sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dataset.c

   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) 2011 by Delphix. All rights reserved.
  24  * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
  25  * All rights reserved.
  26  */
  27
  28 #include <sys/dmu_objset.h>
  29 #include <sys/dsl_dataset.h>
  30 #include <sys/dsl_dir.h>
  31 #include <sys/dsl_prop.h>
  32 #include <sys/dsl_synctask.h>
  33 #include <sys/dmu_traverse.h>
  34 #include <sys/dmu_tx.h>
  35 #include <sys/arc.h>
  36 #include <sys/zio.h>
  37 #include <sys/zap.h>
  38 #include <sys/unique.h>
  39 #include <sys/zfs_context.h>
  40 #include <sys/zfs_ioctl.h>
  41 #include <sys/spa.h>
  42 #include <sys/zfs_znode.h>
  43 #include <sys/zfs_onexit.h>
  44 #include <sys/zvol.h>
  45 #include <sys/dsl_scan.h>
  46 #include <sys/dsl_deadlist.h>
  47
  48 static char *dsl_reaper = "the grim reaper";
  49
  50 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
  51 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
  52 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
  53
  54 #define SWITCH64(x, y) \
  55         { \
  56                 uint64_t __tmp = (x); \
  57                 (x) = (y); \
  58                 (y) = __tmp; \
  59         }
  60
  61 #define DS_REF_MAX      (1ULL << 62)
  62
  63 #define DSL_DEADLIST_BLOCKSIZE  SPA_MAXBLOCKSIZE
  64
  65 #define DSL_DATASET_IS_DESTROYED(ds)    ((ds)->ds_owner == dsl_reaper)
  66
  67
  68 /*
  69  * Figure out how much of this delta should be propogated to the dsl_dir
  70  * layer.  If there's a refreservation, that space has already been
  71  * partially accounted for in our ancestors.
  72  */
  73 static int64_t
  74 parent_delta(dsl_dataset_t *ds, int64_t delta)
  75 {
  76         uint64_t old_bytes, new_bytes;
  77
  78         if (ds->ds_reserved == 0)
  79                 return (delta);
  80
  81         old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
  82         new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
  83
  84         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
  85         return (new_bytes - old_bytes);
  86 }
  87
  88 void
  89 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
  90 {
  91         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
  92         int compressed = BP_GET_PSIZE(bp);
  93         int uncompressed = BP_GET_UCSIZE(bp);
  94         int64_t delta;
  95
  96         dprintf_bp(bp, "ds=%p", ds);
  97
  98         ASSERT(dmu_tx_is_syncing(tx));
  99         /* It could have been compressed away to nothing */
 100         if (BP_IS_HOLE(bp))
 101                 return;
 102         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 103         ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
 104         if (ds == NULL) {
 105                 /*
 106                  * Account for the meta-objset space in its placeholder
 107                  * dsl_dir.
 108                  */
 109                 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
 110                 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
 111                     used, compressed, uncompressed, tx);
 112                 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
 113                 return;
 114         }
 115         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 116
 117         mutex_enter(&ds->ds_dir->dd_lock);
 118         mutex_enter(&ds->ds_lock);
 119         delta = parent_delta(ds, used);
 120         ds->ds_phys->ds_used_bytes += used;
 121         ds->ds_phys->ds_compressed_bytes += compressed;
 122         ds->ds_phys->ds_uncompressed_bytes += uncompressed;
 123         ds->ds_phys->ds_unique_bytes += used;
 124         mutex_exit(&ds->ds_lock);
 125         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 126             compressed, uncompressed, tx);
 127         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 128             DD_USED_REFRSRV, DD_USED_HEAD, tx);
 129         mutex_exit(&ds->ds_dir->dd_lock);
 130 }
 131
 132 int
 133 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 134     boolean_t async)
 135 {
 136         if (BP_IS_HOLE(bp))
 137                 return (0);
 138
 139         ASSERT(dmu_tx_is_syncing(tx));
 140         ASSERT(bp->blk_birth <= tx->tx_txg);
 141
 142         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 143         int compressed = BP_GET_PSIZE(bp);
 144         int uncompressed = BP_GET_UCSIZE(bp);
 145
 146         ASSERT(used > 0);
 147         if (ds == NULL) {
 148                 /*
 149                  * Account for the meta-objset space in its placeholder
 150                  * dataset.
 151                  */
 152                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 153
 154                 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
 155                     -used, -compressed, -uncompressed, tx);
 156                 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
 157                 return (used);
 158         }
 159         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 160
 161         ASSERT(!dsl_dataset_is_snapshot(ds));
 162         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 163
 164         if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
 165                 int64_t delta;
 166
 167                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 168                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 169
 170                 mutex_enter(&ds->ds_dir->dd_lock);
 171                 mutex_enter(&ds->ds_lock);
 172                 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
 173                     !DS_UNIQUE_IS_ACCURATE(ds));
 174                 delta = parent_delta(ds, -used);
 175                 ds->ds_phys->ds_unique_bytes -= used;
 176                 mutex_exit(&ds->ds_lock);
 177                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 178                     delta, -compressed, -uncompressed, tx);
 179                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 180                     DD_USED_REFRSRV, DD_USED_HEAD, tx);
 181                 mutex_exit(&ds->ds_dir->dd_lock);
 182         } else {
 183                 dprintf_bp(bp, "putting on dead list: %s", "");
 184                 if (async) {
 185                         /*
 186                          * We are here as part of zio's write done callback,
 187                          * which means we're a zio interrupt thread.  We can't
 188                          * call dsl_deadlist_insert() now because it may block
 189                          * waiting for I/O.  Instead, put bp on the deferred
 190                          * queue and let dsl_pool_sync() finish the job.
 191                          */
 192                         bplist_append(&ds->ds_pending_deadlist, bp);
 193                 } else {
 194                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 195                 }
 196                 ASSERT3U(ds->ds_prev->ds_object, ==,
 197                     ds->ds_phys->ds_prev_snap_obj);
 198                 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
 199                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 200                 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
 201                     ds->ds_object && bp->blk_birth >
 202                     ds->ds_prev->ds_phys->ds_prev_snap_txg) {
 203                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 204                         mutex_enter(&ds->ds_prev->ds_lock);
 205                         ds->ds_prev->ds_phys->ds_unique_bytes += used;
 206                         mutex_exit(&ds->ds_prev->ds_lock);
 207                 }
 208                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 209                         dsl_dir_transfer_space(ds->ds_dir, used,
 210                             DD_USED_HEAD, DD_USED_SNAP, tx);
 211                 }
 212         }
 213         mutex_enter(&ds->ds_lock);
 214         ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
 215         ds->ds_phys->ds_used_bytes -= used;
 216         ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
 217         ds->ds_phys->ds_compressed_bytes -= compressed;
 218         ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
 219         ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
 220         mutex_exit(&ds->ds_lock);
 221
 222         return (used);
 223 }
 224
 225 uint64_t
 226 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
 227 {
 228         uint64_t trysnap = 0;
 229
 230         if (ds == NULL)
 231                 return (0);
 232         /*
 233          * The snapshot creation could fail, but that would cause an
 234          * incorrect FALSE return, which would only result in an
 235          * overestimation of the amount of space that an operation would
 236          * consume, which is OK.
 237          *
 238          * There's also a small window where we could miss a pending
 239          * snapshot, because we could set the sync task in the quiescing
 240          * phase.  So this should only be used as a guess.
 241          */
 242         if (ds->ds_trysnap_txg >
 243             spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
 244                 trysnap = ds->ds_trysnap_txg;
 245         return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
 246 }
 247
 248 boolean_t
 249 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
 250     uint64_t blk_birth)
 251 {
 252         if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
 253                 return (B_FALSE);
 254
 255         ddt_prefetch(dsl_dataset_get_spa(ds), bp);
 256
 257         return (B_TRUE);
 258 }
 259
 260 /* ARGSUSED */
 261 static void
 262 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
 263 {
 264         dsl_dataset_t *ds = dsv;
 265
 266         ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
 267
 268         unique_remove(ds->ds_fsid_guid);
 269
 270         if (ds->ds_objset != NULL)
 271                 dmu_objset_evict(ds->ds_objset);
 272
 273         if (ds->ds_prev) {
 274                 dsl_dataset_drop_ref(ds->ds_prev, ds);
 275                 ds->ds_prev = NULL;
 276         }
 277
 278         bplist_destroy(&ds->ds_pending_deadlist);
 279         if (db != NULL) {
 280                 dsl_deadlist_close(&ds->ds_deadlist);
 281         } else {
 282                 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
 283                 ASSERT(!ds->ds_deadlist.dl_oldfmt);
 284         }
 285         if (ds->ds_dir)
 286                 dsl_dir_close(ds->ds_dir, ds);
 287
 288         ASSERT(!list_link_active(&ds->ds_synced_link));
 289
 290         if (mutex_owned(&ds->ds_lock))
 291                 mutex_exit(&ds->ds_lock);
 292         mutex_destroy(&ds->ds_lock);
 293         mutex_destroy(&ds->ds_recvlock);
 294         if (mutex_owned(&ds->ds_opening_lock))
 295                 mutex_exit(&ds->ds_opening_lock);
 296         mutex_destroy(&ds->ds_opening_lock);
 297         rw_destroy(&ds->ds_rwlock);
 298         cv_destroy(&ds->ds_exclusive_cv);
 299
 300         kmem_free(ds, sizeof (dsl_dataset_t));
 301 }
 302
 303 static int
 304 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 305 {
 306         dsl_dataset_phys_t *headphys;
 307         int err;
 308         dmu_buf_t *headdbuf;
 309         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 310         objset_t *mos = dp->dp_meta_objset;
 311
 312         if (ds->ds_snapname[0])
 313                 return (0);
 314         if (ds->ds_phys->ds_next_snap_obj == 0)
 315                 return (0);
 316
 317         err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
 318             FTAG, &headdbuf);
 319         if (err)
 320                 return (err);
 321         headphys = headdbuf->db_data;
 322         err = zap_value_search(dp->dp_meta_objset,
 323             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 324         dmu_buf_rele(headdbuf, FTAG);
 325         return (err);
 326 }
 327
 328 static int
 329 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 330 {
 331         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 332         uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
 333         matchtype_t mt;
 334         int err;
 335
 336         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 337                 mt = MT_FIRST;
 338         else
 339                 mt = MT_EXACT;
 340
 341         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 342             value, mt, NULL, 0, NULL);
 343         if (err == ENOTSUP && mt == MT_FIRST)
 344                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 345         return (err);
 346 }
 347
 348 static int
 349 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
 350 {
 351         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 352         uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
 353         matchtype_t mt;
 354         int err;
 355
 356         dsl_dir_snap_cmtime_update(ds->ds_dir);
 357
 358         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 359                 mt = MT_FIRST;
 360         else
 361                 mt = MT_EXACT;
 362
 363         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 364         if (err == ENOTSUP && mt == MT_FIRST)
 365                 err = zap_remove(mos, snapobj, name, tx);
 366         return (err);
 367 }
 368
 369 static int
 370 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 371     dsl_dataset_t **dsp)
 372 {
 373         objset_t *mos = dp->dp_meta_objset;
 374         dmu_buf_t *dbuf;
 375         dsl_dataset_t *ds;
 376         int err;
 377         dmu_object_info_t doi;
 378
 379         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
 380             dsl_pool_sync_context(dp));
 381
 382         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 383         if (err)
 384                 return (err);
 385
 386         /* Make sure dsobj has the correct object type. */
 387         dmu_object_info_from_db(dbuf, &doi);
 388         if (doi.doi_type != DMU_OT_DSL_DATASET)
 389                 return (EINVAL);
 390
 391         ds = dmu_buf_get_user(dbuf);
 392         if (ds == NULL) {
 393                 dsl_dataset_t *winner;
 394
 395                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 396                 ds->ds_dbuf = dbuf;
 397                 ds->ds_object = dsobj;
 398                 ds->ds_phys = dbuf->db_data;
 399
 400                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 401                 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
 402                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 403                 rw_init(&ds->ds_rwlock, 0, 0, 0);
 404                 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
 405
 406                 bplist_create(&ds->ds_pending_deadlist);
 407                 dsl_deadlist_open(&ds->ds_deadlist,
 408                     mos, ds->ds_phys->ds_deadlist_obj);
 409
 410                 if (err == 0) {
 411                         err = dsl_dir_open_obj(dp,
 412                             ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
 413                 }
 414                 if (err) {
 415                         mutex_destroy(&ds->ds_lock);
 416                         mutex_destroy(&ds->ds_recvlock);
 417                         mutex_destroy(&ds->ds_opening_lock);
 418                         rw_destroy(&ds->ds_rwlock);
 419                         cv_destroy(&ds->ds_exclusive_cv);
 420                         bplist_destroy(&ds->ds_pending_deadlist);
 421                         dsl_deadlist_close(&ds->ds_deadlist);
 422                         kmem_free(ds, sizeof (dsl_dataset_t));
 423                         dmu_buf_rele(dbuf, tag);
 424                         return (err);
 425                 }
 426
 427                 if (!dsl_dataset_is_snapshot(ds)) {
 428                         ds->ds_snapname[0] = '\0';
 429                         if (ds->ds_phys->ds_prev_snap_obj) {
 430                                 err = dsl_dataset_get_ref(dp,
 431                                     ds->ds_phys->ds_prev_snap_obj,
 432                                     ds, &ds->ds_prev);
 433                         }
 434                 } else {
 435                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 436                                 err = dsl_dataset_get_snapname(ds);
 437                         if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
 438                                 err = zap_count(
 439                                     ds->ds_dir->dd_pool->dp_meta_objset,
 440                                     ds->ds_phys->ds_userrefs_obj,
 441                                     &ds->ds_userrefs);
 442                         }
 443                 }
 444
 445                 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
 446                         /*
 447                          * In sync context, we're called with either no lock
 448                          * or with the write lock.  If we're not syncing,
 449                          * we're always called with the read lock held.
