1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2012 by Delphix. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  26  */
  27 
  28 /*
  29  * The objective of this program is to provide a DMU/ZAP/SPA stress test
  30  * that runs entirely in userland, is easy to use, and easy to extend.
  31  *
  32  * The overall design of the ztest program is as follows:
  33  *
  34  * (1) For each major functional area (e.g. adding vdevs to a pool,
  35  *     creating and destroying datasets, reading and writing objects, etc)
  36  *     we have a simple routine to test that functionality.  These
  37  *     individual routines do not have to do anything "stressful".
  38  *
  39  * (2) We turn these simple functionality tests into a stress test by
  40  *     running them all in parallel, with as many threads as desired,
  41  *     and spread across as many datasets, objects, and vdevs as desired.
  42  *
  43  * (3) While all this is happening, we inject faults into the pool to
  44  *     verify that self-healing data really works.
  45  *
  46  * (4) Every time we open a dataset, we change its checksum and compression
  47  *     functions.  Thus even individual objects vary from block to block
  48  *     in which checksum they use and whether they're compressed.
  49  *
  50  * (5) To verify that we never lose on-disk consistency after a crash,
  51  *     we run the entire test in a child of the main process.
  52  *     At random times, the child self-immolates with a SIGKILL.
  53  *     This is the software equivalent of pulling the power cord.
  54  *     The parent then runs the test again, using the existing
  55  *     storage pool, as many times as desired. If backwards compatability
  56  *     testing is enabled ztest will sometimes run the "older" version
  57  *     of ztest after a SIGKILL.
  58  *
  59  * (6) To verify that we don't have future leaks or temporal incursions,
  60  *     many of the functional tests record the transaction group number
  61  *     as part of their data.  When reading old data, they verify that
  62  *     the transaction group number is less than the current, open txg.
  63  *     If you add a new test, please do this if applicable.
  64  *
  65  * When run with no arguments, ztest runs for about five minutes and
  66  * produces no output if successful.  To get a little bit of information,
  67  * specify -V.  To get more information, specify -VV, and so on.
  68  *
  69  * To turn this into an overnight stress test, use -T to specify run time.
  70  *
  71  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
  72  * to increase the pool capacity, fanout, and overall stress level.
  73  *
  74  * Use the -k option to set the desired frequency of kills.
  75  *
  76  * When ztest invokes itself it passes all relevant information through a
  77  * temporary file which is mmap-ed in the child process. This allows shared
  78  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
  79  * stored at offset 0 of this file and contains information on the size and
  80  * number of shared structures in the file. The information stored in this file
  81  * must remain backwards compatible with older versions of ztest so that
  82  * ztest can invoke them during backwards compatibility testing (-B).
  83  */
  84 
  85 #include <sys/zfs_context.h>
  86 #include <sys/spa.h>
  87 #include <sys/dmu.h>
  88 #include <sys/txg.h>
  89 #include <sys/dbuf.h>
  90 #include <sys/zap.h>
  91 #include <sys/dmu_objset.h>
  92 #include <sys/poll.h>
  93 #include <sys/stat.h>
  94 #include <sys/time.h>
  95 #include <sys/wait.h>
  96 #include <sys/mman.h>
  97 #include <sys/resource.h>
  98 #include <sys/zio.h>
  99 #include <sys/zil.h>
 100 #include <sys/zil_impl.h>
 101 #include <sys/vdev_impl.h>
 102 #include <sys/vdev_file.h>
 103 #include <sys/spa_impl.h>
 104 #include <sys/metaslab_impl.h>
 105 #include <sys/dsl_prop.h>
 106 #include <sys/dsl_dataset.h>
 107 #include <sys/dsl_destroy.h>
 108 #include <sys/dsl_scan.h>
 109 #include <sys/zio_checksum.h>
 110 #include <sys/refcount.h>
 111 #include <sys/zfeature.h>
 112 #include <sys/dsl_userhold.h>
 113 #include <stdio.h>
 114 #include <stdio_ext.h>
 115 #include <stdlib.h>
 116 #include <unistd.h>
 117 #include <signal.h>
 118 #include <umem.h>
 119 #include <dlfcn.h>
 120 #include <ctype.h>
 121 #include <math.h>
 122 #include <sys/fs/zfs.h>
 123 #include <libnvpair.h>
 124 
 125 static int ztest_fd_data = -1;
 126 static int ztest_fd_rand = -1;
 127 
 128 typedef struct ztest_shared_hdr {
 129         uint64_t        zh_hdr_size;
 130         uint64_t        zh_opts_size;
 131         uint64_t        zh_size;
 132         uint64_t        zh_stats_size;
 133         uint64_t        zh_stats_count;
 134         uint64_t        zh_ds_size;
 135         uint64_t        zh_ds_count;
 136 } ztest_shared_hdr_t;
 137 
 138 static ztest_shared_hdr_t *ztest_shared_hdr;
 139 
 140 typedef struct ztest_shared_opts {
 141         char zo_pool[MAXNAMELEN];
 142         char zo_dir[MAXNAMELEN];
 143         char zo_alt_ztest[MAXNAMELEN];
 144         char zo_alt_libpath[MAXNAMELEN];
 145         uint64_t zo_vdevs;
 146         uint64_t zo_vdevtime;
 147         size_t zo_vdev_size;
 148         int zo_ashift;
 149         int zo_mirrors;
 150         int zo_raidz;
 151         int zo_raidz_parity;
 152         int zo_datasets;
 153         int zo_threads;
 154         uint64_t zo_passtime;
 155         uint64_t zo_killrate;
 156         int zo_verbose;
 157         int zo_init;
 158         uint64_t zo_time;
 159         uint64_t zo_maxloops;
 160         uint64_t zo_metaslab_gang_bang;
 161 } ztest_shared_opts_t;
 162 
 163 static const ztest_shared_opts_t ztest_opts_defaults = {
 164         .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
 165         .zo_dir = { '/', 't', 'm', 'p', '\0' },
 166         .zo_alt_ztest = { '\0' },
 167         .zo_alt_libpath = { '\0' },
 168         .zo_vdevs = 5,
 169         .zo_ashift = SPA_MINBLOCKSHIFT,
 170         .zo_mirrors = 2,
 171         .zo_raidz = 4,
 172         .zo_raidz_parity = 1,
 173         .zo_vdev_size = SPA_MINDEVSIZE,
 174         .zo_datasets = 7,
 175         .zo_threads = 23,
 176         .zo_passtime = 60,              /* 60 seconds */
 177         .zo_killrate = 70,              /* 70% kill rate */
 178         .zo_verbose = 0,
 179         .zo_init = 1,
 180         .zo_time = 300,                 /* 5 minutes */
 181         .zo_maxloops = 50,              /* max loops during spa_freeze() */
 182         .zo_metaslab_gang_bang = 32 << 10
 183 };
 184 
 185 extern uint64_t metaslab_gang_bang;
 186 extern uint64_t metaslab_df_alloc_threshold;
 187 extern uint64_t zfs_deadman_synctime;
 188 
 189 static ztest_shared_opts_t *ztest_shared_opts;
 190 static ztest_shared_opts_t ztest_opts;
 191 
 192 typedef struct ztest_shared_ds {
 193         uint64_t        zd_seq;
 194 } ztest_shared_ds_t;
 195 
 196 static ztest_shared_ds_t *ztest_shared_ds;
 197 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
 198 
 199 #define BT_MAGIC        0x123456789abcdefULL
 200 #define MAXFAULTS() \
 201         (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
 202 
 203 enum ztest_io_type {
 204         ZTEST_IO_WRITE_TAG,
 205         ZTEST_IO_WRITE_PATTERN,
 206         ZTEST_IO_WRITE_ZEROES,
 207         ZTEST_IO_TRUNCATE,
 208         ZTEST_IO_SETATTR,
 209         ZTEST_IO_REWRITE,
 210         ZTEST_IO_TYPES
 211 };
 212 
 213 typedef struct ztest_block_tag {
 214         uint64_t        bt_magic;
 215         uint64_t        bt_objset;
 216         uint64_t        bt_object;
 217         uint64_t        bt_offset;
 218         uint64_t        bt_gen;
 219         uint64_t        bt_txg;
 220         uint64_t        bt_crtxg;
 221 } ztest_block_tag_t;
 222 
 223 typedef struct bufwad {
 224         uint64_t        bw_index;
 225         uint64_t        bw_txg;
 226         uint64_t        bw_data;
 227 } bufwad_t;
 228 
 229 /*
 230  * XXX -- fix zfs range locks to be generic so we can use them here.
 231  */
 232 typedef enum {
 233         RL_READER,
 234         RL_WRITER,
 235         RL_APPEND
 236 } rl_type_t;
 237 
 238 typedef struct rll {
 239         void            *rll_writer;
 240         int             rll_readers;
 241         mutex_t         rll_lock;
 242         cond_t          rll_cv;
 243 } rll_t;
 244 
 245 typedef struct rl {
 246         uint64_t        rl_object;
 247         uint64_t        rl_offset;
 248         uint64_t        rl_size;
 249         rll_t           *rl_lock;
 250 } rl_t;
 251 
 252 #define ZTEST_RANGE_LOCKS       64
 253 #define ZTEST_OBJECT_LOCKS      64
 254 
 255 /*
 256  * Object descriptor.  Used as a template for object lookup/create/remove.
 257  */
 258 typedef struct ztest_od {
 259         uint64_t        od_dir;
 260         uint64_t        od_object;
 261         dmu_object_type_t od_type;
 262         dmu_object_type_t od_crtype;
 263         uint64_t        od_blocksize;
 264         uint64_t        od_crblocksize;
 265         uint64_t        od_gen;
 266         uint64_t        od_crgen;
 267         char            od_name[MAXNAMELEN];
 268 } ztest_od_t;
 269 
 270 /*
 271  * Per-dataset state.
 272  */
 273 typedef struct ztest_ds {
 274         ztest_shared_ds_t *zd_shared;
 275         objset_t        *zd_os;
 276         rwlock_t        zd_zilog_lock;
 277         zilog_t         *zd_zilog;
 278         ztest_od_t      *zd_od;         /* debugging aid */
 279         char            zd_name[MAXNAMELEN];
 280         mutex_t         zd_dirobj_lock;
 281         rll_t           zd_object_lock[ZTEST_OBJECT_LOCKS];
 282         rll_t           zd_range_lock[ZTEST_RANGE_LOCKS];
 283 } ztest_ds_t;
 284 
 285 /*
 286  * Per-iteration state.
 287  */
 288 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
 289 
 290 typedef struct ztest_info {
 291         ztest_func_t    *zi_func;       /* test function */
 292         uint64_t        zi_iters;       /* iterations per execution */
 293         uint64_t        *zi_interval;   /* execute every <interval> seconds */
 294 } ztest_info_t;
 295 
 296 typedef struct ztest_shared_callstate {
 297         uint64_t        zc_count;       /* per-pass count */
 298         uint64_t        zc_time;        /* per-pass time */
 299         uint64_t        zc_next;        /* next time to call this function */
 300 } ztest_shared_callstate_t;
 301 
 302 static ztest_shared_callstate_t *ztest_shared_callstate;
 303 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
 304 
 305 /*
 306  * Note: these aren't static because we want dladdr() to work.
 307  */
 308 ztest_func_t ztest_dmu_read_write;
 309 ztest_func_t ztest_dmu_write_parallel;
 310 ztest_func_t ztest_dmu_object_alloc_free;
 311 ztest_func_t ztest_dmu_commit_callbacks;
 312 ztest_func_t ztest_zap;
 313 ztest_func_t ztest_zap_parallel;
 314 ztest_func_t ztest_zil_commit;
 315 ztest_func_t ztest_zil_remount;
 316 ztest_func_t ztest_dmu_read_write_zcopy;
 317 ztest_func_t ztest_dmu_objset_create_destroy;
 318 ztest_func_t ztest_dmu_prealloc;
 319 ztest_func_t ztest_fzap;
 320 ztest_func_t ztest_dmu_snapshot_create_destroy;
 321 ztest_func_t ztest_dsl_prop_get_set;
 322 ztest_func_t ztest_spa_prop_get_set;
 323 ztest_func_t ztest_spa_create_destroy;
 324 ztest_func_t ztest_fault_inject;
 325 ztest_func_t ztest_ddt_repair;
 326 ztest_func_t ztest_dmu_snapshot_hold;
 327 ztest_func_t ztest_spa_rename;
 328 ztest_func_t ztest_scrub;
 329 ztest_func_t ztest_dsl_dataset_promote_busy;
 330 ztest_func_t ztest_vdev_attach_detach;
 331 ztest_func_t ztest_vdev_LUN_growth;
 332 ztest_func_t ztest_vdev_add_remove;
 333 ztest_func_t ztest_vdev_aux_add_remove;
 334 ztest_func_t ztest_split_pool;
 335 ztest_func_t ztest_reguid;
 336 ztest_func_t ztest_spa_upgrade;
 337 
 338 uint64_t zopt_always = 0ULL * NANOSEC;          /* all the time */
 339 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;  /* every 1/10 second */
 340 uint64_t zopt_often = 1ULL * NANOSEC;           /* every second */
 341 uint64_t zopt_sometimes = 10ULL * NANOSEC;      /* every 10 seconds */
 342 uint64_t zopt_rarely = 60ULL * NANOSEC;         /* every 60 seconds */
 343 
 344 ztest_info_t ztest_info[] = {
 345         { ztest_dmu_read_write,                 1,      &zopt_always        },
 346         { ztest_dmu_write_parallel,             10,     &zopt_always        },
 347         { ztest_dmu_object_alloc_free,          1,      &zopt_always        },
 348         { ztest_dmu_commit_callbacks,           1,      &zopt_always        },
 349         { ztest_zap,                            30,     &zopt_always        },
 350         { ztest_zap_parallel,                   100,    &zopt_always        },
 351         { ztest_split_pool,                     1,      &zopt_always        },
 352         { ztest_zil_commit,                     1,      &zopt_incessant     },
 353         { ztest_zil_remount,                    1,      &zopt_sometimes     },
 354         { ztest_dmu_read_write_zcopy,           1,      &zopt_often },
 355         { ztest_dmu_objset_create_destroy,      1,      &zopt_often },
 356         { ztest_dsl_prop_get_set,               1,      &zopt_often },
 357         { ztest_spa_prop_get_set,               1,      &zopt_sometimes     },
 358 #if 0
 359         { ztest_dmu_prealloc,                   1,      &zopt_sometimes     },
 360 #endif
 361         { ztest_fzap,                           1,      &zopt_sometimes     },
 362         { ztest_dmu_snapshot_create_destroy,    1,      &zopt_sometimes     },
 363         { ztest_spa_create_destroy,             1,      &zopt_sometimes     },
 364         { ztest_fault_inject,                   1,      &zopt_sometimes     },
 365         { ztest_ddt_repair,                     1,      &zopt_sometimes     },
 366         { ztest_dmu_snapshot_hold,              1,      &zopt_sometimes     },
 367         { ztest_reguid,                         1,      &zopt_rarely        },
 368         { ztest_spa_rename,                     1,      &zopt_rarely        },
 369         { ztest_scrub,                          1,      &zopt_rarely        },
 370         { ztest_spa_upgrade,                    1,      &zopt_rarely        },
 371         { ztest_dsl_dataset_promote_busy,       1,      &zopt_rarely        },
 372         { ztest_vdev_attach_detach,             1,      &zopt_sometimes     },
 373         { ztest_vdev_LUN_growth,                1,      &zopt_rarely        },
 374         { ztest_vdev_add_remove,                1,
 375             &ztest_opts.zo_vdevtime                         },
 376         { ztest_vdev_aux_add_remove,            1,
 377             &ztest_opts.zo_vdevtime                         },
 378 };
 379 
 380 #define ZTEST_FUNCS     (sizeof (ztest_info) / sizeof (ztest_info_t))
 381 
 382 /*
 383  * The following struct is used to hold a list of uncalled commit callbacks.
 384  * The callbacks are ordered by txg number.
 385  */
 386 typedef struct ztest_cb_list {
 387         mutex_t zcl_callbacks_lock;
 388         list_t  zcl_callbacks;
 389 } ztest_cb_list_t;
 390 
 391 /*
 392  * Stuff we need to share writably between parent and child.
 393  */
 394 typedef struct ztest_shared {
 395         boolean_t       zs_do_init;
 396         hrtime_t        zs_proc_start;
 397         hrtime_t        zs_proc_stop;
 398         hrtime_t        zs_thread_start;
 399         hrtime_t        zs_thread_stop;
 400         hrtime_t        zs_thread_kill;
 401         uint64_t        zs_enospc_count;
 402         uint64_t        zs_vdev_next_leaf;
 403         uint64_t        zs_vdev_aux;
 404         uint64_t        zs_alloc;
 405         uint64_t        zs_space;
 406         uint64_t        zs_splits;
 407         uint64_t        zs_mirrors;
 408         uint64_t        zs_metaslab_sz;
 409         uint64_t        zs_metaslab_df_alloc_threshold;
 410         uint64_t        zs_guid;
 411 } ztest_shared_t;
 412 
 413 #define ID_PARALLEL     -1ULL
 414 
 415 static char ztest_dev_template[] = "%s/%s.%llua";
 416 static char ztest_aux_template[] = "%s/%s.%s.%llu";
 417 ztest_shared_t *ztest_shared;
 418 
 419 static spa_t *ztest_spa = NULL;
 420 static ztest_ds_t *ztest_ds;
 421 
 422 static mutex_t ztest_vdev_lock;
 423 
 424 /*
 425  * The ztest_name_lock protects the pool and dataset namespace used by
 426  * the individual tests. To modify the namespace, consumers must grab
 427  * this lock as writer. Grabbing the lock as reader will ensure that the
 428  * namespace does not change while the lock is held.
 429  */
 430 static rwlock_t ztest_name_lock;
 431 
 432 static boolean_t ztest_dump_core = B_TRUE;
 433 static boolean_t ztest_exiting;
 434 
 435 /* Global commit callback list */
 436 static ztest_cb_list_t zcl;
 437 
 438 enum ztest_object {
 439         ZTEST_META_DNODE = 0,
 440         ZTEST_DIROBJ,
 441         ZTEST_OBJECTS
 442 };
 443 
 444 static void usage(boolean_t) __NORETURN;
 445 
 446 /*
 447  * These libumem hooks provide a reasonable set of defaults for the allocator's
 448  * debugging facilities.
 449  */
 450 const char *
 451 _umem_debug_init()
 452 {
 453         return ("default,verbose"); /* $UMEM_DEBUG setting */
 454 }
 455 
 456 const char *
 457 _umem_logging_init(void)
 458 {
 459         return ("fail,contents"); /* $UMEM_LOGGING setting */
 460 }
 461 
 462 #define FATAL_MSG_SZ    1024
 463 
 464 char *fatal_msg;
 465 
 466 static void
 467 fatal(int do_perror, char *message, ...)
 468 {
 469         va_list args;
 470         int save_errno = errno;
 471         char buf[FATAL_MSG_SZ];
 472 
 473         (void) fflush(stdout);
 474 
 475         va_start(args, message);
 476         (void) sprintf(buf, "ztest: ");
 477         /* LINTED */
 478         (void) vsprintf(buf + strlen(buf), message, args);
 479         va_end(args);
 480         if (do_perror) {
 481                 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
 482                     ": %s", strerror(save_errno));
 483         }
 484         (void) fprintf(stderr, "%s\n", buf);
 485         fatal_msg = buf;                        /* to ease debugging */
 486         if (ztest_dump_core)
 487                 abort();
 488         exit(3);
 489 }
 490 
 491 static int
 492 str2shift(const char *buf)
 493 {
 494         const char *ends = "BKMGTPEZ";
 495         int i;
 496 
 497         if (buf[0] == '\0')
 498                 return (0);
 499         for (i = 0; i < strlen(ends); i++) {
 500                 if (toupper(buf[0]) == ends[i])
 501                         break;
 502         }
 503         if (i == strlen(ends)) {
 504                 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
 505                     buf);
 506                 usage(B_FALSE);
 507         }
 508         if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
 509                 return (10*i);
 510         }
 511         (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
 512         usage(B_FALSE);
 513         /* NOTREACHED */
 514 }
 515 
 516 static uint64_t
 517 nicenumtoull(const char *buf)
 518 {
 519         char *end;
 520         uint64_t val;
 521 
 522         val = strtoull(buf, &end, 0);
 523         if (end == buf) {
 524                 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
 525                 usage(B_FALSE);
 526         } else if (end[0] == '.') {
 527                 double fval = strtod(buf, &end);
 528                 fval *= pow(2, str2shift(end));
 529                 if (fval > UINT64_MAX) {
 530                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 531                             buf);
 532                         usage(B_FALSE);
 533                 }
 534                 val = (uint64_t)fval;
 535         } else {
 536                 int shift = str2shift(end);
 537                 if (shift >= 64 || (val << shift) >> shift != val) {
 538                         (void) fprintf(stderr, "ztest: value too large: %s\n",
 539                             buf);
 540                         usage(B_FALSE);
 541                 }
 542                 val <<= shift;
 543         }
 544         return (val);
 545 }
 546 
 547 static void
 548 usage(boolean_t requested)
 549 {
 550         const ztest_shared_opts_t *zo = &ztest_opts_defaults;
 551 
 552         char nice_vdev_size[10];
 553         char nice_gang_bang[10];
 554         FILE *fp = requested ? stdout : stderr;
 555 
 556         nicenum(zo->zo_vdev_size, nice_vdev_size);
 557         nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
 558 
 559         (void) fprintf(fp, "Usage: %s\n"
 560             "\t[-v vdevs (default: %llu)]\n"
 561             "\t[-s size_of_each_vdev (default: %s)]\n"
 562             "\t[-a alignment_shift (default: %d)] use 0 for random\n"
 563             "\t[-m mirror_copies (default: %d)]\n"
 564             "\t[-r raidz_disks (default: %d)]\n"
 565             "\t[-R raidz_parity (default: %d)]\n"
 566             "\t[-d datasets (default: %d)]\n"
 567             "\t[-t threads (default: %d)]\n"
 568             "\t[-g gang_block_threshold (default: %s)]\n"
 569             "\t[-i init_count (default: %d)] initialize pool i times\n"
 570             "\t[-k kill_percentage (default: %llu%%)]\n"
 571             "\t[-p pool_name (default: %s)]\n"
 572             "\t[-f dir (default: %s)] file directory for vdev files\n"
 573             "\t[-V] verbose (use multiple times for ever more blather)\n"
 574             "\t[-E] use existing pool instead of creating new one\n"
 575             "\t[-T time (default: %llu sec)] total run time\n"
 576             "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
 577             "\t[-P passtime (default: %llu sec)] time per pass\n"
 578             "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
 579             "\t[-h] (print help)\n"
 580             "",
 581             zo->zo_pool,
 582             (u_longlong_t)zo->zo_vdevs,                      /* -v */
 583             nice_vdev_size,                             /* -s */
 584             zo->zo_ashift,                           /* -a */
 585             zo->zo_mirrors,                          /* -m */
 586             zo->zo_raidz,                            /* -r */
 587             zo->zo_raidz_parity,                     /* -R */
 588             zo->zo_datasets,                         /* -d */
 589             zo->zo_threads,                          /* -t */
 590             nice_gang_bang,                             /* -g */
 591             zo->zo_init,                             /* -i */
 592             (u_longlong_t)zo->zo_killrate,           /* -k */
 593             zo->zo_pool,                             /* -p */
 594             zo->zo_dir,                                      /* -f */
 595             (u_longlong_t)zo->zo_time,                       /* -T */
 596             (u_longlong_t)zo->zo_maxloops,           /* -F */
 597             (u_longlong_t)zo->zo_passtime);
 598         exit(requested ? 0 : 1);
 599 }
 600 
 601 static void
 602 process_options(int argc, char **argv)
 603 {
 604         char *path;
 605         ztest_shared_opts_t *zo = &ztest_opts;
 606 
 607         int opt;
 608         uint64_t value;
 609         char altdir[MAXNAMELEN] = { 0 };
 610 
 611         bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
 612 
 613         while ((opt = getopt(argc, argv,
 614             "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
 615                 value = 0;
 616                 switch (opt) {
 617                 case 'v':
 618                 case 's':
 619                 case 'a':
 620                 case 'm':
 621                 case 'r':
 622                 case 'R':
 623                 case 'd':
 624                 case 't':
 625                 case 'g':
 626                 case 'i':
 627                 case 'k':
 628                 case 'T':
 629                 case 'P':
 630                 case 'F':
 631                         value = nicenumtoull(optarg);
 632                 }
 633                 switch (opt) {
 634                 case 'v':
 635                         zo->zo_vdevs = value;
 636                         break;
 637                 case 's':
 638                         zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
 639                         break;
 640                 case 'a':
 641                         zo->zo_ashift = value;
 642                         break;
 643                 case 'm':
 644                         zo->zo_mirrors = value;
 645                         break;
 646                 case 'r':
 647                         zo->zo_raidz = MAX(1, value);
 648                         break;
 649                 case 'R':
 650                         zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
 651                         break;
 652                 case 'd':
 653                         zo->zo_datasets = MAX(1, value);
 654                         break;
 655                 case 't':
 656                         zo->zo_threads = MAX(1, value);
 657                         break;
 658                 case 'g':
 659                         zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
 660                             value);
 661                         break;
 662                 case 'i':
 663                         zo->zo_init = value;
 664                         break;
 665                 case 'k':
 666                         zo->zo_killrate = value;
 667                         break;
 668                 case 'p':
 669                         (void) strlcpy(zo->zo_pool, optarg,
 670                             sizeof (zo->zo_pool));
 671                         break;
 672                 case 'f':
 673                         path = realpath(optarg, NULL);
 674                         if (path == NULL) {
 675                                 (void) fprintf(stderr, "error: %s: %s\n",
 676                                     optarg, strerror(errno));
 677                                 usage(B_FALSE);
 678                         } else {
 679                                 (void) strlcpy(zo->zo_dir, path,
 680                                     sizeof (zo->zo_dir));
 681                         }
 682                         break;
 683                 case 'V':
 684                         zo->zo_verbose++;
 685                         break;
 686                 case 'E':
 687                         zo->zo_init = 0;
 688                         break;
 689                 case 'T':
 690                         zo->zo_time = value;
 691                         break;
 692                 case 'P':
 693                         zo->zo_passtime = MAX(1, value);
 694                         break;
 695                 case 'F':
 696                         zo->zo_maxloops = MAX(1, value);
 697                         break;
 698                 case 'B':
 699                         (void) strlcpy(altdir, optarg, sizeof (altdir));
 700                         break;
 701                 case 'h':
 702                         usage(B_TRUE);
 703                         break;
 704                 case '?':
 705                 default:
 706                         usage(B_FALSE);
 707                         break;
 708                 }
 709         }
 710 
 711         zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
 712 
 713         zo->zo_vdevtime =
 714             (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
 715             UINT64_MAX >> 2);
 716 
 717         if (strlen(altdir) > 0) {
 718                 char *cmd;
 719                 char *realaltdir;
 720                 char *bin;
 721                 char *ztest;
 722                 char *isa;
 723                 int isalen;
 724 
 725                 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
 726                 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
 727 
 728                 VERIFY(NULL != realpath(getexecname(), cmd));
 729                 if (0 != access(altdir, F_OK)) {
 730                         ztest_dump_core = B_FALSE;
 731                         fatal(B_TRUE, "invalid alternate ztest path: %s",
 732                             altdir);
 733                 }
 734                 VERIFY(NULL != realpath(altdir, realaltdir));
 735 
 736                 /*
 737                  * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
 738                  * We want to extract <isa> to determine if we should use
 739                  * 32 or 64 bit binaries.
