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 2008 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25 
  26 /*
  27  * The objective of this program is to provide a DMU/ZAP/SPA stress test
  28  * that runs entirely in userland, is easy to use, and easy to extend.
  29  *
  30  * The overall design of the ztest program is as follows:
  31  *
  32  * (1) For each major functional area (e.g. adding vdevs to a pool,
  33  *     creating and destroying datasets, reading and writing objects, etc)
  34  *     we have a simple routine to test that functionality.  These
  35  *     individual routines do not have to do anything "stressful".
  36  *
  37  * (2) We turn these simple functionality tests into a stress test by
  38  *     running them all in parallel, with as many threads as desired,
  39  *     and spread across as many datasets, objects, and vdevs as desired.
  40  *
  41  * (3) While all this is happening, we inject faults into the pool to
  42  *     verify that self-healing data really works.
  43  *
  44  * (4) Every time we open a dataset, we change its checksum and compression
  45  *     functions.  Thus even individual objects vary from block to block
  46  *     in which checksum they use and whether they're compressed.
  47  *
  48  * (5) To verify that we never lose on-disk consistency after a crash,
  49  *     we run the entire test in a child of the main process.
  50  *     At random times, the child self-immolates with a SIGKILL.
  51  *     This is the software equivalent of pulling the power cord.
  52  *     The parent then runs the test again, using the existing
  53  *     storage pool, as many times as desired.
  54  *
  55  * (6) To verify that we don't have future leaks or temporal incursions,
  56  *     many of the functional tests record the transaction group number
  57  *     as part of their data.  When reading old data, they verify that
  58  *     the transaction group number is less than the current, open txg.
  59  *     If you add a new test, please do this if applicable.
  60  *
  61  * When run with no arguments, ztest runs for about five minutes and
  62  * produces no output if successful.  To get a little bit of information,
  63  * specify -V.  To get more information, specify -VV, and so on.
  64  *
  65  * To turn this into an overnight stress test, use -T to specify run time.
  66  *
  67  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
  68  * to increase the pool capacity, fanout, and overall stress level.
  69  *
  70  * The -N(okill) option will suppress kills, so each child runs to completion.
  71  * This can be useful when you're trying to distinguish temporal incursions
  72  * from plain old race conditions.
  73  */
  74 
  75 #include <sys/zfs_context.h>
  76 #include <sys/spa.h>
  77 #include <sys/dmu.h>
  78 #include <sys/txg.h>
  79 #include <sys/zap.h>
  80 #include <sys/dmu_objset.h>
  81 #include <sys/poll.h>
  82 #include <sys/stat.h>
  83 #include <sys/time.h>
  84 #include <sys/wait.h>
  85 #include <sys/mman.h>
  86 #include <sys/resource.h>
  87 #include <sys/zio.h>
  88 #include <sys/zio_checksum.h>
  89 #include <sys/zio_compress.h>
  90 #include <sys/zil.h>
  91 #include <sys/vdev_impl.h>
  92 #include <sys/vdev_file.h>
  93 #include <sys/spa_impl.h>
  94 #include <sys/dsl_prop.h>
  95 #include <sys/refcount.h>
  96 #include <stdio.h>
  97 #include <stdio_ext.h>
  98 #include <stdlib.h>
  99 #include <unistd.h>
 100 #include <signal.h>
 101 #include <umem.h>
 102 #include <dlfcn.h>
 103 #include <ctype.h>
 104 #include <math.h>
 105 #include <sys/fs/zfs.h>
 106 
 107 static char cmdname[] = "ztest";
 108 static char *zopt_pool = cmdname;
 109 
 110 static uint64_t zopt_vdevs = 5;
 111 static uint64_t zopt_vdevtime;
 112 static int zopt_ashift = SPA_MINBLOCKSHIFT;
 113 static int zopt_mirrors = 2;
 114 static int zopt_raidz = 4;
 115 static int zopt_raidz_parity = 1;
 116 static size_t zopt_vdev_size = SPA_MINDEVSIZE;
 117 static int zopt_datasets = 7;
 118 static int zopt_threads = 23;
 119 static uint64_t zopt_passtime = 60;     /* 60 seconds */
 120 static uint64_t zopt_killrate = 70;     /* 70% kill rate */
 121 static int zopt_verbose = 0;
 122 static int zopt_init = 1;
 123 static char *zopt_dir = "/tmp";
 124 static uint64_t zopt_time = 300;        /* 5 minutes */
 125 static int zopt_maxfaults;
 126 
 127 typedef struct ztest_block_tag {
 128         uint64_t        bt_objset;
 129         uint64_t        bt_object;
 130         uint64_t        bt_offset;
 131         uint64_t        bt_txg;
 132         uint64_t        bt_thread;
 133         uint64_t        bt_seq;
 134 } ztest_block_tag_t;
 135 
 136 typedef struct ztest_args {
 137         char            za_pool[MAXNAMELEN];
 138         spa_t           *za_spa;
 139         objset_t        *za_os;
 140         zilog_t         *za_zilog;
 141         thread_t        za_thread;
 142         uint64_t        za_instance;
 143         uint64_t        za_random;
 144         uint64_t        za_diroff;
 145         uint64_t        za_diroff_shared;
 146         uint64_t        za_zil_seq;
 147         hrtime_t        za_start;
 148         hrtime_t        za_stop;
 149         hrtime_t        za_kill;
 150         /*
 151          * Thread-local variables can go here to aid debugging.
 152          */
 153         ztest_block_tag_t za_rbt;
 154         ztest_block_tag_t za_wbt;
 155         dmu_object_info_t za_doi;
 156         dmu_buf_t       *za_dbuf;
 157 } ztest_args_t;
 158 
 159 typedef void ztest_func_t(ztest_args_t *);
 160 
 161 /*
 162  * Note: these aren't static because we want dladdr() to work.
 163  */
 164 ztest_func_t ztest_dmu_read_write;
 165 ztest_func_t ztest_dmu_write_parallel;
 166 ztest_func_t ztest_dmu_object_alloc_free;
 167 ztest_func_t ztest_zap;
 168 ztest_func_t ztest_zap_parallel;
 169 ztest_func_t ztest_traverse;
 170 ztest_func_t ztest_dsl_prop_get_set;
 171 ztest_func_t ztest_dmu_objset_create_destroy;
 172 ztest_func_t ztest_dmu_snapshot_create_destroy;
 173 ztest_func_t ztest_spa_create_destroy;
 174 ztest_func_t ztest_fault_inject;
 175 ztest_func_t ztest_spa_rename;
 176 ztest_func_t ztest_vdev_attach_detach;
 177 ztest_func_t ztest_vdev_LUN_growth;
 178 ztest_func_t ztest_vdev_add_remove;
 179 ztest_func_t ztest_vdev_aux_add_remove;
 180 ztest_func_t ztest_scrub;
 181 
 182 typedef struct ztest_info {
 183         ztest_func_t    *zi_func;       /* test function */
 184         uint64_t        zi_iters;       /* iterations per execution */
 185         uint64_t        *zi_interval;   /* execute every <interval> seconds */
 186         uint64_t        zi_calls;       /* per-pass count */
 187         uint64_t        zi_call_time;   /* per-pass time */
 188         uint64_t        zi_call_total;  /* cumulative total */
 189         uint64_t        zi_call_target; /* target cumulative total */
 190 } ztest_info_t;
 191 
 192 uint64_t zopt_always = 0;               /* all the time */
 193 uint64_t zopt_often = 1;                /* every second */
 194 uint64_t zopt_sometimes = 10;           /* every 10 seconds */
 195 uint64_t zopt_rarely = 60;              /* every 60 seconds */
 196 
 197 ztest_info_t ztest_info[] = {
 198         { ztest_dmu_read_write,                 1,      &zopt_always        },
 199         { ztest_dmu_write_parallel,             30,     &zopt_always        },
 200         { ztest_dmu_object_alloc_free,          1,      &zopt_always        },
 201         { ztest_zap,                            30,     &zopt_always        },
 202         { ztest_zap_parallel,                   100,    &zopt_always        },
 203         { ztest_dsl_prop_get_set,               1,      &zopt_sometimes     },
 204         { ztest_dmu_objset_create_destroy,      1,      &zopt_sometimes },
 205         { ztest_dmu_snapshot_create_destroy,    1,      &zopt_sometimes },
 206         { ztest_spa_create_destroy,             1,      &zopt_sometimes },
 207         { ztest_fault_inject,                   1,      &zopt_sometimes     },
 208         { ztest_spa_rename,                     1,      &zopt_rarely        },
 209         { ztest_vdev_attach_detach,             1,      &zopt_rarely        },
 210         { ztest_vdev_LUN_growth,                1,      &zopt_rarely        },
 211         { ztest_vdev_add_remove,                1,      &zopt_vdevtime      },
 212         { ztest_vdev_aux_add_remove,            1,      &zopt_vdevtime      },
 213         { ztest_scrub,                          1,      &zopt_vdevtime      },
 214 };
 215 
 216 #define ZTEST_FUNCS     (sizeof (ztest_info) / sizeof (ztest_info_t))
 217 
 218 #define ZTEST_SYNC_LOCKS        16
 219 
 220 /*
 221  * Stuff we need to share writably between parent and child.
 222  */
 223 typedef struct ztest_shared {
 224         mutex_t         zs_vdev_lock;
 225         rwlock_t        zs_name_lock;
 226         uint64_t        zs_vdev_primaries;
 227         uint64_t        zs_vdev_aux;
 228         uint64_t        zs_enospc_count;
 229         hrtime_t        zs_start_time;
 230         hrtime_t        zs_stop_time;
 231         uint64_t        zs_alloc;
 232         uint64_t        zs_space;
 233         ztest_info_t    zs_info[ZTEST_FUNCS];
 234         mutex_t         zs_sync_lock[ZTEST_SYNC_LOCKS];
 235         uint64_t        zs_seq[ZTEST_SYNC_LOCKS];
 236 } ztest_shared_t;
 237 
 238 static char ztest_dev_template[] = "%s/%s.%llua";
 239 static char ztest_aux_template[] = "%s/%s.%s.%llu";
 240 static ztest_shared_t *ztest_shared;
 241 
 242 static int ztest_random_fd;
 243 static int ztest_dump_core = 1;
 244 
 245 static boolean_t ztest_exiting;
 246 
 247 extern uint64_t metaslab_gang_bang;
 248 
 249 #define ZTEST_DIROBJ            1
 250 #define ZTEST_MICROZAP_OBJ      2
 251 #define ZTEST_FATZAP_OBJ        3
 252 
 253 #define ZTEST_DIROBJ_BLOCKSIZE  (1 << 10)
 254 #define ZTEST_DIRSIZE           256
 255 
 256 static void usage(boolean_t) __NORETURN;
 257 
 258 /*
 259  * These libumem hooks provide a reasonable set of defaults for the allocator's
 260  * debugging facilities.
 261  */
 262 const char *
 263 _umem_debug_init()
 264 {
 265         return ("default,verbose"); /* $UMEM_DEBUG setting */
 266 }
 267 
 268 const char *
 269 _umem_logging_init(void)
 270 {
 271         return ("fail,contents"); /* $UMEM_LOGGING setting */
 272 }
 273 
 274 #define FATAL_MSG_SZ    1024
 275 
 276 char *fatal_msg;
 277 
 278 static void
 279 fatal(int do_perror, char *message, ...)
