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/zio_crypt.h>
  91 #include <sys/zil.h>
  92 #include <sys/vdev_impl.h>
  93 #include <sys/vdev_file.h>
  94 #include <sys/spa_impl.h>
  95 #include <sys/dsl_prop.h>
  96 #include <sys/refcount.h>
  97 #include <stdio.h>
  98 #include <stdio_ext.h>
  99 #include <stdlib.h>
 100 #include <unistd.h>
 101 #include <signal.h>
 102 #include <umem.h>
 103 #include <dlfcn.h>
 104 #include <ctype.h>
 105 #include <math.h>
 106 #include <sys/fs/zfs.h>
 107 
 108 static char cmdname[] = "ztest";
 109 static char *zopt_pool = cmdname;
 110 
 111 static uint64_t zopt_vdevs = 5;
 112 static uint64_t zopt_vdevtime;
 113 static int zopt_ashift = SPA_MINBLOCKSHIFT;
 114 static int zopt_mirrors = 2;
 115 static int zopt_raidz = 4;
 116 static int zopt_raidz_parity = 1;
 117 static size_t zopt_vdev_size = SPA_MINDEVSIZE;
 118 static int zopt_datasets = 7;
 119 static int zopt_threads = 23;
 120 static uint64_t zopt_passtime = 60;     /* 60 seconds */
 121 static uint64_t zopt_killrate = 70;     /* 70% kill rate */
 122 static int zopt_verbose = 0;
 123 static int zopt_init = 1;
 124 static char *zopt_dir = "/tmp";
 125 static uint64_t zopt_time = 300;        /* 5 minutes */
 126 static int zopt_maxfaults;
 127 
 128 typedef struct ztest_block_tag {
 129         uint64_t        bt_objset;
 130         uint64_t        bt_object;
 131         uint64_t        bt_offset;
 132         uint64_t        bt_txg;
 133         uint64_t        bt_thread;
 134         uint64_t        bt_seq;
 135 } ztest_block_tag_t;
 136 
 137 typedef struct ztest_args {
 138         char            za_pool[MAXNAMELEN];
 139         spa_t           *za_spa;
 140         objset_t        *za_os;
 141         zilog_t         *za_zilog;
 142         thread_t        za_thread;
 143         uint64_t        za_instance;
 144         uint64_t        za_random;
 145         uint64_t        za_diroff;
 146         uint64_t        za_diroff_shared;
 147         uint64_t        za_zil_seq;
 148         hrtime_t        za_start;
 149         hrtime_t        za_stop;
 150         hrtime_t        za_kill;
 151         /*
 152          * Thread-local variables can go here to aid debugging.
 153          */
 154         ztest_block_tag_t za_rbt;
 155         ztest_block_tag_t za_wbt;
 156         dmu_object_info_t za_doi;
 157         dmu_buf_t       *za_dbuf;
 158 } ztest_args_t;
 159 
 160 typedef void ztest_func_t(ztest_args_t *);
 161 
 162 /*
 163  * Note: these aren't static because we want dladdr() to work.
 164  */
 165 ztest_func_t ztest_dmu_read_write;
 166 ztest_func_t ztest_dmu_write_parallel;
 167 ztest_func_t ztest_dmu_object_alloc_free;
 168 ztest_func_t ztest_zap;
 169 ztest_func_t ztest_zap_parallel;
 170 ztest_func_t ztest_traverse;
 171 ztest_func_t ztest_dsl_prop_get_set;
 172 ztest_func_t ztest_dmu_objset_create_destroy;
 173 ztest_func_t ztest_dmu_snapshot_create_destroy;
 174 ztest_func_t ztest_spa_create_destroy;
 175 ztest_func_t ztest_fault_inject;
 176 ztest_func_t ztest_spa_rename;
 177 ztest_func_t ztest_vdev_attach_detach;
 178 ztest_func_t ztest_vdev_LUN_growth;
 179 ztest_func_t ztest_vdev_add_remove;
 180 ztest_func_t ztest_vdev_aux_add_remove;
 181 ztest_func_t ztest_scrub;
 182 
 183 typedef struct ztest_info {
 184         ztest_func_t    *zi_func;       /* test function */
 185         uint64_t        zi_iters;       /* iterations per execution */
 186         uint64_t        *zi_interval;   /* execute every <interval> seconds */
 187         uint64_t        zi_calls;       /* per-pass count */
 188         uint64_t        zi_call_time;   /* per-pass time */
 189         uint64_t        zi_call_total;  /* cumulative total */
 190         uint64_t        zi_call_target; /* target cumulative total */
 191 } ztest_info_t;
 192 
 193 uint64_t zopt_always = 0;               /* all the time */
 194 uint64_t zopt_often = 1;                /* every second */
 195 uint64_t zopt_sometimes = 10;           /* every 10 seconds */
 196 uint64_t zopt_rarely = 60;              /* every 60 seconds */
 197 
 198 ztest_info_t ztest_info[] = {
 199         { ztest_dmu_read_write,                 1,      &zopt_always        },
 200         { ztest_dmu_write_parallel,             30,     &zopt_always        },
 201         { ztest_dmu_object_alloc_free,          1,      &zopt_always        },
 202         { ztest_zap,                            30,     &zopt_always        },
 203         { ztest_zap_parallel,                   100,    &zopt_always        },
 204         { ztest_dsl_prop_get_set,               1,      &zopt_sometimes     },
 205         { ztest_dmu_objset_create_destroy,      1,      &zopt_sometimes },
 206         { ztest_dmu_snapshot_create_destroy,    1,      &zopt_sometimes },
 207         { ztest_spa_create_destroy,             1,      &zopt_sometimes },
 208         { ztest_fault_inject,                   1,      &zopt_sometimes     },
 209         { ztest_spa_rename,                     1,      &zopt_rarely        },
 210         { ztest_vdev_attach_detach,             1,      &zopt_rarely        },
 211         { ztest_vdev_LUN_growth,                1,      &zopt_rarely        },
 212         { ztest_vdev_add_remove,                1,      &zopt_vdevtime      },
 213         { ztest_vdev_aux_add_remove,            1,      &zopt_vdevtime      },
 214         { ztest_scrub,                          1,      &zopt_vdevtime      },
 215 };
 216 
 217 #define ZTEST_FUNCS     (sizeof (ztest_info) / sizeof (ztest_info_t))
 218 
 219 #define ZTEST_SYNC_LOCKS        16
 220 
 221 /*
 222  * Stuff we need to share writably between parent and child.
 223  */
 224 typedef struct ztest_shared {
 225         mutex_t         zs_vdev_lock;
 226         rwlock_t        zs_name_lock;
 227         uint64_t        zs_vdev_primaries;
 228         uint64_t        zs_vdev_aux;
 229         uint64_t        zs_enospc_count;
 230         hrtime_t        zs_start_time;
 231         hrtime_t        zs_stop_time;
 232         uint64_t        zs_alloc;
 233         uint64_t        zs_space;
 234         ztest_info_t    zs_info[ZTEST_FUNCS];
 235         mutex_t         zs_sync_lock[ZTEST_SYNC_LOCKS];
 236         uint64_t        zs_seq[ZTEST_SYNC_LOCKS];
 237 } ztest_shared_t;
 238 
 239 static char ztest_dev_template[] = "%s/%s.%llua";
 240 static char ztest_aux_template[] = "%s/%s.%s.%llu";
 241 static ztest_shared_t *ztest_shared;
 242 
 243 static int ztest_random_fd;
 244 static int ztest_dump_core = 1;
 245 
 246 static boolean_t ztest_exiting;
 247 
 248 extern uint64_t metaslab_gang_bang;
 249 
 250 #define ZTEST_DIROBJ            1
 251 #define ZTEST_MICROZAP_OBJ      2
 252 #define ZTEST_FATZAP_OBJ        3
 253 
 254 #define ZTEST_DIROBJ_BLOCKSIZE  (1 << 10)
 255 #define ZTEST_DIRSIZE           256
 256 
 257 static void usage(boolean_t) __NORETURN;
 258 
 259 /*
 260  * These libumem hooks provide a reasonable set of defaults for the allocator's
 261  * debugging facilities.
 262  */
 263 const char *
 264 _umem_debug_init()
 265 {
 266         return ("default,verbose"); /* $UMEM_DEBUG setting */
 267 }
 268 
 269 const char *
 270 _umem_logging_init(void)
 271 {
 272         return ("fail,contents"); /* $UMEM_LOGGING setting */
 273 }
 274 
 275 #define FATAL_MSG_SZ    1024
 276 
 277 char *fatal_msg;
 278 
 279 static void
 280 fatal(int do_perror, char *message, ...)
 281 {
 282         va_list args;
 283         int save_errno = errno;
 284         char buf[FATAL_MSG_SZ];
 285 
 286         (void) fflush(stdout);
 287 
 288         va_start(args, message);
 289         (void) sprintf(buf, "ztest: ");
 290         /* LINTED */
 291         (void) vsprintf(buf + strlen(buf), message, args);
 292         va_end(args);
 293         if (do_perror) {
 294                 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
 295                     ": %s", strerror(save_errno));
 296         }
 297         (void) fprintf(stderr, "%s\n", buf);
 298         fatal_msg = buf;                        /* to ease debugging */
 299         if (ztest_dump_core)
 300                 abort();
 301         exit(3);
 302 }
 303 
 304 static int
 305 str2shift(const char *buf)
 306 {
 307         const char *ends = "BKMGTPEZ";
 308         int i;
 309 
 310         if (buf[0] == '\0')
 311                 return (0);
 312         for (i = 0; i < strlen(ends); i++) {
 313                 if (toupper(buf[0]) == ends[i])
 314                         break;
 315         }
 316         if (i == strlen(ends)) {
 317                 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
 318                     buf);
 319                 usage(B_FALSE);
 320         }
 321         if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
 322                 return (10*i);
 323         }
 324         (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
 325         usage(B_FALSE);
 326         /* NOTREACHED */
 327 }
 328 
 329 static uint64_t
 330 nicenumtoull(const char *buf)
 331 {
 332         char *end;
 333         uint64_t val;
 334 
 335         val = strtoull(buf, &end, 0);
 336         if (end == buf) {
 337                 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
 338                 usage(B_FALSE);
 339         } else if (end[0] == '.') {
 340                 double fval = strtod(buf, &end);
 341                 fval *= pow(2, str2shift(end));
 342                 if (fval > UINT64_MAX) {
 343                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 344                             buf);
 345                         usage(B_FALSE);
 346                 }
 347                 val = (uint64_t)fval;
 348         } else {
 349                 int shift = str2shift(end);
 350                 if (shift >= 64 || (val << shift) >> shift != val) {
 351                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 352                             buf);
 353                         usage(B_FALSE);
 354                 }
 355                 val <<= shift;
 356         }
 357         return (val);
 358 }
 359 
 360 static void
 361 usage(boolean_t requested)
 362 {
 363         char nice_vdev_size[10];
 364         char nice_gang_bang[10];
 365         FILE *fp = requested ? stdout : stderr;
 366 
 367         nicenum(zopt_vdev_size, nice_vdev_size);
 368         nicenum(metaslab_gang_bang, nice_gang_bang);
 369 
 370         (void) fprintf(fp, "Usage: %s\n"
 371             "\t[-v vdevs (default: %llu)]\n"
 372             "\t[-s size_of_each_vdev (default: %s)]\n"
 373             "\t[-a alignment_shift (default: %d) (use 0 for random)]\n"
 374             "\t[-m mirror_copies (default: %d)]\n"
 375             "\t[-r raidz_disks (default: %d)]\n"
 376             "\t[-R raidz_parity (default: %d)]\n"
 377             "\t[-d datasets (default: %d)]\n"
 378             "\t[-t threads (default: %d)]\n"
 379             "\t[-g gang_block_threshold (default: %s)]\n"
 380             "\t[-i initialize pool i times (default: %d)]\n"
 381             "\t[-k kill percentage (default: %llu%%)]\n"
 382             "\t[-p pool_name (default: %s)]\n"
 383             "\t[-f file directory for vdev files (default: %s)]\n"
 384             "\t[-V(erbose)] (use multiple times for ever more blather)\n"
 385             "\t[-E(xisting)] (use existing pool instead of creating new one)\n"
 386             "\t[-T time] total run time (default: %llu sec)\n"
 387             "\t[-P passtime] time per pass (default: %llu sec)\n"
 388             "\t[-h] (print help)\n"
 389             "",
 390             cmdname,
 391             (u_longlong_t)zopt_vdevs,                   /* -v */
 392             nice_vdev_size,                             /* -s */
 393             zopt_ashift,                                /* -a */
 394             zopt_mirrors,                               /* -m */
 395             zopt_raidz,                                 /* -r */
 396             zopt_raidz_parity,                          /* -R */
 397             zopt_datasets,                              /* -d */
 398             zopt_threads,                               /* -t */
 399             nice_gang_bang,                             /* -g */
 400             zopt_init,                                  /* -i */
 401             (u_longlong_t)zopt_killrate,                /* -k */
 402             zopt_pool,                                  /* -p */
 403             zopt_dir,                                   /* -f */
 404             (u_longlong_t)zopt_time,                    /* -T */
 405             (u_longlong_t)zopt_passtime);               /* -P */
 406         exit(requested ? 0 : 1);
 407 }
 408 
 409 static uint64_t
 410 ztest_random(uint64_t range)
 411 {
 412         uint64_t r;
 413 
 414         if (range == 0)
 415                 return (0);
 416 
 417         if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r))
 418                 fatal(1, "short read from /dev/urandom");
 419 
 420         return (r % range);
 421 }
 422 
 423 /* ARGSUSED */
 424 static void
 425 ztest_record_enospc(char *s)
 426 {
 427         ztest_shared->zs_enospc_count++;
 428 }
 429 
 430 static void
 431 process_options(int argc, char **argv)
 432 {
 433         int opt;
 434         uint64_t value;
 435 
 436         /* By default, test gang blocks for blocks 32K and greater */
 437         metaslab_gang_bang = 32 << 10;
 438 
 439         while ((opt = getopt(argc, argv,
 440             "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:h")) != EOF) {
 441                 value = 0;
 442                 switch (opt) {
 443                 case 'v':
 444                 case 's':
 445                 case 'a':
 446                 case 'm':
 447                 case 'r':
 448                 case 'R':
 449                 case 'd':
 450                 case 't':
 451                 case 'g':
 452                 case 'i':
 453                 case 'k':
 454                 case 'T':
 455                 case 'P':
 456                         value = nicenumtoull(optarg);
 457                 }
 458                 switch (opt) {
 459                 case 'v':
 460                         zopt_vdevs = value;
 461                         break;
 462                 case 's':
 463                         zopt_vdev_size = MAX(SPA_MINDEVSIZE, value);
 464                         break;
 465                 case 'a':
 466                         zopt_ashift = value;
 467                         break;
 468                 case 'm':
 469                         zopt_mirrors = value;
 470                         break;
 471                 case 'r':
 472                         zopt_raidz = MAX(1, value);
 473                         break;
 474                 case 'R':
 475                         zopt_raidz_parity = MIN(MAX(value, 1), 2);
 476                         break;
 477                 case 'd':
 478                         zopt_datasets = MAX(1, value);
 479                         break;
 480                 case 't':
 481                         zopt_threads = MAX(1, value);
 482                         break;
 483                 case 'g':
 484                         metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value);
 485                         break;
 486                 case 'i':
 487                         zopt_init = value;
 488                         break;
 489                 case 'k':
 490                         zopt_killrate = value;
 491                         break;
 492                 case 'p':
 493                         zopt_pool = strdup(optarg);
 494                         break;
 495                 case 'f':
 496                         zopt_dir = strdup(optarg);
 497                         break;
 498                 case 'V':
 499                         zopt_verbose++;
 500                         break;
 501                 case 'E':
 502                         zopt_init = 0;
 503                         break;
 504                 case 'T':
 505                         zopt_time = value;
 506                         break;
 507                 case 'P':
 508                         zopt_passtime = MAX(1, value);
 509                         break;