 740                  */
 741                 bin = strstr(cmd, "/usr/bin/");
 742                 ztest = strstr(bin, "/ztest");
 743                 isa = bin + 9;
 744                 isalen = ztest - isa;
 745                 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
 746                     "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
 747                 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
 748                     "%s/usr/lib/%.*s", realaltdir, isalen, isa);
 749 
 750                 if (0 != access(zo->zo_alt_ztest, X_OK)) {
 751                         ztest_dump_core = B_FALSE;
 752                         fatal(B_TRUE, "invalid alternate ztest: %s",
 753                             zo->zo_alt_ztest);
 754                 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
 755                         ztest_dump_core = B_FALSE;
 756                         fatal(B_TRUE, "invalid alternate lib directory %s",
 757                             zo->zo_alt_libpath);
 758                 }
 759 
 760                 umem_free(cmd, MAXPATHLEN);
 761                 umem_free(realaltdir, MAXPATHLEN);
 762         }
 763 }
 764 
 765 static void
 766 ztest_kill(ztest_shared_t *zs)
 767 {
 768         zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
 769         zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
 770         (void) kill(getpid(), SIGKILL);
 771 }
 772 
 773 static uint64_t
 774 ztest_random(uint64_t range)
 775 {
 776         uint64_t r;
 777 
 778         ASSERT3S(ztest_fd_rand, >=, 0);
 779 
 780         if (range == 0)
 781                 return (0);
 782 
 783         if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
 784                 fatal(1, "short read from /dev/urandom");
 785 
 786         return (r % range);
 787 }
 788 
 789 /* ARGSUSED */
 790 static void
 791 ztest_record_enospc(const char *s)
 792 {
 793         ztest_shared->zs_enospc_count++;
 794 }
 795 
 796 static uint64_t
 797 ztest_get_ashift(void)
 798 {
 799         if (ztest_opts.zo_ashift == 0)
 800                 return (SPA_MINBLOCKSHIFT + ztest_random(3));
 801         return (ztest_opts.zo_ashift);
 802 }
 803 
 804 static nvlist_t *
 805 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
 806 {
 807         char pathbuf[MAXPATHLEN];
 808         uint64_t vdev;
 809         nvlist_t *file;
 810 
 811         if (ashift == 0)
 812                 ashift = ztest_get_ashift();
 813 
 814         if (path == NULL) {
 815                 path = pathbuf;
 816 
 817                 if (aux != NULL) {
 818                         vdev = ztest_shared->zs_vdev_aux;
 819                         (void) snprintf(path, sizeof (pathbuf),
 820                             ztest_aux_template, ztest_opts.zo_dir,
 821                             pool == NULL ? ztest_opts.zo_pool : pool,
 822                             aux, vdev);
 823                 } else {
 824                         vdev = ztest_shared->zs_vdev_next_leaf++;
 825                         (void) snprintf(path, sizeof (pathbuf),
 826                             ztest_dev_template, ztest_opts.zo_dir,
 827                             pool == NULL ? ztest_opts.zo_pool : pool, vdev);
 828                 }
 829         }
 830 
 831         if (size != 0) {
 832                 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
 833                 if (fd == -1)
 834                         fatal(1, "can't open %s", path);
 835                 if (ftruncate(fd, size) != 0)
 836                         fatal(1, "can't ftruncate %s", path);
 837                 (void) close(fd);
 838         }
 839 
 840         VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
 841         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
 842         VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
 843         VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
 844 
 845         return (file);
 846 }
 847 
 848 static nvlist_t *
 849 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
 850     uint64_t ashift, int r)
 851 {
 852         nvlist_t *raidz, **child;
 853         int c;
 854 
 855         if (r < 2)
 856                 return (make_vdev_file(path, aux, pool, size, ashift));
 857         child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
 858 
 859         for (c = 0; c < r; c++)
 860                 child[c] = make_vdev_file(path, aux, pool, size, ashift);
 861 
 862         VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
 863         VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
 864             VDEV_TYPE_RAIDZ) == 0);
 865         VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
 866             ztest_opts.zo_raidz_parity) == 0);
 867         VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
 868             child, r) == 0);
 869 
 870         for (c = 0; c < r; c++)
 871                 nvlist_free(child[c]);
 872 
 873         umem_free(child, r * sizeof (nvlist_t *));
 874 
 875         return (raidz);
 876 }
 877 
 878 static nvlist_t *
 879 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
 880     uint64_t ashift, int r, int m)
 881 {
 882         nvlist_t *mirror, **child;
 883         int c;
 884 
 885         if (m < 1)
 886                 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
 887 
 888         child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
 889 
 890         for (c = 0; c < m; c++)
 891                 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
 892 
 893         VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
 894         VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
 895             VDEV_TYPE_MIRROR) == 0);
 896         VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
 897             child, m) == 0);
 898 
 899         for (c = 0; c < m; c++)
 900                 nvlist_free(child[c]);
 901 
 902         umem_free(child, m * sizeof (nvlist_t *));
 903 
 904         return (mirror);
 905 }
 906 
 907 static nvlist_t *
 908 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
 909     int log, int r, int m, int t)
 910 {
 911         nvlist_t *root, **child;
 912         int c;
 913 
 914         ASSERT(t > 0);
 915 
 916         child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
 917 
 918         for (c = 0; c < t; c++) {
 919                 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
 920                     r, m);
 921                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
 922                     log) == 0);
 923         }
 924 
 925         VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
 926         VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
 927         VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
 928             child, t) == 0);
 929 
 930         for (c = 0; c < t; c++)
 931                 nvlist_free(child[c]);
 932 
 933         umem_free(child, t * sizeof (nvlist_t *));
 934 
 935         return (root);
 936 }
 937 
 938 /*
 939  * Find a random spa version. Returns back a random spa version in the
 940  * range [initial_version, SPA_VERSION_FEATURES].
 941  */
 942 static uint64_t
 943 ztest_random_spa_version(uint64_t initial_version)
 944 {
 945         uint64_t version = initial_version;
 946 
 947         if (version <= SPA_VERSION_BEFORE_FEATURES) {
 948                 version = version +
 949                     ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
 950         }
 951 
 952         if (version > SPA_VERSION_BEFORE_FEATURES)
 953                 version = SPA_VERSION_FEATURES;
 954 
 955         ASSERT(SPA_VERSION_IS_SUPPORTED(version));
 956         return (version);
 957 }
 958 
 959 static int
 960 ztest_random_blocksize(void)
 961 {
 962         return (1 << (SPA_MINBLOCKSHIFT +
 963             ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
 964 }
 965 
 966 static int
 967 ztest_random_ibshift(void)
 968 {
 969         return (DN_MIN_INDBLKSHIFT +
 970             ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
 971 }
 972 
 973 static uint64_t
 974 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
 975 {
 976         uint64_t top;
 977         vdev_t *rvd = spa->spa_root_vdev;
 978         vdev_t *tvd;
 979 
 980         ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
 981 
 982         do {
 983                 top = ztest_random(rvd->vdev_children);
 984                 tvd = rvd->vdev_child[top];
 985         } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
 986             tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
 987 
 988         return (top);
 989 }
 990 
 991 static uint64_t
 992 ztest_random_dsl_prop(zfs_prop_t prop)
 993 {
 994         uint64_t value;
 995 
 996         do {
 997                 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
 998         } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
 999 
1000         return (value);
1001 }
1002 
1003 static int
1004 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1005     boolean_t inherit)
1006 {
1007         const char *propname = zfs_prop_to_name(prop);
1008         const char *valname;
1009         char setpoint[MAXPATHLEN];
1010         uint64_t curval;
1011         int error;
1012 
1013         error = dsl_prop_set_int(osname, propname,
1014             (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1015 
1016         if (error == ENOSPC) {
1017                 ztest_record_enospc(FTAG);
1018                 return (error);
1019         }
1020         ASSERT0(error);
1021 
1022         VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1023 
1024         if (ztest_opts.zo_verbose >= 6) {
1025                 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1026                 (void) printf("%s %s = %s at '%s'\n",
1027                     osname, propname, valname, setpoint);
1028         }
1029 
1030         return (error);
1031 }
1032 
1033 static int
1034 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1035 {
1036         spa_t *spa = ztest_spa;
1037         nvlist_t *props = NULL;
1038         int error;
1039 
1040         VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1041         VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1042 
1043         error = spa_prop_set(spa, props);
1044 
1045         nvlist_free(props);
1046 
1047         if (error == ENOSPC) {
1048                 ztest_record_enospc(FTAG);
1049                 return (error);
1050         }
1051         ASSERT0(error);
1052 
1053         return (error);
1054 }
1055 
1056 static void
1057 ztest_rll_init(rll_t *rll)
1058 {
1059         rll->rll_writer = NULL;
1060         rll->rll_readers = 0;
1061         VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1062         VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1063 }
1064 
1065 static void
1066 ztest_rll_destroy(rll_t *rll)
1067 {
1068         ASSERT(rll->rll_writer == NULL);
1069         ASSERT(rll->rll_readers == 0);
1070         VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1071         VERIFY(cond_destroy(&rll->rll_cv) == 0);
1072 }
1073 
1074 static void
1075 ztest_rll_lock(rll_t *rll, rl_type_t type)
1076 {
1077         VERIFY(mutex_lock(&rll->rll_lock) == 0);
1078 
1079         if (type == RL_READER) {
1080                 while (rll->rll_writer != NULL)
1081                         (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1082                 rll->rll_readers++;
1083         } else {
1084                 while (rll->rll_writer != NULL || rll->rll_readers)
1085                         (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1086                 rll->rll_writer = curthread;
1087         }
1088 
1089         VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1090 }
1091 
1092 static void
1093 ztest_rll_unlock(rll_t *rll)
1094 {
1095         VERIFY(mutex_lock(&rll->rll_lock) == 0);
1096 
1097         if (rll->rll_writer) {
1098                 ASSERT(rll->rll_readers == 0);
1099                 rll->rll_writer = NULL;
1100         } else {
1101                 ASSERT(rll->rll_readers != 0);
1102                 ASSERT(rll->rll_writer == NULL);
1103                 rll->rll_readers--;
1104         }
1105 
1106         if (rll->rll_writer == NULL && rll->rll_readers == 0)
1107                 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1108 
1109         VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1110 }
1111 
1112 static void
1113 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1114 {
1115         rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1116 
1117         ztest_rll_lock(rll, type);
1118 }
1119 
1120 static void
1121 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1122 {
1123         rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1124 
1125         ztest_rll_unlock(rll);
1126 }
1127 
1128 static rl_t *
1129 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1130     uint64_t size, rl_type_t type)
1131 {
1132         uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1133         rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1134         rl_t *rl;
1135 
1136         rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1137         rl->rl_object = object;
1138         rl->rl_offset = offset;
1139         rl->rl_size = size;
1140         rl->rl_lock = rll;
1141 
1142         ztest_rll_lock(rll, type);
1143 
1144         return (rl);
1145 }
1146 
1147 static void
1148 ztest_range_unlock(rl_t *rl)
1149 {
1150         rll_t *rll = rl->rl_lock;
1151 
1152         ztest_rll_unlock(rll);
1153 
1154         umem_free(rl, sizeof (*rl));
1155 }
1156 
1157 static void
1158 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1159 {
1160         zd->zd_os = os;
1161         zd->zd_zilog = dmu_objset_zil(os);
1162         zd->zd_shared = szd;
1163         dmu_objset_name(os, zd->zd_name);
1164 
1165         if (zd->zd_shared != NULL)
1166                 zd->zd_shared->zd_seq = 0;
1167 
1168         VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1169         VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1170 
1171         for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1172                 ztest_rll_init(&zd->zd_object_lock[l]);
1173 
1174         for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1175                 ztest_rll_init(&zd->zd_range_lock[l]);
1176 }
1177 
1178 static void
1179 ztest_zd_fini(ztest_ds_t *zd)
1180 {
1181         VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1182 
1183         for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1184                 ztest_rll_destroy(&zd->zd_object_lock[l]);
1185 
1186         for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1187                 ztest_rll_destroy(&zd->zd_range_lock[l]);
1188 }
1189 
1190 #define TXG_MIGHTWAIT   (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1191 
1192 static uint64_t
1193 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1194 {
1195         uint64_t txg;
1196         int error;
1197 
1198         /*
1199          * Attempt to assign tx to some transaction group.
1200          */
1201         error = dmu_tx_assign(tx, txg_how);
1202         if (error) {
1203                 if (error == ERESTART) {
1204                         ASSERT(txg_how == TXG_NOWAIT);
1205                         dmu_tx_wait(tx);
1206                 } else {
1207                         ASSERT3U(error, ==, ENOSPC);
1208                         ztest_record_enospc(tag);
1209                 }
1210                 dmu_tx_abort(tx);
1211                 return (0);
1212         }
1213         txg = dmu_tx_get_txg(tx);
1214         ASSERT(txg != 0);
1215         return (txg);
1216 }
1217 
1218 static void
1219 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1220 {
1221         uint64_t *ip = buf;
1222         uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1223 
1224         while (ip < ip_end)
1225                 *ip++ = value;
1226 }
1227 
1228 static boolean_t
1229 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1230 {
1231         uint64_t *ip = buf;
1232         uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1233         uint64_t diff = 0;
1234 
1235         while (ip < ip_end)
1236                 diff |= (value - *ip++);
1237 
1238         return (diff == 0);
1239 }
1240 
1241 static void
1242 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1243     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1244 {
1245         bt->bt_magic = BT_MAGIC;
1246         bt->bt_objset = dmu_objset_id(os);
1247         bt->bt_object = object;
1248         bt->bt_offset = offset;
1249         bt->bt_gen = gen;
1250         bt->bt_txg = txg;
1251         bt->bt_crtxg = crtxg;
1252 }
1253 
1254 static void
1255 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1256     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1257 {
1258         ASSERT(bt->bt_magic == BT_MAGIC);
1259         ASSERT(bt->bt_objset == dmu_objset_id(os));
1260         ASSERT(bt->bt_object == object);
1261         ASSERT(bt->bt_offset == offset);
1262         ASSERT(bt->bt_gen <= gen);
1263         ASSERT(bt->bt_txg <= txg);
1264         ASSERT(bt->bt_crtxg == crtxg);
1265 }
1266 
1267 static ztest_block_tag_t *
1268 ztest_bt_bonus(dmu_buf_t *db)
1269 {
1270         dmu_object_info_t doi;
1271         ztest_block_tag_t *bt;
1272 
1273         dmu_object_info_from_db(db, &doi);
1274         ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1275         ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1276         bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1277 
1278         return (bt);
1279 }
1280 
1281 /*
1282  * ZIL logging ops
1283  */
1284 
1285 #define lrz_type        lr_mode
1286 #define lrz_blocksize   lr_uid
1287 #define lrz_ibshift     lr_gid
1288 #define lrz_bonustype   lr_rdev
1289 #define lrz_bonuslen    lr_crtime[1]
1290 
1291 static void
1292 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1293 {
1294         char *name = (void *)(lr + 1);          /* name follows lr */
1295         size_t namesize = strlen(name) + 1;
1296         itx_t *itx;
1297 
1298         if (zil_replaying(zd->zd_zilog, tx))
1299                 return;
1300 
1301         itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1302         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1303             sizeof (*lr) + namesize - sizeof (lr_t));
1304 
1305         zil_itx_assign(zd->zd_zilog, itx, tx);
1306 }
1307 
1308 static void
1309 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1310 {
1311         char *name = (void *)(lr + 1);          /* name follows lr */
1312         size_t namesize = strlen(name) + 1;
1313         itx_t *itx;
1314 
1315         if (zil_replaying(zd->zd_zilog, tx))
1316                 return;
1317 
1318         itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1319         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1320             sizeof (*lr) + namesize - sizeof (lr_t));
1321 
1322         itx->itx_oid = object;
1323         zil_itx_assign(zd->zd_zilog, itx, tx);
1324 }
1325 
1326 static void
1327 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1328 {
1329         itx_t *itx;
1330         itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1331 
1332         if (zil_replaying(zd->zd_zilog, tx))
1333                 return;
1334 
1335         if (lr->lr_length > ZIL_MAX_LOG_DATA)
1336                 write_state = WR_INDIRECT;
1337 
1338         itx = zil_itx_create(TX_WRITE,
1339             sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1340 
1341         if (write_state == WR_COPIED &&
1342             dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1343             ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1344                 zil_itx_destroy(itx);
1345                 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1346                 write_state = WR_NEED_COPY;
1347         }
1348         itx->itx_private = zd;
1349         itx->itx_wr_state = write_state;
1350         itx->itx_sync = (ztest_random(8) == 0);
1351         itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1352 
1353         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1354             sizeof (*lr) - sizeof (lr_t));
1355 
1356         zil_itx_assign(zd->zd_zilog, itx, tx);
1357 }
1358 
1359 static void
1360 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1361 {
1362         itx_t *itx;
1363 
1364         if (zil_replaying(zd->zd_zilog, tx))
1365                 return;
1366 
1367         itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1368         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1369             sizeof (*lr) - sizeof (lr_t));
1370 
1371         itx->itx_sync = B_FALSE;
1372         zil_itx_assign(zd->zd_zilog, itx, tx);
1373 }
1374 
1375 static void
1376 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1377 {
1378         itx_t *itx;
1379 
1380         if (zil_replaying(zd->zd_zilog, tx))
1381                 return;
1382 
1383         itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1384         bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1385             sizeof (*lr) - sizeof (lr_t));
1386 
1387         itx->itx_sync = B_FALSE;
1388         zil_itx_assign(zd->zd_zilog, itx, tx);
1389 }
1390 
1391 /*
1392  * ZIL replay ops
1393  */
1394 static int
1395 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1396 {
1397         char *name = (void *)(lr + 1);          /* name follows lr */
1398         objset_t *os = zd->zd_os;
1399         ztest_block_tag_t *bbt;
1400         dmu_buf_t *db;
1401         dmu_tx_t *tx;
1402         uint64_t txg;
1403         int error = 0;
1404 
1405         if (byteswap)
1406                 byteswap_uint64_array(lr, sizeof (*lr));
1407 
1408         ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1409         ASSERT(name[0] != '\0');
1410 
1411         tx = dmu_tx_create(os);
1412 
1413         dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1414 
1415         if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1416                 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1417         } else {
1418                 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1419         }
1420 
1421         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1422         if (txg == 0)
1423                 return (ENOSPC);
1424 
1425         ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1426 
1427         if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1428                 if (lr->lr_foid == 0) {
1429                         lr->lr_foid = zap_create(os,
1430                             lr->lrz_type, lr->lrz_bonustype,
1431                             lr->lrz_bonuslen, tx);
1432                 } else {
1433                         error = zap_create_claim(os, lr->lr_foid,
1434                             lr->lrz_type, lr->lrz_bonustype,
1435                             lr->lrz_bonuslen, tx);
1436                 }
1437         } else {
1438                 if (lr->lr_foid == 0) {
1439                         lr->lr_foid = dmu_object_alloc(os,
1440                             lr->lrz_type, 0, lr->lrz_bonustype,
1441                             lr->lrz_bonuslen, tx);
1442                 } else {
1443                         error = dmu_object_claim(os, lr->lr_foid,
1444                             lr->lrz_type, 0, lr->lrz_bonustype,
1445                             lr->lrz_bonuslen, tx);
1446                 }
1447         }
1448 
1449         if (error) {
1450                 ASSERT3U(error, ==, EEXIST);
1451                 ASSERT(zd->zd_zilog->zl_replay);
1452                 dmu_tx_commit(tx);
1453                 return (error);
1454         }
1455 
1456         ASSERT(lr->lr_foid != 0);
1457 
1458         if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1459                 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1460                     lr->lrz_blocksize, lr->lrz_ibshift, tx));
1461 
1462         VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1463         bbt = ztest_bt_bonus(db);
1464         dmu_buf_will_dirty(db, tx);
1465         ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1466         dmu_buf_rele(db, FTAG);
1467 
1468         VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1469             &lr->lr_foid, tx));
1470 
1471         (void) ztest_log_create(zd, tx, lr);
1472 
1473         dmu_tx_commit(tx);
1474 
1475         return (0);
1476 }
1477 
1478 static int
1479 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1480 {
1481         char *name = (void *)(lr + 1);          /* name follows lr */
1482         objset_t *os = zd->zd_os;
1483         dmu_object_info_t doi;
1484         dmu_tx_t *tx;
1485         uint64_t object, txg;
1486 
1487         if (byteswap)
1488                 byteswap_uint64_array(lr, sizeof (*lr));
1489 
1490         ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1491         ASSERT(name[0] != '\0');
1492 
1493         VERIFY3U(0, ==,
1494             zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1495         ASSERT(object != 0);
1496 
1497         ztest_object_lock(zd, object, RL_WRITER);
1498 
1499         VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1500 
1501         tx = dmu_tx_create(os);
1502 
1503         dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1504         dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1505 
1506         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1507         if (txg == 0) {
1508                 ztest_object_unlock(zd, object);
1509                 return (ENOSPC);
1510         }
1511 
1512         if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1513                 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1514         } else {
1515                 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1516         }
1517 
1518         VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1519 
1520         (void) ztest_log_remove(zd, tx, lr, object);
1521 
1522         dmu_tx_commit(tx);
1523 
1524         ztest_object_unlock(zd, object);
1525 
1526         return (0);
1527 }
1528 
1529 static int
1530 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1531 {
1532         objset_t *os = zd->zd_os;
1533         void *data = lr + 1;                    /* data follows lr */
1534         uint64_t offset, length;
1535         ztest_block_tag_t *bt = data;
1536         ztest_block_tag_t *bbt;
1537         uint64_t gen, txg, lrtxg, crtxg;
1538         dmu_object_info_t doi;
1539         dmu_tx_t *tx;
1540         dmu_buf_t *db;
1541         arc_buf_t *abuf = NULL;
1542         rl_t *rl;
1543 
1544         if (byteswap)
1545                 byteswap_uint64_array(lr, sizeof (*lr));
1546 
1547         offset = lr->lr_offset;
1548         length = lr->lr_length;
1549 
1550         /* If it's a dmu_sync() block, write the whole block */
1551         if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1552                 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1553                 if (length < blocksize) {
1554                         offset -= offset % blocksize;
1555                         length = blocksize;
1556                 }
1557         }
1558 
1559         if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1560                 byteswap_uint64_array(bt, sizeof (*bt));
1561 
1562         if (bt->bt_magic != BT_MAGIC)
1563                 bt = NULL;
1564 
1565         ztest_object_lock(zd, lr->lr_foid, RL_READER);
1566         rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1567 
1568         VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1569 
1570         dmu_object_info_from_db(db, &doi);
1571 
1572         bbt = ztest_bt_bonus(db);
1573         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1574         gen = bbt->bt_gen;
1575         crtxg = bbt->bt_crtxg;
1576         lrtxg = lr->lr_common.lrc_txg;
1577 
1578         tx = dmu_tx_create(os);
1579 
1580         dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1581 
1582         if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1583             P2PHASE(offset, length) == 0)
1584                 abuf = dmu_request_arcbuf(db, length);
1585 
1586         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1587         if (txg == 0) {
1588                 if (abuf != NULL)
1589                         dmu_return_arcbuf(abuf);
1590                 dmu_buf_rele(db, FTAG);
1591                 ztest_range_unlock(rl);
1592                 ztest_object_unlock(zd, lr->lr_foid);
1593                 return (ENOSPC);
1594         }
1595 
1596         if (bt != NULL) {
1597                 /*
1598                  * Usually, verify the old data before writing new data --
1599                  * but not always, because we also want to verify correct
1600                  * behavior when the data was not recently read into cache.
