dmu_prealloc;
 314 ztest_func_t ztest_fzap;
 315 ztest_func_t ztest_dmu_snapshot_create_destroy;
 316 ztest_func_t ztest_dsl_prop_get_set;
 317 ztest_func_t ztest_spa_prop_get_set;
 318 ztest_func_t ztest_spa_create_destroy;
 319 ztest_func_t ztest_fault_inject;
 320 ztest_func_t ztest_ddt_repair;
 321 ztest_func_t ztest_dmu_snapshot_hold;
 322 ztest_func_t ztest_spa_rename;
 323 ztest_func_t ztest_scrub;
 324 ztest_func_t ztest_dsl_dataset_promote_busy;
 325 ztest_func_t ztest_vdev_attach_detach;
 326 ztest_func_t ztest_vdev_LUN_growth;
 327 ztest_func_t ztest_vdev_add_remove;
 328 ztest_func_t ztest_vdev_aux_add_remove;
 329 ztest_func_t ztest_split_pool;
 330 ztest_func_t ztest_reguid;
 331 
 332 uint64_t zopt_always = 0ULL * NANOSEC;          /* all the time */
 333 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;  /* every 1/10 second */
 334 uint64_t zopt_often = 1ULL * NANOSEC;           /* every second */
 335 uint64_t zopt_sometimes = 10ULL * NANOSEC;      /* every 10 seconds */
 336 uint64_t zopt_rarely = 60ULL * NANOSEC;         /* every 60 seconds */
 337 
 338 ztest_info_t ztest_info[] = {
 339         { ztest_dmu_read_write,                 1,      &zopt_always        },
 340         { ztest_dmu_write_parallel,             10,     &zopt_always        },
 341         { ztest_dmu_object_alloc_free,          1,      &zopt_always        },
 342         { ztest_dmu_commit_callbacks,           1,      &zopt_always        },
 343         { ztest_zap,                            30,     &zopt_always        },
 344         { ztest_zap_parallel,                   100,    &zopt_always        },
 345         { ztest_split_pool,                     1,      &zopt_always        },
 346         { ztest_zil_commit,                     1,      &zopt_incessant     },
 347         { ztest_zil_remount,                    1,      &zopt_sometimes     },
 348         { ztest_dmu_read_write_zcopy,           1,      &zopt_often },
 349         { ztest_dmu_objset_create_destroy,      1,      &zopt_often },
 350         { ztest_dsl_prop_get_set,               1,      &zopt_often },
 351         { ztest_spa_prop_get_set,               1,      &zopt_sometimes     },
 352 #if 0
 353         { ztest_dmu_prealloc,                   1,      &zopt_sometimes     },
 354 #endif
 355         { ztest_fzap,                           1,      &zopt_sometimes     },
 356         { ztest_dmu_snapshot_create_destroy,    1,      &zopt_sometimes     },
 357         { ztest_spa_create_destroy,             1,      &zopt_sometimes     },
 358         { ztest_fault_inject,                   1,      &zopt_sometimes     },
 359         { ztest_ddt_repair,                     1,      &zopt_sometimes     },
 360         { ztest_dmu_snapshot_hold,              1,      &zopt_sometimes     },
 361         { ztest_reguid,                         1,      &zopt_sometimes },
 362         { ztest_spa_rename,                     1,      &zopt_rarely        },
 363         { ztest_scrub,                          1,      &zopt_rarely        },
 364         { ztest_dsl_dataset_promote_busy,       1,      &zopt_rarely        },
 365         { ztest_vdev_attach_detach,             1,      &zopt_rarely },
 366         { ztest_vdev_LUN_growth,                1,      &zopt_rarely        },
 367         { ztest_vdev_add_remove,                1,
 368             &ztest_opts.zo_vdevtime                         },
 369         { ztest_vdev_aux_add_remove,            1,
 370             &ztest_opts.zo_vdevtime                         },
 371 };
 372 
 373 #define ZTEST_FUNCS     (sizeof (ztest_info) / sizeof (ztest_info_t))
 374 
 375 /*
 376  * The following struct is used to hold a list of uncalled commit callbacks.
 377  * The callbacks are ordered by txg number.
 378  */
 379 typedef struct ztest_cb_list {
 380         mutex_t zcl_callbacks_lock;
 381         list_t  zcl_callbacks;
 382 } ztest_cb_list_t;
 383 
 384 /*
 385  * Stuff we need to share writably between parent and child.
 386  */
 387 typedef struct ztest_shared {
 388         boolean_t       zs_do_init;
 389         hrtime_t        zs_proc_start;
 390         hrtime_t        zs_proc_stop;
 391         hrtime_t        zs_thread_start;
 392         hrtime_t        zs_thread_stop;
 393         hrtime_t        zs_thread_kill;
 394         uint64_t        zs_enospc_count;
 395         uint64_t        zs_vdev_next_leaf;
 396         uint64_t        zs_vdev_aux;
 397         uint64_t        zs_alloc;
 398         uint64_t        zs_space;
 399         uint64_t        zs_splits;
 400         uint64_t        zs_mirrors;
 401         uint64_t        zs_metaslab_sz;
 402         uint64_t        zs_metaslab_df_alloc_threshold;
 403         uint64_t        zs_guid;
 404 } ztest_shared_t;
 405 
 406 #define ID_PARALLEL     -1ULL
 407 
 408 static char ztest_dev_template[] = "%s/%s.%llua";
 409 static char ztest_aux_template[] = "%s/%s.%s.%llu";
 410 ztest_shared_t *ztest_shared;
 411 
 412 static spa_t *ztest_spa = NULL;
 413 static ztest_ds_t *ztest_ds;
 414 
 415 static mutex_t ztest_vdev_lock;
 416 static rwlock_t ztest_name_lock;
 417 
 418 static boolean_t ztest_dump_core = B_TRUE;
 419 static boolean_t ztest_exiting;
 420 
 421 /* Global commit callback list */
 422 static ztest_cb_list_t zcl;
 423 
 424 enum ztest_object {
 425         ZTEST_META_DNODE = 0,
 426         ZTEST_DIROBJ,
 427         ZTEST_OBJECTS
 428 };
 429 
 430 static void usage(boolean_t) __NORETURN;
 431 
 432 /*
 433  * These libumem hooks provide a reasonable set of defaults for the allocator's
 434  * debugging facilities.
 435  */
 436 const char *
 437 _umem_debug_init()
 438 {
 439         return ("default,verbose"); /* $UMEM_DEBUG setting */
 440 }
 441 
 442 const char *
 443 _umem_logging_init(void)
 444 {
 445         return ("fail,contents"); /* $UMEM_LOGGING setting */
 446 }
 447 
 448 #define FATAL_MSG_SZ    1024
 449 
 450 char *fatal_msg;
 451 
 452 static void
 453 fatal(int do_perror, char *message, ...)
 454 {
 455         va_list args;
 456         int save_errno = errno;
 457         char buf[FATAL_MSG_SZ];
 458 
 459         (void) fflush(stdout);
 460 
 461         va_start(args, message);
 462         (void) sprintf(buf, "ztest: ");
 463         /* LINTED */
 464         (void) vsprintf(buf + strlen(buf), message, args);
 465         va_end(args);
 466         if (do_perror) {
 467                 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
 468                     ": %s", strerror(save_errno));
 469         }
 470         (void) fprintf(stderr, "%s\n", buf);
 471         fatal_msg = buf;                        /* to ease debugging */
 472         if (ztest_dump_core)
 473                 abort();
 474         exit(3);
 475 }
 476 
 477 static int
 478 str2shift(const char *buf)
 479 {
 480         const char *ends = "BKMGTPEZ";
 481         int i;
 482 
 483         if (buf[0] == '\0')
 484                 return (0);
 485         for (i = 0; i < strlen(ends); i++) {
 486                 if (toupper(buf[0]) == ends[i])
 487                         break;
 488         }
 489         if (i == strlen(ends)) {
 490                 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
 491                     buf);
 492                 usage(B_FALSE);
 493         }
 494         if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
 495                 return (10*i);
 496         }
 497         (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
 498         usage(B_FALSE);
 499         /* NOTREACHED */
 500 }
 501 
 502 static uint64_t
 503 nicenumtoull(const char *buf)
 504 {
 505         char *end;
 506         uint64_t val;
 507 
 508         val = strtoull(buf, &end, 0);
 509         if (end == buf) {
 510                 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
 511                 usage(B_FALSE);
 512         } else if (end[0] == '.') {
 513                 double fval = strtod(buf, &end);
 514                 fval *= pow(2, str2shift(end));
 515                 if (fval > UINT64_MAX) {
 516                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 517                             buf);
 518                         usage(B_FALSE);
 519                 }
 520                 val = (uint64_t)fval;
 521         } else {
 522                 int shift = str2shift(end);
 523                 if (shift >= 64 || (val << shift) >> shift != val) {
 524                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 525                             buf);
 526                         usage(B_FALSE);
 527                 }
 528                 val <<= shift;
 529         }
 530         return (val);
 531 }
 532 
 533 static void
 534 usage(boolean_t requested)
 535 {
 536         const ztest_shared_opts_t *zo = &ztest_opts_defaults;
 537 
 538         char nice_vdev_size[10];
 539         char nice_gang_bang[10];
 540         FILE *fp = requested ? stdout : stderr;
 541 
 542         nicenum(zo->zo_vdev_size, nice_vdev_size);
 543         nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
 544 
 545         (void) fprintf(fp, "Usage: %s\n"
 546             "\t[-v vdevs (default: %llu)]\n"
 547             "\t[-s size_of_each_vdev (default: %s)]\n"
 548             "\t[-a alignment_shift (default: %d)] use 0 for random\n"
 549             "\t[-m mirror_copies (default: %d)]\n"
 550             "\t[-r raidz_disks (default: %d)]\n"
 551             "\t[-R raidz_parity (default: %d)]\n"
 552             "\t[-d datasets (default: %d)]\n"
 553             "\t[-t threads (default: %d)]\n"
 554             "\t[-g gang_block_threshold (default: %s)]\n"
 555             "\t[-i init_count (default: %d)] initialize pool i times\n"
 556             "\t[-k kill_percentage (default: %llu%%)]\n"
 557             "\t[-p pool_name (default: %s)]\n"
 558             "\t[-f dir (default: %s)] file directory for vdev files\n"
 559             "\t[-V] verbose (use multiple times for ever more blather)\n"
 560             "\t[-E] use existing pool instead of creating new one\n"
 561             "\t[-T time (default: %llu sec)] total run time\n"
 562             "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
 563             "\t[-P passtime (default: %llu sec)] time per pass\n"
 564             "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
 565             "\t[-h] (print help)\n"
 566             "",
 567             zo->zo_pool,
 568             (u_longlong_t)zo->zo_vdevs,                      /* -v */
 569             nice_vdev_size,                             /* -s */
 570             zo->zo_ashift,                           /* -a */
 571             zo->zo_mirrors,                          /* -m */
 572             zo->zo_raidz,                            /* -r */
 573             zo->zo_raidz_parity,                     /* -R */
 574             zo->zo_datasets,                         /* -d */
 575             zo->zo_threads,                          /* -t */
 576             nice_gang_bang,                             /* -g */
 577             zo->zo_init,                             /* -i */
 578             (u_longlong_t)zo->zo_killrate,           /* -k */
 579             zo->zo_pool,                             /* -p */
 580             zo->zo_dir,                                      /* -f */
 581             (u_longlong_t)zo->zo_time,                       /* -T */
 582             (u_longlong_t)zo->zo_maxloops,           /* -F */
 583             (u_longlong_t)zo->zo_passtime);
 584         exit(requested ? 0 : 1);
 585 }
 586 
 587 static void
 588 process_options(int argc, char **argv)
 589 {
 590         char *path;
 591         ztest_shared_opts_t *zo = &ztest_opts;
 592 
 593         int opt;
 594         uint64_t value;
 595         char altdir[MAXNAMELEN] = { 0 };
 596 
 597         bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
 598 
 599         while ((opt = getopt(argc, argv,
 600             "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
 601                 value = 0;
 602                 switch (opt) {
 603                 case 'v':
 604                 case 's':
 605                 case 'a':
 606                 case 'm':
 607                 case 'r':
 608                 case 'R':
 609                 case 'd':
 610                 case 't':
 611                 case 'g':
 612                 case 'i':
 613                 case 'k':
 614                 case 'T':
 615                 case 'P':
 616                 case 'F':
 617                         value = nicenumtoull(optarg);
 618                 }
 619                 switch (opt) {
 620                 case 'v':
 621                         zo->zo_vdevs = value;
 622                         break;
 623                 case 's':
 624                         zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
 625                         break;
 626                 case 'a':
 627                         zo->zo_ashift = value;
 628                         break;
 629                 case 'm':
 630                         zo->zo_mirrors = value;
 631                         break;
 632                 case 'r':
 633                         zo->zo_raidz = MAX(1, value);
 634                         break;
 635                 case 'R':
 636                         zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
 637                         break;
 638                 case 'd':
 639                         zo->zo_datasets = MAX(1, value);
 640                         break;
 641                 case 't':
 642                         zo->zo_threads = MAX(1, value);
 643                         break;
 644                 case 'g':
 645                         zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
 646                             value);
 647                         break;
 648                 case 'i':
 649                         zo->zo_init = value;
 650                         break;
 651                 case 'k':
 652                         zo->zo_killrate = value;
 653                         break;
 654                 case 'p':
 655                         (void) strlcpy(zo->zo_pool, optarg,
 656                             sizeof (zo->zo_pool));
 657                         break;
 658                 case 'f':
 659                         path = realpath(optarg, NULL);
 660                         if (path == NULL) {
 661                                 (void) fprintf(stderr, "error: %s: %s\n",
 662                                     optarg, strerror(errno));
 663                                 usage(B_FALSE);
 664                         } else {
 665                                 (void) strlcpy(zo->zo_dir, path,
 666                                     sizeof (zo->zo_dir));
 667                         }
 668                         break;
 669                 case 'V':
 670                         zo->zo_verbose++;
 671                         break;
 672                 case 'E':
 673                         zo->zo_init = 0;
 674                         break;
 675                 case 'T':
 676                         zo->zo_time = value;
 677                         break;
 678                 case 'P':
 679                         zo->zo_passtime = MAX(1, value);
 680                         break;
 681                 case 'F':
 682                         zo->zo_maxloops = MAX(1, value);
 683                         break;
 684                 case 'B':
 685                         (void) strlcpy(altdir, optarg, sizeof (altdir));
 686                         break;
 687                 case 'h':
 688                         usage(B_TRUE);
 689                         break;
 690                 case '?':
 691                 default:
 692                         usage(B_FALSE);
 693                         break;
 694                 }
 695         }
 696 
 697         zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
 698 
 699         zo->zo_vdevtime =
 700             (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
 701             UINT64_MAX >> 2);
 702 
 703         if (strlen(altdir) > 0) {
 704                 char cmd[MAXNAMELEN];
 705                 char realaltdir[MAXNAMELEN];
 706                 char *bin;
 707                 char *ztest;
 708                 char *isa;
 709                 int isalen;
 710 
 711                 (void) realpath(getexecname(), cmd);
 712                 if (0 != access(altdir, F_OK)) {
 713                         ztest_dump_core = B_FALSE;
 714                         fatal(B_TRUE, "invalid alternate ztest path: %s",
 715                             altdir);
 716                 }
 717                 VERIFY(NULL != realpath(altdir, realaltdir));
 718 
 719                 /*
 720                  * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
 721                  * We want to extract <isa> to determine if we should use
 722                  * 32 or 64 bit binaries.
 723                  */
 724                 bin = strstr(cmd, "/usr/bin/");
 725                 ztest = strstr(bin, "/ztest");
 726                 isa = bin + 9;
 727                 isalen = ztest - isa;
 728                 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
 729                     "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
 730                 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
 731                     "%s/usr/lib/%.*s", realaltdir, isalen, isa);
 732 
 733                 if (0 != access(zo->zo_alt_ztest, X_OK)) {
 734                         ztest_dump_core = B_FALSE;
 735                         fatal(B_TRUE, "invalid alternate ztest: %s",
 736                             zo->zo_alt_ztest);
 737                 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
 738                         ztest_dump_core = B_FALSE;
 739                         fatal(B_TRUE, "invalid alternate lib directory %s",
 740                             zo->zo_alt_libpath);
 741                 }
 742         }
 743 }
 744 
 745 static void
 746 ztest_kill(ztest_shared_t *zs)
 747 {
 748         zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
 749         zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
 750         (void) kill(getpid(), SIGKILL);
 751 }
 752 
 753 static uint64_t
 754 ztest_random(uint64_t range)
 755 {
 756         uint64_t r;
 757 
 758         if (range == 0)
 759                 return (0);
 760 
 761         if (read(ZTEST_FD_RAND, &r, sizeof (r)) != sizeof (r))
 762                 fatal(1, "short read from /dev/urandom");
 763 
 764         return (r % range);
 765 }
 766 
 767 /* ARGSUSED */
 768 static void
 769 ztest_record_enospc(const char *s)
 770 {
 771         ztest_shared->zs_enospc_count++;
 772 }
 773 
 774 static uint64_t
 775 ztest_get_ashift(void)
 776 {
 777         if (ztest_opts.zo_ashift == 0)
 778                 return (SPA_MINBLOCKSHIFT + ztest_random(3));
 779         return (ztest_opts.zo_ashift);
 780 }
 781 
 782 static nvlist_t *
 783 make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift)
 784 {
 785         char pathbuf[MAXPATHLEN];
 786         uint64_t vdev;
 787         nvlist_t *file;
 788 
 789         if (ashift == 0)
 790                 ashift = ztest_get_ashift();
 791 
 792         if (path == NULL) {
 793                 path = pathbuf;
 794 
 795                 if (aux != NULL) {
 796                         vdev = ztest_shared->zs_vdev_aux;
 797                         (void) snprintf(path, sizeof (pathbuf),
 798                             ztest_aux_template, ztest_opts.zo_dir,
 799                             ztest_opts.zo_pool, aux, vdev);
 800                 } else {
 801                         vdev = ztest_shared->zs_vdev_next_leaf++;
 802                         (void) snprintf(path, sizeof (pathbuf),
 803                             ztest_dev_template, ztest_opts.zo_dir,
 804                             ztest_opts.zo_pool, vdev);
 805                 }
 806         }
 807 
 808         if (size != 0) {
 809                 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
 810                 if (fd == -1)
 811                         fatal(1, "can't open %s", path);
 812                 if (ftruncate(fd, size) != 0)
 813                         fatal(1, "can't ftruncate %s", path);
 814                 (void) close(fd);
 815         }
 816 
 817         VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
 818         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
 819         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
 820         VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
 821 
 822         return (file);
 823 }
 824 
 825 static nvlist_t *
 826 make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r)
 827 {
 828         nvlist_t *raidz, **child;
 829         int c;
 830 
 831         if (r < 2)
 832                 return (make_vdev_file(path, aux, size, ashift));
 833         child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
 834 
 835         for (c = 0; c < r; c++)
 836                 child[c] = make_vdev_file(path, aux, size, ashift);
 837 
 838         VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
 839         VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
 840             VDEV_TYPE_RAIDZ) == 0);
 841         VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
 842             ztest_opts.zo_raidz_parity) == 0);
 843         VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
 844             child, r) == 0);
 845 
 846         for (c = 0; c < r; c++)
 847                 nvlist_free(child[c]);
 848 
 849         umem_free(child, r * sizeof (nvlist_t *));
 850 
 851         return (raidz);
 852 }
 853 
 854 static nvlist_t *
 855 make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift,
 856         int r, int m)
 857 {
 858         nvlist_t *mirror, **child;
 859         int c;
 860 
 861         if (m < 1)
 862                 return (make_vdev_raidz(path, aux, size, ashift, r));
 863 
 864         child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
 865 
 866         for (c = 0; c < m; c++)
 867                 child[c] = make_vdev_raidz(path, aux, size, ashift, r);
 868 
 869         VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
 870         VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
 871             VDEV_TYPE_MIRROR) == 0);
 872         VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
 873             child, m) == 0);
 874 
 875         for (c = 0; c < m; c++)
 876                 nvlist_free(child[c]);
 877 
 878         umem_free(child, m * sizeof (nvlist_t *));
 879 
 880         return (mirror);
 881 }
 882 
 883 static nvlist_t *
 884 make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift,
 885         int log, int r, int m, int t)
 886 {
 887         nvlist_t *root, **child;
 888         int c;
 889 
 890         ASSERT(t > 0);
 891 
 892         child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
 893 
 894         for (c = 0; c < t; c++) {
 895                 child[c] = make_vdev_mirror(path, aux, size, ashift, r, m);
 896                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
 897                     log) == 0);
 898         }
 899 
 900         VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
 901         VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
 902         VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
 903             child, t) == 0);
 904 
 905         for (c = 0; c < t; c++)
 906                 nvlist_free(child[c]);
 907 
 908         umem_free(child, t * sizeof (nvlist_t *));
 909 
 910         return (root);
 911 }
 912 
 913 static int
 914 ztest_random_blocksize(void)
 915 {
 916         return (1 << (SPA_MINBLOCKSHIFT +
 917             ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
 918 }
 919 
 920 static int
 921 ztest_random_ibshift(void)
 922 {
 923         return (DN_MIN_INDBLKSHIFT +
 924             ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
 925 }
 926 
 927 static uint64_t
 928 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
 929 {
 930         uint64_t top;
 931         vdev_t *rvd = spa->spa_root_vdev;
 932         vdev_t *tvd;
 933 
 934         ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
 935 
 936         do {
 937                 top = ztest_random(rvd->vdev_children);
 938                 tvd = rvd->vdev_child[top];
 939         } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
 940             tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
 941 
 942         return (top);
 943 }
 944 
 945 static uint64_t
 946 ztest_random_dsl_prop(zfs_prop_t prop)
 947 {
 948         uint64_t value;
 949 
 950         do {
 951                 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
 952         } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
 953 
 954         return (value);
 955 }
 956 
 957 static int
 958 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
 959     boolean_t inherit)
 960 {
 961         const char *propname = zfs_prop_to_name(prop);
 962         const char *valname;
 963         char setpoint[MAXPATHLEN];
 964         uint64_t curval;
 965         int error;
 966 
 967         error = dsl_prop_set(osname, propname,
 968             (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL),
 969             sizeof (value), 1, &value);
 970 
 971         if (error == ENOSPC) {
 972                 ztest_record_enospc(FTAG);
 973                 return (error);
 974         }
 975         ASSERT0(error);
 976 
 977         VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval),
 978             1, &curval, setpoint), ==, 0);
 979 
 980         if (ztest_opts.zo_verbose >= 6) {
 981                 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
 982                 (void) printf("%s %s = %s at '%s'\n",
 983                     osname, propname, valname, setpoint);
 984         }
 985 
 986         return (error);
 987 }
 988 
 989 static int
 990 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
 991 {
 992         spa_t *spa = ztest_spa;
 993         nvlist_t *props = NULL;
 994         int error;
 995 
 996         VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
 997         VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
 998 
 999         error = spa_prop_set(spa, props);
1000 
1001         nvlist_free(props);
1002 
1003         if (error == ENOSPC) {
1004                 ztest_record_enospc(FTAG);
1005                 return (error);
1006         }
1007         ASSERT0(error);
1008 
1009         return (error);
1010 }
1011 
1012 static void
1013 ztest_rll_init(rll_t *rll)
1014 {
1015         rll->rll_writer = NULL;
1016         rll->rll_readers = 0;
1017         VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1018         VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1019 }
1020 
1021 static void
1022 ztest_rll_destroy(rll_t *rll)
1023 {
1024         ASSERT(rll->rll_writer == NULL);
1025         ASSERT(rll->rll_readers == 0);
1026         VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1027         VERIFY(cond_destroy(&rll->rll_cv) == 0);
1028 }
1029 
1030 static void
1031 ztest_rll_lock(rll_t *rll, rl_type_t type)
1032 {
1033         VERIFY(mutex_lock(&rll->rll_lock) == 0);
1034 
1035         if (type == RL_READER) {
1036                 while (rll->rll_writer != NULL)
1037                         (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1038                 rll->rll_readers++;
1039         } else {
1040                 while (rll->rll_writer != NULL || rll->rll_readers)
1041                         (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1042                 rll->rll_writer = curthread;
1043         }
1044 
1045         VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1046 }
1047 
1048 static void
1049 ztest_rll_unlock(rll_t *rll)
1050 {
1051         VERIFY(mutex_lock(&rll->rll_lock) == 0);
1052 
1053         if (rll->rll_writer) {
1054                 ASSERT(rll->rll_readers == 0);
1055                 rll->rll_writer = NULL;
1056         } else {
1057                 ASSERT(rll->rll_readers != 0);
1058                 ASSERT(rll->rll_writer == NULL);
1059                 rll->rll_readers--;
1060         }
1061 
1062         if (rll->rll_writer == NULL && rll->rll_readers == 0)
1063                 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1064 
1065         VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1066 }
1067 
1068 static void
1069 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1070 {
1071         rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1072 
1073         ztest_rll_lock(rll, type);
1074 }
1075 
1076 static void
1077 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1078 {
1079         rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1080 
1081         ztest_rll_unlock(rll);
1082 }
1083 
1084 static rl_t *
1085 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1086     uint64_t size, rl_type_t type)
1087 {
1088         uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1089         rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1090         rl_t *rl;
1091 
1092         rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1093         rl->rl_object = object;
1094         rl->rl_offset = offset;
1095         rl->rl_size = size;
1096         rl->rl_lock = rll;
1097 
1098         ztest_rll_lock(rll, type);
1099 
1100         return (rl);
1101 }
1102 
1103 static void
1104 ztest_range_unlock(rl_t *rl)
1105 {
1106         rll_t *rll = rl->rl_lock;
1107 
1108         ztest_rll_unlock(rll);
1109 
1110         umem_free(rl, sizeof (*rl));
1111 }
1112 
1113 static void
1114 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1115 {
1116         zd->zd_os = os;
1117         zd->zd_zilog = dmu_objset_zil(os);
1118         zd->zd_shared = szd;
1119         dmu_objset_name(os, zd->zd_name);
1120 
1121         if (zd->zd_shared != NULL)
1122                 zd->zd_shared->zd_seq = 0;
1123 
1124         VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1125         VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1126 
1127         for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1128                 ztest_rll_init(&zd->zd_object_lock[l]);
1129 
1130         for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1131                 ztest_rll_init(&zd->zd_range_lock[l]);
1132 }
1133 
1134 static void
1135 ztest_zd_fini(ztest_ds_t *zd)
1136 {
1137         VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1138 
1139         for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1140                 ztest_rll_destroy(&zd->zd_object_lock[l]);
1141 
1142         for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1143                 ztest_rll_destroy(&zd->zd_range_lock[l]);
1144 }
1145 
1146 #define TXG_MIGHTWAIT   (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1147 
1148 static uint64_t
1149 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1150 {
1151         uint64_t txg;
1152         int error;
1153 
1154         /*
1155          * Attempt to assign tx to some transaction group.