 510                 case 'h':
 511                         usage(B_TRUE);
 512                         break;
 513                 case '?':
 514                 default:
 515                         usage(B_FALSE);
 516                         break;
 517                 }
 518         }
 519 
 520         zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1);
 521 
 522         zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time / zopt_vdevs : UINT64_MAX);
 523         zopt_maxfaults = MAX(zopt_mirrors, 1) * (zopt_raidz_parity + 1) - 1;
 524 }
 525 
 526 static uint64_t
 527 ztest_get_ashift(void)
 528 {
 529         if (zopt_ashift == 0)
 530                 return (SPA_MINBLOCKSHIFT + ztest_random(3));
 531         return (zopt_ashift);
 532 }
 533 
 534 static nvlist_t *
 535 make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift)
 536 {
 537         char pathbuf[MAXPATHLEN];
 538         uint64_t vdev;
 539         nvlist_t *file;
 540 
 541         if (ashift == 0)
 542                 ashift = ztest_get_ashift();
 543 
 544         if (path == NULL) {
 545                 path = pathbuf;
 546 
 547                 if (aux != NULL) {
 548                         vdev = ztest_shared->zs_vdev_aux;
 549                         (void) sprintf(path, ztest_aux_template,
 550                             zopt_dir, zopt_pool, aux, vdev);
 551                 } else {
 552                         vdev = ztest_shared->zs_vdev_primaries++;
 553                         (void) sprintf(path, ztest_dev_template,
 554                             zopt_dir, zopt_pool, vdev);
 555                 }
 556         }
 557 
 558         if (size != 0) {
 559                 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
 560                 if (fd == -1)
 561                         fatal(1, "can't open %s", path);
 562                 if (ftruncate(fd, size) != 0)
 563                         fatal(1, "can't ftruncate %s", path);
 564                 (void) close(fd);
 565         }
 566 
 567         VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
 568         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
 569         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
 570         VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
 571 
 572         return (file);
 573 }
 574 
 575 static nvlist_t *
 576 make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r)
 577 {
 578         nvlist_t *raidz, **child;
 579         int c;
 580 
 581         if (r < 2)
 582                 return (make_vdev_file(path, aux, size, ashift));
 583         child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
 584 
 585         for (c = 0; c < r; c++)
 586                 child[c] = make_vdev_file(path, aux, size, ashift);
 587 
 588         VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
 589         VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
 590             VDEV_TYPE_RAIDZ) == 0);
 591         VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
 592             zopt_raidz_parity) == 0);
 593         VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
 594             child, r) == 0);
 595 
 596         for (c = 0; c < r; c++)
 597                 nvlist_free(child[c]);
 598 
 599         umem_free(child, r * sizeof (nvlist_t *));
 600 
 601         return (raidz);
 602 }
 603 
 604 static nvlist_t *
 605 make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift,
 606         int r, int m)
 607 {
 608         nvlist_t *mirror, **child;
 609         int c;
 610 
 611         if (m < 1)
 612                 return (make_vdev_raidz(path, aux, size, ashift, r));
 613 
 614         child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
 615 
 616         for (c = 0; c < m; c++)
 617                 child[c] = make_vdev_raidz(path, aux, size, ashift, r);
 618 
 619         VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
 620         VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
 621             VDEV_TYPE_MIRROR) == 0);
 622         VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
 623             child, m) == 0);
 624 
 625         for (c = 0; c < m; c++)
 626                 nvlist_free(child[c]);
 627 
 628         umem_free(child, m * sizeof (nvlist_t *));
 629 
 630         return (mirror);
 631 }
 632 
 633 static nvlist_t *
 634 make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift,
 635         int log, int r, int m, int t)
 636 {
 637         nvlist_t *root, **child;
 638         int c;
 639 
 640         ASSERT(t > 0);
 641 
 642         child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
 643 
 644         for (c = 0; c < t; c++) {
 645                 child[c] = make_vdev_mirror(path, aux, size, ashift, r, m);
 646                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
 647                     log) == 0);
 648         }
 649 
 650         VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
 651         VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
 652         VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
 653             child, t) == 0);
 654 
 655         for (c = 0; c < t; c++)
 656                 nvlist_free(child[c]);
 657 
 658         umem_free(child, t * sizeof (nvlist_t *));
 659 
 660         return (root);
 661 }
 662 
 663 static void
 664 ztest_set_random_blocksize(objset_t *os, uint64_t object, dmu_tx_t *tx)
 665 {
 666         int bs = SPA_MINBLOCKSHIFT +
 667             ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1);
 668         int ibs = DN_MIN_INDBLKSHIFT +
 669             ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1);
 670         int error;
 671 
 672         error = dmu_object_set_blocksize(os, object, 1ULL << bs, ibs, tx);
 673         if (error) {
 674                 char osname[300];
 675                 dmu_objset_name(os, osname);
 676                 fatal(0, "dmu_object_set_blocksize('%s', %llu, %d, %d) = %d",
 677                     osname, object, 1 << bs, ibs, error);
 678         }
 679 }
 680 
 681 static uint8_t
 682 ztest_random_checksum(void)
 683 {
 684         uint8_t checksum;
 685 
 686         do {
 687                 checksum = ztest_random(ZIO_CHECKSUM_FUNCTIONS);
 688         } while (zio_checksum_table[checksum].ci_zbt);
 689 
 690         if (checksum == ZIO_CHECKSUM_OFF)
 691                 checksum = ZIO_CHECKSUM_ON;
 692 
 693         return (checksum);
 694 }
 695 
 696 static uint8_t
 697 ztest_random_compress(void)
 698 {
 699         return ((uint8_t)ztest_random(ZIO_COMPRESS_FUNCTIONS));
 700 }
 701 
 702 static uint8_t
 703 ztest_random_crypt(void)
 704 {
 705         return ((uint8_t)ztest_random(ZIO_CRYPT_FUNCTIONS));
 706 }
 707 
 708 static int
 709 ztest_replay_create(objset_t *os, lr_create_t *lr, boolean_t byteswap)
 710 {
 711         dmu_tx_t *tx;
 712         int error;
 713 
 714         if (byteswap)
 715                 byteswap_uint64_array(lr, sizeof (*lr));
 716 
 717         tx = dmu_tx_create(os);
 718         dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
 719         error = dmu_tx_assign(tx, TXG_WAIT);
 720         if (error) {
 721                 dmu_tx_abort(tx);
 722                 return (error);
 723         }
 724 
 725         error = dmu_object_claim(os, lr->lr_doid, lr->lr_mode, 0,
 726             DMU_OT_NONE, 0, tx);
 727         ASSERT3U(error, ==, 0);
 728         dmu_tx_commit(tx);
 729 
 730         if (zopt_verbose >= 5) {
 731                 char osname[MAXNAMELEN];
 732                 dmu_objset_name(os, osname);
 733                 (void) printf("replay create of %s object %llu"
 734                     " in txg %llu = %d\n",
 735                     osname, (u_longlong_t)lr->lr_doid,
 736                     (u_longlong_t)dmu_tx_get_txg(tx), error);
 737         }
 738 
 739         return (error);
 740 }
 741 
 742 static int
 743 ztest_replay_remove(objset_t *os, lr_remove_t *lr, boolean_t byteswap)
 744 {
 745         dmu_tx_t *tx;
 746         int error;
 747 
 748         if (byteswap)
 749                 byteswap_uint64_array(lr, sizeof (*lr));
 750 
 751         tx = dmu_tx_create(os);
 752         dmu_tx_hold_free(tx, lr->lr_doid, 0, DMU_OBJECT_END);
 753         error = dmu_tx_assign(tx, TXG_WAIT);
 754         if (error) {
 755                 dmu_tx_abort(tx);
 756                 return (error);
 757         }
 758 
 759         error = dmu_object_free(os, lr->lr_doid, tx);
 760         dmu_tx_commit(tx);
 761 
 762         return (error);
 763 }
 764 
 765 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
 766         NULL,                   /* 0 no such transaction type */
 767         ztest_replay_create,    /* TX_CREATE */
 768         NULL,                   /* TX_MKDIR */
 769         NULL,                   /* TX_MKXATTR */
 770         NULL,                   /* TX_SYMLINK */
 771         ztest_replay_remove,    /* TX_REMOVE */
 772         NULL,                   /* TX_RMDIR */
 773         NULL,                   /* TX_LINK */
 774         NULL,                   /* TX_RENAME */
 775         NULL,                   /* TX_WRITE */
 776         NULL,                   /* TX_TRUNCATE */
 777         NULL,                   /* TX_SETATTR */
 778         NULL,                   /* TX_ACL */
 779 };
 780 
 781 /*
 782  * Verify that we can't destroy an active pool, create an existing pool,
 783  * or create a pool with a bad vdev spec.
 784  */
 785 void
 786 ztest_spa_create_destroy(ztest_args_t *za)
 787 {
 788         int error;
 789         spa_t *spa;
 790         nvlist_t *nvroot;
 791 
 792         /*
 793          * Attempt to create using a bad file.
 794          */
 795         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
 796         error = spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL);
 797         nvlist_free(nvroot);
 798         if (error != ENOENT)
 799                 fatal(0, "spa_create(bad_file) = %d", error);
 800 
 801         /*
 802          * Attempt to create using a bad mirror.
 803          */
 804         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1);
 805         error = spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL);
 806         nvlist_free(nvroot);
 807         if (error != ENOENT)
 808                 fatal(0, "spa_create(bad_mirror) = %d", error);
 809 
 810         /*
 811          * Attempt to create an existing pool.  It shouldn't matter
 812          * what's in the nvroot; we should fail with EEXIST.
 813          */
 814         (void) rw_rdlock(&ztest_shared->zs_name_lock);
 815         nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
 816         error = spa_create(za->za_pool, nvroot, NULL, NULL, NULL);
 817         nvlist_free(nvroot);
 818         if (error != EEXIST)
 819                 fatal(0, "spa_create(whatever) = %d", error);
 820 
 821         error = spa_open(za->za_pool, &spa, FTAG);
 822         if (error)
 823                 fatal(0, "spa_open() = %d", error);
 824 
 825         error = spa_destroy(za->za_pool);
 826         if (error != EBUSY)
 827                 fatal(0, "spa_destroy() = %d", error);
 828 
 829         spa_close(spa, FTAG);
 830         (void) rw_unlock(&ztest_shared->zs_name_lock);
 831 }
 832 
 833 static vdev_t *
 834 vdev_lookup_by_path(vdev_t *vd, const char *path)
 835 {
 836         vdev_t *mvd;
 837 
 838         if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
 839                 return (vd);
 840 
 841         for (int c = 0; c < vd->vdev_children; c++)
 842                 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
 843                     NULL)
 844                         return (mvd);
 845 
 846         return (NULL);
 847 }
 848 
 849 /*
 850  * Verify that vdev_add() works as expected.
 851  */
 852 void
 853 ztest_vdev_add_remove(ztest_args_t *za)
 854 {
 855         spa_t *spa = za->za_spa;
 856         uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
 857         nvlist_t *nvroot;
 858         int error;
 859 
 860         (void) mutex_lock(&ztest_shared->zs_vdev_lock);
 861 
 862         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 863 
 864         ztest_shared->zs_vdev_primaries =
 865             spa->spa_root_vdev->vdev_children * leaves;
 866 
 867         spa_config_exit(spa, SCL_VDEV, FTAG);
 868 
 869         /*
 870          * Make 1/4 of the devices be log devices.
 871          */
 872         nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0,
 873             ztest_random(4) == 0, zopt_raidz, zopt_mirrors, 1);
 874 
 875         error = spa_vdev_add(spa, nvroot);
 876         nvlist_free(nvroot);
 877 
 878         (void) mutex_unlock(&ztest_shared->zs_vdev_lock);
 879 
 880         if (error == ENOSPC)
 881                 ztest_record_enospc("spa_vdev_add");
 882         else if (error != 0)
 883                 fatal(0, "spa_vdev_add() = %d", error);
 884 }
 885 
 886 /*
 887  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
 888  */
 889 void
 890 ztest_vdev_aux_add_remove(ztest_args_t *za)
 891 {
 892         spa_t *spa = za->za_spa;
 893         vdev_t *rvd = spa->spa_root_vdev;
 894         spa_aux_vdev_t *sav;
 895         char *aux;
 896         uint64_t guid = 0;
 897         int error;
 898 
 899         if (ztest_random(2) == 0) {
 900                 sav = &spa->spa_spares;
 901                 aux = ZPOOL_CONFIG_SPARES;
 902         } else {
 903                 sav = &spa->spa_l2cache;
 904                 aux = ZPOOL_CONFIG_L2CACHE;
 905         }
 906 
 907         (void) mutex_lock(&ztest_shared->zs_vdev_lock);
 908 
 909         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 910 
 911         if (sav->sav_count != 0 && ztest_random(4) == 0) {
 912                 /*
 913                  * Pick a random device to remove.
 914                  */
 915                 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
 916         } else {
 917                 /*
 918                  * Find an unused device we can add.
 919                  */
 920                 ztest_shared->zs_vdev_aux = 0;
 921                 for (;;) {
 922                         char path[MAXPATHLEN];
 923                         int c;
 924                         (void) sprintf(path, ztest_aux_template, zopt_dir,
 925                             zopt_pool, aux, ztest_shared->zs_vdev_aux);
 926                         for (c = 0; c < sav->sav_count; c++)
 927                                 if (strcmp(sav->sav_vdevs[c]->vdev_path,
 928                                     path) == 0)
 929                                         break;
 930                         if (c == sav->sav_count &&
 931                             vdev_lookup_by_path(rvd, path) == NULL)
 932                                 break;
 933                         ztest_shared->zs_vdev_aux++;
 934                 }
 935         }
 936 
 937         spa_config_exit(spa, SCL_VDEV, FTAG);
 938 
 939         if (guid == 0) {
 940                 /*
 941                  * Add a new device.
 942                  */
 943                 nvlist_t *nvroot = make_vdev_root(NULL, aux,
 944                     (zopt_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
 945                 error = spa_vdev_add(spa, nvroot);
 946                 if (error != 0)
 947                         fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
 948                 nvlist_free(nvroot);
 949         } else {
 950                 /*
 951                  * Remove an existing device.  Sometimes, dirty its
 952                  * vdev state first to make sure we handle removal
 953                  * of devices that have pending state changes.
 954                  */
 955                 if (ztest_random(2) == 0)
 956                         (void) vdev_online(spa, guid, B_FALSE, NULL);
 957 
 958                 error = spa_vdev_remove(spa, guid, B_FALSE);
 959                 if (error != 0 && error != EBUSY)
 960                         fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
 961         }
 962 
 963         (void) mutex_unlock(&ztest_shared->zs_vdev_lock);
 964 }
 965 
 966 /*
 967  * Verify that we can attach and detach devices.