1601                  */
1602                 ASSERT(offset % doi.doi_data_block_size == 0);
1603                 if (ztest_random(4) != 0) {
1604                         int prefetch = ztest_random(2) ?
1605                             DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1606                         ztest_block_tag_t rbt;
1607 
1608                         VERIFY(dmu_read(os, lr->lr_foid, offset,
1609                             sizeof (rbt), &rbt, prefetch) == 0);
1610                         if (rbt.bt_magic == BT_MAGIC) {
1611                                 ztest_bt_verify(&rbt, os, lr->lr_foid,
1612                                     offset, gen, txg, crtxg);
1613                         }
1614                 }
1615 
1616                 /*
1617                  * Writes can appear to be newer than the bonus buffer because
1618                  * the ztest_get_data() callback does a dmu_read() of the
1619                  * open-context data, which may be different than the data
1620                  * as it was when the write was generated.
1621                  */
1622                 if (zd->zd_zilog->zl_replay) {
1623                         ztest_bt_verify(bt, os, lr->lr_foid, offset,
1624                             MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1625                             bt->bt_crtxg);
1626                 }
1627 
1628                 /*
1629                  * Set the bt's gen/txg to the bonus buffer's gen/txg
1630                  * so that all of the usual ASSERTs will work.
1631                  */
1632                 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1633         }
1634 
1635         if (abuf == NULL) {
1636                 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1637         } else {
1638                 bcopy(data, abuf->b_data, length);
1639                 dmu_assign_arcbuf(db, offset, abuf, tx);
1640         }
1641 
1642         (void) ztest_log_write(zd, tx, lr);
1643 
1644         dmu_buf_rele(db, FTAG);
1645 
1646         dmu_tx_commit(tx);
1647 
1648         ztest_range_unlock(rl);
1649         ztest_object_unlock(zd, lr->lr_foid);
1650 
1651         return (0);
1652 }
1653 
1654 static int
1655 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1656 {
1657         objset_t *os = zd->zd_os;
1658         dmu_tx_t *tx;
1659         uint64_t txg;
1660         rl_t *rl;
1661 
1662         if (byteswap)
1663                 byteswap_uint64_array(lr, sizeof (*lr));
1664 
1665         ztest_object_lock(zd, lr->lr_foid, RL_READER);
1666         rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1667             RL_WRITER);
1668 
1669         tx = dmu_tx_create(os);
1670 
1671         dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1672 
1673         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1674         if (txg == 0) {
1675                 ztest_range_unlock(rl);
1676                 ztest_object_unlock(zd, lr->lr_foid);
1677                 return (ENOSPC);
1678         }
1679 
1680         VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1681             lr->lr_length, tx) == 0);
1682 
1683         (void) ztest_log_truncate(zd, tx, lr);
1684 
1685         dmu_tx_commit(tx);
1686 
1687         ztest_range_unlock(rl);
1688         ztest_object_unlock(zd, lr->lr_foid);
1689 
1690         return (0);
1691 }
1692 
1693 static int
1694 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1695 {
1696         objset_t *os = zd->zd_os;
1697         dmu_tx_t *tx;
1698         dmu_buf_t *db;
1699         ztest_block_tag_t *bbt;
1700         uint64_t txg, lrtxg, crtxg;
1701 
1702         if (byteswap)
1703                 byteswap_uint64_array(lr, sizeof (*lr));
1704 
1705         ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1706 
1707         VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1708 
1709         tx = dmu_tx_create(os);
1710         dmu_tx_hold_bonus(tx, lr->lr_foid);
1711 
1712         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1713         if (txg == 0) {
1714                 dmu_buf_rele(db, FTAG);
1715                 ztest_object_unlock(zd, lr->lr_foid);
1716                 return (ENOSPC);
1717         }
1718 
1719         bbt = ztest_bt_bonus(db);
1720         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1721         crtxg = bbt->bt_crtxg;
1722         lrtxg = lr->lr_common.lrc_txg;
1723 
1724         if (zd->zd_zilog->zl_replay) {
1725                 ASSERT(lr->lr_size != 0);
1726                 ASSERT(lr->lr_mode != 0);
1727                 ASSERT(lrtxg != 0);
1728         } else {
1729                 /*
1730                  * Randomly change the size and increment the generation.
1731                  */
1732                 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1733                     sizeof (*bbt);
1734                 lr->lr_mode = bbt->bt_gen + 1;
1735                 ASSERT(lrtxg == 0);
1736         }
1737 
1738         /*
1739          * Verify that the current bonus buffer is not newer than our txg.
1740          */
1741         ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1742             MAX(txg, lrtxg), crtxg);
1743 
1744         dmu_buf_will_dirty(db, tx);
1745 
1746         ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1747         ASSERT3U(lr->lr_size, <=, db->db_size);
1748         VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1749         bbt = ztest_bt_bonus(db);
1750 
1751         ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1752 
1753         dmu_buf_rele(db, FTAG);
1754 
1755         (void) ztest_log_setattr(zd, tx, lr);
1756 
1757         dmu_tx_commit(tx);
1758 
1759         ztest_object_unlock(zd, lr->lr_foid);
1760 
1761         return (0);
1762 }
1763 
1764 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1765         NULL,                   /* 0 no such transaction type */
1766         ztest_replay_create,    /* TX_CREATE */
1767         NULL,                   /* TX_MKDIR */
1768         NULL,                   /* TX_MKXATTR */
1769         NULL,                   /* TX_SYMLINK */
1770         ztest_replay_remove,    /* TX_REMOVE */
1771         NULL,                   /* TX_RMDIR */
1772         NULL,                   /* TX_LINK */
1773         NULL,                   /* TX_RENAME */
1774         ztest_replay_write,     /* TX_WRITE */
1775         ztest_replay_truncate,  /* TX_TRUNCATE */
1776         ztest_replay_setattr,   /* TX_SETATTR */
1777         NULL,                   /* TX_ACL */
1778         NULL,                   /* TX_CREATE_ACL */
1779         NULL,                   /* TX_CREATE_ATTR */
1780         NULL,                   /* TX_CREATE_ACL_ATTR */
1781         NULL,                   /* TX_MKDIR_ACL */
1782         NULL,                   /* TX_MKDIR_ATTR */
1783         NULL,                   /* TX_MKDIR_ACL_ATTR */
1784         NULL,                   /* TX_WRITE2 */
1785 };
1786 
1787 /*
1788  * ZIL get_data callbacks
1789  */
1790 
1791 static void
1792 ztest_get_done(zgd_t *zgd, int error)
1793 {
1794         ztest_ds_t *zd = zgd->zgd_private;
1795         uint64_t object = zgd->zgd_rl->rl_object;
1796 
1797         if (zgd->zgd_db)
1798                 dmu_buf_rele(zgd->zgd_db, zgd);
1799 
1800         ztest_range_unlock(zgd->zgd_rl);
1801         ztest_object_unlock(zd, object);
1802 
1803         if (error == 0 && zgd->zgd_bp)
1804                 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1805 
1806         umem_free(zgd, sizeof (*zgd));
1807 }
1808 
1809 static int
1810 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1811 {
1812         ztest_ds_t *zd = arg;
1813         objset_t *os = zd->zd_os;
1814         uint64_t object = lr->lr_foid;
1815         uint64_t offset = lr->lr_offset;
1816         uint64_t size = lr->lr_length;
1817         blkptr_t *bp = &lr->lr_blkptr;
1818         uint64_t txg = lr->lr_common.lrc_txg;
1819         uint64_t crtxg;
1820         dmu_object_info_t doi;
1821         dmu_buf_t *db;
1822         zgd_t *zgd;
1823         int error;
1824 
1825         ztest_object_lock(zd, object, RL_READER);
1826         error = dmu_bonus_hold(os, object, FTAG, &db);
1827         if (error) {
1828                 ztest_object_unlock(zd, object);
1829                 return (error);
1830         }
1831 
1832         crtxg = ztest_bt_bonus(db)->bt_crtxg;
1833 
1834         if (crtxg == 0 || crtxg > txg) {
1835                 dmu_buf_rele(db, FTAG);
1836                 ztest_object_unlock(zd, object);
1837                 return (ENOENT);
1838         }
1839 
1840         dmu_object_info_from_db(db, &doi);
1841         dmu_buf_rele(db, FTAG);
1842         db = NULL;
1843 
1844         zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1845         zgd->zgd_zilog = zd->zd_zilog;
1846         zgd->zgd_private = zd;
1847 
1848         if (buf != NULL) {      /* immediate write */
1849                 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1850                     RL_READER);
1851 
1852                 error = dmu_read(os, object, offset, size, buf,
1853                     DMU_READ_NO_PREFETCH);
1854                 ASSERT(error == 0);
1855         } else {
1856                 size = doi.doi_data_block_size;
1857                 if (ISP2(size)) {
1858                         offset = P2ALIGN(offset, size);
1859                 } else {
1860                         ASSERT(offset < size);
1861                         offset = 0;
1862                 }
1863 
1864                 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1865                     RL_READER);
1866 
1867                 error = dmu_buf_hold(os, object, offset, zgd, &db,
1868                     DMU_READ_NO_PREFETCH);
1869 
1870                 if (error == 0) {
1871                         blkptr_t *obp = dmu_buf_get_blkptr(db);
1872                         if (obp) {
1873                                 ASSERT(BP_IS_HOLE(bp));
1874                                 *bp = *obp;
1875                         }
1876 
1877                         zgd->zgd_db = db;
1878                         zgd->zgd_bp = bp;
1879 
1880                         ASSERT(db->db_offset == offset);
1881                         ASSERT(db->db_size == size);
1882 
1883                         error = dmu_sync(zio, lr->lr_common.lrc_txg,
1884                             ztest_get_done, zgd);
1885 
1886                         if (error == 0)
1887                                 return (0);
1888                 }
1889         }
1890 
1891         ztest_get_done(zgd, error);
1892 
1893         return (error);
1894 }
1895 
1896 static void *
1897 ztest_lr_alloc(size_t lrsize, char *name)
1898 {
1899         char *lr;
1900         size_t namesize = name ? strlen(name) + 1 : 0;
1901 
1902         lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1903 
1904         if (name)
1905                 bcopy(name, lr + lrsize, namesize);
1906 
1907         return (lr);
1908 }
1909 
1910 void
1911 ztest_lr_free(void *lr, size_t lrsize, char *name)
1912 {
1913         size_t namesize = name ? strlen(name) + 1 : 0;
1914 
1915         umem_free(lr, lrsize + namesize);
1916 }
1917 
1918 /*
1919  * Lookup a bunch of objects.  Returns the number of objects not found.
1920  */
1921 static int
1922 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1923 {
1924         int missing = 0;
1925         int error;
1926 
1927         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1928 
1929         for (int i = 0; i < count; i++, od++) {
1930                 od->od_object = 0;
1931                 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1932                     sizeof (uint64_t), 1, &od->od_object);
1933                 if (error) {
1934                         ASSERT(error == ENOENT);
1935                         ASSERT(od->od_object == 0);
1936                         missing++;
1937                 } else {
1938                         dmu_buf_t *db;
1939                         ztest_block_tag_t *bbt;
1940                         dmu_object_info_t doi;
1941 
1942                         ASSERT(od->od_object != 0);
1943                         ASSERT(missing == 0);   /* there should be no gaps */
1944 
1945                         ztest_object_lock(zd, od->od_object, RL_READER);
1946                         VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1947                             od->od_object, FTAG, &db));
1948                         dmu_object_info_from_db(db, &doi);
1949                         bbt = ztest_bt_bonus(db);
1950                         ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1951                         od->od_type = doi.doi_type;
1952                         od->od_blocksize = doi.doi_data_block_size;
1953                         od->od_gen = bbt->bt_gen;
1954                         dmu_buf_rele(db, FTAG);
1955                         ztest_object_unlock(zd, od->od_object);
1956                 }
1957         }
1958 
1959         return (missing);
1960 }
1961 
1962 static int
1963 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1964 {
1965         int missing = 0;
1966 
1967         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1968 
1969         for (int i = 0; i < count; i++, od++) {
1970                 if (missing) {
1971                         od->od_object = 0;
1972                         missing++;
1973                         continue;
1974                 }
1975 
1976                 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1977 
1978                 lr->lr_doid = od->od_dir;
1979                 lr->lr_foid = 0;     /* 0 to allocate, > 0 to claim */
1980                 lr->lrz_type = od->od_crtype;
1981                 lr->lrz_blocksize = od->od_crblocksize;
1982                 lr->lrz_ibshift = ztest_random_ibshift();
1983                 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1984                 lr->lrz_bonuslen = dmu_bonus_max();
1985                 lr->lr_gen = od->od_crgen;
1986                 lr->lr_crtime[0] = time(NULL);
1987 
1988                 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1989                         ASSERT(missing == 0);
1990                         od->od_object = 0;
1991                         missing++;
1992                 } else {
1993                         od->od_object = lr->lr_foid;
1994                         od->od_type = od->od_crtype;
1995                         od->od_blocksize = od->od_crblocksize;
1996                         od->od_gen = od->od_crgen;
1997                         ASSERT(od->od_object != 0);
1998                 }
1999 
2000                 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2001         }
2002 
2003         return (missing);
2004 }
2005 
2006 static int
2007 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2008 {
2009         int missing = 0;
2010         int error;
2011 
2012         ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2013 
2014         od += count - 1;
2015 
2016         for (int i = count - 1; i >= 0; i--, od--) {
2017                 if (missing) {
2018                         missing++;
2019                         continue;
2020                 }
2021 
2022                 /*
2023                  * No object was found.
2024                  */
2025                 if (od->od_object == 0)
2026                         continue;
2027 
2028                 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2029 
2030                 lr->lr_doid = od->od_dir;
2031 
2032                 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2033                         ASSERT3U(error, ==, ENOSPC);
2034                         missing++;
2035                 } else {
2036                         od->od_object = 0;
2037                 }
2038                 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2039         }
2040 
2041         return (missing);
2042 }
2043 
2044 static int
2045 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2046     void *data)
2047 {
2048         lr_write_t *lr;
2049         int error;
2050 
2051         lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2052 
2053         lr->lr_foid = object;
2054         lr->lr_offset = offset;
2055         lr->lr_length = size;
2056         lr->lr_blkoff = 0;
2057         BP_ZERO(&lr->lr_blkptr);
2058 
2059         bcopy(data, lr + 1, size);
2060 
2061         error = ztest_replay_write(zd, lr, B_FALSE);
2062 
2063         ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2064 
2065         return (error);
2066 }
2067 
2068 static int
2069 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2070 {
2071         lr_truncate_t *lr;
2072         int error;
2073 
2074         lr = ztest_lr_alloc(sizeof (*lr), NULL);
2075 
2076         lr->lr_foid = object;
2077         lr->lr_offset = offset;
2078         lr->lr_length = size;
2079 
2080         error = ztest_replay_truncate(zd, lr, B_FALSE);
2081 
2082         ztest_lr_free(lr, sizeof (*lr), NULL);
2083 
2084         return (error);
2085 }
2086 
2087 static int
2088 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2089 {
2090         lr_setattr_t *lr;
2091         int error;
2092 
2093         lr = ztest_lr_alloc(sizeof (*lr), NULL);
2094 
2095         lr->lr_foid = object;
2096         lr->lr_size = 0;
2097         lr->lr_mode = 0;
2098 
2099         error = ztest_replay_setattr(zd, lr, B_FALSE);
2100 
2101         ztest_lr_free(lr, sizeof (*lr), NULL);
2102 
2103         return (error);
2104 }
2105 
2106 static void
2107 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2108 {
2109         objset_t *os = zd->zd_os;
2110         dmu_tx_t *tx;
2111         uint64_t txg;
2112         rl_t *rl;
2113 
2114         txg_wait_synced(dmu_objset_pool(os), 0);
2115 
2116         ztest_object_lock(zd, object, RL_READER);
2117         rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2118 
2119         tx = dmu_tx_create(os);
2120 
2121         dmu_tx_hold_write(tx, object, offset, size);
2122 
2123         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2124 
2125         if (txg != 0) {
2126                 dmu_prealloc(os, object, offset, size, tx);
2127                 dmu_tx_commit(tx);
2128                 txg_wait_synced(dmu_objset_pool(os), txg);
2129         } else {
2130                 (void) dmu_free_long_range(os, object, offset, size);
2131         }
2132 
2133         ztest_range_unlock(rl);
2134         ztest_object_unlock(zd, object);
2135 }
2136 
2137 static void
2138 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2139 {
2140         int err;
2141         ztest_block_tag_t wbt;
2142         dmu_object_info_t doi;
2143         enum ztest_io_type io_type;
2144         uint64_t blocksize;
2145         void *data;
2146 
2147         VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2148         blocksize = doi.doi_data_block_size;
2149         data = umem_alloc(blocksize, UMEM_NOFAIL);
2150 
2151         /*
2152          * Pick an i/o type at random, biased toward writing block tags.
2153          */
2154         io_type = ztest_random(ZTEST_IO_TYPES);
2155         if (ztest_random(2) == 0)
2156                 io_type = ZTEST_IO_WRITE_TAG;
2157 
2158         (void) rw_rdlock(&zd->zd_zilog_lock);
2159 
2160         switch (io_type) {
2161 
2162         case ZTEST_IO_WRITE_TAG:
2163                 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2164                 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2165                 break;
2166 
2167         case ZTEST_IO_WRITE_PATTERN:
2168                 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2169                 if (ztest_random(2) == 0) {
2170                         /*
2171                          * Induce fletcher2 collisions to ensure that
2172                          * zio_ddt_collision() detects and resolves them
2173                          * when using fletcher2-verify for deduplication.
2174                          */
2175                         ((uint64_t *)data)[0] ^= 1ULL << 63;
2176                         ((uint64_t *)data)[4] ^= 1ULL << 63;
2177                 }
2178                 (void) ztest_write(zd, object, offset, blocksize, data);
2179                 break;
2180 
2181         case ZTEST_IO_WRITE_ZEROES:
2182                 bzero(data, blocksize);
2183                 (void) ztest_write(zd, object, offset, blocksize, data);
2184                 break;
2185 
2186         case ZTEST_IO_TRUNCATE:
2187                 (void) ztest_truncate(zd, object, offset, blocksize);
2188                 break;
2189 
2190         case ZTEST_IO_SETATTR:
2191                 (void) ztest_setattr(zd, object);
2192                 break;
2193 
2194         case ZTEST_IO_REWRITE:
2195                 (void) rw_rdlock(&ztest_name_lock);
2196                 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2197                     ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2198                     B_FALSE);
2199                 VERIFY(err == 0 || err == ENOSPC);
2200                 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2201                     ZFS_PROP_COMPRESSION,
2202                     ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2203                     B_FALSE);
2204                 VERIFY(err == 0 || err == ENOSPC);
2205                 (void) rw_unlock(&ztest_name_lock);
2206 
2207                 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2208                     DMU_READ_NO_PREFETCH));
2209 
2210                 (void) ztest_write(zd, object, offset, blocksize, data);
2211                 break;
2212         }
2213 
2214         (void) rw_unlock(&zd->zd_zilog_lock);
2215 
2216         umem_free(data, blocksize);
2217 }
2218 
2219 /*
2220  * Initialize an object description template.
2221  */
2222 static void
2223 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2224     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2225 {
2226         od->od_dir = ZTEST_DIROBJ;
2227         od->od_object = 0;
2228 
2229         od->od_crtype = type;
2230         od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2231         od->od_crgen = gen;
2232 
2233         od->od_type = DMU_OT_NONE;
2234         od->od_blocksize = 0;
2235         od->od_gen = 0;
2236 
2237         (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2238             tag, (int64_t)id, index);
2239 }
2240 
2241 /*
2242  * Lookup or create the objects for a test using the od template.
2243  * If the objects do not all exist, or if 'remove' is specified,
2244  * remove any existing objects and create new ones.  Otherwise,
2245  * use the existing objects.
2246  */
2247 static int
2248 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2249 {
2250         int count = size / sizeof (*od);
2251         int rv = 0;
2252 
2253         VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2254         if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2255             (ztest_remove(zd, od, count) != 0 ||
2256             ztest_create(zd, od, count) != 0))
2257                 rv = -1;
2258         zd->zd_od = od;
2259         VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2260 
2261         return (rv);
2262 }
2263 
2264 /* ARGSUSED */
2265 void
2266 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2267 {
2268         zilog_t *zilog = zd->zd_zilog;
2269 
2270         (void) rw_rdlock(&zd->zd_zilog_lock);
2271 
2272         zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2273 
2274         /*
2275          * Remember the committed values in zd, which is in parent/child
2276          * shared memory.  If we die, the next iteration of ztest_run()
2277          * will verify that the log really does contain this record.
2278          */
2279         mutex_enter(&zilog->zl_lock);
2280         ASSERT(zd->zd_shared != NULL);
2281         ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2282         zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2283         mutex_exit(&zilog->zl_lock);
2284 
2285         (void) rw_unlock(&zd->zd_zilog_lock);
2286 }
2287 
2288 /*
2289  * This function is designed to simulate the operations that occur during a
2290  * mount/unmount operation.  We hold the dataset across these operations in an
2291  * attempt to expose any implicit assumptions about ZIL management.
2292  */
2293 /* ARGSUSED */
2294 void
2295 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2296 {
2297         objset_t *os = zd->zd_os;
2298 
2299         /*
2300          * We grab the zd_dirobj_lock to ensure that no other thread is
2301          * updating the zil (i.e. adding in-memory log records) and the
2302          * zd_zilog_lock to block any I/O.