1156          */
1157         error = dmu_tx_assign(tx, txg_how);
1158         if (error) {
1159                 if (error == ERESTART) {
1160                         ASSERT(txg_how == TXG_NOWAIT);
1161                         dmu_tx_wait(tx);
1162                 } else {
1163                         ASSERT3U(error, ==, ENOSPC);
1164                         ztest_record_enospc(tag);
1165                 }
1166                 dmu_tx_abort(tx);
1167                 return (0);
1168         }
1169         txg = dmu_tx_get_txg(tx);
1170         ASSERT(txg != 0);
1171         return (txg);
1172 }
1173 
1174 static void
1175 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1176 {
1177         uint64_t *ip = buf;
1178         uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1179 
1180         while (ip < ip_end)
1181                 *ip++ = value;
1182 }
1183 
1184 static boolean_t
1185 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1186 {
1187         uint64_t *ip = buf;
1188         uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1189         uint64_t diff = 0;
1190 
1191         while (ip < ip_end)
1192                 diff |= (value - *ip++);
1193 
1194         return (diff == 0);
1195 }
1196 
1197 static void
1198 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1199     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1200 {
1201         bt->bt_magic = BT_MAGIC;
1202         bt->bt_objset = dmu_objset_id(os);
1203         bt->bt_object = object;
1204         bt->bt_offset = offset;
1205         bt->bt_gen = gen;
1206         bt->bt_txg = txg;
1207         bt->bt_crtxg = crtxg;
1208 }
1209 
1210 static void
1211 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1212     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1213 {
1214         ASSERT(bt->bt_magic == BT_MAGIC);
1215         ASSERT(bt->bt_objset == dmu_objset_id(os));
1216         ASSERT(bt->bt_object == object);
1217         ASSERT(bt->bt_offset == offset);
1218         ASSERT(bt->bt_gen <= gen);
1219         ASSERT(bt->bt_txg <= txg);
1220         ASSERT(bt->bt_crtxg == crtxg);
1221 }
1222 
1223 static ztest_block_tag_t *
1224 ztest_bt_bonus(dmu_buf_t *db)
1225 {
1226         dmu_object_info_t doi;
1227         ztest_block_tag_t *bt;
1228 
1229         dmu_object_info_from_db(db, &doi);
1230         ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1231         ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1232         bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1233 
1234         return (bt);
1235 }
1236 
1237 /*
1238  * ZIL logging ops
1239  */
1240 
1241 #define lrz_type        lr_mode
1242 #define lrz_blocksize   lr_uid
1243 #define lrz_ibshift     lr_gid
1244 #define lrz_bonustype   lr_rdev
1245 #define lrz_bonuslen    lr_crtime[1]
1246 
1247 static void
1248 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1249 {
1250         char *name = (void *)(lr + 1);          /* name follows lr */
1251         size_t namesize = strlen(name) + 1;
1252         itx_t *itx;
1253 
1254         if (zil_replaying(zd->zd_zilog, tx))
1255                 return;
1256 
1257         itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1258         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1259             sizeof (*lr) + namesize - sizeof (lr_t));
1260 
1261         zil_itx_assign(zd->zd_zilog, itx, tx);
1262 }
1263 
1264 static void
1265 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1266 {
1267         char *name = (void *)(lr + 1);          /* name follows lr */
1268         size_t namesize = strlen(name) + 1;
1269         itx_t *itx;
1270 
1271         if (zil_replaying(zd->zd_zilog, tx))
1272                 return;
1273 
1274         itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1275         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1276             sizeof (*lr) + namesize - sizeof (lr_t));
1277 
1278         itx->itx_oid = object;
1279         zil_itx_assign(zd->zd_zilog, itx, tx);
1280 }
1281 
1282 static void
1283 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1284 {
1285         itx_t *itx;
1286         itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1287 
1288         if (zil_replaying(zd->zd_zilog, tx))
1289                 return;
1290 
1291         if (lr->lr_length > ZIL_MAX_LOG_DATA)
1292                 write_state = WR_INDIRECT;
1293 
1294         itx = zil_itx_create(TX_WRITE,
1295             sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1296 
1297         if (write_state == WR_COPIED &&
1298             dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1299             ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1300                 zil_itx_destroy(itx);
1301                 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1302                 write_state = WR_NEED_COPY;
1303         }
1304         itx->itx_private = zd;
1305         itx->itx_wr_state = write_state;
1306         itx->itx_sync = (ztest_random(8) == 0);
1307         itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1308 
1309         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1310             sizeof (*lr) - sizeof (lr_t));
1311 
1312         zil_itx_assign(zd->zd_zilog, itx, tx);
1313 }
1314 
1315 static void
1316 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1317 {
1318         itx_t *itx;
1319 
1320         if (zil_replaying(zd->zd_zilog, tx))
1321                 return;
1322 
1323         itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1324         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1325             sizeof (*lr) - sizeof (lr_t));
1326 
1327         itx->itx_sync = B_FALSE;
1328         zil_itx_assign(zd->zd_zilog, itx, tx);
1329 }
1330 
1331 static void
1332 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1333 {
1334         itx_t *itx;
1335 
1336         if (zil_replaying(zd->zd_zilog, tx))
1337                 return;
1338 
1339         itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1340         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1341             sizeof (*lr) - sizeof (lr_t));
1342 
1343         itx->itx_sync = B_FALSE;
1344         zil_itx_assign(zd->zd_zilog, itx, tx);
1345 }
1346 
1347 /*
1348  * ZIL replay ops
1349  */
1350 static int
1351 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1352 {
1353         char *name = (void *)(lr + 1);          /* name follows lr */
1354         objset_t *os = zd->zd_os;
1355         ztest_block_tag_t *bbt;
1356         dmu_buf_t *db;
1357         dmu_tx_t *tx;
1358         uint64_t txg;
1359         int error = 0;
1360 
1361         if (byteswap)
1362                 byteswap_uint64_array(lr, sizeof (*lr));
1363 
1364         ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1365         ASSERT(name[0] != '\0');
1366 
1367         tx = dmu_tx_create(os);
1368 
1369         dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1370 
1371         if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1372                 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1373         } else {
1374                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1375         }
1376 
1377         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1378         if (txg == 0)
1379                 return (ENOSPC);
1380 
1381         ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1382 
1383         if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1384                 if (lr->lr_foid == 0) {
1385                         lr->lr_foid = zap_create(os,
1386                             lr->lrz_type, lr->lrz_bonustype,
1387                             lr->lrz_bonuslen, tx);
1388                 } else {
1389                         error = zap_create_claim(os, lr->lr_foid,
1390                             lr->lrz_type, lr->lrz_bonustype,
1391                             lr->lrz_bonuslen, tx);
1392                 }
1393         } else {
1394                 if (lr->lr_foid == 0) {
1395                         lr->lr_foid = dmu_object_alloc(os,
1396                             lr->lrz_type, 0, lr->lrz_bonustype,
1397                             lr->lrz_bonuslen, tx);
1398                 } else {
1399                         error = dmu_object_claim(os, lr->lr_foid,
1400                             lr->lrz_type, 0, lr->lrz_bonustype,
1401                             lr->lrz_bonuslen, tx);
1402                 }
1403         }
1404 
1405         if (error) {
1406                 ASSERT3U(error, ==, EEXIST);
1407                 ASSERT(zd->zd_zilog->zl_replay);
1408                 dmu_tx_commit(tx);
1409                 return (error);
1410         }
1411 
1412         ASSERT(lr->lr_foid != 0);
1413 
1414         if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1415                 VERIFY0(dmu_object_set_blocksize(os, lr->lr_foid,
1416                     lr->lrz_blocksize, lr->lrz_ibshift, tx));
1417 
1418         VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1419         bbt = ztest_bt_bonus(db);
1420         dmu_buf_will_dirty(db, tx);
1421         ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1422         dmu_buf_rele(db, FTAG);
1423 
1424         VERIFY0(zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1425             &lr->lr_foid, tx));
1426 
1427         (void) ztest_log_create(zd, tx, lr);
1428 
1429         dmu_tx_commit(tx);
1430 
1431         return (0);
1432 }
1433 
1434 static int
1435 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1436 {
1437         char *name = (void *)(lr + 1);          /* name follows lr */
1438         objset_t *os = zd->zd_os;
1439         dmu_object_info_t doi;
1440         dmu_tx_t *tx;
1441         uint64_t object, txg;
1442 
1443         if (byteswap)
1444                 byteswap_uint64_array(lr, sizeof (*lr));
1445 
1446         ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1447         ASSERT(name[0] != '\0');
1448 
1449         VERIFY0(
1450             zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1451         ASSERT(object != 0);
1452 
1453         ztest_object_lock(zd, object, RL_WRITER);
1454 
1455         VERIFY0(dmu_object_info(os, object, &doi));
1456 
1457         tx = dmu_tx_create(os);
1458 
1459         dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1460         dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1461 
1462         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1463         if (txg == 0) {
1464                 ztest_object_unlock(zd, object);
1465                 return (ENOSPC);
1466         }
1467 
1468         if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1469                 VERIFY0(zap_destroy(os, object, tx));
1470         } else {
1471                 VERIFY0(dmu_object_free(os, object, tx));
1472         }
1473 
1474         VERIFY0(zap_remove(os, lr->lr_doid, name, tx));
1475 
1476         (void) ztest_log_remove(zd, tx, lr, object);
1477 
1478         dmu_tx_commit(tx);
1479 
1480         ztest_object_unlock(zd, object);
1481 
1482         return (0);
1483 }
1484 
1485 static int
1486 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1487 {
1488         objset_t *os = zd->zd_os;
1489         void *data = lr + 1;                    /* data follows lr */
1490         uint64_t offset, length;
1491         ztest_block_tag_t *bt = data;
1492         ztest_block_tag_t *bbt;
1493         uint64_t gen, txg, lrtxg, crtxg;
1494         dmu_object_info_t doi;
1495         dmu_tx_t *tx;
1496         dmu_buf_t *db;
1497         arc_buf_t *abuf = NULL;
1498         rl_t *rl;
1499 
1500         if (byteswap)
1501                 byteswap_uint64_array(lr, sizeof (*lr));
1502 
1503         offset = lr->lr_offset;
1504         length = lr->lr_length;
1505 
1506         /* If it's a dmu_sync() block, write the whole block */
1507         if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1508                 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1509                 if (length < blocksize) {
1510                         offset -= offset % blocksize;
1511                         length = blocksize;
1512                 }
1513         }
1514 
1515         if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1516                 byteswap_uint64_array(bt, sizeof (*bt));
1517 
1518         if (bt->bt_magic != BT_MAGIC)
1519                 bt = NULL;
1520 
1521         ztest_object_lock(zd, lr->lr_foid, RL_READER);
1522         rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1523 
1524         VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1525 
1526         dmu_object_info_from_db(db, &doi);
1527 
1528         bbt = ztest_bt_bonus(db);
1529         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1530         gen = bbt->bt_gen;
1531         crtxg = bbt->bt_crtxg;
1532         lrtxg = lr->lr_common.lrc_txg;
1533 
1534         tx = dmu_tx_create(os);
1535 
1536         dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1537 
1538         if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1539             P2PHASE(offset, length) == 0)
1540                 abuf = dmu_request_arcbuf(db, length);
1541 
1542         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1543         if (txg == 0) {
1544                 if (abuf != NULL)
1545                         dmu_return_arcbuf(abuf);
1546                 dmu_buf_rele(db, FTAG);
1547                 ztest_range_unlock(rl);
1548                 ztest_object_unlock(zd, lr->lr_foid);
1549                 return (ENOSPC);
1550         }
1551 
1552         if (bt != NULL) {
1553                 /*
1554                  * Usually, verify the old data before writing new data --
1555                  * but not always, because we also want to verify correct
1556                  * behavior when the data was not recently read into cache.
1557                  */
1558                 ASSERT(offset % doi.doi_data_block_size == 0);
1559                 if (ztest_random(4) != 0) {
1560                         int prefetch = ztest_random(2) ?
1561                             DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1562                         ztest_block_tag_t rbt;
1563 
1564                         VERIFY(dmu_read(os, lr->lr_foid, offset,
1565                             sizeof (rbt), &rbt, prefetch) == 0);
1566                         if (rbt.bt_magic == BT_MAGIC) {
1567                                 ztest_bt_verify(&rbt, os, lr->lr_foid,
1568                                     offset, gen, txg, crtxg);
1569                         }
1570                 }
1571 
1572                 /*
1573                  * Writes can appear to be newer than the bonus buffer because
1574                  * the ztest_get_data() callback does a dmu_read() of the
1575                  * open-context data, which may be different than the data
1576                  * as it was when the write was generated.
1577                  */
1578                 if (zd->zd_zilog->zl_replay) {
1579                         ztest_bt_verify(bt, os, lr->lr_foid, offset,
1580                             MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1581                             bt->bt_crtxg);
1582                 }
1583 
1584                 /*
1585                  * Set the bt's gen/txg to the bonus buffer's gen/txg
1586                  * so that all of the usual ASSERTs will work.
1587                  */
1588                 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1589         }
1590 
1591         if (abuf == NULL) {
1592                 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1593         } else {
1594                 bcopy(data, abuf->b_data, length);
1595                 dmu_assign_arcbuf(db, offset, abuf, tx);
1596         }
1597 
1598         (void) ztest_log_write(zd, tx, lr);
1599 
1600         dmu_buf_rele(db, FTAG);
1601 
1602         dmu_tx_commit(tx);
1603 
1604         ztest_range_unlock(rl);
1605         ztest_object_unlock(zd, lr->lr_foid);
1606 
1607         return (0);
1608 }
1609 
1610 static int
1611 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1612 {
1613         objset_t *os = zd->zd_os;
1614         dmu_tx_t *tx;
1615         uint64_t txg;
1616         rl_t *rl;
1617 
1618         if (byteswap)
1619                 byteswap_uint64_array(lr, sizeof (*lr));
1620 
1621         ztest_object_lock(zd, lr->lr_foid, RL_READER);
1622         rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1623             RL_WRITER);
1624 
1625         tx = dmu_tx_create(os);
1626 
1627         dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1628 
1629         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1630         if (txg == 0) {
1631                 ztest_range_unlock(rl);
1632                 ztest_object_unlock(zd, lr->lr_foid);
1633                 return (ENOSPC);
1634         }
1635 
1636         VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1637             lr->lr_length, tx) == 0);
1638 
1639         (void) ztest_log_truncate(zd, tx, lr);
1640 
1641         dmu_tx_commit(tx);
1642 
1643         ztest_range_unlock(rl);
1644         ztest_object_unlock(zd, lr->lr_foid);
1645 
1646         return (0);
1647 }
1648 
1649 static int
1650 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1651 {
1652         objset_t *os = zd->zd_os;
1653         dmu_tx_t *tx;
1654         dmu_buf_t *db;
1655         ztest_block_tag_t *bbt;
1656         uint64_t txg, lrtxg, crtxg;
1657 
1658         if (byteswap)
1659                 byteswap_uint64_array(lr, sizeof (*lr));
1660 
1661         ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1662 
1663         VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1664 
1665         tx = dmu_tx_create(os);
1666         dmu_tx_hold_bonus(tx, lr->lr_foid);
1667 
1668         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1669         if (txg == 0) {
1670                 dmu_buf_rele(db, FTAG);
1671                 ztest_object_unlock(zd, lr->lr_foid);
1672                 return (ENOSPC);
1673         }
1674 
1675         bbt = ztest_bt_bonus(db);
1676         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1677         crtxg = bbt->bt_crtxg;
1678         lrtxg = lr->lr_common.lrc_txg;
1679 
1680         if (zd->zd_zilog->zl_replay) {
1681                 ASSERT(lr->lr_size != 0);
1682                 ASSERT(lr->lr_mode != 0);
1683                 ASSERT(lrtxg != 0);
1684         } else {
1685                 /*
1686                  * Randomly change the size and increment the generation.
1687                  */
1688                 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1689                     sizeof (*bbt);
1690                 lr->lr_mode = bbt->bt_gen + 1;
1691                 ASSERT(lrtxg == 0);
1692         }
1693 
1694         /*
1695          * Verify that the current bonus buffer is not newer than our txg.