 450                          */
 451                         boolean_t need_lock =
 452                             !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
 453                             dsl_pool_sync_context(dp);
 454
 455                         if (need_lock)
 456                                 rw_enter(&dp->dp_config_rwlock, RW_READER);
 457
 458                         err = dsl_prop_get_ds(ds,
 459                             "refreservation", sizeof (uint64_t), 1,
 460                             &ds->ds_reserved, NULL);
 461                         if (err == 0) {
 462                                 err = dsl_prop_get_ds(ds,
 463                                     "refquota", sizeof (uint64_t), 1,
 464                                     &ds->ds_quota, NULL);
 465                         }
 466
 467                         if (need_lock)
 468                                 rw_exit(&dp->dp_config_rwlock);
 469                 } else {
 470                         ds->ds_reserved = ds->ds_quota = 0;
 471                 }
 472
 473                 if (err == 0) {
 474                         winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
 475                             dsl_dataset_evict);
 476                 }
 477                 if (err || winner) {
 478                         bplist_destroy(&ds->ds_pending_deadlist);
 479                         dsl_deadlist_close(&ds->ds_deadlist);
 480                         if (ds->ds_prev)
 481                                 dsl_dataset_drop_ref(ds->ds_prev, ds);
 482                         dsl_dir_close(ds->ds_dir, ds);
 483                         mutex_destroy(&ds->ds_lock);
 484                         mutex_destroy(&ds->ds_recvlock);
 485                         mutex_destroy(&ds->ds_opening_lock);
 486                         rw_destroy(&ds->ds_rwlock);
 487                         cv_destroy(&ds->ds_exclusive_cv);
 488                         kmem_free(ds, sizeof (dsl_dataset_t));
 489                         if (err) {
 490                                 dmu_buf_rele(dbuf, tag);
 491                                 return (err);
 492                         }
 493                         ds = winner;
 494                 } else {
 495                         ds->ds_fsid_guid =
 496                             unique_insert(ds->ds_phys->ds_fsid_guid);
 497                 }
 498         }
 499         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 500         ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
 501         ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
 502             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 503             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 504         mutex_enter(&ds->ds_lock);
 505         if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
 506                 mutex_exit(&ds->ds_lock);
 507                 dmu_buf_rele(ds->ds_dbuf, tag);
 508                 return (ENOENT);
 509         }
 510         mutex_exit(&ds->ds_lock);
 511         *dsp = ds;
 512         return (0);
 513 }
 514
 515 static int
 516 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
 517 {
 518         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 519
 520         /*
 521          * In syncing context we don't want the rwlock lock: there
 522          * may be an existing writer waiting for sync phase to
 523          * finish.  We don't need to worry about such writers, since
 524          * sync phase is single-threaded, so the writer can't be
 525          * doing anything while we are active.
 526          */
 527         if (dsl_pool_sync_context(dp)) {
 528                 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
 529                 return (0);
 530         }
 531
 532         /*
 533          * Normal users will hold the ds_rwlock as a READER until they
 534          * are finished (i.e., call dsl_dataset_rele()).  "Owners" will
 535          * drop their READER lock after they set the ds_owner field.
 536          *
 537          * If the dataset is being destroyed, the destroy thread will
 538          * obtain a WRITER lock for exclusive access after it's done its
 539          * open-context work and then change the ds_owner to
 540          * dsl_reaper once destruction is assured.  So threads
 541          * may block here temporarily, until the "destructability" of
 542          * the dataset is determined.
 543          */
 544         ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
 545         mutex_enter(&ds->ds_lock);
 546         while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
 547                 rw_exit(&dp->dp_config_rwlock);
 548                 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
 549                 if (DSL_DATASET_IS_DESTROYED(ds)) {
 550                         mutex_exit(&ds->ds_lock);
 551                         dsl_dataset_drop_ref(ds, tag);
 552                         rw_enter(&dp->dp_config_rwlock, RW_READER);
 553                         return (ENOENT);
 554                 }
 555                 /*
 556                  * The dp_config_rwlock lives above the ds_lock. And
 557                  * we need to check DSL_DATASET_IS_DESTROYED() while
 558                  * holding the ds_lock, so we have to drop and reacquire
 559                  * the ds_lock here.
 560                  */
 561                 mutex_exit(&ds->ds_lock);
 562                 rw_enter(&dp->dp_config_rwlock, RW_READER);
 563                 mutex_enter(&ds->ds_lock);
 564         }
 565         mutex_exit(&ds->ds_lock);
 566         return (0);
 567 }
 568
 569 int
 570 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 571     dsl_dataset_t **dsp)
 572 {
 573         int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
 574
 575         if (err)
 576                 return (err);
 577         return (dsl_dataset_hold_ref(*dsp, tag));
 578 }
 579
 580 int
 581 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
 582     void *tag, dsl_dataset_t **dsp)
 583 {
 584         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 585         if (err)
 586                 return (err);
 587         if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
 588                 dsl_dataset_rele(*dsp, tag);
 589                 *dsp = NULL;
 590                 return (EBUSY);
 591         }
 592         return (0);
 593 }
 594
 595 int
 596 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
 597 {
 598         dsl_dir_t *dd;
 599         dsl_pool_t *dp;
 600         const char *snapname;
 601         uint64_t obj;
 602         int err = 0;
 603
 604         err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
 605         if (err)
 606                 return (err);
 607
 608         dp = dd->dd_pool;
 609         obj = dd->dd_phys->dd_head_dataset_obj;
 610         rw_enter(&dp->dp_config_rwlock, RW_READER);
 611         if (obj)
 612                 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
 613         else
 614                 err = ENOENT;
 615         if (err)
 616                 goto out;
 617
 618         err = dsl_dataset_hold_ref(*dsp, tag);
 619
 620         /* we may be looking for a snapshot */
 621         if (err == 0 && snapname != NULL) {
 622                 dsl_dataset_t *ds = NULL;
 623
 624                 if (*snapname++ != '@') {
 625                         dsl_dataset_rele(*dsp, tag);
 626                         err = ENOENT;
 627                         goto out;
 628                 }
 629
 630                 dprintf("looking for snapshot '%s'\n", snapname);
 631                 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
 632                 if (err == 0)
 633                         err = dsl_dataset_get_ref(dp, obj, tag, &ds);
 634                 dsl_dataset_rele(*dsp, tag);
 635
 636                 ASSERT3U((err == 0), ==, (ds != NULL));
 637
 638                 if (ds) {
 639                         mutex_enter(&ds->ds_lock);
 640                         if (ds->ds_snapname[0] == 0)
 641                                 (void) strlcpy(ds->ds_snapname, snapname,
 642                                     sizeof (ds->ds_snapname));
 643                         mutex_exit(&ds->ds_lock);
 644                         err = dsl_dataset_hold_ref(ds, tag);
 645                         *dsp = err ? NULL : ds;
 646                 }
 647         }
 648 out:
 649         rw_exit(&dp->dp_config_rwlock);
 650         dsl_dir_close(dd, FTAG);
 651         return (err);
 652 }
 653
 654 int
 655 dsl_dataset_own(const char *name, boolean_t inconsistentok,
 656     void *tag, dsl_dataset_t **dsp)
 657 {
 658         int err = dsl_dataset_hold(name, tag, dsp);
 659         if (err)
 660                 return (err);
 661         if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
 662                 dsl_dataset_rele(*dsp, tag);
 663                 return (EBUSY);
 664         }
 665         return (0);
 666 }
 667
 668 void
 669 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 670 {
 671         if (ds == NULL) {
 672                 (void) strcpy(name, "mos");
 673         } else {
 674                 dsl_dir_name(ds->ds_dir, name);
 675                 VERIFY(0 == dsl_dataset_get_snapname(ds));
 676                 if (ds->ds_snapname[0]) {
 677                         (void) strcat(name, "@");
 678                         /*
 679                          * We use a "recursive" mutex so that we
 680                          * can call dprintf_ds() with ds_lock held.
 681                          */
 682                         if (!MUTEX_HELD(&ds->ds_lock)) {
 683                                 mutex_enter(&ds->ds_lock);
 684                                 (void) strcat(name, ds->ds_snapname);
 685                                 mutex_exit(&ds->ds_lock);
 686                         } else {
 687                                 (void) strcat(name, ds->ds_snapname);
 688                         }
 689                 }
 690         }
 691 }
 692
 693 static int
 694 dsl_dataset_namelen(dsl_dataset_t *ds)
 695 {
 696         int result;
 697
 698         if (ds == NULL) {
 699                 result = 3;     /* "mos" */
 700         } else {
 701                 result = dsl_dir_namelen(ds->ds_dir);
 702                 VERIFY(0 == dsl_dataset_get_snapname(ds));
 703                 if (ds->ds_snapname[0]) {
 704                         ++result;       /* adding one for the @-sign */
 705                         if (!MUTEX_HELD(&ds->ds_lock)) {
 706                                 mutex_enter(&ds->ds_lock);
 707                                 result += strlen(ds->ds_snapname);
 708                                 mutex_exit(&ds->ds_lock);
 709                         } else {
 710                                 result += strlen(ds->ds_snapname);
 711                         }
 712                 }
 713         }
 714
 715         return (result);
 716 }
 717
 718 void
 719 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
 720 {
 721         dmu_buf_rele(ds->ds_dbuf, tag);
 722 }
 723
 724 void
 725 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 726 {
 727         if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
 728                 rw_exit(&ds->ds_rwlock);
 729         }
 730         dsl_dataset_drop_ref(ds, tag);
 731 }
 732
 733 void
 734 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 735 {
 736         ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
 737             (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
 738
 739         mutex_enter(&ds->ds_lock);
 740         ds->ds_owner = NULL;
 741         if (RW_WRITE_HELD(&ds->ds_rwlock)) {
 742                 rw_exit(&ds->ds_rwlock);
 743                 cv_broadcast(&ds->ds_exclusive_cv);
 744         }
 745         mutex_exit(&ds->ds_lock);
 746         if (ds->ds_dbuf)
 747                 dsl_dataset_drop_ref(ds, tag);
 748         else
 749                 dsl_dataset_evict(NULL, ds);
 750 }
 751
 752 boolean_t
 753 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
 754 {
 755         boolean_t gotit = FALSE;
 756
 757         mutex_enter(&ds->ds_lock);
 758         if (ds->ds_owner == NULL &&
 759             (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
 760                 ds->ds_owner = tag;
 761                 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
 762                         rw_exit(&ds->ds_rwlock);
 763                 gotit = TRUE;
 764         }
 765         mutex_exit(&ds->ds_lock);
 766         return (gotit);
 767 }
 768
 769 void
 770 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
 771 {
 772         ASSERT3P(owner, ==, ds->ds_owner);
 773         if (!RW_WRITE_HELD(&ds->ds_rwlock))
 774                 rw_enter(&ds->ds_rwlock, RW_WRITER);
 775 }
 776
 777 uint64_t
 778 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 779     uint64_t flags, dmu_tx_t *tx)
 780 {
 781         dsl_pool_t *dp = dd->dd_pool;
 782         dmu_buf_t *dbuf;
 783         dsl_dataset_phys_t *dsphys;
 784         uint64_t dsobj;
 785         objset_t *mos = dp->dp_meta_objset;
 786
 787         if (origin == NULL)
 788                 origin = dp->dp_origin_snap;
 789
 790         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 791         ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
 792         ASSERT(dmu_tx_is_syncing(tx));
 793         ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
 794
 795         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 796             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 797         VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 798         dmu_buf_will_dirty(dbuf, tx);
 799         dsphys = dbuf->db_data;
 800         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 801         dsphys->ds_dir_obj = dd->dd_object;
 802         dsphys->ds_flags = flags;
 803         dsphys->ds_fsid_guid = unique_create();
 804         (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 805             sizeof (dsphys->ds_guid));
 806         dsphys->ds_snapnames_zapobj =
 807             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 808             DMU_OT_NONE, 0, tx);
 809         dsphys->ds_creation_time = gethrestime_sec();
 810         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 811
 812         if (origin == NULL) {
 813                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 814         } else {
 815                 dsl_dataset_t *ohds;
 816
 817                 dsphys->ds_prev_snap_obj = origin->ds_object;
 818                 dsphys->ds_prev_snap_txg =
 819                     origin->ds_phys->ds_creation_txg;
 820                 dsphys->ds_used_bytes =
 821                     origin->ds_phys->ds_used_bytes;
 822                 dsphys->ds_compressed_bytes =
 823                     origin->ds_phys->ds_compressed_bytes;
 824                 dsphys->ds_uncompressed_bytes =
 825                     origin->ds_phys->ds_uncompressed_bytes;
 826                 dsphys->ds_bp = origin->ds_phys->ds_bp;
 827                 dsphys->ds_flags |= origin->ds_phys->ds_flags;
 828
 829                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 830                 origin->ds_phys->ds_num_children++;
 831
 832                 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
 833                     origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
 834                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 835                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 836                 dsl_dataset_rele(ohds, FTAG);
 837
 838                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 839                         if (origin->ds_phys->ds_next_clones_obj == 0) {
 840                                 origin->ds_phys->ds_next_clones_obj =
 841                                     zap_create(mos,
 842                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 843                         }
 844                         VERIFY(0 == zap_add_int(mos,
 845                             origin->ds_phys->ds_next_clones_obj,
 846                             dsobj, tx));
 847                 }
 848
 849                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 850                 dd->dd_phys->dd_origin_obj = origin->ds_object;
 851                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 852                         if (origin->ds_dir->dd_phys->dd_clones == 0) {
 853                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 854                                 origin->ds_dir->dd_phys->dd_clones =
 855                                     zap_create(mos,
 856                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 857                         }
 858                         VERIFY3U(0, ==, zap_add_int(mos,
 859                             origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
 860                 }
 861         }
 862
 863         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 864                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 865
 866         dmu_buf_rele(dbuf, FTAG);
 867
 868         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 869         dd->dd_phys->dd_head_dataset_obj = dsobj;
 870
 871         return (dsobj);
 872 }
 873
 874 uint64_t
 875 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 876     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 877 {
 878         dsl_pool_t *dp = pdd->dd_pool;
 879         uint64_t dsobj, ddobj;
 880         dsl_dir_t *dd;
 881
 882         ASSERT(lastname[0] != '@');
 883
 884         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 885         VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
 886
 887         dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
 888
 889         dsl_deleg_set_create_perms(dd, tx, cr);
 890
 891         dsl_dir_close(dd, FTAG);
 892
 893         /*
 894          * If we are creating a clone, make sure we zero out any stale
 895          * data from the origin snapshots zil header.