 968  */
 969 void
 970 ztest_vdev_attach_detach(ztest_args_t *za)
 971 {
 972         spa_t *spa = za->za_spa;
 973         spa_aux_vdev_t *sav = &spa->spa_spares;
 974         vdev_t *rvd = spa->spa_root_vdev;
 975         vdev_t *oldvd, *newvd, *pvd;
 976         nvlist_t *root;
 977         uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
 978         uint64_t leaf, top;
 979         uint64_t ashift = ztest_get_ashift();
 980         uint64_t oldguid, pguid;
 981         size_t oldsize, newsize;
 982         char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
 983         int replacing;
 984         int oldvd_has_siblings = B_FALSE;
 985         int newvd_is_spare = B_FALSE;
 986         int oldvd_is_log;
 987         int error, expected_error;
 988 
 989         (void) mutex_lock(&ztest_shared->zs_vdev_lock);
 990 
 991         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 992 
 993         /*
 994          * Decide whether to do an attach or a replace.
 995          */
 996         replacing = ztest_random(2);
 997 
 998         /*
 999          * Pick a random top-level vdev.
1000          */
1001         top = ztest_random(rvd->vdev_children);
1002 
1003         /*
1004          * Pick a random leaf within it.
1005          */
1006         leaf = ztest_random(leaves);
1007 
1008         /*
1009          * Locate this vdev.
1010          */
1011         oldvd = rvd->vdev_child[top];
1012         if (zopt_mirrors >= 1) {
1013                 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
1014                 ASSERT(oldvd->vdev_children >= zopt_mirrors);
1015                 oldvd = oldvd->vdev_child[leaf / zopt_raidz];
1016         }
1017         if (zopt_raidz > 1) {
1018                 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
1019                 ASSERT(oldvd->vdev_children == zopt_raidz);
1020                 oldvd = oldvd->vdev_child[leaf % zopt_raidz];
1021         }
1022 
1023         /*
1024          * If we're already doing an attach or replace, oldvd may be a
1025          * mirror vdev -- in which case, pick a random child.
1026          */
1027         while (oldvd->vdev_children != 0) {
1028                 oldvd_has_siblings = B_TRUE;
1029                 ASSERT(oldvd->vdev_children >= 2);
1030                 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
1031         }
1032 
1033         oldguid = oldvd->vdev_guid;
1034         oldsize = vdev_get_rsize(oldvd);
1035         oldvd_is_log = oldvd->vdev_top->vdev_islog;
1036         (void) strcpy(oldpath, oldvd->vdev_path);
1037         pvd = oldvd->vdev_parent;
1038         pguid = pvd->vdev_guid;
1039 
1040         /*
1041          * If oldvd has siblings, then half of the time, detach it.
1042          */
1043         if (oldvd_has_siblings && ztest_random(2) == 0) {
1044                 spa_config_exit(spa, SCL_VDEV, FTAG);
1045                 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
1046                 if (error != 0 && error != ENODEV && error != EBUSY &&
1047                     error != ENOTSUP)
1048                         fatal(0, "detach (%s) returned %d", oldpath, error);
1049                 (void) mutex_unlock(&ztest_shared->zs_vdev_lock);
1050                 return;
1051         }
1052 
1053         /*
1054          * For the new vdev, choose with equal probability between the two
1055          * standard paths (ending in either 'a' or 'b') or a random hot spare.
1056          */
1057         if (sav->sav_count != 0 && ztest_random(3) == 0) {
1058                 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
1059                 newvd_is_spare = B_TRUE;
1060                 (void) strcpy(newpath, newvd->vdev_path);
1061         } else {
1062                 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
1063                     zopt_dir, zopt_pool, top * leaves + leaf);
1064                 if (ztest_random(2) == 0)
1065                         newpath[strlen(newpath) - 1] = 'b';
1066                 newvd = vdev_lookup_by_path(rvd, newpath);
1067         }
1068 
1069         if (newvd) {
1070                 newsize = vdev_get_rsize(newvd);
1071         } else {
1072                 /*
1073                  * Make newsize a little bigger or smaller than oldsize.
1074                  * If it's smaller, the attach should fail.
1075                  * If it's larger, and we're doing a replace,
1076                  * we should get dynamic LUN growth when we're done.
1077                  */
1078                 newsize = 10 * oldsize / (9 + ztest_random(3));
1079         }
1080 
1081         /*
1082          * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
1083          * unless it's a replace; in that case any non-replacing parent is OK.
1084          *
1085          * If newvd is already part of the pool, it should fail with EBUSY.
1086          *
1087          * If newvd is too small, it should fail with EOVERFLOW.
1088          */
1089         if (pvd->vdev_ops != &vdev_mirror_ops &&
1090             pvd->vdev_ops != &vdev_root_ops && (!replacing ||
1091             pvd->vdev_ops == &vdev_replacing_ops ||
1092             pvd->vdev_ops == &vdev_spare_ops))
1093                 expected_error = ENOTSUP;
1094         else if (newvd_is_spare && (!replacing || oldvd_is_log))
1095                 expected_error = ENOTSUP;
1096         else if (newvd == oldvd)
1097                 expected_error = replacing ? 0 : EBUSY;
1098         else if (vdev_lookup_by_path(rvd, newpath) != NULL)
1099                 expected_error = EBUSY;
1100         else if (newsize < oldsize)
1101                 expected_error = EOVERFLOW;
1102         else if (ashift > oldvd->vdev_top->vdev_ashift)
1103                 expected_error = EDOM;
1104         else
1105                 expected_error = 0;
1106 
1107         spa_config_exit(spa, SCL_VDEV, FTAG);
1108 
1109         /*
1110          * Build the nvlist describing newpath.
1111          */
1112         root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0,
1113             ashift, 0, 0, 0, 1);
1114 
1115         error = spa_vdev_attach(spa, oldguid, root, replacing);
1116 
1117         nvlist_free(root);
1118 
1119         /*
1120          * If our parent was the replacing vdev, but the replace completed,
1121          * then instead of failing with ENOTSUP we may either succeed,
1122          * fail with ENODEV, or fail with EOVERFLOW.
1123          */
1124         if (expected_error == ENOTSUP &&
1125             (error == 0 || error == ENODEV || error == EOVERFLOW))
1126                 expected_error = error;
1127 
1128         /*
1129          * If someone grew the LUN, the replacement may be too small.
1130          */
1131         if (error == EOVERFLOW || error == EBUSY)
1132                 expected_error = error;
1133 
1134         /* XXX workaround 6690467 */
1135         if (error != expected_error && expected_error != EBUSY) {
1136                 fatal(0, "attach (%s %llu, %s %llu, %d) "
1137                     "returned %d, expected %d",
1138                     oldpath, (longlong_t)oldsize, newpath,
1139                     (longlong_t)newsize, replacing, error, expected_error);
1140         }
1141 
1142         (void) mutex_unlock(&ztest_shared->zs_vdev_lock);
1143 }
1144 
1145 /*
1146  * Verify that dynamic LUN growth works as expected.
1147  */
1148 void
1149 ztest_vdev_LUN_growth(ztest_args_t *za)
1150 {
1151         spa_t *spa = za->za_spa;
1152         char dev_name[MAXPATHLEN];
1153         uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
1154         uint64_t vdev;
1155         size_t fsize;
1156         int fd;
1157 
1158         (void) mutex_lock(&ztest_shared->zs_vdev_lock);
1159 
1160         /*
1161          * Pick a random leaf vdev.
1162          */
1163         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1164         vdev = ztest_random(spa->spa_root_vdev->vdev_children * leaves);
1165         spa_config_exit(spa, SCL_VDEV, FTAG);
1166 
1167         (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
1168 
1169         if ((fd = open(dev_name, O_RDWR)) != -1) {
1170                 /*
1171                  * Determine the size.
1172                  */
1173                 fsize = lseek(fd, 0, SEEK_END);
1174 
1175                 /*
1176                  * If it's less than 2x the original size, grow by around 3%.
1177                  */
1178                 if (fsize < 2 * zopt_vdev_size) {
1179                         size_t newsize = fsize + ztest_random(fsize / 32);
1180                         (void) ftruncate(fd, newsize);
1181                         if (zopt_verbose >= 6) {
1182                                 (void) printf("%s grew from %lu to %lu bytes\n",
1183                                     dev_name, (ulong_t)fsize, (ulong_t)newsize);
1184                         }
1185                 }
1186                 (void) close(fd);
1187         }
1188 
1189         (void) mutex_unlock(&ztest_shared->zs_vdev_lock);
1190 }
1191 
1192 /* ARGSUSED */
1193 static void
1194 ztest_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
1195 {
1196         /*
1197          * Create the directory object.
1198          */
1199         VERIFY(dmu_object_claim(os, ZTEST_DIROBJ,
1200             DMU_OT_UINT64_OTHER, ZTEST_DIROBJ_BLOCKSIZE,
1201             DMU_OT_UINT64_OTHER, 5 * sizeof (ztest_block_tag_t), tx) == 0);
1202 
1203         VERIFY(zap_create_claim(os, ZTEST_MICROZAP_OBJ,
1204             DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
1205 
1206         VERIFY(zap_create_claim(os, ZTEST_FATZAP_OBJ,
1207             DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
1208 }
1209 
1210 static int
1211 ztest_destroy_cb(char *name, void *arg)
1212 {
1213         ztest_args_t *za = arg;
1214         objset_t *os;
1215         dmu_object_info_t *doi = &za->za_doi;
1216         int error;
1217 
1218         /*
1219          * Verify that the dataset contains a directory object.
1220          */
1221         error = dmu_objset_open(name, DMU_OST_OTHER,
1222             DS_MODE_USER | DS_MODE_READONLY, &os);
1223         ASSERT3U(error, ==, 0);
1224         error = dmu_object_info(os, ZTEST_DIROBJ, doi);
1225         if (error != ENOENT) {
1226                 /* We could have crashed in the middle of destroying it */
1227                 ASSERT3U(error, ==, 0);
1228                 ASSERT3U(doi->doi_type, ==, DMU_OT_UINT64_OTHER);
1229                 ASSERT3S(doi->doi_physical_blks, >=, 0);
1230         }
1231         dmu_objset_close(os);
1232 
1233         /*
1234          * Destroy the dataset.
1235          */
1236         error = dmu_objset_destroy(name);
1237         if (error) {
1238                 (void) dmu_objset_open(name, DMU_OST_OTHER,
1239                     DS_MODE_USER | DS_MODE_READONLY, &os);
1240                 fatal(0, "dmu_objset_destroy(os=%p) = %d\n", &os, error);
1241         }
1242         return (0);
1243 }
1244 
1245 /*
1246  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
1247  */
1248 static uint64_t
1249 ztest_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t object, int mode)
1250 {
1251         itx_t *itx;
1252         lr_create_t *lr;
1253         size_t namesize;
1254         char name[24];
1255 
1256         (void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object);
1257         namesize = strlen(name) + 1;
1258 
1259         itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize +
1260             ztest_random(ZIL_MAX_BLKSZ));
1261         lr = (lr_create_t *)&itx->itx_lr;
1262         bzero(lr + 1, lr->lr_common.lrc_reclen - sizeof (*lr));
1263         lr->lr_doid = object;
1264         lr->lr_foid = 0;
1265         lr->lr_mode = mode;
1266         lr->lr_uid = 0;
1267         lr->lr_gid = 0;
1268         lr->lr_gen = dmu_tx_get_txg(tx);
1269         lr->lr_crtime[0] = time(NULL);
1270         lr->lr_crtime[1] = 0;
1271         lr->lr_rdev = 0;
1272         bcopy(name, (char *)(lr + 1), namesize);
1273 
1274         return (zil_itx_assign(zilog, itx, tx));
1275 }
1276 
1277 void
1278 ztest_dmu_objset_create_destroy(ztest_args_t *za)
1279 {
1280         int error;
1281         objset_t *os, *os2;
1282         char name[100];
1283         int basemode, expected_error;
1284         zilog_t *zilog;
1285         uint64_t seq;
1286         uint64_t objects;
1287 
1288         (void) rw_rdlock(&ztest_shared->zs_name_lock);
1289         (void) snprintf(name, 100, "%s/%s_temp_%llu", za->za_pool, za->za_pool,
1290             (u_longlong_t)za->za_instance);
1291 
1292         basemode = DS_MODE_TYPE(za->za_instance);
1293         if (basemode != DS_MODE_USER && basemode != DS_MODE_OWNER)
1294                 basemode = DS_MODE_USER;
1295 
1296         /*
1297          * If this dataset exists from a previous run, process its replay log
1298          * half of the time.  If we don't replay it, then dmu_objset_destroy()
1299          * (invoked from ztest_destroy_cb() below) should just throw it away.
1300          */
1301         if (ztest_random(2) == 0 &&
1302             dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_OWNER, &os) == 0) {
1303                 zil_replay(os, os, ztest_replay_vector);
1304                 dmu_objset_close(os);
1305         }
1306 
1307         /*
1308          * There may be an old instance of the dataset we're about to
1309          * create lying around from a previous run.  If so, destroy it
1310          * and all of its snapshots.
1311          */
1312         (void) dmu_objset_find(name, ztest_destroy_cb, za,
1313             DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1314 
1315         /*
1316          * Verify that the destroyed dataset is no longer in the namespace.
1317          */
1318         error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os);
1319         if (error != ENOENT)
1320                 fatal(1, "dmu_objset_open(%s) found destroyed dataset %p",
1321                     name, os);
1322 
1323         /*
1324          * Verify that we can create a new dataset.
1325          */
1326         error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0,
1327             ztest_create_cb, NULL);
1328         if (error) {
1329                 if (error == ENOSPC) {
1330                         ztest_record_enospc("dmu_objset_create");
1331                         (void) rw_unlock(&ztest_shared->zs_name_lock);
1332                         return;
1333                 }
1334                 fatal(0, "dmu_objset_create(%s) = %d", name, error);
1335         }
1336 
1337         error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os);
1338         if (error) {
1339                 fatal(0, "dmu_objset_open(%s) = %d", name, error);
1340         }
1341 
1342         /*
1343          * Open the intent log for it.
1344          */
1345         zilog = zil_open(os, NULL);
1346 
1347         /*
1348          * Put a random number of objects in there.
1349          */
1350         objects = ztest_random(20);
1351         seq = 0;
1352         while (objects-- != 0) {
1353                 uint64_t object;
1354                 dmu_tx_t *tx = dmu_tx_create(os);
1355                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, sizeof (name));
1356                 error = dmu_tx_assign(tx, TXG_WAIT);
1357                 if (error) {
1358                         dmu_tx_abort(tx);
1359                 } else {
1360                         object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1361                             DMU_OT_NONE, 0, tx);
1362                         ztest_set_random_blocksize(os, object, tx);
1363                         seq = ztest_log_create(zilog, tx, object,
1364                             DMU_OT_UINT64_OTHER);
1365                         dmu_write(os, object, 0, sizeof (name), name, tx);
1366                         dmu_tx_commit(tx);
1367                 }
1368                 if (ztest_random(5) == 0) {
1369                         zil_commit(zilog, seq, object);
1370                 }
1371                 if (ztest_random(100) == 0) {
1372                         error = zil_suspend(zilog);
1373                         if (error == 0) {
1374                                 zil_resume(zilog);
1375                         }
1376                 }
1377         }
1378 
1379         /*
1380          * Verify that we cannot create an existing dataset.
1381          */
1382         error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0, NULL, NULL);
1383         if (error != EEXIST)
1384                 fatal(0, "created existing dataset, error = %d", error);
1385 
1386         /*
1387          * Verify that multiple dataset holds are allowed, but only when
1388          * the new access mode is compatible with the base mode.