 280 {
 281         va_list args;
 282         int save_errno = errno;
 283         char buf[FATAL_MSG_SZ];
 284 
 285         (void) fflush(stdout);
 286 
 287         va_start(args, message);
 288         (void) sprintf(buf, "ztest: ");
 289         /* LINTED */
 290         (void) vsprintf(buf + strlen(buf), message, args);
 291         va_end(args);
 292         if (do_perror) {
 293                 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
 294                     ": %s", strerror(save_errno));
 295         }
 296         (void) fprintf(stderr, "%s\n", buf);
 297         fatal_msg = buf;                        /* to ease debugging */
 298         if (ztest_dump_core)
 299                 abort();
 300         exit(3);
 301 }
 302 
 303 static int
 304 str2shift(const char *buf)
 305 {
 306         const char *ends = "BKMGTPEZ";
 307         int i;
 308 
 309         if (buf[0] == '\0')
 310                 return (0);
 311         for (i = 0; i < strlen(ends); i++) {
 312                 if (toupper(buf[0]) == ends[i])
 313                         break;
 314         }
 315         if (i == strlen(ends)) {
 316                 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
 317                     buf);
 318                 usage(B_FALSE);
 319         }
 320         if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
 321                 return (10*i);
 322         }
 323         (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
 324         usage(B_FALSE);
 325         /* NOTREACHED */
 326 }
 327 
 328 static uint64_t
 329 nicenumtoull(const char *buf)
 330 {
 331         char *end;
 332         uint64_t val;
 333 
 334         val = strtoull(buf, &end, 0);
 335         if (end == buf) {
 336                 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
 337                 usage(B_FALSE);
 338         } else if (end[0] == '.') {
 339                 double fval = strtod(buf, &end);
 340                 fval *= pow(2, str2shift(end));
 341                 if (fval > UINT64_MAX) {
 342                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 343                             buf);
 344                         usage(B_FALSE);
 345                 }
 346                 val = (uint64_t)fval;
 347         } else {
 348                 int shift = str2shift(end);
 349                 if (shift >= 64 || (val << shift) >> shift != val) {
 350                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 351                             buf);
 352                         usage(B_FALSE);
 353                 }
 354                 val <<= shift;
 355         }
 356         return (val);
 357 }
 358 
 359 static void
 360 usage(boolean_t requested)
 361 {
 362         char nice_vdev_size[10];
 363         char nice_gang_bang[10];
 364         FILE *fp = requested ? stdout : stderr;
 365 
 366         nicenum(zopt_vdev_size, nice_vdev_size);
 367         nicenum(metaslab_gang_bang, nice_gang_bang);
 368 
 369         (void) fprintf(fp, "Usage: %s\n"
 370             "\t[-v vdevs (default: %llu)]\n"
 371             "\t[-s size_of_each_vdev (default: %s)]\n"
 372             "\t[-a alignment_shift (default: %d) (use 0 for random)]\n"
 373             "\t[-m mirror_copies (default: %d)]\n"
 374             "\t[-r raidz_disks (default: %d)]\n"
 375             "\t[-R raidz_parity (default: %d)]\n"
 376             "\t[-d datasets (default: %d)]\n"
 377             "\t[-t threads (default: %d)]\n"
 378             "\t[-g gang_block_threshold (default: %s)]\n"
 379             "\t[-i initialize pool i times (default: %d)]\n"
 380             "\t[-k kill percentage (default: %llu%%)]\n"
 381             "\t[-p pool_name (default: %s)]\n"
 382             "\t[-f file directory for vdev files (default: %s)]\n"
 383             "\t[-V(erbose)] (use multiple times for ever more blather)\n"
 384             "\t[-E(xisting)] (use existing pool instead of creating new one)\n"
 385             "\t[-T time] total run time (default: %llu sec)\n"
 386             "\t[-P passtime] time per pass (default: %llu sec)\n"
 387             "\t[-h] (print help)\n"
 388             "",
 389             cmdname,
 390             (u_longlong_t)zopt_vdevs,                   /* -v */
 391             nice_vdev_size,                             /* -s */
 392             zopt_ashift,                                /* -a */
 393             zopt_mirrors,                               /* -m */
 394             zopt_raidz,                                 /* -r */
 395             zopt_raidz_parity,                          /* -R */
 396             zopt_datasets,                              /* -d */
 397             zopt_threads,                               /* -t */
 398             nice_gang_bang,                             /* -g */
 399             zopt_init,                                  /* -i */
 400             (u_longlong_t)zopt_killrate,                /* -k */
 401             zopt_pool,                                  /* -p */
 402             zopt_dir,                                   /* -f */
 403             (u_longlong_t)zopt_time,                    /* -T */
 404             (u_longlong_t)zopt_passtime);               /* -P */
 405         exit(requested ? 0 : 1);
 406 }
 407 
 408 static uint64_t
 409 ztest_random(uint64_t range)
 410 {
 411         uint64_t r;
 412 
 413         if (range == 0)
 414                 return (0);
 415 
 416         if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r))
 417                 fatal(1, "short read from /dev/urandom");
 418 
 419         return (r % range);
 420 }
 421 
 422 /* ARGSUSED */
 423 static void
 424 ztest_record_enospc(char *s)
 425 {
 426         ztest_shared->zs_enospc_count++;
 427 }
 428 
 429 static void
 430 process_options(int argc, char **argv)
 431 {
 432         int opt;
 433         uint64_t value;
 434 
 435         /* By default, test gang blocks for blocks 32K and greater */
 436         metaslab_gang_bang = 32 << 10;
 437 
 438         while ((opt = getopt(argc, argv,
 439             "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:h")) != EOF) {
 440                 value = 0;
 441                 switch (opt) {
 442                 case 'v':
 443                 case 's':
 444                 case 'a':
 445                 case 'm':
 446                 case 'r':
 447                 case 'R':
 448                 case 'd':
 449                 case 't':
 450                 case 'g':
 451                 case 'i':
 452                 case 'k':
 453                 case 'T':
 454                 case 'P':
 455                         value = nicenumtoull(optarg);
 456                 }
 457                 switch (opt) {
 458                 case 'v':
 459                         zopt_vdevs = value;
 460                         break;
 461                 case 's':
 462                         zopt_vdev_size = MAX(SPA_MINDEVSIZE, value);
 463                         break;
 464                 case 'a':
 465                         zopt_ashift = value;
 466                         break;
 467                 case 'm':
 468                         zopt_mirrors = value;
 469                         break;
 470                 case 'r':
 471                         zopt_raidz = MAX(1, value);
 472                         break;
 473                 case 'R':
 474                         zopt_raidz_parity = MIN(MAX(value, 1), 2);
 475                         break;
 476                 case 'd':
 477                         zopt_datasets = MAX(1, value);
 478                         break;
 479                 case 't':
 480                         zopt_threads = MAX(1, value);
 481                         break;
 482                 case 'g':
 483                         metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value);
 484                         break;
 485                 case 'i':
 486                         zopt_init = value;
 487                         break;
 488                 case 'k':
 489                         zopt_killrate = value;
 490                         break;
 491                 case 'p':
 492                         zopt_pool = strdup(optarg);
 493                         break;
 494                 case 'f':
 495                         zopt_dir = strdup(optarg);
 496                         break;
 497                 case 'V':
 498                         zopt_verbose++;
 499                         break;
 500                 case 'E':
 501                         zopt_init = 0;
 502                         break;
 503                 case 'T':
 504                         zopt_time = value;
 505                         break;
 506                 case 'P':
 507                         zopt_passtime = MAX(1, value);
 508                         break;
 509                 case 'h':
 510                         usage(B_TRUE);
 511                         break;
 512                 case '?':
 513                 default:
 514                         usage(B_FALSE);
 515                         break;
 516                 }
 517         }
 518 
 519         zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1);
 520 
 521         zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time / zopt_vdevs : UINT64_MAX);
 522         zopt_maxfaults = MAX(zopt_mirrors, 1) * (zopt_raidz_parity + 1) - 1;
 523 }
 524 
 525 static uint64_t
 526 ztest_get_ashift(void)
 527 {
 528         if (zopt_ashift == 0)
 529                 return (SPA_MINBLOCKSHIFT + ztest_random(3));
 530         return (zopt_ashift);
 531 }
 532 
 533 static nvlist_t *
 534 make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift)
 535 {
 536         char pathbuf[MAXPATHLEN];
 537         uint64_t vdev;
 538         nvlist_t *file;
 539 
 540         if (ashift == 0)
 541                 ashift = ztest_get_ashift();
 542 
 543         if (path == NULL) {
 544                 path = pathbuf;
 545 
 546                 if (aux != NULL) {
 547                         vdev = ztest_shared->zs_vdev_aux;
 548                         (void) sprintf(path, ztest_aux_template,
 549                             zopt_dir, zopt_pool, aux, vdev);
 550                 } else {
 551                         vdev = ztest_shared->zs_vdev_primaries++;
 552                         (void) sprintf(path, ztest_dev_template,
 553                             zopt_dir, zopt_pool, vdev);
 554                 }
 555         }
 556 
 557         if (size != 0) {
 558                 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
 559                 if (fd == -1)
 560                         fatal(1, "can't open %s", path);
 561                 if (ftruncate(fd, size) != 0)
 562                         fatal(1, "can't ftruncate %s", path);
 563                 (void) close(fd);
 564         }
 565 
 566         VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
 567         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
 568         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
 569         VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
 570 
 571         return (file);
 572 }
 573 
 574 static nvlist_t *
 575 make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r)
 576 {
 577         nvlist_t *raidz, **child;
 578         int c;
 579 
 580         if (r < 2)
 581                 return (make_vdev_file(path, aux, size, ashift));
 582         child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
 583 
 584         for (c = 0; c < r; c++)
 585                 child[c] = make_vdev_file(path, aux, size, ashift);
 586 
 587         VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
 588         VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
 589             VDEV_TYPE_RAIDZ) == 0);
 590         VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
 591             zopt_raidz_parity) == 0);
 592         VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
 593             child, r) == 0);
 594 
 595         for (c = 0; c < r; c++)
 596                 nvlist_free(child[c]);
 597 
 598         umem_free(child, r * sizeof (nvlist_t *));
 599 
 600         return (raidz);
 601 }
 602 
 603 static nvlist_t *
 604 make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift,
 605         int r, int m)
 606 {
 607         nvlist_t *mirror, **child;
 608         int c;
 609 
 610         if (m < 1)
 611                 return (make_vdev_raidz(path, aux, size, ashift, r));
 612 
 613         child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
 614 
 615         for (c = 0; c < m; c++)
 616                 child[c] = make_vdev_raidz(path, aux, size, ashift, r);
 617 
 618         VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
 619         VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
 620             VDEV_TYPE_MIRROR) == 0);
 621         VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
 622             child, m) == 0);
 623 
 624         for (c = 0; c < m; c++)
 625                 nvlist_free(child[c]);
 626 
 627         umem_free(child, m * sizeof (nvlist_t *));
 628 
 629         return (mirror);
 630 }
 631 
 632 static nvlist_t *
 633 make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift,
 634         int log, int r, int m, int t)
 635 {
 636         nvlist_t *root, **child;
 637         int c;
 638 
 639         ASSERT(t > 0);
 640 
 641         child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
 642 
 643         for (c = 0; c < t; c++) {
 644                 child[c] = make_vdev_mirror(path, aux, size, ashift, r, m);
 645                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
 646                     log) == 0);
 647         }
 648 
 649         VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
 650         VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
 651         VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
 652             child, t) == 0);
 653 
 654         for (c = 0; c < t; c++)
 655                 nvlist_free(child[c]);
 656 
 657         umem_free(child, t * sizeof (nvlist_t *));
 658 
 659         return (root);
 660 }
 661 
 662 static void
 663 ztest_set_random_blocksize(objset_t *os, uint64_t object, dmu_tx_t *tx)
 664 {
 665         int bs = SPA_MINBLOCKSHIFT +
 666             ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1);
 667         int ibs = DN_MIN_INDBLKSHIFT +
 668             ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1);
 669         int error;
 670 
 671         error = dmu_object_set_blocksize(os, object, 1ULL << bs, ibs, tx);
 672         if (error) {
 673                 char osname[300];
 674                 dmu_objset_name(os, osname);
 675                 fatal(0, "dmu_object_set_blocksize('%s', %llu, %d, %d) = %d",
 676                     osname, object, 1 << bs, ibs, error);
 677         }
 678 }
 679 
 680 static uint8_t
 681 ztest_random_checksum(void)
 682 {
 683         uint8_t checksum;
 684 
 685         do {
 686                 checksum = ztest_random(ZIO_CHECKSUM_FUNCTIONS);
 687         } while (zio_checksum_table[checksum].ci_zbt);
 688 
 689         if (checksum == ZIO_CHECKSUM_OFF)
 690                 checksum = ZIO_CHECKSUM_ON;
 691 
 692         return (checksum);
 693 }
 694 
 695 static uint8_t
 696 ztest_random_compress(void)
 697 {
 698         return ((uint8_t)ztest_random(ZIO_COMPRESS_FUNCTIONS));
 699 }
 700 
 701 static int
 702 ztest_replay_create(objset_t *os, lr_create_t *lr, boolean_t byteswap)
 703 {
 704         dmu_tx_t *tx;
 705         int error;
 706 
 707         if (byteswap)
 708                 byteswap_uint64_array(lr, sizeof (*lr));
 709 
 710         tx = dmu_tx_create(os);
 711         dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
 712         error = dmu_tx_assign(tx, TXG_WAIT);
 713         if (error) {
 714                 dmu_tx_abort(tx);
 715                 return (error);
 716         }
 717 
 718         error = dmu_object_claim(os, lr->lr_doid, lr->lr_mode, 0,
 719             DMU_OT_NONE, 0, tx);
 720         ASSERT3U(error, ==, 0);
 721         dmu_tx_commit(tx);
 722 
 723         if (zopt_verbose >= 5) {
 724                 char osname[MAXNAMELEN];
 725                 dmu_objset_name(os, osname);
 726                 (void) printf("replay create of %s object %llu"
 727                     " in txg %llu = %d\n",
 728                     osname, (u_longlong_t)lr->lr_doid,
 729                     (u_longlong_t)dmu_tx_get_txg(tx), error);
 730         }
 731 
 732         return (error);
 733 }
 734 
 735 static int
 736 ztest_replay_remove(objset_t *os, lr_remove_t *lr, boolean_t byteswap)
 737 {
 738         dmu_tx_t *tx;
 739         int error;
 740 
 741         if (byteswap)
 742                 byteswap_uint64_array(lr, sizeof (*lr));
 743 
 744         tx = dmu_tx_create(os);
 745         dmu_tx_hold_free(tx, lr->lr_doid, 0, DMU_OBJECT_END);
 746         error = dmu_tx_assign(tx, TXG_WAIT);
 747         if (error) {
 748                 dmu_tx_abort(tx);
 749                 return (error);
 750         }
 751 
 752         error = dmu_object_free(os, lr->lr_doid, tx);
 753         dmu_tx_commit(tx);
 754 
 755         return (error);
 756 }
 757 
 758 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
 759         NULL,                   /* 0 no such transaction type */
 760         ztest_replay_create,    /* TX_CREATE */
 761         NULL,                   /* TX_MKDIR */
 762         NULL,                   /* TX_MKXATTR */
 763         NULL,                   /* TX_SYMLINK */
 764         ztest_replay_remove,    /* TX_REMOVE */
 765         NULL,                   /* TX_RMDIR */
 766         NULL,                   /* TX_LINK */
 767         NULL,                   /* TX_RENAME */
 768         NULL,                   /* TX_WRITE */
 769         NULL,                   /* TX_TRUNCATE */
 770         NULL,                   /* TX_SETATTR */
 771         NULL,                   /* TX_ACL */
 772 };
 773 
 774 /*
 775  * Verify that we can't destroy an active pool, create an existing pool,
 776  * or create a pool with a bad vdev spec.
 777  */
 778 void
 779 ztest_spa_create_destroy(ztest_args_t *za)
 780 {
 781         int error;
 782         spa_t *spa;
 783         nvlist_t *nvroot;
 784 
 785         /*
 786          * Attempt to create using a bad file.
 787          */
 788         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
 789         error = spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL);
 790         nvlist_free(nvroot);
 791         if (error != ENOENT)
 792                 fatal(0, "spa_create(bad_file) = %d", error);
 793 
 794         /*
 795          * Attempt to create using a bad mirror.
 796          */
 797         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1);
 798         error = spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL);
 799         nvlist_free(nvroot);
 800         if (error != ENOENT)
 801                 fatal(0, "spa_create(bad_mirror) = %d", error);
 802 
 803         /*
 804          * Attempt to create an existing pool.  It shouldn't matter
 805          * what's in the nvroot; we should fail with EEXIST.
 806          */
 807         (void) rw_rdlock(&ztest_shared->zs_name_lock);
 808         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
 809         error = spa_create(za->za_pool, nvroot, NULL, NULL, NULL);
 810         nvlist_free(nvroot);
 811         if (error != EEXIST)
 812                 fatal(0, "spa_create(whatever) = %d", error);
 813 
 814         error = spa_open(za->za_pool, &spa, FTAG);
 815         if (error)
 816                 fatal(0, "spa_open() = %d", error);
 817 
 818         error = spa_destroy(za->za_pool);
 819         if (error != EBUSY)
 820                 fatal(0, "spa_destroy() = %d", error);
 821 
 822         spa_close(spa, FTAG);
 823         (void) rw_unlock(&ztest_shared->zs_name_lock);
 824 }
 825 
 826 static vdev_t *
 827 vdev_lookup_by_path(vdev_t *vd, const char *path)
 828 {
 829         vdev_t *mvd;
 830 
 831         if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
 832                 return (vd);
 833 
 834         for (int c = 0; c < vd->vdev_children; c++)
 835                 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
 836                     NULL)
 837                         return (mvd);
 838 
 839         return (NULL);
 840 }
 841 
 842 /*
 843  * Verify that vdev_add() works as expected.
 844  */
 845 void
 846 ztest_vdev_add_remove(ztest_args_t *za)
 847 {
 848         spa_t *spa = za->za_spa;
 849         uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
 850         nvlist_t *nvroot;
 851         int error;
 852 
 853         (void) mutex_lock(&ztest_shared->zs_vdev_lock);
 854 
 855         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 856 
 857         ztest_shared->zs_vdev_primaries =
 858             spa->spa_root_vdev->vdev_children * leaves;
 859 
 860         spa_config_exit(spa, SCL_VDEV, FTAG);
 861 
 862         /*
 863          * Make 1/4 of the devices be log devices.
 864          */
 865         nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0,
 866             ztest_random(4) == 0, zopt_raidz, zopt_mirrors, 1);
 867 
 868         error = spa_vdev_add(spa, nvroot);
 869         nvlist_free(nvroot);
 870 
 871         (void) mutex_unlock(&ztest_shared->zs_vdev_lock);
 872 
 873         if (error == ENOSPC)
 874                 ztest_record_enospc("spa_vdev_add");
 875         else if (error != 0)
 876                 fatal(0, "spa_vdev_add() = %d", error);
 877 }
 878 
 879 /*
 880  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
 881  */
 882 void
 883 ztest_vdev_aux_add_remove(ztest_args_t *za)
 884 {
 885         spa_t *spa = za->za_spa;
 886         vdev_t *rvd = spa->spa_root_vdev;
 887         spa_aux_vdev_t *sav;
 888         char *aux;
 889         uint64_t guid = 0;
 890         int error;
 891 
 892         if (ztest_random(2) == 0) {
 893                 sav = &spa->spa_spares;
 894                 aux = ZPOOL_CONFIG_SPARES;
 895         } else {
 896                 sav = &spa->spa_l2cache;
 897                 aux = ZPOOL_CONFIG_L2CACHE;
 898         }
 899 
 900         (void) mutex_lock(&ztest_shared->zs_vdev_lock);
 901 
 902         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 903 
 904         if (sav->sav_count != 0 && ztest_random(4) == 0) {
 905                 /*
 906                  * Pick a random device to remove.