2303          */
2304         VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2305         (void) rw_wrlock(&zd->zd_zilog_lock);
2306 
2307         /* zfsvfs_teardown() */
2308         zil_close(zd->zd_zilog);
2309 
2310         /* zfsvfs_setup() */
2311         VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2312         zil_replay(os, zd, ztest_replay_vector);
2313 
2314         (void) rw_unlock(&zd->zd_zilog_lock);
2315         VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2316 }
2317 
2318 /*
2319  * Verify that we can't destroy an active pool, create an existing pool,
2320  * or create a pool with a bad vdev spec.
2321  */
2322 /* ARGSUSED */
2323 void
2324 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2325 {
2326         ztest_shared_opts_t *zo = &ztest_opts;
2327         spa_t *spa;
2328         nvlist_t *nvroot;
2329 
2330         /*
2331          * Attempt to create using a bad file.
2332          */
2333         nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2334         VERIFY3U(ENOENT, ==,
2335             spa_create("ztest_bad_file", nvroot, NULL, NULL));
2336         nvlist_free(nvroot);
2337 
2338         /*
2339          * Attempt to create using a bad mirror.
2340          */
2341         nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2342         VERIFY3U(ENOENT, ==,
2343             spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2344         nvlist_free(nvroot);
2345 
2346         /*
2347          * Attempt to create an existing pool.  It shouldn't matter
2348          * what's in the nvroot; we should fail with EEXIST.
2349          */
2350         (void) rw_rdlock(&ztest_name_lock);
2351         nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2352         VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2353         nvlist_free(nvroot);
2354         VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2355         VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2356         spa_close(spa, FTAG);
2357 
2358         (void) rw_unlock(&ztest_name_lock);
2359 }
2360 
2361 /* ARGSUSED */
2362 void
2363 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2364 {
2365         spa_t *spa;
2366         uint64_t initial_version = SPA_VERSION_INITIAL;
2367         uint64_t version, newversion;
2368         nvlist_t *nvroot, *props;
2369         char *name;
2370 
2371         VERIFY0(mutex_lock(&ztest_vdev_lock));
2372         name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2373 
2374         /*
2375          * Clean up from previous runs.
2376          */
2377         (void) spa_destroy(name);
2378 
2379         nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2380             0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2381 
2382         /*
2383          * If we're configuring a RAIDZ device then make sure that the
2384          * the initial version is capable of supporting that feature.
2385          */
2386         switch (ztest_opts.zo_raidz_parity) {
2387         case 0:
2388         case 1:
2389                 initial_version = SPA_VERSION_INITIAL;
2390                 break;
2391         case 2:
2392                 initial_version = SPA_VERSION_RAIDZ2;
2393                 break;
2394         case 3:
2395                 initial_version = SPA_VERSION_RAIDZ3;
2396                 break;
2397         }
2398 
2399         /*
2400          * Create a pool with a spa version that can be upgraded. Pick
2401          * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2402          */
2403         do {
2404                 version = ztest_random_spa_version(initial_version);
2405         } while (version > SPA_VERSION_BEFORE_FEATURES);
2406 
2407         props = fnvlist_alloc();
2408         fnvlist_add_uint64(props,
2409             zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2410         VERIFY0(spa_create(name, nvroot, props, NULL));
2411         fnvlist_free(nvroot);
2412         fnvlist_free(props);
2413 
2414         VERIFY0(spa_open(name, &spa, FTAG));
2415         VERIFY3U(spa_version(spa), ==, version);
2416         newversion = ztest_random_spa_version(version + 1);
2417 
2418         if (ztest_opts.zo_verbose >= 4) {
2419                 (void) printf("upgrading spa version from %llu to %llu\n",
2420                     (u_longlong_t)version, (u_longlong_t)newversion);
2421         }
2422 
2423         spa_upgrade(spa, newversion);
2424         VERIFY3U(spa_version(spa), >, version);
2425         VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2426             zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2427         spa_close(spa, FTAG);
2428 
2429         strfree(name);
2430         VERIFY0(mutex_unlock(&ztest_vdev_lock));
2431 }
2432 
2433 static vdev_t *
2434 vdev_lookup_by_path(vdev_t *vd, const char *path)
2435 {
2436         vdev_t *mvd;
2437 
2438         if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2439                 return (vd);
2440 
2441         for (int c = 0; c < vd->vdev_children; c++)
2442                 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2443                     NULL)
2444                         return (mvd);
2445 
2446         return (NULL);
2447 }
2448 
2449 /*
2450  * Find the first available hole which can be used as a top-level.
2451  */
2452 int
2453 find_vdev_hole(spa_t *spa)
2454 {
2455         vdev_t *rvd = spa->spa_root_vdev;
2456         int c;
2457 
2458         ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2459 
2460         for (c = 0; c < rvd->vdev_children; c++) {
2461                 vdev_t *cvd = rvd->vdev_child[c];
2462 
2463                 if (cvd->vdev_ishole)
2464                         break;
2465         }
2466         return (c);
2467 }
2468 
2469 /*
2470  * Verify that vdev_add() works as expected.
2471  */
2472 /* ARGSUSED */
2473 void
2474 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2475 {
2476         ztest_shared_t *zs = ztest_shared;
2477         spa_t *spa = ztest_spa;
2478         uint64_t leaves;
2479         uint64_t guid;
2480         nvlist_t *nvroot;
2481         int error;
2482 
2483         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2484         leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2485 
2486         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2487 
2488         ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2489 
2490         /*
2491          * If we have slogs then remove them 1/4 of the time.
2492          */
2493         if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2494                 /*
2495                  * Grab the guid from the head of the log class rotor.
2496                  */
2497                 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2498 
2499                 spa_config_exit(spa, SCL_VDEV, FTAG);
2500 
2501                 /*
2502                  * We have to grab the zs_name_lock as writer to
2503                  * prevent a race between removing a slog (dmu_objset_find)
2504                  * and destroying a dataset. Removing the slog will
2505                  * grab a reference on the dataset which may cause
2506                  * dmu_objset_destroy() to fail with EBUSY thus
2507                  * leaving the dataset in an inconsistent state.
2508                  */
2509                 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2510                 error = spa_vdev_remove(spa, guid, B_FALSE);
2511                 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2512 
2513                 if (error && error != EEXIST)
2514                         fatal(0, "spa_vdev_remove() = %d", error);
2515         } else {
2516                 spa_config_exit(spa, SCL_VDEV, FTAG);
2517 
2518                 /*
2519                  * Make 1/4 of the devices be log devices.
2520                  */
2521                 nvroot = make_vdev_root(NULL, NULL, NULL,
2522                     ztest_opts.zo_vdev_size, 0,
2523                     ztest_random(4) == 0, ztest_opts.zo_raidz,
2524                     zs->zs_mirrors, 1);
2525 
2526                 error = spa_vdev_add(spa, nvroot);
2527                 nvlist_free(nvroot);
2528 
2529                 if (error == ENOSPC)
2530                         ztest_record_enospc("spa_vdev_add");
2531                 else if (error != 0)
2532                         fatal(0, "spa_vdev_add() = %d", error);
2533         }
2534 
2535         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2536 }
2537 
2538 /*
2539  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2540  */
2541 /* ARGSUSED */
2542 void
2543 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2544 {
2545         ztest_shared_t *zs = ztest_shared;
2546         spa_t *spa = ztest_spa;
2547         vdev_t *rvd = spa->spa_root_vdev;
2548         spa_aux_vdev_t *sav;
2549         char *aux;
2550         uint64_t guid = 0;
2551         int error;
2552 
2553         if (ztest_random(2) == 0) {
2554                 sav = &spa->spa_spares;
2555                 aux = ZPOOL_CONFIG_SPARES;
2556         } else {
2557                 sav = &spa->spa_l2cache;
2558                 aux = ZPOOL_CONFIG_L2CACHE;
2559         }
2560 
2561         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2562 
2563         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2564 
2565         if (sav->sav_count != 0 && ztest_random(4) == 0) {
2566                 /*
2567                  * Pick a random device to remove.
2568                  */
2569                 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2570         } else {
2571                 /*
2572                  * Find an unused device we can add.
2573                  */
2574                 zs->zs_vdev_aux = 0;
2575                 for (;;) {
2576                         char path[MAXPATHLEN];
2577                         int c;
2578                         (void) snprintf(path, sizeof (path), ztest_aux_template,
2579                             ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2580                             zs->zs_vdev_aux);
2581                         for (c = 0; c < sav->sav_count; c++)
2582                                 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2583                                     path) == 0)
2584                                         break;
2585                         if (c == sav->sav_count &&
2586                             vdev_lookup_by_path(rvd, path) == NULL)
2587                                 break;
2588                         zs->zs_vdev_aux++;
2589                 }
2590         }
2591 
2592         spa_config_exit(spa, SCL_VDEV, FTAG);
2593 
2594         if (guid == 0) {
2595                 /*
2596                  * Add a new device.
2597                  */
2598                 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2599                     (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2600                 error = spa_vdev_add(spa, nvroot);
2601                 if (error != 0)
2602                         fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2603                 nvlist_free(nvroot);
2604         } else {
2605                 /*
2606                  * Remove an existing device.  Sometimes, dirty its
2607                  * vdev state first to make sure we handle removal
2608                  * of devices that have pending state changes.
2609                  */
2610                 if (ztest_random(2) == 0)
2611                         (void) vdev_online(spa, guid, 0, NULL);
2612 
2613                 error = spa_vdev_remove(spa, guid, B_FALSE);
2614                 if (error != 0 && error != EBUSY)
2615                         fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2616         }
2617 
2618         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2619 }
2620 
2621 /*
2622  * split a pool if it has mirror tlvdevs
2623  */
2624 /* ARGSUSED */
2625 void
2626 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2627 {
2628         ztest_shared_t *zs = ztest_shared;
2629         spa_t *spa = ztest_spa;
2630         vdev_t *rvd = spa->spa_root_vdev;
2631         nvlist_t *tree, **child, *config, *split, **schild;
2632         uint_t c, children, schildren = 0, lastlogid = 0;
2633         int error = 0;
2634 
2635         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2636 
2637         /* ensure we have a useable config; mirrors of raidz aren't supported */
2638         if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2639                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2640                 return;
2641         }
2642 
2643         /* clean up the old pool, if any */
2644         (void) spa_destroy("splitp");
2645 
2646         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2647 
2648         /* generate a config from the existing config */
2649         mutex_enter(&spa->spa_props_lock);
2650         VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2651             &tree) == 0);
2652         mutex_exit(&spa->spa_props_lock);
2653 
2654         VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2655             &children) == 0);
2656 
2657         schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2658         for (c = 0; c < children; c++) {
2659                 vdev_t *tvd = rvd->vdev_child[c];
2660                 nvlist_t **mchild;
2661                 uint_t mchildren;
2662 
2663                 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2664                         VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2665                             0) == 0);
2666                         VERIFY(nvlist_add_string(schild[schildren],
2667                             ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2668                         VERIFY(nvlist_add_uint64(schild[schildren],
2669                             ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2670                         if (lastlogid == 0)
2671                                 lastlogid = schildren;
2672                         ++schildren;
2673                         continue;
2674                 }
2675                 lastlogid = 0;
2676                 VERIFY(nvlist_lookup_nvlist_array(child[c],
2677                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2678                 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2679         }
2680 
2681         /* OK, create a config that can be used to split */
2682         VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2683         VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2684             VDEV_TYPE_ROOT) == 0);
2685         VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2686             lastlogid != 0 ? lastlogid : schildren) == 0);
2687 
2688         VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2689         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2690 
2691         for (c = 0; c < schildren; c++)
2692                 nvlist_free(schild[c]);
2693         free(schild);
2694         nvlist_free(split);
2695 
2696         spa_config_exit(spa, SCL_VDEV, FTAG);
2697 
2698         (void) rw_wrlock(&ztest_name_lock);
2699         error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2700         (void) rw_unlock(&ztest_name_lock);
2701 
2702         nvlist_free(config);
2703 
2704         if (error == 0) {
2705                 (void) printf("successful split - results:\n");
2706                 mutex_enter(&spa_namespace_lock);
2707                 show_pool_stats(spa);
2708                 show_pool_stats(spa_lookup("splitp"));
2709                 mutex_exit(&spa_namespace_lock);
2710                 ++zs->zs_splits;
2711                 --zs->zs_mirrors;
2712         }
2713         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2714 
2715 }
2716 
2717 /*
2718  * Verify that we can attach and detach devices.
2719  */
2720 /* ARGSUSED */
2721 void
2722 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2723 {
2724         ztest_shared_t *zs = ztest_shared;
2725         spa_t *spa = ztest_spa;
2726         spa_aux_vdev_t *sav = &spa->spa_spares;
2727         vdev_t *rvd = spa->spa_root_vdev;
2728         vdev_t *oldvd, *newvd, *pvd;
2729         nvlist_t *root;
2730         uint64_t leaves;
2731         uint64_t leaf, top;
2732         uint64_t ashift = ztest_get_ashift();
2733         uint64_t oldguid, pguid;
2734         size_t oldsize, newsize;
2735         char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2736         int replacing;
2737         int oldvd_has_siblings = B_FALSE;
2738         int newvd_is_spare = B_FALSE;
2739         int oldvd_is_log;
2740         int error, expected_error;
2741 
2742         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2743         leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2744 
2745         spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2746 
2747         /*
2748          * Decide whether to do an attach or a replace.
2749          */
2750         replacing = ztest_random(2);
2751 
2752         /*
2753          * Pick a random top-level vdev.
2754          */
2755         top = ztest_random_vdev_top(spa, B_TRUE);
2756 
2757         /*
2758          * Pick a random leaf within it.
2759          */
2760         leaf = ztest_random(leaves);
2761 
2762         /*
2763          * Locate this vdev.
2764          */
2765         oldvd = rvd->vdev_child[top];
2766         if (zs->zs_mirrors >= 1) {
2767                 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2768                 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2769                 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2770         }
2771         if (ztest_opts.zo_raidz > 1) {
2772                 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2773                 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2774                 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2775         }
2776 
2777         /*
2778          * If we're already doing an attach or replace, oldvd may be a
2779          * mirror vdev -- in which case, pick a random child.
2780          */
2781         while (oldvd->vdev_children != 0) {
2782                 oldvd_has_siblings = B_TRUE;
2783                 ASSERT(oldvd->vdev_children >= 2);
2784                 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2785         }
2786 
2787         oldguid = oldvd->vdev_guid;
2788         oldsize = vdev_get_min_asize(oldvd);
2789         oldvd_is_log = oldvd->vdev_top->vdev_islog;
2790         (void) strcpy(oldpath, oldvd->vdev_path);
2791         pvd = oldvd->vdev_parent;
2792         pguid = pvd->vdev_guid;
2793 
2794         /*
2795          * If oldvd has siblings, then half of the time, detach it.
2796          */
2797         if (oldvd_has_siblings && ztest_random(2) == 0) {
2798                 spa_config_exit(spa, SCL_VDEV, FTAG);
2799                 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2800                 if (error != 0 && error != ENODEV && error != EBUSY &&
2801                     error != ENOTSUP)
2802                         fatal(0, "detach (%s) returned %d", oldpath, error);
2803                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2804                 return;
2805         }
2806 
2807         /*
2808          * For the new vdev, choose with equal probability between the two
2809          * standard paths (ending in either 'a' or 'b') or a random hot spare.
2810          */
2811         if (sav->sav_count != 0 && ztest_random(3) == 0) {
2812                 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2813                 newvd_is_spare = B_TRUE;
2814                 (void) strcpy(newpath, newvd->vdev_path);
2815         } else {
2816                 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2817                     ztest_opts.zo_dir, ztest_opts.zo_pool,
2818                     top * leaves + leaf);
2819                 if (ztest_random(2) == 0)
2820                         newpath[strlen(newpath) - 1] = 'b';
2821                 newvd = vdev_lookup_by_path(rvd, newpath);
2822         }
2823 
2824         if (newvd) {
2825                 newsize = vdev_get_min_asize(newvd);
2826         } else {
2827                 /*
2828                  * Make newsize a little bigger or smaller than oldsize.
2829                  * If it's smaller, the attach should fail.
2830                  * If it's larger, and we're doing a replace,
2831                  * we should get dynamic LUN growth when we're done.
2832                  */
2833                 newsize = 10 * oldsize / (9 + ztest_random(3));
2834         }
2835 
2836         /*
2837          * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2838          * unless it's a replace; in that case any non-replacing parent is OK.
2839          *
2840          * If newvd is already part of the pool, it should fail with EBUSY.
2841          *
2842          * If newvd is too small, it should fail with EOVERFLOW.
2843          */
2844         if (pvd->vdev_ops != &vdev_mirror_ops &&
2845             pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2846             pvd->vdev_ops == &vdev_replacing_ops ||
2847             pvd->vdev_ops == &vdev_spare_ops))
2848                 expected_error = ENOTSUP;
2849         else if (newvd_is_spare && (!replacing || oldvd_is_log))
2850                 expected_error = ENOTSUP;
2851         else if (newvd == oldvd)
2852                 expected_error = replacing ? 0 : EBUSY;
2853         else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2854                 expected_error = EBUSY;
2855         else if (newsize < oldsize)
2856                 expected_error = EOVERFLOW;
2857         else if (ashift > oldvd->vdev_top->vdev_ashift)
2858                 expected_error = EDOM;
2859         else
2860                 expected_error = 0;
2861 
2862         spa_config_exit(spa, SCL_VDEV, FTAG);
2863 
2864         /*
2865          * Build the nvlist describing newpath.
2866          */
2867         root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2868             ashift, 0, 0, 0, 1);
2869 
2870         error = spa_vdev_attach(spa, oldguid, root, replacing);
2871 
2872         nvlist_free(root);
2873 
2874         /*
2875          * If our parent was the replacing vdev, but the replace completed,
2876          * then instead of failing with ENOTSUP we may either succeed,
2877          * fail with ENODEV, or fail with EOVERFLOW.
2878          */
2879         if (expected_error == ENOTSUP &&
2880             (error == 0 || error == ENODEV || error == EOVERFLOW))
2881                 expected_error = error;
2882 
2883         /*
2884          * If someone grew the LUN, the replacement may be too small.
2885          */
2886         if (error == EOVERFLOW || error == EBUSY)
2887                 expected_error = error;
2888 
2889         /* XXX workaround 6690467 */
2890         if (error != expected_error && expected_error != EBUSY) {
2891                 fatal(0, "attach (%s %llu, %s %llu, %d) "
2892                     "returned %d, expected %d",
2893                     oldpath, (longlong_t)oldsize, newpath,
2894                     (longlong_t)newsize, replacing, error, expected_error);
2895         }
2896 
2897         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2898 }
2899 
2900 /*
2901  * Callback function which expands the physical size of the vdev.
2902  */
2903 vdev_t *
2904 grow_vdev(vdev_t *vd, void *arg)
2905 {
2906         spa_t *spa = vd->vdev_spa;
2907         size_t *newsize = arg;
2908         size_t fsize;
2909         int fd;
2910 
2911         ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2912         ASSERT(vd->vdev_ops->vdev_op_leaf);
2913 
2914         if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2915                 return (vd);
2916 
2917         fsize = lseek(fd, 0, SEEK_END);
2918         (void) ftruncate(fd, *newsize);
2919 
2920         if (ztest_opts.zo_verbose >= 6) {
2921                 (void) printf("%s grew from %lu to %lu bytes\n",
2922                     vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2923         }
2924         (void) close(fd);
2925         return (NULL);
2926 }
2927 
2928 /*
2929  * Callback function which expands a given vdev by calling vdev_online().
2930  */
2931 /* ARGSUSED */
2932 vdev_t *
2933 online_vdev(vdev_t *vd, void *arg)
2934 {
2935         spa_t *spa = vd->vdev_spa;
2936         vdev_t *tvd = vd->vdev_top;
2937         uint64_t guid = vd->vdev_guid;
2938         uint64_t generation = spa->spa_config_generation + 1;
2939         vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2940         int error;
2941 
2942         ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2943         ASSERT(vd->vdev_ops->vdev_op_leaf);
2944 
2945         /* Calling vdev_online will initialize the new metaslabs */
2946         spa_config_exit(spa, SCL_STATE, spa);
2947         error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2948         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2949 
2950         /*
2951          * If vdev_online returned an error or the underlying vdev_open
2952          * failed then we abort the expand. The only way to know that
2953          * vdev_open fails is by checking the returned newstate.
2954          */
2955         if (error || newstate != VDEV_STATE_HEALTHY) {
2956                 if (ztest_opts.zo_verbose >= 5) {
2957                         (void) printf("Unable to expand vdev, state %llu, "
2958                             "error %d\n", (u_longlong_t)newstate, error);
2959                 }
2960                 return (vd);
2961         }
2962         ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2963 
2964         /*
2965          * Since we dropped the lock we need to ensure that we're
2966          * still talking to the original vdev. It's possible this
2967          * vdev may have been detached/replaced while we were
2968          * trying to online it.
2969          */
2970         if (generation != spa->spa_config_generation) {
2971                 if (ztest_opts.zo_verbose >= 5) {
2972                         (void) printf("vdev configuration has changed, "
2973                             "guid %llu, state %llu, expected gen %llu, "
2974                             "got gen %llu\n",
2975                             (u_longlong_t)guid,
2976                             (u_longlong_t)tvd->vdev_state,
2977                             (u_longlong_t)generation,
2978                             (u_longlong_t)spa->spa_config_generation);
2979                 }
2980                 return (vd);
2981         }
2982         return (NULL);
2983 }
2984 
2985 /*
2986  * Traverse the vdev tree calling the supplied function.
2987  * We continue to walk the tree until we either have walked all
2988  * children or we receive a non-NULL return from the callback.
2989  * If a NULL callback is passed, then we just return back the first
2990  * leaf vdev we encounter.
2991  */
2992 vdev_t *
2993 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2994 {
2995         if (vd->vdev_ops->vdev_op_leaf) {
2996                 if (func == NULL)
2997                         return (vd);
2998                 else
2999                         return (func(vd, arg));
3000         }
3001 
3002         for (uint_t c = 0; c < vd->vdev_children; c++) {
3003                 vdev_t *cvd = vd->vdev_child[c];
3004                 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3005                         return (cvd);
3006         }
3007         return (NULL);
3008 }
3009 
3010 /*
3011  * Verify that dynamic LUN growth works as expected.
3012  */
3013 /* ARGSUSED */
3014 void
3015 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3016 {
3017         spa_t *spa = ztest_spa;
3018         vdev_t *vd, *tvd;
3019         metaslab_class_t *mc;
3020         metaslab_group_t *mg;
3021         size_t psize, newsize;
3022         uint64_t top;
3023         uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3024 
3025         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3026         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3027 
3028         top = ztest_random_vdev_top(spa, B_TRUE);
3029 
3030         tvd = spa->spa_root_vdev->vdev_child[top];
3031         mg = tvd->vdev_mg;
3032         mc = mg->mg_class;
3033         old_ms_count = tvd->vdev_ms_count;
3034         old_class_space = metaslab_class_get_space(mc);
3035 
3036         /*
3037          * Determine the size of the first leaf vdev associated with
3038          * our top-level device.
3039          */
3040         vd = vdev_walk_tree(tvd, NULL, NULL);
3041         ASSERT3P(vd, !=, NULL);
3042         ASSERT(vd->vdev_ops->vdev_op_leaf);
3043 
3044         psize = vd->vdev_psize;
3045 
3046         /*
3047          * We only try to expand the vdev if it's healthy, less than 4x its
3048          * original size, and it has a valid psize.
3049          */
3050         if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3051             psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3052                 spa_config_exit(spa, SCL_STATE, spa);
3053                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3054                 return;
3055         }
3056         ASSERT(psize > 0);
3057         newsize = psize + psize / 8;
3058         ASSERT3U(newsize, >, psize);
3059 
3060         if (ztest_opts.zo_verbose >= 6) {
3061                 (void) printf("Expanding LUN %s from %lu to %lu\n",
3062                     vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3063         }
3064 
3065         /*
3066          * Growing the vdev is a two step process:
3067          *      1). expand the physical size (i.e. relabel)
3068          *      2). online the vdev to create the new metaslabs
3069          */
3070         if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3071             vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3072             tvd->vdev_state != VDEV_STATE_HEALTHY) {
3073                 if (ztest_opts.zo_verbose >= 5) {
3074                         (void) printf("Could not expand LUN because "
3075                             "the vdev configuration changed.\n");
3076                 }
3077                 spa_config_exit(spa, SCL_STATE, spa);
3078                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3079                 return;
3080         }
3081 
3082         spa_config_exit(spa, SCL_STATE, spa);
3083 
3084         /*
3085          * Expanding the LUN will update the config asynchronously,
3086          * thus we must wait for the async thread to complete any
3087          * pending tasks before proceeding.