1696          */
1697         ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1698             MAX(txg, lrtxg), crtxg);
1699 
1700         dmu_buf_will_dirty(db, tx);
1701 
1702         ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1703         ASSERT3U(lr->lr_size, <=, db->db_size);
1704         VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1705         bbt = ztest_bt_bonus(db);
1706 
1707         ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1708 
1709         dmu_buf_rele(db, FTAG);
1710 
1711         (void) ztest_log_setattr(zd, tx, lr);
1712 
1713         dmu_tx_commit(tx);
1714 
1715         ztest_object_unlock(zd, lr->lr_foid);
1716 
1717         return (0);
1718 }
1719 
1720 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1721         NULL,                   /* 0 no such transaction type */
1722         ztest_replay_create,    /* TX_CREATE */
1723         NULL,                   /* TX_MKDIR */
1724         NULL,                   /* TX_MKXATTR */
1725         NULL,                   /* TX_SYMLINK */
1726         ztest_replay_remove,    /* TX_REMOVE */
1727         NULL,                   /* TX_RMDIR */
1728         NULL,                   /* TX_LINK */
1729         NULL,                   /* TX_RENAME */
1730         ztest_replay_write,     /* TX_WRITE */
1731         ztest_replay_truncate,  /* TX_TRUNCATE */
1732         ztest_replay_setattr,   /* TX_SETATTR */
1733         NULL,                   /* TX_ACL */
1734         NULL,                   /* TX_CREATE_ACL */
1735         NULL,                   /* TX_CREATE_ATTR */
1736         NULL,                   /* TX_CREATE_ACL_ATTR */
1737         NULL,                   /* TX_MKDIR_ACL */
1738         NULL,                   /* TX_MKDIR_ATTR */
1739         NULL,                   /* TX_MKDIR_ACL_ATTR */
1740         NULL,                   /* TX_WRITE2 */
1741 };
1742 
1743 /*
1744  * ZIL get_data callbacks
1745  */
1746 
1747 static void
1748 ztest_get_done(zgd_t *zgd, int error)
1749 {
1750         ztest_ds_t *zd = zgd->zgd_private;
1751         uint64_t object = zgd->zgd_rl->rl_object;
1752 
1753         if (zgd->zgd_db)
1754                 dmu_buf_rele(zgd->zgd_db, zgd);
1755 
1756         ztest_range_unlock(zgd->zgd_rl);
1757         ztest_object_unlock(zd, object);
1758 
1759         if (error == 0 && zgd->zgd_bp)
1760                 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1761 
1762         umem_free(zgd, sizeof (*zgd));
1763 }
1764 
1765 static int
1766 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1767 {
1768         ztest_ds_t *zd = arg;
1769         objset_t *os = zd->zd_os;
1770         uint64_t object = lr->lr_foid;
1771         uint64_t offset = lr->lr_offset;
1772         uint64_t size = lr->lr_length;
1773         blkptr_t *bp = &lr->lr_blkptr;
1774         uint64_t txg = lr->lr_common.lrc_txg;
1775         uint64_t crtxg;
1776         dmu_object_info_t doi;
1777         dmu_buf_t *db;
1778         zgd_t *zgd;
1779         int error;
1780 
1781         ztest_object_lock(zd, object, RL_READER);
1782         error = dmu_bonus_hold(os, object, FTAG, &db);
1783         if (error) {
1784                 ztest_object_unlock(zd, object);
1785                 return (error);
1786         }
1787 
1788         crtxg = ztest_bt_bonus(db)->bt_crtxg;
1789 
1790         if (crtxg == 0 || crtxg > txg) {
1791                 dmu_buf_rele(db, FTAG);
1792                 ztest_object_unlock(zd, object);
1793                 return (ENOENT);
1794         }
1795 
1796         dmu_object_info_from_db(db, &doi);
1797         dmu_buf_rele(db, FTAG);
1798         db = NULL;
1799 
1800         zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1801         zgd->zgd_zilog = zd->zd_zilog;
1802         zgd->zgd_private = zd;
1803 
1804         if (buf != NULL) {      /* immediate write */
1805                 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1806                     RL_READER);
1807 
1808                 error = dmu_read(os, object, offset, size, buf,
1809                     DMU_READ_NO_PREFETCH);
1810                 ASSERT(error == 0);
1811         } else {
1812                 size = doi.doi_data_block_size;
1813                 if (ISP2(size)) {
1814                         offset = P2ALIGN(offset, size);
1815                 } else {
1816                         ASSERT(offset < size);
1817                         offset = 0;
1818                 }
1819 
1820                 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1821                     RL_READER);
1822 
1823                 error = dmu_buf_hold(os, object, offset, zgd, &db,
1824                     DMU_READ_NO_PREFETCH);
1825 
1826                 if (error == 0) {
1827                         zgd->zgd_db = db;
1828                         zgd->zgd_bp = bp;
1829 
1830                         ASSERT(db->db_offset == offset);
1831                         ASSERT(db->db_size == size);
1832 
1833                         error = dmu_sync(zio, lr->lr_common.lrc_txg,
1834                             ztest_get_done, zgd);
1835 
1836                         if (error == 0)
1837                                 return (0);
1838                 }
1839         }
1840 
1841         ztest_get_done(zgd, error);
1842 
1843         return (error);
1844 }
1845 
1846 static void *
1847 ztest_lr_alloc(size_t lrsize, char *name)
1848 {
1849         char *lr;
1850         size_t namesize = name ? strlen(name) + 1 : 0;
1851 
1852         lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1853 
1854         if (name)
1855                 bcopy(name, lr + lrsize, namesize);
1856 
1857         return (lr);
1858 }
1859 
1860 void
1861 ztest_lr_free(void *lr, size_t lrsize, char *name)
1862 {
1863         size_t namesize = name ? strlen(name) + 1 : 0;
1864 
1865         umem_free(lr, lrsize + namesize);
1866 }
1867 
1868 /*
1869  * Lookup a bunch of objects.  Returns the number of objects not found.
1870  */
1871 static int
1872 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1873 {
1874         int missing = 0;
1875         int error;
1876 
1877         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1878 
1879         for (int i = 0; i < count; i++, od++) {
1880                 od->od_object = 0;
1881                 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1882                     sizeof (uint64_t), 1, &od->od_object);
1883                 if (error) {
1884                         ASSERT(error == ENOENT);
1885                         ASSERT(od->od_object == 0);
1886                         missing++;
1887                 } else {
1888                         dmu_buf_t *db;
1889                         ztest_block_tag_t *bbt;
1890                         dmu_object_info_t doi;
1891 
1892                         ASSERT(od->od_object != 0);
1893                         ASSERT(missing == 0);   /* there should be no gaps */
1894 
1895                         ztest_object_lock(zd, od->od_object, RL_READER);
1896                         VERIFY0(dmu_bonus_hold(zd->zd_os,
1897                             od->od_object, FTAG, &db));
1898                         dmu_object_info_from_db(db, &doi);
1899                         bbt = ztest_bt_bonus(db);
1900                         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1901                         od->od_type = doi.doi_type;
1902                         od->od_blocksize = doi.doi_data_block_size;
1903                         od->od_gen = bbt->bt_gen;
1904                         dmu_buf_rele(db, FTAG);
1905                         ztest_object_unlock(zd, od->od_object);
1906                 }
1907         }
1908 
1909         return (missing);
1910 }
1911 
1912 static int
1913 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1914 {
1915         int missing = 0;
1916 
1917         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1918 
1919         for (int i = 0; i < count; i++, od++) {
1920                 if (missing) {
1921                         od->od_object = 0;
1922                         missing++;
1923                         continue;
1924                 }
1925 
1926                 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1927 
1928                 lr->lr_doid = od->od_dir;
1929                 lr->lr_foid = 0;     /* 0 to allocate, > 0 to claim */
1930                 lr->lrz_type = od->od_crtype;
1931                 lr->lrz_blocksize = od->od_crblocksize;
1932                 lr->lrz_ibshift = ztest_random_ibshift();
1933                 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1934                 lr->lrz_bonuslen = dmu_bonus_max();
1935                 lr->lr_gen = od->od_crgen;
1936                 lr->lr_crtime[0] = time(NULL);
1937 
1938                 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1939                         ASSERT(missing == 0);
1940                         od->od_object = 0;
1941                         missing++;
1942                 } else {
1943                         od->od_object = lr->lr_foid;
1944                         od->od_type = od->od_crtype;
1945                         od->od_blocksize = od->od_crblocksize;
1946                         od->od_gen = od->od_crgen;
1947                         ASSERT(od->od_object != 0);
1948                 }
1949 
1950                 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1951         }
1952 
1953         return (missing);
1954 }
1955 
1956 static int
1957 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
1958 {
1959         int missing = 0;
1960         int error;
1961 
1962         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1963 
1964         od += count - 1;
1965 
1966         for (int i = count - 1; i >= 0; i--, od--) {
1967                 if (missing) {
1968                         missing++;
1969                         continue;
1970                 }
1971 
1972                 if (od->od_object == 0)
1973                         continue;
1974 
1975                 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1976 
1977                 lr->lr_doid = od->od_dir;
1978 
1979                 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
1980                         ASSERT3U(error, ==, ENOSPC);
1981                         missing++;
1982                 } else {
1983                         od->od_object = 0;
1984                 }
1985                 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1986         }
1987 
1988         return (missing);
1989 }
1990 
1991 static int
1992 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
1993     void *data)
1994 {
1995         lr_write_t *lr;
1996         int error;
1997 
1998         lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
1999 
2000         lr->lr_foid = object;
2001         lr->lr_offset = offset;
2002         lr->lr_length = size;
2003         lr->lr_blkoff = 0;
2004         BP_ZERO(&lr->lr_blkptr);
2005 
2006         bcopy(data, lr + 1, size);
2007 
2008         error = ztest_replay_write(zd, lr, B_FALSE);
2009 
2010         ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2011 
2012         return (error);
2013 }
2014 
2015 static int
2016 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2017 {
2018         lr_truncate_t *lr;
2019         int error;
2020 
2021         lr = ztest_lr_alloc(sizeof (*lr), NULL);
2022 
2023         lr->lr_foid = object;
2024         lr->lr_offset = offset;
2025         lr->lr_length = size;
2026 
2027         error = ztest_replay_truncate(zd, lr, B_FALSE);
2028 
2029         ztest_lr_free(lr, sizeof (*lr), NULL);
2030 
2031         return (error);
2032 }
2033 
2034 static int
2035 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2036 {
2037         lr_setattr_t *lr;
2038         int error;
2039 
2040         lr = ztest_lr_alloc(sizeof (*lr), NULL);
2041 
2042         lr->lr_foid = object;
2043         lr->lr_size = 0;
2044         lr->lr_mode = 0;
2045 
2046         error = ztest_replay_setattr(zd, lr, B_FALSE);
2047 
2048         ztest_lr_free(lr, sizeof (*lr), NULL);
2049 
2050         return (error);
2051 }
2052 
2053 static void
2054 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2055 {
2056         objset_t *os = zd->zd_os;
2057         dmu_tx_t *tx;
2058         uint64_t txg;
2059         rl_t *rl;
2060 
2061         txg_wait_synced(dmu_objset_pool(os), 0);
2062 
2063         ztest_object_lock(zd, object, RL_READER);
2064         rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2065 
2066         tx = dmu_tx_create(os);
2067 
2068         dmu_tx_hold_write(tx, object, offset, size);
2069 
2070         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2071 
2072         if (txg != 0) {
2073                 dmu_prealloc(os, object, offset, size, tx);
2074                 dmu_tx_commit(tx);
2075                 txg_wait_synced(dmu_objset_pool(os), txg);
2076         } else {
2077                 (void) dmu_free_long_range(os, object, offset, size);
2078         }
2079 
2080         ztest_range_unlock(rl);
2081         ztest_object_unlock(zd, object);
2082 }
2083 
2084 static void
2085 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2086 {
2087         ztest_block_tag_t wbt;
2088         dmu_object_info_t doi;
2089         enum ztest_io_type io_type;
2090         uint64_t blocksize;
2091         void *data;
2092 
2093         VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2094         blocksize = doi.doi_data_block_size;
2095         data = umem_alloc(blocksize, UMEM_NOFAIL);
2096 
2097         /*
2098          * Pick an i/o type at random, biased toward writing block tags.
2099          */
2100         io_type = ztest_random(ZTEST_IO_TYPES);
2101         if (ztest_random(2) == 0)
2102                 io_type = ZTEST_IO_WRITE_TAG;
2103 
2104         (void) rw_rdlock(&zd->zd_zilog_lock);
2105 
2106         switch (io_type) {
2107 
2108         case ZTEST_IO_WRITE_TAG:
2109                 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2110                 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2111                 break;
2112 
2113         case ZTEST_IO_WRITE_PATTERN:
2114                 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2115                 if (ztest_random(2) == 0) {
2116                         /*
2117                          * Induce fletcher2 collisions to ensure that
2118                          * zio_ddt_collision() detects and resolves them
2119                          * when using fletcher2-verify for deduplication.
2120                          */
2121                         ((uint64_t *)data)[0] ^= 1ULL << 63;
2122                         ((uint64_t *)data)[4] ^= 1ULL << 63;
2123                 }
2124                 (void) ztest_write(zd, object, offset, blocksize, data);
2125                 break;
2126 
2127         case ZTEST_IO_WRITE_ZEROES:
2128                 bzero(data, blocksize);
2129                 (void) ztest_write(zd, object, offset, blocksize, data);
2130                 break;
2131 
2132         case ZTEST_IO_TRUNCATE:
2133                 (void) ztest_truncate(zd, object, offset, blocksize);
2134                 break;
2135 
2136         case ZTEST_IO_SETATTR:
2137                 (void) ztest_setattr(zd, object);
2138                 break;
2139         }
2140 
2141         (void) rw_unlock(&zd->zd_zilog_lock);
2142 
2143         umem_free(data, blocksize);
2144 }
2145 
2146 /*
2147  * Initialize an object description template.
2148  */
2149 static void
2150 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2151     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2152 {
2153         od->od_dir = ZTEST_DIROBJ;
2154         od->od_object = 0;
2155 
2156         od->od_crtype = type;
2157         od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2158         od->od_crgen = gen;
2159 
2160         od->od_type = DMU_OT_NONE;
2161         od->od_blocksize = 0;
2162         od->od_gen = 0;
2163 
2164         (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2165             tag, (int64_t)id, index);
2166 }
2167 
2168 /*
2169  * Lookup or create the objects for a test using the od template.
2170  * If the objects do not all exist, or if 'remove' is specified,
2171  * remove any existing objects and create new ones.  Otherwise,
2172  * use the existing objects.
2173  */
2174 static int
2175 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2176 {
2177         int count = size / sizeof (*od);
2178         int rv = 0;
2179 
2180         VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2181         if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2182             (ztest_remove(zd, od, count) != 0 ||
2183             ztest_create(zd, od, count) != 0))
2184                 rv = -1;
2185         zd->zd_od = od;
2186         VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2187 
2188         return (rv);
2189 }
2190 
2191 /* ARGSUSED */
2192 void
2193 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2194 {
2195         zilog_t *zilog = zd->zd_zilog;
2196 
2197         (void) rw_rdlock(&zd->zd_zilog_lock);
2198 
2199         zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2200 
2201         /*
2202          * Remember the committed values in zd, which is in parent/child
2203          * shared memory.  If we die, the next iteration of ztest_run()
2204          * will verify that the log really does contain this record.
2205          */
2206         mutex_enter(&zilog->zl_lock);
2207         ASSERT(zd->zd_shared != NULL);
2208         ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2209         zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2210         mutex_exit(&zilog->zl_lock);
2211 
2212         (void) rw_unlock(&zd->zd_zilog_lock);
2213 }
2214 
2215 /*
2216  * This function is designed to simulate the operations that occur during a
2217  * mount/unmount operation.  We hold the dataset across these operations in an
2218  * attempt to expose any implicit assumptions about ZIL management.
2219  */
2220 /* ARGSUSED */
2221 void
2222 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2223 {
2224         objset_t *os = zd->zd_os;
2225 
2226         (void) rw_wrlock(&zd->zd_zilog_lock);
2227 
2228         /* zfsvfs_teardown() */
2229         zil_close(zd->zd_zilog);
2230 
2231         /* zfsvfs_setup() */
2232         VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2233         zil_replay(os, zd, ztest_replay_vector);
2234 
2235         (void) rw_unlock(&zd->zd_zilog_lock);
2236 }
2237 
2238 /*
2239  * Verify that we can't destroy an active pool, create an existing pool,
2240  * or create a pool with a bad vdev spec.
2241  */
2242 /* ARGSUSED */
2243 void
2244 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2245 {
2246         ztest_shared_opts_t *zo = &ztest_opts;
2247         spa_t *spa;
2248         nvlist_t *nvroot;
2249 
2250         /*
2251          * Attempt to create using a bad file.
2252          */
2253         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2254         VERIFY3U(ENOENT, ==,
2255             spa_create("ztest_bad_file", nvroot, NULL, NULL));
2256         nvlist_free(nvroot);
2257 
2258         /*
2259          * Attempt to create using a bad mirror.
2260          */
2261         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1);
2262         VERIFY3U(ENOENT, ==,
2263             spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2264         nvlist_free(nvroot);
2265 
2266         /*
2267          * Attempt to create an existing pool.  It shouldn't matter
2268          * what's in the nvroot; we should fail with EEXIST.
2269          */
2270         (void) rw_rdlock(&ztest_name_lock);
2271         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2272         VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2273         nvlist_free(nvroot);
2274         VERIFY0(spa_open(zo->zo_pool, &spa, FTAG));
2275         VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2276         spa_close(spa, FTAG);
2277 
2278         (void) rw_unlock(&ztest_name_lock);
2279 }
2280 
2281 static vdev_t *
2282 vdev_lookup_by_path(vdev_t *vd, const char *path)
2283 {
2284         vdev_t *mvd;
2285 
2286         if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2287                 return (vd);
2288 
2289         for (int c = 0; c < vd->vdev_children; c++)
2290                 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2291                     NULL)
2292                         return (mvd);
2293 
2294         return (NULL);
2295 }
2296 
2297 /*
2298  * Find the first available hole which can be used as a top-level.
2299  */
2300 int
2301 find_vdev_hole(spa_t *spa)
2302 {
2303         vdev_t *rvd = spa->spa_root_vdev;
2304         int c;
2305 
2306         ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2307 
2308         for (c = 0; c < rvd->vdev_children; c++) {
2309                 vdev_t *cvd = rvd->vdev_child[c];
2310 
2311                 if (cvd->vdev_ishole)
2312                         break;
2313         }
2314         return (c);
2315 }
2316 
2317 /*
2318  * Verify that vdev_add() works as expected.
2319  */
2320 /* ARGSUSED */
2321 void
2322 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2323 {
2324         ztest_shared_t *zs = ztest_shared;
2325         spa_t *spa = ztest_spa;
2326         uint64_t leaves;
2327         uint64_t guid;
2328         nvlist_t *nvroot;
2329         int error;
2330 
2331         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2332         leaves =
2333             MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2334 
2335         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2336 
2337         ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2338 
2339         /*
2340          * If we have slogs then remove them 1/4 of the time.
2341          */
2342         if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2343                 /*
2344                  * Grab the guid from the head of the log class rotor.
2345                  */
2346                 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2347 
2348                 spa_config_exit(spa, SCL_VDEV, FTAG);
2349 
2350                 /*
2351                  * We have to grab the zs_name_lock as writer to
2352                  * prevent a race between removing a slog (dmu_objset_find)
2353                  * and destroying a dataset. Removing the slog will
2354                  * grab a reference on the dataset which may cause
2355                  * dmu_objset_destroy() to fail with EBUSY thus
2356                  * leaving the dataset in an inconsistent state.
2357                  */
2358                 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2359                 error = spa_vdev_remove(spa, guid, B_FALSE);
2360                 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2361 
2362                 if (error && error != EEXIST)
2363                         fatal(0, "spa_vdev_remove() = %d", error);
2364         } else {
2365                 spa_config_exit(spa, SCL_VDEV, FTAG);
2366 
2367                 /*
2368                  * Make 1/4 of the devices be log devices.
2369                  */
2370                 nvroot = make_vdev_root(NULL, NULL,
2371                     ztest_opts.zo_vdev_size, 0,
2372                     ztest_random(4) == 0, ztest_opts.zo_raidz,
2373                     zs->zs_mirrors, 1);
2374 
2375                 error = spa_vdev_add(spa, nvroot);
2376                 nvlist_free(nvroot);
2377 
2378                 if (error == ENOSPC)
2379                         ztest_record_enospc("spa_vdev_add");
2380                 else if (error != 0)
2381                         fatal(0, "spa_vdev_add() = %d", error);
2382         }
2383 
2384         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2385 }
2386 
2387 /*
2388  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2389  */
2390 /* ARGSUSED */
2391 void
2392 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2393 {
2394         ztest_shared_t *zs = ztest_shared;
2395         spa_t *spa = ztest_spa;
2396         vdev_t *rvd = spa->spa_root_vdev;
2397         spa_aux_vdev_t *sav;
2398         char *aux;
2399         uint64_t guid = 0;
2400         int error;
2401 
2402         if (ztest_random(2) == 0) {
2403                 sav = &spa->spa_spares;
2404                 aux = ZPOOL_CONFIG_SPARES;
2405         } else {
2406                 sav = &spa->spa_l2cache;
2407                 aux = ZPOOL_CONFIG_L2CACHE;
2408         }
2409 
2410         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2411 
2412         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2413 
2414         if (sav->sav_count != 0 && ztest_random(4) == 0) {
2415                 /*
2416                  * Pick a random device to remove.
2417                  */
2418                 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2419         } else {
2420                 /*
2421                  * Find an unused device we can add.
2422                  */
2423                 zs->zs_vdev_aux = 0;
2424                 for (;;) {
2425                         char path[MAXPATHLEN];
2426                         int c;
2427                         (void) snprintf(path, sizeof (path), ztest_aux_template,
2428                             ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2429                             zs->zs_vdev_aux);
2430                         for (c = 0; c < sav->sav_count; c++)
2431                                 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2432                                     path) == 0)
2433                                         break;
2434                         if (c == sav->sav_count &&
2435                             vdev_lookup_by_path(rvd, path) == NULL)
2436                                 break;
2437                         zs->zs_vdev_aux++;
2438                 }
2439         }
2440 
2441         spa_config_exit(spa, SCL_VDEV, FTAG);
2442 
2443         if (guid == 0) {
2444                 /*
2445                  * Add a new device.
2446                  */
2447                 nvlist_t *nvroot = make_vdev_root(NULL, aux,
2448                     (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2449                 error = spa_vdev_add(spa, nvroot);
2450                 if (error != 0)
2451                         fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2452                 nvlist_free(nvroot);
2453         } else {
2454                 /*
2455                  * Remove an existing device.  Sometimes, dirty its
2456                  * vdev state first to make sure we handle removal
2457                  * of devices that have pending state changes.