 907                  */
 908                 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
 909         } else {
 910                 /*
 911                  * Find an unused device we can add.
 912                  */
 913                 ztest_shared->zs_vdev_aux = 0;
 914                 for (;;) {
 915                         char path[MAXPATHLEN];
 916                         int c;
 917                         (void) sprintf(path, ztest_aux_template, zopt_dir,
 918                             zopt_pool, aux, ztest_shared->zs_vdev_aux);
 919                         for (c = 0; c < sav->sav_count; c++)
 920                                 if (strcmp(sav->sav_vdevs[c]->vdev_path,
 921                                     path) == 0)
 922                                         break;
 923                         if (c == sav->sav_count &&
 924                             vdev_lookup_by_path(rvd, path) == NULL)
 925                                 break;
 926                         ztest_shared->zs_vdev_aux++;
 927                 }
 928         }
 929 
 930         spa_config_exit(spa, SCL_VDEV, FTAG);
 931 
 932         if (guid == 0) {
 933                 /*
 934                  * Add a new device.
 935                  */
 936                 nvlist_t *nvroot = make_vdev_root(NULL, aux,
 937                     (zopt_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
 938                 error = spa_vdev_add(spa, nvroot);
 939                 if (error != 0)
 940                         fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
 941                 nvlist_free(nvroot);
 942         } else {
 943                 /*
 944                  * Remove an existing device.  Sometimes, dirty its
 945                  * vdev state first to make sure we handle removal
 946                  * of devices that have pending state changes.
 947                  */
 948                 if (ztest_random(2) == 0)
 949                         (void) vdev_online(spa, guid, B_FALSE, NULL);
 950 
 951                 error = spa_vdev_remove(spa, guid, B_FALSE);
 952                 if (error != 0 && error != EBUSY)
 953                         fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
 954         }
 955 
 956         (void) mutex_unlock(&ztest_shared->zs_vdev_lock);
 957 }
 958 
 959 /*
 960  * Verify that we can attach and detach devices.
 961  */
 962 void
 963 ztest_vdev_attach_detach(ztest_args_t *za)
 964 {
 965         spa_t *spa = za->za_spa;
 966         spa_aux_vdev_t *sav = &spa->spa_spares;
 967         vdev_t *rvd = spa->spa_root_vdev;
 968         vdev_t *oldvd, *newvd, *pvd;
 969         nvlist_t *root;
 970         uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
 971         uint64_t leaf, top;
 972         uint64_t ashift = ztest_get_ashift();
 973         uint64_t oldguid, pguid;
 974         size_t oldsize, newsize;
 975         char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
 976         int replacing;
 977         int oldvd_has_siblings = B_FALSE;
 978         int newvd_is_spare = B_FALSE;
 979         int oldvd_is_log;
 980         int error, expected_error;
 981 
 982         (void) mutex_lock(&ztest_shared->zs_vdev_lock);
 983 
 984         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 985 
 986         /*
 987          * Decide whether to do an attach or a replace.
 988          */
 989         replacing = ztest_random(2);
 990 
 991         /*
 992          * Pick a random top-level vdev.
 993          */
 994         top = ztest_random(rvd->vdev_children);
 995 
 996         /*
 997          * Pick a random leaf within it.
 998          */
 999         leaf = ztest_random(leaves);
1000 
1001         /*
1002          * Locate this vdev.
1003          */
1004         oldvd = rvd->vdev_child[top];
1005         if (zopt_mirrors >= 1) {
1006                 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
1007                 ASSERT(oldvd->vdev_children >= zopt_mirrors);
1008                 oldvd = oldvd->vdev_child[leaf / zopt_raidz];
1009         }
1010         if (zopt_raidz > 1) {
1011                 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
1012                 ASSERT(oldvd->vdev_children == zopt_raidz);
1013                 oldvd = oldvd->vdev_child[leaf % zopt_raidz];
1014         }
1015 
1016         /*
1017          * If we're already doing an attach or replace, oldvd may be a
1018          * mirror vdev -- in which case, pick a random child.
1019          */
1020         while (oldvd->vdev_children != 0) {
1021                 oldvd_has_siblings = B_TRUE;
1022                 ASSERT(oldvd->vdev_children >= 2);
1023                 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
1024         }
1025 
1026         oldguid = oldvd->vdev_guid;
1027         oldsize = vdev_get_rsize(oldvd);
1028         oldvd_is_log = oldvd->vdev_top->vdev_islog;
1029         (void) strcpy(oldpath, oldvd->vdev_path);
1030         pvd = oldvd->vdev_parent;
1031         pguid = pvd->vdev_guid;
1032 
1033         /*
1034          * If oldvd has siblings, then half of the time, detach it.
1035          */
1036         if (oldvd_has_siblings && ztest_random(2) == 0) {
1037                 spa_config_exit(spa, SCL_VDEV, FTAG);
1038                 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
1039                 if (error != 0 && error != ENODEV && error != EBUSY &&
1040                     error != ENOTSUP)
1041                         fatal(0, "detach (%s) returned %d", oldpath, error);
1042                 (void) mutex_unlock(&ztest_shared->zs_vdev_lock);
1043                 return;
1044         }
1045 
1046         /*
1047          * For the new vdev, choose with equal probability between the two
1048          * standard paths (ending in either 'a' or 'b') or a random hot spare.
1049          */
1050         if (sav->sav_count != 0 && ztest_random(3) == 0) {
1051                 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
1052                 newvd_is_spare = B_TRUE;
1053                 (void) strcpy(newpath, newvd->vdev_path);
1054         } else {
1055                 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
1056                     zopt_dir, zopt_pool, top * leaves + leaf);
1057                 if (ztest_random(2) == 0)
1058                         newpath[strlen(newpath) - 1] = 'b';
1059                 newvd = vdev_lookup_by_path(rvd, newpath);
1060         }
1061 
1062         if (newvd) {
1063                 newsize = vdev_get_rsize(newvd);
1064         } else {
1065                 /*
1066                  * Make newsize a little bigger or smaller than oldsize.
1067                  * If it's smaller, the attach should fail.
1068                  * If it's larger, and we're doing a replace,
1069                  * we should get dynamic LUN growth when we're done.
1070                  */
1071                 newsize = 10 * oldsize / (9 + ztest_random(3));
1072         }
1073 
1074         /*
1075          * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
1076          * unless it's a replace; in that case any non-replacing parent is OK.
1077          *
1078          * If newvd is already part of the pool, it should fail with EBUSY.
1079          *
1080          * If newvd is too small, it should fail with EOVERFLOW.
1081          */
1082         if (pvd->vdev_ops != &vdev_mirror_ops &&
1083             pvd->vdev_ops != &vdev_root_ops && (!replacing ||
1084             pvd->vdev_ops == &vdev_replacing_ops ||
1085             pvd->vdev_ops == &vdev_spare_ops))
1086                 expected_error = ENOTSUP;
1087         else if (newvd_is_spare && (!replacing || oldvd_is_log))
1088                 expected_error = ENOTSUP;
1089         else if (newvd == oldvd)
1090                 expected_error = replacing ? 0 : EBUSY;
1091         else if (vdev_lookup_by_path(rvd, newpath) != NULL)
1092                 expected_error = EBUSY;
1093         else if (newsize < oldsize)
1094                 expected_error = EOVERFLOW;
1095         else if (ashift > oldvd->vdev_top->vdev_ashift)
1096                 expected_error = EDOM;
1097         else
1098                 expected_error = 0;
1099 
1100         spa_config_exit(spa, SCL_VDEV, FTAG);
1101 
1102         /*
1103          * Build the nvlist describing newpath.
1104          */
1105         root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0,
1106             ashift, 0, 0, 0, 1);
1107 
1108         error = spa_vdev_attach(spa, oldguid, root, replacing);
1109 
1110         nvlist_free(root);
1111 
1112         /*
1113          * If our parent was the replacing vdev, but the replace completed,
1114          * then instead of failing with ENOTSUP we may either succeed,
1115          * fail with ENODEV, or fail with EOVERFLOW.
1116          */
1117         if (expected_error == ENOTSUP &&
1118             (error == 0 || error == ENODEV || error == EOVERFLOW))
1119                 expected_error = error;
1120 
1121         /*
1122          * If someone grew the LUN, the replacement may be too small.
1123          */
1124         if (error == EOVERFLOW || error == EBUSY)
1125                 expected_error = error;
1126 
1127         /* XXX workaround 6690467 */
1128         if (error != expected_error && expected_error != EBUSY) {
1129                 fatal(0, "attach (%s %llu, %s %llu, %d) "
1130                     "returned %d, expected %d",
1131                     oldpath, (longlong_t)oldsize, newpath,
1132                     (longlong_t)newsize, replacing, error, expected_error);
1133         }
1134 
1135         (void) mutex_unlock(&ztest_shared->zs_vdev_lock);
1136 }
1137 
1138 /*
1139  * Verify that dynamic LUN growth works as expected.
1140  */
1141 void
1142 ztest_vdev_LUN_growth(ztest_args_t *za)
1143 {
1144         spa_t *spa = za->za_spa;
1145         char dev_name[MAXPATHLEN];
1146         uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
1147         uint64_t vdev;
1148         size_t fsize;
1149         int fd;
1150 
1151         (void) mutex_lock(&ztest_shared->zs_vdev_lock);
1152 
1153         /*
1154          * Pick a random leaf vdev.
1155          */
1156         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1157         vdev = ztest_random(spa->spa_root_vdev->vdev_children * leaves);
1158         spa_config_exit(spa, SCL_VDEV, FTAG);
1159 
1160         (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
1161 
1162         if ((fd = open(dev_name, O_RDWR)) != -1) {
1163                 /*
1164                  * Determine the size.
1165                  */
1166                 fsize = lseek(fd, 0, SEEK_END);
1167 
1168                 /*
1169                  * If it's less than 2x the original size, grow by around 3%.
1170                  */
1171                 if (fsize < 2 * zopt_vdev_size) {
1172                         size_t newsize = fsize + ztest_random(fsize / 32);
1173                         (void) ftruncate(fd, newsize);
1174                         if (zopt_verbose >= 6) {
1175                                 (void) printf("%s grew from %lu to %lu bytes\n",
1176                                     dev_name, (ulong_t)fsize, (ulong_t)newsize);
1177                         }
1178                 }
1179                 (void) close(fd);
1180         }
1181 
1182         (void) mutex_unlock(&ztest_shared->zs_vdev_lock);
1183 }
1184 
1185 /* ARGSUSED */
1186 static void
1187 ztest_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
1188 {
1189         /*
1190          * Create the directory object.
1191          */
1192         VERIFY(dmu_object_claim(os, ZTEST_DIROBJ,
1193             DMU_OT_UINT64_OTHER, ZTEST_DIROBJ_BLOCKSIZE,
1194             DMU_OT_UINT64_OTHER, 5 * sizeof (ztest_block_tag_t), tx) == 0);
1195 
1196         VERIFY(zap_create_claim(os, ZTEST_MICROZAP_OBJ,
1197             DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
1198 
1199         VERIFY(zap_create_claim(os, ZTEST_FATZAP_OBJ,
1200             DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
1201 }
1202 
1203 static int
1204 ztest_destroy_cb(char *name, void *arg)
1205 {
1206         ztest_args_t *za = arg;
1207         objset_t *os;
1208         dmu_object_info_t *doi = &za->za_doi;
1209         int error;
1210 
1211         /*
1212          * Verify that the dataset contains a directory object.
1213          */
1214         error = dmu_objset_open(name, DMU_OST_OTHER,
1215             DS_MODE_USER | DS_MODE_READONLY, &os);
1216         ASSERT3U(error, ==, 0);
1217         error = dmu_object_info(os, ZTEST_DIROBJ, doi);
1218         if (error != ENOENT) {
1219                 /* We could have crashed in the middle of destroying it */
1220                 ASSERT3U(error, ==, 0);
1221                 ASSERT3U(doi->doi_type, ==, DMU_OT_UINT64_OTHER);
1222                 ASSERT3S(doi->doi_physical_blks, >=, 0);
1223         }
1224         dmu_objset_close(os);
1225 
1226         /*
1227          * Destroy the dataset.
1228          */
1229         error = dmu_objset_destroy(name);
1230         if (error) {
1231                 (void) dmu_objset_open(name, DMU_OST_OTHER,
1232                     DS_MODE_USER | DS_MODE_READONLY, &os);
1233                 fatal(0, "dmu_objset_destroy(os=%p) = %d\n", &os, error);
1234         }
1235         return (0);
1236 }
1237 
1238 /*
1239  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
1240  */
1241 static uint64_t
1242 ztest_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t object, int mode)
1243 {
1244         itx_t *itx;
1245         lr_create_t *lr;
1246         size_t namesize;
1247         char name[24];
1248 
1249         (void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object);
1250         namesize = strlen(name) + 1;
1251 
1252         itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize +
1253             ztest_random(ZIL_MAX_BLKSZ));
1254         lr = (lr_create_t *)&itx->itx_lr;
1255         bzero(lr + 1, lr->lr_common.lrc_reclen - sizeof (*lr));
1256         lr->lr_doid = object;
1257         lr->lr_foid = 0;
1258         lr->lr_mode = mode;
1259         lr->lr_uid = 0;
1260         lr->lr_gid = 0;
1261         lr->lr_gen = dmu_tx_get_txg(tx);
1262         lr->lr_crtime[0] = time(NULL);
1263         lr->lr_crtime[1] = 0;
1264         lr->lr_rdev = 0;
1265         bcopy(name, (char *)(lr + 1), namesize);
1266 
1267         return (zil_itx_assign(zilog, itx, tx));
1268 }
1269 
1270 void
1271 ztest_dmu_objset_create_destroy(ztest_args_t *za)
1272 {
1273         int error;
1274         objset_t *os, *os2;
1275         char name[100];
1276         int basemode, expected_error;
1277         zilog_t *zilog;
1278         uint64_t seq;
1279         uint64_t objects;
1280 
1281         (void) rw_rdlock(&ztest_shared->zs_name_lock);
1282         (void) snprintf(name, 100, "%s/%s_temp_%llu", za->za_pool, za->za_pool,
1283             (u_longlong_t)za->za_instance);
1284 
1285         basemode = DS_MODE_TYPE(za->za_instance);
1286         if (basemode != DS_MODE_USER && basemode != DS_MODE_OWNER)
1287                 basemode = DS_MODE_USER;
1288 
1289         /*
1290          * If this dataset exists from a previous run, process its replay log
1291          * half of the time.  If we don't replay it, then dmu_objset_destroy()
1292          * (invoked from ztest_destroy_cb() below) should just throw it away.
1293          */
1294         if (ztest_random(2) == 0 &&
1295             dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_OWNER, &os) == 0) {
1296                 zil_replay(os, os, ztest_replay_vector);
1297                 dmu_objset_close(os);
1298         }
1299 
1300         /*
1301          * There may be an old instance of the dataset we're about to
1302          * create lying around from a previous run.  If so, destroy it
1303          * and all of its snapshots.
1304          */
1305         (void) dmu_objset_find(name, ztest_destroy_cb, za,
1306             DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1307 
1308         /*
1309          * Verify that the destroyed dataset is no longer in the namespace.
1310          */
1311         error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os);
1312         if (error != ENOENT)
1313                 fatal(1, "dmu_objset_open(%s) found destroyed dataset %p",
1314                     name, os);
1315 
1316         /*
1317          * Verify that we can create a new dataset.