3088          */
3089         for (;;) {
3090                 boolean_t done;
3091                 mutex_enter(&spa->spa_async_lock);
3092                 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3093                 mutex_exit(&spa->spa_async_lock);
3094                 if (done)
3095                         break;
3096                 txg_wait_synced(spa_get_dsl(spa), 0);
3097                 (void) poll(NULL, 0, 100);
3098         }
3099 
3100         spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3101 
3102         tvd = spa->spa_root_vdev->vdev_child[top];
3103         new_ms_count = tvd->vdev_ms_count;
3104         new_class_space = metaslab_class_get_space(mc);
3105 
3106         if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3107                 if (ztest_opts.zo_verbose >= 5) {
3108                         (void) printf("Could not verify LUN expansion due to "
3109                             "intervening vdev offline or remove.\n");
3110                 }
3111                 spa_config_exit(spa, SCL_STATE, spa);
3112                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3113                 return;
3114         }
3115 
3116         /*
3117          * Make sure we were able to grow the vdev.
3118          */
3119         if (new_ms_count <= old_ms_count)
3120                 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3121                     old_ms_count, new_ms_count);
3122 
3123         /*
3124          * Make sure we were able to grow the pool.
3125          */
3126         if (new_class_space <= old_class_space)
3127                 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3128                     old_class_space, new_class_space);
3129 
3130         if (ztest_opts.zo_verbose >= 5) {
3131                 char oldnumbuf[6], newnumbuf[6];
3132 
3133                 nicenum(old_class_space, oldnumbuf);
3134                 nicenum(new_class_space, newnumbuf);
3135                 (void) printf("%s grew from %s to %s\n",
3136                     spa->spa_name, oldnumbuf, newnumbuf);
3137         }
3138 
3139         spa_config_exit(spa, SCL_STATE, spa);
3140         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3141 }
3142 
3143 /*
3144  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3145  */
3146 /* ARGSUSED */
3147 static void
3148 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3149 {
3150         /*
3151          * Create the objects common to all ztest datasets.
3152          */
3153         VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3154             DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3155 }
3156 
3157 static int
3158 ztest_dataset_create(char *dsname)
3159 {
3160         uint64_t zilset = ztest_random(100);
3161         int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3162             ztest_objset_create_cb, NULL);
3163 
3164         if (err || zilset < 80)
3165                 return (err);
3166 
3167         if (ztest_opts.zo_verbose >= 6)
3168                 (void) printf("Setting dataset %s to sync always\n", dsname);
3169         return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3170             ZFS_SYNC_ALWAYS, B_FALSE));
3171 }
3172 
3173 /* ARGSUSED */
3174 static int
3175 ztest_objset_destroy_cb(const char *name, void *arg)
3176 {
3177         objset_t *os;
3178         dmu_object_info_t doi;
3179         int error;
3180 
3181         /*
3182          * Verify that the dataset contains a directory object.
3183          */
3184         VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3185         error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3186         if (error != ENOENT) {
3187                 /* We could have crashed in the middle of destroying it */
3188                 ASSERT0(error);
3189                 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3190                 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3191         }
3192         dmu_objset_disown(os, FTAG);
3193 
3194         /*
3195          * Destroy the dataset.
3196          */
3197         if (strchr(name, '@') != NULL) {
3198                 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3199         } else {
3200                 VERIFY0(dsl_destroy_head(name));
3201         }
3202         return (0);
3203 }
3204 
3205 static boolean_t
3206 ztest_snapshot_create(char *osname, uint64_t id)
3207 {
3208         char snapname[MAXNAMELEN];
3209         int error;
3210 
3211         (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3212 
3213         error = dmu_objset_snapshot_one(osname, snapname);
3214         if (error == ENOSPC) {
3215                 ztest_record_enospc(FTAG);
3216                 return (B_FALSE);
3217         }
3218         if (error != 0 && error != EEXIST) {
3219                 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3220                     snapname, error);
3221         }
3222         return (B_TRUE);
3223 }
3224 
3225 static boolean_t
3226 ztest_snapshot_destroy(char *osname, uint64_t id)
3227 {
3228         char snapname[MAXNAMELEN];
3229         int error;
3230 
3231         (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3232             (u_longlong_t)id);
3233 
3234         error = dsl_destroy_snapshot(snapname, B_FALSE);
3235         if (error != 0 && error != ENOENT)
3236                 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3237         return (B_TRUE);
3238 }
3239 
3240 /* ARGSUSED */
3241 void
3242 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3243 {
3244         ztest_ds_t zdtmp;
3245         int iters;
3246         int error;
3247         objset_t *os, *os2;
3248         char name[MAXNAMELEN];
3249         zilog_t *zilog;
3250 
3251         (void) rw_rdlock(&ztest_name_lock);
3252 
3253         (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3254             ztest_opts.zo_pool, (u_longlong_t)id);
3255 
3256         /*
3257          * If this dataset exists from a previous run, process its replay log
3258          * half of the time.  If we don't replay it, then dmu_objset_destroy()
3259          * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3260          */
3261         if (ztest_random(2) == 0 &&
3262             dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3263                 ztest_zd_init(&zdtmp, NULL, os);
3264                 zil_replay(os, &zdtmp, ztest_replay_vector);
3265                 ztest_zd_fini(&zdtmp);
3266                 dmu_objset_disown(os, FTAG);
3267         }
3268 
3269         /*
3270          * There may be an old instance of the dataset we're about to
3271          * create lying around from a previous run.  If so, destroy it
3272          * and all of its snapshots.
3273          */
3274         (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3275             DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3276 
3277         /*
3278          * Verify that the destroyed dataset is no longer in the namespace.
3279          */
3280         VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3281             FTAG, &os));
3282 
3283         /*
3284          * Verify that we can create a new dataset.
3285          */
3286         error = ztest_dataset_create(name);
3287         if (error) {
3288                 if (error == ENOSPC) {
3289                         ztest_record_enospc(FTAG);
3290                         (void) rw_unlock(&ztest_name_lock);
3291                         return;
3292                 }
3293                 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3294         }
3295 
3296         VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3297 
3298         ztest_zd_init(&zdtmp, NULL, os);
3299 
3300         /*
3301          * Open the intent log for it.
3302          */
3303         zilog = zil_open(os, ztest_get_data);
3304 
3305         /*
3306          * Put some objects in there, do a little I/O to them,
3307          * and randomly take a couple of snapshots along the way.
3308          */
3309         iters = ztest_random(5);
3310         for (int i = 0; i < iters; i++) {
3311                 ztest_dmu_object_alloc_free(&zdtmp, id);
3312                 if (ztest_random(iters) == 0)
3313                         (void) ztest_snapshot_create(name, i);
3314         }
3315 
3316         /*
3317          * Verify that we cannot create an existing dataset.
3318          */
3319         VERIFY3U(EEXIST, ==,
3320             dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3321 
3322         /*
3323          * Verify that we can hold an objset that is also owned.
3324          */
3325         VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3326         dmu_objset_rele(os2, FTAG);
3327 
3328         /*
3329          * Verify that we cannot own an objset that is already owned.
3330          */
3331         VERIFY3U(EBUSY, ==,
3332             dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3333 
3334         zil_close(zilog);
3335         dmu_objset_disown(os, FTAG);
3336         ztest_zd_fini(&zdtmp);
3337 
3338         (void) rw_unlock(&ztest_name_lock);
3339 }
3340 
3341 /*
3342  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3343  */
3344 void
3345 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3346 {
3347         (void) rw_rdlock(&ztest_name_lock);
3348         (void) ztest_snapshot_destroy(zd->zd_name, id);
3349         (void) ztest_snapshot_create(zd->zd_name, id);
3350         (void) rw_unlock(&ztest_name_lock);
3351 }
3352 
3353 /*
3354  * Cleanup non-standard snapshots and clones.
3355  */
3356 void
3357 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3358 {
3359         char snap1name[MAXNAMELEN];
3360         char clone1name[MAXNAMELEN];
3361         char snap2name[MAXNAMELEN];
3362         char clone2name[MAXNAMELEN];
3363         char snap3name[MAXNAMELEN];
3364         int error;
3365 
3366         (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3367         (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3368         (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3369         (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3370         (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3371 
3372         error = dsl_destroy_head(clone2name);
3373         if (error && error != ENOENT)
3374                 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3375         error = dsl_destroy_snapshot(snap3name, B_FALSE);
3376         if (error && error != ENOENT)
3377                 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3378         error = dsl_destroy_snapshot(snap2name, B_FALSE);
3379         if (error && error != ENOENT)
3380                 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3381         error = dsl_destroy_head(clone1name);
3382         if (error && error != ENOENT)
3383                 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3384         error = dsl_destroy_snapshot(snap1name, B_FALSE);
3385         if (error && error != ENOENT)
3386                 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3387 }
3388 
3389 /*
3390  * Verify dsl_dataset_promote handles EBUSY
3391  */
3392 void
3393 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3394 {
3395         objset_t *os;
3396         char snap1name[MAXNAMELEN];
3397         char clone1name[MAXNAMELEN];
3398         char snap2name[MAXNAMELEN];
3399         char clone2name[MAXNAMELEN];
3400         char snap3name[MAXNAMELEN];
3401         char *osname = zd->zd_name;
3402         int error;
3403 
3404         (void) rw_rdlock(&ztest_name_lock);
3405 
3406         ztest_dsl_dataset_cleanup(osname, id);
3407 
3408         (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3409         (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3410         (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3411         (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3412         (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3413 
3414         error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3415         if (error && error != EEXIST) {
3416                 if (error == ENOSPC) {
3417                         ztest_record_enospc(FTAG);
3418                         goto out;
3419                 }
3420                 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3421         }
3422 
3423         error = dmu_objset_clone(clone1name, snap1name);
3424         if (error) {
3425                 if (error == ENOSPC) {
3426                         ztest_record_enospc(FTAG);
3427                         goto out;
3428                 }
3429                 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3430         }
3431 
3432         error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3433         if (error && error != EEXIST) {
3434                 if (error == ENOSPC) {
3435                         ztest_record_enospc(FTAG);
3436                         goto out;
3437                 }
3438                 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3439         }
3440 
3441         error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3442         if (error && error != EEXIST) {
3443                 if (error == ENOSPC) {
3444                         ztest_record_enospc(FTAG);
3445                         goto out;
3446                 }
3447                 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3448         }
3449 
3450         error = dmu_objset_clone(clone2name, snap3name);
3451         if (error) {
3452                 if (error == ENOSPC) {
3453                         ztest_record_enospc(FTAG);
3454                         goto out;
3455                 }
3456                 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3457         }
3458 
3459         error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3460         if (error)
3461                 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3462         error = dsl_dataset_promote(clone2name, NULL);
3463         if (error != EBUSY)
3464                 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3465                     error);
3466         dmu_objset_disown(os, FTAG);
3467 
3468 out:
3469         ztest_dsl_dataset_cleanup(osname, id);
3470 
3471         (void) rw_unlock(&ztest_name_lock);
3472 }
3473 
3474 /*
3475  * Verify that dmu_object_{alloc,free} work as expected.
3476  */
3477 void
3478 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3479 {
3480         ztest_od_t od[4];
3481         int batchsize = sizeof (od) / sizeof (od[0]);
3482 
3483         for (int b = 0; b < batchsize; b++)
3484                 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3485 
3486         /*
3487          * Destroy the previous batch of objects, create a new batch,
3488          * and do some I/O on the new objects.
3489          */
3490         if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3491                 return;
3492 
3493         while (ztest_random(4 * batchsize) != 0)
3494                 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3495                     ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3496 }
3497 
3498 /*
3499  * Verify that dmu_{read,write} work as expected.
3500  */
3501 void
3502 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3503 {
3504         objset_t *os = zd->zd_os;
3505         ztest_od_t od[2];
3506         dmu_tx_t *tx;
3507         int i, freeit, error;
3508         uint64_t n, s, txg;
3509         bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3510         uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3511         uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3512         uint64_t regions = 997;
3513         uint64_t stride = 123456789ULL;
3514         uint64_t width = 40;
3515         int free_percent = 5;
3516 
3517         /*
3518          * This test uses two objects, packobj and bigobj, that are always
3519          * updated together (i.e. in the same tx) so that their contents are
3520          * in sync and can be compared.  Their contents relate to each other
3521          * in a simple way: packobj is a dense array of 'bufwad' structures,
3522          * while bigobj is a sparse array of the same bufwads.  Specifically,
3523          * for any index n, there are three bufwads that should be identical:
3524          *
3525          *      packobj, at offset n * sizeof (bufwad_t)
3526          *      bigobj, at the head of the nth chunk
3527          *      bigobj, at the tail of the nth chunk
3528          *
3529          * The chunk size is arbitrary. It doesn't have to be a power of two,
3530          * and it doesn't have any relation to the object blocksize.
3531          * The only requirement is that it can hold at least two bufwads.
3532          *
3533          * Normally, we write the bufwad to each of these locations.
3534          * However, free_percent of the time we instead write zeroes to
3535          * packobj and perform a dmu_free_range() on bigobj.  By comparing
3536          * bigobj to packobj, we can verify that the DMU is correctly
3537          * tracking which parts of an object are allocated and free,
3538          * and that the contents of the allocated blocks are correct.
3539          */
3540 
3541         /*
3542          * Read the directory info.  If it's the first time, set things up.
3543          */
3544         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3545         ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3546 
3547         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3548                 return;
3549 
3550         bigobj = od[0].od_object;
3551         packobj = od[1].od_object;
3552         chunksize = od[0].od_gen;
3553         ASSERT(chunksize == od[1].od_gen);
3554 
3555         /*
3556          * Prefetch a random chunk of the big object.
3557          * Our aim here is to get some async reads in flight
3558          * for blocks that we may free below; the DMU should
3559          * handle this race correctly.
3560          */
3561         n = ztest_random(regions) * stride + ztest_random(width);
3562         s = 1 + ztest_random(2 * width - 1);
3563         dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3564 
3565         /*
3566          * Pick a random index and compute the offsets into packobj and bigobj.
3567          */
3568         n = ztest_random(regions) * stride + ztest_random(width);
3569         s = 1 + ztest_random(width - 1);
3570 
3571         packoff = n * sizeof (bufwad_t);
3572         packsize = s * sizeof (bufwad_t);
3573 
3574         bigoff = n * chunksize;
3575         bigsize = s * chunksize;
3576 
3577         packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3578         bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3579 
3580         /*
3581          * free_percent of the time, free a range of bigobj rather than
3582          * overwriting it.
3583          */
3584         freeit = (ztest_random(100) < free_percent);
3585 
3586         /*
3587          * Read the current contents of our objects.
3588          */
3589         error = dmu_read(os, packobj, packoff, packsize, packbuf,
3590             DMU_READ_PREFETCH);
3591         ASSERT0(error);
3592         error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3593             DMU_READ_PREFETCH);
3594         ASSERT0(error);
3595 
3596         /*
3597          * Get a tx for the mods to both packobj and bigobj.
3598          */
3599         tx = dmu_tx_create(os);
3600 
3601         dmu_tx_hold_write(tx, packobj, packoff, packsize);
3602 
3603         if (freeit)
3604                 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3605         else
3606                 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3607 
3608         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3609         if (txg == 0) {
3610                 umem_free(packbuf, packsize);
3611                 umem_free(bigbuf, bigsize);
3612                 return;
3613         }
3614 
3615         dmu_object_set_checksum(os, bigobj,
3616             (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3617 
3618         dmu_object_set_compress(os, bigobj,
3619             (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3620 
3621         /*
3622          * For each index from n to n + s, verify that the existing bufwad
3623          * in packobj matches the bufwads at the head and tail of the
3624          * corresponding chunk in bigobj.  Then update all three bufwads
3625          * with the new values we want to write out.
3626          */
3627         for (i = 0; i < s; i++) {
3628                 /* LINTED */
3629                 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3630                 /* LINTED */
3631                 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3632                 /* LINTED */
3633                 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3634 
3635                 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3636                 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3637 
3638                 if (pack->bw_txg > txg)
3639                         fatal(0, "future leak: got %llx, open txg is %llx",
3640                             pack->bw_txg, txg);
3641 
3642                 if (pack->bw_data != 0 && pack->bw_index != n + i)
3643                         fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3644                             pack->bw_index, n, i);
3645 
3646                 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3647                         fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3648 
3649                 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3650                         fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3651 
3652                 if (freeit) {
3653                         bzero(pack, sizeof (bufwad_t));
3654                 } else {
3655                         pack->bw_index = n + i;
3656                         pack->bw_txg = txg;
3657                         pack->bw_data = 1 + ztest_random(-2ULL);
3658                 }
3659                 *bigH = *pack;
3660                 *bigT = *pack;
3661         }
3662 
3663         /*
3664          * We've verified all the old bufwads, and made new ones.
3665          * Now write them out.
3666          */
3667         dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3668 
3669         if (freeit) {
3670                 if (ztest_opts.zo_verbose >= 7) {
3671                         (void) printf("freeing offset %llx size %llx"
3672                             " txg %llx\n",
3673                             (u_longlong_t)bigoff,
3674                             (u_longlong_t)bigsize,
3675                             (u_longlong_t)txg);
3676                 }
3677                 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3678         } else {
3679                 if (ztest_opts.zo_verbose >= 7) {
3680                         (void) printf("writing offset %llx size %llx"
3681                             " txg %llx\n",
3682                             (u_longlong_t)bigoff,
3683                             (u_longlong_t)bigsize,
3684                             (u_longlong_t)txg);
3685                 }
3686                 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3687         }
3688 
3689         dmu_tx_commit(tx);
3690 
3691         /*
3692          * Sanity check the stuff we just wrote.
3693          */
3694         {
3695                 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3696                 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3697 
3698                 VERIFY(0 == dmu_read(os, packobj, packoff,
3699                     packsize, packcheck, DMU_READ_PREFETCH));
3700                 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3701                     bigsize, bigcheck, DMU_READ_PREFETCH));
3702 
3703                 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3704                 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3705 
3706                 umem_free(packcheck, packsize);
3707                 umem_free(bigcheck, bigsize);
3708         }
3709 
3710         umem_free(packbuf, packsize);
3711         umem_free(bigbuf, bigsize);
3712 }
3713 
3714 void
3715 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3716     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3717 {
3718         uint64_t i;
3719         bufwad_t *pack;
3720         bufwad_t *bigH;
3721         bufwad_t *bigT;
3722 
3723         /*
3724          * For each index from n to n + s, verify that the existing bufwad
3725          * in packobj matches the bufwads at the head and tail of the
3726          * corresponding chunk in bigobj.  Then update all three bufwads
3727          * with the new values we want to write out.
3728          */
3729         for (i = 0; i < s; i++) {
3730                 /* LINTED */
3731                 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3732                 /* LINTED */
3733                 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3734                 /* LINTED */
3735                 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3736 
3737                 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3738                 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3739 
3740                 if (pack->bw_txg > txg)
3741                         fatal(0, "future leak: got %llx, open txg is %llx",
3742                             pack->bw_txg, txg);
3743 
3744                 if (pack->bw_data != 0 && pack->bw_index != n + i)
3745                         fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3746                             pack->bw_index, n, i);
3747 
3748                 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3749                         fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3750 
3751                 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3752                         fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3753 
3754                 pack->bw_index = n + i;
3755                 pack->bw_txg = txg;
3756                 pack->bw_data = 1 + ztest_random(-2ULL);
3757 
3758                 *bigH = *pack;
3759                 *bigT = *pack;
3760         }
3761 }
3762 
3763 void
3764 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3765 {
3766         objset_t *os = zd->zd_os;
3767         ztest_od_t od[2];
3768         dmu_tx_t *tx;
3769         uint64_t i;
3770         int error;
3771         uint64_t n, s, txg;
3772         bufwad_t *packbuf, *bigbuf;
3773         uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3774         uint64_t blocksize = ztest_random_blocksize();
3775         uint64_t chunksize = blocksize;
3776         uint64_t regions = 997;
3777         uint64_t stride = 123456789ULL;
3778         uint64_t width = 9;
3779         dmu_buf_t *bonus_db;
3780         arc_buf_t **bigbuf_arcbufs;
3781         dmu_object_info_t doi;
3782 
3783         /*
3784          * This test uses two objects, packobj and bigobj, that are always
3785          * updated together (i.e. in the same tx) so that their contents are
3786          * in sync and can be compared.  Their contents relate to each other
3787          * in a simple way: packobj is a dense array of 'bufwad' structures,
3788          * while bigobj is a sparse array of the same bufwads.  Specifically,
3789          * for any index n, there are three bufwads that should be identical:
3790          *
3791          *      packobj, at offset n * sizeof (bufwad_t)
3792          *      bigobj, at the head of the nth chunk
3793          *      bigobj, at the tail of the nth chunk
3794          *
3795          * The chunk size is set equal to bigobj block size so that
3796          * dmu_assign_arcbuf() can be tested for object updates.
3797          */
3798 
3799         /*
3800          * Read the directory info.  If it's the first time, set things up.
3801          */
3802         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3803         ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3804 
3805         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3806                 return;
3807 
3808         bigobj = od[0].od_object;
3809         packobj = od[1].od_object;
3810         blocksize = od[0].od_blocksize;
3811         chunksize = blocksize;
3812         ASSERT(chunksize == od[1].od_gen);
3813 
3814         VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3815         VERIFY(ISP2(doi.doi_data_block_size));
3816         VERIFY(chunksize == doi.doi_data_block_size);
3817         VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3818 
3819         /*
3820          * Pick a random index and compute the offsets into packobj and bigobj.
3821          */
3822         n = ztest_random(regions) * stride + ztest_random(width);
3823         s = 1 + ztest_random(width - 1);
3824 
3825         packoff = n * sizeof (bufwad_t);
3826         packsize = s * sizeof (bufwad_t);
3827 
3828         bigoff = n * chunksize;
3829         bigsize = s * chunksize;
3830 
3831         packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3832         bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3833 
3834         VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3835 
3836         bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3837 
3838         /*
3839          * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3840          * Iteration 1 test zcopy to already referenced dbufs.
3841          * Iteration 2 test zcopy to dirty dbuf in the same txg.
3842          * Iteration 3 test zcopy to dbuf dirty in previous txg.
3843          * Iteration 4 test zcopy when dbuf is no longer dirty.
3844          * Iteration 5 test zcopy when it can't be done.
3845          * Iteration 6 one more zcopy write.
3846          */
3847         for (i = 0; i < 7; i++) {
3848                 uint64_t j;
3849                 uint64_t off;
3850 
3851                 /*
3852                  * In iteration 5 (i == 5) use arcbufs
3853                  * that don't match bigobj blksz to test
3854                  * dmu_assign_arcbuf() when it can't directly
3855                  * assign an arcbuf to a dbuf.
3856                  */
3857                 for (j = 0; j < s; j++) {
3858                         if (i != 5) {
3859                                 bigbuf_arcbufs[j] =
3860                                     dmu_request_arcbuf(bonus_db, chunksize);
3861                         } else {
3862                                 bigbuf_arcbufs[2 * j] =
3863                                     dmu_request_arcbuf(bonus_db, chunksize / 2);
3864                                 bigbuf_arcbufs[2 * j + 1] =
3865                                     dmu_request_arcbuf(bonus_db, chunksize / 2);
3866                         }
3867                 }
3868 
3869                 /*
3870                  * Get a tx for the mods to both packobj and bigobj.
3871                  */
3872                 tx = dmu_tx_create(os);
3873 
3874                 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3875                 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3876 
3877                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3878                 if (txg == 0) {
3879                         umem_free(packbuf, packsize);
3880                         umem_free(bigbuf, bigsize);
3881                         for (j = 0; j < s; j++) {
3882                                 if (i != 5) {
3883                                         dmu_return_arcbuf(bigbuf_arcbufs[j]);
3884                                 } else {
3885                                         dmu_return_arcbuf(
3886                                             bigbuf_arcbufs[2 * j]);
3887                                         dmu_return_arcbuf(
3888                                             bigbuf_arcbufs[2 * j + 1]);
3889                                 }
3890                         }
3891                         umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3892                         dmu_buf_rele(bonus_db, FTAG);
3893                         return;
3894                 }
3895 
3896                 /*
3897                  * 50% of the time don't read objects in the 1st iteration to
3898                  * test dmu_assign_arcbuf() for the case when there're no
3899                  * existing dbufs for the specified offsets.