1389          */
1390         if (basemode == DS_MODE_OWNER) {
1391                 error = dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_USER,
1392                     &os2);
1393                 if (error)
1394                         fatal(0, "dmu_objset_open('%s') = %d", name, error);
1395                 else
1396                         dmu_objset_close(os2);
1397         }
1398         error = dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_OWNER, &os2);
1399         expected_error = (basemode == DS_MODE_OWNER) ? EBUSY : 0;
1400         if (error != expected_error)
1401                 fatal(0, "dmu_objset_open('%s') = %d, expected %d",
1402                     name, error, expected_error);
1403         if (error == 0)
1404                 dmu_objset_close(os2);
1405 
1406         zil_close(zilog);
1407         dmu_objset_close(os);
1408 
1409         error = dmu_objset_destroy(name);
1410         if (error)
1411                 fatal(0, "dmu_objset_destroy(%s) = %d", name, error);
1412 
1413         (void) rw_unlock(&ztest_shared->zs_name_lock);
1414 }
1415 
1416 /*
1417  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
1418  */
1419 void
1420 ztest_dmu_snapshot_create_destroy(ztest_args_t *za)
1421 {
1422         int error;
1423         objset_t *os = za->za_os;
1424         char snapname[100];
1425         char osname[MAXNAMELEN];
1426 
1427         (void) rw_rdlock(&ztest_shared->zs_name_lock);
1428         dmu_objset_name(os, osname);
1429         (void) snprintf(snapname, 100, "%s@%llu", osname,
1430             (u_longlong_t)za->za_instance);
1431 
1432         error = dmu_objset_destroy(snapname);
1433         if (error != 0 && error != ENOENT)
1434                 fatal(0, "dmu_objset_destroy() = %d", error);
1435         error = dmu_objset_snapshot(osname, strchr(snapname, '@')+1, FALSE);
1436         if (error == ENOSPC)
1437                 ztest_record_enospc("dmu_take_snapshot");
1438         else if (error != 0 && error != EEXIST)
1439                 fatal(0, "dmu_take_snapshot() = %d", error);
1440         (void) rw_unlock(&ztest_shared->zs_name_lock);
1441 }
1442 
1443 /*
1444  * Verify that dmu_object_{alloc,free} work as expected.
1445  */
1446 void
1447 ztest_dmu_object_alloc_free(ztest_args_t *za)
1448 {
1449         objset_t *os = za->za_os;
1450         dmu_buf_t *db;
1451         dmu_tx_t *tx;
1452         uint64_t batchobj, object, batchsize, endoff, temp;
1453         int b, c, error, bonuslen;
1454         dmu_object_info_t *doi = &za->za_doi;
1455         char osname[MAXNAMELEN];
1456 
1457         dmu_objset_name(os, osname);
1458 
1459         endoff = -8ULL;
1460         batchsize = 2;
1461 
1462         /*
1463          * Create a batch object if necessary, and record it in the directory.
1464          */
1465         VERIFY3U(0, ==, dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
1466             sizeof (uint64_t), &batchobj));
1467         if (batchobj == 0) {
1468                 tx = dmu_tx_create(os);
1469                 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff,
1470                     sizeof (uint64_t));
1471                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1472                 error = dmu_tx_assign(tx, TXG_WAIT);
1473                 if (error) {
1474                         ztest_record_enospc("create a batch object");
1475                         dmu_tx_abort(tx);
1476                         return;
1477                 }
1478                 batchobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1479                     DMU_OT_NONE, 0, tx);
1480                 ztest_set_random_blocksize(os, batchobj, tx);
1481                 dmu_write(os, ZTEST_DIROBJ, za->za_diroff,
1482                     sizeof (uint64_t), &batchobj, tx);
1483                 dmu_tx_commit(tx);
1484         }
1485 
1486         /*
1487          * Destroy the previous batch of objects.
1488          */
1489         for (b = 0; b < batchsize; b++) {
1490                 VERIFY3U(0, ==, dmu_read(os, batchobj, b * sizeof (uint64_t),
1491                     sizeof (uint64_t), &object));
1492                 if (object == 0)
1493                         continue;
1494                 /*
1495                  * Read and validate contents.
1496                  * We expect the nth byte of the bonus buffer to be n.
1497                  */
1498                 VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db));
1499                 za->za_dbuf = db;
1500 
1501                 dmu_object_info_from_db(db, doi);
1502                 ASSERT(doi->doi_type == DMU_OT_UINT64_OTHER);
1503                 ASSERT(doi->doi_bonus_type == DMU_OT_PLAIN_OTHER);
1504                 ASSERT3S(doi->doi_physical_blks, >=, 0);
1505 
1506                 bonuslen = doi->doi_bonus_size;
1507 
1508                 for (c = 0; c < bonuslen; c++) {
1509                         if (((uint8_t *)db->db_data)[c] !=
1510                             (uint8_t)(c + bonuslen)) {
1511                                 fatal(0,
1512                                     "bad bonus: %s, obj %llu, off %d: %u != %u",
1513                                     osname, object, c,
1514                                     ((uint8_t *)db->db_data)[c],
1515                                     (uint8_t)(c + bonuslen));
1516                         }
1517                 }
1518 
1519                 dmu_buf_rele(db, FTAG);
1520                 za->za_dbuf = NULL;
1521 
1522                 /*
1523                  * We expect the word at endoff to be our object number.
1524                  */
1525                 VERIFY(0 == dmu_read(os, object, endoff,
1526                     sizeof (uint64_t), &temp));
1527 
1528                 if (temp != object) {
1529                         fatal(0, "bad data in %s, got %llu, expected %llu",
1530                             osname, temp, object);
1531                 }
1532 
1533                 /*
1534                  * Destroy old object and clear batch entry.
1535                  */
1536                 tx = dmu_tx_create(os);
1537                 dmu_tx_hold_write(tx, batchobj,
1538                     b * sizeof (uint64_t), sizeof (uint64_t));
1539                 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1540                 error = dmu_tx_assign(tx, TXG_WAIT);
1541                 if (error) {
1542                         ztest_record_enospc("free object");
1543                         dmu_tx_abort(tx);
1544                         return;
1545                 }
1546                 error = dmu_object_free(os, object, tx);
1547                 if (error) {
1548                         fatal(0, "dmu_object_free('%s', %llu) = %d",
1549                             osname, object, error);
1550                 }
1551                 object = 0;
1552 
1553                 dmu_object_set_checksum(os, batchobj,
1554                     ztest_random_checksum(), tx);
1555                 dmu_object_set_compress(os, batchobj,
1556                     ztest_random_compress(), tx);
1557                 dmu_object_set_crypt(os, batchobj,
1558                     ztest_random_crypt(), tx);
1559 
1560                 dmu_write(os, batchobj, b * sizeof (uint64_t),
1561                     sizeof (uint64_t), &object, tx);
1562 
1563                 dmu_tx_commit(tx);
1564         }
1565 
1566         /*
1567          * Before creating the new batch of objects, generate a bunch of churn.
1568          */
1569         for (b = ztest_random(100); b > 0; b--) {
1570                 tx = dmu_tx_create(os);
1571                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1572                 error = dmu_tx_assign(tx, TXG_WAIT);
1573                 if (error) {
1574                         ztest_record_enospc("churn objects");
1575                         dmu_tx_abort(tx);
1576                         return;
1577                 }
1578                 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1579                     DMU_OT_NONE, 0, tx);
1580                 ztest_set_random_blocksize(os, object, tx);
1581                 error = dmu_object_free(os, object, tx);
1582                 if (error) {
1583                         fatal(0, "dmu_object_free('%s', %llu) = %d",
1584                             osname, object, error);
1585                 }
1586                 dmu_tx_commit(tx);
1587         }
1588 
1589         /*
1590          * Create a new batch of objects with randomly chosen
1591          * blocksizes and record them in the batch directory.
1592          */
1593         for (b = 0; b < batchsize; b++) {
1594                 uint32_t va_blksize;
1595                 u_longlong_t va_nblocks;
1596 
1597                 tx = dmu_tx_create(os);
1598                 dmu_tx_hold_write(tx, batchobj, b * sizeof (uint64_t),
1599                     sizeof (uint64_t));
1600                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1601                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, endoff,
1602                     sizeof (uint64_t));
1603                 error = dmu_tx_assign(tx, TXG_WAIT);
1604                 if (error) {
1605                         ztest_record_enospc("create batchobj");
1606                         dmu_tx_abort(tx);
1607                         return;
1608                 }
1609                 bonuslen = (int)ztest_random(dmu_bonus_max()) + 1;
1610 
1611                 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1612                     DMU_OT_PLAIN_OTHER, bonuslen, tx);
1613 
1614                 ztest_set_random_blocksize(os, object, tx);
1615 
1616                 dmu_object_set_checksum(os, object,
1617                     ztest_random_checksum(), tx);
1618                 dmu_object_set_compress(os, object,
1619                     ztest_random_compress(), tx);
1620                 dmu_object_set_crypt(os, object,
1621                     ztest_random_crypt(), tx);
1622 
1623                 dmu_write(os, batchobj, b * sizeof (uint64_t),
1624                     sizeof (uint64_t), &object, tx);
1625 
1626                 /*
1627                  * Write to both the bonus buffer and the regular data.
1628                  */
1629                 VERIFY(dmu_bonus_hold(os, object, FTAG, &db) == 0);
1630                 za->za_dbuf = db;
1631                 ASSERT3U(bonuslen, <=, db->db_size);
1632 
1633                 dmu_object_size_from_db(db, &va_blksize, &va_nblocks);
1634                 ASSERT3S(va_nblocks, >=, 0);
1635 
1636                 dmu_buf_will_dirty(db, tx);
1637 
1638                 /*
1639                  * See comments above regarding the contents of
1640                  * the bonus buffer and the word at endoff.
1641                  */
1642                 for (c = 0; c < bonuslen; c++)
1643                         ((uint8_t *)db->db_data)[c] = (uint8_t)(c + bonuslen);
1644 
1645                 dmu_buf_rele(db, FTAG);
1646                 za->za_dbuf = NULL;
1647 
1648                 /*
1649                  * Write to a large offset to increase indirection.
1650                  */
1651                 dmu_write(os, object, endoff, sizeof (uint64_t), &object, tx);
1652 
1653                 dmu_tx_commit(tx);
1654         }
1655 }
1656 
1657 /*
1658  * Verify that dmu_{read,write} work as expected.
1659  */
1660 typedef struct bufwad {
1661         uint64_t        bw_index;
1662         uint64_t        bw_txg;
1663         uint64_t        bw_data;
1664 } bufwad_t;
1665 
1666 typedef struct dmu_read_write_dir {
1667         uint64_t        dd_packobj;
1668         uint64_t        dd_bigobj;
1669         uint64_t        dd_chunk;
1670 } dmu_read_write_dir_t;
1671 
1672 void
1673 ztest_dmu_read_write(ztest_args_t *za)
1674 {
1675         objset_t *os = za->za_os;
1676         dmu_read_write_dir_t dd;
1677         dmu_tx_t *tx;
1678         int i, freeit, error;
1679         uint64_t n, s, txg;
1680         bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
1681         uint64_t packoff, packsize, bigoff, bigsize;
1682         uint64_t regions = 997;
1683         uint64_t stride = 123456789ULL;
1684         uint64_t width = 40;
1685         int free_percent = 5;
1686 
1687         /*
1688          * This test uses two objects, packobj and bigobj, that are always
1689          * updated together (i.e. in the same tx) so that their contents are
1690          * in sync and can be compared.  Their contents relate to each other
1691          * in a simple way: packobj is a dense array of 'bufwad' structures,
1692          * while bigobj is a sparse array of the same bufwads.  Specifically,
1693          * for any index n, there are three bufwads that should be identical:
1694          *
1695          *      packobj, at offset n * sizeof (bufwad_t)
1696          *      bigobj, at the head of the nth chunk
1697          *      bigobj, at the tail of the nth chunk
1698          *
1699          * The chunk size is arbitrary. It doesn't have to be a power of two,
1700          * and it doesn't have any relation to the object blocksize.
1701          * The only requirement is that it can hold at least two bufwads.
1702          *
1703          * Normally, we write the bufwad to each of these locations.
1704          * However, free_percent of the time we instead write zeroes to
1705          * packobj and perform a dmu_free_range() on bigobj.  By comparing
1706          * bigobj to packobj, we can verify that the DMU is correctly
1707          * tracking which parts of an object are allocated and free,
1708          * and that the contents of the allocated blocks are correct.
1709          */
1710 
1711         /*
1712          * Read the directory info.  If it's the first time, set things up.
1713          */
1714         VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
1715             sizeof (dd), &dd));
1716         if (dd.dd_chunk == 0) {
1717                 ASSERT(dd.dd_packobj == 0);
1718                 ASSERT(dd.dd_bigobj == 0);
1719                 tx = dmu_tx_create(os);
1720                 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (dd));
1721                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1722                 error = dmu_tx_assign(tx, TXG_WAIT);
1723                 if (error) {
1724                         ztest_record_enospc("create r/w directory");
1725                         dmu_tx_abort(tx);
1726                         return;
1727                 }
1728 
1729                 dd.dd_packobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1730                     DMU_OT_NONE, 0, tx);
1731                 dd.dd_bigobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1732                     DMU_OT_NONE, 0, tx);
1733                 dd.dd_chunk = (1000 + ztest_random(1000)) * sizeof (uint64_t);
1734 
1735                 ztest_set_random_blocksize(os, dd.dd_packobj, tx);
1736                 ztest_set_random_blocksize(os, dd.dd_bigobj, tx);
1737 
1738                 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (dd), &dd,
1739                     tx);
1740                 dmu_tx_commit(tx);
1741         }
1742 
1743         /*
1744          * Prefetch a random chunk of the big object.
1745          * Our aim here is to get some async reads in flight
1746          * for blocks that we may free below; the DMU should
1747          * handle this race correctly.
1748          */
1749         n = ztest_random(regions) * stride + ztest_random(width);
1750         s = 1 + ztest_random(2 * width - 1);
1751         dmu_prefetch(os, dd.dd_bigobj, n * dd.dd_chunk, s * dd.dd_chunk);
1752 
1753         /*
1754          * Pick a random index and compute the offsets into packobj and bigobj.
1755          */
1756         n = ztest_random(regions) * stride + ztest_random(width);
1757         s = 1 + ztest_random(width - 1);
1758 
1759         packoff = n * sizeof (bufwad_t);
1760         packsize = s * sizeof (bufwad_t);
1761 
1762         bigoff = n * dd.dd_chunk;
1763         bigsize = s * dd.dd_chunk;
1764 
1765         packbuf = umem_alloc(packsize, UMEM_NOFAIL);
1766         bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
1767 
1768         /*
1769          * free_percent of the time, free a range of bigobj rather than
1770          * overwriting it.
1771          */
1772         freeit = (ztest_random(100) < free_percent);
1773 
1774         /*
1775          * Read the current contents of our objects.
1776          */
1777         error = dmu_read(os, dd.dd_packobj, packoff, packsize, packbuf);
1778         ASSERT3U(error, ==, 0);
1779         error = dmu_read(os, dd.dd_bigobj, bigoff, bigsize, bigbuf);
1780         ASSERT3U(error, ==, 0);
1781 
1782         /*
1783          * Get a tx for the mods to both packobj and bigobj.