1318          */
1319         error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0,
1320             ztest_create_cb, NULL);
1321         if (error) {
1322                 if (error == ENOSPC) {
1323                         ztest_record_enospc("dmu_objset_create");
1324                         (void) rw_unlock(&ztest_shared->zs_name_lock);
1325                         return;
1326                 }
1327                 fatal(0, "dmu_objset_create(%s) = %d", name, error);
1328         }
1329 
1330         error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os);
1331         if (error) {
1332                 fatal(0, "dmu_objset_open(%s) = %d", name, error);
1333         }
1334 
1335         /*
1336          * Open the intent log for it.
1337          */
1338         zilog = zil_open(os, NULL);
1339 
1340         /*
1341          * Put a random number of objects in there.
1342          */
1343         objects = ztest_random(20);
1344         seq = 0;
1345         while (objects-- != 0) {
1346                 uint64_t object;
1347                 dmu_tx_t *tx = dmu_tx_create(os);
1348                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, sizeof (name));
1349                 error = dmu_tx_assign(tx, TXG_WAIT);
1350                 if (error) {
1351                         dmu_tx_abort(tx);
1352                 } else {
1353                         object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1354                             DMU_OT_NONE, 0, tx);
1355                         ztest_set_random_blocksize(os, object, tx);
1356                         seq = ztest_log_create(zilog, tx, object,
1357                             DMU_OT_UINT64_OTHER);
1358                         dmu_write(os, object, 0, sizeof (name), name, tx);
1359                         dmu_tx_commit(tx);
1360                 }
1361                 if (ztest_random(5) == 0) {
1362                         zil_commit(zilog, seq, object);
1363                 }
1364                 if (ztest_random(100) == 0) {
1365                         error = zil_suspend(zilog);
1366                         if (error == 0) {
1367                                 zil_resume(zilog);
1368                         }
1369                 }
1370         }
1371 
1372         /*
1373          * Verify that we cannot create an existing dataset.
1374          */
1375         error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0, NULL, NULL);
1376         if (error != EEXIST)
1377                 fatal(0, "created existing dataset, error = %d", error);
1378 
1379         /*
1380          * Verify that multiple dataset holds are allowed, but only when
1381          * the new access mode is compatible with the base mode.
1382          */
1383         if (basemode == DS_MODE_OWNER) {
1384                 error = dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_USER,
1385                     &os2);
1386                 if (error)
1387                         fatal(0, "dmu_objset_open('%s') = %d", name, error);
1388                 else
1389                         dmu_objset_close(os2);
1390         }
1391         error = dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_OWNER, &os2);
1392         expected_error = (basemode == DS_MODE_OWNER) ? EBUSY : 0;
1393         if (error != expected_error)
1394                 fatal(0, "dmu_objset_open('%s') = %d, expected %d",
1395                     name, error, expected_error);
1396         if (error == 0)
1397                 dmu_objset_close(os2);
1398 
1399         zil_close(zilog);
1400         dmu_objset_close(os);
1401 
1402         error = dmu_objset_destroy(name);
1403         if (error)
1404                 fatal(0, "dmu_objset_destroy(%s) = %d", name, error);
1405 
1406         (void) rw_unlock(&ztest_shared->zs_name_lock);
1407 }
1408 
1409 /*
1410  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
1411  */
1412 void
1413 ztest_dmu_snapshot_create_destroy(ztest_args_t *za)
1414 {
1415         int error;
1416         objset_t *os = za->za_os;
1417         char snapname[100];
1418         char osname[MAXNAMELEN];
1419 
1420         (void) rw_rdlock(&ztest_shared->zs_name_lock);
1421         dmu_objset_name(os, osname);
1422         (void) snprintf(snapname, 100, "%s@%llu", osname,
1423             (u_longlong_t)za->za_instance);
1424 
1425         error = dmu_objset_destroy(snapname);
1426         if (error != 0 && error != ENOENT)
1427                 fatal(0, "dmu_objset_destroy() = %d", error);
1428         error = dmu_objset_snapshot(osname, strchr(snapname, '@')+1, FALSE);
1429         if (error == ENOSPC)
1430                 ztest_record_enospc("dmu_take_snapshot");
1431         else if (error != 0 && error != EEXIST)
1432                 fatal(0, "dmu_take_snapshot() = %d", error);
1433         (void) rw_unlock(&ztest_shared->zs_name_lock);
1434 }
1435 
1436 /*
1437  * Verify that dmu_object_{alloc,free} work as expected.
1438  */
1439 void
1440 ztest_dmu_object_alloc_free(ztest_args_t *za)
1441 {
1442         objset_t *os = za->za_os;
1443         dmu_buf_t *db;
1444         dmu_tx_t *tx;
1445         uint64_t batchobj, object, batchsize, endoff, temp;
1446         int b, c, error, bonuslen;
1447         dmu_object_info_t *doi = &za->za_doi;
1448         char osname[MAXNAMELEN];
1449 
1450         dmu_objset_name(os, osname);
1451 
1452         endoff = -8ULL;
1453         batchsize = 2;
1454 
1455         /*
1456          * Create a batch object if necessary, and record it in the directory.
1457          */
1458         VERIFY3U(0, ==, dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
1459             sizeof (uint64_t), &batchobj));
1460         if (batchobj == 0) {
1461                 tx = dmu_tx_create(os);
1462                 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff,
1463                     sizeof (uint64_t));
1464                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1465                 error = dmu_tx_assign(tx, TXG_WAIT);
1466                 if (error) {
1467                         ztest_record_enospc("create a batch object");
1468                         dmu_tx_abort(tx);
1469                         return;
1470                 }
1471                 batchobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1472                     DMU_OT_NONE, 0, tx);
1473                 ztest_set_random_blocksize(os, batchobj, tx);
1474                 dmu_write(os, ZTEST_DIROBJ, za->za_diroff,
1475                     sizeof (uint64_t), &batchobj, tx);
1476                 dmu_tx_commit(tx);
1477         }
1478 
1479         /*
1480          * Destroy the previous batch of objects.
1481          */
1482         for (b = 0; b < batchsize; b++) {
1483                 VERIFY3U(0, ==, dmu_read(os, batchobj, b * sizeof (uint64_t),
1484                     sizeof (uint64_t), &object));
1485                 if (object == 0)
1486                         continue;
1487                 /*
1488                  * Read and validate contents.
1489                  * We expect the nth byte of the bonus buffer to be n.
1490                  */
1491                 VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db));
1492                 za->za_dbuf = db;
1493 
1494                 dmu_object_info_from_db(db, doi);
1495                 ASSERT(doi->doi_type == DMU_OT_UINT64_OTHER);
1496                 ASSERT(doi->doi_bonus_type == DMU_OT_PLAIN_OTHER);
1497                 ASSERT3S(doi->doi_physical_blks, >=, 0);
1498 
1499                 bonuslen = doi->doi_bonus_size;
1500 
1501                 for (c = 0; c < bonuslen; c++) {
1502                         if (((uint8_t *)db->db_data)[c] !=
1503                             (uint8_t)(c + bonuslen)) {
1504                                 fatal(0,
1505                                     "bad bonus: %s, obj %llu, off %d: %u != %u",
1506                                     osname, object, c,
1507                                     ((uint8_t *)db->db_data)[c],
1508                                     (uint8_t)(c + bonuslen));
1509                         }
1510                 }
1511 
1512                 dmu_buf_rele(db, FTAG);
1513                 za->za_dbuf = NULL;
1514 
1515                 /*
1516                  * We expect the word at endoff to be our object number.
1517                  */
1518                 VERIFY(0 == dmu_read(os, object, endoff,
1519                     sizeof (uint64_t), &temp));
1520 
1521                 if (temp != object) {
1522                         fatal(0, "bad data in %s, got %llu, expected %llu",
1523                             osname, temp, object);
1524                 }
1525 
1526                 /*
1527                  * Destroy old object and clear batch entry.
1528                  */
1529                 tx = dmu_tx_create(os);
1530                 dmu_tx_hold_write(tx, batchobj,
1531                     b * sizeof (uint64_t), sizeof (uint64_t));
1532                 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1533                 error = dmu_tx_assign(tx, TXG_WAIT);
1534                 if (error) {
1535                         ztest_record_enospc("free object");
1536                         dmu_tx_abort(tx);
1537                         return;
1538                 }
1539                 error = dmu_object_free(os, object, tx);
1540                 if (error) {
1541                         fatal(0, "dmu_object_free('%s', %llu) = %d",
1542                             osname, object, error);
1543                 }
1544                 object = 0;
1545 
1546                 dmu_object_set_checksum(os, batchobj,
1547                     ztest_random_checksum(), tx);
1548                 dmu_object_set_compress(os, batchobj,
1549                     ztest_random_compress(), tx);
1550 
1551                 dmu_write(os, batchobj, b * sizeof (uint64_t),
1552                     sizeof (uint64_t), &object, tx);
1553 
1554                 dmu_tx_commit(tx);
1555         }
1556 
1557         /*
1558          * Before creating the new batch of objects, generate a bunch of churn.
1559          */
1560         for (b = ztest_random(100); b > 0; b--) {
1561                 tx = dmu_tx_create(os);
1562                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1563                 error = dmu_tx_assign(tx, TXG_WAIT);
1564                 if (error) {
1565                         ztest_record_enospc("churn objects");
1566                         dmu_tx_abort(tx);
1567                         return;
1568                 }
1569                 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1570                     DMU_OT_NONE, 0, tx);
1571                 ztest_set_random_blocksize(os, object, tx);
1572                 error = dmu_object_free(os, object, tx);
1573                 if (error) {
1574                         fatal(0, "dmu_object_free('%s', %llu) = %d",
1575                             osname, object, error);
1576                 }
1577                 dmu_tx_commit(tx);
1578         }
1579 
1580         /*
1581          * Create a new batch of objects with randomly chosen
1582          * blocksizes and record them in the batch directory.
1583          */
1584         for (b = 0; b < batchsize; b++) {
1585                 uint32_t va_blksize;
1586                 u_longlong_t va_nblocks;
1587 
1588                 tx = dmu_tx_create(os);
1589                 dmu_tx_hold_write(tx, batchobj, b * sizeof (uint64_t),
1590                     sizeof (uint64_t));
1591                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1592                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, endoff,
1593                     sizeof (uint64_t));
1594                 error = dmu_tx_assign(tx, TXG_WAIT);
1595                 if (error) {
1596                         ztest_record_enospc("create batchobj");
1597                         dmu_tx_abort(tx);
1598                         return;
1599                 }
1600                 bonuslen = (int)ztest_random(dmu_bonus_max()) + 1;
1601 
1602                 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1603                     DMU_OT_PLAIN_OTHER, bonuslen, tx);
1604 
1605                 ztest_set_random_blocksize(os, object, tx);
1606 
1607                 dmu_object_set_checksum(os, object,
1608                     ztest_random_checksum(), tx);
1609                 dmu_object_set_compress(os, object,
1610                     ztest_random_compress(), tx);
1611 
1612                 dmu_write(os, batchobj, b * sizeof (uint64_t),
1613                     sizeof (uint64_t), &object, tx);
1614 
1615                 /*
1616                  * Write to both the bonus buffer and the regular data.
1617                  */
1618                 VERIFY(dmu_bonus_hold(os, object, FTAG, &db) == 0);
1619                 za->za_dbuf = db;
1620                 ASSERT3U(bonuslen, <=, db->db_size);
1621 
1622                 dmu_object_size_from_db(db, &va_blksize, &va_nblocks);
1623                 ASSERT3S(va_nblocks, >=, 0);
1624 
1625                 dmu_buf_will_dirty(db, tx);
1626 
1627                 /*
1628                  * See comments above regarding the contents of
1629                  * the bonus buffer and the word at endoff.
1630                  */
1631                 for (c = 0; c < bonuslen; c++)
1632                         ((uint8_t *)db->db_data)[c] = (uint8_t)(c + bonuslen);
1633 
1634                 dmu_buf_rele(db, FTAG);
1635                 za->za_dbuf = NULL;
1636 
1637                 /*
1638                  * Write to a large offset to increase indirection.
1639                  */
1640                 dmu_write(os, object, endoff, sizeof (uint64_t), &object, tx);
1641 
1642                 dmu_tx_commit(tx);
1643         }
1644 }
1645 
1646 /*
1647  * Verify that dmu_{read,write} work as expected.
1648  */
1649 typedef struct bufwad {
1650         uint64_t        bw_index;
1651         uint64_t        bw_txg;
1652         uint64_t        bw_data;
1653 } bufwad_t;
1654 
1655 typedef struct dmu_read_write_dir {
1656         uint64_t        dd_packobj;
1657         uint64_t        dd_bigobj;
1658         uint64_t        dd_chunk;
1659 } dmu_read_write_dir_t;
1660 
1661 void
1662 ztest_dmu_read_write(ztest_args_t *za)
1663 {
1664         objset_t *os = za->za_os;
1665         dmu_read_write_dir_t dd;
1666         dmu_tx_t *tx;
1667         int i, freeit, error;
1668         uint64_t n, s, txg;
1669         bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
1670         uint64_t packoff, packsize, bigoff, bigsize;
1671         uint64_t regions = 997;
1672         uint64_t stride = 123456789ULL;
1673         uint64_t width = 40;
1674         int free_percent = 5;
1675 
1676         /*
1677          * This test uses two objects, packobj and bigobj, that are always
1678          * updated together (i.e. in the same tx) so that their contents are
1679          * in sync and can be compared.  Their contents relate to each other
1680          * in a simple way: packobj is a dense array of 'bufwad' structures,
1681          * while bigobj is a sparse array of the same bufwads.  Specifically,
1682          * for any index n, there are three bufwads that should be identical:
1683          *
1684          *      packobj, at offset n * sizeof (bufwad_t)
1685          *      bigobj, at the head of the nth chunk
1686          *      bigobj, at the tail of the nth chunk
1687          *
1688          * The chunk size is arbitrary. It doesn't have to be a power of two,
1689          * and it doesn't have any relation to the object blocksize.
1690          * The only requirement is that it can hold at least two bufwads.
1691          *
1692          * Normally, we write the bufwad to each of these locations.
1693          * However, free_percent of the time we instead write zeroes to
1694          * packobj and perform a dmu_free_range() on bigobj.  By comparing
1695          * bigobj to packobj, we can verify that the DMU is correctly
1696          * tracking which parts of an object are allocated and free,
1697          * and that the contents of the allocated blocks are correct.
1698          */
1699 
1700         /*
1701          * Read the directory info.  If it's the first time, set things up.
1702          */
1703         VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
1704             sizeof (dd), &dd));
1705         if (dd.dd_chunk == 0) {
1706                 ASSERT(dd.dd_packobj == 0);
1707                 ASSERT(dd.dd_bigobj == 0);
1708                 tx = dmu_tx_create(os);
1709                 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (dd));
1710                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1711                 error = dmu_tx_assign(tx, TXG_WAIT);
1712                 if (error) {
1713                         ztest_record_enospc("create r/w directory");
1714                         dmu_tx_abort(tx);
1715                         return;
1716                 }
1717 
1718                 dd.dd_packobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1719                     DMU_OT_NONE, 0, tx);
1720                 dd.dd_bigobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1721                     DMU_OT_NONE, 0, tx);
1722                 dd.dd_chunk = (1000 + ztest_random(1000)) * sizeof (uint64_t);
1723 
1724                 ztest_set_random_blocksize(os, dd.dd_packobj, tx);
1725                 ztest_set_random_blocksize(os, dd.dd_bigobj, tx);
1726 
1727                 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (dd), &dd,
1728                     tx);
1729                 dmu_tx_commit(tx);
1730         }
1731 
1732         /*
1733          * Prefetch a random chunk of the big object.
1734          * Our aim here is to get some async reads in flight
1735          * for blocks that we may free below; the DMU should
1736          * handle this race correctly.