 896          */
 897         if (origin != NULL) {
 898                 dsl_dataset_t *ds;
 899                 objset_t *os;
 900
 901                 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 902                 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
 903                 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 904                 dsl_dataset_dirty(ds, tx);
 905                 dsl_dataset_rele(ds, FTAG);
 906         }
 907
 908         return (dsobj);
 909 }
 910
 911 struct destroyarg {
 912         dsl_sync_task_group_t *dstg;
 913         char *snapname;
 914         char *failed;
 915         boolean_t defer;
 916 };
 917
 918 static int
 919 dsl_snapshot_destroy_one(const char *name, void *arg)
 920 {
 921         struct destroyarg *da = arg;
 922         dsl_dataset_t *ds;
 923         int err;
 924         char *dsname;
 925
 926         dsname = kmem_asprintf("%s@%s", name, da->snapname);
 927         err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds);
 928         strfree(dsname);
 929         if (err == 0) {
 930                 struct dsl_ds_destroyarg *dsda;
 931
 932                 dsl_dataset_make_exclusive(ds, da->dstg);
 933                 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP);
 934                 dsda->ds = ds;
 935                 dsda->defer = da->defer;
 936                 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
 937                     dsl_dataset_destroy_sync, dsda, da->dstg, 0);
 938         } else if (err == ENOENT) {
 939                 err = 0;
 940         } else {
 941                 (void) strcpy(da->failed, name);
 942         }
 943         return (err);
 944 }
 945
 946 /*
 947  * Destroy 'snapname' in all descendants of 'fsname'.
 948  */
 949 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
 950 int
 951 dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer)
 952 {
 953         int err;
 954         struct destroyarg da;
 955         dsl_sync_task_t *dst;
 956         spa_t *spa;
 957
 958         err = spa_open(fsname, &spa, FTAG);
 959         if (err)
 960                 return (err);
 961         da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
 962         da.snapname = snapname;
 963         da.failed = fsname;
 964         da.defer = defer;
 965
 966         err = dmu_objset_find(fsname,
 967             dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
 968
 969         if (err == 0)
 970                 err = dsl_sync_task_group_wait(da.dstg);
 971
 972         for (dst = list_head(&da.dstg->dstg_tasks); dst;
 973             dst = list_next(&da.dstg->dstg_tasks, dst)) {
 974                 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
 975                 dsl_dataset_t *ds = dsda->ds;
 976
 977                 /*
 978                  * Return the file system name that triggered the error
 979                  */
 980                 if (dst->dst_err) {
 981                         dsl_dataset_name(ds, fsname);
 982                         *strchr(fsname, '@') = '\0';
 983                 }
 984                 ASSERT3P(dsda->rm_origin, ==, NULL);
 985                 dsl_dataset_disown(ds, da.dstg);
 986                 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
 987         }
 988
 989         dsl_sync_task_group_destroy(da.dstg);
 990         spa_close(spa, FTAG);
 991         return (err);
 992 }
 993
 994 static boolean_t
 995 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
 996 {
 997         boolean_t might_destroy = B_FALSE;
 998
 999         mutex_enter(&ds->ds_lock);
1000         if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
1001             DS_IS_DEFER_DESTROY(ds))
1002                 might_destroy = B_TRUE;
1003         mutex_exit(&ds->ds_lock);
1004
1005         return (might_destroy);
1006 }
1007
1008 /*
1009  * If we're removing a clone, and these three conditions are true:
1010  *      1) the clone's origin has no other children
1011  *      2) the clone's origin has no user references
1012  *      3) the clone's origin has been marked for deferred destruction
1013  * Then, prepare to remove the origin as part of this sync task group.
1014  */
1015 static int
1016 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1017 {
1018         dsl_dataset_t *ds = dsda->ds;
1019         dsl_dataset_t *origin = ds->ds_prev;
1020
1021         if (dsl_dataset_might_destroy_origin(origin)) {
1022                 char *name;
1023                 int namelen;
1024                 int error;
1025
1026                 namelen = dsl_dataset_namelen(origin) + 1;
1027                 name = kmem_alloc(namelen, KM_SLEEP);
1028                 dsl_dataset_name(origin, name);
1029 #ifdef _KERNEL
1030                 error = zfs_unmount_snap(name, NULL);
1031                 if (error) {
1032                         kmem_free(name, namelen);
1033                         return (error);
1034                 }
1035 #endif
1036                 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1037                 kmem_free(name, namelen);
1038                 if (error)
1039                         return (error);
1040                 dsda->rm_origin = origin;
1041                 dsl_dataset_make_exclusive(origin, tag);
1042         }
1043
1044         return (0);
1045 }
1046
1047 /*
1048  * ds must be opened as OWNER.  On return (whether successful or not),
1049  * ds will be closed and caller can no longer dereference it.
1050  */
1051 int
1052 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1053 {
1054         int err;
1055         dsl_sync_task_group_t *dstg;
1056         objset_t *os;
1057         dsl_dir_t *dd;
1058         uint64_t obj;
1059         struct dsl_ds_destroyarg dsda = { 0 };
1060         dsl_dataset_t dummy_ds = { 0 };
1061
1062         dsda.ds = ds;
1063
1064         if (dsl_dataset_is_snapshot(ds)) {
1065                 /* Destroying a snapshot is simpler */
1066                 dsl_dataset_make_exclusive(ds, tag);
1067
1068                 dsda.defer = defer;
1069                 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1070                     dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1071                     &dsda, tag, 0);
1072                 ASSERT3P(dsda.rm_origin, ==, NULL);
1073                 goto out;
1074         } else if (defer) {
1075                 err = EINVAL;
1076                 goto out;
1077         }
1078
1079         dd = ds->ds_dir;
1080         dummy_ds.ds_dir = dd;
1081         dummy_ds.ds_object = ds->ds_object;
1082
1083         /*
1084          * Check for errors and mark this ds as inconsistent, in
1085          * case we crash while freeing the objects.
1086          */
1087         err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1088             dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1089         if (err)
1090                 goto out;
1091
1092         err = dmu_objset_from_ds(ds, &os);
1093         if (err)
1094                 goto out;
1095
1096         /*
1097          * remove the objects in open context, so that we won't
1098          * have too much to do in syncing context.
1099          */
1100         for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1101             ds->ds_phys->ds_prev_snap_txg)) {
1102                 /*
1103                  * Ignore errors, if there is not enough disk space
1104                  * we will deal with it in dsl_dataset_destroy_sync().
1105                  */
1106                 (void) dmu_free_object(os, obj);
1107         }
1108         if (err != ESRCH)
1109                 goto out;
1110
1111         /*
1112          * Only the ZIL knows how to free log blocks.
1113          */
1114         zil_destroy(dmu_objset_zil(os), B_FALSE);
1115
1116         /*
1117          * Sync out all in-flight IO.
1118          */
1119         txg_wait_synced(dd->dd_pool, 0);
1120
1121         /*
1122          * If we managed to free all the objects in open
1123          * context, the user space accounting should be zero.
1124          */
1125         if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1126             dmu_objset_userused_enabled(os)) {
1127                 uint64_t count;
1128
1129                 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1130                     count == 0);
1131                 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1132                     count == 0);
1133         }
1134
1135         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1136         err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1137         rw_exit(&dd->dd_pool->dp_config_rwlock);
1138
1139         if (err)
1140                 goto out;
1141
1142         /*
1143          * Blow away the dsl_dir + head dataset.
1144          */
1145         dsl_dataset_make_exclusive(ds, tag);
1146         /*
1147          * If we're removing a clone, we might also need to remove its
1148          * origin.
1149          */
1150         do {
1151                 dsda.need_prep = B_FALSE;
1152                 if (dsl_dir_is_clone(dd)) {
1153                         err = dsl_dataset_origin_rm_prep(&dsda, tag);
1154                         if (err) {
1155                                 dsl_dir_close(dd, FTAG);
1156                                 goto out;
1157                         }
1158                 }
1159
1160                 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1161                 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1162                     dsl_dataset_destroy_sync, &dsda, tag, 0);
1163                 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1164                     dsl_dir_destroy_sync, &dummy_ds, FTAG, 0);
1165                 err = dsl_sync_task_group_wait(dstg);
1166                 dsl_sync_task_group_destroy(dstg);
1167
1168                 /*
1169                  * We could be racing against 'zfs release' or 'zfs destroy -d'
1170                  * on the origin snap, in which case we can get EBUSY if we
1171                  * needed to destroy the origin snap but were not ready to
1172                  * do so.
1173                  */
1174                 if (dsda.need_prep) {
1175                         ASSERT(err == EBUSY);
1176                         ASSERT(dsl_dir_is_clone(dd));
1177                         ASSERT(dsda.rm_origin == NULL);
1178                 }
1179         } while (dsda.need_prep);
1180
1181         if (dsda.rm_origin != NULL)
1182                 dsl_dataset_disown(dsda.rm_origin, tag);
1183
1184         /* if it is successful, dsl_dir_destroy_sync will close the dd */
1185         if (err)
1186                 dsl_dir_close(dd, FTAG);
1187 out:
1188         dsl_dataset_disown(ds, tag);
1189         return (err);
1190 }
1191
1192 blkptr_t *
1193 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1194 {
1195         return (&ds->ds_phys->ds_bp);
1196 }
1197
1198 void
1199 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1200 {
1201         ASSERT(dmu_tx_is_syncing(tx));
1202         /* If it's the meta-objset, set dp_meta_rootbp */
1203         if (ds == NULL) {
1204                 tx->tx_pool->dp_meta_rootbp = *bp;
1205         } else {
1206                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1207                 ds->ds_phys->ds_bp = *bp;
1208         }
1209 }
1210
1211 spa_t *
1212 dsl_dataset_get_spa(dsl_dataset_t *ds)
1213 {
1214         return (ds->ds_dir->dd_pool->dp_spa);
1215 }
1216
1217 void
1218 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1219 {
1220         dsl_pool_t *dp;
1221
1222         if (ds == NULL) /* this is the meta-objset */
1223                 return;
1224
1225         ASSERT(ds->ds_objset != NULL);
1226
1227         if (ds->ds_phys->ds_next_snap_obj != 0)
1228                 panic("dirtying snapshot!");
1229
1230         dp = ds->ds_dir->dd_pool;
1231
1232         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1233                 /* up the hold count until we can be written out */
1234                 dmu_buf_add_ref(ds->ds_dbuf, ds);
1235         }
1236 }
1237
1238 /*
1239  * The unique space in the head dataset can be calculated by subtracting
1240  * the space used in the most recent snapshot, that is still being used
1241  * in this file system, from the space currently in use.  To figure out
1242  * the space in the most recent snapshot still in use, we need to take
1243  * the total space used in the snapshot and subtract out the space that
1244  * has been freed up since the snapshot was taken.
1245  */
1246 static void
1247 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1248 {
1249         uint64_t mrs_used;
1250         uint64_t dlused, dlcomp, dluncomp;
1251
1252         ASSERT(!dsl_dataset_is_snapshot(ds));
1253
1254         if (ds->ds_phys->ds_prev_snap_obj != 0)
1255                 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1256         else
1257                 mrs_used = 0;
1258
1259         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1260
1261         ASSERT3U(dlused, <=, mrs_used);
1262         ds->ds_phys->ds_unique_bytes =
1263             ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1264
1265         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1266             SPA_VERSION_UNIQUE_ACCURATE)
1267                 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1268 }
1269
1270 struct killarg {
1271         dsl_dataset_t *ds;
1272         dmu_tx_t *tx;
1273 };
1274
1275 /* ARGSUSED */
1276 static int
1277 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1278     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1279 {
1280         struct killarg *ka = arg;
1281         dmu_tx_t *tx = ka->tx;
1282
1283         if (bp == NULL)
1284                 return (0);
1285
1286         if (zb->zb_level == ZB_ZIL_LEVEL) {
1287                 ASSERT(zilog != NULL);
1288                 /*
1289                  * It's a block in the intent log.  It has no
1290                  * accounting, so just free it.
1291                  */
1292                 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1293         } else {
1294                 ASSERT(zilog == NULL);
1295                 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1296                 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1297         }
1298
1299         return (0);
1300 }
1301
1302 /* ARGSUSED */
1303 static int
1304 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1305 {
1306         dsl_dataset_t *ds = arg1;
1307         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1308         uint64_t count;
1309         int err;
1310
1311         /*
1312          * Can't delete a head dataset if there are snapshots of it.