1784          */
1785         tx = dmu_tx_create(os);
1786 
1787         dmu_tx_hold_write(tx, dd.dd_packobj, packoff, packsize);
1788 
1789         if (freeit)
1790                 dmu_tx_hold_free(tx, dd.dd_bigobj, bigoff, bigsize);
1791         else
1792                 dmu_tx_hold_write(tx, dd.dd_bigobj, bigoff, bigsize);
1793 
1794         error = dmu_tx_assign(tx, TXG_WAIT);
1795 
1796         if (error) {
1797                 ztest_record_enospc("dmu r/w range");
1798                 dmu_tx_abort(tx);
1799                 umem_free(packbuf, packsize);
1800                 umem_free(bigbuf, bigsize);
1801                 return;
1802         }
1803 
1804         txg = dmu_tx_get_txg(tx);
1805 
1806         /*
1807          * For each index from n to n + s, verify that the existing bufwad
1808          * in packobj matches the bufwads at the head and tail of the
1809          * corresponding chunk in bigobj.  Then update all three bufwads
1810          * with the new values we want to write out.
1811          */
1812         for (i = 0; i < s; i++) {
1813                 /* LINTED */
1814                 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
1815                 /* LINTED */
1816                 bigH = (bufwad_t *)((char *)bigbuf + i * dd.dd_chunk);
1817                 /* LINTED */
1818                 bigT = (bufwad_t *)((char *)bigH + dd.dd_chunk) - 1;
1819 
1820                 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
1821                 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
1822 
1823                 if (pack->bw_txg > txg)
1824                         fatal(0, "future leak: got %llx, open txg is %llx",
1825                             pack->bw_txg, txg);
1826 
1827                 if (pack->bw_data != 0 && pack->bw_index != n + i)
1828                         fatal(0, "wrong index: got %llx, wanted %llx+%llx",
1829                             pack->bw_index, n, i);
1830 
1831                 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
1832                         fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
1833 
1834                 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
1835                         fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
1836 
1837                 if (freeit) {
1838                         bzero(pack, sizeof (bufwad_t));
1839                 } else {
1840                         pack->bw_index = n + i;
1841                         pack->bw_txg = txg;
1842                         pack->bw_data = 1 + ztest_random(-2ULL);
1843                 }
1844                 *bigH = *pack;
1845                 *bigT = *pack;
1846         }
1847 
1848         /*
1849          * We've verified all the old bufwads, and made new ones.
1850          * Now write them out.
1851          */
1852         dmu_write(os, dd.dd_packobj, packoff, packsize, packbuf, tx);
1853 
1854         if (freeit) {
1855                 if (zopt_verbose >= 6) {
1856                         (void) printf("freeing offset %llx size %llx"
1857                             " txg %llx\n",
1858                             (u_longlong_t)bigoff,
1859                             (u_longlong_t)bigsize,
1860                             (u_longlong_t)txg);
1861                 }
1862                 VERIFY(0 == dmu_free_range(os, dd.dd_bigobj, bigoff,
1863                     bigsize, tx));
1864         } else {
1865                 if (zopt_verbose >= 6) {
1866                         (void) printf("writing offset %llx size %llx"
1867                             " txg %llx\n",
1868                             (u_longlong_t)bigoff,
1869                             (u_longlong_t)bigsize,
1870                             (u_longlong_t)txg);
1871                 }
1872                 dmu_write(os, dd.dd_bigobj, bigoff, bigsize, bigbuf, tx);
1873         }
1874 
1875         dmu_tx_commit(tx);
1876 
1877         /*
1878          * Sanity check the stuff we just wrote.
1879          */
1880         {
1881                 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
1882                 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
1883 
1884                 VERIFY(0 == dmu_read(os, dd.dd_packobj, packoff,
1885                     packsize, packcheck));
1886                 VERIFY(0 == dmu_read(os, dd.dd_bigobj, bigoff,
1887                     bigsize, bigcheck));
1888 
1889                 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
1890                 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
1891 
1892                 umem_free(packcheck, packsize);
1893                 umem_free(bigcheck, bigsize);
1894         }
1895 
1896         umem_free(packbuf, packsize);
1897         umem_free(bigbuf, bigsize);
1898 }
1899 
1900 void
1901 ztest_dmu_check_future_leak(ztest_args_t *za)
1902 {
1903         objset_t *os = za->za_os;
1904         dmu_buf_t *db;
1905         ztest_block_tag_t *bt;
1906         dmu_object_info_t *doi = &za->za_doi;
1907 
1908         /*
1909          * Make sure that, if there is a write record in the bonus buffer
1910          * of the ZTEST_DIROBJ, that the txg for this record is <= the
1911          * last synced txg of the pool.
1912          */
1913         VERIFY(dmu_bonus_hold(os, ZTEST_DIROBJ, FTAG, &db) == 0);
1914         za->za_dbuf = db;
1915         VERIFY(dmu_object_info(os, ZTEST_DIROBJ, doi) == 0);
1916         ASSERT3U(doi->doi_bonus_size, >=, sizeof (*bt));
1917         ASSERT3U(doi->doi_bonus_size, <=, db->db_size);
1918         ASSERT3U(doi->doi_bonus_size % sizeof (*bt), ==, 0);
1919         bt = (void *)((char *)db->db_data + doi->doi_bonus_size - sizeof (*bt));
1920         if (bt->bt_objset != 0) {
1921                 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1922                 ASSERT3U(bt->bt_object, ==, ZTEST_DIROBJ);
1923                 ASSERT3U(bt->bt_offset, ==, -1ULL);
1924                 ASSERT3U(bt->bt_txg, <, spa_first_txg(za->za_spa));
1925         }
1926         dmu_buf_rele(db, FTAG);
1927         za->za_dbuf = NULL;
1928 }
1929 
1930 void
1931 ztest_dmu_write_parallel(ztest_args_t *za)
1932 {
1933         objset_t *os = za->za_os;
1934         ztest_block_tag_t *rbt = &za->za_rbt;
1935         ztest_block_tag_t *wbt = &za->za_wbt;
1936         const size_t btsize = sizeof (ztest_block_tag_t);
1937         dmu_buf_t *db;
1938         int b, error;
1939         int bs = ZTEST_DIROBJ_BLOCKSIZE;
1940         int do_free = 0;
1941         uint64_t off, txg, txg_how;
1942         mutex_t *lp;
1943         char osname[MAXNAMELEN];
1944         char iobuf[SPA_MAXBLOCKSIZE];
1945         blkptr_t blk = { 0 };
1946         uint64_t blkoff;
1947         zbookmark_t zb;
1948         dmu_tx_t *tx = dmu_tx_create(os);
1949 
1950         dmu_objset_name(os, osname);
1951 
1952         /*
1953          * Have multiple threads write to large offsets in ZTEST_DIROBJ
1954          * to verify that having multiple threads writing to the same object
1955          * in parallel doesn't cause any trouble.
1956          */
1957         if (ztest_random(4) == 0) {
1958                 /*
1959                  * Do the bonus buffer instead of a regular block.
1960                  * We need a lock to serialize resize vs. others,
1961                  * so we hash on the objset ID.
1962                  */
1963                 b = dmu_objset_id(os) % ZTEST_SYNC_LOCKS;
1964                 off = -1ULL;
1965                 dmu_tx_hold_bonus(tx, ZTEST_DIROBJ);
1966         } else {
1967                 b = ztest_random(ZTEST_SYNC_LOCKS);
1968                 off = za->za_diroff_shared + (b << SPA_MAXBLOCKSHIFT);
1969                 if (ztest_random(4) == 0) {
1970                         do_free = 1;
1971                         dmu_tx_hold_free(tx, ZTEST_DIROBJ, off, bs);
1972                 } else {
1973                         dmu_tx_hold_write(tx, ZTEST_DIROBJ, off, bs);
1974                 }
1975         }
1976 
1977         txg_how = ztest_random(2) == 0 ? TXG_WAIT : TXG_NOWAIT;
1978         error = dmu_tx_assign(tx, txg_how);
1979         if (error) {
1980                 if (error == ERESTART) {
1981                         ASSERT(txg_how == TXG_NOWAIT);
1982                         dmu_tx_wait(tx);
1983                 } else {
1984                         ztest_record_enospc("dmu write parallel");
1985                 }
1986                 dmu_tx_abort(tx);
1987                 return;
1988         }
1989         txg = dmu_tx_get_txg(tx);
1990 
1991         lp = &ztest_shared->zs_sync_lock[b];
1992         (void) mutex_lock(lp);
1993 
1994         wbt->bt_objset = dmu_objset_id(os);
1995         wbt->bt_object = ZTEST_DIROBJ;
1996         wbt->bt_offset = off;
1997         wbt->bt_txg = txg;
1998         wbt->bt_thread = za->za_instance;
1999         wbt->bt_seq = ztest_shared->zs_seq[b]++;  /* protected by lp */
2000 
2001         /*
2002          * Occasionally, write an all-zero block to test the behavior
2003          * of blocks that compress into holes.
2004          */
2005         if (off != -1ULL && ztest_random(8) == 0)
2006                 bzero(wbt, btsize);
2007 
2008         if (off == -1ULL) {
2009                 dmu_object_info_t *doi = &za->za_doi;
2010                 char *dboff;
2011 
2012                 VERIFY(dmu_bonus_hold(os, ZTEST_DIROBJ, FTAG, &db) == 0);
2013                 za->za_dbuf = db;
2014                 dmu_object_info_from_db(db, doi);
2015                 ASSERT3U(doi->doi_bonus_size, <=, db->db_size);
2016                 ASSERT3U(doi->doi_bonus_size, >=, btsize);
2017                 ASSERT3U(doi->doi_bonus_size % btsize, ==, 0);
2018                 dboff = (char *)db->db_data + doi->doi_bonus_size - btsize;
2019                 bcopy(dboff, rbt, btsize);
2020                 if (rbt->bt_objset != 0) {
2021                         ASSERT3U(rbt->bt_objset, ==, wbt->bt_objset);
2022                         ASSERT3U(rbt->bt_object, ==, wbt->bt_object);
2023                         ASSERT3U(rbt->bt_offset, ==, wbt->bt_offset);
2024                         ASSERT3U(rbt->bt_txg, <=, wbt->bt_txg);
2025                 }
2026                 if (ztest_random(10) == 0) {
2027                         int newsize = (ztest_random(db->db_size /
2028                             btsize) + 1) * btsize;
2029 
2030                         ASSERT3U(newsize, >=, btsize);
2031                         ASSERT3U(newsize, <=, db->db_size);
2032                         VERIFY3U(dmu_set_bonus(db, newsize, tx), ==, 0);
2033                         dboff = (char *)db->db_data + newsize - btsize;
2034                 }
2035                 dmu_buf_will_dirty(db, tx);
2036                 bcopy(wbt, dboff, btsize);
2037                 dmu_buf_rele(db, FTAG);
2038                 za->za_dbuf = NULL;
2039         } else if (do_free) {
2040                 VERIFY(dmu_free_range(os, ZTEST_DIROBJ, off, bs, tx) == 0);
2041         } else {
2042                 dmu_write(os, ZTEST_DIROBJ, off, btsize, wbt, tx);
2043         }
2044 
2045         (void) mutex_unlock(lp);
2046 
2047         if (ztest_random(1000) == 0)
2048                 (void) poll(NULL, 0, 1); /* open dn_notxholds window */
2049 
2050         dmu_tx_commit(tx);
2051 
2052         if (ztest_random(10000) == 0)
2053                 txg_wait_synced(dmu_objset_pool(os), txg);
2054 
2055         if (off == -1ULL || do_free)
2056                 return;
2057 
2058         if (ztest_random(2) != 0)
2059                 return;
2060 
2061         /*
2062          * dmu_sync() the block we just wrote.
2063          */
2064         (void) mutex_lock(lp);
2065 
2066         blkoff = P2ALIGN_TYPED(off, bs, uint64_t);
2067         error = dmu_buf_hold(os, ZTEST_DIROBJ, blkoff, FTAG, &db);
2068         za->za_dbuf = db;
2069         if (error) {
2070                 (void) mutex_unlock(lp);
2071                 return;
2072         }
2073         blkoff = off - blkoff;
2074         error = dmu_sync(NULL, db, &blk, txg, NULL, NULL);
2075         dmu_buf_rele(db, FTAG);
2076         za->za_dbuf = NULL;
2077 
2078         (void) mutex_unlock(lp);
2079 
2080         if (error)
2081                 return;
2082 
2083         if (blk.blk_birth == 0)         /* concurrent free */
2084                 return;
2085 
2086         txg_suspend(dmu_objset_pool(os));
2087 
2088         ASSERT(blk.blk_fill == 1);
2089         ASSERT3U(BP_GET_TYPE(&blk), ==, DMU_OT_UINT64_OTHER);
2090         ASSERT3U(BP_GET_LEVEL(&blk), ==, 0);
2091         ASSERT3U(BP_GET_LSIZE(&blk), ==, bs);
2092 
2093         /*
2094          * Read the block that dmu_sync() returned to make sure its contents
2095          * match what we wrote.  We do this while still txg_suspend()ed
2096          * to ensure that the block can't be reused before we read it.
2097          */
2098         zb.zb_objset = dmu_objset_id(os);
2099         zb.zb_object = ZTEST_DIROBJ;
2100         zb.zb_level = 0;
2101         zb.zb_blkid = off / bs;
2102         error = zio_wait(zio_read(NULL, za->za_spa, &blk, iobuf, bs,
2103             NULL, NULL, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_MUSTSUCCEED, &zb));
2104         ASSERT3U(error, ==, 0);
2105 
2106         txg_resume(dmu_objset_pool(os));
2107 
2108         bcopy(&iobuf[blkoff], rbt, btsize);
2109 
2110         if (rbt->bt_objset == 0)             /* concurrent free */
2111                 return;
2112 
2113         if (wbt->bt_objset == 0)             /* all-zero overwrite */
2114                 return;
2115 
2116         ASSERT3U(rbt->bt_objset, ==, wbt->bt_objset);
2117         ASSERT3U(rbt->bt_object, ==, wbt->bt_object);
2118         ASSERT3U(rbt->bt_offset, ==, wbt->bt_offset);
2119 
2120         /*
2121          * The semantic of dmu_sync() is that we always push the most recent
2122          * version of the data, so in the face of concurrent updates we may
2123          * see a newer version of the block.  That's OK.
2124          */
2125         ASSERT3U(rbt->bt_txg, >=, wbt->bt_txg);
2126         if (rbt->bt_thread == wbt->bt_thread)
2127                 ASSERT3U(rbt->bt_seq, ==, wbt->bt_seq);
2128         else
2129                 ASSERT3U(rbt->bt_seq, >, wbt->bt_seq);
2130 }
2131 
2132 /*
2133  * Verify that zap_{create,destroy,add,remove,update} work as expected.
2134  */
2135 #define ZTEST_ZAP_MIN_INTS      1
2136 #define ZTEST_ZAP_MAX_INTS      4
2137 #define ZTEST_ZAP_MAX_PROPS     1000
2138 
2139 void
2140 ztest_zap(ztest_args_t *za)
2141 {
2142         objset_t *os = za->za_os;
2143         uint64_t object;
2144         uint64_t txg, last_txg;
2145         uint64_t value[ZTEST_ZAP_MAX_INTS];
2146         uint64_t zl_ints, zl_intsize, prop;
2147         int i, ints;
2148         dmu_tx_t *tx;
2149         char propname[100], txgname[100];
2150         int error;
2151         char osname[MAXNAMELEN];
2152         char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
2153 
2154         dmu_objset_name(os, osname);
2155 
2156         /*
2157          * Create a new object if necessary, and record it in the directory.