1737          */
1738         n = ztest_random(regions) * stride + ztest_random(width);
1739         s = 1 + ztest_random(2 * width - 1);
1740         dmu_prefetch(os, dd.dd_bigobj, n * dd.dd_chunk, s * dd.dd_chunk);
1741 
1742         /*
1743          * Pick a random index and compute the offsets into packobj and bigobj.
1744          */
1745         n = ztest_random(regions) * stride + ztest_random(width);
1746         s = 1 + ztest_random(width - 1);
1747 
1748         packoff = n * sizeof (bufwad_t);
1749         packsize = s * sizeof (bufwad_t);
1750 
1751         bigoff = n * dd.dd_chunk;
1752         bigsize = s * dd.dd_chunk;
1753 
1754         packbuf = umem_alloc(packsize, UMEM_NOFAIL);
1755         bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
1756 
1757         /*
1758          * free_percent of the time, free a range of bigobj rather than
1759          * overwriting it.
1760          */
1761         freeit = (ztest_random(100) < free_percent);
1762 
1763         /*
1764          * Read the current contents of our objects.
1765          */
1766         error = dmu_read(os, dd.dd_packobj, packoff, packsize, packbuf);
1767         ASSERT3U(error, ==, 0);
1768         error = dmu_read(os, dd.dd_bigobj, bigoff, bigsize, bigbuf);
1769         ASSERT3U(error, ==, 0);
1770 
1771         /*
1772          * Get a tx for the mods to both packobj and bigobj.
1773          */
1774         tx = dmu_tx_create(os);
1775 
1776         dmu_tx_hold_write(tx, dd.dd_packobj, packoff, packsize);
1777 
1778         if (freeit)
1779                 dmu_tx_hold_free(tx, dd.dd_bigobj, bigoff, bigsize);
1780         else
1781                 dmu_tx_hold_write(tx, dd.dd_bigobj, bigoff, bigsize);
1782 
1783         error = dmu_tx_assign(tx, TXG_WAIT);
1784 
1785         if (error) {
1786                 ztest_record_enospc("dmu r/w range");
1787                 dmu_tx_abort(tx);
1788                 umem_free(packbuf, packsize);
1789                 umem_free(bigbuf, bigsize);
1790                 return;
1791         }
1792 
1793         txg = dmu_tx_get_txg(tx);
1794 
1795         /*
1796          * For each index from n to n + s, verify that the existing bufwad
1797          * in packobj matches the bufwads at the head and tail of the
1798          * corresponding chunk in bigobj.  Then update all three bufwads
1799          * with the new values we want to write out.
1800          */
1801         for (i = 0; i < s; i++) {
1802                 /* LINTED */
1803                 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
1804                 /* LINTED */
1805                 bigH = (bufwad_t *)((char *)bigbuf + i * dd.dd_chunk);
1806                 /* LINTED */
1807                 bigT = (bufwad_t *)((char *)bigH + dd.dd_chunk) - 1;
1808 
1809                 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
1810                 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
1811 
1812                 if (pack->bw_txg > txg)
1813                         fatal(0, "future leak: got %llx, open txg is %llx",
1814                             pack->bw_txg, txg);
1815 
1816                 if (pack->bw_data != 0 && pack->bw_index != n + i)
1817                         fatal(0, "wrong index: got %llx, wanted %llx+%llx",
1818                             pack->bw_index, n, i);
1819 
1820                 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
1821                         fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
1822 
1823                 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
1824                         fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
1825 
1826                 if (freeit) {
1827                         bzero(pack, sizeof (bufwad_t));
1828                 } else {
1829                         pack->bw_index = n + i;
1830                         pack->bw_txg = txg;
1831                         pack->bw_data = 1 + ztest_random(-2ULL);
1832                 }
1833                 *bigH = *pack;
1834                 *bigT = *pack;
1835         }
1836 
1837         /*
1838          * We've verified all the old bufwads, and made new ones.
1839          * Now write them out.
1840          */
1841         dmu_write(os, dd.dd_packobj, packoff, packsize, packbuf, tx);
1842 
1843         if (freeit) {
1844                 if (zopt_verbose >= 6) {
1845                         (void) printf("freeing offset %llx size %llx"
1846                             " txg %llx\n",
1847                             (u_longlong_t)bigoff,
1848                             (u_longlong_t)bigsize,
1849                             (u_longlong_t)txg);
1850                 }
1851                 VERIFY(0 == dmu_free_range(os, dd.dd_bigobj, bigoff,
1852                     bigsize, tx));
1853         } else {
1854                 if (zopt_verbose >= 6) {
1855                         (void) printf("writing offset %llx size %llx"
1856                             " txg %llx\n",
1857                             (u_longlong_t)bigoff,
1858                             (u_longlong_t)bigsize,
1859                             (u_longlong_t)txg);
1860                 }
1861                 dmu_write(os, dd.dd_bigobj, bigoff, bigsize, bigbuf, tx);
1862         }
1863 
1864         dmu_tx_commit(tx);
1865 
1866         /*
1867          * Sanity check the stuff we just wrote.
1868          */
1869         {
1870                 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
1871                 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
1872 
1873                 VERIFY(0 == dmu_read(os, dd.dd_packobj, packoff,
1874                     packsize, packcheck));
1875                 VERIFY(0 == dmu_read(os, dd.dd_bigobj, bigoff,
1876                     bigsize, bigcheck));
1877 
1878                 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
1879                 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
1880 
1881                 umem_free(packcheck, packsize);
1882                 umem_free(bigcheck, bigsize);
1883         }
1884 
1885         umem_free(packbuf, packsize);
1886         umem_free(bigbuf, bigsize);
1887 }
1888 
1889 void
1890 ztest_dmu_check_future_leak(ztest_args_t *za)
1891 {
1892         objset_t *os = za->za_os;
1893         dmu_buf_t *db;
1894         ztest_block_tag_t *bt;
1895         dmu_object_info_t *doi = &za->za_doi;
1896 
1897         /*
1898          * Make sure that, if there is a write record in the bonus buffer
1899          * of the ZTEST_DIROBJ, that the txg for this record is <= the
1900          * last synced txg of the pool.
1901          */
1902         VERIFY(dmu_bonus_hold(os, ZTEST_DIROBJ, FTAG, &db) == 0);
1903         za->za_dbuf = db;
1904         VERIFY(dmu_object_info(os, ZTEST_DIROBJ, doi) == 0);
1905         ASSERT3U(doi->doi_bonus_size, >=, sizeof (*bt));
1906         ASSERT3U(doi->doi_bonus_size, <=, db->db_size);
1907         ASSERT3U(doi->doi_bonus_size % sizeof (*bt), ==, 0);
1908         bt = (void *)((char *)db->db_data + doi->doi_bonus_size - sizeof (*bt));
1909         if (bt->bt_objset != 0) {
1910                 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1911                 ASSERT3U(bt->bt_object, ==, ZTEST_DIROBJ);
1912                 ASSERT3U(bt->bt_offset, ==, -1ULL);
1913                 ASSERT3U(bt->bt_txg, <, spa_first_txg(za->za_spa));
1914         }
1915         dmu_buf_rele(db, FTAG);
1916         za->za_dbuf = NULL;
1917 }
1918 
1919 void
1920 ztest_dmu_write_parallel(ztest_args_t *za)
1921 {
1922         objset_t *os = za->za_os;
1923         ztest_block_tag_t *rbt = &za->za_rbt;
1924         ztest_block_tag_t *wbt = &za->za_wbt;
1925         const size_t btsize = sizeof (ztest_block_tag_t);
1926         dmu_buf_t *db;
1927         int b, error;
1928         int bs = ZTEST_DIROBJ_BLOCKSIZE;
1929         int do_free = 0;
1930         uint64_t off, txg, txg_how;
1931         mutex_t *lp;
1932         char osname[MAXNAMELEN];
1933         char iobuf[SPA_MAXBLOCKSIZE];
1934         blkptr_t blk = { 0 };
1935         uint64_t blkoff;
1936         zbookmark_t zb;
1937         dmu_tx_t *tx = dmu_tx_create(os);
1938 
1939         dmu_objset_name(os, osname);
1940 
1941         /*
1942          * Have multiple threads write to large offsets in ZTEST_DIROBJ
1943          * to verify that having multiple threads writing to the same object
1944          * in parallel doesn't cause any trouble.
1945          */
1946         if (ztest_random(4) == 0) {
1947                 /*
1948                  * Do the bonus buffer instead of a regular block.
1949                  * We need a lock to serialize resize vs. others,
1950                  * so we hash on the objset ID.
1951                  */
1952                 b = dmu_objset_id(os) % ZTEST_SYNC_LOCKS;
1953                 off = -1ULL;
1954                 dmu_tx_hold_bonus(tx, ZTEST_DIROBJ);
1955         } else {
1956                 b = ztest_random(ZTEST_SYNC_LOCKS);
1957                 off = za->za_diroff_shared + (b << SPA_MAXBLOCKSHIFT);
1958                 if (ztest_random(4) == 0) {
1959                         do_free = 1;
1960                         dmu_tx_hold_free(tx, ZTEST_DIROBJ, off, bs);
1961                 } else {
1962                         dmu_tx_hold_write(tx, ZTEST_DIROBJ, off, bs);
1963                 }
1964         }
1965 
1966         txg_how = ztest_random(2) == 0 ? TXG_WAIT : TXG_NOWAIT;
1967         error = dmu_tx_assign(tx, txg_how);
1968         if (error) {
1969                 if (error == ERESTART) {
1970                         ASSERT(txg_how == TXG_NOWAIT);
1971                         dmu_tx_wait(tx);
1972                 } else {
1973                         ztest_record_enospc("dmu write parallel");
1974                 }
1975                 dmu_tx_abort(tx);
1976                 return;
1977         }
1978         txg = dmu_tx_get_txg(tx);
1979 
1980         lp = &ztest_shared->zs_sync_lock[b];
1981         (void) mutex_lock(lp);
1982 
1983         wbt->bt_objset = dmu_objset_id(os);
1984         wbt->bt_object = ZTEST_DIROBJ;
1985         wbt->bt_offset = off;
1986         wbt->bt_txg = txg;
1987         wbt->bt_thread = za->za_instance;
1988         wbt->bt_seq = ztest_shared->zs_seq[b]++;  /* protected by lp */
1989 
1990         /*
1991          * Occasionally, write an all-zero block to test the behavior
1992          * of blocks that compress into holes.
1993          */
1994         if (off != -1ULL && ztest_random(8) == 0)
1995                 bzero(wbt, btsize);
1996 
1997         if (off == -1ULL) {
1998                 dmu_object_info_t *doi = &za->za_doi;
1999                 char *dboff;
2000 
2001                 VERIFY(dmu_bonus_hold(os, ZTEST_DIROBJ, FTAG, &db) == 0);
2002                 za->za_dbuf = db;
2003                 dmu_object_info_from_db(db, doi);
2004                 ASSERT3U(doi->doi_bonus_size, <=, db->db_size);
2005                 ASSERT3U(doi->doi_bonus_size, >=, btsize);
2006                 ASSERT3U(doi->doi_bonus_size % btsize, ==, 0);
2007                 dboff = (char *)db->db_data + doi->doi_bonus_size - btsize;
2008                 bcopy(dboff, rbt, btsize);
2009                 if (rbt->bt_objset != 0) {
2010                         ASSERT3U(rbt->bt_objset, ==, wbt->bt_objset);
2011                         ASSERT3U(rbt->bt_object, ==, wbt->bt_object);
2012                         ASSERT3U(rbt->bt_offset, ==, wbt->bt_offset);
2013                         ASSERT3U(rbt->bt_txg, <=, wbt->bt_txg);
2014                 }
2015                 if (ztest_random(10) == 0) {
2016                         int newsize = (ztest_random(db->db_size /
2017                             btsize) + 1) * btsize;
2018 
2019                         ASSERT3U(newsize, >=, btsize);
2020                         ASSERT3U(newsize, <=, db->db_size);
2021                         VERIFY3U(dmu_set_bonus(db, newsize, tx), ==, 0);
2022                         dboff = (char *)db->db_data + newsize - btsize;
2023                 }
2024                 dmu_buf_will_dirty(db, tx);
2025                 bcopy(wbt, dboff, btsize);
2026                 dmu_buf_rele(db, FTAG);
2027                 za->za_dbuf = NULL;
2028         } else if (do_free) {
2029                 VERIFY(dmu_free_range(os, ZTEST_DIROBJ, off, bs, tx) == 0);
2030         } else {
2031                 dmu_write(os, ZTEST_DIROBJ, off, btsize, wbt, tx);
2032         }
2033 
2034         (void) mutex_unlock(lp);
2035 
2036         if (ztest_random(1000) == 0)
2037                 (void) poll(NULL, 0, 1); /* open dn_notxholds window */
2038 
2039         dmu_tx_commit(tx);
2040 
2041         if (ztest_random(10000) == 0)
2042                 txg_wait_synced(dmu_objset_pool(os), txg);
2043 
2044         if (off == -1ULL || do_free)
2045                 return;
2046 
2047         if (ztest_random(2) != 0)
2048                 return;
2049 
2050         /*
2051          * dmu_sync() the block we just wrote.
2052          */
2053         (void) mutex_lock(lp);
2054 
2055         blkoff = P2ALIGN_TYPED(off, bs, uint64_t);
2056         error = dmu_buf_hold(os, ZTEST_DIROBJ, blkoff, FTAG, &db);
2057         za->za_dbuf = db;
2058         if (error) {
2059                 (void) mutex_unlock(lp);
2060                 return;
2061         }
2062         blkoff = off - blkoff;
2063         error = dmu_sync(NULL, db, &blk, txg, NULL, NULL);
2064         dmu_buf_rele(db, FTAG);
2065         za->za_dbuf = NULL;
2066 
2067         (void) mutex_unlock(lp);
2068 
2069         if (error)
2070                 return;
2071 
2072         if (blk.blk_birth == 0)         /* concurrent free */
2073                 return;
2074 
2075         txg_suspend(dmu_objset_pool(os));
2076 
2077         ASSERT(blk.blk_fill == 1);
2078         ASSERT3U(BP_GET_TYPE(&blk), ==, DMU_OT_UINT64_OTHER);
2079         ASSERT3U(BP_GET_LEVEL(&blk), ==, 0);
2080         ASSERT3U(BP_GET_LSIZE(&blk), ==, bs);
2081 
2082         /*
2083          * Read the block that dmu_sync() returned to make sure its contents
2084          * match what we wrote.  We do this while still txg_suspend()ed
2085          * to ensure that the block can't be reused before we read it.
2086          */
2087         zb.zb_objset = dmu_objset_id(os);
2088         zb.zb_object = ZTEST_DIROBJ;
2089         zb.zb_level = 0;
2090         zb.zb_blkid = off / bs;
2091         error = zio_wait(zio_read(NULL, za->za_spa, &blk, iobuf, bs,
2092             NULL, NULL, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_MUSTSUCCEED, &zb));
2093         ASSERT3U(error, ==, 0);
2094 
2095         txg_resume(dmu_objset_pool(os));
2096 
2097         bcopy(&iobuf[blkoff], rbt, btsize);
2098 
2099         if (rbt->bt_objset == 0)             /* concurrent free */
2100                 return;
2101 
2102         if (wbt->bt_objset == 0)             /* all-zero overwrite */
2103                 return;
2104 
2105         ASSERT3U(rbt->bt_objset, ==, wbt->bt_objset);
2106         ASSERT3U(rbt->bt_object, ==, wbt->bt_object);
2107         ASSERT3U(rbt->bt_offset, ==, wbt->bt_offset);
2108 
2109         /*
2110          * The semantic of dmu_sync() is that we always push the most recent
2111          * version of the data, so in the face of concurrent updates we may
2112          * see a newer version of the block.  That's OK.