3900                  */
3901                 if (i != 0 || ztest_random(2) != 0) {
3902                         error = dmu_read(os, packobj, packoff,
3903                             packsize, packbuf, DMU_READ_PREFETCH);
3904                         ASSERT0(error);
3905                         error = dmu_read(os, bigobj, bigoff, bigsize,
3906                             bigbuf, DMU_READ_PREFETCH);
3907                         ASSERT0(error);
3908                 }
3909                 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3910                     n, chunksize, txg);
3911 
3912                 /*
3913                  * We've verified all the old bufwads, and made new ones.
3914                  * Now write them out.
3915                  */
3916                 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3917                 if (ztest_opts.zo_verbose >= 7) {
3918                         (void) printf("writing offset %llx size %llx"
3919                             " txg %llx\n",
3920                             (u_longlong_t)bigoff,
3921                             (u_longlong_t)bigsize,
3922                             (u_longlong_t)txg);
3923                 }
3924                 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3925                         dmu_buf_t *dbt;
3926                         if (i != 5) {
3927                                 bcopy((caddr_t)bigbuf + (off - bigoff),
3928                                     bigbuf_arcbufs[j]->b_data, chunksize);
3929                         } else {
3930                                 bcopy((caddr_t)bigbuf + (off - bigoff),
3931                                     bigbuf_arcbufs[2 * j]->b_data,
3932                                     chunksize / 2);
3933                                 bcopy((caddr_t)bigbuf + (off - bigoff) +
3934                                     chunksize / 2,
3935                                     bigbuf_arcbufs[2 * j + 1]->b_data,
3936                                     chunksize / 2);
3937                         }
3938 
3939                         if (i == 1) {
3940                                 VERIFY(dmu_buf_hold(os, bigobj, off,
3941                                     FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3942                         }
3943                         if (i != 5) {
3944                                 dmu_assign_arcbuf(bonus_db, off,
3945                                     bigbuf_arcbufs[j], tx);
3946                         } else {
3947                                 dmu_assign_arcbuf(bonus_db, off,
3948                                     bigbuf_arcbufs[2 * j], tx);
3949                                 dmu_assign_arcbuf(bonus_db,
3950                                     off + chunksize / 2,
3951                                     bigbuf_arcbufs[2 * j + 1], tx);
3952                         }
3953                         if (i == 1) {
3954                                 dmu_buf_rele(dbt, FTAG);
3955                         }
3956                 }
3957                 dmu_tx_commit(tx);
3958 
3959                 /*
3960                  * Sanity check the stuff we just wrote.
3961                  */
3962                 {
3963                         void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3964                         void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3965 
3966                         VERIFY(0 == dmu_read(os, packobj, packoff,
3967                             packsize, packcheck, DMU_READ_PREFETCH));
3968                         VERIFY(0 == dmu_read(os, bigobj, bigoff,
3969                             bigsize, bigcheck, DMU_READ_PREFETCH));
3970 
3971                         ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3972                         ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3973 
3974                         umem_free(packcheck, packsize);
3975                         umem_free(bigcheck, bigsize);
3976                 }
3977                 if (i == 2) {
3978                         txg_wait_open(dmu_objset_pool(os), 0);
3979                 } else if (i == 3) {
3980                         txg_wait_synced(dmu_objset_pool(os), 0);
3981                 }
3982         }
3983 
3984         dmu_buf_rele(bonus_db, FTAG);
3985         umem_free(packbuf, packsize);
3986         umem_free(bigbuf, bigsize);
3987         umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3988 }
3989 
3990 /* ARGSUSED */
3991 void
3992 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3993 {
3994         ztest_od_t od[1];
3995         uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3996             (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3997 
3998         /*
3999          * Have multiple threads write to large offsets in an object
4000          * to verify that parallel writes to an object -- even to the
4001          * same blocks within the object -- doesn't cause any trouble.
4002          */
4003         ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4004 
4005         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4006                 return;
4007 
4008         while (ztest_random(10) != 0)
4009                 ztest_io(zd, od[0].od_object, offset);
4010 }
4011 
4012 void
4013 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4014 {
4015         ztest_od_t od[1];
4016         uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4017             (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4018         uint64_t count = ztest_random(20) + 1;
4019         uint64_t blocksize = ztest_random_blocksize();
4020         void *data;
4021 
4022         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4023 
4024         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4025                 return;
4026 
4027         if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4028                 return;
4029 
4030         ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4031 
4032         data = umem_zalloc(blocksize, UMEM_NOFAIL);
4033 
4034         while (ztest_random(count) != 0) {
4035                 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4036                 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4037                     data) != 0)
4038                         break;
4039                 while (ztest_random(4) != 0)
4040                         ztest_io(zd, od[0].od_object, randoff);
4041         }
4042 
4043         umem_free(data, blocksize);
4044 }
4045 
4046 /*
4047  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4048  */
4049 #define ZTEST_ZAP_MIN_INTS      1
4050 #define ZTEST_ZAP_MAX_INTS      4
4051 #define ZTEST_ZAP_MAX_PROPS     1000
4052 
4053 void
4054 ztest_zap(ztest_ds_t *zd, uint64_t id)
4055 {
4056         objset_t *os = zd->zd_os;
4057         ztest_od_t od[1];
4058         uint64_t object;
4059         uint64_t txg, last_txg;
4060         uint64_t value[ZTEST_ZAP_MAX_INTS];
4061         uint64_t zl_ints, zl_intsize, prop;
4062         int i, ints;
4063         dmu_tx_t *tx;
4064         char propname[100], txgname[100];
4065         int error;
4066         char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4067 
4068         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4069 
4070         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4071                 return;
4072 
4073         object = od[0].od_object;
4074 
4075         /*
4076          * Generate a known hash collision, and verify that
4077          * we can lookup and remove both entries.
4078          */
4079         tx = dmu_tx_create(os);
4080         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4081         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4082         if (txg == 0)
4083                 return;
4084         for (i = 0; i < 2; i++) {
4085                 value[i] = i;
4086                 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4087                     1, &value[i], tx));
4088         }
4089         for (i = 0; i < 2; i++) {
4090                 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4091                     sizeof (uint64_t), 1, &value[i], tx));
4092                 VERIFY3U(0, ==,
4093                     zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4094                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4095                 ASSERT3U(zl_ints, ==, 1);
4096         }
4097         for (i = 0; i < 2; i++) {
4098                 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4099         }
4100         dmu_tx_commit(tx);
4101 
4102         /*
4103          * Generate a buch of random entries.
4104          */
4105         ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4106 
4107         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4108         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4109         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4110         bzero(value, sizeof (value));
4111         last_txg = 0;
4112 
4113         /*
4114          * If these zap entries already exist, validate their contents.
4115          */
4116         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4117         if (error == 0) {
4118                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4119                 ASSERT3U(zl_ints, ==, 1);
4120 
4121                 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4122                     zl_ints, &last_txg) == 0);
4123 
4124                 VERIFY(zap_length(os, object, propname, &zl_intsize,
4125                     &zl_ints) == 0);
4126 
4127                 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4128                 ASSERT3U(zl_ints, ==, ints);
4129 
4130                 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4131                     zl_ints, value) == 0);
4132 
4133                 for (i = 0; i < ints; i++) {
4134                         ASSERT3U(value[i], ==, last_txg + object + i);
4135                 }
4136         } else {
4137                 ASSERT3U(error, ==, ENOENT);
4138         }
4139 
4140         /*
4141          * Atomically update two entries in our zap object.
4142          * The first is named txg_%llu, and contains the txg
4143          * in which the property was last updated.  The second
4144          * is named prop_%llu, and the nth element of its value
4145          * should be txg + object + n.
4146          */
4147         tx = dmu_tx_create(os);
4148         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4149         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4150         if (txg == 0)
4151                 return;
4152 
4153         if (last_txg > txg)
4154                 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4155 
4156         for (i = 0; i < ints; i++)
4157                 value[i] = txg + object + i;
4158 
4159         VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4160             1, &txg, tx));
4161         VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4162             ints, value, tx));
4163 
4164         dmu_tx_commit(tx);
4165 
4166         /*
4167          * Remove a random pair of entries.
4168          */
4169         prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4170         (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4171         (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4172 
4173         error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4174 
4175         if (error == ENOENT)
4176                 return;
4177 
4178         ASSERT0(error);
4179 
4180         tx = dmu_tx_create(os);
4181         dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4182         txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4183         if (txg == 0)
4184                 return;
4185         VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4186         VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4187         dmu_tx_commit(tx);
4188 }
4189 
4190 /*
4191  * Testcase to test the upgrading of a microzap to fatzap.
4192  */
4193 void
4194 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4195 {
4196         objset_t *os = zd->zd_os;
4197         ztest_od_t od[1];
4198         uint64_t object, txg;
4199 
4200         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4201 
4202         if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4203                 return;
4204 
4205         object = od[0].od_object;
4206 
4207         /*
4208          * Add entries to this ZAP and make sure it spills over
4209          * and gets upgraded to a fatzap. Also, since we are adding
4210          * 2050 entries we should see ptrtbl growth and leaf-block split.
4211          */
4212         for (int i = 0; i < 2050; i++) {
4213                 char name[MAXNAMELEN];
4214                 uint64_t value = i;
4215                 dmu_tx_t *tx;
4216                 int error;
4217 
4218                 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4219                     id, value);
4220 
4221                 tx = dmu_tx_create(os);
4222                 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4223                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4224                 if (txg == 0)
4225                         return;
4226                 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4227                     &value, tx);
4228                 ASSERT(error == 0 || error == EEXIST);
4229                 dmu_tx_commit(tx);
4230         }
4231 }
4232 
4233 /* ARGSUSED */
4234 void
4235 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4236 {
4237         objset_t *os = zd->zd_os;
4238         ztest_od_t od[1];
4239         uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4240         dmu_tx_t *tx;
4241         int i, namelen, error;
4242         int micro = ztest_random(2);
4243         char name[20], string_value[20];
4244         void *data;
4245 
4246         ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4247 
4248         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4249                 return;
4250 
4251         object = od[0].od_object;
4252 
4253         /*
4254          * Generate a random name of the form 'xxx.....' where each
4255          * x is a random printable character and the dots are dots.
4256          * There are 94 such characters, and the name length goes from
4257          * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4258          */
4259         namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4260 
4261         for (i = 0; i < 3; i++)
4262                 name[i] = '!' + ztest_random('~' - '!' + 1);
4263         for (; i < namelen - 1; i++)
4264                 name[i] = '.';
4265         name[i] = '\0';
4266 
4267         if ((namelen & 1) || micro) {
4268                 wsize = sizeof (txg);
4269                 wc = 1;
4270                 data = &txg;
4271         } else {
4272                 wsize = 1;
4273                 wc = namelen;
4274                 data = string_value;
4275         }
4276 
4277         count = -1ULL;
4278         VERIFY0(zap_count(os, object, &count));
4279         ASSERT(count != -1ULL);
4280 
4281         /*
4282          * Select an operation: length, lookup, add, update, remove.
4283          */
4284         i = ztest_random(5);
4285 
4286         if (i >= 2) {
4287                 tx = dmu_tx_create(os);
4288                 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4289                 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4290                 if (txg == 0)
4291                         return;
4292                 bcopy(name, string_value, namelen);
4293         } else {
4294                 tx = NULL;
4295                 txg = 0;
4296                 bzero(string_value, namelen);
4297         }
4298 
4299         switch (i) {
4300 
4301         case 0:
4302                 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4303                 if (error == 0) {
4304                         ASSERT3U(wsize, ==, zl_wsize);
4305                         ASSERT3U(wc, ==, zl_wc);
4306                 } else {
4307                         ASSERT3U(error, ==, ENOENT);
4308                 }
4309                 break;
4310 
4311         case 1:
4312                 error = zap_lookup(os, object, name, wsize, wc, data);
4313                 if (error == 0) {
4314                         if (data == string_value &&
4315                             bcmp(name, data, namelen) != 0)
4316                                 fatal(0, "name '%s' != val '%s' len %d",
4317                                     name, data, namelen);
4318                 } else {
4319                         ASSERT3U(error, ==, ENOENT);
4320                 }
4321                 break;
4322 
4323         case 2:
4324                 error = zap_add(os, object, name, wsize, wc, data, tx);
4325                 ASSERT(error == 0 || error == EEXIST);
4326                 break;
4327 
4328         case 3:
4329                 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4330                 break;
4331 
4332         case 4:
4333                 error = zap_remove(os, object, name, tx);
4334                 ASSERT(error == 0 || error == ENOENT);
4335                 break;
4336         }
4337 
4338         if (tx != NULL)
4339                 dmu_tx_commit(tx);
4340 }
4341 
4342 /*
4343  * Commit callback data.
4344  */
4345 typedef struct ztest_cb_data {
4346         list_node_t             zcd_node;
4347         uint64_t                zcd_txg;
4348         int                     zcd_expected_err;
4349         boolean_t               zcd_added;
4350         boolean_t               zcd_called;
4351         spa_t                   *zcd_spa;
4352 } ztest_cb_data_t;
4353 
4354 /* This is the actual commit callback function */
4355 static void
4356 ztest_commit_callback(void *arg, int error)
4357 {
4358         ztest_cb_data_t *data = arg;
4359         uint64_t synced_txg;
4360 
4361         VERIFY(data != NULL);
4362         VERIFY3S(data->zcd_expected_err, ==, error);
4363         VERIFY(!data->zcd_called);
4364 
4365         synced_txg = spa_last_synced_txg(data->zcd_spa);
4366         if (data->zcd_txg > synced_txg)
4367                 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4368                     ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4369                     synced_txg);
4370 
4371         data->zcd_called = B_TRUE;
4372 
4373         if (error == ECANCELED) {
4374                 ASSERT0(data->zcd_txg);
4375                 ASSERT(!data->zcd_added);
4376 
4377                 /*
4378                  * The private callback data should be destroyed here, but
4379                  * since we are going to check the zcd_called field after
4380                  * dmu_tx_abort(), we will destroy it there.
4381                  */
4382                 return;
4383         }
4384 
4385         /* Was this callback added to the global callback list? */
4386         if (!data->zcd_added)
4387                 goto out;
4388 
4389         ASSERT3U(data->zcd_txg, !=, 0);
4390 
4391         /* Remove our callback from the list */
4392         (void) mutex_lock(&zcl.zcl_callbacks_lock);
4393         list_remove(&zcl.zcl_callbacks, data);
4394         (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4395 
4396 out:
4397         umem_free(data, sizeof (ztest_cb_data_t));
4398 }
4399 
4400 /* Allocate and initialize callback data structure */
4401 static ztest_cb_data_t *
4402 ztest_create_cb_data(objset_t *os, uint64_t txg)
4403 {
4404         ztest_cb_data_t *cb_data;
4405 
4406         cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4407 
4408         cb_data->zcd_txg = txg;
4409         cb_data->zcd_spa = dmu_objset_spa(os);
4410 
4411         return (cb_data);
4412 }
4413 
4414 /*
4415  * If a number of txgs equal to this threshold have been created after a commit
4416  * callback has been registered but not called, then we assume there is an
4417  * implementation bug.
4418  */
4419 #define ZTEST_COMMIT_CALLBACK_THRESH    (TXG_CONCURRENT_STATES + 2)
4420 
4421 /*
4422  * Commit callback test.
4423  */
4424 void
4425 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4426 {
4427         objset_t *os = zd->zd_os;
4428         ztest_od_t od[1];
4429         dmu_tx_t *tx;
4430         ztest_cb_data_t *cb_data[3], *tmp_cb;
4431         uint64_t old_txg, txg;
4432         int i, error;
4433 
4434         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4435 
4436         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4437                 return;
4438 
4439         tx = dmu_tx_create(os);
4440 
4441         cb_data[0] = ztest_create_cb_data(os, 0);
4442         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4443 
4444         dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4445 
4446         /* Every once in a while, abort the transaction on purpose */
4447         if (ztest_random(100) == 0)
4448                 error = -1;
4449 
4450         if (!error)
4451                 error = dmu_tx_assign(tx, TXG_NOWAIT);
4452 
4453         txg = error ? 0 : dmu_tx_get_txg(tx);
4454 
4455         cb_data[0]->zcd_txg = txg;
4456         cb_data[1] = ztest_create_cb_data(os, txg);
4457         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4458 
4459         if (error) {
4460                 /*
4461                  * It's not a strict requirement to call the registered
4462                  * callbacks from inside dmu_tx_abort(), but that's what
4463                  * it's supposed to happen in the current implementation
4464                  * so we will check for that.
4465                  */
4466                 for (i = 0; i < 2; i++) {
4467                         cb_data[i]->zcd_expected_err = ECANCELED;
4468                         VERIFY(!cb_data[i]->zcd_called);
4469                 }
4470 
4471                 dmu_tx_abort(tx);
4472 
4473                 for (i = 0; i < 2; i++) {
4474                         VERIFY(cb_data[i]->zcd_called);
4475                         umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4476                 }
4477 
4478                 return;
4479         }
4480 
4481         cb_data[2] = ztest_create_cb_data(os, txg);
4482         dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4483 
4484         /*
4485          * Read existing data to make sure there isn't a future leak.
4486          */
4487         VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4488             &old_txg, DMU_READ_PREFETCH));
4489 
4490         if (old_txg > txg)
4491                 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4492                     old_txg, txg);
4493 
4494         dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4495 
4496         (void) mutex_lock(&zcl.zcl_callbacks_lock);
4497 
4498         /*
4499          * Since commit callbacks don't have any ordering requirement and since
4500          * it is theoretically possible for a commit callback to be called
4501          * after an arbitrary amount of time has elapsed since its txg has been
4502          * synced, it is difficult to reliably determine whether a commit
4503          * callback hasn't been called due to high load or due to a flawed
4504          * implementation.
4505          *
4506          * In practice, we will assume that if after a certain number of txgs a
4507          * commit callback hasn't been called, then most likely there's an
4508          * implementation bug..
4509          */
4510         tmp_cb = list_head(&zcl.zcl_callbacks);
4511         if (tmp_cb != NULL &&
4512             (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4513                 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4514                     PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4515         }
4516 
4517         /*
4518          * Let's find the place to insert our callbacks.
4519          *
4520          * Even though the list is ordered by txg, it is possible for the
4521          * insertion point to not be the end because our txg may already be
4522          * quiescing at this point and other callbacks in the open txg
4523          * (from other objsets) may have sneaked in.
4524          */
4525         tmp_cb = list_tail(&zcl.zcl_callbacks);
4526         while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4527                 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4528 
4529         /* Add the 3 callbacks to the list */
4530         for (i = 0; i < 3; i++) {
4531                 if (tmp_cb == NULL)
4532                         list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4533                 else
4534                         list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4535                             cb_data[i]);
4536 
4537                 cb_data[i]->zcd_added = B_TRUE;
4538                 VERIFY(!cb_data[i]->zcd_called);
4539 
4540                 tmp_cb = cb_data[i];
4541         }
4542 
4543         (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4544 
4545         dmu_tx_commit(tx);
4546 }
4547 
4548 /* ARGSUSED */
4549 void
4550 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4551 {
4552         zfs_prop_t proplist[] = {
4553                 ZFS_PROP_CHECKSUM,
4554                 ZFS_PROP_COMPRESSION,
4555                 ZFS_PROP_COPIES,
4556                 ZFS_PROP_DEDUP
4557         };
4558 
4559         (void) rw_rdlock(&ztest_name_lock);
4560 
4561         for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4562                 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4563                     ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4564 
4565         (void) rw_unlock(&ztest_name_lock);
4566 }
4567 
4568 /* ARGSUSED */
4569 void
4570 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4571 {
4572         nvlist_t *props = NULL;
4573 
4574         (void) rw_rdlock(&ztest_name_lock);
4575 
4576         (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4577             ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4578 
4579         VERIFY0(spa_prop_get(ztest_spa, &props));
4580 
4581         if (ztest_opts.zo_verbose >= 6)
4582                 dump_nvlist(props, 4);
4583 
4584         nvlist_free(props);
4585 
4586         (void) rw_unlock(&ztest_name_lock);
4587 }
4588 
4589 static int
4590 user_release_one(const char *snapname, const char *holdname)
4591 {
4592         nvlist_t *snaps, *holds;
4593         int error;
4594 
4595         snaps = fnvlist_alloc();
4596         holds = fnvlist_alloc();
4597         fnvlist_add_boolean(holds, holdname);
4598         fnvlist_add_nvlist(snaps, snapname, holds);
4599         fnvlist_free(holds);
4600         error = dsl_dataset_user_release(snaps, NULL);
4601         fnvlist_free(snaps);
4602         return (error);
4603 }
4604 
4605 /*
4606  * Test snapshot hold/release and deferred destroy.
4607  */
4608 void
4609 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4610 {
4611         int error;
4612         objset_t *os = zd->zd_os;
4613         objset_t *origin;
4614         char snapname[100];
4615         char fullname[100];
4616         char clonename[100];
4617         char tag[100];
4618         char osname[MAXNAMELEN];
4619         nvlist_t *holds;
4620 
4621         (void) rw_rdlock(&ztest_name_lock);
4622 
4623         dmu_objset_name(os, osname);
4624 
4625         (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4626         (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4627         (void) snprintf(clonename, sizeof (clonename),
4628             "%s/ch1_%llu", osname, id);
4629         (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4630 
4631         /*
4632          * Clean up from any previous run.
4633          */
4634         error = dsl_destroy_head(clonename);
4635         if (error != ENOENT)
4636                 ASSERT0(error);
4637         error = user_release_one(fullname, tag);
4638         if (error != ESRCH && error != ENOENT)
4639                 ASSERT0(error);
4640         error = dsl_destroy_snapshot(fullname, B_FALSE);
4641         if (error != ENOENT)
4642                 ASSERT0(error);
4643 
4644         /*
4645          * Create snapshot, clone it, mark snap for deferred destroy,
4646          * destroy clone, verify snap was also destroyed.
4647          */
4648         error = dmu_objset_snapshot_one(osname, snapname);
4649         if (error) {
4650                 if (error == ENOSPC) {
4651                         ztest_record_enospc("dmu_objset_snapshot");
4652                         goto out;
4653                 }
4654                 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4655         }
4656 
4657         error = dmu_objset_clone(clonename, fullname);
4658         if (error) {
4659                 if (error == ENOSPC) {
4660                         ztest_record_enospc("dmu_objset_clone");
4661                         goto out;
4662                 }
4663                 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4664         }
4665 
4666         error = dsl_destroy_snapshot(fullname, B_TRUE);
4667         if (error) {
4668                 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4669                     fullname, error);
4670         }
4671 
4672         error = dsl_destroy_head(clonename);
4673         if (error)
4674                 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4675 
4676         error = dmu_objset_hold(fullname, FTAG, &origin);
4677         if (error != ENOENT)
4678                 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4679 
4680         /*
4681          * Create snapshot, add temporary hold, verify that we can't
4682          * destroy a held snapshot, mark for deferred destroy,
4683          * release hold, verify snapshot was destroyed.
4684          */
4685         error = dmu_objset_snapshot_one(osname, snapname);
4686         if (error) {
4687                 if (error == ENOSPC) {
4688                         ztest_record_enospc("dmu_objset_snapshot");
4689                         goto out;
4690                 }
4691                 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4692         }
4693 
4694         holds = fnvlist_alloc();
4695         fnvlist_add_string(holds, fullname, tag);
4696         error = dsl_dataset_user_hold(holds, 0, NULL);
4697         fnvlist_free(holds);
4698 
4699         if (error)
4700                 fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4701 
4702         error = dsl_destroy_snapshot(fullname, B_FALSE);
4703         if (error != EBUSY) {
4704                 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4705                     fullname, error);
4706         }
4707 
4708         error = dsl_destroy_snapshot(fullname, B_TRUE);
4709         if (error) {
4710                 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4711                     fullname, error);
4712         }
4713 
4714         error = user_release_one(fullname, tag);
4715         if (error)
4716                 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4717 
4718         VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4719 
4720 out:
4721         (void) rw_unlock(&ztest_name_lock);
4722 }
4723 
4724 /*
4725  * Inject random faults into the on-disk data.