2158          */
2159         VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
2160             sizeof (uint64_t), &object));
2161 
2162         if (object == 0) {
2163                 tx = dmu_tx_create(os);
2164                 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff,
2165                     sizeof (uint64_t));
2166                 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL);
2167                 error = dmu_tx_assign(tx, TXG_WAIT);
2168                 if (error) {
2169                         ztest_record_enospc("create zap test obj");
2170                         dmu_tx_abort(tx);
2171                         return;
2172                 }
2173                 object = zap_create(os, DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx);
2174                 if (error) {
2175                         fatal(0, "zap_create('%s', %llu) = %d",
2176                             osname, object, error);
2177                 }
2178                 ASSERT(object != 0);
2179                 dmu_write(os, ZTEST_DIROBJ, za->za_diroff,
2180                     sizeof (uint64_t), &object, tx);
2181                 /*
2182                  * Generate a known hash collision, and verify that
2183                  * we can lookup and remove both entries.
2184                  */
2185                 for (i = 0; i < 2; i++) {
2186                         value[i] = i;
2187                         error = zap_add(os, object, hc[i], sizeof (uint64_t),
2188                             1, &value[i], tx);
2189                         ASSERT3U(error, ==, 0);
2190                 }
2191                 for (i = 0; i < 2; i++) {
2192                         error = zap_add(os, object, hc[i], sizeof (uint64_t),
2193                             1, &value[i], tx);
2194                         ASSERT3U(error, ==, EEXIST);
2195                         error = zap_length(os, object, hc[i],
2196                             &zl_intsize, &zl_ints);
2197                         ASSERT3U(error, ==, 0);
2198                         ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2199                         ASSERT3U(zl_ints, ==, 1);
2200                 }
2201                 for (i = 0; i < 2; i++) {
2202                         error = zap_remove(os, object, hc[i], tx);
2203                         ASSERT3U(error, ==, 0);
2204                 }
2205 
2206                 dmu_tx_commit(tx);
2207         }
2208 
2209         ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
2210 
2211         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
2212         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
2213         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
2214         bzero(value, sizeof (value));
2215         last_txg = 0;
2216 
2217         /*
2218          * If these zap entries already exist, validate their contents.
2219          */
2220         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
2221         if (error == 0) {
2222                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2223                 ASSERT3U(zl_ints, ==, 1);
2224 
2225                 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
2226                     zl_ints, &last_txg) == 0);
2227 
2228                 VERIFY(zap_length(os, object, propname, &zl_intsize,
2229                     &zl_ints) == 0);
2230 
2231                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2232                 ASSERT3U(zl_ints, ==, ints);
2233 
2234                 VERIFY(zap_lookup(os, object, propname, zl_intsize,
2235                     zl_ints, value) == 0);
2236 
2237                 for (i = 0; i < ints; i++) {
2238                         ASSERT3U(value[i], ==, last_txg + object + i);
2239                 }
2240         } else {
2241                 ASSERT3U(error, ==, ENOENT);
2242         }
2243 
2244         /*
2245          * Atomically update two entries in our zap object.
2246          * The first is named txg_%llu, and contains the txg
2247          * in which the property was last updated.  The second
2248          * is named prop_%llu, and the nth element of its value
2249          * should be txg + object + n.
2250          */
2251         tx = dmu_tx_create(os);
2252         dmu_tx_hold_zap(tx, object, TRUE, NULL);
2253         error = dmu_tx_assign(tx, TXG_WAIT);
2254         if (error) {
2255                 ztest_record_enospc("create zap entry");
2256                 dmu_tx_abort(tx);
2257                 return;
2258         }
2259         txg = dmu_tx_get_txg(tx);
2260 
2261         if (last_txg > txg)
2262                 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
2263 
2264         for (i = 0; i < ints; i++)
2265                 value[i] = txg + object + i;
2266 
2267         error = zap_update(os, object, txgname, sizeof (uint64_t), 1, &txg, tx);
2268         if (error)
2269                 fatal(0, "zap_update('%s', %llu, '%s') = %d",
2270                     osname, object, txgname, error);
2271 
2272         error = zap_update(os, object, propname, sizeof (uint64_t),
2273             ints, value, tx);
2274         if (error)
2275                 fatal(0, "zap_update('%s', %llu, '%s') = %d",
2276                     osname, object, propname, error);
2277 
2278         dmu_tx_commit(tx);
2279 
2280         /*
2281          * Remove a random pair of entries.
2282          */
2283         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
2284         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
2285         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
2286 
2287         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
2288 
2289         if (error == ENOENT)
2290                 return;
2291 
2292         ASSERT3U(error, ==, 0);
2293 
2294         tx = dmu_tx_create(os);
2295         dmu_tx_hold_zap(tx, object, TRUE, NULL);
2296         error = dmu_tx_assign(tx, TXG_WAIT);
2297         if (error) {
2298                 ztest_record_enospc("remove zap entry");
2299                 dmu_tx_abort(tx);
2300                 return;
2301         }
2302         error = zap_remove(os, object, txgname, tx);
2303         if (error)
2304                 fatal(0, "zap_remove('%s', %llu, '%s') = %d",
2305                     osname, object, txgname, error);
2306 
2307         error = zap_remove(os, object, propname, tx);
2308         if (error)
2309                 fatal(0, "zap_remove('%s', %llu, '%s') = %d",
2310                     osname, object, propname, error);
2311 
2312         dmu_tx_commit(tx);
2313 
2314         /*
2315          * Once in a while, destroy the object.
2316          */
2317         if (ztest_random(1000) != 0)
2318                 return;
2319 
2320         tx = dmu_tx_create(os);
2321         dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t));
2322         dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
2323         error = dmu_tx_assign(tx, TXG_WAIT);
2324         if (error) {
2325                 ztest_record_enospc("destroy zap object");
2326                 dmu_tx_abort(tx);
2327                 return;
2328         }
2329         error = zap_destroy(os, object, tx);
2330         if (error)
2331                 fatal(0, "zap_destroy('%s', %llu) = %d",
2332                     osname, object, error);
2333         object = 0;
2334         dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t),
2335             &object, tx);
2336         dmu_tx_commit(tx);
2337 }
2338 
2339 void
2340 ztest_zap_parallel(ztest_args_t *za)
2341 {
2342         objset_t *os = za->za_os;
2343         uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
2344         dmu_tx_t *tx;
2345         int i, namelen, error;
2346         char name[20], string_value[20];
2347         void *data;
2348 
2349         /*
2350          * Generate a random name of the form 'xxx.....' where each
2351          * x is a random printable character and the dots are dots.
2352          * There are 94 such characters, and the name length goes from
2353          * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
2354          */
2355         namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
2356 
2357         for (i = 0; i < 3; i++)
2358                 name[i] = '!' + ztest_random('~' - '!' + 1);
2359         for (; i < namelen - 1; i++)
2360                 name[i] = '.';
2361         name[i] = '\0';
2362 
2363         if (ztest_random(2) == 0)
2364                 object = ZTEST_MICROZAP_OBJ;
2365         else
2366                 object = ZTEST_FATZAP_OBJ;
2367 
2368         if ((namelen & 1) || object == ZTEST_MICROZAP_OBJ) {
2369                 wsize = sizeof (txg);
2370                 wc = 1;
2371                 data = &txg;
2372         } else {
2373                 wsize = 1;
2374                 wc = namelen;
2375                 data = string_value;
2376         }
2377 
2378         count = -1ULL;
2379         VERIFY(zap_count(os, object, &count) == 0);
2380         ASSERT(count != -1ULL);
2381 
2382         /*
2383          * Select an operation: length, lookup, add, update, remove.
2384          */
2385         i = ztest_random(5);
2386 
2387         if (i >= 2) {
2388                 tx = dmu_tx_create(os);
2389                 dmu_tx_hold_zap(tx, object, TRUE, NULL);
2390                 error = dmu_tx_assign(tx, TXG_WAIT);
2391                 if (error) {
2392                         ztest_record_enospc("zap parallel");
2393                         dmu_tx_abort(tx);
2394                         return;
2395                 }
2396                 txg = dmu_tx_get_txg(tx);
2397                 bcopy(name, string_value, namelen);
2398         } else {
2399                 tx = NULL;
2400                 txg = 0;
2401                 bzero(string_value, namelen);
2402         }
2403 
2404         switch (i) {
2405 
2406         case 0:
2407                 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
2408                 if (error == 0) {
2409                         ASSERT3U(wsize, ==, zl_wsize);
2410                         ASSERT3U(wc, ==, zl_wc);
2411                 } else {
2412                         ASSERT3U(error, ==, ENOENT);
2413                 }
2414                 break;
2415 
2416         case 1:
2417                 error = zap_lookup(os, object, name, wsize, wc, data);
2418                 if (error == 0) {
2419                         if (data == string_value &&
2420                             bcmp(name, data, namelen) != 0)
2421                                 fatal(0, "name '%s' != val '%s' len %d",
2422                                     name, data, namelen);
2423                 } else {
2424                         ASSERT3U(error, ==, ENOENT);
2425                 }
2426                 break;
2427 
2428         case 2:
2429                 error = zap_add(os, object, name, wsize, wc, data, tx);
2430                 ASSERT(error == 0 || error == EEXIST);
2431                 break;
2432 
2433         case 3:
2434                 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
2435                 break;
2436 
2437         case 4:
2438                 error = zap_remove(os, object, name, tx);
2439                 ASSERT(error == 0 || error == ENOENT);
2440                 break;
2441         }
2442 
2443         if (tx != NULL)
2444                 dmu_tx_commit(tx);
2445 }
2446 
2447 void
2448 ztest_dsl_prop_get_set(ztest_args_t *za)
2449 {
2450         objset_t *os = za->za_os;
2451         int i, inherit;
2452         uint64_t value;
2453         const char *prop, *valname;
2454         char setpoint[MAXPATHLEN];
2455         char osname[MAXNAMELEN];
2456         int error;
2457 
2458         (void) rw_rdlock(&ztest_shared->zs_name_lock);
2459 
2460         dmu_objset_name(os, osname);
2461 
2462         for (i = 0; i < 3; i++) {
2463                 if (i == 0) {
2464                         prop = "checksum";
2465                         value = ztest_random_checksum();
2466                         inherit = (value == ZIO_CHECKSUM_INHERIT);
2467                 } else if (i == 1) {
2468                         prop = "compression";
2469                         value = ztest_random_compress();
2470                         inherit = (value == ZIO_COMPRESS_INHERIT);
2471                 } else {
2472                         prop = "crypt";
2473                         value = ztest_random_crypt();
2474                         inherit = (value == ZIO_CRYPT_INHERIT);
2475                 }
2476 
2477                 error = dsl_prop_set(osname, prop, sizeof (value),
2478                     !inherit, &value);
2479 
2480                 if (error == ENOSPC) {
2481                         ztest_record_enospc("dsl_prop_set");
2482                         break;
2483                 }
2484 
2485                 ASSERT3U(error, ==, 0);
2486 
2487                 VERIFY3U(dsl_prop_get(osname, prop, sizeof (value),
2488                     1, &value, setpoint), ==, 0);
2489 
2490                 if (i == 0) {
2491                         valname = zio_checksum_table[value].ci_name;
2492                 } else if (i == 1) {
2493                         valname = zio_compress_table[value].ci_name;
2494                 } else {
2495                         valname = zio_crypt_table[value].ci_name;
2496                 }
2497 
2498 
2499                 if (zopt_verbose >= 6) {
2500                         (void) printf("%s %s = %s for '%s'\n",
2501                             osname, prop, valname, setpoint);
2502                 }
2503         }
2504 
2505         (void) rw_unlock(&ztest_shared->zs_name_lock);
2506 }
2507 
2508 /*
2509  * Inject random faults into the on-disk data.
2510  */
2511 void
2512 ztest_fault_inject(ztest_args_t *za)
2513 {
2514         int fd;
2515         uint64_t offset;
2516         uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
2517         uint64_t bad = 0x1990c0ffeedecade;
2518         uint64_t top, leaf;
2519         char path0[MAXPATHLEN];
2520         char pathrand[MAXPATHLEN];
2521         size_t fsize;
2522         spa_t *spa = za->za_spa;
2523         int bshift = SPA_MAXBLOCKSHIFT + 2;     /* don't scrog all labels */
2524         int iters = 1000;
2525         int maxfaults = zopt_maxfaults;
2526         vdev_t *vd0 = NULL;
2527         uint64_t guid0 = 0;
2528 
2529         ASSERT(leaves >= 1);
2530 
2531         /*
2532          * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
2533          */
2534         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
2535 
2536         if (ztest_random(2) == 0) {
2537                 /*
2538                  * Inject errors on a normal data device.
2539                  */
2540                 top = ztest_random(spa->spa_root_vdev->vdev_children);
2541                 leaf = ztest_random(leaves);
2542 
2543                 /*
2544                  * Generate paths to the first leaf in this top-level vdev,
2545                  * and to the random leaf we selected.  We'll induce transient
2546                  * write failures and random online/offline activity on leaf 0,
2547                  * and we'll write random garbage to the randomly chosen leaf.
2548                  */
2549                 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
2550                     zopt_dir, zopt_pool, top * leaves + 0);
2551                 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
2552                     zopt_dir, zopt_pool, top * leaves + leaf);
2553 
2554                 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
2555                 if (vd0 != NULL && maxfaults != 1) {
2556                         /*
2557                          * Make vd0 explicitly claim to be unreadable,
2558                          * or unwriteable, or reach behind its back
2559                          * and close the underlying fd.  We can do this if
2560                          * maxfaults == 0 because we'll fail and reexecute,
2561                          * and we can do it if maxfaults >= 2 because we'll
2562                          * have enough redundancy.  If maxfaults == 1, the
2563                          * combination of this with injection of random data
2564                          * corruption below exceeds the pool's fault tolerance.
2565                          */
2566                         vdev_file_t *vf = vd0->vdev_tsd;
2567 
2568                         if (vf != NULL && ztest_random(3) == 0) {
2569                                 (void) close(vf->vf_vnode->v_fd);
2570                                 vf->vf_vnode->v_fd = -1;
2571                         } else if (ztest_random(2) == 0) {
2572                                 vd0->vdev_cant_read = B_TRUE;
2573                         } else {
2574                                 vd0->vdev_cant_write = B_TRUE;
2575                         }
2576                         guid0 = vd0->vdev_guid;
2577                 }
2578         } else {
2579                 /*
2580                  * Inject errors on an l2cache device.
2581                  */
2582                 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2583 
2584                 if (sav->sav_count == 0) {
2585                         spa_config_exit(spa, SCL_STATE, FTAG);
2586                         return;
2587                 }
2588                 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
2589                 guid0 = vd0->vdev_guid;
2590                 (void) strcpy(path0, vd0->vdev_path);
2591                 (void) strcpy(pathrand, vd0->vdev_path);
2592 
2593                 leaf = 0;
2594                 leaves = 1;
2595                 maxfaults = INT_MAX;    /* no limit on cache devices */
2596         }
2597 
2598         spa_config_exit(spa, SCL_STATE, FTAG);
2599 
2600         if (maxfaults == 0)
2601                 return;
2602 
2603         /*
2604          * If we can tolerate two or more faults, randomly online/offline vd0.