1313          * (Except if the only snapshots are from the branch we cloned
1314          * from.)
1315          */
1316         if (ds->ds_prev != NULL &&
1317             ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1318                 return (EBUSY);
1319
1320         /*
1321          * This is really a dsl_dir thing, but check it here so that
1322          * we'll be less likely to leave this dataset inconsistent &
1323          * nearly destroyed.
1324          */
1325         err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1326         if (err)
1327                 return (err);
1328         if (count != 0)
1329                 return (EEXIST);
1330
1331         return (0);
1332 }
1333
1334 /* ARGSUSED */
1335 static void
1336 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1337 {
1338         dsl_dataset_t *ds = arg1;
1339         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1340
1341         /* Mark it as inconsistent on-disk, in case we crash */
1342         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1343         ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1344
1345         spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1346             "dataset = %llu", ds->ds_object);
1347 }
1348
1349 static int
1350 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1351     dmu_tx_t *tx)
1352 {
1353         dsl_dataset_t *ds = dsda->ds;
1354         dsl_dataset_t *ds_prev = ds->ds_prev;
1355
1356         if (dsl_dataset_might_destroy_origin(ds_prev)) {
1357                 struct dsl_ds_destroyarg ndsda = {0};
1358
1359                 /*
1360                  * If we're not prepared to remove the origin, don't remove
1361                  * the clone either.
1362                  */
1363                 if (dsda->rm_origin == NULL) {
1364                         dsda->need_prep = B_TRUE;
1365                         return (EBUSY);
1366                 }
1367
1368                 ndsda.ds = ds_prev;
1369                 ndsda.is_origin_rm = B_TRUE;
1370                 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1371         }
1372
1373         /*
1374          * If we're not going to remove the origin after all,
1375          * undo the open context setup.
1376          */
1377         if (dsda->rm_origin != NULL) {
1378                 dsl_dataset_disown(dsda->rm_origin, tag);
1379                 dsda->rm_origin = NULL;
1380         }
1381
1382         return (0);
1383 }
1384
1385 /*
1386  * If you add new checks here, you may need to add
1387  * additional checks to the "temporary" case in
1388  * snapshot_check() in dmu_objset.c.
1389  */
1390 /* ARGSUSED */
1391 int
1392 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1393 {
1394         struct dsl_ds_destroyarg *dsda = arg1;
1395         dsl_dataset_t *ds = dsda->ds;
1396
1397         /* we have an owner hold, so noone else can destroy us */
1398         ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1399
1400         /*
1401          * Only allow deferred destroy on pools that support it.
1402          * NOTE: deferred destroy is only supported on snapshots.
1403          */
1404         if (dsda->defer) {
1405                 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1406                     SPA_VERSION_USERREFS)
1407                         return (ENOTSUP);
1408                 ASSERT(dsl_dataset_is_snapshot(ds));
1409                 return (0);
1410         }
1411
1412         /*
1413          * Can't delete a head dataset if there are snapshots of it.
1414          * (Except if the only snapshots are from the branch we cloned
1415          * from.)
1416          */
1417         if (ds->ds_prev != NULL &&
1418             ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1419                 return (EBUSY);
1420
1421         /*
1422          * If we made changes this txg, traverse_dsl_dataset won't find
1423          * them.  Try again.
1424          */
1425         if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1426                 return (EAGAIN);
1427
1428         if (dsl_dataset_is_snapshot(ds)) {
1429                 /*
1430                  * If this snapshot has an elevated user reference count,
1431                  * we can't destroy it yet.
1432                  */
1433                 if (ds->ds_userrefs > 0 && !dsda->releasing)
1434                         return (EBUSY);
1435
1436                 mutex_enter(&ds->ds_lock);
1437                 /*
1438                  * Can't delete a branch point. However, if we're destroying
1439                  * a clone and removing its origin due to it having a user
1440                  * hold count of 0 and having been marked for deferred destroy,
1441                  * it's OK for the origin to have a single clone.
1442                  */
1443                 if (ds->ds_phys->ds_num_children >
1444                     (dsda->is_origin_rm ? 2 : 1)) {
1445                         mutex_exit(&ds->ds_lock);
1446                         return (EEXIST);
1447                 }
1448                 mutex_exit(&ds->ds_lock);
1449         } else if (dsl_dir_is_clone(ds->ds_dir)) {
1450                 return (dsl_dataset_origin_check(dsda, arg2, tx));
1451         }
1452
1453         /* XXX we should do some i/o error checking... */
1454         return (0);
1455 }
1456
1457 struct refsarg {
1458         kmutex_t lock;
1459         boolean_t gone;
1460         kcondvar_t cv;
1461 };
1462
1463 /* ARGSUSED */
1464 static void
1465 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1466 {
1467         struct refsarg *arg = argv;
1468
1469         mutex_enter(&arg->lock);
1470         arg->gone = TRUE;
1471         cv_signal(&arg->cv);
1472         mutex_exit(&arg->lock);
1473 }
1474
1475 static void
1476 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1477 {
1478         struct refsarg arg;
1479
1480         bzero(&arg, sizeof(arg));
1481         mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1482         cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1483         arg.gone = FALSE;
1484         (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1485             dsl_dataset_refs_gone);
1486         dmu_buf_rele(ds->ds_dbuf, tag);
1487         mutex_enter(&arg.lock);
1488         while (!arg.gone)
1489                 cv_wait(&arg.cv, &arg.lock);
1490         ASSERT(arg.gone);
1491         mutex_exit(&arg.lock);
1492         ds->ds_dbuf = NULL;
1493         ds->ds_phys = NULL;
1494         mutex_destroy(&arg.lock);
1495         cv_destroy(&arg.cv);
1496 }
1497
1498 static void
1499 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1500 {
1501         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1502         uint64_t count;
1503         int err;
1504
1505         ASSERT(ds->ds_phys->ds_num_children >= 2);
1506         err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1507         /*
1508          * The err should not be ENOENT, but a bug in a previous version
1509          * of the code could cause upgrade_clones_cb() to not set
1510          * ds_next_snap_obj when it should, leading to a missing entry.
1511          * If we knew that the pool was created after
1512          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1513          * ENOENT.  However, at least we can check that we don't have
1514          * too many entries in the next_clones_obj even after failing to
1515          * remove this one.
1516          */
1517         if (err != ENOENT) {
1518                 VERIFY3U(err, ==, 0);
1519         }
1520         ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1521             &count));
1522         ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1523 }
1524
1525 static void
1526 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1527 {
1528         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1529         zap_cursor_t zc;
1530         zap_attribute_t za;
1531
1532         /*
1533          * If it is the old version, dd_clones doesn't exist so we can't
1534          * find the clones, but deadlist_remove_key() is a no-op so it
1535          * doesn't matter.
1536          */
1537         if (ds->ds_dir->dd_phys->dd_clones == 0)
1538                 return;
1539
1540         for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1541             zap_cursor_retrieve(&zc, &za) == 0;
1542             zap_cursor_advance(&zc)) {
1543                 dsl_dataset_t *clone;
1544
1545                 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1546                     za.za_first_integer, FTAG, &clone));
1547                 if (clone->ds_dir->dd_origin_txg > mintxg) {
1548                         dsl_deadlist_remove_key(&clone->ds_deadlist,
1549                             mintxg, tx);
1550                         dsl_dataset_remove_clones_key(clone, mintxg, tx);
1551                 }
1552                 dsl_dataset_rele(clone, FTAG);
1553         }
1554         zap_cursor_fini(&zc);
1555 }
1556
1557 struct process_old_arg {
1558         dsl_dataset_t *ds;
1559         dsl_dataset_t *ds_prev;
1560         boolean_t after_branch_point;
1561         zio_t *pio;
1562         uint64_t used, comp, uncomp;
1563 };
1564
1565 static int
1566 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1567 {
1568         struct process_old_arg *poa = arg;
1569         dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1570
1571         if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1572                 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1573                 if (poa->ds_prev && !poa->after_branch_point &&
1574                     bp->blk_birth >
1575                     poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1576                         poa->ds_prev->ds_phys->ds_unique_bytes +=
1577                             bp_get_dsize_sync(dp->dp_spa, bp);
1578                 }
1579         } else {
1580                 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1581                 poa->comp += BP_GET_PSIZE(bp);
1582                 poa->uncomp += BP_GET_UCSIZE(bp);
1583                 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1584         }
1585         return (0);
1586 }
1587
1588 static void
1589 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1590     dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1591 {
1592         struct process_old_arg poa = { 0 };
1593         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1594         objset_t *mos = dp->dp_meta_objset;
1595
1596         ASSERT(ds->ds_deadlist.dl_oldfmt);
1597         ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1598
1599         poa.ds = ds;
1600         poa.ds_prev = ds_prev;
1601         poa.after_branch_point = after_branch_point;
1602         poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1603         VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1604             process_old_cb, &poa, tx));
1605         VERIFY3U(zio_wait(poa.pio), ==, 0);
1606         ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1607
1608         /* change snapused */
1609         dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1610             -poa.used, -poa.comp, -poa.uncomp, tx);
1611
1612         /* swap next's deadlist to our deadlist */
1613         dsl_deadlist_close(&ds->ds_deadlist);
1614         dsl_deadlist_close(&ds_next->ds_deadlist);
1615         SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1616             ds->ds_phys->ds_deadlist_obj);
1617         dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1618         dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1619             ds_next->ds_phys->ds_deadlist_obj);
1620 }
1621
1622 void
1623 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1624 {
1625         struct dsl_ds_destroyarg *dsda = arg1;
1626         dsl_dataset_t *ds = dsda->ds;
1627         int err;
1628         int after_branch_point = FALSE;
1629         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1630         objset_t *mos = dp->dp_meta_objset;
1631         dsl_dataset_t *ds_prev = NULL;
1632         boolean_t wont_destroy;
1633         uint64_t obj;
1634
1635         wont_destroy = (dsda->defer &&
1636             (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1637
1638         ASSERT(ds->ds_owner || wont_destroy);
1639         ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1640         ASSERT(ds->ds_prev == NULL ||
1641             ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1642         ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1643
1644         if (wont_destroy) {
1645                 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1646                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1647                 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1648                 return;
1649         }
1650
1651         /* signal any waiters that this dataset is going away */
1652         mutex_enter(&ds->ds_lock);
1653         ds->ds_owner = dsl_reaper;
1654         cv_broadcast(&ds->ds_exclusive_cv);
1655         mutex_exit(&ds->ds_lock);
1656
1657         /* Remove our reservation */
1658         if (ds->ds_reserved != 0) {
1659                 dsl_prop_setarg_t psa;
1660                 uint64_t value = 0;
1661
1662                 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1663                     (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1664                     &value);
1665                 psa.psa_effective_value = 0;    /* predict default value */
1666
1667                 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1668                 ASSERT3U(ds->ds_reserved, ==, 0);
1669         }
1670
1671         ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1672
1673         dsl_scan_ds_destroyed(ds, tx);
1674
1675         obj = ds->ds_object;
1676
1677         if (ds->ds_phys->ds_prev_snap_obj != 0) {
1678                 if (ds->ds_prev) {
1679                         ds_prev = ds->ds_prev;
1680                 } else {
1681                         VERIFY(0 == dsl_dataset_hold_obj(dp,
1682                             ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1683                 }
1684                 after_branch_point =
1685                     (ds_prev->ds_phys->ds_next_snap_obj != obj);
1686
1687                 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1688                 if (after_branch_point &&
1689                     ds_prev->ds_phys->ds_next_clones_obj != 0) {
1690                         remove_from_next_clones(ds_prev, obj, tx);
1691                         if (ds->ds_phys->ds_next_snap_obj != 0) {
1692                                 VERIFY(0 == zap_add_int(mos,
1693                                     ds_prev->ds_phys->ds_next_clones_obj,
1694                                     ds->ds_phys->ds_next_snap_obj, tx));
1695                         }
1696                 }
1697                 if (after_branch_point &&
1698                     ds->ds_phys->ds_next_snap_obj == 0) {
1699                         /* This clone is toast. */
1700                         ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1701                         ds_prev->ds_phys->ds_num_children--;
1702
1703                         /*
1704                          * If the clone's origin has no other clones, no
1705                          * user holds, and has been marked for deferred
1706                          * deletion, then we should have done the necessary
1707                          * destroy setup for it.