2458                  */
2459                 if (ztest_random(2) == 0)
2460                         (void) vdev_online(spa, guid, 0, NULL);
2461 
2462                 error = spa_vdev_remove(spa, guid, B_FALSE);
2463                 if (error != 0 && error != EBUSY)
2464                         fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2465         }
2466 
2467         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2468 }
2469 
2470 /*
2471  * split a pool if it has mirror tlvdevs
2472  */
2473 /* ARGSUSED */
2474 void
2475 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2476 {
2477         ztest_shared_t *zs = ztest_shared;
2478         spa_t *spa = ztest_spa;
2479         vdev_t *rvd = spa->spa_root_vdev;
2480         nvlist_t *tree, **child, *config, *split, **schild;
2481         uint_t c, children, schildren = 0, lastlogid = 0;
2482         int error = 0;
2483 
2484         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2485 
2486         /* ensure we have a useable config; mirrors of raidz aren't supported */
2487         if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2488                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2489                 return;
2490         }
2491 
2492         /* clean up the old pool, if any */
2493         (void) spa_destroy("splitp");
2494 
2495         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2496 
2497         /* generate a config from the existing config */
2498         mutex_enter(&spa->spa_props_lock);
2499         VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2500             &tree) == 0);
2501         mutex_exit(&spa->spa_props_lock);
2502 
2503         VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2504             &children) == 0);
2505 
2506         schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2507         for (c = 0; c < children; c++) {
2508                 vdev_t *tvd = rvd->vdev_child[c];
2509                 nvlist_t **mchild;
2510                 uint_t mchildren;
2511 
2512                 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2513                         VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2514                             0) == 0);
2515                         VERIFY(nvlist_add_string(schild[schildren],
2516                             ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2517                         VERIFY(nvlist_add_uint64(schild[schildren],
2518                             ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2519                         if (lastlogid == 0)
2520                                 lastlogid = schildren;
2521                         ++schildren;
2522                         continue;
2523                 }
2524                 lastlogid = 0;
2525                 VERIFY(nvlist_lookup_nvlist_array(child[c],
2526                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2527                 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2528         }
2529 
2530         /* OK, create a config that can be used to split */
2531         VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2532         VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2533             VDEV_TYPE_ROOT) == 0);
2534         VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2535             lastlogid != 0 ? lastlogid : schildren) == 0);
2536 
2537         VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2538         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2539 
2540         for (c = 0; c < schildren; c++)
2541                 nvlist_free(schild[c]);
2542         free(schild);
2543         nvlist_free(split);
2544 
2545         spa_config_exit(spa, SCL_VDEV, FTAG);
2546 
2547         (void) rw_wrlock(&ztest_name_lock);
2548         error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2549         (void) rw_unlock(&ztest_name_lock);
2550 
2551         nvlist_free(config);
2552 
2553         if (error == 0) {
2554                 (void) printf("successful split - results:\n");
2555                 mutex_enter(&spa_namespace_lock);
2556                 show_pool_stats(spa);
2557                 show_pool_stats(spa_lookup("splitp"));
2558                 mutex_exit(&spa_namespace_lock);
2559                 ++zs->zs_splits;
2560                 --zs->zs_mirrors;
2561         }
2562         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2563 
2564 }
2565 
2566 /*
2567  * Verify that we can attach and detach devices.
2568  */
2569 /* ARGSUSED */
2570 void
2571 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2572 {
2573         ztest_shared_t *zs = ztest_shared;
2574         spa_t *spa = ztest_spa;
2575         spa_aux_vdev_t *sav = &spa->spa_spares;
2576         vdev_t *rvd = spa->spa_root_vdev;
2577         vdev_t *oldvd, *newvd, *pvd;
2578         nvlist_t *root;
2579         uint64_t leaves;
2580         uint64_t leaf, top;
2581         uint64_t ashift = ztest_get_ashift();
2582         uint64_t oldguid, pguid;
2583         size_t oldsize, newsize;
2584         char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2585         int replacing;
2586         int oldvd_has_siblings = B_FALSE;
2587         int newvd_is_spare = B_FALSE;
2588         int oldvd_is_log;
2589         int error, expected_error;
2590 
2591         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2592         leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2593 
2594         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2595 
2596         /*
2597          * Decide whether to do an attach or a replace.
2598          */
2599         replacing = ztest_random(2);
2600 
2601         /*
2602          * Pick a random top-level vdev.
2603          */
2604         top = ztest_random_vdev_top(spa, B_TRUE);
2605 
2606         /*
2607          * Pick a random leaf within it.
2608          */
2609         leaf = ztest_random(leaves);
2610 
2611         /*
2612          * Locate this vdev.
2613          */
2614         oldvd = rvd->vdev_child[top];
2615         if (zs->zs_mirrors >= 1) {
2616                 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2617                 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2618                 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2619         }
2620         if (ztest_opts.zo_raidz > 1) {
2621                 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2622                 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2623                 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2624         }
2625 
2626         /*
2627          * If we're already doing an attach or replace, oldvd may be a
2628          * mirror vdev -- in which case, pick a random child.
2629          */
2630         while (oldvd->vdev_children != 0) {
2631                 oldvd_has_siblings = B_TRUE;
2632                 ASSERT(oldvd->vdev_children >= 2);
2633                 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2634         }
2635 
2636         oldguid = oldvd->vdev_guid;
2637         oldsize = vdev_get_min_asize(oldvd);
2638         oldvd_is_log = oldvd->vdev_top->vdev_islog;
2639         (void) strcpy(oldpath, oldvd->vdev_path);
2640         pvd = oldvd->vdev_parent;
2641         pguid = pvd->vdev_guid;
2642 
2643         /*
2644          * If oldvd has siblings, then half of the time, detach it.
2645          */
2646         if (oldvd_has_siblings && ztest_random(2) == 0) {
2647                 spa_config_exit(spa, SCL_VDEV, FTAG);
2648                 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2649                 if (error != 0 && error != ENODEV && error != EBUSY &&
2650                     error != ENOTSUP)
2651                         fatal(0, "detach (%s) returned %d", oldpath, error);
2652                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2653                 return;
2654         }
2655 
2656         /*
2657          * For the new vdev, choose with equal probability between the two
2658          * standard paths (ending in either 'a' or 'b') or a random hot spare.
2659          */
2660         if (sav->sav_count != 0 && ztest_random(3) == 0) {
2661                 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2662                 newvd_is_spare = B_TRUE;
2663                 (void) strcpy(newpath, newvd->vdev_path);
2664         } else {
2665                 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2666                     ztest_opts.zo_dir, ztest_opts.zo_pool,
2667                     top * leaves + leaf);
2668                 if (ztest_random(2) == 0)
2669                         newpath[strlen(newpath) - 1] = 'b';
2670                 newvd = vdev_lookup_by_path(rvd, newpath);
2671         }
2672 
2673         if (newvd) {
2674                 newsize = vdev_get_min_asize(newvd);
2675         } else {
2676                 /*
2677                  * Make newsize a little bigger or smaller than oldsize.
2678                  * If it's smaller, the attach should fail.
2679                  * If it's larger, and we're doing a replace,
2680                  * we should get dynamic LUN growth when we're done.
2681                  */
2682                 newsize = 10 * oldsize / (9 + ztest_random(3));
2683         }
2684 
2685         /*
2686          * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2687          * unless it's a replace; in that case any non-replacing parent is OK.
2688          *
2689          * If newvd is already part of the pool, it should fail with EBUSY.
2690          *
2691          * If newvd is too small, it should fail with EOVERFLOW.
2692          */
2693         if (pvd->vdev_ops != &vdev_mirror_ops &&
2694             pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2695             pvd->vdev_ops == &vdev_replacing_ops ||
2696             pvd->vdev_ops == &vdev_spare_ops))
2697                 expected_error = ENOTSUP;
2698         else if (newvd_is_spare && (!replacing || oldvd_is_log))
2699                 expected_error = ENOTSUP;
2700         else if (newvd == oldvd)
2701                 expected_error = replacing ? 0 : EBUSY;
2702         else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2703                 expected_error = EBUSY;
2704         else if (newsize < oldsize)
2705                 expected_error = EOVERFLOW;
2706         else if (ashift > oldvd->vdev_top->vdev_ashift)
2707                 expected_error = EDOM;
2708         else
2709                 expected_error = 0;
2710 
2711         spa_config_exit(spa, SCL_VDEV, FTAG);
2712 
2713         /*
2714          * Build the nvlist describing newpath.
2715          */
2716         root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0,
2717             ashift, 0, 0, 0, 1);
2718 
2719         error = spa_vdev_attach(spa, oldguid, root, replacing);
2720 
2721         nvlist_free(root);
2722 
2723         /*
2724          * If our parent was the replacing vdev, but the replace completed,
2725          * then instead of failing with ENOTSUP we may either succeed,
2726          * fail with ENODEV, or fail with EOVERFLOW.
2727          */
2728         if (expected_error == ENOTSUP &&
2729             (error == 0 || error == ENODEV || error == EOVERFLOW))
2730                 expected_error = error;
2731 
2732         /*
2733          * If someone grew the LUN, the replacement may be too small.
2734          */
2735         if (error == EOVERFLOW || error == EBUSY)
2736                 expected_error = error;
2737 
2738         /* XXX workaround 6690467 */
2739         if (error != expected_error && expected_error != EBUSY) {
2740                 fatal(0, "attach (%s %llu, %s %llu, %d) "
2741                     "returned %d, expected %d",
2742                     oldpath, (longlong_t)oldsize, newpath,
2743                     (longlong_t)newsize, replacing, error, expected_error);
2744         }
2745 
2746         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2747 }
2748 
2749 /*
2750  * Callback function which expands the physical size of the vdev.
2751  */
2752 vdev_t *
2753 grow_vdev(vdev_t *vd, void *arg)
2754 {
2755         spa_t *spa = vd->vdev_spa;
2756         size_t *newsize = arg;
2757         size_t fsize;
2758         int fd;
2759 
2760         ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2761         ASSERT(vd->vdev_ops->vdev_op_leaf);
2762 
2763         if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2764                 return (vd);
2765 
2766         fsize = lseek(fd, 0, SEEK_END);
2767         (void) ftruncate(fd, *newsize);
2768 
2769         if (ztest_opts.zo_verbose >= 6) {
2770                 (void) printf("%s grew from %lu to %lu bytes\n",
2771                     vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2772         }
2773         (void) close(fd);
2774         return (NULL);
2775 }
2776 
2777 /*
2778  * Callback function which expands a given vdev by calling vdev_online().
2779  */
2780 /* ARGSUSED */
2781 vdev_t *
2782 online_vdev(vdev_t *vd, void *arg)
2783 {
2784         spa_t *spa = vd->vdev_spa;
2785         vdev_t *tvd = vd->vdev_top;
2786         uint64_t guid = vd->vdev_guid;
2787         uint64_t generation = spa->spa_config_generation + 1;
2788         vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2789         int error;
2790 
2791         ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2792         ASSERT(vd->vdev_ops->vdev_op_leaf);
2793 
2794         /* Calling vdev_online will initialize the new metaslabs */
2795         spa_config_exit(spa, SCL_STATE, spa);
2796         error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2797         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2798 
2799         /*
2800          * If vdev_online returned an error or the underlying vdev_open
2801          * failed then we abort the expand. The only way to know that
2802          * vdev_open fails is by checking the returned newstate.
2803          */
2804         if (error || newstate != VDEV_STATE_HEALTHY) {
2805                 if (ztest_opts.zo_verbose >= 5) {
2806                         (void) printf("Unable to expand vdev, state %llu, "
2807                             "error %d\n", (u_longlong_t)newstate, error);
2808                 }
2809                 return (vd);
2810         }
2811         ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2812 
2813         /*
2814          * Since we dropped the lock we need to ensure that we're
2815          * still talking to the original vdev. It's possible this
2816          * vdev may have been detached/replaced while we were
2817          * trying to online it.
2818          */
2819         if (generation != spa->spa_config_generation) {
2820                 if (ztest_opts.zo_verbose >= 5) {
2821                         (void) printf("vdev configuration has changed, "
2822                             "guid %llu, state %llu, expected gen %llu, "
2823                             "got gen %llu\n",
2824                             (u_longlong_t)guid,
2825                             (u_longlong_t)tvd->vdev_state,
2826                             (u_longlong_t)generation,
2827                             (u_longlong_t)spa->spa_config_generation);
2828                 }
2829                 return (vd);
2830         }
2831         return (NULL);
2832 }
2833 
2834 /*
2835  * Traverse the vdev tree calling the supplied function.
2836  * We continue to walk the tree until we either have walked all
2837  * children or we receive a non-NULL return from the callback.
2838  * If a NULL callback is passed, then we just return back the first
2839  * leaf vdev we encounter.
2840  */
2841 vdev_t *
2842 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2843 {
2844         if (vd->vdev_ops->vdev_op_leaf) {
2845                 if (func == NULL)
2846                         return (vd);
2847                 else
2848                         return (func(vd, arg));
2849         }
2850 
2851         for (uint_t c = 0; c < vd->vdev_children; c++) {
2852                 vdev_t *cvd = vd->vdev_child[c];
2853                 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
2854                         return (cvd);
2855         }
2856         return (NULL);
2857 }
2858 
2859 /*
2860  * Verify that dynamic LUN growth works as expected.
2861  */
2862 /* ARGSUSED */
2863 void
2864 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
2865 {
2866         spa_t *spa = ztest_spa;
2867         vdev_t *vd, *tvd;
2868         metaslab_class_t *mc;
2869         metaslab_group_t *mg;
2870         size_t psize, newsize;
2871         uint64_t top;
2872         uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
2873 
2874         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2875         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2876 
2877         top = ztest_random_vdev_top(spa, B_TRUE);
2878 
2879         tvd = spa->spa_root_vdev->vdev_child[top];
2880         mg = tvd->vdev_mg;
2881         mc = mg->mg_class;
2882         old_ms_count = tvd->vdev_ms_count;
2883         old_class_space = metaslab_class_get_space(mc);
2884 
2885         /*
2886          * Determine the size of the first leaf vdev associated with
2887          * our top-level device.
2888          */
2889         vd = vdev_walk_tree(tvd, NULL, NULL);
2890         ASSERT3P(vd, !=, NULL);
2891         ASSERT(vd->vdev_ops->vdev_op_leaf);
2892 
2893         psize = vd->vdev_psize;
2894 
2895         /*
2896          * We only try to expand the vdev if it's healthy, less than 4x its
2897          * original size, and it has a valid psize.
2898          */
2899         if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
2900             psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
2901                 spa_config_exit(spa, SCL_STATE, spa);
2902                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2903                 return;
2904         }
2905         ASSERT(psize > 0);
2906         newsize = psize + psize / 8;
2907         ASSERT3U(newsize, >, psize);
2908 
2909         if (ztest_opts.zo_verbose >= 6) {
2910                 (void) printf("Expanding LUN %s from %lu to %lu\n",
2911                     vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
2912         }
2913 
2914         /*
2915          * Growing the vdev is a two step process:
2916          *      1). expand the physical size (i.e. relabel)
2917          *      2). online the vdev to create the new metaslabs
2918          */
2919         if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
2920             vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
2921             tvd->vdev_state != VDEV_STATE_HEALTHY) {
2922                 if (ztest_opts.zo_verbose >= 5) {
2923                         (void) printf("Could not expand LUN because "
2924                             "the vdev configuration changed.\n");
2925                 }
2926                 spa_config_exit(spa, SCL_STATE, spa);
2927                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2928                 return;
2929         }
2930 
2931         spa_config_exit(spa, SCL_STATE, spa);
2932 
2933         /*
2934          * Expanding the LUN will update the config asynchronously,
2935          * thus we must wait for the async thread to complete any
2936          * pending tasks before proceeding.
2937          */
2938         for (;;) {
2939                 boolean_t done;
2940                 mutex_enter(&spa->spa_async_lock);
2941                 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
2942                 mutex_exit(&spa->spa_async_lock);
2943                 if (done)
2944                         break;
2945                 txg_wait_synced(spa_get_dsl(spa), 0);
2946                 (void) poll(NULL, 0, 100);
2947         }
2948 
2949         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2950 
2951         tvd = spa->spa_root_vdev->vdev_child[top];
2952         new_ms_count = tvd->vdev_ms_count;
2953         new_class_space = metaslab_class_get_space(mc);
2954 
2955         if (tvd->vdev_mg != mg || mg->mg_class != mc) {
2956                 if (ztest_opts.zo_verbose >= 5) {
2957                         (void) printf("Could not verify LUN expansion due to "
2958                             "intervening vdev offline or remove.\n");
2959                 }
2960                 spa_config_exit(spa, SCL_STATE, spa);
2961                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2962                 return;
2963         }
2964 
2965         /*
2966          * Make sure we were able to grow the vdev.
2967          */
2968         if (new_ms_count <= old_ms_count)
2969                 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
2970                     old_ms_count, new_ms_count);
2971 
2972         /*
2973          * Make sure we were able to grow the pool.
2974          */
2975         if (new_class_space <= old_class_space)
2976                 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
2977                     old_class_space, new_class_space);
2978 
2979         if (ztest_opts.zo_verbose >= 5) {
2980                 char oldnumbuf[6], newnumbuf[6];
2981 
2982                 nicenum(old_class_space, oldnumbuf);
2983                 nicenum(new_class_space, newnumbuf);
2984                 (void) printf("%s grew from %s to %s\n",
2985                     spa->spa_name, oldnumbuf, newnumbuf);
2986         }
2987 
2988         spa_config_exit(spa, SCL_STATE, spa);
2989         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2990 }
2991 
2992 /*
2993  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
2994  */
2995 /* ARGSUSED */
2996 static void
2997 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
2998 {
2999         /*
3000          * Create the objects common to all ztest datasets.
3001          */
3002         VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3003             DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3004 }
3005 
3006 static int
3007 ztest_dataset_create(char *dsname)
3008 {
3009         uint64_t zilset = ztest_random(100);
3010         int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3011             ztest_objset_create_cb, NULL);
3012 
3013         if (err || zilset < 80)
3014                 return (err);
3015 
3016         if (ztest_opts.zo_verbose >= 6)
3017                 (void) printf("Setting dataset %s to sync always\n", dsname);
3018         return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3019             ZFS_SYNC_ALWAYS, B_FALSE));
3020 }
3021 
3022 /* ARGSUSED */
3023 static int
3024 ztest_objset_destroy_cb(const char *name, void *arg)
3025 {
3026         objset_t *os;
3027         dmu_object_info_t doi;
3028         int error;
3029 
3030         /*
3031          * Verify that the dataset contains a directory object.
3032          */
3033         VERIFY0(dmu_objset_hold(name, FTAG, &os));
3034         error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3035         if (error != ENOENT) {
3036                 /* We could have crashed in the middle of destroying it */
3037                 ASSERT0(error);
3038                 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3039                 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3040         }
3041         dmu_objset_rele(os, FTAG);
3042 
3043         /*
3044          * Destroy the dataset.
3045          */
3046         VERIFY0(dmu_objset_destroy(name, B_FALSE));
3047         return (0);
3048 }
3049 
3050 static boolean_t
3051 ztest_snapshot_create(char *osname, uint64_t id)
3052 {
3053         char snapname[MAXNAMELEN];
3054         int error;
3055 
3056         (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3057             (u_longlong_t)id);
3058 
3059         error = dmu_objset_snapshot_one(osname, strchr(snapname, '@') + 1);
3060         if (error == ENOSPC) {
3061                 ztest_record_enospc(FTAG);
3062                 return (B_FALSE);
3063         }
3064         if (error != 0 && error != EEXIST)
3065                 fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error);
3066         return (B_TRUE);
3067 }
3068 
3069 static boolean_t
3070 ztest_snapshot_destroy(char *osname, uint64_t id)
3071 {
3072         char snapname[MAXNAMELEN];
3073         int error;
3074 
3075         (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3076             (u_longlong_t)id);
3077 
3078         error = dmu_objset_destroy(snapname, B_FALSE);
3079         if (error != 0 && error != ENOENT)
3080                 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3081         return (B_TRUE);
3082 }
3083 
3084 /* ARGSUSED */
3085 void
3086 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3087 {
3088         ztest_ds_t zdtmp;
3089         int iters;
3090         int error;
3091         objset_t *os, *os2;
3092         char name[MAXNAMELEN];
3093         zilog_t *zilog;
3094 
3095         (void) rw_rdlock(&ztest_name_lock);
3096 
3097         (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3098             ztest_opts.zo_pool, (u_longlong_t)id);
3099 
3100         /*
3101          * If this dataset exists from a previous run, process its replay log
3102          * half of the time.  If we don't replay it, then dmu_objset_destroy()
3103          * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3104          */
3105         if (ztest_random(2) == 0 &&
3106             dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3107                 ztest_zd_init(&zdtmp, NULL, os);
3108                 zil_replay(os, &zdtmp, ztest_replay_vector);
3109                 ztest_zd_fini(&zdtmp);
3110                 dmu_objset_disown(os, FTAG);
3111         }
3112 
3113         /*
3114          * There may be an old instance of the dataset we're about to
3115          * create lying around from a previous run.  If so, destroy it
3116          * and all of its snapshots.
3117          */
3118         (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3119             DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3120 
3121         /*
3122          * Verify that the destroyed dataset is no longer in the namespace.
3123          */
3124         VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os));
3125 
3126         /*
3127          * Verify that we can create a new dataset.
3128          */
3129         error = ztest_dataset_create(name);
3130         if (error) {
3131                 if (error == ENOSPC) {
3132                         ztest_record_enospc(FTAG);
3133                         (void) rw_unlock(&ztest_name_lock);
3134                         return;
3135                 }
3136                 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3137         }
3138 
3139         VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3140 
3141         ztest_zd_init(&zdtmp, NULL, os);
3142 
3143         /*
3144          * Open the intent log for it.
3145          */
3146         zilog = zil_open(os, ztest_get_data);
3147 
3148         /*
3149          * Put some objects in there, do a little I/O to them,
3150          * and randomly take a couple of snapshots along the way.
3151          */
3152         iters = ztest_random(5);
3153         for (int i = 0; i < iters; i++) {
3154                 ztest_dmu_object_alloc_free(&zdtmp, id);
3155                 if (ztest_random(iters) == 0)
3156                         (void) ztest_snapshot_create(name, i);
3157         }
3158 
3159         /*
3160          * Verify that we cannot create an existing dataset.
3161          */
3162         VERIFY3U(EEXIST, ==,
3163             dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3164 
3165         /*
3166          * Verify that we can hold an objset that is also owned.
3167          */
3168         VERIFY0(dmu_objset_hold(name, FTAG, &os2));
3169         dmu_objset_rele(os2, FTAG);
3170 
3171         /*
3172          * Verify that we cannot own an objset that is already owned.