4726  */
4727 /* ARGSUSED */
4728 void
4729 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4730 {
4731         ztest_shared_t *zs = ztest_shared;
4732         spa_t *spa = ztest_spa;
4733         int fd;
4734         uint64_t offset;
4735         uint64_t leaves;
4736         uint64_t bad = 0x1990c0ffeedecade;
4737         uint64_t top, leaf;
4738         char path0[MAXPATHLEN];
4739         char pathrand[MAXPATHLEN];
4740         size_t fsize;
4741         int bshift = SPA_MAXBLOCKSHIFT + 2;     /* don't scrog all labels */
4742         int iters = 1000;
4743         int maxfaults;
4744         int mirror_save;
4745         vdev_t *vd0 = NULL;
4746         uint64_t guid0 = 0;
4747         boolean_t islog = B_FALSE;
4748 
4749         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4750         maxfaults = MAXFAULTS();
4751         leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4752         mirror_save = zs->zs_mirrors;
4753         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4754 
4755         ASSERT(leaves >= 1);
4756 
4757         /*
4758          * Grab the name lock as reader. There are some operations
4759          * which don't like to have their vdevs changed while
4760          * they are in progress (i.e. spa_change_guid). Those
4761          * operations will have grabbed the name lock as writer.
4762          */
4763         (void) rw_rdlock(&ztest_name_lock);
4764 
4765         /*
4766          * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4767          */
4768         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4769 
4770         if (ztest_random(2) == 0) {
4771                 /*
4772                  * Inject errors on a normal data device or slog device.
4773                  */
4774                 top = ztest_random_vdev_top(spa, B_TRUE);
4775                 leaf = ztest_random(leaves) + zs->zs_splits;
4776 
4777                 /*
4778                  * Generate paths to the first leaf in this top-level vdev,
4779                  * and to the random leaf we selected.  We'll induce transient
4780                  * write failures and random online/offline activity on leaf 0,
4781                  * and we'll write random garbage to the randomly chosen leaf.
4782                  */
4783                 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4784                     ztest_opts.zo_dir, ztest_opts.zo_pool,
4785                     top * leaves + zs->zs_splits);
4786                 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4787                     ztest_opts.zo_dir, ztest_opts.zo_pool,
4788                     top * leaves + leaf);
4789 
4790                 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4791                 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4792                         islog = B_TRUE;
4793 
4794                 /*
4795                  * If the top-level vdev needs to be resilvered
4796                  * then we only allow faults on the device that is
4797                  * resilvering.
4798                  */
4799                 if (vd0 != NULL && maxfaults != 1 &&
4800                     (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4801                     vd0->vdev_resilvering)) {
4802                         /*
4803                          * Make vd0 explicitly claim to be unreadable,
4804                          * or unwriteable, or reach behind its back
4805                          * and close the underlying fd.  We can do this if
4806                          * maxfaults == 0 because we'll fail and reexecute,
4807                          * and we can do it if maxfaults >= 2 because we'll
4808                          * have enough redundancy.  If maxfaults == 1, the
4809                          * combination of this with injection of random data
4810                          * corruption below exceeds the pool's fault tolerance.
4811                          */
4812                         vdev_file_t *vf = vd0->vdev_tsd;
4813 
4814                         if (vf != NULL && ztest_random(3) == 0) {
4815                                 (void) close(vf->vf_vnode->v_fd);
4816                                 vf->vf_vnode->v_fd = -1;
4817                         } else if (ztest_random(2) == 0) {
4818                                 vd0->vdev_cant_read = B_TRUE;
4819                         } else {
4820                                 vd0->vdev_cant_write = B_TRUE;
4821                         }
4822                         guid0 = vd0->vdev_guid;
4823                 }
4824         } else {
4825                 /*
4826                  * Inject errors on an l2cache device.
4827                  */
4828                 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4829 
4830                 if (sav->sav_count == 0) {
4831                         spa_config_exit(spa, SCL_STATE, FTAG);
4832                         (void) rw_unlock(&ztest_name_lock);
4833                         return;
4834                 }
4835                 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4836                 guid0 = vd0->vdev_guid;
4837                 (void) strcpy(path0, vd0->vdev_path);
4838                 (void) strcpy(pathrand, vd0->vdev_path);
4839 
4840                 leaf = 0;
4841                 leaves = 1;
4842                 maxfaults = INT_MAX;    /* no limit on cache devices */
4843         }
4844 
4845         spa_config_exit(spa, SCL_STATE, FTAG);
4846         (void) rw_unlock(&ztest_name_lock);
4847 
4848         /*
4849          * If we can tolerate two or more faults, or we're dealing
4850          * with a slog, randomly online/offline vd0.
4851          */
4852         if ((maxfaults >= 2 || islog) && guid0 != 0) {
4853                 if (ztest_random(10) < 6) {
4854                         int flags = (ztest_random(2) == 0 ?
4855                             ZFS_OFFLINE_TEMPORARY : 0);
4856 
4857                         /*
4858                          * We have to grab the zs_name_lock as writer to
4859                          * prevent a race between offlining a slog and
4860                          * destroying a dataset. Offlining the slog will
4861                          * grab a reference on the dataset which may cause
4862                          * dmu_objset_destroy() to fail with EBUSY thus
4863                          * leaving the dataset in an inconsistent state.
4864                          */
4865                         if (islog)
4866                                 (void) rw_wrlock(&ztest_name_lock);
4867 
4868                         VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4869 
4870                         if (islog)
4871                                 (void) rw_unlock(&ztest_name_lock);
4872                 } else {
4873                         /*
4874                          * Ideally we would like to be able to randomly
4875                          * call vdev_[on|off]line without holding locks
4876                          * to force unpredictable failures but the side
4877                          * effects of vdev_[on|off]line prevent us from
4878                          * doing so. We grab the ztest_vdev_lock here to
4879                          * prevent a race between injection testing and
4880                          * aux_vdev removal.
4881                          */
4882                         VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4883                         (void) vdev_online(spa, guid0, 0, NULL);
4884                         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4885                 }
4886         }
4887 
4888         if (maxfaults == 0)
4889                 return;
4890 
4891         /*
4892          * We have at least single-fault tolerance, so inject data corruption.
4893          */
4894         fd = open(pathrand, O_RDWR);
4895 
4896         if (fd == -1)   /* we hit a gap in the device namespace */
4897                 return;
4898 
4899         fsize = lseek(fd, 0, SEEK_END);
4900 
4901         while (--iters != 0) {
4902                 offset = ztest_random(fsize / (leaves << bshift)) *
4903                     (leaves << bshift) + (leaf << bshift) +
4904                     (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4905 
4906                 if (offset >= fsize)
4907                         continue;
4908 
4909                 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4910                 if (mirror_save != zs->zs_mirrors) {
4911                         VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4912                         (void) close(fd);
4913                         return;
4914                 }
4915 
4916                 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4917                         fatal(1, "can't inject bad word at 0x%llx in %s",
4918                             offset, pathrand);
4919 
4920                 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4921 
4922                 if (ztest_opts.zo_verbose >= 7)
4923                         (void) printf("injected bad word into %s,"
4924                             " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4925         }
4926 
4927         (void) close(fd);
4928 }
4929 
4930 /*
4931  * Verify that DDT repair works as expected.
4932  */
4933 void
4934 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4935 {
4936         ztest_shared_t *zs = ztest_shared;
4937         spa_t *spa = ztest_spa;
4938         objset_t *os = zd->zd_os;
4939         ztest_od_t od[1];
4940         uint64_t object, blocksize, txg, pattern, psize;
4941         enum zio_checksum checksum = spa_dedup_checksum(spa);
4942         dmu_buf_t *db;
4943         dmu_tx_t *tx;
4944         void *buf;
4945         blkptr_t blk;
4946         int copies = 2 * ZIO_DEDUPDITTO_MIN;
4947 
4948         blocksize = ztest_random_blocksize();
4949         blocksize = MIN(blocksize, 2048);       /* because we write so many */
4950 
4951         ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4952 
4953         if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4954                 return;
4955 
4956         /*
4957          * Take the name lock as writer to prevent anyone else from changing
4958          * the pool and dataset properies we need to maintain during this test.
4959          */
4960         (void) rw_wrlock(&ztest_name_lock);
4961 
4962         if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4963             B_FALSE) != 0 ||
4964             ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4965             B_FALSE) != 0) {
4966                 (void) rw_unlock(&ztest_name_lock);
4967                 return;
4968         }
4969 
4970         object = od[0].od_object;
4971         blocksize = od[0].od_blocksize;
4972         pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4973 
4974         ASSERT(object != 0);
4975 
4976         tx = dmu_tx_create(os);
4977         dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4978         txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4979         if (txg == 0) {
4980                 (void) rw_unlock(&ztest_name_lock);
4981                 return;
4982         }
4983 
4984         /*
4985          * Write all the copies of our block.
4986          */
4987         for (int i = 0; i < copies; i++) {
4988                 uint64_t offset = i * blocksize;
4989                 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
4990                     DMU_READ_NO_PREFETCH);
4991                 if (error != 0) {
4992                         fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
4993                             os, (long long)object, (long long) offset, error);
4994                 }
4995                 ASSERT(db->db_offset == offset);
4996                 ASSERT(db->db_size == blocksize);
4997                 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4998                     ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4999                 dmu_buf_will_fill(db, tx);
5000                 ztest_pattern_set(db->db_data, db->db_size, pattern);
5001                 dmu_buf_rele(db, FTAG);
5002         }
5003 
5004         dmu_tx_commit(tx);
5005         txg_wait_synced(spa_get_dsl(spa), txg);
5006 
5007         /*
5008          * Find out what block we got.
5009          */
5010         VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5011             DMU_READ_NO_PREFETCH));
5012         blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5013         dmu_buf_rele(db, FTAG);
5014 
5015         /*
5016          * Damage the block.  Dedup-ditto will save us when we read it later.
5017          */
5018         psize = BP_GET_PSIZE(&blk);
5019         buf = zio_buf_alloc(psize);
5020         ztest_pattern_set(buf, psize, ~pattern);
5021 
5022         (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5023             buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5024             ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5025 
5026         zio_buf_free(buf, psize);
5027 
5028         (void) rw_unlock(&ztest_name_lock);
5029 }
5030 
5031 /*
5032  * Scrub the pool.
5033  */
5034 /* ARGSUSED */
5035 void
5036 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5037 {
5038         spa_t *spa = ztest_spa;
5039 
5040         (void) spa_scan(spa, POOL_SCAN_SCRUB);
5041         (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5042         (void) spa_scan(spa, POOL_SCAN_SCRUB);
5043 }
5044 
5045 /*
5046  * Change the guid for the pool.
5047  */
5048 /* ARGSUSED */
5049 void
5050 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5051 {
5052         spa_t *spa = ztest_spa;
5053         uint64_t orig, load;
5054         int error;
5055 
5056         orig = spa_guid(spa);
5057         load = spa_load_guid(spa);
5058 
5059         (void) rw_wrlock(&ztest_name_lock);
5060         error = spa_change_guid(spa);
5061         (void) rw_unlock(&ztest_name_lock);
5062 
5063         if (error != 0)
5064                 return;
5065 
5066         if (ztest_opts.zo_verbose >= 4) {
5067                 (void) printf("Changed guid old %llu -> %llu\n",
5068                     (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5069         }
5070 
5071         VERIFY3U(orig, !=, spa_guid(spa));
5072         VERIFY3U(load, ==, spa_load_guid(spa));
5073 }
5074 
5075 /*
5076  * Rename the pool to a different name and then rename it back.
5077  */
5078 /* ARGSUSED */
5079 void
5080 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5081 {
5082         char *oldname, *newname;
5083         spa_t *spa;
5084 
5085         (void) rw_wrlock(&ztest_name_lock);
5086 
5087         oldname = ztest_opts.zo_pool;
5088         newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5089         (void) strcpy(newname, oldname);
5090         (void) strcat(newname, "_tmp");
5091 
5092         /*
5093          * Do the rename
5094          */
5095         VERIFY3U(0, ==, spa_rename(oldname, newname));
5096 
5097         /*
5098          * Try to open it under the old name, which shouldn't exist
5099          */
5100         VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5101 
5102         /*
5103          * Open it under the new name and make sure it's still the same spa_t.
5104          */
5105         VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5106 
5107         ASSERT(spa == ztest_spa);
5108         spa_close(spa, FTAG);
5109 
5110         /*
5111          * Rename it back to the original
5112          */
5113         VERIFY3U(0, ==, spa_rename(newname, oldname));
5114 
5115         /*
5116          * Make sure it can still be opened
5117          */
5118         VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5119 
5120         ASSERT(spa == ztest_spa);
5121         spa_close(spa, FTAG);
5122 
5123         umem_free(newname, strlen(newname) + 1);
5124 
5125         (void) rw_unlock(&ztest_name_lock);
5126 }
5127 
5128 /*
5129  * Verify pool integrity by running zdb.
5130  */
5131 static void
5132 ztest_run_zdb(char *pool)
5133 {
5134         int status;
5135         char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5136         char zbuf[1024];
5137         char *bin;
5138         char *ztest;
5139         char *isa;
5140         int isalen;
5141         FILE *fp;
5142 
5143         (void) realpath(getexecname(), zdb);
5144 
5145         /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5146         bin = strstr(zdb, "/usr/bin/");
5147         ztest = strstr(bin, "/ztest");
5148         isa = bin + 8;
5149         isalen = ztest - isa;
5150         isa = strdup(isa);
5151         /* LINTED */
5152         (void) sprintf(bin,
5153             "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
5154             isalen,
5155             isa,
5156             ztest_opts.zo_verbose >= 3 ? "s" : "",
5157             ztest_opts.zo_verbose >= 4 ? "v" : "",
5158             spa_config_path,
5159             pool);
5160         free(isa);
5161 
5162         if (ztest_opts.zo_verbose >= 5)
5163                 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5164 
5165         fp = popen(zdb, "r");
5166 
5167         while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5168                 if (ztest_opts.zo_verbose >= 3)
5169                         (void) printf("%s", zbuf);
5170 
5171         status = pclose(fp);
5172 
5173         if (status == 0)
5174                 return;
5175 
5176         ztest_dump_core = 0;
5177         if (WIFEXITED(status))
5178                 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5179         else
5180                 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5181 }
5182 
5183 static void
5184 ztest_walk_pool_directory(char *header)
5185 {
5186         spa_t *spa = NULL;
5187 
5188         if (ztest_opts.zo_verbose >= 6)
5189                 (void) printf("%s\n", header);
5190 
5191         mutex_enter(&spa_namespace_lock);
5192         while ((spa = spa_next(spa)) != NULL)
5193                 if (ztest_opts.zo_verbose >= 6)
5194                         (void) printf("\t%s\n", spa_name(spa));
5195         mutex_exit(&spa_namespace_lock);
5196 }
5197 
5198 static void
5199 ztest_spa_import_export(char *oldname, char *newname)
5200 {
5201         nvlist_t *config, *newconfig;
5202         uint64_t pool_guid;
5203         spa_t *spa;
5204         int error;
5205 
5206         if (ztest_opts.zo_verbose >= 4) {
5207                 (void) printf("import/export: old = %s, new = %s\n",
5208                     oldname, newname);
5209         }
5210 
5211         /*
5212          * Clean up from previous runs.
5213          */
5214         (void) spa_destroy(newname);
5215 
5216         /*
5217          * Get the pool's configuration and guid.
5218          */
5219         VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5220 
5221         /*
5222          * Kick off a scrub to tickle scrub/export races.
5223          */
5224         if (ztest_random(2) == 0)
5225                 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5226 
5227         pool_guid = spa_guid(spa);
5228         spa_close(spa, FTAG);
5229 
5230         ztest_walk_pool_directory("pools before export");
5231 
5232         /*
5233          * Export it.
5234          */
5235         VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5236 
5237         ztest_walk_pool_directory("pools after export");
5238 
5239         /*
5240          * Try to import it.
5241          */
5242         newconfig = spa_tryimport(config);
5243         ASSERT(newconfig != NULL);
5244         nvlist_free(newconfig);
5245 
5246         /*
5247          * Import it under the new name.
5248          */
5249         error = spa_import(newname, config, NULL, 0);
5250         if (error != 0) {
5251                 dump_nvlist(config, 0);
5252                 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5253                     oldname, newname, error);
5254         }
5255 
5256         ztest_walk_pool_directory("pools after import");
5257 
5258         /*
5259          * Try to import it again -- should fail with EEXIST.
5260          */
5261         VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5262 
5263         /*
5264          * Try to import it under a different name -- should fail with EEXIST.
5265          */
5266         VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5267 
5268         /*
5269          * Verify that the pool is no longer visible under the old name.
5270          */
5271         VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5272 
5273         /*
5274          * Verify that we can open and close the pool using the new name.
5275          */
5276         VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5277         ASSERT(pool_guid == spa_guid(spa));
5278         spa_close(spa, FTAG);
5279 
5280         nvlist_free(config);
5281 }
5282 
5283 static void
5284 ztest_resume(spa_t *spa)
5285 {
5286         if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5287                 (void) printf("resuming from suspended state\n");
5288         spa_vdev_state_enter(spa, SCL_NONE);
5289         vdev_clear(spa, NULL);
5290         (void) spa_vdev_state_exit(spa, NULL, 0);
5291         (void) zio_resume(spa);
5292 }
5293 
5294 static void *
5295 ztest_resume_thread(void *arg)
5296 {
5297         spa_t *spa = arg;
5298 
5299         while (!ztest_exiting) {
5300                 if (spa_suspended(spa))
5301                         ztest_resume(spa);
5302                 (void) poll(NULL, 0, 100);
5303         }
5304         return (NULL);
5305 }
5306 
5307 static void *
5308 ztest_deadman_thread(void *arg)
5309 {
5310         ztest_shared_t *zs = arg;
5311         spa_t *spa = ztest_spa;
5312         hrtime_t delta, total = 0;
5313 
5314         for (;;) {
5315                 delta = (zs->zs_thread_stop - zs->zs_thread_start) /
5316                     NANOSEC + zfs_deadman_synctime;
5317 
5318                 (void) poll(NULL, 0, (int)(1000 * delta));
5319 
5320                 /*
5321                  * If the pool is suspended then fail immediately. Otherwise,
5322                  * check to see if the pool is making any progress. If
5323                  * vdev_deadman() discovers that there hasn't been any recent
5324                  * I/Os then it will end up aborting the tests.
5325                  */
5326                 if (spa_suspended(spa)) {
5327                         fatal(0, "aborting test after %llu seconds because "
5328                             "pool has transitioned to a suspended state.",
5329                             zfs_deadman_synctime);
5330                         return (NULL);
5331                 }
5332                 vdev_deadman(spa->spa_root_vdev);
5333 
5334                 total += zfs_deadman_synctime;
5335                 (void) printf("ztest has been running for %lld seconds\n",
5336                     total);
5337         }
5338 }
5339 
5340 static void
5341 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5342 {
5343         ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5344         ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5345         hrtime_t functime = gethrtime();
5346 
5347         for (int i = 0; i < zi->zi_iters; i++)
5348                 zi->zi_func(zd, id);
5349 
5350         functime = gethrtime() - functime;
5351 
5352         atomic_add_64(&zc->zc_count, 1);
5353         atomic_add_64(&zc->zc_time, functime);
5354 
5355         if (ztest_opts.zo_verbose >= 4) {
5356                 Dl_info dli;
5357                 (void) dladdr((void *)zi->zi_func, &dli);
5358                 (void) printf("%6.2f sec in %s\n",
5359                     (double)functime / NANOSEC, dli.dli_sname);
5360         }
5361 }
5362 
5363 static void *
5364 ztest_thread(void *arg)
5365 {
5366         int rand;
5367         uint64_t id = (uintptr_t)arg;
5368         ztest_shared_t *zs = ztest_shared;
5369         uint64_t call_next;
5370         hrtime_t now;
5371         ztest_info_t *zi;
5372         ztest_shared_callstate_t *zc;
5373 
5374         while ((now = gethrtime()) < zs->zs_thread_stop) {
5375                 /*
5376                  * See if it's time to force a crash.
5377                  */
5378                 if (now > zs->zs_thread_kill)
5379                         ztest_kill(zs);
5380 
5381                 /*
5382                  * If we're getting ENOSPC with some regularity, stop.
5383                  */
5384                 if (zs->zs_enospc_count > 10)
5385                         break;
5386 
5387                 /*
5388                  * Pick a random function to execute.
5389                  */
5390                 rand = ztest_random(ZTEST_FUNCS);
5391                 zi = &ztest_info[rand];
5392                 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5393                 call_next = zc->zc_next;
5394 
5395                 if (now >= call_next &&
5396                     atomic_cas_64(&zc->zc_next, call_next, call_next +
5397                     ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5398                         ztest_execute(rand, zi, id);
5399                 }
5400         }
5401 
5402         return (NULL);
5403 }
5404 
5405 static void
5406 ztest_dataset_name(char *dsname, char *pool, int d)
5407 {
5408         (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5409 }
5410 
5411 static void
5412 ztest_dataset_destroy(int d)
5413 {
5414         char name[MAXNAMELEN];
5415 
5416         ztest_dataset_name(name, ztest_opts.zo_pool, d);
5417 
5418         if (ztest_opts.zo_verbose >= 3)
5419                 (void) printf("Destroying %s to free up space\n", name);
5420 
5421         /*
5422          * Cleanup any non-standard clones and snapshots.  In general,
5423          * ztest thread t operates on dataset (t % zopt_datasets),
5424          * so there may be more than one thing to clean up.
5425          */
5426         for (int t = d; t < ztest_opts.zo_threads;
5427             t += ztest_opts.zo_datasets) {
5428                 ztest_dsl_dataset_cleanup(name, t);
5429         }
5430 
5431         (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5432             DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5433 }
5434 
5435 static void
5436 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5437 {
5438         uint64_t usedobjs, dirobjs, scratch;
5439 
5440         /*
5441          * ZTEST_DIROBJ is the object directory for the entire dataset.
5442          * Therefore, the number of objects in use should equal the
5443          * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5444          * If not, we have an object leak.
5445          *
5446          * Note that we can only check this in ztest_dataset_open(),
5447          * when the open-context and syncing-context values agree.
5448          * That's because zap_count() returns the open-context value,
5449          * while dmu_objset_space() returns the rootbp fill count.
5450          */
5451         VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5452         dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5453         ASSERT3U(dirobjs + 1, ==, usedobjs);
5454 }
5455 
5456 static int
5457 ztest_dataset_open(int d)
5458 {
5459         ztest_ds_t *zd = &ztest_ds[d];
5460         uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5461         objset_t *os;
5462         zilog_t *zilog;
5463         char name[MAXNAMELEN];
5464         int error;
5465 
5466         ztest_dataset_name(name, ztest_opts.zo_pool, d);
5467 
5468         (void) rw_rdlock(&ztest_name_lock);
5469 
5470         error = ztest_dataset_create(name);
5471         if (error == ENOSPC) {
5472                 (void) rw_unlock(&ztest_name_lock);
5473                 ztest_record_enospc(FTAG);
5474                 return (error);
5475         }
5476         ASSERT(error == 0 || error == EEXIST);
5477 
5478         VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5479         (void) rw_unlock(&ztest_name_lock);
5480 
5481         ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5482 
5483         zilog = zd->zd_zilog;
5484 
5485         if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5486             zilog->zl_header->zh_claim_lr_seq < committed_seq)
5487                 fatal(0, "missing log records: claimed %llu < committed %llu",
5488                     zilog->zl_header->zh_claim_lr_seq, committed_seq);
5489 
5490         ztest_dataset_dirobj_verify(zd);
5491 
5492         zil_replay(os, zd, ztest_replay_vector);
5493 
5494         ztest_dataset_dirobj_verify(zd);
5495 
5496         if (ztest_opts.zo_verbose >= 6)
5497                 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5498                     zd->zd_name,
5499                     (u_longlong_t)zilog->zl_parse_blk_count,
5500                     (u_longlong_t)zilog->zl_parse_lr_count,
5501                     (u_longlong_t)zilog->zl_replaying_seq);
5502 
5503         zilog = zil_open(os, ztest_get_data);
5504 
5505         if (zilog->zl_replaying_seq != 0 &&
5506             zilog->zl_replaying_seq < committed_seq)
5507                 fatal(0, "missing log records: replayed %llu < committed %llu",
5508                     zilog->zl_replaying_seq, committed_seq);
5509 
5510         return (0);
5511 }
5512 
5513 static void
5514 ztest_dataset_close(int d)
5515 {
5516         ztest_ds_t *zd = &ztest_ds[d];
5517 
5518         zil_close(zd->zd_zilog);
5519         dmu_objset_disown(zd->zd_os, zd);
5520 
5521         ztest_zd_fini(zd);
5522 }
5523 
5524 /*
5525  * Kick off threads to run tests on all datasets in parallel.