1708                          */
1709                         if (ds_prev->ds_phys->ds_num_children == 1 &&
1710                             ds_prev->ds_userrefs == 0 &&
1711                             DS_IS_DEFER_DESTROY(ds_prev)) {
1712                                 ASSERT3P(dsda->rm_origin, !=, NULL);
1713                         } else {
1714                                 ASSERT3P(dsda->rm_origin, ==, NULL);
1715                         }
1716                 } else if (!after_branch_point) {
1717                         ds_prev->ds_phys->ds_next_snap_obj =
1718                             ds->ds_phys->ds_next_snap_obj;
1719                 }
1720         }
1721
1722         if (dsl_dataset_is_snapshot(ds)) {
1723                 dsl_dataset_t *ds_next;
1724                 uint64_t old_unique;
1725                 uint64_t used = 0, comp = 0, uncomp = 0;
1726
1727                 VERIFY(0 == dsl_dataset_hold_obj(dp,
1728                     ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1729                 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1730
1731                 old_unique = ds_next->ds_phys->ds_unique_bytes;
1732
1733                 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1734                 ds_next->ds_phys->ds_prev_snap_obj =
1735                     ds->ds_phys->ds_prev_snap_obj;
1736                 ds_next->ds_phys->ds_prev_snap_txg =
1737                     ds->ds_phys->ds_prev_snap_txg;
1738                 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1739                     ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1740
1741
1742                 if (ds_next->ds_deadlist.dl_oldfmt) {
1743                         process_old_deadlist(ds, ds_prev, ds_next,
1744                             after_branch_point, tx);
1745                 } else {
1746                         /* Adjust prev's unique space. */
1747                         if (ds_prev && !after_branch_point) {
1748                                 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1749                                     ds_prev->ds_phys->ds_prev_snap_txg,
1750                                     ds->ds_phys->ds_prev_snap_txg,
1751                                     &used, &comp, &uncomp);
1752                                 ds_prev->ds_phys->ds_unique_bytes += used;
1753                         }
1754
1755                         /* Adjust snapused. */
1756                         dsl_deadlist_space_range(&ds_next->ds_deadlist,
1757                             ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1758                             &used, &comp, &uncomp);
1759                         dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1760                             -used, -comp, -uncomp, tx);
1761
1762                         /* Move blocks to be freed to pool's free list. */
1763                         dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1764                             &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1765                             tx);
1766                         dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1767                             DD_USED_HEAD, used, comp, uncomp, tx);
1768                         dsl_dir_dirty(tx->tx_pool->dp_free_dir, tx);
1769
1770                         /* Merge our deadlist into next's and free it. */
1771                         dsl_deadlist_merge(&ds_next->ds_deadlist,
1772                             ds->ds_phys->ds_deadlist_obj, tx);
1773                 }
1774                 dsl_deadlist_close(&ds->ds_deadlist);
1775                 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1776
1777                 /* Collapse range in clone heads */
1778                 dsl_dataset_remove_clones_key(ds,
1779                     ds->ds_phys->ds_creation_txg, tx);
1780
1781                 if (dsl_dataset_is_snapshot(ds_next)) {
1782                         dsl_dataset_t *ds_nextnext;
1783
1784                         /*
1785                          * Update next's unique to include blocks which
1786                          * were previously shared by only this snapshot
1787                          * and it.  Those blocks will be born after the
1788                          * prev snap and before this snap, and will have
1789                          * died after the next snap and before the one
1790                          * after that (ie. be on the snap after next's
1791                          * deadlist).
1792                          */
1793                         VERIFY(0 == dsl_dataset_hold_obj(dp,
1794                             ds_next->ds_phys->ds_next_snap_obj,
1795                             FTAG, &ds_nextnext));
1796                         dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1797                             ds->ds_phys->ds_prev_snap_txg,
1798                             ds->ds_phys->ds_creation_txg,
1799                             &used, &comp, &uncomp);
1800                         ds_next->ds_phys->ds_unique_bytes += used;
1801                         dsl_dataset_rele(ds_nextnext, FTAG);
1802                         ASSERT3P(ds_next->ds_prev, ==, NULL);
1803
1804                         /* Collapse range in this head. */
1805                         dsl_dataset_t *hds;
1806                         VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1807                             ds->ds_dir->dd_phys->dd_head_dataset_obj,
1808                             FTAG, &hds));
1809                         dsl_deadlist_remove_key(&hds->ds_deadlist,
1810                             ds->ds_phys->ds_creation_txg, tx);
1811                         dsl_dataset_rele(hds, FTAG);
1812
1813                 } else {
1814                         ASSERT3P(ds_next->ds_prev, ==, ds);
1815                         dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1816                         ds_next->ds_prev = NULL;
1817                         if (ds_prev) {
1818                                 VERIFY(0 == dsl_dataset_get_ref(dp,
1819                                     ds->ds_phys->ds_prev_snap_obj,
1820                                     ds_next, &ds_next->ds_prev));
1821                         }
1822
1823                         dsl_dataset_recalc_head_uniq(ds_next);
1824
1825                         /*
1826                          * Reduce the amount of our unconsmed refreservation
1827                          * being charged to our parent by the amount of
1828                          * new unique data we have gained.
1829                          */
1830                         if (old_unique < ds_next->ds_reserved) {
1831                                 int64_t mrsdelta;
1832                                 uint64_t new_unique =
1833                                     ds_next->ds_phys->ds_unique_bytes;
1834
1835                                 ASSERT(old_unique <= new_unique);
1836                                 mrsdelta = MIN(new_unique - old_unique,
1837                                     ds_next->ds_reserved - old_unique);
1838                                 dsl_dir_diduse_space(ds->ds_dir,
1839                                     DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1840                         }
1841                 }
1842                 dsl_dataset_rele(ds_next, FTAG);
1843         } else {
1844                 /*
1845                  * There's no next snapshot, so this is a head dataset.
1846                  * Destroy the deadlist.  Unless it's a clone, the
1847                  * deadlist should be empty.  (If it's a clone, it's
1848                  * safe to ignore the deadlist contents.)
1849                  */
1850                 struct killarg ka;
1851
1852                 dsl_deadlist_close(&ds->ds_deadlist);
1853                 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1854                 ds->ds_phys->ds_deadlist_obj = 0;
1855
1856                 /*
1857                  * Free everything that we point to (that's born after
1858                  * the previous snapshot, if we are a clone)
1859                  *
1860                  * NB: this should be very quick, because we already
1861                  * freed all the objects in open context.
1862                  */
1863                 ka.ds = ds;
1864                 ka.tx = tx;
1865                 err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1866                     TRAVERSE_POST, kill_blkptr, &ka);
1867                 ASSERT3U(err, ==, 0);
1868                 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1869                     ds->ds_phys->ds_unique_bytes == 0);
1870
1871                 if (ds->ds_prev != NULL) {
1872                         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1873                                 VERIFY3U(0, ==, zap_remove_int(mos,
1874                                     ds->ds_prev->ds_dir->dd_phys->dd_clones,
1875                                     ds->ds_object, tx));
1876                         }
1877                         dsl_dataset_rele(ds->ds_prev, ds);
1878                         ds->ds_prev = ds_prev = NULL;
1879                 }
1880         }
1881
1882         /*
1883          * This must be done after the dsl_traverse(), because it will
1884          * re-open the objset.
1885          */
1886         if (ds->ds_objset) {
1887                 dmu_objset_evict(ds->ds_objset);
1888                 ds->ds_objset = NULL;
1889         }
1890
1891         if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1892                 /* Erase the link in the dir */
1893                 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1894                 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1895                 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1896                 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1897                 ASSERT(err == 0);
1898         } else {
1899                 /* remove from snapshot namespace */
1900                 dsl_dataset_t *ds_head;
1901                 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1902                 VERIFY(0 == dsl_dataset_hold_obj(dp,
1903                     ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1904                 VERIFY(0 == dsl_dataset_get_snapname(ds));
1905 #ifdef ZFS_DEBUG
1906                 {
1907                         uint64_t val;
1908
1909                         err = dsl_dataset_snap_lookup(ds_head,
1910                             ds->ds_snapname, &val);
1911                         ASSERT3U(err, ==, 0);
1912                         ASSERT3U(val, ==, obj);
1913                 }
1914 #endif
1915                 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1916                 ASSERT(err == 0);
1917                 dsl_dataset_rele(ds_head, FTAG);
1918         }
1919
1920         if (ds_prev && ds->ds_prev != ds_prev)
1921                 dsl_dataset_rele(ds_prev, FTAG);
1922
1923         spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1924         spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
1925             "dataset = %llu", ds->ds_object);
1926
1927         if (ds->ds_phys->ds_next_clones_obj != 0) {
1928                 uint64_t count;
1929                 ASSERT(0 == zap_count(mos,
1930                     ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1931                 VERIFY(0 == dmu_object_free(mos,
1932                     ds->ds_phys->ds_next_clones_obj, tx));
1933         }
1934         if (ds->ds_phys->ds_props_obj != 0)
1935                 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1936         if (ds->ds_phys->ds_userrefs_obj != 0)
1937                 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1938         dsl_dir_close(ds->ds_dir, ds);
1939         ds->ds_dir = NULL;
1940         dsl_dataset_drain_refs(ds, tag);
1941         VERIFY(0 == dmu_object_free(mos, obj, tx));
1942
1943         if (dsda->rm_origin) {
1944                 /*
1945                  * Remove the origin of the clone we just destroyed.
1946                  */
1947                 struct dsl_ds_destroyarg ndsda = {0};
1948
1949                 ndsda.ds = dsda->rm_origin;
1950                 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1951         }
1952 }
1953
1954 static int
1955 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1956 {
1957         uint64_t asize;
1958
1959         if (!dmu_tx_is_syncing(tx))
1960                 return (0);
1961
1962         /*
1963          * If there's an fs-only reservation, any blocks that might become
1964          * owned by the snapshot dataset must be accommodated by space
1965          * outside of the reservation.
1966          */
1967         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1968         asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1969         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1970                 return (ENOSPC);
1971
1972         /*
1973          * Propogate any reserved space for this snapshot to other
1974          * snapshot checks in this sync group.
1975          */
1976         if (asize > 0)
1977                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1978
1979         return (0);
1980 }
1981
1982 int
1983 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1984 {
1985         dsl_dataset_t *ds = arg1;
1986         const char *snapname = arg2;
1987         int err;
1988         uint64_t value;
1989
1990         /*
1991          * We don't allow multiple snapshots of the same txg.  If there
1992          * is already one, try again.
1993          */
1994         if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1995                 return (EAGAIN);
1996
1997         /*
1998          * Check for conflicting name snapshot name.
1999          */
2000         err = dsl_dataset_snap_lookup(ds, snapname, &value);
2001         if (err == 0)
2002                 return (EEXIST);
2003         if (err != ENOENT)
2004                 return (err);
2005
2006         /*
2007          * Check that the dataset's name is not too long.  Name consists
2008          * of the dataset's length + 1 for the @-sign + snapshot name's length
2009          */
2010         if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2011                 return (ENAMETOOLONG);
2012
2013         err = dsl_dataset_snapshot_reserve_space(ds, tx);
2014         if (err)
2015                 return (err);
2016
2017         ds->ds_trysnap_txg = tx->tx_txg;
2018         return (0);
2019 }
2020
2021 void
2022 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2023 {
2024         dsl_dataset_t *ds = arg1;
2025         const char *snapname = arg2;
2026         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2027         dmu_buf_t *dbuf;
2028         dsl_dataset_phys_t *dsphys;
2029         uint64_t dsobj, crtxg;
2030         objset_t *mos = dp->dp_meta_objset;
2031         int err;
2032
2033         ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2034
2035         /*
2036          * The origin's ds_creation_txg has to be < TXG_INITIAL
2037          */
2038         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2039                 crtxg = 1;
2040         else
2041                 crtxg = tx->tx_txg;
2042
2043         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2044             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2045         VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2046         dmu_buf_will_dirty(dbuf, tx);
2047         dsphys = dbuf->db_data;
2048         bzero(dsphys, sizeof (dsl_dataset_phys_t));
2049         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2050         dsphys->ds_fsid_guid = unique_create();
2051         (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2052             sizeof (dsphys->ds_guid));
2053         dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2054         dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2055         dsphys->ds_next_snap_obj = ds->ds_object;
2056         dsphys->ds_num_children = 1;
2057         dsphys->ds_creation_time = gethrestime_sec();
2058         dsphys->ds_creation_txg = crtxg;
2059         dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2060         dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
2061         dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2062         dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2063         dsphys->ds_flags = ds->ds_phys->ds_flags;
2064         dsphys->ds_bp = ds->ds_phys->ds_bp;
2065         dmu_buf_rele(dbuf, FTAG);
2066
2067         ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2068         if (ds->ds_prev) {
2069                 uint64_t next_clones_obj =
2070                     ds->ds_prev->ds_phys->ds_next_clones_obj;
2071                 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2072                     ds->ds_object ||
2073                     ds->ds_prev->ds_phys->ds_num_children > 1);
2074                 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2075                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2076                         ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2077                             ds->ds_prev->ds_phys->ds_creation_txg);
2078                         ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2079                 } else if (next_clones_obj != 0) {
2080                         remove_from_next_clones(ds->ds_prev,
2081                             dsphys->ds_next_snap_obj, tx);
2082                         VERIFY3U(0, ==, zap_add_int(mos,
2083                             next_clones_obj, dsobj, tx));
2084                 }
2085         }
2086
2087         /*
2088          * If we have a reference-reservation on this dataset, we will
2089          * need to increase the amount of refreservation being charged
2090          * since our unique space is going to zero.
2091          */
2092         if (ds->ds_reserved) {
2093                 int64_t delta;
2094                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2095                 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2096                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2097                     delta, 0, 0, tx);
2098         }
2099
2100         dmu_buf_will_dirty(ds->ds_dbuf, tx);
2101         zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2102             ds->ds_dir->dd_myname, snapname, dsobj,
2103             ds->ds_phys->ds_prev_snap_txg);
2104         ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2105             UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2106         dsl_deadlist_close(&ds->ds_deadlist);
2107         dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2108         dsl_deadlist_add_key(&ds->ds_deadlist,
2109             ds->ds_phys->ds_prev_snap_txg, tx);
2110
2111         ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2112         ds->ds_phys->ds_prev_snap_obj = dsobj;
2113         ds->ds_phys->ds_prev_snap_txg = crtxg;
2114         ds->ds_phys->ds_unique_bytes = 0;
2115         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2116                 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2117
2118         err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2119             snapname, 8, 1, &dsobj, tx);
2120         ASSERT(err == 0);
2121
2122         if (ds->ds_prev)
2123                 dsl_dataset_drop_ref(ds->ds_prev, ds);
2124         VERIFY(0 == dsl_dataset_get_ref(dp,
2125             ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2126
2127         dsl_scan_ds_snapshotted(ds, tx);
2128
2129         dsl_dir_snap_cmtime_update(ds->ds_dir);
2130
2131         spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
2132             "dataset = %llu", dsobj);
2133 }
2134
2135 void
2136 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2137 {
2138         ASSERT(dmu_tx_is_syncing(tx));
2139         ASSERT(ds->ds_objset != NULL);
2140         ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2141
2142         /*
2143          * in case we had to change ds_fsid_guid when we opened it,
2144          * sync it out now.