2605          */
2606         if (maxfaults >= 2 && guid0 != 0) {
2607                 if (ztest_random(10) < 6) {
2608                         int flags = (ztest_random(2) == 0 ?
2609                             ZFS_OFFLINE_TEMPORARY : 0);
2610                         VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
2611                 } else {
2612                         (void) vdev_online(spa, guid0, 0, NULL);
2613                 }
2614         }
2615 
2616         /*
2617          * We have at least single-fault tolerance, so inject data corruption.
2618          */
2619         fd = open(pathrand, O_RDWR);
2620 
2621         if (fd == -1)   /* we hit a gap in the device namespace */
2622                 return;
2623 
2624         fsize = lseek(fd, 0, SEEK_END);
2625 
2626         while (--iters != 0) {
2627                 offset = ztest_random(fsize / (leaves << bshift)) *
2628                     (leaves << bshift) + (leaf << bshift) +
2629                     (ztest_random(1ULL << (bshift - 1)) & -8ULL);
2630 
2631                 if (offset >= fsize)
2632                         continue;
2633 
2634                 if (zopt_verbose >= 6)
2635                         (void) printf("injecting bad word into %s,"
2636                             " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
2637 
2638                 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
2639                         fatal(1, "can't inject bad word at 0x%llx in %s",
2640                             offset, pathrand);
2641         }
2642 
2643         (void) close(fd);
2644 }
2645 
2646 /*
2647  * Scrub the pool.
2648  */
2649 void
2650 ztest_scrub(ztest_args_t *za)
2651 {
2652         spa_t *spa = za->za_spa;
2653 
2654         (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING);
2655         (void) poll(NULL, 0, 1000); /* wait a second, then force a restart */
2656         (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING);
2657 }
2658 
2659 /*
2660  * Rename the pool to a different name and then rename it back.
2661  */
2662 void
2663 ztest_spa_rename(ztest_args_t *za)
2664 {
2665         char *oldname, *newname;
2666         int error;
2667         spa_t *spa;
2668 
2669         (void) rw_wrlock(&ztest_shared->zs_name_lock);
2670 
2671         oldname = za->za_pool;
2672         newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
2673         (void) strcpy(newname, oldname);
2674         (void) strcat(newname, "_tmp");
2675 
2676         /*
2677          * Do the rename
2678          */
2679         error = spa_rename(oldname, newname);
2680         if (error)
2681                 fatal(0, "spa_rename('%s', '%s') = %d", oldname,
2682                     newname, error);
2683 
2684         /*
2685          * Try to open it under the old name, which shouldn't exist
2686          */
2687         error = spa_open(oldname, &spa, FTAG);
2688         if (error != ENOENT)
2689                 fatal(0, "spa_open('%s') = %d", oldname, error);
2690 
2691         /*
2692          * Open it under the new name and make sure it's still the same spa_t.
2693          */
2694         error = spa_open(newname, &spa, FTAG);
2695         if (error != 0)
2696                 fatal(0, "spa_open('%s') = %d", newname, error);
2697 
2698         ASSERT(spa == za->za_spa);
2699         spa_close(spa, FTAG);
2700 
2701         /*
2702          * Rename it back to the original
2703          */
2704         error = spa_rename(newname, oldname);
2705         if (error)
2706                 fatal(0, "spa_rename('%s', '%s') = %d", newname,
2707                     oldname, error);
2708 
2709         /*
2710          * Make sure it can still be opened
2711          */
2712         error = spa_open(oldname, &spa, FTAG);
2713         if (error != 0)
2714                 fatal(0, "spa_open('%s') = %d", oldname, error);
2715 
2716         ASSERT(spa == za->za_spa);
2717         spa_close(spa, FTAG);
2718 
2719         umem_free(newname, strlen(newname) + 1);
2720 
2721         (void) rw_unlock(&ztest_shared->zs_name_lock);
2722 }
2723 
2724 
2725 /*
2726  * Completely obliterate one disk.
2727  */
2728 static void
2729 ztest_obliterate_one_disk(uint64_t vdev)
2730 {
2731         int fd;
2732         char dev_name[MAXPATHLEN], copy_name[MAXPATHLEN];
2733         size_t fsize;
2734 
2735         if (zopt_maxfaults < 2)
2736                 return;
2737 
2738         (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
2739         (void) snprintf(copy_name, MAXPATHLEN, "%s.old", dev_name);
2740 
2741         fd = open(dev_name, O_RDWR);
2742 
2743         if (fd == -1)
2744                 fatal(1, "can't open %s", dev_name);
2745 
2746         /*
2747          * Determine the size.
2748          */
2749         fsize = lseek(fd, 0, SEEK_END);
2750 
2751         (void) close(fd);
2752 
2753         /*
2754          * Rename the old device to dev_name.old (useful for debugging).
2755          */
2756         VERIFY(rename(dev_name, copy_name) == 0);
2757 
2758         /*
2759          * Create a new one.
2760          */
2761         VERIFY((fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666)) >= 0);
2762         VERIFY(ftruncate(fd, fsize) == 0);
2763         (void) close(fd);
2764 }
2765 
2766 static void
2767 ztest_replace_one_disk(spa_t *spa, uint64_t vdev)
2768 {
2769         char dev_name[MAXPATHLEN];
2770         nvlist_t *root;
2771         int error;
2772         uint64_t guid;
2773         vdev_t *vd;
2774 
2775         (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
2776 
2777         /*
2778          * Build the nvlist describing dev_name.
2779          */
2780         root = make_vdev_root(dev_name, NULL, 0, 0, 0, 0, 0, 1);
2781 
2782         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2783         if ((vd = vdev_lookup_by_path(spa->spa_root_vdev, dev_name)) == NULL)
2784                 guid = 0;
2785         else
2786                 guid = vd->vdev_guid;
2787         spa_config_exit(spa, SCL_VDEV, FTAG);
2788         error = spa_vdev_attach(spa, guid, root, B_TRUE);
2789         if (error != 0 &&
2790             error != EBUSY &&
2791             error != ENOTSUP &&
2792             error != ENODEV &&
2793             error != EDOM)
2794                 fatal(0, "spa_vdev_attach(in-place) = %d", error);
2795 
2796         nvlist_free(root);
2797 }
2798 
2799 static void
2800 ztest_verify_blocks(char *pool)
2801 {
2802         int status;
2803         char zdb[MAXPATHLEN + MAXNAMELEN + 20];
2804         char zbuf[1024];
2805         char *bin;
2806         char *ztest;
2807         char *isa;
2808         int isalen;
2809         FILE *fp;
2810 
2811         (void) realpath(getexecname(), zdb);
2812 
2813         /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
2814         bin = strstr(zdb, "/usr/bin/");
2815         ztest = strstr(bin, "/ztest");
2816         isa = bin + 8;
2817         isalen = ztest - isa;
2818         isa = strdup(isa);
2819         /* LINTED */
2820         (void) sprintf(bin,
2821             "/usr/sbin%.*s/zdb -bc%s%s -U /tmp/zpool.cache %s",
2822             isalen,
2823             isa,
2824             zopt_verbose >= 3 ? "s" : "",
2825             zopt_verbose >= 4 ? "v" : "",
2826             pool);
2827         free(isa);
2828 
2829         if (zopt_verbose >= 5)
2830                 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
2831 
2832         fp = popen(zdb, "r");
2833 
2834         while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
2835                 if (zopt_verbose >= 3)
2836                         (void) printf("%s", zbuf);
2837 
2838         status = pclose(fp);
2839 
2840         if (status == 0)
2841                 return;
2842 
2843         ztest_dump_core = 0;
2844         if (WIFEXITED(status))
2845                 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
2846         else
2847                 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
2848 }
2849 
2850 static void
2851 ztest_walk_pool_directory(char *header)
2852 {
2853         spa_t *spa = NULL;
2854 
2855         if (zopt_verbose >= 6)
2856                 (void) printf("%s\n", header);
2857 
2858         mutex_enter(&spa_namespace_lock);
2859         while ((spa = spa_next(spa)) != NULL)
2860                 if (zopt_verbose >= 6)
2861                         (void) printf("\t%s\n", spa_name(spa));
2862         mutex_exit(&spa_namespace_lock);
2863 }
2864 
2865 static void
2866 ztest_spa_import_export(char *oldname, char *newname)
2867 {
2868         nvlist_t *config, *newconfig;
2869         uint64_t pool_guid;
2870         spa_t *spa;
2871         int error;
2872 
2873         if (zopt_verbose >= 4) {
2874                 (void) printf("import/export: old = %s, new = %s\n",
2875                     oldname, newname);
2876         }
2877 
2878         /*
2879          * Clean up from previous runs.
2880          */
2881         (void) spa_destroy(newname);
2882 
2883         /*
2884          * Get the pool's configuration and guid.
2885          */
2886         error = spa_open(oldname, &spa, FTAG);
2887         if (error)
2888                 fatal(0, "spa_open('%s') = %d", oldname, error);
2889 
2890         /*
2891          * Kick off a scrub to tickle scrub/export races.
2892          */
2893         if (ztest_random(2) == 0)
2894                 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING);
2895 
2896         pool_guid = spa_guid(spa);
2897         spa_close(spa, FTAG);
2898 
2899         ztest_walk_pool_directory("pools before export");
2900 
2901         /*
2902          * Export it.
2903          */
2904         error = spa_export(oldname, &config, B_FALSE, B_FALSE);
2905         if (error)
2906                 fatal(0, "spa_export('%s') = %d", oldname, error);
2907 
2908         ztest_walk_pool_directory("pools after export");
2909 
2910         /*
2911          * Try to import it.
2912          */
2913         newconfig = spa_tryimport(config);
2914         ASSERT(newconfig != NULL);
2915         nvlist_free(newconfig);
2916 
2917         /*
2918          * Import it under the new name.
2919          */
2920         error = spa_import(newname, config, NULL);
2921         if (error)
2922                 fatal(0, "spa_import('%s') = %d", newname, error);
2923 
2924         ztest_walk_pool_directory("pools after import");
2925 
2926         /*
2927          * Try to import it again -- should fail with EEXIST.
2928          */
2929         error = spa_import(newname, config, NULL);
2930         if (error != EEXIST)
2931                 fatal(0, "spa_import('%s') twice", newname);
2932 
2933         /*
2934          * Try to import it under a different name -- should fail with EEXIST.
2935          */
2936         error = spa_import(oldname, config, NULL);
2937         if (error != EEXIST)
2938                 fatal(0, "spa_import('%s') under multiple names", newname);
2939 
2940         /*
2941          * Verify that the pool is no longer visible under the old name.
2942          */
2943         error = spa_open(oldname, &spa, FTAG);
2944         if (error != ENOENT)
2945                 fatal(0, "spa_open('%s') = %d", newname, error);
2946 
2947         /*
2948          * Verify that we can open and close the pool using the new name.
2949          */
2950         error = spa_open(newname, &spa, FTAG);
2951         if (error)
2952                 fatal(0, "spa_open('%s') = %d", newname, error);
2953         ASSERT(pool_guid == spa_guid(spa));
2954         spa_close(spa, FTAG);
2955 
2956         nvlist_free(config);
2957 }
2958 
2959 static void
2960 ztest_resume(spa_t *spa)
2961 {
2962         if (spa_suspended(spa)) {
2963                 spa_vdev_state_enter(spa);
2964                 vdev_clear(spa, NULL);
2965                 (void) spa_vdev_state_exit(spa, NULL, 0);
2966                 zio_resume(spa);
2967         }
2968 }
2969 
2970 static void *
2971 ztest_resume_thread(void *arg)
2972 {
2973         spa_t *spa = arg;
2974 
2975         while (!ztest_exiting) {
2976                 (void) poll(NULL, 0, 1000);
2977                 ztest_resume(spa);
2978         }
2979         return (NULL);
2980 }
2981 
2982 static void *
2983 ztest_thread(void *arg)
2984 {
2985         ztest_args_t *za = arg;
2986         ztest_shared_t *zs = ztest_shared;
2987         hrtime_t now, functime;
2988         ztest_info_t *zi;
2989         int f, i;
2990 
2991         while ((now = gethrtime()) < za->za_stop) {
2992                 /*
2993                  * See if it's time to force a crash.
2994                  */
2995                 if (now > za->za_kill) {
2996                         zs->zs_alloc = spa_get_alloc(za->za_spa);
2997                         zs->zs_space = spa_get_space(za->za_spa);
2998                         (void) kill(getpid(), SIGKILL);
2999                 }
3000 
3001                 /*
3002                  * Pick a random function.
3003                  */
3004                 f = ztest_random(ZTEST_FUNCS);
3005                 zi = &zs->zs_info[f];
3006 
3007                 /*
3008                  * Decide whether to call it, based on the requested frequency.
3009                  */
3010                 if (zi->zi_call_target == 0 ||
3011                     (double)zi->zi_call_total / zi->zi_call_target >
3012                     (double)(now - zs->zs_start_time) / (zopt_time * NANOSEC))
3013                         continue;
3014 
3015                 atomic_add_64(&zi->zi_calls, 1);
3016                 atomic_add_64(&zi->zi_call_total, 1);
3017 
3018                 za->za_diroff = (za->za_instance * ZTEST_FUNCS + f) *
3019                     ZTEST_DIRSIZE;
3020                 za->za_diroff_shared = (1ULL << 63);
3021 
3022                 for (i = 0; i < zi->zi_iters; i++)
3023                         zi->zi_func(za);
3024 
3025                 functime = gethrtime() - now;
3026 
3027                 atomic_add_64(&zi->zi_call_time, functime);
3028 
3029                 if (zopt_verbose >= 4) {
3030                         Dl_info dli;
3031                         (void) dladdr((void *)zi->zi_func, &dli);
3032                         (void) printf("%6.2f sec in %s\n",
3033                             (double)functime / NANOSEC, dli.dli_sname);
3034                 }
3035 
3036                 /*
3037                  * If we're getting ENOSPC with some regularity, stop.
3038                  */
3039                 if (zs->zs_enospc_count > 10)
3040                         break;
3041         }
3042 
3043         return (NULL);
3044 }
3045 
3046 /*
3047  * Kick off threads to run tests on all datasets in parallel.
3048  */
3049 static void
3050 ztest_run(char *pool)
3051 {
3052         int t, d, error;
3053         ztest_shared_t *zs = ztest_shared;
3054         ztest_args_t *za;
3055         spa_t *spa;
3056         char name[100];
3057         thread_t resume_tid;
3058 
3059         ztest_exiting = B_FALSE;
3060 
3061         (void) _mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL);
3062         (void) rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL);
3063 
3064         for (t = 0; t < ZTEST_SYNC_LOCKS; t++)
3065                 (void) _mutex_init(&zs->zs_sync_lock[t], USYNC_THREAD, NULL);
3066 
3067         /*
3068          * Destroy one disk before we even start.
3069          * It's mirrored, so everything should work just fine.
3070          * This makes us exercise fault handling very early in spa_load().
3071          */
3072         ztest_obliterate_one_disk(0);
3073 
3074         /*
3075          * Verify that the sum of the sizes of all blocks in the pool
3076          * equals the SPA's allocated space total.
3077          */
3078         ztest_verify_blocks(pool);
3079 
3080         /*
3081          * Kick off a replacement of the disk we just obliterated.