5526  */
5527 static void
5528 ztest_run(ztest_shared_t *zs)
5529 {
5530         thread_t *tid;
5531         spa_t *spa;
5532         objset_t *os;
5533         thread_t resume_tid;
5534         int error;
5535 
5536         ztest_exiting = B_FALSE;
5537 
5538         /*
5539          * Initialize parent/child shared state.
5540          */
5541         VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5542         VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5543 
5544         zs->zs_thread_start = gethrtime();
5545         zs->zs_thread_stop =
5546             zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5547         zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5548         zs->zs_thread_kill = zs->zs_thread_stop;
5549         if (ztest_random(100) < ztest_opts.zo_killrate) {
5550                 zs->zs_thread_kill -=
5551                     ztest_random(ztest_opts.zo_passtime * NANOSEC);
5552         }
5553 
5554         (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5555 
5556         list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5557             offsetof(ztest_cb_data_t, zcd_node));
5558 
5559         /*
5560          * Open our pool.
5561          */
5562         kernel_init(FREAD | FWRITE);
5563         VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5564         spa->spa_debug = B_TRUE;
5565         ztest_spa = spa;
5566 
5567         VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5568             DMU_OST_ANY, B_TRUE, FTAG, &os));
5569         zs->zs_guid = dmu_objset_fsid_guid(os);
5570         dmu_objset_disown(os, FTAG);
5571 
5572         spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5573 
5574         /*
5575          * We don't expect the pool to suspend unless maxfaults == 0,
5576          * in which case ztest_fault_inject() temporarily takes away
5577          * the only valid replica.
5578          */
5579         if (MAXFAULTS() == 0)
5580                 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5581         else
5582                 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5583 
5584         /*
5585          * Create a thread to periodically resume suspended I/O.
5586          */
5587         VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5588             &resume_tid) == 0);
5589 
5590         /*
5591          * Create a deadman thread to abort() if we hang.
5592          */
5593         VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5594             NULL) == 0);
5595 
5596         /*
5597          * Verify that we can safely inquire about about any object,
5598          * whether it's allocated or not.  To make it interesting,
5599          * we probe a 5-wide window around each power of two.
5600          * This hits all edge cases, including zero and the max.
5601          */
5602         for (int t = 0; t < 64; t++) {
5603                 for (int d = -5; d <= 5; d++) {
5604                         error = dmu_object_info(spa->spa_meta_objset,
5605                             (1ULL << t) + d, NULL);
5606                         ASSERT(error == 0 || error == ENOENT ||
5607                             error == EINVAL);
5608                 }
5609         }
5610 
5611         /*
5612          * If we got any ENOSPC errors on the previous run, destroy something.
5613          */
5614         if (zs->zs_enospc_count != 0) {
5615                 int d = ztest_random(ztest_opts.zo_datasets);
5616                 ztest_dataset_destroy(d);
5617         }
5618         zs->zs_enospc_count = 0;
5619 
5620         tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5621             UMEM_NOFAIL);
5622 
5623         if (ztest_opts.zo_verbose >= 4)
5624                 (void) printf("starting main threads...\n");
5625 
5626         /*
5627          * Kick off all the tests that run in parallel.
5628          */
5629         for (int t = 0; t < ztest_opts.zo_threads; t++) {
5630                 if (t < ztest_opts.zo_datasets &&
5631                     ztest_dataset_open(t) != 0)
5632                         return;
5633                 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5634                     THR_BOUND, &tid[t]) == 0);
5635         }
5636 
5637         /*
5638          * Wait for all of the tests to complete.  We go in reverse order
5639          * so we don't close datasets while threads are still using them.
5640          */
5641         for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5642                 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5643                 if (t < ztest_opts.zo_datasets)
5644                         ztest_dataset_close(t);
5645         }
5646 
5647         txg_wait_synced(spa_get_dsl(spa), 0);
5648 
5649         zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5650         zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5651 
5652         umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5653 
5654         /* Kill the resume thread */
5655         ztest_exiting = B_TRUE;
5656         VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5657         ztest_resume(spa);
5658 
5659         /*
5660          * Right before closing the pool, kick off a bunch of async I/O;
5661          * spa_close() should wait for it to complete.
5662          */
5663         for (uint64_t object = 1; object < 50; object++)
5664                 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5665 
5666         spa_close(spa, FTAG);
5667 
5668         /*
5669          * Verify that we can loop over all pools.
5670          */
5671         mutex_enter(&spa_namespace_lock);
5672         for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5673                 if (ztest_opts.zo_verbose > 3)
5674                         (void) printf("spa_next: found %s\n", spa_name(spa));
5675         mutex_exit(&spa_namespace_lock);
5676 
5677         /*
5678          * Verify that we can export the pool and reimport it under a
5679          * different name.
5680          */
5681         if (ztest_random(2) == 0) {
5682                 char name[MAXNAMELEN];
5683                 (void) snprintf(name, MAXNAMELEN, "%s_import",
5684                     ztest_opts.zo_pool);
5685                 ztest_spa_import_export(ztest_opts.zo_pool, name);
5686                 ztest_spa_import_export(name, ztest_opts.zo_pool);
5687         }
5688 
5689         kernel_fini();
5690 
5691         list_destroy(&zcl.zcl_callbacks);
5692 
5693         (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5694 
5695         (void) rwlock_destroy(&ztest_name_lock);
5696         (void) _mutex_destroy(&ztest_vdev_lock);
5697 }
5698 
5699 static void
5700 ztest_freeze(void)
5701 {
5702         ztest_ds_t *zd = &ztest_ds[0];
5703         spa_t *spa;
5704         int numloops = 0;
5705 
5706         if (ztest_opts.zo_verbose >= 3)
5707                 (void) printf("testing spa_freeze()...\n");
5708 
5709         kernel_init(FREAD | FWRITE);
5710         VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5711         VERIFY3U(0, ==, ztest_dataset_open(0));
5712         spa->spa_debug = B_TRUE;
5713         ztest_spa = spa;
5714 
5715         /*
5716          * Force the first log block to be transactionally allocated.
5717          * We have to do this before we freeze the pool -- otherwise
5718          * the log chain won't be anchored.
5719          */
5720         while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5721                 ztest_dmu_object_alloc_free(zd, 0);
5722                 zil_commit(zd->zd_zilog, 0);
5723         }
5724 
5725         txg_wait_synced(spa_get_dsl(spa), 0);
5726 
5727         /*
5728          * Freeze the pool.  This stops spa_sync() from doing anything,
5729          * so that the only way to record changes from now on is the ZIL.
5730          */
5731         spa_freeze(spa);
5732 
5733         /*
5734          * Run tests that generate log records but don't alter the pool config
5735          * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5736          * We do a txg_wait_synced() after each iteration to force the txg
5737          * to increase well beyond the last synced value in the uberblock.
5738          * The ZIL should be OK with that.
5739          */
5740         while (ztest_random(10) != 0 &&
5741             numloops++ < ztest_opts.zo_maxloops) {
5742                 ztest_dmu_write_parallel(zd, 0);
5743                 ztest_dmu_object_alloc_free(zd, 0);
5744                 txg_wait_synced(spa_get_dsl(spa), 0);
5745         }
5746 
5747         /*
5748          * Commit all of the changes we just generated.
5749          */
5750         zil_commit(zd->zd_zilog, 0);
5751         txg_wait_synced(spa_get_dsl(spa), 0);
5752 
5753         /*
5754          * Close our dataset and close the pool.
5755          */
5756         ztest_dataset_close(0);
5757         spa_close(spa, FTAG);
5758         kernel_fini();
5759 
5760         /*
5761          * Open and close the pool and dataset to induce log replay.
5762          */
5763         kernel_init(FREAD | FWRITE);
5764         VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5765         ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5766         VERIFY3U(0, ==, ztest_dataset_open(0));
5767         ztest_dataset_close(0);
5768 
5769         spa->spa_debug = B_TRUE;
5770         ztest_spa = spa;
5771         txg_wait_synced(spa_get_dsl(spa), 0);
5772         ztest_reguid(NULL, 0);
5773 
5774         spa_close(spa, FTAG);
5775         kernel_fini();
5776 }
5777 
5778 void
5779 print_time(hrtime_t t, char *timebuf)
5780 {
5781         hrtime_t s = t / NANOSEC;
5782         hrtime_t m = s / 60;
5783         hrtime_t h = m / 60;
5784         hrtime_t d = h / 24;
5785 
5786         s -= m * 60;
5787         m -= h * 60;
5788         h -= d * 24;
5789 
5790         timebuf[0] = '\0';
5791 
5792         if (d)
5793                 (void) sprintf(timebuf,
5794                     "%llud%02lluh%02llum%02llus", d, h, m, s);
5795         else if (h)
5796                 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5797         else if (m)
5798                 (void) sprintf(timebuf, "%llum%02llus", m, s);
5799         else
5800                 (void) sprintf(timebuf, "%llus", s);
5801 }
5802 
5803 static nvlist_t *
5804 make_random_props()
5805 {
5806         nvlist_t *props;
5807 
5808         VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5809         if (ztest_random(2) == 0)
5810                 return (props);
5811         VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5812 
5813         return (props);
5814 }
5815 
5816 /*
5817  * Create a storage pool with the given name and initial vdev size.
5818  * Then test spa_freeze() functionality.
5819  */
5820 static void
5821 ztest_init(ztest_shared_t *zs)
5822 {
5823         spa_t *spa;
5824         nvlist_t *nvroot, *props;
5825 
5826         VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5827         VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5828 
5829         kernel_init(FREAD | FWRITE);
5830 
5831         /*
5832          * Create the storage pool.
5833          */
5834         (void) spa_destroy(ztest_opts.zo_pool);
5835         ztest_shared->zs_vdev_next_leaf = 0;
5836         zs->zs_splits = 0;
5837         zs->zs_mirrors = ztest_opts.zo_mirrors;
5838         nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5839             0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5840         props = make_random_props();
5841         for (int i = 0; i < SPA_FEATURES; i++) {
5842                 char buf[1024];
5843                 (void) snprintf(buf, sizeof (buf), "feature@%s",
5844                     spa_feature_table[i].fi_uname);
5845                 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5846         }
5847         VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5848         nvlist_free(nvroot);
5849 
5850         VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5851         zs->zs_metaslab_sz =
5852             1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5853 
5854         spa_close(spa, FTAG);
5855 
5856         kernel_fini();
5857 
5858         ztest_run_zdb(ztest_opts.zo_pool);
5859 
5860         ztest_freeze();
5861 
5862         ztest_run_zdb(ztest_opts.zo_pool);
5863 
5864         (void) rwlock_destroy(&ztest_name_lock);
5865         (void) _mutex_destroy(&ztest_vdev_lock);
5866 }
5867 
5868 static void
5869 setup_data_fd(void)
5870 {
5871         static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5872 
5873         ztest_fd_data = mkstemp(ztest_name_data);
5874         ASSERT3S(ztest_fd_data, >=, 0);
5875         (void) unlink(ztest_name_data);
5876 }
5877 
5878 
5879 static int
5880 shared_data_size(ztest_shared_hdr_t *hdr)
5881 {
5882         int size;
5883 
5884         size = hdr->zh_hdr_size;
5885         size += hdr->zh_opts_size;
5886         size += hdr->zh_size;
5887         size += hdr->zh_stats_size * hdr->zh_stats_count;
5888         size += hdr->zh_ds_size * hdr->zh_ds_count;
5889 
5890         return (size);
5891 }
5892 
5893 static void
5894 setup_hdr(void)
5895 {
5896         int size;
5897         ztest_shared_hdr_t *hdr;
5898 
5899         hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5900             PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5901         ASSERT(hdr != MAP_FAILED);
5902 
5903         VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5904 
5905         hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5906         hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5907         hdr->zh_size = sizeof (ztest_shared_t);
5908         hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5909         hdr->zh_stats_count = ZTEST_FUNCS;
5910         hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5911         hdr->zh_ds_count = ztest_opts.zo_datasets;
5912 
5913         size = shared_data_size(hdr);
5914         VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5915 
5916         (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5917 }
5918 
5919 static void
5920 setup_data(void)
5921 {
5922         int size, offset;
5923         ztest_shared_hdr_t *hdr;
5924         uint8_t *buf;
5925 
5926         hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5927             PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5928         ASSERT(hdr != MAP_FAILED);
5929 
5930         size = shared_data_size(hdr);
5931 
5932         (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5933         hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5934             PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5935         ASSERT(hdr != MAP_FAILED);
5936         buf = (uint8_t *)hdr;
5937 
5938         offset = hdr->zh_hdr_size;
5939         ztest_shared_opts = (void *)&buf[offset];
5940         offset += hdr->zh_opts_size;
5941         ztest_shared = (void *)&buf[offset];
5942         offset += hdr->zh_size;
5943         ztest_shared_callstate = (void *)&buf[offset];
5944         offset += hdr->zh_stats_size * hdr->zh_stats_count;
5945         ztest_shared_ds = (void *)&buf[offset];
5946 }
5947 
5948 static boolean_t
5949 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5950 {
5951         pid_t pid;
5952         int status;
5953         char *cmdbuf = NULL;
5954 
5955         pid = fork();
5956 
5957         if (cmd == NULL) {
5958                 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5959                 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
5960                 cmd = cmdbuf;
5961         }
5962 
5963         if (pid == -1)
5964                 fatal(1, "fork failed");
5965 
5966         if (pid == 0) { /* child */
5967                 char *emptyargv[2] = { cmd, NULL };
5968                 char fd_data_str[12];
5969 
5970                 struct rlimit rl = { 1024, 1024 };
5971                 (void) setrlimit(RLIMIT_NOFILE, &rl);
5972 
5973                 (void) close(ztest_fd_rand);
5974                 VERIFY3U(11, >=,
5975                     snprintf(fd_data_str, 12, "%d", ztest_fd_data));
5976                 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
5977 
5978                 (void) enable_extended_FILE_stdio(-1, -1);
5979                 if (libpath != NULL)
5980                         VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5981                 (void) execv(cmd, emptyargv);
5982                 ztest_dump_core = B_FALSE;
5983                 fatal(B_TRUE, "exec failed: %s", cmd);
5984         }
5985 
5986         if (cmdbuf != NULL) {
5987                 umem_free(cmdbuf, MAXPATHLEN);
5988                 cmd = NULL;
5989         }
5990 
5991         while (waitpid(pid, &status, 0) != pid)
5992                 continue;
5993         if (statusp != NULL)
5994                 *statusp = status;
5995 
5996         if (WIFEXITED(status)) {
5997                 if (WEXITSTATUS(status) != 0) {
5998                         (void) fprintf(stderr, "child exited with code %d\n",
5999                             WEXITSTATUS(status));
6000                         exit(2);
6001                 }
6002                 return (B_FALSE);
6003         } else if (WIFSIGNALED(status)) {
6004                 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6005                         (void) fprintf(stderr, "child died with signal %d\n",
6006                             WTERMSIG(status));
6007                         exit(3);
6008                 }
6009                 return (B_TRUE);
6010         } else {
6011                 (void) fprintf(stderr, "something strange happened to child\n");
6012                 exit(4);
6013                 /* NOTREACHED */
6014         }
6015 }
6016 
6017 static void
6018 ztest_run_init(void)
6019 {
6020         ztest_shared_t *zs = ztest_shared;
6021 
6022         ASSERT(ztest_opts.zo_init != 0);
6023 
6024         /*
6025          * Blow away any existing copy of zpool.cache
6026          */
6027         (void) remove(spa_config_path);
6028 
6029         /*
6030          * Create and initialize our storage pool.
6031          */
6032         for (int i = 1; i <= ztest_opts.zo_init; i++) {
6033                 bzero(zs, sizeof (ztest_shared_t));
6034                 if (ztest_opts.zo_verbose >= 3 &&
6035                     ztest_opts.zo_init != 1) {
6036                         (void) printf("ztest_init(), pass %d\n", i);
6037                 }
6038                 ztest_init(zs);
6039         }
6040 }
6041 
6042 int
6043 main(int argc, char **argv)
6044 {
6045         int kills = 0;
6046         int iters = 0;
6047         int older = 0;
6048         int newer = 0;
6049         ztest_shared_t *zs;
6050         ztest_info_t *zi;
6051         ztest_shared_callstate_t *zc;
6052         char timebuf[100];
6053         char numbuf[6];
6054         spa_t *spa;
6055         char *cmd;
6056         boolean_t hasalt;
6057         char *fd_data_str = getenv("ZTEST_FD_DATA");
6058 
6059         (void) setvbuf(stdout, NULL, _IOLBF, 0);
6060 
6061         dprintf_setup(&argc, argv);
6062         zfs_deadman_synctime = 300;
6063 
6064         ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6065         ASSERT3S(ztest_fd_rand, >=, 0);
6066 
6067         if (!fd_data_str) {
6068                 process_options(argc, argv);
6069 
6070                 setup_data_fd();
6071                 setup_hdr();
6072                 setup_data();
6073                 bcopy(&ztest_opts, ztest_shared_opts,
6074                     sizeof (*ztest_shared_opts));
6075         } else {
6076                 ztest_fd_data = atoi(fd_data_str);
6077                 setup_data();
6078                 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6079         }
6080         ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6081 
6082         /* Override location of zpool.cache */
6083         VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6084             ztest_opts.zo_dir), !=, -1);
6085 
6086         ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6087             UMEM_NOFAIL);
6088         zs = ztest_shared;
6089 
6090         if (fd_data_str) {
6091                 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6092                 metaslab_df_alloc_threshold =
6093                     zs->zs_metaslab_df_alloc_threshold;
6094 
6095                 if (zs->zs_do_init)
6096                         ztest_run_init();
6097                 else
6098                         ztest_run(zs);
6099                 exit(0);
6100         }
6101 
6102         hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6103 
6104         if (ztest_opts.zo_verbose >= 1) {
6105                 (void) printf("%llu vdevs, %d datasets, %d threads,"
6106                     " %llu seconds...\n",
6107                     (u_longlong_t)ztest_opts.zo_vdevs,
6108                     ztest_opts.zo_datasets,
6109                     ztest_opts.zo_threads,
6110                     (u_longlong_t)ztest_opts.zo_time);
6111         }
6112 
6113         cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6114         (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6115 
6116         zs->zs_do_init = B_TRUE;
6117         if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6118                 if (ztest_opts.zo_verbose >= 1) {
6119                         (void) printf("Executing older ztest for "
6120                             "initialization: %s\n", ztest_opts.zo_alt_ztest);
6121                 }
6122                 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6123                     ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6124         } else {
6125                 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6126         }
6127         zs->zs_do_init = B_FALSE;
6128 
6129         zs->zs_proc_start = gethrtime();
6130         zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6131 
6132         for (int f = 0; f < ZTEST_FUNCS; f++) {
6133                 zi = &ztest_info[f];
6134                 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6135                 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6136                         zc->zc_next = UINT64_MAX;
6137                 else
6138                         zc->zc_next = zs->zs_proc_start +
6139                             ztest_random(2 * zi->zi_interval[0] + 1);
6140         }
6141 
6142         /*
6143          * Run the tests in a loop.  These tests include fault injection
6144          * to verify that self-healing data works, and forced crashes
6145          * to verify that we never lose on-disk consistency.
6146          */
6147         while (gethrtime() < zs->zs_proc_stop) {
6148                 int status;
6149                 boolean_t killed;
6150 
6151                 /*
6152                  * Initialize the workload counters for each function.
6153                  */
6154                 for (int f = 0; f < ZTEST_FUNCS; f++) {
6155                         zc = ZTEST_GET_SHARED_CALLSTATE(f);
6156                         zc->zc_count = 0;
6157                         zc->zc_time = 0;
6158                 }
6159 
6160                 /* Set the allocation switch size */
6161                 zs->zs_metaslab_df_alloc_threshold =
6162                     ztest_random(zs->zs_metaslab_sz / 4) + 1;
6163 
6164                 if (!hasalt || ztest_random(2) == 0) {
6165                         if (hasalt && ztest_opts.zo_verbose >= 1) {
6166                                 (void) printf("Executing newer ztest: %s\n",
6167                                     cmd);
6168                         }
6169                         newer++;
6170                         killed = exec_child(cmd, NULL, B_TRUE, &status);
6171                 } else {
6172                         if (hasalt && ztest_opts.zo_verbose >= 1) {
6173                                 (void) printf("Executing older ztest: %s\n",
6174                                     ztest_opts.zo_alt_ztest);
6175                         }
6176                         older++;
6177                         killed = exec_child(ztest_opts.zo_alt_ztest,
6178                             ztest_opts.zo_alt_libpath, B_TRUE, &status);
6179                 }
6180 
6181                 if (killed)
6182                         kills++;
6183                 iters++;
6184 
6185                 if (ztest_opts.zo_verbose >= 1) {
6186                         hrtime_t now = gethrtime();
6187 
6188                         now = MIN(now, zs->zs_proc_stop);
6189                         print_time(zs->zs_proc_stop - now, timebuf);
6190                         nicenum(zs->zs_space, numbuf);
6191 
6192                         (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6193                             "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6194                             iters,
6195                             WIFEXITED(status) ? "Complete" : "SIGKILL",
6196                             (u_longlong_t)zs->zs_enospc_count,
6197                             100.0 * zs->zs_alloc / zs->zs_space,
6198                             numbuf,
6199                             100.0 * (now - zs->zs_proc_start) /
6200                             (ztest_opts.zo_time * NANOSEC), timebuf);
6201                 }
6202 
6203                 if (ztest_opts.zo_verbose >= 2) {
6204                         (void) printf("\nWorkload summary:\n\n");
6205                         (void) printf("%7s %9s   %s\n",
6206                             "Calls", "Time", "Function");
6207                         (void) printf("%7s %9s   %s\n",
6208                             "-----", "----", "--------");
6209                         for (int f = 0; f < ZTEST_FUNCS; f++) {
6210                                 Dl_info dli;
6211 
6212                                 zi = &ztest_info[f];
6213                                 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6214                                 print_time(zc->zc_time, timebuf);
6215                                 (void) dladdr((void *)zi->zi_func, &dli);
6216                                 (void) printf("%7llu %9s   %s\n",
6217                                     (u_longlong_t)zc->zc_count, timebuf,
6218                                     dli.dli_sname);
6219                         }
6220                         (void) printf("\n");
6221                 }
6222 
6223                 /*
6224                  * It's possible that we killed a child during a rename test,
6225                  * in which case we'll have a 'ztest_tmp' pool lying around
6226                  * instead of 'ztest'.  Do a blind rename in case this happened.
6227                  */
6228                 kernel_init(FREAD);
6229                 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6230                         spa_close(spa, FTAG);
6231                 } else {
6232                         char tmpname[MAXNAMELEN];
6233                         kernel_fini();
6234                         kernel_init(FREAD | FWRITE);
6235                         (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6236                             ztest_opts.zo_pool);
6237                         (void) spa_rename(tmpname, ztest_opts.zo_pool);
6238                 }
6239                 kernel_fini();
6240 
6241                 ztest_run_zdb(ztest_opts.zo_pool);
6242         }
6243 
6244         if (ztest_opts.zo_verbose >= 1) {
6245                 if (hasalt) {
6246                         (void) printf("%d runs of older ztest: %s\n", older,
6247                             ztest_opts.zo_alt_ztest);
6248                         (void) printf("%d runs of newer ztest: %s\n", newer,
6249                             cmd);
6250                 }
6251                 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6252                     kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6253         }
6254 
6255         umem_free(cmd, MAXNAMELEN);
6256 
6257         return (0);
6258 }