2113          */
2114         ASSERT3U(rbt->bt_txg, >=, wbt->bt_txg);
2115         if (rbt->bt_thread == wbt->bt_thread)
2116                 ASSERT3U(rbt->bt_seq, ==, wbt->bt_seq);
2117         else
2118                 ASSERT3U(rbt->bt_seq, >, wbt->bt_seq);
2119 }
2120 
2121 /*
2122  * Verify that zap_{create,destroy,add,remove,update} work as expected.
2123  */
2124 #define ZTEST_ZAP_MIN_INTS      1
2125 #define ZTEST_ZAP_MAX_INTS      4
2126 #define ZTEST_ZAP_MAX_PROPS     1000
2127 
2128 void
2129 ztest_zap(ztest_args_t *za)
2130 {
2131         objset_t *os = za->za_os;
2132         uint64_t object;
2133         uint64_t txg, last_txg;
2134         uint64_t value[ZTEST_ZAP_MAX_INTS];
2135         uint64_t zl_ints, zl_intsize, prop;
2136         int i, ints;
2137         dmu_tx_t *tx;
2138         char propname[100], txgname[100];
2139         int error;
2140         char osname[MAXNAMELEN];
2141         char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
2142 
2143         dmu_objset_name(os, osname);
2144 
2145         /*
2146          * Create a new object if necessary, and record it in the directory.
2147          */
2148         VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
2149             sizeof (uint64_t), &object));
2150 
2151         if (object == 0) {
2152                 tx = dmu_tx_create(os);
2153                 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff,
2154                     sizeof (uint64_t));
2155                 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL);
2156                 error = dmu_tx_assign(tx, TXG_WAIT);
2157                 if (error) {
2158                         ztest_record_enospc("create zap test obj");
2159                         dmu_tx_abort(tx);
2160                         return;
2161                 }
2162                 object = zap_create(os, DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx);
2163                 if (error) {
2164                         fatal(0, "zap_create('%s', %llu) = %d",
2165                             osname, object, error);
2166                 }
2167                 ASSERT(object != 0);
2168                 dmu_write(os, ZTEST_DIROBJ, za->za_diroff,
2169                     sizeof (uint64_t), &object, tx);
2170                 /*
2171                  * Generate a known hash collision, and verify that
2172                  * we can lookup and remove both entries.
2173                  */
2174                 for (i = 0; i < 2; i++) {
2175                         value[i] = i;
2176                         error = zap_add(os, object, hc[i], sizeof (uint64_t),
2177                             1, &value[i], tx);
2178                         ASSERT3U(error, ==, 0);
2179                 }
2180                 for (i = 0; i < 2; i++) {
2181                         error = zap_add(os, object, hc[i], sizeof (uint64_t),
2182                             1, &value[i], tx);
2183                         ASSERT3U(error, ==, EEXIST);
2184                         error = zap_length(os, object, hc[i],
2185                             &zl_intsize, &zl_ints);
2186                         ASSERT3U(error, ==, 0);
2187                         ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2188                         ASSERT3U(zl_ints, ==, 1);
2189                 }
2190                 for (i = 0; i < 2; i++) {
2191                         error = zap_remove(os, object, hc[i], tx);
2192                         ASSERT3U(error, ==, 0);
2193                 }
2194 
2195                 dmu_tx_commit(tx);
2196         }
2197 
2198         ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
2199 
2200         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
2201         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
2202         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
2203         bzero(value, sizeof (value));
2204         last_txg = 0;
2205 
2206         /*
2207          * If these zap entries already exist, validate their contents.
2208          */
2209         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
2210         if (error == 0) {
2211                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2212                 ASSERT3U(zl_ints, ==, 1);
2213 
2214                 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
2215                     zl_ints, &last_txg) == 0);
2216 
2217                 VERIFY(zap_length(os, object, propname, &zl_intsize,
2218                     &zl_ints) == 0);
2219 
2220                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2221                 ASSERT3U(zl_ints, ==, ints);
2222 
2223                 VERIFY(zap_lookup(os, object, propname, zl_intsize,
2224                     zl_ints, value) == 0);
2225 
2226                 for (i = 0; i < ints; i++) {
2227                         ASSERT3U(value[i], ==, last_txg + object + i);
2228                 }
2229         } else {
2230                 ASSERT3U(error, ==, ENOENT);
2231         }
2232 
2233         /*
2234          * Atomically update two entries in our zap object.
2235          * The first is named txg_%llu, and contains the txg
2236          * in which the property was last updated.  The second
2237          * is named prop_%llu, and the nth element of its value
2238          * should be txg + object + n.
2239          */
2240         tx = dmu_tx_create(os);
2241         dmu_tx_hold_zap(tx, object, TRUE, NULL);
2242         error = dmu_tx_assign(tx, TXG_WAIT);
2243         if (error) {
2244                 ztest_record_enospc("create zap entry");
2245                 dmu_tx_abort(tx);
2246                 return;
2247         }
2248         txg = dmu_tx_get_txg(tx);
2249 
2250         if (last_txg > txg)
2251                 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
2252 
2253         for (i = 0; i < ints; i++)
2254                 value[i] = txg + object + i;
2255 
2256         error = zap_update(os, object, txgname, sizeof (uint64_t), 1, &txg, tx);
2257         if (error)
2258                 fatal(0, "zap_update('%s', %llu, '%s') = %d",
2259                     osname, object, txgname, error);
2260 
2261         error = zap_update(os, object, propname, sizeof (uint64_t),
2262             ints, value, tx);
2263         if (error)
2264                 fatal(0, "zap_update('%s', %llu, '%s') = %d",
2265                     osname, object, propname, error);
2266 
2267         dmu_tx_commit(tx);
2268 
2269         /*
2270          * Remove a random pair of entries.
2271          */
2272         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
2273         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
2274         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
2275 
2276         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
2277 
2278         if (error == ENOENT)
2279                 return;
2280 
2281         ASSERT3U(error, ==, 0);
2282 
2283         tx = dmu_tx_create(os);
2284         dmu_tx_hold_zap(tx, object, TRUE, NULL);
2285         error = dmu_tx_assign(tx, TXG_WAIT);
2286         if (error) {
2287                 ztest_record_enospc("remove zap entry");
2288                 dmu_tx_abort(tx);
2289                 return;
2290         }
2291         error = zap_remove(os, object, txgname, tx);
2292         if (error)
2293                 fatal(0, "zap_remove('%s', %llu, '%s') = %d",
2294                     osname, object, txgname, error);
2295 
2296         error = zap_remove(os, object, propname, tx);
2297         if (error)
2298                 fatal(0, "zap_remove('%s', %llu, '%s') = %d",
2299                     osname, object, propname, error);
2300 
2301         dmu_tx_commit(tx);
2302 
2303         /*
2304          * Once in a while, destroy the object.
2305          */
2306         if (ztest_random(1000) != 0)
2307                 return;
2308 
2309         tx = dmu_tx_create(os);
2310         dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t));
2311         dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
2312         error = dmu_tx_assign(tx, TXG_WAIT);
2313         if (error) {
2314                 ztest_record_enospc("destroy zap object");
2315                 dmu_tx_abort(tx);
2316                 return;
2317         }
2318         error = zap_destroy(os, object, tx);
2319         if (error)
2320                 fatal(0, "zap_destroy('%s', %llu) = %d",
2321                     osname, object, error);
2322         object = 0;
2323         dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t),
2324             &object, tx);
2325         dmu_tx_commit(tx);
2326 }
2327 
2328 void
2329 ztest_zap_parallel(ztest_args_t *za)
2330 {
2331         objset_t *os = za->za_os;
2332         uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
2333         dmu_tx_t *tx;
2334         int i, namelen, error;
2335         char name[20], string_value[20];
2336         void *data;
2337 
2338         /*
2339          * Generate a random name of the form 'xxx.....' where each
2340          * x is a random printable character and the dots are dots.
2341          * There are 94 such characters, and the name length goes from
2342          * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
2343          */
2344         namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
2345 
2346         for (i = 0; i < 3; i++)
2347                 name[i] = '!' + ztest_random('~' - '!' + 1);
2348         for (; i < namelen - 1; i++)
2349                 name[i] = '.';
2350         name[i] = '\0';
2351 
2352         if (ztest_random(2) == 0)
2353                 object = ZTEST_MICROZAP_OBJ;
2354         else
2355                 object = ZTEST_FATZAP_OBJ;
2356 
2357         if ((namelen & 1) || object == ZTEST_MICROZAP_OBJ) {
2358                 wsize = sizeof (txg);
2359                 wc = 1;
2360                 data = &txg;
2361         } else {
2362                 wsize = 1;
2363                 wc = namelen;
2364                 data = string_value;
2365         }
2366 
2367         count = -1ULL;
2368         VERIFY(zap_count(os, object, &count) == 0);
2369         ASSERT(count != -1ULL);
2370 
2371         /*
2372          * Select an operation: length, lookup, add, update, remove.
2373          */
2374         i = ztest_random(5);
2375 
2376         if (i >= 2) {
2377                 tx = dmu_tx_create(os);
2378                 dmu_tx_hold_zap(tx, object, TRUE, NULL);
2379                 error = dmu_tx_assign(tx, TXG_WAIT);
2380                 if (error) {
2381                         ztest_record_enospc("zap parallel");
2382                         dmu_tx_abort(tx);
2383                         return;
2384                 }
2385                 txg = dmu_tx_get_txg(tx);
2386                 bcopy(name, string_value, namelen);
2387         } else {
2388                 tx = NULL;
2389                 txg = 0;
2390                 bzero(string_value, namelen);
2391         }
2392 
2393         switch (i) {
2394 
2395         case 0:
2396                 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
2397                 if (error == 0) {
2398                         ASSERT3U(wsize, ==, zl_wsize);
2399                         ASSERT3U(wc, ==, zl_wc);
2400                 } else {
2401                         ASSERT3U(error, ==, ENOENT);
2402                 }
2403                 break;
2404 
2405         case 1:
2406                 error = zap_lookup(os, object, name, wsize, wc, data);
2407                 if (error == 0) {
2408                         if (data == string_value &&
2409                             bcmp(name, data, namelen) != 0)
2410                                 fatal(0, "name '%s' != val '%s' len %d",
2411                                     name, data, namelen);
2412                 } else {
2413                         ASSERT3U(error, ==, ENOENT);
2414                 }
2415                 break;
2416 
2417         case 2:
2418                 error = zap_add(os, object, name, wsize, wc, data, tx);
2419                 ASSERT(error == 0 || error == EEXIST);
2420                 break;
2421 
2422         case 3:
2423                 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
2424                 break;
2425 
2426         case 4:
2427                 error = zap_remove(os, object, name, tx);
2428                 ASSERT(error == 0 || error == ENOENT);
2429                 break;
2430         }
2431 
2432         if (tx != NULL)
2433                 dmu_tx_commit(tx);
2434 }
2435 
2436 void
2437 ztest_dsl_prop_get_set(ztest_args_t *za)
2438 {
2439         objset_t *os = za->za_os;
2440         int i, inherit;
2441         uint64_t value;
2442         const char *prop, *valname;
2443         char setpoint[MAXPATHLEN];
2444         char osname[MAXNAMELEN];
2445         int error;
2446 
2447         (void) rw_rdlock(&ztest_shared->zs_name_lock);
2448 
2449         dmu_objset_name(os, osname);
2450 
2451         for (i = 0; i < 2; i++) {
2452                 if (i == 0) {
2453                         prop = "checksum";
2454                         value = ztest_random_checksum();
2455                         inherit = (value == ZIO_CHECKSUM_INHERIT);
2456                 } else {
2457                         prop = "compression";
2458                         value = ztest_random_compress();
2459                         inherit = (value == ZIO_COMPRESS_INHERIT);
2460                 }
2461 
2462                 error = dsl_prop_set(osname, prop, sizeof (value),
2463                     !inherit, &value);
2464 
2465                 if (error == ENOSPC) {
2466                         ztest_record_enospc("dsl_prop_set");
2467                         break;
2468                 }
2469 
2470                 ASSERT3U(error, ==, 0);
2471 
2472                 VERIFY3U(dsl_prop_get(osname, prop, sizeof (value),
2473                     1, &value, setpoint), ==, 0);
2474 
2475                 if (i == 0)
2476                         valname = zio_checksum_table[value].ci_name;
2477                 else
2478                         valname = zio_compress_table[value].ci_name;
2479 
2480                 if (zopt_verbose >= 6) {
2481                         (void) printf("%s %s = %s for '%s'\n",
2482                             osname, prop, valname, setpoint);
2483                 }
2484         }
2485 
2486         (void) rw_unlock(&ztest_shared->zs_name_lock);
2487 }
2488 
2489 /*
2490  * Inject random faults into the on-disk data.
2491  */
2492 void
2493 ztest_fault_inject(ztest_args_t *za)
2494 {
2495         int fd;
2496         uint64_t offset;
2497         uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
2498         uint64_t bad = 0x1990c0ffeedecade;
2499         uint64_t top, leaf;
2500         char path0[MAXPATHLEN];
2501         char pathrand[MAXPATHLEN];
2502         size_t fsize;
2503         spa_t *spa = za->za_spa;
2504         int bshift = SPA_MAXBLOCKSHIFT + 2;     /* don't scrog all labels */
2505         int iters = 1000;
2506         int maxfaults = zopt_maxfaults;
2507         vdev_t *vd0 = NULL;
2508         uint64_t guid0 = 0;
2509 
2510         ASSERT(leaves >= 1);
2511 
2512         /*
2513          * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
2514          */
2515         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
2516 
2517         if (ztest_random(2) == 0) {
2518                 /*
2519                  * Inject errors on a normal data device.
2520                  */
2521                 top = ztest_random(spa->spa_root_vdev->vdev_children);
2522                 leaf = ztest_random(leaves);
2523 
2524                 /*
2525                  * Generate paths to the first leaf in this top-level vdev,
2526                  * and to the random leaf we selected.  We'll induce transient
2527                  * write failures and random online/offline activity on leaf 0,
2528                  * and we'll write random garbage to the randomly chosen leaf.
2529                  */
2530                 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
2531                     zopt_dir, zopt_pool, top * leaves + 0);
2532                 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
2533                     zopt_dir, zopt_pool, top * leaves + leaf);
2534 
2535                 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
2536                 if (vd0 != NULL && maxfaults != 1) {
2537                         /*
2538                          * Make vd0 explicitly claim to be unreadable,
2539                          * or unwriteable, or reach behind its back
2540                          * and close the underlying fd.  We can do this if
2541                          * maxfaults == 0 because we'll fail and reexecute,
2542                          * and we can do it if maxfaults >= 2 because we'll
2543                          * have enough redundancy.  If maxfaults == 1, the
2544                          * combination of this with injection of random data
2545                          * corruption below exceeds the pool's fault tolerance.
2546                          */
2547                         vdev_file_t *vf = vd0->vdev_tsd;
2548 
2549                         if (vf != NULL && ztest_random(3) == 0) {
2550                                 (void) close(vf->vf_vnode->v_fd);
2551                                 vf->vf_vnode->v_fd = -1;
2552                         } else if (ztest_random(2) == 0) {
2553                                 vd0->vdev_cant_read = B_TRUE;
2554                         } else {
2555                                 vd0->vdev_cant_write = B_TRUE;
2556                         }
2557                         guid0 = vd0->vdev_guid;
2558                 }
2559         } else {
2560                 /*
2561                  * Inject errors on an l2cache device.