3082          */
3083         kernel_init(FREAD | FWRITE);
3084         VERIFY(spa_open(pool, &spa, FTAG) == 0);
3085         ztest_replace_one_disk(spa, 0);
3086         if (zopt_verbose >= 5)
3087                 show_pool_stats(spa);
3088         spa_close(spa, FTAG);
3089         kernel_fini();
3090 
3091         kernel_init(FREAD | FWRITE);
3092 
3093         /*
3094          * Verify that we can export the pool and reimport it under a
3095          * different name.
3096          */
3097         if (ztest_random(2) == 0) {
3098                 (void) snprintf(name, 100, "%s_import", pool);
3099                 ztest_spa_import_export(pool, name);
3100                 ztest_spa_import_export(name, pool);
3101         }
3102 
3103         /*
3104          * Verify that we can loop over all pools.
3105          */
3106         mutex_enter(&spa_namespace_lock);
3107         for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) {
3108                 if (zopt_verbose > 3) {
3109                         (void) printf("spa_next: found %s\n", spa_name(spa));
3110                 }
3111         }
3112         mutex_exit(&spa_namespace_lock);
3113 
3114         /*
3115          * Open our pool.
3116          */
3117         VERIFY(spa_open(pool, &spa, FTAG) == 0);
3118 
3119         /*
3120          * We don't expect the pool to suspend unless maxfaults == 0,
3121          * in which case ztest_fault_inject() temporarily takes away
3122          * the only valid replica.
3123          */
3124         if (zopt_maxfaults == 0)
3125                 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
3126         else
3127                 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
3128 
3129         /*
3130          * Create a thread to periodically resume suspended I/O.
3131          */
3132         VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
3133             &resume_tid) == 0);
3134 
3135         /*
3136          * Verify that we can safely inquire about about any object,
3137          * whether it's allocated or not.  To make it interesting,
3138          * we probe a 5-wide window around each power of two.
3139          * This hits all edge cases, including zero and the max.
3140          */
3141         for (t = 0; t < 64; t++) {
3142                 for (d = -5; d <= 5; d++) {
3143                         error = dmu_object_info(spa->spa_meta_objset,
3144                             (1ULL << t) + d, NULL);
3145                         ASSERT(error == 0 || error == ENOENT ||
3146                             error == EINVAL);
3147                 }
3148         }
3149 
3150         /*
3151          * Now kick off all the tests that run in parallel.
3152          */
3153         zs->zs_enospc_count = 0;
3154 
3155         za = umem_zalloc(zopt_threads * sizeof (ztest_args_t), UMEM_NOFAIL);
3156 
3157         if (zopt_verbose >= 4)
3158                 (void) printf("starting main threads...\n");
3159 
3160         za[0].za_start = gethrtime();
3161         za[0].za_stop = za[0].za_start + zopt_passtime * NANOSEC;
3162         za[0].za_stop = MIN(za[0].za_stop, zs->zs_stop_time);
3163         za[0].za_kill = za[0].za_stop;
3164         if (ztest_random(100) < zopt_killrate)
3165                 za[0].za_kill -= ztest_random(zopt_passtime * NANOSEC);
3166 
3167         for (t = 0; t < zopt_threads; t++) {
3168                 d = t % zopt_datasets;
3169 
3170                 (void) strcpy(za[t].za_pool, pool);
3171                 za[t].za_os = za[d].za_os;
3172                 za[t].za_spa = spa;
3173                 za[t].za_zilog = za[d].za_zilog;
3174                 za[t].za_instance = t;
3175                 za[t].za_random = ztest_random(-1ULL);
3176                 za[t].za_start = za[0].za_start;
3177                 za[t].za_stop = za[0].za_stop;
3178                 za[t].za_kill = za[0].za_kill;
3179 
3180                 if (t < zopt_datasets) {
3181                         int test_future = FALSE;
3182                         (void) rw_rdlock(&ztest_shared->zs_name_lock);
3183                         (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d);
3184                         error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0,
3185                             ztest_create_cb, NULL);
3186                         if (error == EEXIST) {
3187                                 test_future = TRUE;
3188                         } else if (error == ENOSPC) {
3189                                 zs->zs_enospc_count++;
3190                                 (void) rw_unlock(&ztest_shared->zs_name_lock);
3191                                 break;
3192                         } else if (error != 0) {
3193                                 fatal(0, "dmu_objset_create(%s) = %d",
3194                                     name, error);
3195                         }
3196                         error = dmu_objset_open(name, DMU_OST_OTHER,
3197                             DS_MODE_USER, &za[d].za_os);
3198                         if (error)
3199                                 fatal(0, "dmu_objset_open('%s') = %d",
3200                                     name, error);
3201                         (void) rw_unlock(&ztest_shared->zs_name_lock);
3202                         if (test_future)
3203                                 ztest_dmu_check_future_leak(&za[t]);
3204                         zil_replay(za[d].za_os, za[d].za_os,
3205                             ztest_replay_vector);
3206                         za[d].za_zilog = zil_open(za[d].za_os, NULL);
3207                 }
3208 
3209                 VERIFY(thr_create(0, 0, ztest_thread, &za[t], THR_BOUND,
3210                     &za[t].za_thread) == 0);
3211         }
3212 
3213         while (--t >= 0) {
3214                 VERIFY(thr_join(za[t].za_thread, NULL, NULL) == 0);
3215                 if (t < zopt_datasets) {
3216                         zil_close(za[t].za_zilog);
3217                         dmu_objset_close(za[t].za_os);
3218                 }
3219         }
3220 
3221         if (zopt_verbose >= 3)
3222                 show_pool_stats(spa);
3223 
3224         txg_wait_synced(spa_get_dsl(spa), 0);
3225 
3226         zs->zs_alloc = spa_get_alloc(spa);
3227         zs->zs_space = spa_get_space(spa);
3228 
3229         /*
3230          * If we had out-of-space errors, destroy a random objset.
3231          */
3232         if (zs->zs_enospc_count != 0) {
3233                 (void) rw_rdlock(&ztest_shared->zs_name_lock);
3234                 d = (int)ztest_random(zopt_datasets);
3235                 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d);
3236                 if (zopt_verbose >= 3)
3237                         (void) printf("Destroying %s to free up space\n", name);
3238                 (void) dmu_objset_find(name, ztest_destroy_cb, &za[d],
3239                     DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
3240                 (void) rw_unlock(&ztest_shared->zs_name_lock);
3241         }
3242 
3243         txg_wait_synced(spa_get_dsl(spa), 0);
3244 
3245         umem_free(za, zopt_threads * sizeof (ztest_args_t));
3246 
3247         /* Kill the resume thread */
3248         ztest_exiting = B_TRUE;
3249         VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
3250         ztest_resume(spa);
3251 
3252         /*
3253          * Right before closing the pool, kick off a bunch of async I/O;
3254          * spa_close() should wait for it to complete.
3255          */
3256         for (t = 1; t < 50; t++)
3257                 dmu_prefetch(spa->spa_meta_objset, t, 0, 1 << 15);
3258 
3259         spa_close(spa, FTAG);
3260 
3261         kernel_fini();
3262 }
3263 
3264 void
3265 print_time(hrtime_t t, char *timebuf)
3266 {
3267         hrtime_t s = t / NANOSEC;
3268         hrtime_t m = s / 60;
3269         hrtime_t h = m / 60;
3270         hrtime_t d = h / 24;
3271 
3272         s -= m * 60;
3273         m -= h * 60;
3274         h -= d * 24;
3275 
3276         timebuf[0] = '\0';
3277 
3278         if (d)
3279                 (void) sprintf(timebuf,
3280                     "%llud%02lluh%02llum%02llus", d, h, m, s);
3281         else if (h)
3282                 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
3283         else if (m)
3284                 (void) sprintf(timebuf, "%llum%02llus", m, s);
3285         else
3286                 (void) sprintf(timebuf, "%llus", s);
3287 }
3288 
3289 /*
3290  * Create a storage pool with the given name and initial vdev size.
3291  * Then create the specified number of datasets in the pool.
3292  */
3293 static void
3294 ztest_init(char *pool)
3295 {
3296         spa_t *spa;
3297         int error;
3298         nvlist_t *nvroot;
3299 
3300         kernel_init(FREAD | FWRITE);
3301 
3302         /*
3303          * Create the storage pool.
3304          */
3305         (void) spa_destroy(pool);
3306         ztest_shared->zs_vdev_primaries = 0;
3307         nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0,
3308             0, zopt_raidz, zopt_mirrors, 1);
3309         error = spa_create(pool, nvroot, NULL, NULL, NULL);
3310         nvlist_free(nvroot);
3311 
3312         if (error)
3313                 fatal(0, "spa_create() = %d", error);
3314         error = spa_open(pool, &spa, FTAG);
3315         if (error)
3316                 fatal(0, "spa_open() = %d", error);
3317 
3318         if (zopt_verbose >= 3)
3319                 show_pool_stats(spa);
3320 
3321         spa_close(spa, FTAG);
3322 
3323         kernel_fini();
3324 }
3325 
3326 int
3327 main(int argc, char **argv)
3328 {
3329         int kills = 0;
3330         int iters = 0;
3331         int i, f;
3332         ztest_shared_t *zs;
3333         ztest_info_t *zi;
3334         char timebuf[100];
3335         char numbuf[6];
3336 
3337         (void) setvbuf(stdout, NULL, _IOLBF, 0);
3338 
3339         /* Override location of zpool.cache */
3340         spa_config_path = "/tmp/zpool.cache";
3341 
3342         ztest_random_fd = open("/dev/urandom", O_RDONLY);
3343 
3344         process_options(argc, argv);
3345 
3346         /*
3347          * Blow away any existing copy of zpool.cache
3348          */
3349         if (zopt_init != 0)
3350                 (void) remove("/tmp/zpool.cache");
3351 
3352         zs = ztest_shared = (void *)mmap(0,
3353             P2ROUNDUP(sizeof (ztest_shared_t), getpagesize()),
3354             PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
3355 
3356         if (zopt_verbose >= 1) {
3357                 (void) printf("%llu vdevs, %d datasets, %d threads,"
3358                     " %llu seconds...\n",
3359                     (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads,
3360                     (u_longlong_t)zopt_time);
3361         }
3362 
3363         /*
3364          * Create and initialize our storage pool.
3365          */
3366         for (i = 1; i <= zopt_init; i++) {
3367                 bzero(zs, sizeof (ztest_shared_t));
3368                 if (zopt_verbose >= 3 && zopt_init != 1)
3369                         (void) printf("ztest_init(), pass %d\n", i);
3370                 ztest_init(zopt_pool);
3371         }
3372 
3373         /*
3374          * Initialize the call targets for each function.
3375          */
3376         for (f = 0; f < ZTEST_FUNCS; f++) {
3377                 zi = &zs->zs_info[f];
3378 
3379                 *zi = ztest_info[f];
3380 
3381                 if (*zi->zi_interval == 0)
3382                         zi->zi_call_target = UINT64_MAX;
3383                 else
3384                         zi->zi_call_target = zopt_time / *zi->zi_interval;
3385         }
3386 
3387         zs->zs_start_time = gethrtime();
3388         zs->zs_stop_time = zs->zs_start_time + zopt_time * NANOSEC;
3389 
3390         /*
3391          * Run the tests in a loop.  These tests include fault injection
3392          * to verify that self-healing data works, and forced crashes
3393          * to verify that we never lose on-disk consistency.
3394          */
3395         while (gethrtime() < zs->zs_stop_time) {
3396                 int status;
3397                 pid_t pid;
3398                 char *tmp;
3399 
3400                 /*
3401                  * Initialize the workload counters for each function.
3402                  */
3403                 for (f = 0; f < ZTEST_FUNCS; f++) {
3404                         zi = &zs->zs_info[f];
3405                         zi->zi_calls = 0;
3406                         zi->zi_call_time = 0;
3407                 }
3408 
3409                 pid = fork();
3410 
3411                 if (pid == -1)
3412                         fatal(1, "fork failed");
3413 
3414                 if (pid == 0) { /* child */
3415                         struct rlimit rl = { 1024, 1024 };
3416                         (void) setrlimit(RLIMIT_NOFILE, &rl);
3417                         (void) enable_extended_FILE_stdio(-1, -1);
3418                         ztest_run(zopt_pool);
3419                         exit(0);
3420                 }
3421 
3422                 while (waitpid(pid, &status, 0) != pid)
3423                         continue;
3424 
3425                 if (WIFEXITED(status)) {
3426                         if (WEXITSTATUS(status) != 0) {
3427                                 (void) fprintf(stderr,
3428                                     "child exited with code %d\n",
3429                                     WEXITSTATUS(status));
3430                                 exit(2);
3431                         }
3432                 } else if (WIFSIGNALED(status)) {
3433                         if (WTERMSIG(status) != SIGKILL) {
3434                                 (void) fprintf(stderr,
3435                                     "child died with signal %d\n",
3436                                     WTERMSIG(status));
3437                                 exit(3);
3438                         }
3439                         kills++;
3440                 } else {
3441                         (void) fprintf(stderr, "something strange happened "
3442                             "to child\n");
3443                         exit(4);
3444                 }
3445 
3446                 iters++;
3447 
3448                 if (zopt_verbose >= 1) {
3449                         hrtime_t now = gethrtime();
3450 
3451                         now = MIN(now, zs->zs_stop_time);
3452                         print_time(zs->zs_stop_time - now, timebuf);
3453                         nicenum(zs->zs_space, numbuf);
3454 
3455                         (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
3456                             "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
3457                             iters,
3458                             WIFEXITED(status) ? "Complete" : "SIGKILL",
3459                             (u_longlong_t)zs->zs_enospc_count,
3460                             100.0 * zs->zs_alloc / zs->zs_space,
3461                             numbuf,
3462                             100.0 * (now - zs->zs_start_time) /
3463                             (zopt_time * NANOSEC), timebuf);
3464                 }
3465 
3466                 if (zopt_verbose >= 2) {
3467                         (void) printf("\nWorkload summary:\n\n");
3468                         (void) printf("%7s %9s   %s\n",
3469                             "Calls", "Time", "Function");
3470                         (void) printf("%7s %9s   %s\n",
3471                             "-----", "----", "--------");
3472                         for (f = 0; f < ZTEST_FUNCS; f++) {
3473                                 Dl_info dli;
3474 
3475                                 zi = &zs->zs_info[f];
3476                                 print_time(zi->zi_call_time, timebuf);
3477                                 (void) dladdr((void *)zi->zi_func, &dli);
3478                                 (void) printf("%7llu %9s   %s\n",
3479                                     (u_longlong_t)zi->zi_calls, timebuf,
3480                                     dli.dli_sname);
3481                         }
3482                         (void) printf("\n");
3483                 }
3484 
3485                 /*
3486                  * It's possible that we killed a child during a rename test, in
3487                  * which case we'll have a 'ztest_tmp' pool lying around instead
3488                  * of 'ztest'.  Do a blind rename in case this happened.
3489                  */
3490                 tmp = umem_alloc(strlen(zopt_pool) + 5, UMEM_NOFAIL);
3491                 (void) strcpy(tmp, zopt_pool);
3492                 (void) strcat(tmp, "_tmp");
3493                 kernel_init(FREAD | FWRITE);
3494                 (void) spa_rename(tmp, zopt_pool);
3495                 kernel_fini();
3496                 umem_free(tmp, strlen(tmp) + 1);
3497         }
3498 
3499         ztest_verify_blocks(zopt_pool);
3500 
3501         if (zopt_verbose >= 1) {
3502                 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
3503                     kills, iters - kills, (100.0 * kills) / MAX(1, iters));
3504         }
3505 
3506         return (0);
3507 }