2145          */
2146         dmu_buf_will_dirty(ds->ds_dbuf, tx);
2147         ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2148
2149         dsl_dir_dirty(ds->ds_dir, tx);
2150         dmu_objset_sync(ds->ds_objset, zio, tx);
2151 }
2152
2153 void
2154 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2155 {
2156         uint64_t refd, avail, uobjs, aobjs, ratio;
2157
2158         dsl_dir_stats(ds->ds_dir, nv);
2159
2160         dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2161         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2162         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2163
2164         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2165             ds->ds_phys->ds_creation_time);
2166         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2167             ds->ds_phys->ds_creation_txg);
2168         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2169             ds->ds_quota);
2170         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2171             ds->ds_reserved);
2172         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2173             ds->ds_phys->ds_guid);
2174         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2175             ds->ds_phys->ds_unique_bytes);
2176         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2177             ds->ds_object);
2178         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2179             ds->ds_userrefs);
2180         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2181             DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2182
2183         ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2184             (ds->ds_phys->ds_uncompressed_bytes * 100 /
2185             ds->ds_phys->ds_compressed_bytes);
2186         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2187
2188         if (ds->ds_phys->ds_next_snap_obj) {
2189                 /*
2190                  * This is a snapshot; override the dd's space used with
2191                  * our unique space and compression ratio.
2192                  */
2193                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2194                     ds->ds_phys->ds_unique_bytes);
2195                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2196         }
2197 }
2198
2199 void
2200 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2201 {
2202         stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2203         stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2204         stat->dds_guid = ds->ds_phys->ds_guid;
2205         if (ds->ds_phys->ds_next_snap_obj) {
2206                 stat->dds_is_snapshot = B_TRUE;
2207                 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2208         } else {
2209                 stat->dds_is_snapshot = B_FALSE;
2210                 stat->dds_num_clones = 0;
2211         }
2212
2213         /* clone origin is really a dsl_dir thing... */
2214         rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2215         if (dsl_dir_is_clone(ds->ds_dir)) {
2216                 dsl_dataset_t *ods;
2217
2218                 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2219                     ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2220                 dsl_dataset_name(ods, stat->dds_origin);
2221                 dsl_dataset_drop_ref(ods, FTAG);
2222         } else {
2223                 stat->dds_origin[0] = '\0';
2224         }
2225         rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2226 }
2227
2228 uint64_t
2229 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2230 {
2231         return (ds->ds_fsid_guid);
2232 }
2233
2234 void
2235 dsl_dataset_space(dsl_dataset_t *ds,
2236     uint64_t *refdbytesp, uint64_t *availbytesp,
2237     uint64_t *usedobjsp, uint64_t *availobjsp)
2238 {
2239         *refdbytesp = ds->ds_phys->ds_used_bytes;
2240         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2241         if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2242                 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2243         if (ds->ds_quota != 0) {
2244                 /*
2245                  * Adjust available bytes according to refquota
2246                  */
2247                 if (*refdbytesp < ds->ds_quota)
2248                         *availbytesp = MIN(*availbytesp,
2249                             ds->ds_quota - *refdbytesp);
2250                 else
2251                         *availbytesp = 0;
2252         }
2253         *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2254         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2255 }
2256
2257 boolean_t
2258 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2259 {
2260         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2261
2262         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2263             dsl_pool_sync_context(dp));
2264         if (ds->ds_prev == NULL)
2265                 return (B_FALSE);
2266         if (ds->ds_phys->ds_bp.blk_birth >
2267             ds->ds_prev->ds_phys->ds_creation_txg) {
2268                 objset_t *os, *os_prev;
2269                 /*
2270                  * It may be that only the ZIL differs, because it was
2271                  * reset in the head.  Don't count that as being
2272                  * modified.
2273                  */
2274                 if (dmu_objset_from_ds(ds, &os) != 0)
2275                         return (B_TRUE);
2276                 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2277                         return (B_TRUE);
2278                 return (bcmp(&os->os_phys->os_meta_dnode,
2279                     &os_prev->os_phys->os_meta_dnode,
2280                     sizeof (os->os_phys->os_meta_dnode)) != 0);
2281         }
2282         return (B_FALSE);
2283 }
2284
2285 /* ARGSUSED */
2286 static int
2287 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2288 {
2289         dsl_dataset_t *ds = arg1;
2290         char *newsnapname = arg2;
2291         dsl_dir_t *dd = ds->ds_dir;
2292         dsl_dataset_t *hds;
2293         uint64_t val;
2294         int err;
2295
2296         err = dsl_dataset_hold_obj(dd->dd_pool,
2297             dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2298         if (err)
2299                 return (err);
2300
2301         /* new name better not be in use */
2302         err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2303         dsl_dataset_rele(hds, FTAG);
2304
2305         if (err == 0)
2306                 err = EEXIST;
2307         else if (err == ENOENT)
2308                 err = 0;
2309
2310         /* dataset name + 1 for the "@" + the new snapshot name must fit */
2311         if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2312                 err = ENAMETOOLONG;
2313
2314         return (err);
2315 }
2316
2317 static void
2318 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2319 {
2320         char oldname[MAXPATHLEN], newname[MAXPATHLEN];
2321         dsl_dataset_t *ds = arg1;
2322         const char *newsnapname = arg2;
2323         dsl_dir_t *dd = ds->ds_dir;
2324         objset_t *mos = dd->dd_pool->dp_meta_objset;
2325         dsl_dataset_t *hds;
2326         int err;
2327
2328         ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2329
2330         VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2331             dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2332
2333         VERIFY(0 == dsl_dataset_get_snapname(ds));
2334         err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2335         ASSERT3U(err, ==, 0);
2336         dsl_dataset_name(ds, oldname);
2337         mutex_enter(&ds->ds_lock);
2338         (void) strcpy(ds->ds_snapname, newsnapname);
2339         mutex_exit(&ds->ds_lock);
2340         err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2341             ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2342         ASSERT3U(err, ==, 0);
2343         dsl_dataset_name(ds, newname);
2344 #ifdef _KERNEL
2345         zvol_rename_minors(oldname, newname);
2346 #endif
2347
2348         spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2349             "dataset = %llu", ds->ds_object);
2350         dsl_dataset_rele(hds, FTAG);
2351 }
2352
2353 struct renamesnaparg {
2354         dsl_sync_task_group_t *dstg;
2355         char failed[MAXPATHLEN];
2356         char *oldsnap;
2357         char *newsnap;
2358 };
2359
2360 static int
2361 dsl_snapshot_rename_one(const char *name, void *arg)
2362 {
2363         struct renamesnaparg *ra = arg;
2364         dsl_dataset_t *ds = NULL;
2365         char *snapname;
2366         int err;
2367
2368         snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2369         (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2370
2371         /*
2372          * For recursive snapshot renames the parent won't be changing
2373          * so we just pass name for both the to/from argument.
2374          */
2375         err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2376         if (err != 0) {
2377                 strfree(snapname);
2378                 return (err == ENOENT ? 0 : err);
2379         }
2380
2381 #ifdef _KERNEL
2382         /*
2383          * For all filesystems undergoing rename, we'll need to unmount it.
2384          */
2385         (void) zfs_unmount_snap(snapname, NULL);
2386 #endif
2387         err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2388         strfree(snapname);
2389         if (err != 0)
2390                 return (err == ENOENT ? 0 : err);
2391
2392         dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2393             dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2394
2395         return (0);
2396 }
2397
2398 static int
2399 dsl_recursive_rename(char *oldname, const char *newname)
2400 {
2401         int err;
2402         struct renamesnaparg *ra;
2403         dsl_sync_task_t *dst;
2404         spa_t *spa;
2405         char *cp, *fsname = spa_strdup(oldname);
2406         int len = strlen(oldname) + 1;
2407
2408         /* truncate the snapshot name to get the fsname */
2409         cp = strchr(fsname, '@');
2410         *cp = '\0';
2411
2412         err = spa_open(fsname, &spa, FTAG);
2413         if (err) {
2414                 kmem_free(fsname, len);
2415                 return (err);
2416         }
2417         ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2418         ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2419
2420         ra->oldsnap = strchr(oldname, '@') + 1;
2421         ra->newsnap = strchr(newname, '@') + 1;
2422         *ra->failed = '\0';
2423
2424         err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2425             DS_FIND_CHILDREN);
2426         kmem_free(fsname, len);
2427
2428         if (err == 0) {
2429                 err = dsl_sync_task_group_wait(ra->dstg);
2430         }
2431
2432         for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2433             dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2434                 dsl_dataset_t *ds = dst->dst_arg1;
2435                 if (dst->dst_err) {
2436                         dsl_dir_name(ds->ds_dir, ra->failed);
2437                         (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2438                         (void) strlcat(ra->failed, ra->newsnap,
2439                             sizeof (ra->failed));
2440                 }
2441                 dsl_dataset_rele(ds, ra->dstg);
2442         }
2443
2444         if (err)
2445                 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2446
2447         dsl_sync_task_group_destroy(ra->dstg);
2448         kmem_free(ra, sizeof (struct renamesnaparg));
2449         spa_close(spa, FTAG);
2450         return (err);
2451 }
2452
2453 static int
2454 dsl_valid_rename(const char *oldname, void *arg)
2455 {
2456         int delta = *(int *)arg;
2457
2458         if (strlen(oldname) + delta >= MAXNAMELEN)
2459                 return (ENAMETOOLONG);
2460
2461         return (0);
2462 }
2463
2464 #pragma weak dmu_objset_rename = dsl_dataset_rename
2465 int
2466 dsl_dataset_rename(char *oldname, const char *newname, int flags)
2467 {
2468         dsl_dir_t *dd;
2469         dsl_dataset_t *ds;
2470         const char *tail;
2471         int err;
2472
2473         err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2474         if (err)
2475                 return (err);
2476
2477         if (tail == NULL) {
2478                 int delta = strlen(newname) - strlen(oldname);
2479
2480                 /* if we're growing, validate child name lengths */
2481                 if (delta > 0)
2482                         err = dmu_objset_find(oldname, dsl_valid_rename,
2483                             &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2484
2485                 if (err == 0)
2486                         err = dsl_dir_rename(dd, newname, flags);
2487                 dsl_dir_close(dd, FTAG);
2488                 return (err);
2489         }
2490
2491         if (tail[0] != '@') {
2492                 /* the name ended in a nonexistent component */
2493                 dsl_dir_close(dd, FTAG);
2494                 return (ENOENT);
2495         }
2496
2497         dsl_dir_close(dd, FTAG);
2498
2499         /* new name must be snapshot in same filesystem */
2500         tail = strchr(newname, '@');
2501         if (tail == NULL)
2502                 return (EINVAL);
2503         tail++;
2504         if (strncmp(oldname, newname, tail - newname) != 0)
2505                 return (EXDEV);
2506
2507         if (flags & ZFS_RENAME_RECURSIVE) {
2508                 err = dsl_recursive_rename(oldname, newname);
2509         } else {
2510                 err = dsl_dataset_hold(oldname, FTAG, &ds);
2511                 if (err)
2512                         return (err);
2513
2514                 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2515                     dsl_dataset_snapshot_rename_check,
2516                     dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2517
2518                 dsl_dataset_rele(ds, FTAG);
2519         }
2520
2521         return (err);
2522 }
2523
2524 struct promotenode {
2525         list_node_t link;
2526         dsl_dataset_t *ds;
2527 };
2528
2529 struct promotearg {
2530         list_t shared_snaps, origin_snaps, clone_snaps;
2531         dsl_dataset_t *origin_origin;
2532         uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2533         char *err_ds;
2534 };
2535
2536 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2537 static boolean_t snaplist_unstable(list_t *l);
2538
2539 static int
2540 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2541 {
2542         dsl_dataset_t *hds = arg1;
2543         struct promotearg *pa = arg2;
2544         struct promotenode *snap = list_head(&pa->shared_snaps);
2545         dsl_dataset_t *origin_ds = snap->ds;
2546         int err;
2547         uint64_t unused;
2548
2549         /* Check that it is a real clone */
2550         if (!dsl_dir_is_clone(hds->ds_dir))
2551                 return (EINVAL);
2552
2553         /* Since this is so expensive, don't do the preliminary check */
2554         if (!dmu_tx_is_syncing(tx))
2555                 return (0);
2556
2557         if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2558                 return (EXDEV);
2559
2560         /* compute origin's new unique space */
2561         snap = list_tail(&pa->clone_snaps);
2562         ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2563         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2564             origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2565             &pa->unique, &unused, &unused);
2566
2567         /*
2568          * Walk the snapshots that we are moving
2569          *
2570          * Compute space to transfer.  Consider the incremental changes
2571          * to used for each snapshot:
2572          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2573          * So each snapshot gave birth to:
2574          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2575          * So a sequence would look like:
2576          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2577          * Which simplifies to:
2578          * uN + kN + kN-1 + ... + k1 + k0
2579          * Note however, if we stop before we reach the ORIGIN we get:
2580          * uN + kN + kN-1 + ... + kM - uM-1
2581          */
2582         pa->used = origin_ds->ds_phys->ds_used_bytes;
2583         pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2584         pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2585         for (snap = list_head(&pa->shared_snaps); snap;
2586             snap = list_next(&pa->shared_snaps, snap)) {
2587                 uint64_t val, dlused, dlcomp, dluncomp;
2588                 dsl_dataset_t *ds = snap->ds;
2589
2590                 /* Check that the snapshot name does not conflict */
2591                 VERIFY(0 == dsl_dataset_get_snapname(ds));
2592                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2593                 if (err == 0) {
2594                         err = EEXIST;
2595                         goto out;
2596                 }
2597                 if (err != ENOENT)
2598                         goto out;
2599
2600                 /* The very first snapshot does not have a deadlist */
2601                 if (ds->ds_phys->ds_prev_snap_obj == 0)
2602                         continue;
2603
2604                 dsl_deadlist_space(&ds->ds_deadlist,
2605                     &dlused, &dlcomp, &dluncomp);
2606                 pa->used += dlused;
2607                 pa->comp += dlcomp;
2608                 pa->uncomp += dluncomp;
2609         }
2610
2611         /*
2612          * If we are a clone of a clone then we never reached ORIGIN,
2613          * so we need to subtract out the clone origin's used space.