3173          */
3174         VERIFY3U(EBUSY, ==,
3175             dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3176 
3177         zil_close(zilog);
3178         dmu_objset_disown(os, FTAG);
3179         ztest_zd_fini(&zdtmp);
3180 
3181         (void) rw_unlock(&ztest_name_lock);
3182 }
3183 
3184 /*
3185  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3186  */
3187 void
3188 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3189 {
3190         (void) rw_rdlock(&ztest_name_lock);
3191         (void) ztest_snapshot_destroy(zd->zd_name, id);
3192         (void) ztest_snapshot_create(zd->zd_name, id);
3193         (void) rw_unlock(&ztest_name_lock);
3194 }
3195 
3196 /*
3197  * Cleanup non-standard snapshots and clones.
3198  */
3199 void
3200 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3201 {
3202         char snap1name[MAXNAMELEN];
3203         char clone1name[MAXNAMELEN];
3204         char snap2name[MAXNAMELEN];
3205         char clone2name[MAXNAMELEN];
3206         char snap3name[MAXNAMELEN];
3207         int error;
3208 
3209         (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3210         (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3211         (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3212         (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3213         (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3214 
3215         error = dmu_objset_destroy(clone2name, B_FALSE);
3216         if (error && error != ENOENT)
3217                 fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error);
3218         error = dmu_objset_destroy(snap3name, B_FALSE);
3219         if (error && error != ENOENT)
3220                 fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error);
3221         error = dmu_objset_destroy(snap2name, B_FALSE);
3222         if (error && error != ENOENT)
3223                 fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error);
3224         error = dmu_objset_destroy(clone1name, B_FALSE);
3225         if (error && error != ENOENT)
3226                 fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error);
3227         error = dmu_objset_destroy(snap1name, B_FALSE);
3228         if (error && error != ENOENT)
3229                 fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error);
3230 }
3231 
3232 /*
3233  * Verify dsl_dataset_promote handles EBUSY
3234  */
3235 void
3236 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3237 {
3238         objset_t *clone;
3239         dsl_dataset_t *ds;
3240         char snap1name[MAXNAMELEN];
3241         char clone1name[MAXNAMELEN];
3242         char snap2name[MAXNAMELEN];
3243         char clone2name[MAXNAMELEN];
3244         char snap3name[MAXNAMELEN];
3245         char *osname = zd->zd_name;
3246         int error;
3247 
3248         (void) rw_rdlock(&ztest_name_lock);
3249 
3250         ztest_dsl_dataset_cleanup(osname, id);
3251 
3252         (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3253         (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3254         (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3255         (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3256         (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3257 
3258         error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3259         if (error && error != EEXIST) {
3260                 if (error == ENOSPC) {
3261                         ztest_record_enospc(FTAG);
3262                         goto out;
3263                 }
3264                 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3265         }
3266 
3267         error = dmu_objset_hold(snap1name, FTAG, &clone);
3268         if (error)
3269                 fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error);
3270 
3271         error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0);
3272         dmu_objset_rele(clone, FTAG);
3273         if (error) {
3274                 if (error == ENOSPC) {
3275                         ztest_record_enospc(FTAG);
3276                         goto out;
3277                 }
3278                 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3279         }
3280 
3281         error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3282         if (error && error != EEXIST) {
3283                 if (error == ENOSPC) {
3284                         ztest_record_enospc(FTAG);
3285                         goto out;
3286                 }
3287                 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3288         }
3289 
3290         error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3291         if (error && error != EEXIST) {
3292                 if (error == ENOSPC) {
3293                         ztest_record_enospc(FTAG);
3294                         goto out;
3295                 }
3296                 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3297         }
3298 
3299         error = dmu_objset_hold(snap3name, FTAG, &clone);
3300         if (error)
3301                 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3302 
3303         error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0);
3304         dmu_objset_rele(clone, FTAG);
3305         if (error) {
3306                 if (error == ENOSPC) {
3307                         ztest_record_enospc(FTAG);
3308                         goto out;
3309                 }
3310                 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3311         }
3312 
3313         error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds);
3314         if (error)
3315                 fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error);
3316         error = dsl_dataset_promote(clone2name, NULL);
3317         if (error != EBUSY)
3318                 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3319                     error);
3320         dsl_dataset_disown(ds, FTAG);
3321 
3322 out:
3323         ztest_dsl_dataset_cleanup(osname, id);
3324 
3325         (void) rw_unlock(&ztest_name_lock);
3326 }
3327 
3328 /*
3329  * Verify that dmu_object_{alloc,free} work as expected.
3330  */
3331 void
3332 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3333 {
3334         ztest_od_t od[4];
3335         int batchsize = sizeof (od) / sizeof (od[0]);
3336 
3337         for (int b = 0; b < batchsize; b++)
3338                 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3339 
3340         /*
3341          * Destroy the previous batch of objects, create a new batch,
3342          * and do some I/O on the new objects.
3343          */
3344         if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3345                 return;
3346 
3347         while (ztest_random(4 * batchsize) != 0)
3348                 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3349                     ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3350 }
3351 
3352 /*
3353  * Verify that dmu_{read,write} work as expected.
3354  */
3355 void
3356 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3357 {
3358         objset_t *os = zd->zd_os;
3359         ztest_od_t od[2];
3360         dmu_tx_t *tx;
3361         int i, freeit, error;
3362         uint64_t n, s, txg;
3363         bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3364         uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3365         uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3366         uint64_t regions = 997;
3367         uint64_t stride = 123456789ULL;
3368         uint64_t width = 40;
3369         int free_percent = 5;
3370 
3371         /*
3372          * This test uses two objects, packobj and bigobj, that are always
3373          * updated together (i.e. in the same tx) so that their contents are
3374          * in sync and can be compared.  Their contents relate to each other
3375          * in a simple way: packobj is a dense array of 'bufwad' structures,
3376          * while bigobj is a sparse array of the same bufwads.  Specifically,
3377          * for any index n, there are three bufwads that should be identical:
3378          *
3379          *      packobj, at offset n * sizeof (bufwad_t)
3380          *      bigobj, at the head of the nth chunk
3381          *      bigobj, at the tail of the nth chunk
3382          *
3383          * The chunk size is arbitrary. It doesn't have to be a power of two,
3384          * and it doesn't have any relation to the object blocksize.
3385          * The only requirement is that it can hold at least two bufwads.
3386          *
3387          * Normally, we write the bufwad to each of these locations.
3388          * However, free_percent of the time we instead write zeroes to
3389          * packobj and perform a dmu_free_range() on bigobj.  By comparing
3390          * bigobj to packobj, we can verify that the DMU is correctly
3391          * tracking which parts of an object are allocated and free,
3392          * and that the contents of the allocated blocks are correct.
3393          */
3394 
3395         /*
3396          * Read the directory info.  If it's the first time, set things up.
3397          */
3398         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3399         ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3400 
3401         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3402                 return;
3403 
3404         bigobj = od[0].od_object;
3405         packobj = od[1].od_object;
3406         chunksize = od[0].od_gen;
3407         ASSERT(chunksize == od[1].od_gen);
3408 
3409         /*
3410          * Prefetch a random chunk of the big object.
3411          * Our aim here is to get some async reads in flight
3412          * for blocks that we may free below; the DMU should
3413          * handle this race correctly.
3414          */
3415         n = ztest_random(regions) * stride + ztest_random(width);
3416         s = 1 + ztest_random(2 * width - 1);
3417         dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3418 
3419         /*
3420          * Pick a random index and compute the offsets into packobj and bigobj.
3421          */
3422         n = ztest_random(regions) * stride + ztest_random(width);
3423         s = 1 + ztest_random(width - 1);
3424 
3425         packoff = n * sizeof (bufwad_t);
3426         packsize = s * sizeof (bufwad_t);
3427 
3428         bigoff = n * chunksize;
3429         bigsize = s * chunksize;
3430 
3431         packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3432         bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3433 
3434         /*
3435          * free_percent of the time, free a range of bigobj rather than
3436          * overwriting it.
3437          */
3438         freeit = (ztest_random(100) < free_percent);
3439 
3440         /*
3441          * Read the current contents of our objects.
3442          */
3443         error = dmu_read(os, packobj, packoff, packsize, packbuf,
3444             DMU_READ_PREFETCH);
3445         ASSERT0(error);
3446         error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3447             DMU_READ_PREFETCH);
3448         ASSERT0(error);
3449 
3450         /*
3451          * Get a tx for the mods to both packobj and bigobj.
3452          */
3453         tx = dmu_tx_create(os);
3454 
3455         dmu_tx_hold_write(tx, packobj, packoff, packsize);
3456 
3457         if (freeit)
3458                 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3459         else
3460                 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3461 
3462         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3463         if (txg == 0) {
3464                 umem_free(packbuf, packsize);
3465                 umem_free(bigbuf, bigsize);
3466                 return;
3467         }
3468 
3469         dmu_object_set_checksum(os, bigobj,
3470             (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3471 
3472         dmu_object_set_compress(os, bigobj,
3473             (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3474 
3475         /*
3476          * For each index from n to n + s, verify that the existing bufwad
3477          * in packobj matches the bufwads at the head and tail of the
3478          * corresponding chunk in bigobj.  Then update all three bufwads
3479          * with the new values we want to write out.
3480          */
3481         for (i = 0; i < s; i++) {
3482                 /* LINTED */
3483                 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3484                 /* LINTED */
3485                 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3486                 /* LINTED */
3487                 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3488 
3489                 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3490                 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3491 
3492                 if (pack->bw_txg > txg)
3493                         fatal(0, "future leak: got %llx, open txg is %llx",
3494                             pack->bw_txg, txg);
3495 
3496                 if (pack->bw_data != 0 && pack->bw_index != n + i)
3497                         fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3498                             pack->bw_index, n, i);
3499 
3500                 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3501                         fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3502 
3503                 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3504                         fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3505 
3506                 if (freeit) {
3507                         bzero(pack, sizeof (bufwad_t));
3508                 } else {
3509                         pack->bw_index = n + i;
3510                         pack->bw_txg = txg;
3511                         pack->bw_data = 1 + ztest_random(-2ULL);
3512                 }
3513                 *bigH = *pack;
3514                 *bigT = *pack;
3515         }
3516 
3517         /*
3518          * We've verified all the old bufwads, and made new ones.
3519          * Now write them out.
3520          */
3521         dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3522 
3523         if (freeit) {
3524                 if (ztest_opts.zo_verbose >= 7) {
3525                         (void) printf("freeing offset %llx size %llx"
3526                             " txg %llx\n",
3527                             (u_longlong_t)bigoff,
3528                             (u_longlong_t)bigsize,
3529                             (u_longlong_t)txg);
3530                 }
3531                 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3532         } else {
3533                 if (ztest_opts.zo_verbose >= 7) {
3534                         (void) printf("writing offset %llx size %llx"
3535                             " txg %llx\n",
3536                             (u_longlong_t)bigoff,
3537                             (u_longlong_t)bigsize,
3538                             (u_longlong_t)txg);
3539                 }
3540                 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3541         }
3542 
3543         dmu_tx_commit(tx);
3544 
3545         /*
3546          * Sanity check the stuff we just wrote.
3547          */
3548         {
3549                 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3550                 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3551 
3552                 VERIFY(0 == dmu_read(os, packobj, packoff,
3553                     packsize, packcheck, DMU_READ_PREFETCH));
3554                 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3555                     bigsize, bigcheck, DMU_READ_PREFETCH));
3556 
3557                 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3558                 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3559 
3560                 umem_free(packcheck, packsize);
3561                 umem_free(bigcheck, bigsize);
3562         }
3563 
3564         umem_free(packbuf, packsize);
3565         umem_free(bigbuf, bigsize);
3566 }
3567 
3568 void
3569 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3570     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3571 {
3572         uint64_t i;
3573         bufwad_t *pack;
3574         bufwad_t *bigH;
3575         bufwad_t *bigT;
3576 
3577         /*
3578          * For each index from n to n + s, verify that the existing bufwad
3579          * in packobj matches the bufwads at the head and tail of the
3580          * corresponding chunk in bigobj.  Then update all three bufwads
3581          * with the new values we want to write out.
3582          */
3583         for (i = 0; i < s; i++) {
3584                 /* LINTED */
3585                 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3586                 /* LINTED */
3587                 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3588                 /* LINTED */
3589                 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3590 
3591                 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3592                 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3593 
3594                 if (pack->bw_txg > txg)
3595                         fatal(0, "future leak: got %llx, open txg is %llx",
3596                             pack->bw_txg, txg);
3597 
3598                 if (pack->bw_data != 0 && pack->bw_index != n + i)
3599                         fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3600                             pack->bw_index, n, i);
3601 
3602                 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3603                         fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3604 
3605                 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3606                         fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3607 
3608                 pack->bw_index = n + i;
3609                 pack->bw_txg = txg;
3610                 pack->bw_data = 1 + ztest_random(-2ULL);
3611 
3612                 *bigH = *pack;
3613                 *bigT = *pack;
3614         }
3615 }
3616 
3617 void
3618 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3619 {
3620         objset_t *os = zd->zd_os;
3621         ztest_od_t od[2];
3622         dmu_tx_t *tx;
3623         uint64_t i;
3624         int error;
3625         uint64_t n, s, txg;
3626         bufwad_t *packbuf, *bigbuf;
3627         uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3628         uint64_t blocksize = ztest_random_blocksize();
3629         uint64_t chunksize = blocksize;
3630         uint64_t regions = 997;
3631         uint64_t stride = 123456789ULL;
3632         uint64_t width = 9;
3633         dmu_buf_t *bonus_db;
3634         arc_buf_t **bigbuf_arcbufs;
3635         dmu_object_info_t doi;
3636 
3637         /*
3638          * This test uses two objects, packobj and bigobj, that are always
3639          * updated together (i.e. in the same tx) so that their contents are
3640          * in sync and can be compared.  Their contents relate to each other
3641          * in a simple way: packobj is a dense array of 'bufwad' structures,
3642          * while bigobj is a sparse array of the same bufwads.  Specifically,
3643          * for any index n, there are three bufwads that should be identical:
3644          *
3645          *      packobj, at offset n * sizeof (bufwad_t)
3646          *      bigobj, at the head of the nth chunk
3647          *      bigobj, at the tail of the nth chunk
3648          *
3649          * The chunk size is set equal to bigobj block size so that
3650          * dmu_assign_arcbuf() can be tested for object updates.
3651          */
3652 
3653         /*
3654          * Read the directory info.  If it's the first time, set things up.
3655          */
3656         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3657         ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3658 
3659         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3660                 return;
3661 
3662         bigobj = od[0].od_object;
3663         packobj = od[1].od_object;
3664         blocksize = od[0].od_blocksize;
3665         chunksize = blocksize;
3666         ASSERT(chunksize == od[1].od_gen);
3667 
3668         VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3669         VERIFY(ISP2(doi.doi_data_block_size));
3670         VERIFY(chunksize == doi.doi_data_block_size);
3671         VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3672 
3673         /*
3674          * Pick a random index and compute the offsets into packobj and bigobj.
3675          */
3676         n = ztest_random(regions) * stride + ztest_random(width);
3677         s = 1 + ztest_random(width - 1);
3678 
3679         packoff = n * sizeof (bufwad_t);
3680         packsize = s * sizeof (bufwad_t);
3681 
3682         bigoff = n * chunksize;
3683         bigsize = s * chunksize;
3684 
3685         packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3686         bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3687 
3688         VERIFY0(dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3689 
3690         bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3691 
3692         /*
3693          * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3694          * Iteration 1 test zcopy to already referenced dbufs.
3695          * Iteration 2 test zcopy to dirty dbuf in the same txg.
3696          * Iteration 3 test zcopy to dbuf dirty in previous txg.
3697          * Iteration 4 test zcopy when dbuf is no longer dirty.
3698          * Iteration 5 test zcopy when it can't be done.
3699          * Iteration 6 one more zcopy write.
3700          */
3701         for (i = 0; i < 7; i++) {
3702                 uint64_t j;
3703                 uint64_t off;
3704 
3705                 /*
3706                  * In iteration 5 (i == 5) use arcbufs
3707                  * that don't match bigobj blksz to test
3708                  * dmu_assign_arcbuf() when it can't directly
3709                  * assign an arcbuf to a dbuf.
3710                  */
3711                 for (j = 0; j < s; j++) {
3712                         if (i != 5) {
3713                                 bigbuf_arcbufs[j] =
3714                                     dmu_request_arcbuf(bonus_db, chunksize);
3715                         } else {
3716                                 bigbuf_arcbufs[2 * j] =
3717                                     dmu_request_arcbuf(bonus_db, chunksize / 2);
3718                                 bigbuf_arcbufs[2 * j + 1] =
3719                                     dmu_request_arcbuf(bonus_db, chunksize / 2);
3720                         }
3721                 }
3722 
3723                 /*
3724                  * Get a tx for the mods to both packobj and bigobj.
3725                  */
3726                 tx = dmu_tx_create(os);
3727 
3728                 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3729                 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3730 
3731                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3732                 if (txg == 0) {
3733                         umem_free(packbuf, packsize);
3734                         umem_free(bigbuf, bigsize);
3735                         for (j = 0; j < s; j++) {
3736                                 if (i != 5) {
3737                                         dmu_return_arcbuf(bigbuf_arcbufs[j]);
3738                                 } else {
3739                                         dmu_return_arcbuf(
3740                                             bigbuf_arcbufs[2 * j]);
3741                                         dmu_return_arcbuf(
3742                                             bigbuf_arcbufs[2 * j + 1]);
3743                                 }
3744                         }
3745                         umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3746                         dmu_buf_rele(bonus_db, FTAG);
3747                         return;
3748                 }
3749 
3750                 /*
3751                  * 50% of the time don't read objects in the 1st iteration to
3752                  * test dmu_assign_arcbuf() for the case when there're no
3753                  * existing dbufs for the specified offsets.
3754                  */
3755                 if (i != 0 || ztest_random(2) != 0) {
3756                         error = dmu_read(os, packobj, packoff,
3757                             packsize, packbuf, DMU_READ_PREFETCH);
3758                         ASSERT0(error);
3759                         error = dmu_read(os, bigobj, bigoff, bigsize,
3760                             bigbuf, DMU_READ_PREFETCH);
3761                         ASSERT0(error);
3762                 }
3763                 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3764                     n, chunksize, txg);
3765 
3766                 /*
3767                  * We've verified all the old bufwads, and made new ones.
3768                  * Now write them out.
3769                  */
3770                 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3771                 if (ztest_opts.zo_verbose >= 7) {
3772                         (void) printf("writing offset %llx size %llx"
3773                             " txg %llx\n",
3774                             (u_longlong_t)bigoff,
3775                             (u_longlong_t)bigsize,
3776                             (u_longlong_t)txg);
3777                 }
3778                 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3779                         dmu_buf_t *dbt;
3780                         if (i != 5) {
3781                                 bcopy((caddr_t)bigbuf + (off - bigoff),
3782                                     bigbuf_arcbufs[j]->b_data, chunksize);
3783                         } else {
3784                                 bcopy((caddr_t)bigbuf + (off - bigoff),
3785                                     bigbuf_arcbufs[2 * j]->b_data,
3786                                     chunksize / 2);
3787                                 bcopy((caddr_t)bigbuf + (off - bigoff) +
3788                                     chunksize / 2,
3789                                     bigbuf_arcbufs[2 * j + 1]->b_data,
3790                                     chunksize / 2);
3791                         }
3792 
3793                         if (i == 1) {
3794                                 VERIFY(dmu_buf_hold(os, bigobj, off,
3795                                     FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3796                         }
3797                         if (i != 5) {
3798                                 dmu_assign_arcbuf(bonus_db, off,
3799                                     bigbuf_arcbufs[j], tx);
3800                         } else {
3801                                 dmu_assign_arcbuf(bonus_db, off,
3802                                     bigbuf_arcbufs[2 * j], tx);
3803                                 dmu_assign_arcbuf(bonus_db,
3804                                     off + chunksize / 2,
3805                                     bigbuf_arcbufs[2 * j + 1], tx);
3806                         }
3807                         if (i == 1) {
3808                                 dmu_buf_rele(dbt, FTAG);
3809                         }
3810                 }
3811                 dmu_tx_commit(tx);
3812 
3813                 /*
3814                  * Sanity check the stuff we just wrote.
3815                  */
3816                 {
3817                         void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3818                         void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3819 
3820                         VERIFY(0 == dmu_read(os, packobj, packoff,
3821                             packsize, packcheck, DMU_READ_PREFETCH));
3822                         VERIFY(0 == dmu_read(os, bigobj, bigoff,
3823                             bigsize, bigcheck, DMU_READ_PREFETCH));
3824 
3825                         ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3826                         ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3827 
3828                         umem_free(packcheck, packsize);
3829                         umem_free(bigcheck, bigsize);
3830                 }
3831                 if (i == 2) {
3832                         txg_wait_open(dmu_objset_pool(os), 0);
3833                 } else if (i == 3) {
3834                         txg_wait_synced(dmu_objset_pool(os), 0);
3835                 }
3836         }
3837 
3838         dmu_buf_rele(bonus_db, FTAG);
3839         umem_free(packbuf, packsize);
3840         umem_free(bigbuf, bigsize);
3841         umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3842 }
3843 
3844 /* ARGSUSED */
3845 void
3846 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3847 {
3848         ztest_od_t od[1];
3849         uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3850             (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3851 
3852         /*
3853          * Have multiple threads write to large offsets in an object
3854          * to verify that parallel writes to an object -- even to the
3855          * same blocks within the object -- doesn't cause any trouble.
3856          */
3857         ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
3858 
3859         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3860                 return;
3861 
3862         while (ztest_random(10) != 0)
3863                 ztest_io(zd, od[0].od_object, offset);
3864 }
3865 
3866 void
3867 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
3868 {
3869         ztest_od_t od[1];
3870         uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
3871             (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3872         uint64_t count = ztest_random(20) + 1;
3873         uint64_t blocksize = ztest_random_blocksize();
3874         void *data;
3875 
3876         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3877 
3878         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3879                 return;
3880 
3881         if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
3882                 return;
3883 
3884         ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
3885 
3886         data = umem_zalloc(blocksize, UMEM_NOFAIL);
3887 
3888         while (ztest_random(count) != 0) {
3889                 uint64_t randoff = offset + (ztest_random(count) * blocksize);
3890                 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
3891                     data) != 0)
3892                         break;
3893                 while (ztest_random(4) != 0)
3894                         ztest_io(zd, od[0].od_object, randoff);
3895         }
3896 
3897         umem_free(data, blocksize);
3898 }
3899 
3900 /*
3901  * Verify that zap_{create,destroy,add,remove,update} work as expected.