2562                  */
2563                 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2564 
2565                 if (sav->sav_count == 0) {
2566                         spa_config_exit(spa, SCL_STATE, FTAG);
2567                         return;
2568                 }
2569                 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
2570                 guid0 = vd0->vdev_guid;
2571                 (void) strcpy(path0, vd0->vdev_path);
2572                 (void) strcpy(pathrand, vd0->vdev_path);
2573 
2574                 leaf = 0;
2575                 leaves = 1;
2576                 maxfaults = INT_MAX;    /* no limit on cache devices */
2577         }
2578 
2579         spa_config_exit(spa, SCL_STATE, FTAG);
2580 
2581         if (maxfaults == 0)
2582                 return;
2583 
2584         /*
2585          * If we can tolerate two or more faults, randomly online/offline vd0.
2586          */
2587         if (maxfaults >= 2 && guid0 != 0) {
2588                 if (ztest_random(10) < 6) {
2589                         int flags = (ztest_random(2) == 0 ?
2590                             ZFS_OFFLINE_TEMPORARY : 0);
2591                         VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
2592                 } else {
2593                         (void) vdev_online(spa, guid0, 0, NULL);
2594                 }
2595         }
2596 
2597         /*
2598          * We have at least single-fault tolerance, so inject data corruption.
2599          */
2600         fd = open(pathrand, O_RDWR);
2601 
2602         if (fd == -1)   /* we hit a gap in the device namespace */
2603                 return;
2604 
2605         fsize = lseek(fd, 0, SEEK_END);
2606 
2607         while (--iters != 0) {
2608                 offset = ztest_random(fsize / (leaves << bshift)) *
2609                     (leaves << bshift) + (leaf << bshift) +
2610                     (ztest_random(1ULL << (bshift - 1)) & -8ULL);
2611 
2612                 if (offset >= fsize)
2613                         continue;
2614 
2615                 if (zopt_verbose >= 6)
2616                         (void) printf("injecting bad word into %s,"
2617                             " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
2618 
2619                 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
2620                         fatal(1, "can't inject bad word at 0x%llx in %s",
2621                             offset, pathrand);
2622         }
2623 
2624         (void) close(fd);
2625 }
2626 
2627 /*
2628  * Scrub the pool.
2629  */
2630 void
2631 ztest_scrub(ztest_args_t *za)
2632 {
2633         spa_t *spa = za->za_spa;
2634 
2635         (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING);
2636         (void) poll(NULL, 0, 1000); /* wait a second, then force a restart */
2637         (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING);
2638 }
2639 
2640 /*
2641  * Rename the pool to a different name and then rename it back.
2642  */
2643 void
2644 ztest_spa_rename(ztest_args_t *za)
2645 {
2646         char *oldname, *newname;
2647         int error;
2648         spa_t *spa;
2649 
2650         (void) rw_wrlock(&ztest_shared->zs_name_lock);
2651 
2652         oldname = za->za_pool;
2653         newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
2654         (void) strcpy(newname, oldname);
2655         (void) strcat(newname, "_tmp");
2656 
2657         /*
2658          * Do the rename
2659          */
2660         error = spa_rename(oldname, newname);
2661         if (error)
2662                 fatal(0, "spa_rename('%s', '%s') = %d", oldname,
2663                     newname, error);
2664 
2665         /*
2666          * Try to open it under the old name, which shouldn't exist
2667          */
2668         error = spa_open(oldname, &spa, FTAG);
2669         if (error != ENOENT)
2670                 fatal(0, "spa_open('%s') = %d", oldname, error);
2671 
2672         /*
2673          * Open it under the new name and make sure it's still the same spa_t.
2674          */
2675         error = spa_open(newname, &spa, FTAG);
2676         if (error != 0)
2677                 fatal(0, "spa_open('%s') = %d", newname, error);
2678 
2679         ASSERT(spa == za->za_spa);
2680         spa_close(spa, FTAG);
2681 
2682         /*
2683          * Rename it back to the original
2684          */
2685         error = spa_rename(newname, oldname);
2686         if (error)
2687                 fatal(0, "spa_rename('%s', '%s') = %d", newname,
2688                     oldname, error);
2689 
2690         /*
2691          * Make sure it can still be opened
2692          */
2693         error = spa_open(oldname, &spa, FTAG);
2694         if (error != 0)
2695                 fatal(0, "spa_open('%s') = %d", oldname, error);
2696 
2697         ASSERT(spa == za->za_spa);
2698         spa_close(spa, FTAG);
2699 
2700         umem_free(newname, strlen(newname) + 1);
2701 
2702         (void) rw_unlock(&ztest_shared->zs_name_lock);
2703 }
2704 
2705 
2706 /*
2707  * Completely obliterate one disk.
2708  */
2709 static void
2710 ztest_obliterate_one_disk(uint64_t vdev)
2711 {
2712         int fd;
2713         char dev_name[MAXPATHLEN], copy_name[MAXPATHLEN];
2714         size_t fsize;
2715 
2716         if (zopt_maxfaults < 2)
2717                 return;
2718 
2719         (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
2720         (void) snprintf(copy_name, MAXPATHLEN, "%s.old", dev_name);
2721 
2722         fd = open(dev_name, O_RDWR);
2723 
2724         if (fd == -1)
2725                 fatal(1, "can't open %s", dev_name);
2726 
2727         /*
2728          * Determine the size.
2729          */
2730         fsize = lseek(fd, 0, SEEK_END);
2731 
2732         (void) close(fd);
2733 
2734         /*
2735          * Rename the old device to dev_name.old (useful for debugging).
2736          */
2737         VERIFY(rename(dev_name, copy_name) == 0);
2738 
2739         /*
2740          * Create a new one.
2741          */
2742         VERIFY((fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666)) >= 0);
2743         VERIFY(ftruncate(fd, fsize) == 0);
2744         (void) close(fd);
2745 }
2746 
2747 static void
2748 ztest_replace_one_disk(spa_t *spa, uint64_t vdev)
2749 {
2750         char dev_name[MAXPATHLEN];
2751         nvlist_t *root;
2752         int error;
2753         uint64_t guid;
2754         vdev_t *vd;
2755 
2756         (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
2757 
2758         /*
2759          * Build the nvlist describing dev_name.
2760          */
2761         root = make_vdev_root(dev_name, NULL, 0, 0, 0, 0, 0, 1);
2762 
2763         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2764         if ((vd = vdev_lookup_by_path(spa->spa_root_vdev, dev_name)) == NULL)
2765                 guid = 0;
2766         else
2767                 guid = vd->vdev_guid;
2768         spa_config_exit(spa, SCL_VDEV, FTAG);
2769         error = spa_vdev_attach(spa, guid, root, B_TRUE);
2770         if (error != 0 &&
2771             error != EBUSY &&
2772             error != ENOTSUP &&
2773             error != ENODEV &&
2774             error != EDOM)
2775                 fatal(0, "spa_vdev_attach(in-place) = %d", error);
2776 
2777         nvlist_free(root);
2778 }
2779 
2780 static void
2781 ztest_verify_blocks(char *pool)
2782 {
2783         int status;
2784         char zdb[MAXPATHLEN + MAXNAMELEN + 20];
2785         char zbuf[1024];
2786         char *bin;
2787         char *ztest;
2788         char *isa;
2789         int isalen;
2790         FILE *fp;
2791 
2792         (void) realpath(getexecname(), zdb);
2793 
2794         /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
2795         bin = strstr(zdb, "/usr/bin/");
2796         ztest = strstr(bin, "/ztest");
2797         isa = bin + 8;
2798         isalen = ztest - isa;
2799         isa = strdup(isa);
2800         /* LINTED */
2801         (void) sprintf(bin,
2802             "/usr/sbin%.*s/zdb -bc%s%s -U /tmp/zpool.cache %s",
2803             isalen,
2804             isa,
2805             zopt_verbose >= 3 ? "s" : "",
2806             zopt_verbose >= 4 ? "v" : "",
2807             pool);
2808         free(isa);
2809 
2810         if (zopt_verbose >= 5)
2811                 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
2812 
2813         fp = popen(zdb, "r");
2814 
2815         while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
2816                 if (zopt_verbose >= 3)
2817                         (void) printf("%s", zbuf);
2818 
2819         status = pclose(fp);
2820 
2821         if (status == 0)
2822                 return;
2823 
2824         ztest_dump_core = 0;
2825         if (WIFEXITED(status))
2826                 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
2827         else
2828                 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
2829 }
2830 
2831 static void
2832 ztest_walk_pool_directory(char *header)
2833 {
2834         spa_t *spa = NULL;
2835 
2836         if (zopt_verbose >= 6)
2837                 (void) printf("%s\n", header);
2838 
2839         mutex_enter(&spa_namespace_lock);
2840         while ((spa = spa_next(spa)) != NULL)
2841                 if (zopt_verbose >= 6)
2842                         (void) printf("\t%s\n", spa_name(spa));
2843         mutex_exit(&spa_namespace_lock);
2844 }
2845 
2846 static void
2847 ztest_spa_import_export(char *oldname, char *newname)
2848 {
2849         nvlist_t *config, *newconfig;
2850         uint64_t pool_guid;
2851         spa_t *spa;
2852         int error;
2853 
2854         if (zopt_verbose >= 4) {
2855                 (void) printf("import/export: old = %s, new = %s\n",
2856                     oldname, newname);
2857         }
2858 
2859         /*
2860          * Clean up from previous runs.
2861          */
2862         (void) spa_destroy(newname);
2863 
2864         /*
2865          * Get the pool's configuration and guid.
2866          */
2867         error = spa_open(oldname, &spa, FTAG);
2868         if (error)
2869                 fatal(0, "spa_open('%s') = %d", oldname, error);
2870 
2871         /*
2872          * Kick off a scrub to tickle scrub/export races.
2873          */
2874         if (ztest_random(2) == 0)
2875                 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING);
2876 
2877         pool_guid = spa_guid(spa);
2878         spa_close(spa, FTAG);
2879 
2880         ztest_walk_pool_directory("pools before export");
2881 
2882         /*
2883          * Export it.
2884          */
2885         error = spa_export(oldname, &config, B_FALSE, B_FALSE);
2886         if (error)
2887                 fatal(0, "spa_export('%s') = %d", oldname, error);
2888 
2889         ztest_walk_pool_directory("pools after export");
2890 
2891         /*
2892          * Try to import it.
2893          */
2894         newconfig = spa_tryimport(config);
2895         ASSERT(newconfig != NULL);
2896         nvlist_free(newconfig);
2897 
2898         /*
2899          * Import it under the new name.
2900          */
2901         error = spa_import(newname, config, NULL);
2902         if (error)
2903                 fatal(0, "spa_import('%s') = %d", newname, error);
2904 
2905         ztest_walk_pool_directory("pools after import");
2906 
2907         /*
2908          * Try to import it again -- should fail with EEXIST.
2909          */
2910         error = spa_import(newname, config, NULL);
2911         if (error != EEXIST)
2912                 fatal(0, "spa_import('%s') twice", newname);
2913 
2914         /*
2915          * Try to import it under a different name -- should fail with EEXIST.
2916          */
2917         error = spa_import(oldname, config, NULL);
2918         if (error != EEXIST)
2919                 fatal(0, "spa_import('%s') under multiple names", newname);
2920 
2921         /*
2922          * Verify that the pool is no longer visible under the old name.
2923          */
2924         error = spa_open(oldname, &spa, FTAG);
2925         if (error != ENOENT)
2926                 fatal(0, "spa_open('%s') = %d", newname, error);
2927 
2928         /*
2929          * Verify that we can open and close the pool using the new name.
2930          */
2931         error = spa_open(newname, &spa, FTAG);
2932         if (error)
2933                 fatal(0, "spa_open('%s') = %d", newname, error);
2934         ASSERT(pool_guid == spa_guid(spa));
2935         spa_close(spa, FTAG);
2936 
2937         nvlist_free(config);
2938 }
2939 
2940 static void
2941 ztest_resume(spa_t *spa)
2942 {
2943         if (spa_suspended(spa)) {
2944                 spa_vdev_state_enter(spa);
2945                 vdev_clear(spa, NULL);
2946                 (void) spa_vdev_state_exit(spa, NULL, 0);
2947                 zio_resume(spa);
2948         }
2949 }
2950 
2951 static void *
2952 ztest_resume_thread(void *arg)
2953 {
2954         spa_t *spa = arg;
2955 
2956         while (!ztest_exiting) {
2957                 (void) poll(NULL, 0, 1000);
2958                 ztest_resume(spa);
2959         }
2960         return (NULL);
2961 }
2962 
2963 static void *
2964 ztest_thread(void *arg)
2965 {
2966         ztest_args_t *za = arg;
2967         ztest_shared_t *zs = ztest_shared;
2968         hrtime_t now, functime;
2969         ztest_info_t *zi;
2970         int f, i;
2971 
2972         while ((now = gethrtime()) < za->za_stop) {
2973                 /*
2974                  * See if it's time to force a crash.
2975                  */
2976                 if (now > za->za_kill) {
2977                         zs->zs_alloc = spa_get_alloc(za->za_spa);
2978                         zs->zs_space = spa_get_space(za->za_spa);
2979                         (void) kill(getpid(), SIGKILL);
2980                 }
2981 
2982                 /*
2983                  * Pick a random function.
2984                  */
2985                 f = ztest_random(ZTEST_FUNCS);
2986                 zi = &zs->zs_info[f];
2987 
2988                 /*
2989                  * Decide whether to call it, based on the requested frequency.
2990                  */
2991                 if (zi->zi_call_target == 0 ||
2992                     (double)zi->zi_call_total / zi->zi_call_target >
2993                     (double)(now - zs->zs_start_time) / (zopt_time * NANOSEC))
2994                         continue;
2995 
2996                 atomic_add_64(&zi->zi_calls, 1);
2997                 atomic_add_64(&zi->zi_call_total, 1);
2998 
2999                 za->za_diroff = (za->za_instance * ZTEST_FUNCS + f) *
3000                     ZTEST_DIRSIZE;
3001                 za->za_diroff_shared = (1ULL << 63);
3002 
3003                 for (i = 0; i < zi->zi_iters; i++)
3004                         zi->zi_func(za);
3005 
3006                 functime = gethrtime() - now;
3007 
3008                 atomic_add_64(&zi->zi_call_time, functime);
3009 
3010                 if (zopt_verbose >= 4) {
3011                         Dl_info dli;
3012                         (void) dladdr((void *)zi->zi_func, &dli);
3013                         (void) printf("%6.2f sec in %s\n",
3014                             (double)functime / NANOSEC, dli.dli_sname);
3015                 }
3016 
3017                 /*
3018                  * If we're getting ENOSPC with some regularity, stop.
3019                  */
3020                 if (zs->zs_enospc_count > 10)
3021                         break;
3022         }
3023 
3024         return (NULL);
3025 }
3026 
3027 /*
3028  * Kick off threads to run tests on all datasets in parallel.
3029  */
3030 static void
3031 ztest_run(char *pool)
3032 {
3033         int t, d, error;
3034         ztest_shared_t *zs = ztest_shared;
3035         ztest_args_t *za;
3036         spa_t *spa;
3037         char name[100];
3038         thread_t resume_tid;
3039 
3040         ztest_exiting = B_FALSE;
3041 
3042         (void) _mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL);
3043         (void) rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL);
3044 
3045         for (t = 0; t < ZTEST_SYNC_LOCKS; t++)
3046                 (void) _mutex_init(&zs->zs_sync_lock[t], USYNC_THREAD, NULL);
3047 
3048         /*
3049          * Destroy one disk before we even start.
3050          * It's mirrored, so everything should work just fine.
3051          * This makes us exercise fault handling very early in spa_load().
3052          */
3053         ztest_obliterate_one_disk(0);
3054 
3055         /*
3056          * Verify that the sum of the sizes of all blocks in the pool
3057          * equals the SPA's allocated space total.
3058          */
3059         ztest_verify_blocks(pool);
3060 
3061         /*
3062          * Kick off a replacement of the disk we just obliterated.