2614          */
2615         if (pa->origin_origin) {
2616                 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2617                 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2618                 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2619         }
2620
2621         /* Check that there is enough space here */
2622         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2623             pa->used);
2624         if (err)
2625                 return (err);
2626
2627         /*
2628          * Compute the amounts of space that will be used by snapshots
2629          * after the promotion (for both origin and clone).  For each,
2630          * it is the amount of space that will be on all of their
2631          * deadlists (that was not born before their new origin).
2632          */
2633         if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2634                 uint64_t space;
2635
2636                 /*
2637                  * Note, typically this will not be a clone of a clone,
2638                  * so dd_origin_txg will be < TXG_INITIAL, so
2639                  * these snaplist_space() -> dsl_deadlist_space_range()
2640                  * calls will be fast because they do not have to
2641                  * iterate over all bps.
2642                  */
2643                 snap = list_head(&pa->origin_snaps);
2644                 err = snaplist_space(&pa->shared_snaps,
2645                     snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2646                 if (err)
2647                         return (err);
2648
2649                 err = snaplist_space(&pa->clone_snaps,
2650                     snap->ds->ds_dir->dd_origin_txg, &space);
2651                 if (err)
2652                         return (err);
2653                 pa->cloneusedsnap += space;
2654         }
2655         if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2656                 err = snaplist_space(&pa->origin_snaps,
2657                     origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2658                 if (err)
2659                         return (err);
2660         }
2661
2662         return (0);
2663 out:
2664         pa->err_ds =  snap->ds->ds_snapname;
2665         return (err);
2666 }
2667
2668 static void
2669 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2670 {
2671         dsl_dataset_t *hds = arg1;
2672         struct promotearg *pa = arg2;
2673         struct promotenode *snap = list_head(&pa->shared_snaps);
2674         dsl_dataset_t *origin_ds = snap->ds;
2675         dsl_dataset_t *origin_head;
2676         dsl_dir_t *dd = hds->ds_dir;
2677         dsl_pool_t *dp = hds->ds_dir->dd_pool;
2678         dsl_dir_t *odd = NULL;
2679         uint64_t oldnext_obj;
2680         int64_t delta;
2681
2682         ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2683
2684         snap = list_head(&pa->origin_snaps);
2685         origin_head = snap->ds;
2686
2687         /*
2688          * We need to explicitly open odd, since origin_ds's dd will be
2689          * changing.
2690          */
2691         VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2692             NULL, FTAG, &odd));
2693
2694         /* change origin's next snap */
2695         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2696         oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2697         snap = list_tail(&pa->clone_snaps);
2698         ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2699         origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2700
2701         /* change the origin's next clone */
2702         if (origin_ds->ds_phys->ds_next_clones_obj) {
2703                 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2704                 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2705                     origin_ds->ds_phys->ds_next_clones_obj,
2706                     oldnext_obj, tx));
2707         }
2708
2709         /* change origin */
2710         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2711         ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2712         dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2713         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2714         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2715         odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2716         origin_head->ds_dir->dd_origin_txg =
2717             origin_ds->ds_phys->ds_creation_txg;
2718
2719         /* change dd_clone entries */
2720         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2721                 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2722                     odd->dd_phys->dd_clones, hds->ds_object, tx));
2723                 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2724                     pa->origin_origin->ds_dir->dd_phys->dd_clones,
2725                     hds->ds_object, tx));
2726
2727                 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2728                     pa->origin_origin->ds_dir->dd_phys->dd_clones,
2729                     origin_head->ds_object, tx));
2730                 if (dd->dd_phys->dd_clones == 0) {
2731                         dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2732                             DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2733                 }
2734                 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2735                     dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2736
2737         }
2738
2739         /* move snapshots to this dir */
2740         for (snap = list_head(&pa->shared_snaps); snap;
2741             snap = list_next(&pa->shared_snaps, snap)) {
2742                 dsl_dataset_t *ds = snap->ds;
2743
2744                 /* unregister props as dsl_dir is changing */
2745                 if (ds->ds_objset) {
2746                         dmu_objset_evict(ds->ds_objset);
2747                         ds->ds_objset = NULL;
2748                 }
2749                 /* move snap name entry */
2750                 VERIFY(0 == dsl_dataset_get_snapname(ds));
2751                 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2752                     ds->ds_snapname, tx));
2753                 VERIFY(0 == zap_add(dp->dp_meta_objset,
2754                     hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2755                     8, 1, &ds->ds_object, tx));
2756
2757                 /* change containing dsl_dir */
2758                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2759                 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2760                 ds->ds_phys->ds_dir_obj = dd->dd_object;
2761                 ASSERT3P(ds->ds_dir, ==, odd);
2762                 dsl_dir_close(ds->ds_dir, ds);
2763                 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2764                     NULL, ds, &ds->ds_dir));
2765
2766                 /* move any clone references */
2767                 if (ds->ds_phys->ds_next_clones_obj &&
2768                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2769                         zap_cursor_t zc;
2770                         zap_attribute_t za;
2771
2772                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
2773                             ds->ds_phys->ds_next_clones_obj);
2774                             zap_cursor_retrieve(&zc, &za) == 0;
2775                             zap_cursor_advance(&zc)) {
2776                                 dsl_dataset_t *cnds;
2777                                 uint64_t o;
2778
2779                                 if (za.za_first_integer == oldnext_obj) {
2780                                         /*
2781                                          * We've already moved the
2782                                          * origin's reference.
2783                                          */
2784                                         continue;
2785                                 }
2786
2787                                 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2788                                     za.za_first_integer, FTAG, &cnds));
2789                                 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2790
2791                                 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2792                                     odd->dd_phys->dd_clones, o, tx), ==, 0);
2793                                 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2794                                     dd->dd_phys->dd_clones, o, tx), ==, 0);
2795                                 dsl_dataset_rele(cnds, FTAG);
2796                         }
2797                         zap_cursor_fini(&zc);
2798                 }
2799
2800                 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2801         }
2802
2803         /*
2804          * Change space accounting.
2805          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2806          * both be valid, or both be 0 (resulting in delta == 0).  This
2807          * is true for each of {clone,origin} independently.
2808          */
2809
2810         delta = pa->cloneusedsnap -
2811             dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2812         ASSERT3S(delta, >=, 0);
2813         ASSERT3U(pa->used, >=, delta);
2814         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2815         dsl_dir_diduse_space(dd, DD_USED_HEAD,
2816             pa->used - delta, pa->comp, pa->uncomp, tx);
2817
2818         delta = pa->originusedsnap -
2819             odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2820         ASSERT3S(delta, <=, 0);
2821         ASSERT3U(pa->used, >=, -delta);
2822         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2823         dsl_dir_diduse_space(odd, DD_USED_HEAD,
2824             -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2825
2826         origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2827
2828         /* log history record */
2829         spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2830             "dataset = %llu", hds->ds_object);
2831
2832         dsl_dir_close(odd, FTAG);
2833 }
2834
2835 static char *snaplist_tag = "snaplist";
2836 /*
2837  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2838  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2839  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2840  * snapshots back to this dataset's origin.
2841  */
2842 static int
2843 snaplist_make(dsl_pool_t *dp, boolean_t own,
2844     uint64_t first_obj, uint64_t last_obj, list_t *l)
2845 {
2846         uint64_t obj = last_obj;
2847
2848         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2849
2850         list_create(l, sizeof (struct promotenode),
2851             offsetof(struct promotenode, link));
2852
2853         while (obj != first_obj) {
2854                 dsl_dataset_t *ds;
2855                 struct promotenode *snap;
2856                 int err;
2857
2858                 if (own) {
2859                         err = dsl_dataset_own_obj(dp, obj,
2860                             0, snaplist_tag, &ds);
2861                         if (err == 0)
2862                                 dsl_dataset_make_exclusive(ds, snaplist_tag);
2863                 } else {
2864                         err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2865                 }
2866                 if (err == ENOENT) {
2867                         /* lost race with snapshot destroy */
2868                         struct promotenode *last = list_tail(l);
2869                         ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2870                         obj = last->ds->ds_phys->ds_prev_snap_obj;
2871                         continue;
2872                 } else if (err) {
2873                         return (err);
2874                 }
2875
2876                 if (first_obj == 0)
2877                         first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2878
2879                 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2880                 snap->ds = ds;
2881                 list_insert_tail(l, snap);
2882                 obj = ds->ds_phys->ds_prev_snap_obj;
2883         }
2884
2885         return (0);
2886 }
2887
2888 static int
2889 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2890 {
2891         struct promotenode *snap;
2892
2893         *spacep = 0;
2894         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2895                 uint64_t used, comp, uncomp;
2896                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2897                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
2898                 *spacep += used;
2899         }
2900         return (0);
2901 }
2902
2903 static void
2904 snaplist_destroy(list_t *l, boolean_t own)
2905 {
2906         struct promotenode *snap;
2907
2908         if (!l || !list_link_active(&l->list_head))
2909                 return;
2910
2911         while ((snap = list_tail(l)) != NULL) {
2912                 list_remove(l, snap);
2913                 if (own)
2914                         dsl_dataset_disown(snap->ds, snaplist_tag);
2915                 else
2916                         dsl_dataset_rele(snap->ds, snaplist_tag);
2917                 kmem_free(snap, sizeof (struct promotenode));
2918         }
2919         list_destroy(l);
2920 }
2921
2922 /*
2923  * Promote a clone.  Nomenclature note:
2924  * "clone" or "cds": the original clone which is being promoted
2925  * "origin" or "ods": the snapshot which is originally clone's origin
2926  * "origin head" or "ohds": the dataset which is the head
2927  * (filesystem/volume) for the origin
2928  * "origin origin": the origin of the origin's filesystem (typically
2929  * NULL, indicating that the clone is not a clone of a clone).
2930  */
2931 int
2932 dsl_dataset_promote(const char *name, char *conflsnap)
2933 {
2934         dsl_dataset_t *ds;
2935         dsl_dir_t *dd;
2936         dsl_pool_t *dp;
2937         dmu_object_info_t doi;
2938         struct promotearg pa = { 0 };
2939         struct promotenode *snap;
2940         int err;
2941
2942         err = dsl_dataset_hold(name, FTAG, &ds);
2943         if (err)
2944                 return (err);
2945         dd = ds->ds_dir;
2946         dp = dd->dd_pool;
2947
2948         err = dmu_object_info(dp->dp_meta_objset,
2949             ds->ds_phys->ds_snapnames_zapobj, &doi);
2950         if (err) {
2951                 dsl_dataset_rele(ds, FTAG);
2952                 return (err);
2953         }
2954
2955         if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2956                 dsl_dataset_rele(ds, FTAG);
2957                 return (EINVAL);
2958         }
2959
2960         /*
2961          * We are going to inherit all the snapshots taken before our
2962          * origin (i.e., our new origin will be our parent's origin).
2963          * Take ownership of them so that we can rename them into our
2964          * namespace.
2965          */
2966         rw_enter(&dp->dp_config_rwlock, RW_READER);
2967
2968         err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2969             &pa.shared_snaps);
2970         if (err != 0)
2971                 goto out;
2972
2973         err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2974         if (err != 0)
2975                 goto out;
2976
2977         snap = list_head(&pa.shared_snaps);
2978         ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2979         err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2980             snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2981         if (err != 0)
2982                 goto out;
2983
2984         if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2985                 err = dsl_dataset_hold_obj(dp,
2986                     snap->ds->ds_dir->dd_phys->dd_origin_obj,
2987                     FTAG, &pa.origin_origin);
2988                 if (err != 0)
2989                         goto out;
2990         }
2991
2992 out:
2993         rw_exit(&dp->dp_config_rwlock);
2994
2995         /*
2996          * Add in 128x the snapnames zapobj size, since we will be moving
2997          * a bunch of snapnames to the promoted ds, and dirtying their
2998          * bonus buffers.