3902  */
3903 #define ZTEST_ZAP_MIN_INTS      1
3904 #define ZTEST_ZAP_MAX_INTS      4
3905 #define ZTEST_ZAP_MAX_PROPS     1000
3906 
3907 void
3908 ztest_zap(ztest_ds_t *zd, uint64_t id)
3909 {
3910         objset_t *os = zd->zd_os;
3911         ztest_od_t od[1];
3912         uint64_t object;
3913         uint64_t txg, last_txg;
3914         uint64_t value[ZTEST_ZAP_MAX_INTS];
3915         uint64_t zl_ints, zl_intsize, prop;
3916         int i, ints;
3917         dmu_tx_t *tx;
3918         char propname[100], txgname[100];
3919         int error;
3920         char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
3921 
3922         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
3923 
3924         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3925                 return;
3926 
3927         object = od[0].od_object;
3928 
3929         /*
3930          * Generate a known hash collision, and verify that
3931          * we can lookup and remove both entries.
3932          */
3933         tx = dmu_tx_create(os);
3934         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3935         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3936         if (txg == 0)
3937                 return;
3938         for (i = 0; i < 2; i++) {
3939                 value[i] = i;
3940                 VERIFY0(zap_add(os, object, hc[i], sizeof (uint64_t),
3941                     1, &value[i], tx));
3942         }
3943         for (i = 0; i < 2; i++) {
3944                 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
3945                     sizeof (uint64_t), 1, &value[i], tx));
3946                 VERIFY0(zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
3947                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3948                 ASSERT3U(zl_ints, ==, 1);
3949         }
3950         for (i = 0; i < 2; i++) {
3951                 VERIFY0(zap_remove(os, object, hc[i], tx));
3952         }
3953         dmu_tx_commit(tx);
3954 
3955         /*
3956          * Generate a buch of random entries.
3957          */
3958         ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
3959 
3960         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
3961         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
3962         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
3963         bzero(value, sizeof (value));
3964         last_txg = 0;
3965 
3966         /*
3967          * If these zap entries already exist, validate their contents.
3968          */
3969         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
3970         if (error == 0) {
3971                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3972                 ASSERT3U(zl_ints, ==, 1);
3973 
3974                 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
3975                     zl_ints, &last_txg) == 0);
3976 
3977                 VERIFY(zap_length(os, object, propname, &zl_intsize,
3978                     &zl_ints) == 0);
3979 
3980                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3981                 ASSERT3U(zl_ints, ==, ints);
3982 
3983                 VERIFY(zap_lookup(os, object, propname, zl_intsize,
3984                     zl_ints, value) == 0);
3985 
3986                 for (i = 0; i < ints; i++) {
3987                         ASSERT3U(value[i], ==, last_txg + object + i);
3988                 }
3989         } else {
3990                 ASSERT3U(error, ==, ENOENT);
3991         }
3992 
3993         /*
3994          * Atomically update two entries in our zap object.
3995          * The first is named txg_%llu, and contains the txg
3996          * in which the property was last updated.  The second
3997          * is named prop_%llu, and the nth element of its value
3998          * should be txg + object + n.
3999          */
4000         tx = dmu_tx_create(os);
4001         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4002         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4003         if (txg == 0)
4004                 return;
4005 
4006         if (last_txg > txg)
4007                 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4008 
4009         for (i = 0; i < ints; i++)
4010                 value[i] = txg + object + i;
4011 
4012         VERIFY0(zap_update(os, object, txgname, sizeof (uint64_t),
4013             1, &txg, tx));
4014         VERIFY0(zap_update(os, object, propname, sizeof (uint64_t),
4015             ints, value, tx));
4016 
4017         dmu_tx_commit(tx);
4018 
4019         /*
4020          * Remove a random pair of entries.
4021          */
4022         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4023         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4024         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4025 
4026         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4027 
4028         if (error == ENOENT)
4029                 return;
4030 
4031         ASSERT0(error);
4032 
4033         tx = dmu_tx_create(os);
4034         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4035         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4036         if (txg == 0)
4037                 return;
4038         VERIFY0(zap_remove(os, object, txgname, tx));
4039         VERIFY0(zap_remove(os, object, propname, tx));
4040         dmu_tx_commit(tx);
4041 }
4042 
4043 /*
4044  * Testcase to test the upgrading of a microzap to fatzap.
4045  */
4046 void
4047 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4048 {
4049         objset_t *os = zd->zd_os;
4050         ztest_od_t od[1];
4051         uint64_t object, txg;
4052 
4053         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4054 
4055         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4056                 return;
4057 
4058         object = od[0].od_object;
4059 
4060         /*
4061          * Add entries to this ZAP and make sure it spills over
4062          * and gets upgraded to a fatzap. Also, since we are adding
4063          * 2050 entries we should see ptrtbl growth and leaf-block split.
4064          */
4065         for (int i = 0; i < 2050; i++) {
4066                 char name[MAXNAMELEN];
4067                 uint64_t value = i;
4068                 dmu_tx_t *tx;
4069                 int error;
4070 
4071                 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4072                     id, value);
4073 
4074                 tx = dmu_tx_create(os);
4075                 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4076                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4077                 if (txg == 0)
4078                         return;
4079                 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4080                     &value, tx);
4081                 ASSERT(error == 0 || error == EEXIST);
4082                 dmu_tx_commit(tx);
4083         }
4084 }
4085 
4086 /* ARGSUSED */
4087 void
4088 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4089 {
4090         objset_t *os = zd->zd_os;
4091         ztest_od_t od[1];
4092         uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4093         dmu_tx_t *tx;
4094         int i, namelen, error;
4095         int micro = ztest_random(2);
4096         char name[20], string_value[20];
4097         void *data;
4098 
4099         ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4100 
4101         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4102                 return;
4103 
4104         object = od[0].od_object;
4105 
4106         /*
4107          * Generate a random name of the form 'xxx.....' where each
4108          * x is a random printable character and the dots are dots.
4109          * There are 94 such characters, and the name length goes from
4110          * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4111          */
4112         namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4113 
4114         for (i = 0; i < 3; i++)
4115                 name[i] = '!' + ztest_random('~' - '!' + 1);
4116         for (; i < namelen - 1; i++)
4117                 name[i] = '.';
4118         name[i] = '\0';
4119 
4120         if ((namelen & 1) || micro) {
4121                 wsize = sizeof (txg);
4122                 wc = 1;
4123                 data = &txg;
4124         } else {
4125                 wsize = 1;
4126                 wc = namelen;
4127                 data = string_value;
4128         }
4129 
4130         count = -1ULL;
4131         VERIFY(zap_count(os, object, &count) == 0);
4132         ASSERT(count != -1ULL);
4133 
4134         /*
4135          * Select an operation: length, lookup, add, update, remove.
4136          */
4137         i = ztest_random(5);
4138 
4139         if (i >= 2) {
4140                 tx = dmu_tx_create(os);
4141                 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4142                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4143                 if (txg == 0)
4144                         return;
4145                 bcopy(name, string_value, namelen);
4146         } else {
4147                 tx = NULL;
4148                 txg = 0;
4149                 bzero(string_value, namelen);
4150         }
4151 
4152         switch (i) {
4153 
4154         case 0:
4155                 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4156                 if (error == 0) {
4157                         ASSERT3U(wsize, ==, zl_wsize);
4158                         ASSERT3U(wc, ==, zl_wc);
4159                 } else {
4160                         ASSERT3U(error, ==, ENOENT);
4161                 }
4162                 break;
4163 
4164         case 1:
4165                 error = zap_lookup(os, object, name, wsize, wc, data);
4166                 if (error == 0) {
4167                         if (data == string_value &&
4168                             bcmp(name, data, namelen) != 0)
4169                                 fatal(0, "name '%s' != val '%s' len %d",
4170                                     name, data, namelen);
4171                 } else {
4172                         ASSERT3U(error, ==, ENOENT);
4173                 }
4174                 break;
4175 
4176         case 2:
4177                 error = zap_add(os, object, name, wsize, wc, data, tx);
4178                 ASSERT(error == 0 || error == EEXIST);
4179                 break;
4180 
4181         case 3:
4182                 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4183                 break;
4184 
4185         case 4:
4186                 error = zap_remove(os, object, name, tx);
4187                 ASSERT(error == 0 || error == ENOENT);
4188                 break;
4189         }
4190 
4191         if (tx != NULL)
4192                 dmu_tx_commit(tx);
4193 }
4194 
4195 /*
4196  * Commit callback data.
4197  */
4198 typedef struct ztest_cb_data {
4199         list_node_t             zcd_node;
4200         uint64_t                zcd_txg;
4201         int                     zcd_expected_err;
4202         boolean_t               zcd_added;
4203         boolean_t               zcd_called;
4204         spa_t                   *zcd_spa;
4205 } ztest_cb_data_t;
4206 
4207 /* This is the actual commit callback function */
4208 static void
4209 ztest_commit_callback(void *arg, int error)
4210 {
4211         ztest_cb_data_t *data = arg;
4212         uint64_t synced_txg;
4213 
4214         VERIFY(data != NULL);
4215         VERIFY3S(data->zcd_expected_err, ==, error);
4216         VERIFY(!data->zcd_called);
4217 
4218         synced_txg = spa_last_synced_txg(data->zcd_spa);
4219         if (data->zcd_txg > synced_txg)
4220                 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4221                     ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4222                     synced_txg);
4223 
4224         data->zcd_called = B_TRUE;
4225 
4226         if (error == ECANCELED) {
4227                 ASSERT0(data->zcd_txg);
4228                 ASSERT(!data->zcd_added);
4229 
4230                 /*
4231                  * The private callback data should be destroyed here, but
4232                  * since we are going to check the zcd_called field after
4233                  * dmu_tx_abort(), we will destroy it there.
4234                  */
4235                 return;
4236         }
4237 
4238         /* Was this callback added to the global callback list? */
4239         if (!data->zcd_added)
4240                 goto out;
4241 
4242         ASSERT3U(data->zcd_txg, !=, 0);
4243 
4244         /* Remove our callback from the list */
4245         (void) mutex_lock(&zcl.zcl_callbacks_lock);
4246         list_remove(&zcl.zcl_callbacks, data);
4247         (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4248 
4249 out:
4250         umem_free(data, sizeof (ztest_cb_data_t));
4251 }
4252 
4253 /* Allocate and initialize callback data structure */
4254 static ztest_cb_data_t *
4255 ztest_create_cb_data(objset_t *os, uint64_t txg)
4256 {
4257         ztest_cb_data_t *cb_data;
4258 
4259         cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4260 
4261         cb_data->zcd_txg = txg;
4262         cb_data->zcd_spa = dmu_objset_spa(os);
4263 
4264         return (cb_data);
4265 }
4266 
4267 /*
4268  * If a number of txgs equal to this threshold have been created after a commit
4269  * callback has been registered but not called, then we assume there is an
4270  * implementation bug.
4271  */
4272 #define ZTEST_COMMIT_CALLBACK_THRESH    (TXG_CONCURRENT_STATES + 2)
4273 
4274 /*
4275  * Commit callback test.
4276  */
4277 void
4278 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4279 {
4280         objset_t *os = zd->zd_os;
4281         ztest_od_t od[1];
4282         dmu_tx_t *tx;
4283         ztest_cb_data_t *cb_data[3], *tmp_cb;
4284         uint64_t old_txg, txg;
4285         int i, error;
4286 
4287         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4288 
4289         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4290                 return;
4291 
4292         tx = dmu_tx_create(os);
4293 
4294         cb_data[0] = ztest_create_cb_data(os, 0);
4295         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4296 
4297         dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4298 
4299         /* Every once in a while, abort the transaction on purpose */
4300         if (ztest_random(100) == 0)
4301                 error = -1;
4302 
4303         if (!error)
4304                 error = dmu_tx_assign(tx, TXG_NOWAIT);
4305 
4306         txg = error ? 0 : dmu_tx_get_txg(tx);
4307 
4308         cb_data[0]->zcd_txg = txg;
4309         cb_data[1] = ztest_create_cb_data(os, txg);
4310         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4311 
4312         if (error) {
4313                 /*
4314                  * It's not a strict requirement to call the registered
4315                  * callbacks from inside dmu_tx_abort(), but that's what
4316                  * it's supposed to happen in the current implementation
4317                  * so we will check for that.
4318                  */
4319                 for (i = 0; i < 2; i++) {
4320                         cb_data[i]->zcd_expected_err = ECANCELED;
4321                         VERIFY(!cb_data[i]->zcd_called);
4322                 }
4323 
4324                 dmu_tx_abort(tx);
4325 
4326                 for (i = 0; i < 2; i++) {
4327                         VERIFY(cb_data[i]->zcd_called);
4328                         umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4329                 }
4330 
4331                 return;
4332         }
4333 
4334         cb_data[2] = ztest_create_cb_data(os, txg);
4335         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4336 
4337         /*
4338          * Read existing data to make sure there isn't a future leak.
4339          */
4340         VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4341             &old_txg, DMU_READ_PREFETCH));
4342 
4343         if (old_txg > txg)
4344                 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4345                     old_txg, txg);
4346 
4347         dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4348 
4349         (void) mutex_lock(&zcl.zcl_callbacks_lock);
4350 
4351         /*
4352          * Since commit callbacks don't have any ordering requirement and since
4353          * it is theoretically possible for a commit callback to be called
4354          * after an arbitrary amount of time has elapsed since its txg has been
4355          * synced, it is difficult to reliably determine whether a commit
4356          * callback hasn't been called due to high load or due to a flawed
4357          * implementation.
4358          *
4359          * In practice, we will assume that if after a certain number of txgs a
4360          * commit callback hasn't been called, then most likely there's an
4361          * implementation bug..
4362          */
4363         tmp_cb = list_head(&zcl.zcl_callbacks);
4364         if (tmp_cb != NULL &&
4365             tmp_cb->zcd_txg > txg - ZTEST_COMMIT_CALLBACK_THRESH) {
4366                 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4367                     PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4368         }
4369 
4370         /*
4371          * Let's find the place to insert our callbacks.
4372          *
4373          * Even though the list is ordered by txg, it is possible for the
4374          * insertion point to not be the end because our txg may already be
4375          * quiescing at this point and other callbacks in the open txg
4376          * (from other objsets) may have sneaked in.
4377          */
4378         tmp_cb = list_tail(&zcl.zcl_callbacks);
4379         while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4380                 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4381 
4382         /* Add the 3 callbacks to the list */
4383         for (i = 0; i < 3; i++) {
4384                 if (tmp_cb == NULL)
4385                         list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4386                 else
4387                         list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4388                             cb_data[i]);
4389 
4390                 cb_data[i]->zcd_added = B_TRUE;
4391                 VERIFY(!cb_data[i]->zcd_called);
4392 
4393                 tmp_cb = cb_data[i];
4394         }
4395 
4396         (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4397 
4398         dmu_tx_commit(tx);
4399 }
4400 
4401 /* ARGSUSED */
4402 void
4403 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4404 {
4405         zfs_prop_t proplist[] = {
4406                 ZFS_PROP_CHECKSUM,
4407                 ZFS_PROP_COMPRESSION,
4408                 ZFS_PROP_COPIES,
4409                 ZFS_PROP_DEDUP
4410         };
4411 
4412         (void) rw_rdlock(&ztest_name_lock);
4413 
4414         for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4415                 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4416                     ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4417 
4418         (void) rw_unlock(&ztest_name_lock);
4419 }
4420 
4421 /* ARGSUSED */
4422 void
4423 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4424 {
4425         nvlist_t *props = NULL;
4426 
4427         (void) rw_rdlock(&ztest_name_lock);
4428 
4429         (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4430             ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4431 
4432         VERIFY0(spa_prop_get(ztest_spa, &props));
4433 
4434         if (ztest_opts.zo_verbose >= 6)
4435                 dump_nvlist(props, 4);
4436 
4437         nvlist_free(props);
4438 
4439         (void) rw_unlock(&ztest_name_lock);
4440 }
4441 
4442 /*
4443  * Test snapshot hold/release and deferred destroy.
4444  */
4445 void
4446 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4447 {
4448         int error;
4449         objset_t *os = zd->zd_os;
4450         objset_t *origin;
4451         char snapname[100];
4452         char fullname[100];
4453         char clonename[100];
4454         char tag[100];
4455         char osname[MAXNAMELEN];
4456 
4457         (void) rw_rdlock(&ztest_name_lock);
4458 
4459         dmu_objset_name(os, osname);
4460 
4461         (void) snprintf(snapname, 100, "sh1_%llu", id);
4462         (void) snprintf(fullname, 100, "%s@%s", osname, snapname);
4463         (void) snprintf(clonename, 100, "%s/ch1_%llu", osname, id);
4464         (void) snprintf(tag, 100, "%tag_%llu", id);
4465 
4466         /*
4467          * Clean up from any previous run.
4468          */
4469         (void) dmu_objset_destroy(clonename, B_FALSE);
4470         (void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4471         (void) dmu_objset_destroy(fullname, B_FALSE);
4472 
4473         /*
4474          * Create snapshot, clone it, mark snap for deferred destroy,
4475          * destroy clone, verify snap was also destroyed.
4476          */
4477         error = dmu_objset_snapshot_one(osname, snapname);
4478         if (error) {
4479                 if (error == ENOSPC) {
4480                         ztest_record_enospc("dmu_objset_snapshot");
4481                         goto out;
4482                 }
4483                 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4484         }
4485 
4486         error = dmu_objset_hold(fullname, FTAG, &origin);
4487         if (error)
4488                 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4489 
4490         error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0);
4491         dmu_objset_rele(origin, FTAG);
4492         if (error) {
4493                 if (error == ENOSPC) {
4494                         ztest_record_enospc("dmu_objset_clone");
4495                         goto out;
4496                 }
4497                 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4498         }
4499 
4500         error = dmu_objset_destroy(fullname, B_TRUE);
4501         if (error) {
4502                 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4503                     fullname, error);
4504         }
4505 
4506         error = dmu_objset_destroy(clonename, B_FALSE);
4507         if (error)
4508                 fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error);
4509 
4510         error = dmu_objset_hold(fullname, FTAG, &origin);
4511         if (error != ENOENT)
4512                 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4513 
4514         /*
4515          * Create snapshot, add temporary hold, verify that we can't
4516          * destroy a held snapshot, mark for deferred destroy,
4517          * release hold, verify snapshot was destroyed.
4518          */
4519         error = dmu_objset_snapshot_one(osname, snapname);
4520         if (error) {
4521                 if (error == ENOSPC) {
4522                         ztest_record_enospc("dmu_objset_snapshot");
4523                         goto out;
4524                 }
4525                 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4526         }
4527 
4528         error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE,
4529             B_TRUE, -1);
4530         if (error)
4531                 fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4532 
4533         error = dmu_objset_destroy(fullname, B_FALSE);
4534         if (error != EBUSY) {
4535                 fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4536                     fullname, error);
4537         }
4538 
4539         error = dmu_objset_destroy(fullname, B_TRUE);
4540         if (error) {
4541                 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4542                     fullname, error);
4543         }
4544 
4545         error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4546         if (error)
4547                 fatal(0, "dsl_dataset_user_release(%s)", fullname, tag);
4548 
4549         VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT);
4550 
4551 out:
4552         (void) rw_unlock(&ztest_name_lock);
4553 }
4554 
4555 /*
4556  * Inject random faults into the on-disk data.
4557  */
4558 /* ARGSUSED */
4559 void
4560 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4561 {
4562         ztest_shared_t *zs = ztest_shared;
4563         spa_t *spa = ztest_spa;
4564         int fd;
4565         uint64_t offset;
4566         uint64_t leaves;
4567         uint64_t bad = 0x1990c0ffeedecade;
4568         uint64_t top, leaf;
4569         char path0[MAXPATHLEN];
4570         char pathrand[MAXPATHLEN];
4571         size_t fsize;
4572         int bshift = SPA_MAXBLOCKSHIFT + 2;     /* don't scrog all labels */
4573         int iters = 1000;
4574         int maxfaults;
4575         int mirror_save;
4576         vdev_t *vd0 = NULL;
4577         uint64_t guid0 = 0;
4578         boolean_t islog = B_FALSE;
4579 
4580         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4581         maxfaults = MAXFAULTS();
4582         leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4583         mirror_save = zs->zs_mirrors;
4584         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4585 
4586         ASSERT(leaves >= 1);
4587 
4588         /*
4589          * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4590          */
4591         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4592 
4593         if (ztest_random(2) == 0) {
4594                 /*
4595                  * Inject errors on a normal data device or slog device.
4596                  */
4597                 top = ztest_random_vdev_top(spa, B_TRUE);
4598                 leaf = ztest_random(leaves) + zs->zs_splits;
4599 
4600                 /*
4601                  * Generate paths to the first leaf in this top-level vdev,
4602                  * and to the random leaf we selected.  We'll induce transient
4603                  * write failures and random online/offline activity on leaf 0,
4604                  * and we'll write random garbage to the randomly chosen leaf.
4605                  */
4606                 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4607                     ztest_opts.zo_dir, ztest_opts.zo_pool,
4608                     top * leaves + zs->zs_splits);
4609                 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4610                     ztest_opts.zo_dir, ztest_opts.zo_pool,
4611                     top * leaves + leaf);
4612 
4613                 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4614                 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4615                         islog = B_TRUE;
4616 
4617                 if (vd0 != NULL && maxfaults != 1) {
4618                         /*
4619                          * Make vd0 explicitly claim to be unreadable,
4620                          * or unwriteable, or reach behind its back
4621                          * and close the underlying fd.  We can do this if
4622                          * maxfaults == 0 because we'll fail and reexecute,
4623                          * and we can do it if maxfaults >= 2 because we'll
4624                          * have enough redundancy.  If maxfaults == 1, the
4625                          * combination of this with injection of random data
4626                          * corruption below exceeds the pool's fault tolerance.
4627                          */
4628                         vdev_file_t *vf = vd0->vdev_tsd;
4629 
4630                         if (vf != NULL && ztest_random(3) == 0) {
4631                                 (void) close(vf->vf_vnode->v_fd);
4632                                 vf->vf_vnode->v_fd = -1;
4633                         } else if (ztest_random(2) == 0) {
4634                                 vd0->vdev_cant_read = B_TRUE;
4635                         } else {
4636                                 vd0->vdev_cant_write = B_TRUE;
4637                         }
4638                         guid0 = vd0->vdev_guid;
4639                 }
4640         } else {
4641                 /*
4642                  * Inject errors on an l2cache device.