3063          */
3064         kernel_init(FREAD | FWRITE);
3065         VERIFY(spa_open(pool, &spa, FTAG) == 0);
3066         ztest_replace_one_disk(spa, 0);
3067         if (zopt_verbose >= 5)
3068                 show_pool_stats(spa);
3069         spa_close(spa, FTAG);
3070         kernel_fini();
3071 
3072         kernel_init(FREAD | FWRITE);
3073 
3074         /*
3075          * Verify that we can export the pool and reimport it under a
3076          * different name.
3077          */
3078         if (ztest_random(2) == 0) {
3079                 (void) snprintf(name, 100, "%s_import", pool);
3080                 ztest_spa_import_export(pool, name);
3081                 ztest_spa_import_export(name, pool);
3082         }
3083 
3084         /*
3085          * Verify that we can loop over all pools.
3086          */
3087         mutex_enter(&spa_namespace_lock);
3088         for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) {
3089                 if (zopt_verbose > 3) {
3090                         (void) printf("spa_next: found %s\n", spa_name(spa));
3091                 }
3092         }
3093         mutex_exit(&spa_namespace_lock);
3094 
3095         /*
3096          * Open our pool.
3097          */
3098         VERIFY(spa_open(pool, &spa, FTAG) == 0);
3099 
3100         /*
3101          * We don't expect the pool to suspend unless maxfaults == 0,
3102          * in which case ztest_fault_inject() temporarily takes away
3103          * the only valid replica.
3104          */
3105         if (zopt_maxfaults == 0)
3106                 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
3107         else
3108                 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
3109 
3110         /*
3111          * Create a thread to periodically resume suspended I/O.
3112          */
3113         VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
3114             &resume_tid) == 0);
3115 
3116         /*
3117          * Verify that we can safely inquire about about any object,
3118          * whether it's allocated or not.  To make it interesting,
3119          * we probe a 5-wide window around each power of two.
3120          * This hits all edge cases, including zero and the max.
3121          */
3122         for (t = 0; t < 64; t++) {
3123                 for (d = -5; d <= 5; d++) {
3124                         error = dmu_object_info(spa->spa_meta_objset,
3125                             (1ULL << t) + d, NULL);
3126                         ASSERT(error == 0 || error == ENOENT ||
3127                             error == EINVAL);
3128                 }
3129         }
3130 
3131         /*
3132          * Now kick off all the tests that run in parallel.
3133          */
3134         zs->zs_enospc_count = 0;
3135 
3136         za = umem_zalloc(zopt_threads * sizeof (ztest_args_t), UMEM_NOFAIL);
3137 
3138         if (zopt_verbose >= 4)
3139                 (void) printf("starting main threads...\n");
3140 
3141         za[0].za_start = gethrtime();
3142         za[0].za_stop = za[0].za_start + zopt_passtime * NANOSEC;
3143         za[0].za_stop = MIN(za[0].za_stop, zs->zs_stop_time);
3144         za[0].za_kill = za[0].za_stop;
3145         if (ztest_random(100) < zopt_killrate)
3146                 za[0].za_kill -= ztest_random(zopt_passtime * NANOSEC);
3147 
3148         for (t = 0; t < zopt_threads; t++) {
3149                 d = t % zopt_datasets;
3150 
3151                 (void) strcpy(za[t].za_pool, pool);
3152                 za[t].za_os = za[d].za_os;
3153                 za[t].za_spa = spa;
3154                 za[t].za_zilog = za[d].za_zilog;
3155                 za[t].za_instance = t;
3156                 za[t].za_random = ztest_random(-1ULL);
3157                 za[t].za_start = za[0].za_start;
3158                 za[t].za_stop = za[0].za_stop;
3159                 za[t].za_kill = za[0].za_kill;
3160 
3161                 if (t < zopt_datasets) {
3162                         int test_future = FALSE;
3163                         (void) rw_rdlock(&ztest_shared->zs_name_lock);
3164                         (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d);
3165                         error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0,
3166                             ztest_create_cb, NULL);
3167                         if (error == EEXIST) {
3168                                 test_future = TRUE;
3169                         } else if (error == ENOSPC) {
3170                                 zs->zs_enospc_count++;
3171                                 (void) rw_unlock(&ztest_shared->zs_name_lock);
3172                                 break;
3173                         } else if (error != 0) {
3174                                 fatal(0, "dmu_objset_create(%s) = %d",
3175                                     name, error);
3176                         }
3177                         error = dmu_objset_open(name, DMU_OST_OTHER,
3178                             DS_MODE_USER, &za[d].za_os);
3179                         if (error)
3180                                 fatal(0, "dmu_objset_open('%s') = %d",
3181                                     name, error);
3182                         (void) rw_unlock(&ztest_shared->zs_name_lock);
3183                         if (test_future)
3184                                 ztest_dmu_check_future_leak(&za[t]);
3185                         zil_replay(za[d].za_os, za[d].za_os,
3186                             ztest_replay_vector);
3187                         za[d].za_zilog = zil_open(za[d].za_os, NULL);
3188                 }
3189 
3190                 VERIFY(thr_create(0, 0, ztest_thread, &za[t], THR_BOUND,
3191                     &za[t].za_thread) == 0);
3192         }
3193 
3194         while (--t >= 0) {
3195                 VERIFY(thr_join(za[t].za_thread, NULL, NULL) == 0);
3196                 if (t < zopt_datasets) {
3197                         zil_close(za[t].za_zilog);
3198                         dmu_objset_close(za[t].za_os);
3199                 }
3200         }
3201 
3202         if (zopt_verbose >= 3)
3203                 show_pool_stats(spa);
3204 
3205         txg_wait_synced(spa_get_dsl(spa), 0);
3206 
3207         zs->zs_alloc = spa_get_alloc(spa);
3208         zs->zs_space = spa_get_space(spa);
3209 
3210         /*
3211          * If we had out-of-space errors, destroy a random objset.
3212          */
3213         if (zs->zs_enospc_count != 0) {
3214                 (void) rw_rdlock(&ztest_shared->zs_name_lock);
3215                 d = (int)ztest_random(zopt_datasets);
3216                 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d);
3217                 if (zopt_verbose >= 3)
3218                         (void) printf("Destroying %s to free up space\n", name);
3219                 (void) dmu_objset_find(name, ztest_destroy_cb, &za[d],
3220                     DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
3221                 (void) rw_unlock(&ztest_shared->zs_name_lock);
3222         }
3223 
3224         txg_wait_synced(spa_get_dsl(spa), 0);
3225 
3226         umem_free(za, zopt_threads * sizeof (ztest_args_t));
3227 
3228         /* Kill the resume thread */
3229         ztest_exiting = B_TRUE;
3230         VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
3231         ztest_resume(spa);
3232 
3233         /*
3234          * Right before closing the pool, kick off a bunch of async I/O;
3235          * spa_close() should wait for it to complete.
3236          */
3237         for (t = 1; t < 50; t++)
3238                 dmu_prefetch(spa->spa_meta_objset, t, 0, 1 << 15);
3239 
3240         spa_close(spa, FTAG);
3241 
3242         kernel_fini();
3243 }
3244 
3245 void
3246 print_time(hrtime_t t, char *timebuf)
3247 {
3248         hrtime_t s = t / NANOSEC;
3249         hrtime_t m = s / 60;
3250         hrtime_t h = m / 60;
3251         hrtime_t d = h / 24;
3252 
3253         s -= m * 60;
3254         m -= h * 60;
3255         h -= d * 24;
3256 
3257         timebuf[0] = '\0';
3258 
3259         if (d)
3260                 (void) sprintf(timebuf,
3261                     "%llud%02lluh%02llum%02llus", d, h, m, s);
3262         else if (h)
3263                 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
3264         else if (m)
3265                 (void) sprintf(timebuf, "%llum%02llus", m, s);
3266         else
3267                 (void) sprintf(timebuf, "%llus", s);
3268 }
3269 
3270 /*
3271  * Create a storage pool with the given name and initial vdev size.
3272  * Then create the specified number of datasets in the pool.
3273  */
3274 static void
3275 ztest_init(char *pool)
3276 {
3277         spa_t *spa;
3278         int error;
3279         nvlist_t *nvroot;
3280 
3281         kernel_init(FREAD | FWRITE);
3282 
3283         /*
3284          * Create the storage pool.
3285          */
3286         (void) spa_destroy(pool);
3287         ztest_shared->zs_vdev_primaries = 0;
3288         nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0,
3289             0, zopt_raidz, zopt_mirrors, 1);
3290         error = spa_create(pool, nvroot, NULL, NULL, NULL);
3291         nvlist_free(nvroot);
3292 
3293         if (error)
3294                 fatal(0, "spa_create() = %d", error);
3295         error = spa_open(pool, &spa, FTAG);
3296         if (error)
3297                 fatal(0, "spa_open() = %d", error);
3298 
3299         if (zopt_verbose >= 3)
3300                 show_pool_stats(spa);
3301 
3302         spa_close(spa, FTAG);
3303 
3304         kernel_fini();
3305 }
3306 
3307 int
3308 main(int argc, char **argv)
3309 {
3310         int kills = 0;
3311         int iters = 0;
3312         int i, f;
3313         ztest_shared_t *zs;
3314         ztest_info_t *zi;
3315         char timebuf[100];
3316         char numbuf[6];
3317 
3318         (void) setvbuf(stdout, NULL, _IOLBF, 0);
3319 
3320         /* Override location of zpool.cache */
3321         spa_config_path = "/tmp/zpool.cache";
3322 
3323         ztest_random_fd = open("/dev/urandom", O_RDONLY);
3324 
3325         process_options(argc, argv);
3326 
3327         /*
3328          * Blow away any existing copy of zpool.cache
3329          */
3330         if (zopt_init != 0)
3331                 (void) remove("/tmp/zpool.cache");
3332 
3333         zs = ztest_shared = (void *)mmap(0,
3334             P2ROUNDUP(sizeof (ztest_shared_t), getpagesize()),
3335             PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
3336 
3337         if (zopt_verbose >= 1) {
3338                 (void) printf("%llu vdevs, %d datasets, %d threads,"
3339                     " %llu seconds...\n",
3340                     (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads,
3341                     (u_longlong_t)zopt_time);
3342         }
3343 
3344         /*
3345          * Create and initialize our storage pool.
3346          */
3347         for (i = 1; i <= zopt_init; i++) {
3348                 bzero(zs, sizeof (ztest_shared_t));
3349                 if (zopt_verbose >= 3 && zopt_init != 1)
3350                         (void) printf("ztest_init(), pass %d\n", i);
3351                 ztest_init(zopt_pool);
3352         }
3353 
3354         /*
3355          * Initialize the call targets for each function.
3356          */
3357         for (f = 0; f < ZTEST_FUNCS; f++) {
3358                 zi = &zs->zs_info[f];
3359 
3360                 *zi = ztest_info[f];
3361 
3362                 if (*zi->zi_interval == 0)
3363                         zi->zi_call_target = UINT64_MAX;
3364                 else
3365                         zi->zi_call_target = zopt_time / *zi->zi_interval;
3366         }
3367 
3368         zs->zs_start_time = gethrtime();
3369         zs->zs_stop_time = zs->zs_start_time + zopt_time * NANOSEC;
3370 
3371         /*
3372          * Run the tests in a loop.  These tests include fault injection
3373          * to verify that self-healing data works, and forced crashes
3374          * to verify that we never lose on-disk consistency.
3375          */
3376         while (gethrtime() < zs->zs_stop_time) {
3377                 int status;
3378                 pid_t pid;
3379                 char *tmp;
3380 
3381                 /*
3382                  * Initialize the workload counters for each function.
3383                  */
3384                 for (f = 0; f < ZTEST_FUNCS; f++) {
3385                         zi = &zs->zs_info[f];
3386                         zi->zi_calls = 0;
3387                         zi->zi_call_time = 0;
3388                 }
3389 
3390                 pid = fork();
3391 
3392                 if (pid == -1)
3393                         fatal(1, "fork failed");
3394 
3395                 if (pid == 0) { /* child */
3396                         struct rlimit rl = { 1024, 1024 };
3397                         (void) setrlimit(RLIMIT_NOFILE, &rl);
3398                         (void) enable_extended_FILE_stdio(-1, -1);
3399                         ztest_run(zopt_pool);
3400                         exit(0);
3401                 }
3402 
3403                 while (waitpid(pid, &status, 0) != pid)
3404                         continue;
3405 
3406                 if (WIFEXITED(status)) {
3407                         if (WEXITSTATUS(status) != 0) {
3408                                 (void) fprintf(stderr,
3409                                     "child exited with code %d\n",
3410                                     WEXITSTATUS(status));
3411                                 exit(2);
3412                         }
3413                 } else if (WIFSIGNALED(status)) {
3414                         if (WTERMSIG(status) != SIGKILL) {
3415                                 (void) fprintf(stderr,
3416                                     "child died with signal %d\n",
3417                                     WTERMSIG(status));
3418                                 exit(3);
3419                         }
3420                         kills++;
3421                 } else {
3422                         (void) fprintf(stderr, "something strange happened "
3423                             "to child\n");
3424                         exit(4);
3425                 }
3426 
3427                 iters++;
3428 
3429                 if (zopt_verbose >= 1) {
3430                         hrtime_t now = gethrtime();
3431 
3432                         now = MIN(now, zs->zs_stop_time);
3433                         print_time(zs->zs_stop_time - now, timebuf);
3434                         nicenum(zs->zs_space, numbuf);
3435 
3436                         (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
3437                             "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
3438                             iters,
3439                             WIFEXITED(status) ? "Complete" : "SIGKILL",
3440                             (u_longlong_t)zs->zs_enospc_count,
3441                             100.0 * zs->zs_alloc / zs->zs_space,
3442                             numbuf,
3443                             100.0 * (now - zs->zs_start_time) /
3444                             (zopt_time * NANOSEC), timebuf);
3445                 }
3446 
3447                 if (zopt_verbose >= 2) {
3448                         (void) printf("\nWorkload summary:\n\n");
3449                         (void) printf("%7s %9s   %s\n",
3450                             "Calls", "Time", "Function");
3451                         (void) printf("%7s %9s   %s\n",
3452                             "-----", "----", "--------");
3453                         for (f = 0; f < ZTEST_FUNCS; f++) {
3454                                 Dl_info dli;
3455 
3456                                 zi = &zs->zs_info[f];
3457                                 print_time(zi->zi_call_time, timebuf);
3458                                 (void) dladdr((void *)zi->zi_func, &dli);
3459                                 (void) printf("%7llu %9s   %s\n",
3460                                     (u_longlong_t)zi->zi_calls, timebuf,
3461                                     dli.dli_sname);
3462                         }
3463                         (void) printf("\n");
3464                 }
3465 
3466                 /*
3467                  * It's possible that we killed a child during a rename test, in
3468                  * which case we'll have a 'ztest_tmp' pool lying around instead
3469                  * of 'ztest'.  Do a blind rename in case this happened.
3470                  */
3471                 tmp = umem_alloc(strlen(zopt_pool) + 5, UMEM_NOFAIL);
3472                 (void) strcpy(tmp, zopt_pool);
3473                 (void) strcat(tmp, "_tmp");
3474                 kernel_init(FREAD | FWRITE);
3475                 (void) spa_rename(tmp, zopt_pool);
3476                 kernel_fini();
3477                 umem_free(tmp, strlen(tmp) + 1);
3478         }
3479 
3480         ztest_verify_blocks(zopt_pool);
3481 
3482         if (zopt_verbose >= 1) {
3483                 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
3484                     kills, iters - kills, (100.0 * kills) / MAX(1, iters));
3485         }
3486 
3487         return (0);
3488 }