4643                  */
4644                 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4645 
4646                 if (sav->sav_count == 0) {
4647                         spa_config_exit(spa, SCL_STATE, FTAG);
4648                         return;
4649                 }
4650                 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4651                 guid0 = vd0->vdev_guid;
4652                 (void) strcpy(path0, vd0->vdev_path);
4653                 (void) strcpy(pathrand, vd0->vdev_path);
4654 
4655                 leaf = 0;
4656                 leaves = 1;
4657                 maxfaults = INT_MAX;    /* no limit on cache devices */
4658         }
4659 
4660         spa_config_exit(spa, SCL_STATE, FTAG);
4661 
4662         /*
4663          * If we can tolerate two or more faults, or we're dealing
4664          * with a slog, randomly online/offline vd0.
4665          */
4666         if ((maxfaults >= 2 || islog) && guid0 != 0) {
4667                 if (ztest_random(10) < 6) {
4668                         int flags = (ztest_random(2) == 0 ?
4669                             ZFS_OFFLINE_TEMPORARY : 0);
4670 
4671                         /*
4672                          * We have to grab the zs_name_lock as writer to
4673                          * prevent a race between offlining a slog and
4674                          * destroying a dataset. Offlining the slog will
4675                          * grab a reference on the dataset which may cause
4676                          * dmu_objset_destroy() to fail with EBUSY thus
4677                          * leaving the dataset in an inconsistent state.
4678                          */
4679                         if (islog)
4680                                 (void) rw_wrlock(&ztest_name_lock);
4681 
4682                         VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4683 
4684                         if (islog)
4685                                 (void) rw_unlock(&ztest_name_lock);
4686                 } else {
4687                         (void) vdev_online(spa, guid0, 0, NULL);
4688                 }
4689         }
4690 
4691         if (maxfaults == 0)
4692                 return;
4693 
4694         /*
4695          * We have at least single-fault tolerance, so inject data corruption.
4696          */
4697         fd = open(pathrand, O_RDWR);
4698 
4699         if (fd == -1)   /* we hit a gap in the device namespace */
4700                 return;
4701 
4702         fsize = lseek(fd, 0, SEEK_END);
4703 
4704         while (--iters != 0) {
4705                 offset = ztest_random(fsize / (leaves << bshift)) *
4706                     (leaves << bshift) + (leaf << bshift) +
4707                     (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4708 
4709                 if (offset >= fsize)
4710                         continue;
4711 
4712                 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4713                 if (mirror_save != zs->zs_mirrors) {
4714                         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4715                         (void) close(fd);
4716                         return;
4717                 }
4718 
4719                 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4720                         fatal(1, "can't inject bad word at 0x%llx in %s",
4721                             offset, pathrand);
4722 
4723                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4724 
4725                 if (ztest_opts.zo_verbose >= 7)
4726                         (void) printf("injected bad word into %s,"
4727                             " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4728         }
4729 
4730         (void) close(fd);
4731 }
4732 
4733 /*
4734  * Verify that DDT repair works as expected.
4735  */
4736 void
4737 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4738 {
4739         ztest_shared_t *zs = ztest_shared;
4740         spa_t *spa = ztest_spa;
4741         objset_t *os = zd->zd_os;
4742         ztest_od_t od[1];
4743         uint64_t object, blocksize, txg, pattern, psize;
4744         enum zio_checksum checksum = spa_dedup_checksum(spa);
4745         dmu_buf_t *db;
4746         dmu_tx_t *tx;
4747         void *buf;
4748         blkptr_t blk;
4749         int copies = 2 * ZIO_DEDUPDITTO_MIN;
4750 
4751         blocksize = ztest_random_blocksize();
4752         blocksize = MIN(blocksize, 2048);       /* because we write so many */
4753 
4754         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4755 
4756         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4757                 return;
4758 
4759         /*
4760          * Take the name lock as writer to prevent anyone else from changing
4761          * the pool and dataset properies we need to maintain during this test.
4762          */
4763         (void) rw_wrlock(&ztest_name_lock);
4764 
4765         if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4766             B_FALSE) != 0 ||
4767             ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4768             B_FALSE) != 0) {
4769                 (void) rw_unlock(&ztest_name_lock);
4770                 return;
4771         }
4772 
4773         object = od[0].od_object;
4774         blocksize = od[0].od_blocksize;
4775         pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4776 
4777         ASSERT(object != 0);
4778 
4779         tx = dmu_tx_create(os);
4780         dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4781         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4782         if (txg == 0) {
4783                 (void) rw_unlock(&ztest_name_lock);
4784                 return;
4785         }
4786 
4787         /*
4788          * Write all the copies of our block.
4789          */
4790         for (int i = 0; i < copies; i++) {
4791                 uint64_t offset = i * blocksize;
4792                 VERIFY(dmu_buf_hold(os, object, offset, FTAG, &db,
4793                     DMU_READ_NO_PREFETCH) == 0);
4794                 ASSERT(db->db_offset == offset);
4795                 ASSERT(db->db_size == blocksize);
4796                 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4797                     ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4798                 dmu_buf_will_fill(db, tx);
4799                 ztest_pattern_set(db->db_data, db->db_size, pattern);
4800                 dmu_buf_rele(db, FTAG);
4801         }
4802 
4803         dmu_tx_commit(tx);
4804         txg_wait_synced(spa_get_dsl(spa), txg);
4805 
4806         /*
4807          * Find out what block we got.
4808          */
4809         VERIFY(dmu_buf_hold(os, object, 0, FTAG, &db,
4810             DMU_READ_NO_PREFETCH) == 0);
4811         blk = *((dmu_buf_impl_t *)db)->db_blkptr;
4812         dmu_buf_rele(db, FTAG);
4813 
4814         /*
4815          * Damage the block.  Dedup-ditto will save us when we read it later.
4816          */
4817         psize = BP_GET_PSIZE(&blk);
4818         buf = zio_buf_alloc(psize);
4819         ztest_pattern_set(buf, psize, ~pattern);
4820 
4821         (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
4822             buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
4823             ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
4824 
4825         zio_buf_free(buf, psize);
4826 
4827         (void) rw_unlock(&ztest_name_lock);
4828 }
4829 
4830 /*
4831  * Scrub the pool.
4832  */
4833 /* ARGSUSED */
4834 void
4835 ztest_scrub(ztest_ds_t *zd, uint64_t id)
4836 {
4837         spa_t *spa = ztest_spa;
4838 
4839         (void) spa_scan(spa, POOL_SCAN_SCRUB);
4840         (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
4841         (void) spa_scan(spa, POOL_SCAN_SCRUB);
4842 }
4843 
4844 /*
4845  * Change the guid for the pool.
4846  */
4847 /* ARGSUSED */
4848 void
4849 ztest_reguid(ztest_ds_t *zd, uint64_t id)
4850 {
4851         spa_t *spa = ztest_spa;
4852         uint64_t orig, load;
4853 
4854         orig = spa_guid(spa);
4855         load = spa_load_guid(spa);
4856         if (spa_change_guid(spa) != 0)
4857                 return;
4858 
4859         if (ztest_opts.zo_verbose >= 3) {
4860                 (void) printf("Changed guid old %llu -> %llu\n",
4861                     (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
4862         }
4863 
4864         VERIFY3U(orig, !=, spa_guid(spa));
4865         VERIFY3U(load, ==, spa_load_guid(spa));
4866 }
4867 
4868 /*
4869  * Rename the pool to a different name and then rename it back.
4870  */
4871 /* ARGSUSED */
4872 void
4873 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
4874 {
4875         char *oldname, *newname;
4876         spa_t *spa;
4877 
4878         (void) rw_wrlock(&ztest_name_lock);
4879 
4880         oldname = ztest_opts.zo_pool;
4881         newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
4882         (void) strcpy(newname, oldname);
4883         (void) strcat(newname, "_tmp");
4884 
4885         /*
4886          * Do the rename
4887          */
4888         VERIFY0(spa_rename(oldname, newname));
4889 
4890         /*
4891          * Try to open it under the old name, which shouldn't exist
4892          */
4893         VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
4894 
4895         /*
4896          * Open it under the new name and make sure it's still the same spa_t.
4897          */
4898         VERIFY0(spa_open(newname, &spa, FTAG));
4899 
4900         ASSERT(spa == ztest_spa);
4901         spa_close(spa, FTAG);
4902 
4903         /*
4904          * Rename it back to the original
4905          */
4906         VERIFY0(spa_rename(newname, oldname));
4907 
4908         /*
4909          * Make sure it can still be opened
4910          */
4911         VERIFY0(spa_open(oldname, &spa, FTAG));
4912 
4913         ASSERT(spa == ztest_spa);
4914         spa_close(spa, FTAG);
4915 
4916         umem_free(newname, strlen(newname) + 1);
4917 
4918         (void) rw_unlock(&ztest_name_lock);
4919 }
4920 
4921 /*
4922  * Verify pool integrity by running zdb.
4923  */
4924 static void
4925 ztest_run_zdb(char *pool)
4926 {
4927         int status;
4928         char zdb[MAXPATHLEN + MAXNAMELEN + 20];
4929         char zbuf[1024];
4930         char *bin;
4931         char *ztest;
4932         char *isa;
4933         int isalen;
4934         FILE *fp;
4935 
4936         (void) realpath(getexecname(), zdb);
4937 
4938         /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
4939         bin = strstr(zdb, "/usr/bin/");
4940         ztest = strstr(bin, "/ztest");
4941         isa = bin + 8;
4942         isalen = ztest - isa;
4943         isa = strdup(isa);
4944         /* LINTED */
4945         (void) sprintf(bin,
4946             "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
4947             isalen,
4948             isa,
4949             ztest_opts.zo_verbose >= 3 ? "s" : "",
4950             ztest_opts.zo_verbose >= 4 ? "v" : "",
4951             spa_config_path,
4952             pool);
4953         free(isa);
4954 
4955         if (ztest_opts.zo_verbose >= 5)
4956                 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
4957 
4958         fp = popen(zdb, "r");
4959 
4960         while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
4961                 if (ztest_opts.zo_verbose >= 3)
4962                         (void) printf("%s", zbuf);
4963 
4964         status = pclose(fp);
4965 
4966         if (status == 0)
4967                 return;
4968 
4969         ztest_dump_core = 0;
4970         if (WIFEXITED(status))
4971                 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
4972         else
4973                 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
4974 }
4975 
4976 static void
4977 ztest_walk_pool_directory(char *header)
4978 {
4979         spa_t *spa = NULL;
4980 
4981         if (ztest_opts.zo_verbose >= 6)
4982                 (void) printf("%s\n", header);
4983 
4984         mutex_enter(&spa_namespace_lock);
4985         while ((spa = spa_next(spa)) != NULL)
4986                 if (ztest_opts.zo_verbose >= 6)
4987                         (void) printf("\t%s\n", spa_name(spa));
4988         mutex_exit(&spa_namespace_lock);
4989 }
4990 
4991 static void
4992 ztest_spa_import_export(char *oldname, char *newname)
4993 {
4994         nvlist_t *config, *newconfig;
4995         uint64_t pool_guid;
4996         spa_t *spa;
4997 
4998         if (ztest_opts.zo_verbose >= 4) {
4999                 (void) printf("import/export: old = %s, new = %s\n",
5000                     oldname, newname);
5001         }
5002 
5003         /*
5004          * Clean up from previous runs.
5005          */
5006         (void) spa_destroy(newname);
5007 
5008         /*
5009          * Get the pool's configuration and guid.
5010          */
5011         VERIFY0(spa_open(oldname, &spa, FTAG));
5012 
5013         /*
5014          * Kick off a scrub to tickle scrub/export races.
5015          */
5016         if (ztest_random(2) == 0)
5017                 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5018 
5019         pool_guid = spa_guid(spa);
5020         spa_close(spa, FTAG);
5021 
5022         ztest_walk_pool_directory("pools before export");
5023 
5024         /*
5025          * Export it.
5026          */
5027         VERIFY0(spa_export(oldname, &config, B_FALSE, B_FALSE));
5028 
5029         ztest_walk_pool_directory("pools after export");
5030 
5031         /*
5032          * Try to import it.
5033          */
5034         newconfig = spa_tryimport(config);
5035         ASSERT(newconfig != NULL);
5036         nvlist_free(newconfig);
5037 
5038         /*
5039          * Import it under the new name.
5040          */
5041         VERIFY0(spa_import(newname, config, NULL, 0));
5042 
5043         ztest_walk_pool_directory("pools after import");
5044 
5045         /*
5046          * Try to import it again -- should fail with EEXIST.
5047          */
5048         VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5049 
5050         /*
5051          * Try to import it under a different name -- should fail with EEXIST.
5052          */
5053         VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5054 
5055         /*
5056          * Verify that the pool is no longer visible under the old name.
5057          */
5058         VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5059 
5060         /*
5061          * Verify that we can open and close the pool using the new name.
5062          */
5063         VERIFY0(spa_open(newname, &spa, FTAG));
5064         ASSERT(pool_guid == spa_guid(spa));
5065         spa_close(spa, FTAG);
5066 
5067         nvlist_free(config);
5068 }
5069 
5070 static void
5071 ztest_resume(spa_t *spa)
5072 {
5073         if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5074                 (void) printf("resuming from suspended state\n");
5075         spa_vdev_state_enter(spa, SCL_NONE);
5076         vdev_clear(spa, NULL);
5077         (void) spa_vdev_state_exit(spa, NULL, 0);
5078         (void) zio_resume(spa);
5079 }
5080 
5081 static void *
5082 ztest_resume_thread(void *arg)
5083 {
5084         spa_t *spa = arg;
5085 
5086         while (!ztest_exiting) {
5087                 if (spa_suspended(spa))
5088                         ztest_resume(spa);
5089                 (void) poll(NULL, 0, 100);
5090         }
5091         return (NULL);
5092 }
5093 
5094 static void *
5095 ztest_deadman_thread(void *arg)
5096 {
5097         ztest_shared_t *zs = arg;
5098         int grace = 300;
5099         hrtime_t delta;
5100 
5101         delta = (zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + grace;
5102 
5103         (void) poll(NULL, 0, (int)(1000 * delta));
5104 
5105         fatal(0, "failed to complete within %d seconds of deadline", grace);
5106 
5107         return (NULL);
5108 }
5109 
5110 static void
5111 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5112 {
5113         ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5114         ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5115         hrtime_t functime = gethrtime();
5116 
5117         for (int i = 0; i < zi->zi_iters; i++)
5118                 zi->zi_func(zd, id);
5119 
5120         functime = gethrtime() - functime;
5121 
5122         atomic_add_64(&zc->zc_count, 1);
5123         atomic_add_64(&zc->zc_time, functime);
5124 
5125         if (ztest_opts.zo_verbose >= 4) {
5126                 Dl_info dli;
5127                 (void) dladdr((void *)zi->zi_func, &dli);
5128                 (void) printf("%6.2f sec in %s\n",
5129                     (double)functime / NANOSEC, dli.dli_sname);
5130         }
5131 }
5132 
5133 static void *
5134 ztest_thread(void *arg)
5135 {
5136         int rand;
5137         uint64_t id = (uintptr_t)arg;
5138         ztest_shared_t *zs = ztest_shared;
5139         uint64_t call_next;
5140         hrtime_t now;
5141         ztest_info_t *zi;
5142         ztest_shared_callstate_t *zc;
5143 
5144         while ((now = gethrtime()) < zs->zs_thread_stop) {
5145                 /*
5146                  * See if it's time to force a crash.
5147                  */
5148                 if (now > zs->zs_thread_kill)
5149                         ztest_kill(zs);
5150 
5151                 /*
5152                  * If we're getting ENOSPC with some regularity, stop.
5153                  */
5154                 if (zs->zs_enospc_count > 10)
5155                         break;
5156 
5157                 /*
5158                  * Pick a random function to execute.
5159                  */
5160                 rand = ztest_random(ZTEST_FUNCS);
5161                 zi = &ztest_info[rand];
5162                 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5163                 call_next = zc->zc_next;
5164 
5165                 if (now >= call_next &&
5166                     atomic_cas_64(&zc->zc_next, call_next, call_next +
5167                     ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5168                         ztest_execute(rand, zi, id);
5169                 }
5170         }
5171 
5172         return (NULL);
5173 }
5174 
5175 static void
5176 ztest_dataset_name(char *dsname, char *pool, int d)
5177 {
5178         (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5179 }
5180 
5181 static void
5182 ztest_dataset_destroy(int d)
5183 {
5184         char name[MAXNAMELEN];
5185 
5186         ztest_dataset_name(name, ztest_opts.zo_pool, d);
5187 
5188         if (ztest_opts.zo_verbose >= 3)
5189                 (void) printf("Destroying %s to free up space\n", name);
5190 
5191         /*
5192          * Cleanup any non-standard clones and snapshots.  In general,
5193          * ztest thread t operates on dataset (t % zopt_datasets),
5194          * so there may be more than one thing to clean up.
5195          */
5196         for (int t = d; t < ztest_opts.zo_threads;
5197             t += ztest_opts.zo_datasets) {
5198                 ztest_dsl_dataset_cleanup(name, t);
5199         }
5200 
5201         (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5202             DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5203 }
5204 
5205 static void
5206 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5207 {
5208         uint64_t usedobjs, dirobjs, scratch;
5209 
5210         /*
5211          * ZTEST_DIROBJ is the object directory for the entire dataset.
5212          * Therefore, the number of objects in use should equal the
5213          * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5214          * If not, we have an object leak.
5215          *
5216          * Note that we can only check this in ztest_dataset_open(),
5217          * when the open-context and syncing-context values agree.
5218          * That's because zap_count() returns the open-context value,
5219          * while dmu_objset_space() returns the rootbp fill count.
5220          */
5221         VERIFY0(zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5222         dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5223         ASSERT3U(dirobjs + 1, ==, usedobjs);
5224 }
5225 
5226 static int
5227 ztest_dataset_open(int d)
5228 {
5229         ztest_ds_t *zd = &ztest_ds[d];
5230         uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5231         objset_t *os;
5232         zilog_t *zilog;
5233         char name[MAXNAMELEN];
5234         int error;
5235 
5236         ztest_dataset_name(name, ztest_opts.zo_pool, d);
5237 
5238         (void) rw_rdlock(&ztest_name_lock);
5239 
5240         error = ztest_dataset_create(name);
5241         if (error == ENOSPC) {
5242                 (void) rw_unlock(&ztest_name_lock);
5243                 ztest_record_enospc(FTAG);
5244                 return (error);
5245         }
5246         ASSERT(error == 0 || error == EEXIST);
5247 
5248         VERIFY0(dmu_objset_hold(name, zd, &os));
5249         (void) rw_unlock(&ztest_name_lock);
5250 
5251         ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5252 
5253         zilog = zd->zd_zilog;
5254 
5255         if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5256             zilog->zl_header->zh_claim_lr_seq < committed_seq)
5257                 fatal(0, "missing log records: claimed %llu < committed %llu",
5258                     zilog->zl_header->zh_claim_lr_seq, committed_seq);
5259 
5260         ztest_dataset_dirobj_verify(zd);
5261 
5262         zil_replay(os, zd, ztest_replay_vector);
5263 
5264         ztest_dataset_dirobj_verify(zd);
5265 
5266         if (ztest_opts.zo_verbose >= 6)
5267                 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5268                     zd->zd_name,
5269                     (u_longlong_t)zilog->zl_parse_blk_count,
5270                     (u_longlong_t)zilog->zl_parse_lr_count,
5271                     (u_longlong_t)zilog->zl_replaying_seq);
5272 
5273         zilog = zil_open(os, ztest_get_data);
5274 
5275         if (zilog->zl_replaying_seq != 0 &&
5276             zilog->zl_replaying_seq < committed_seq)
5277                 fatal(0, "missing log records: replayed %llu < committed %llu",
5278                     zilog->zl_replaying_seq, committed_seq);
5279 
5280         return (0);
5281 }
5282 
5283 static void
5284 ztest_dataset_close(int d)
5285 {
5286         ztest_ds_t *zd = &ztest_ds[d];
5287 
5288         zil_close(zd->zd_zilog);
5289         dmu_objset_rele(zd->zd_os, zd);
5290 
5291         ztest_zd_fini(zd);
5292 }
5293 
5294 /*
5295  * Kick off threads to run tests on all datasets in parallel.
5296  */
5297 static void
5298 ztest_run(ztest_shared_t *zs)
5299 {
5300         thread_t *tid;
5301         spa_t *spa;
5302         objset_t *os;
5303         thread_t resume_tid;
5304         int error;
5305 
5306         ztest_exiting = B_FALSE;
5307 
5308         /*
5309          * Initialize parent/child shared state.
5310          */
5311         VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5312         VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5313 
5314         zs->zs_thread_start = gethrtime();
5315         zs->zs_thread_stop =
5316             zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5317         zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5318         zs->zs_thread_kill = zs->zs_thread_stop;
5319         if (ztest_random(100) < ztest_opts.zo_killrate) {
5320                 zs->zs_thread_kill -=
5321                     ztest_random(ztest_opts.zo_passtime * NANOSEC);
5322         }
5323 
5324         (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5325 
5326         list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5327             offsetof(ztest_cb_data_t, zcd_node));
5328 
5329         /*
5330          * Open our pool.
5331          */
5332         kernel_init(FREAD | FWRITE);
5333         VERIFY(spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0);
5334         spa->spa_debug = B_TRUE;
5335         ztest_spa = spa;
5336 
5337         VERIFY0(dmu_objset_hold(ztest_opts.zo_pool, FTAG, &os));
5338         zs->zs_guid = dmu_objset_fsid_guid(os);
5339         dmu_objset_rele(os, FTAG);
5340 
5341         spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5342 
5343         /*
5344          * We don't expect the pool to suspend unless maxfaults == 0,
5345          * in which case ztest_fault_inject() temporarily takes away
5346          * the only valid replica.
5347          */
5348         if (MAXFAULTS() == 0)
5349                 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5350         else
5351                 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5352 
5353         /*
5354          * Create a thread to periodically resume suspended I/O.
5355          */
5356         VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5357             &resume_tid) == 0);
5358 
5359         /*
5360          * Create a deadman thread to abort() if we hang.
5361          */
5362         VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5363             NULL) == 0);
5364 
5365         /*
5366          * Verify that we can safely inquire about about any object,
5367          * whether it's allocated or not.  To make it interesting,
5368          * we probe a 5-wide window around each power of two.
5369          * This hits all edge cases, including zero and the max.
5370          */
5371         for (int t = 0; t < 64; t++) {
5372                 for (int d = -5; d <= 5; d++) {
5373                         error = dmu_object_info(spa->spa_meta_objset,
5374                             (1ULL << t) + d, NULL);
5375                         ASSERT(error == 0 || error == ENOENT ||
5376                             error == EINVAL);
5377                 }
5378         }
5379 
5380         /*
5381          * If we got any ENOSPC errors on the previous run, destroy something.
5382          */
5383         if (zs->zs_enospc_count != 0) {
5384                 int d = ztest_random(ztest_opts.zo_datasets);
5385                 ztest_dataset_destroy(d);
5386         }
5387         zs->zs_enospc_count = 0;
5388 
5389         tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5390             UMEM_NOFAIL);
5391 
5392         if (ztest_opts.zo_verbose >= 4)
5393                 (void) printf("starting main threads...\n");
5394 
5395         /*
5396          * Kick off all the tests that run in parallel.
5397          */
5398         for (int t = 0; t < ztest_opts.zo_threads; t++) {
5399                 if (t < ztest_opts.zo_datasets &&
5400                     ztest_dataset_open(t) != 0)
5401                         return;
5402                 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5403                     THR_BOUND, &tid[t]) == 0);
5404         }
5405 
5406         /*
5407          * Wait for all of the tests to complete.  We go in reverse order
5408          * so we don't close datasets while threads are still using them.
5409          */
5410         for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5411                 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5412                 if (t < ztest_opts.zo_datasets)
5413                         ztest_dataset_close(t);
5414         }
5415 
5416         txg_wait_synced(spa_get_dsl(spa), 0);
5417 
5418         zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5419         zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5420 
5421         umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5422 
5423         /* Kill the resume thread */
5424         ztest_exiting = B_TRUE;
5425         VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5426         ztest_resume(spa);
5427 
5428         /*
5429          * Right before closing the pool, kick off a bunch of async I/O;
5430          * spa_close() should wait for it to complete.
5431          */
5432         for (uint64_t object = 1; object < 50; object++)
5433                 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5434 
5435         spa_close(spa, FTAG);
5436 
5437         /*
5438          * Verify that we can loop over all pools.
5439          */
5440         mutex_enter(&spa_namespace_lock);
5441         for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5442                 if (ztest_opts.zo_verbose > 3)
5443                         (void) printf("spa_next: found %s\n", spa_name(spa));
5444         mutex_exit(&spa_namespace_lock);
5445 
5446         /*
5447          * Verify that we can export the pool and reimport it under a
5448          * different name.
5449          */
5450         if (ztest_random(2) == 0) {
5451                 char name[MAXNAMELEN];
5452                 (void) snprintf(name, MAXNAMELEN, "%s_import",
5453                     ztest_opts.zo_pool);
5454                 ztest_spa_import_export(ztest_opts.zo_pool, name);
5455                 ztest_spa_import_export(name, ztest_opts.zo_pool);
5456         }
5457 
5458         kernel_fini();
5459 
5460         list_destroy(&zcl.zcl_callbacks);
5461 
5462         (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5463 
5464         (void) rwlock_destroy(&ztest_name_lock);
5465         (void) _mutex_destroy(&ztest_vdev_lock);
5466 }
5467 
5468 static void
5469 ztest_freeze(void)
5470 {
5471         ztest_ds_t *zd = &ztest_ds[0];
5472         spa_t *spa;
5473         int numloops = 0;
5474 
5475         if (ztest_opts.zo_verbose >= 3)
5476                 (void) printf("testing spa_freeze()...\n");
5477 
5478         kernel_init(FREAD | FWRITE);
5479         VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5480         VERIFY0(ztest_dataset_open(0));
5481 
5482         /*
5483          * Force the first log block to be transactionally allocated.
5484          * We have to do this before we freeze the pool -- otherwise
5485          * the log chain won't be anchored.
5486          */
5487         while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5488                 ztest_dmu_object_alloc_free(zd, 0);
5489                 zil_commit(zd->zd_zilog, 0);
5490         }
5491 
5492         txg_wait_synced(spa_get_dsl(spa), 0);
5493 
5494         /*
5495          * Freeze the pool.  This stops spa_sync() from doing anything,
5496          * so that the only way to record changes from now on is the ZIL.
5497          */
5498         spa_freeze(spa);
5499 
5500         /*
5501          * Run tests that generate log records but don't alter the pool config
5502          * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5503          * We do a txg_wait_synced() after each iteration to force the txg
5504          * to increase well beyond the last synced value in the uberblock.
5505          * The ZIL should be OK with that.
5506          */
5507         while (ztest_random(10) != 0 &&
5508             numloops++ < ztest_opts.zo_maxloops) {
5509                 ztest_dmu_write_parallel(zd, 0);
5510                 ztest_dmu_object_alloc_free(zd, 0);
5511                 txg_wait_synced(spa_get_dsl(spa), 0);
5512         }
5513 
5514         /*
5515          * Commit all of the changes we just generated.
5516          */
5517         zil_commit(zd->zd_zilog, 0);
5518         txg_wait_synced(spa_get_dsl(spa), 0);
5519 
5520         /*
5521          * Close our dataset and close the pool.
5522          */
5523         ztest_dataset_close(0);
5524         spa_close(spa, FTAG);
5525         kernel_fini();
5526 
5527         /*
5528          * Open and close the pool and dataset to induce log replay.
5529          */
5530         kernel_init(FREAD | FWRITE);
5531         VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5532         VERIFY0(ztest_dataset_open(0));
5533         ztest_dataset_close(0);
5534         spa_close(spa, FTAG);
5535         kernel_fini();
5536 }
5537 
5538 void
5539 print_time(hrtime_t t, char *timebuf)
5540 {
5541         hrtime_t s = t / NANOSEC;
5542         hrtime_t m = s / 60;
5543         hrtime_t h = m / 60;
5544         hrtime_t d = h / 24;
5545 
5546         s -= m * 60;
5547         m -= h * 60;
5548         h -= d * 24;
5549 
5550         timebuf[0] = '\0';
5551 
5552         if (d)
5553                 (void) sprintf(timebuf,
5554                     "%llud%02lluh%02llum%02llus", d, h, m, s);
5555         else if (h)
5556                 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5557         else if (m)
5558                 (void) sprintf(timebuf, "%llum%02llus", m, s);
5559         else
5560                 (void) sprintf(timebuf, "%llus", s);
5561 }
5562 
5563 static nvlist_t *
5564 make_random_props()
5565 {
5566         nvlist_t *props;
5567 
5568         VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5569         if (ztest_random(2) == 0)
5570                 return (props);
5571         VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5572 
5573         return (props);
5574 }
5575 
5576 /*
5577  * Create a storage pool with the given name and initial vdev size.
5578  * Then test spa_freeze() functionality.
5579  */
5580 static void
5581 ztest_init(ztest_shared_t *zs)
5582 {
5583         spa_t *spa;
5584         nvlist_t *nvroot, *props;
5585 
5586         VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5587         VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5588 
5589         kernel_init(FREAD | FWRITE);
5590 
5591         /*
5592          * Create the storage pool.
5593          */
5594         (void) spa_destroy(ztest_opts.zo_pool);
5595         ztest_shared->zs_vdev_next_leaf = 0;
5596         zs->zs_splits = 0;
5597         zs->zs_mirrors = ztest_opts.zo_mirrors;
5598         nvroot = make_vdev_root(NULL, NULL, ztest_opts.zo_vdev_size, 0,
5599             0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5600         props = make_random_props();
5601         for (int i = 0; i < SPA_FEATURES; i++) {
5602                 char buf[1024];
5603                 (void) snprintf(buf, sizeof (buf), "feature@%s",
5604                     spa_feature_table[i].fi_uname);
5605                 VERIFY0(nvlist_add_uint64(props, buf, 0));
5606         }
5607         VERIFY0(spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5608         nvlist_free(nvroot);
5609 
5610         VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5611         zs->zs_metaslab_sz =
5612             1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5613 
5614         spa_close(spa, FTAG);
5615 
5616         kernel_fini();
5617 
5618         ztest_run_zdb(ztest_opts.zo_pool);
5619 
5620         ztest_freeze();
5621 
5622         ztest_run_zdb(ztest_opts.zo_pool);
5623 
5624         (void) rwlock_destroy(&ztest_name_lock);
5625         (void) _mutex_destroy(&ztest_vdev_lock);
5626 }
5627 
5628 static void
5629 setup_fds(void)
5630 {
5631         int fd;
5632 
5633         char *tmp = tempnam(NULL, NULL);
5634         fd = open(tmp, O_RDWR | O_CREAT, 0700);
5635         ASSERT3U(fd, ==, ZTEST_FD_DATA);
5636         (void) unlink(tmp);
5637         free(tmp);
5638 
5639         fd = open("/dev/urandom", O_RDONLY);
5640         ASSERT3U(fd, ==, ZTEST_FD_RAND);
5641 }
5642 
5643 static int
5644 shared_data_size(ztest_shared_hdr_t *hdr)
5645 {
5646         int size;
5647 
5648         size = hdr->zh_hdr_size;
5649         size += hdr->zh_opts_size;
5650         size += hdr->zh_size;
5651         size += hdr->zh_stats_size * hdr->zh_stats_count;
5652         size += hdr->zh_ds_size * hdr->zh_ds_count;
5653 
5654         return (size);
5655 }
5656 
5657 static void
5658 setup_hdr(void)
5659 {
5660         int size;
5661         ztest_shared_hdr_t *hdr;
5662 
5663         hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5664             PROT_READ | PROT_WRITE, MAP_SHARED, ZTEST_FD_DATA, 0);
5665         ASSERT(hdr != MAP_FAILED);
5666 
5667         VERIFY0(ftruncate(ZTEST_FD_DATA, sizeof (ztest_shared_hdr_t)));
5668 
5669         hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5670         hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5671         hdr->zh_size = sizeof (ztest_shared_t);
5672         hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5673         hdr->zh_stats_count = ZTEST_FUNCS;
5674         hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5675         hdr->zh_ds_count = ztest_opts.zo_datasets;
5676 
5677         size = shared_data_size(hdr);
5678         VERIFY0(ftruncate(ZTEST_FD_DATA, size));
5679 
5680         (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5681 }
5682 
5683 static void
5684 setup_data(void)
5685 {
5686         int size, offset;
5687         ztest_shared_hdr_t *hdr;
5688         uint8_t *buf;
5689 
5690         hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5691             PROT_READ, MAP_SHARED, ZTEST_FD_DATA, 0);
5692         ASSERT(hdr != MAP_FAILED);
5693 
5694         size = shared_data_size(hdr);
5695 
5696         (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5697         hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5698             PROT_READ | PROT_WRITE, MAP_SHARED, ZTEST_FD_DATA, 0);
5699         ASSERT(hdr != MAP_FAILED);
5700         buf = (uint8_t *)hdr;
5701 
5702         offset = hdr->zh_hdr_size;
5703         ztest_shared_opts = (void *)&buf[offset];
5704         offset += hdr->zh_opts_size;
5705         ztest_shared = (void *)&buf[offset];
5706         offset += hdr->zh_size;
5707         ztest_shared_callstate = (void *)&buf[offset];
5708         offset += hdr->zh_stats_size * hdr->zh_stats_count;
5709         ztest_shared_ds = (void *)&buf[offset];
5710 }
5711 
5712 static boolean_t
5713 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5714 {
5715         pid_t pid;
5716         int status;
5717         char cmdbuf[MAXPATHLEN];
5718 
5719         pid = fork();
5720 
5721         if (cmd == NULL) {
5722                 (void) strlcpy(cmdbuf, getexecname(), sizeof (cmdbuf));
5723                 cmd = cmdbuf;
5724         }
5725 
5726         if (pid == -1)
5727                 fatal(1, "fork failed");
5728 
5729         if (pid == 0) { /* child */
5730                 char *emptyargv[2] = { cmd, NULL };
5731 
5732                 struct rlimit rl = { 1024, 1024 };
5733                 (void) setrlimit(RLIMIT_NOFILE, &rl);
5734                 (void) enable_extended_FILE_stdio(-1, -1);
5735                 if (libpath != NULL)
5736                         VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5737                 (void) execv(cmd, emptyargv);
5738                 ztest_dump_core = B_FALSE;
5739                 fatal(B_TRUE, "exec failed: %s", cmd);
5740         }
5741 
5742         while (waitpid(pid, &status, 0) != pid)
5743                 continue;
5744         if (statusp != NULL)
5745                 *statusp = status;
5746 
5747         if (WIFEXITED(status)) {
5748                 if (WEXITSTATUS(status) != 0) {
5749                         (void) fprintf(stderr, "child exited with code %d\n",
5750                             WEXITSTATUS(status));
5751                         exit(2);
5752                 }
5753                 return (B_FALSE);
5754         } else if (WIFSIGNALED(status)) {
5755                 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
5756                         (void) fprintf(stderr, "child died with signal %d\n",
5757                             WTERMSIG(status));
5758                         exit(3);
5759                 }
5760                 return (B_TRUE);
5761         } else {
5762                 (void) fprintf(stderr, "something strange happened to child\n");
5763                 exit(4);
5764                 /* NOTREACHED */
5765         }
5766 }
5767 
5768 static void
5769 ztest_run_init(void)
5770 {
5771         ztest_shared_t *zs = ztest_shared;
5772 
5773         ASSERT(ztest_opts.zo_init != 0);
5774 
5775         /*
5776          * Blow away any existing copy of zpool.cache
5777          */
5778         (void) remove(spa_config_path);
5779 
5780         /*
5781          * Create and initialize our storage pool.
5782          */
5783         for (int i = 1; i <= ztest_opts.zo_init; i++) {
5784                 bzero(zs, sizeof (ztest_shared_t));
5785                 if (ztest_opts.zo_verbose >= 3 &&
5786                     ztest_opts.zo_init != 1) {
5787                         (void) printf("ztest_init(), pass %d\n", i);
5788                 }
5789                 ztest_init(zs);
5790         }
5791 }
5792 
5793 int
5794 main(int argc, char **argv)
5795 {
5796         int kills = 0;
5797         int iters = 0;
5798         int older = 0;
5799         int newer = 0;
5800         ztest_shared_t *zs;
5801         ztest_info_t *zi;
5802         ztest_shared_callstate_t *zc;
5803         char timebuf[100];
5804         char numbuf[6];
5805         spa_t *spa;
5806         char cmd[MAXNAMELEN];
5807         boolean_t hasalt;
5808 
5809         boolean_t ischild = (0 == lseek(ZTEST_FD_DATA, 0, SEEK_CUR));
5810         ASSERT(ischild || errno == EBADF);
5811 
5812         (void) setvbuf(stdout, NULL, _IOLBF, 0);
5813 
5814         if (!ischild) {
5815                 process_options(argc, argv);
5816 
5817                 setup_fds();
5818                 setup_hdr();
5819                 setup_data();
5820                 bcopy(&ztest_opts, ztest_shared_opts,
5821                     sizeof (*ztest_shared_opts));
5822         } else {
5823                 setup_data();
5824                 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
5825         }
5826         ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
5827 
5828         /* Override location of zpool.cache */
5829         (void) asprintf((char **)&spa_config_path, "%s/zpool.cache",
5830             ztest_opts.zo_dir);
5831 
5832         ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
5833             UMEM_NOFAIL);
5834         zs = ztest_shared;
5835 
5836         if (ischild) {
5837                 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
5838                 metaslab_df_alloc_threshold =
5839                     zs->zs_metaslab_df_alloc_threshold;
5840 
5841                 if (zs->zs_do_init)
5842                         ztest_run_init();
5843                 else
5844                         ztest_run(zs);
5845                 exit(0);
5846         }
5847 
5848         hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
5849 
5850         if (ztest_opts.zo_verbose >= 1) {
5851                 (void) printf("%llu vdevs, %d datasets, %d threads,"
5852                     " %llu seconds...\n",
5853                     (u_longlong_t)ztest_opts.zo_vdevs,
5854                     ztest_opts.zo_datasets,
5855                     ztest_opts.zo_threads,
5856                     (u_longlong_t)ztest_opts.zo_time);
5857         }
5858 
5859         (void) strlcpy(cmd, getexecname(), sizeof (cmd));
5860 
5861         zs->zs_do_init = B_TRUE;
5862         if (strlen(ztest_opts.zo_alt_ztest) != 0) {
5863                 if (ztest_opts.zo_verbose >= 1) {
5864                         (void) printf("Executing older ztest for "
5865                             "initialization: %s\n", ztest_opts.zo_alt_ztest);
5866                 }
5867                 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
5868                     ztest_opts.zo_alt_libpath, B_FALSE, NULL));
5869         } else {
5870                 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
5871         }
5872         zs->zs_do_init = B_FALSE;
5873 
5874         zs->zs_proc_start = gethrtime();
5875         zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
5876 
5877         for (int f = 0; f < ZTEST_FUNCS; f++) {
5878                 zi = &ztest_info[f];
5879                 zc = ZTEST_GET_SHARED_CALLSTATE(f);
5880                 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
5881                         zc->zc_next = UINT64_MAX;
5882                 else
5883                         zc->zc_next = zs->zs_proc_start +
5884                             ztest_random(2 * zi->zi_interval[0] + 1);
5885         }
5886 
5887         /*
5888          * Run the tests in a loop.  These tests include fault injection
5889          * to verify that self-healing data works, and forced crashes
5890          * to verify that we never lose on-disk consistency.
5891          */
5892         while (gethrtime() < zs->zs_proc_stop) {
5893                 int status;
5894                 boolean_t killed;
5895 
5896                 /*
5897                  * Initialize the workload counters for each function.
5898                  */
5899                 for (int f = 0; f < ZTEST_FUNCS; f++) {
5900                         zc = ZTEST_GET_SHARED_CALLSTATE(f);
5901                         zc->zc_count = 0;
5902                         zc->zc_time = 0;
5903                 }
5904 
5905                 /* Set the allocation switch size */
5906                 zs->zs_metaslab_df_alloc_threshold =
5907                     ztest_random(zs->zs_metaslab_sz / 4) + 1;
5908 
5909                 if (!hasalt || ztest_random(2) == 0) {
5910                         if (hasalt && ztest_opts.zo_verbose >= 1) {
5911                                 (void) printf("Executing newer ztest: %s\n",
5912                                     cmd);
5913                         }
5914                         newer++;
5915                         killed = exec_child(cmd, NULL, B_TRUE, &status);
5916                 } else {
5917                         if (hasalt && ztest_opts.zo_verbose >= 1) {
5918                                 (void) printf("Executing older ztest: %s\n",
5919                                     ztest_opts.zo_alt_ztest);
5920                         }
5921                         older++;
5922                         killed = exec_child(ztest_opts.zo_alt_ztest,
5923                             ztest_opts.zo_alt_libpath, B_TRUE, &status);
5924                 }
5925 
5926                 if (killed)
5927                         kills++;
5928                 iters++;
5929 
5930                 if (ztest_opts.zo_verbose >= 1) {
5931                         hrtime_t now = gethrtime();
5932 
5933                         now = MIN(now, zs->zs_proc_stop);
5934                         print_time(zs->zs_proc_stop - now, timebuf);
5935                         nicenum(zs->zs_space, numbuf);
5936 
5937                         (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
5938                             "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
5939                             iters,
5940                             WIFEXITED(status) ? "Complete" : "SIGKILL",
5941                             (u_longlong_t)zs->zs_enospc_count,
5942                             100.0 * zs->zs_alloc / zs->zs_space,
5943                             numbuf,
5944                             100.0 * (now - zs->zs_proc_start) /
5945                             (ztest_opts.zo_time * NANOSEC), timebuf);
5946                 }
5947 
5948                 if (ztest_opts.zo_verbose >= 2) {
5949                         (void) printf("\nWorkload summary:\n\n");
5950                         (void) printf("%7s %9s   %s\n",
5951                             "Calls", "Time", "Function");
5952                         (void) printf("%7s %9s   %s\n",
5953                             "-----", "----", "--------");
5954                         for (int f = 0; f < ZTEST_FUNCS; f++) {
5955                                 Dl_info dli;
5956 
5957                                 zi = &ztest_info[f];
5958                                 zc = ZTEST_GET_SHARED_CALLSTATE(f);
5959                                 print_time(zc->zc_time, timebuf);
5960                                 (void) dladdr((void *)zi->zi_func, &dli);
5961                                 (void) printf("%7llu %9s   %s\n",
5962                                     (u_longlong_t)zc->zc_count, timebuf,
5963                                     dli.dli_sname);
5964                         }
5965                         (void) printf("\n");
5966                 }
5967 
5968                 /*
5969                  * It's possible that we killed a child during a rename test,
5970                  * in which case we'll have a 'ztest_tmp' pool lying around
5971                  * instead of 'ztest'.  Do a blind rename in case this happened.
5972                  */
5973                 kernel_init(FREAD);
5974                 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
5975                         spa_close(spa, FTAG);
5976                 } else {
5977                         char tmpname[MAXNAMELEN];
5978                         kernel_fini();
5979                         kernel_init(FREAD | FWRITE);
5980                         (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
5981                             ztest_opts.zo_pool);
5982                         (void) spa_rename(tmpname, ztest_opts.zo_pool);
5983                 }
5984                 kernel_fini();
5985 
5986                 ztest_run_zdb(ztest_opts.zo_pool);
5987         }
5988 
5989         if (ztest_opts.zo_verbose >= 1) {
5990                 if (hasalt) {
5991                         (void) printf("%d runs of older ztest: %s\n", older,
5992                             ztest_opts.zo_alt_ztest);
5993                         (void) printf("%d runs of newer ztest: %s\n", newer,
5994                             cmd);
5995                 }
5996                 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
5997                     kills, iters - kills, (100.0 * kills) / MAX(1, iters));
5998         }
5999 
6000         return (0);
6001 }