2999          */
3000         if (err == 0) {
3001                 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3002                     dsl_dataset_promote_sync, ds, &pa,
3003                     2 + 2 * doi.doi_physical_blocks_512);
3004                 if (err && pa.err_ds && conflsnap)
3005                         (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3006         }
3007
3008         snaplist_destroy(&pa.shared_snaps, B_TRUE);
3009         snaplist_destroy(&pa.clone_snaps, B_FALSE);
3010         snaplist_destroy(&pa.origin_snaps, B_FALSE);
3011         if (pa.origin_origin)
3012                 dsl_dataset_rele(pa.origin_origin, FTAG);
3013         dsl_dataset_rele(ds, FTAG);
3014         return (err);
3015 }
3016
3017 struct cloneswaparg {
3018         dsl_dataset_t *cds; /* clone dataset */
3019         dsl_dataset_t *ohds; /* origin's head dataset */
3020         boolean_t force;
3021         int64_t unused_refres_delta; /* change in unconsumed refreservation */
3022 };
3023
3024 /* ARGSUSED */
3025 static int
3026 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3027 {
3028         struct cloneswaparg *csa = arg1;
3029
3030         /* they should both be heads */
3031         if (dsl_dataset_is_snapshot(csa->cds) ||
3032             dsl_dataset_is_snapshot(csa->ohds))
3033                 return (EINVAL);
3034
3035         /* the branch point should be just before them */
3036         if (csa->cds->ds_prev != csa->ohds->ds_prev)
3037                 return (EINVAL);
3038
3039         /* cds should be the clone (unless they are unrelated) */
3040         if (csa->cds->ds_prev != NULL &&
3041             csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3042             csa->ohds->ds_object !=
3043             csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3044                 return (EINVAL);
3045
3046         /* the clone should be a child of the origin */
3047         if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3048                 return (EINVAL);
3049
3050         /* ohds shouldn't be modified unless 'force' */
3051         if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3052                 return (ETXTBSY);
3053
3054         /* adjust amount of any unconsumed refreservation */
3055         csa->unused_refres_delta =
3056             (int64_t)MIN(csa->ohds->ds_reserved,
3057             csa->ohds->ds_phys->ds_unique_bytes) -
3058             (int64_t)MIN(csa->ohds->ds_reserved,
3059             csa->cds->ds_phys->ds_unique_bytes);
3060
3061         if (csa->unused_refres_delta > 0 &&
3062             csa->unused_refres_delta >
3063             dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3064                 return (ENOSPC);
3065
3066         if (csa->ohds->ds_quota != 0 &&
3067             csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3068                 return (EDQUOT);
3069
3070         return (0);
3071 }
3072
3073 /* ARGSUSED */
3074 static void
3075 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3076 {
3077         struct cloneswaparg *csa = arg1;
3078         dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3079
3080         ASSERT(csa->cds->ds_reserved == 0);
3081         ASSERT(csa->ohds->ds_quota == 0 ||
3082             csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3083
3084         dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3085         dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3086
3087         if (csa->cds->ds_objset != NULL) {
3088                 dmu_objset_evict(csa->cds->ds_objset);
3089                 csa->cds->ds_objset = NULL;
3090         }
3091
3092         if (csa->ohds->ds_objset != NULL) {
3093                 dmu_objset_evict(csa->ohds->ds_objset);
3094                 csa->ohds->ds_objset = NULL;
3095         }
3096
3097         /*
3098          * Reset origin's unique bytes, if it exists.
3099          */
3100         if (csa->cds->ds_prev) {
3101                 dsl_dataset_t *origin = csa->cds->ds_prev;
3102                 uint64_t comp, uncomp;
3103
3104                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3105                 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3106                     origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3107                     &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3108         }
3109
3110         /* swap blkptrs */
3111         {
3112                 blkptr_t tmp;
3113                 tmp = csa->ohds->ds_phys->ds_bp;
3114                 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3115                 csa->cds->ds_phys->ds_bp = tmp;
3116         }
3117
3118         /* set dd_*_bytes */
3119         {
3120                 int64_t dused, dcomp, duncomp;
3121                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3122                 uint64_t odl_used, odl_comp, odl_uncomp;
3123
3124                 ASSERT3U(csa->cds->ds_dir->dd_phys->
3125                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
3126
3127                 dsl_deadlist_space(&csa->cds->ds_deadlist,
3128                     &cdl_used, &cdl_comp, &cdl_uncomp);
3129                 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3130                     &odl_used, &odl_comp, &odl_uncomp);
3131
3132                 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
3133                     (csa->ohds->ds_phys->ds_used_bytes + odl_used);
3134                 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3135                     (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3136                 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3137                     cdl_uncomp -
3138                     (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3139
3140                 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3141                     dused, dcomp, duncomp, tx);
3142                 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3143                     -dused, -dcomp, -duncomp, tx);
3144
3145                 /*
3146                  * The difference in the space used by snapshots is the
3147                  * difference in snapshot space due to the head's
3148                  * deadlist (since that's the only thing that's
3149                  * changing that affects the snapused).
3150                  */
3151                 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3152                     csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3153                     &cdl_used, &cdl_comp, &cdl_uncomp);
3154                 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3155                     csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3156                     &odl_used, &odl_comp, &odl_uncomp);
3157                 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3158                     DD_USED_HEAD, DD_USED_SNAP, tx);
3159         }
3160
3161         /* swap ds_*_bytes */
3162         SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
3163             csa->cds->ds_phys->ds_used_bytes);
3164         SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3165             csa->cds->ds_phys->ds_compressed_bytes);
3166         SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3167             csa->cds->ds_phys->ds_uncompressed_bytes);
3168         SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3169             csa->cds->ds_phys->ds_unique_bytes);
3170
3171         /* apply any parent delta for change in unconsumed refreservation */
3172         dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3173             csa->unused_refres_delta, 0, 0, tx);
3174
3175         /*
3176          * Swap deadlists.
3177          */
3178         dsl_deadlist_close(&csa->cds->ds_deadlist);
3179         dsl_deadlist_close(&csa->ohds->ds_deadlist);
3180         SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3181             csa->cds->ds_phys->ds_deadlist_obj);
3182         dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3183             csa->cds->ds_phys->ds_deadlist_obj);
3184         dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3185             csa->ohds->ds_phys->ds_deadlist_obj);
3186
3187         dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3188 }
3189
3190 /*
3191  * Swap 'clone' with its origin head datasets.  Used at the end of "zfs
3192  * recv" into an existing fs to swizzle the file system to the new
3193  * version, and by "zfs rollback".  Can also be used to swap two
3194  * independent head datasets if neither has any snapshots.
3195  */
3196 int
3197 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3198     boolean_t force)
3199 {
3200         struct cloneswaparg csa;
3201         int error;
3202
3203         ASSERT(clone->ds_owner);
3204         ASSERT(origin_head->ds_owner);
3205 retry:
3206         /*
3207          * Need exclusive access for the swap. If we're swapping these
3208          * datasets back after an error, we already hold the locks.
3209          */
3210         if (!RW_WRITE_HELD(&clone->ds_rwlock))
3211                 rw_enter(&clone->ds_rwlock, RW_WRITER);
3212         if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3213             !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3214                 rw_exit(&clone->ds_rwlock);
3215                 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3216                 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3217                         rw_exit(&origin_head->ds_rwlock);
3218                         goto retry;
3219                 }
3220         }
3221         csa.cds = clone;
3222         csa.ohds = origin_head;
3223         csa.force = force;
3224         error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3225             dsl_dataset_clone_swap_check,
3226             dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3227         return (error);
3228 }
3229
3230 /*
3231  * Given a pool name and a dataset object number in that pool,
3232  * return the name of that dataset.
3233  */
3234 int
3235 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3236 {
3237         spa_t *spa;
3238         dsl_pool_t *dp;
3239         dsl_dataset_t *ds;
3240         int error;
3241
3242         if ((error = spa_open(pname, &spa, FTAG)) != 0)
3243                 return (error);
3244         dp = spa_get_dsl(spa);
3245         rw_enter(&dp->dp_config_rwlock, RW_READER);
3246         if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3247                 dsl_dataset_name(ds, buf);
3248                 dsl_dataset_rele(ds, FTAG);
3249         }
3250         rw_exit(&dp->dp_config_rwlock);
3251         spa_close(spa, FTAG);
3252
3253         return (error);
3254 }
3255
3256 int
3257 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3258     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3259 {
3260         int error = 0;
3261
3262         ASSERT3S(asize, >, 0);
3263
3264         /*
3265          * *ref_rsrv is the portion of asize that will come from any
3266          * unconsumed refreservation space.
3267          */
3268         *ref_rsrv = 0;
3269
3270         mutex_enter(&ds->ds_lock);
3271         /*
3272          * Make a space adjustment for reserved bytes.
3273          */
3274         if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3275                 ASSERT3U(*used, >=,
3276                     ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3277                 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3278                 *ref_rsrv =
3279                     asize - MIN(asize, parent_delta(ds, asize + inflight));
3280         }
3281
3282         if (!check_quota || ds->ds_quota == 0) {
3283                 mutex_exit(&ds->ds_lock);
3284                 return (0);
3285         }
3286         /*
3287          * If they are requesting more space, and our current estimate
3288          * is over quota, they get to try again unless the actual
3289          * on-disk is over quota and there are no pending changes (which
3290          * may free up space for us).
3291          */
3292         if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
3293                 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
3294                         error = ERESTART;
3295                 else
3296                         error = EDQUOT;
3297         }
3298         mutex_exit(&ds->ds_lock);
3299
3300         return (error);
3301 }
3302
3303 /* ARGSUSED */
3304 static int
3305 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3306 {
3307         dsl_dataset_t *ds = arg1;
3308         dsl_prop_setarg_t *psa = arg2;
3309         int err;
3310
3311         if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3312                 return (ENOTSUP);
3313
3314         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3315                 return (err);
3316
3317         if (psa->psa_effective_value == 0)
3318                 return (0);
3319
3320         if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes ||
3321             psa->psa_effective_value < ds->ds_reserved)
3322                 return (ENOSPC);
3323
3324         return (0);
3325 }
3326
3327 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3328
3329 void
3330 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3331 {
3332         dsl_dataset_t *ds = arg1;
3333         dsl_prop_setarg_t *psa = arg2;
3334         uint64_t effective_value = psa->psa_effective_value;
3335
3336         dsl_prop_set_sync(ds, psa, tx);
3337         DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3338
3339         if (ds->ds_quota != effective_value) {
3340                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3341                 ds->ds_quota = effective_value;
3342
3343                 spa_history_log_internal(LOG_DS_REFQUOTA,
3344                     ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3345                     (longlong_t)ds->ds_quota, ds->ds_object);
3346         }
3347 }
3348
3349 int
3350 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3351 {
3352         dsl_dataset_t *ds;
3353         dsl_prop_setarg_t psa;
3354         int err;
3355
3356         dsl_prop_setarg_init_uint64(&psa, "refquota", source, &quota);
3357
3358         err = dsl_dataset_hold(dsname, FTAG, &ds);
3359         if (err)
3360                 return (err);
3361
3362         /*
3363          * If someone removes a file, then tries to set the quota, we
3364          * want to make sure the file freeing takes effect.
3365          */
3366         txg_wait_open(ds->ds_dir->dd_pool, 0);
3367
3368         err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3369             dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3370             ds, &psa, 0);
3371
3372         dsl_dataset_rele(ds, FTAG);
3373         return (err);
3374 }
3375
3376 static int
3377 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3378 {
3379         dsl_dataset_t *ds = arg1;
3380         dsl_prop_setarg_t *psa = arg2;
3381         uint64_t effective_value;
3382         uint64_t unique;
3383         int err;
3384
3385         if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3386             SPA_VERSION_REFRESERVATION)
3387                 return (ENOTSUP);
3388
3389         if (dsl_dataset_is_snapshot(ds))
3390                 return (EINVAL);
3391
3392         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3393                 return (err);
3394
3395         effective_value = psa->psa_effective_value;
3396
3397         /*
3398          * If we are doing the preliminary check in open context, the
3399          * space estimates may be inaccurate.
3400          */
3401         if (!dmu_tx_is_syncing(tx))
3402                 return (0);
3403
3404         mutex_enter(&ds->ds_lock);
3405         if (!DS_UNIQUE_IS_ACCURATE(ds))
3406                 dsl_dataset_recalc_head_uniq(ds);
3407         unique = ds->ds_phys->ds_unique_bytes;
3408         mutex_exit(&ds->ds_lock);
3409
3410         if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3411                 uint64_t delta = MAX(unique, effective_value) -
3412                     MAX(unique, ds->ds_reserved);
3413
3414                 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3415                         return (ENOSPC);
3416                 if (ds->ds_quota > 0 &&
3417                     effective_value > ds->ds_quota)
3418                         return (ENOSPC);
3419         }
3420
3421         return (0);
3422 }
3423
3424 static void
3425 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3426 {
3427         dsl_dataset_t *ds = arg1;
3428         dsl_prop_setarg_t *psa = arg2;
3429         uint64_t effective_value = psa->psa_effective_value;
3430         uint64_t unique;
3431         int64_t delta;
3432
3433         dsl_prop_set_sync(ds, psa, tx);
3434         DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3435
3436         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3437
3438         mutex_enter(&ds->ds_dir->dd_lock);
3439         mutex_enter(&ds->ds_lock);
3440         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3441         unique = ds->ds_phys->ds_unique_bytes;
3442         delta = MAX(0, (int64_t)(effective_value - unique)) -
3443             MAX(0, (int64_t)(ds->ds_reserved - unique));
3444         ds->ds_reserved = effective_value;
3445         mutex_exit(&ds->ds_lock);
3446
3447         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3448         mutex_exit(&ds->ds_dir->dd_lock);
3449
3450         spa_history_log_internal(LOG_DS_REFRESERV,
3451             ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3452             (longlong_t)effective_value, ds->ds_object);
3453 }
3454
3455 int
3456 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3457     uint64_t reservation)
3458 {
3459         dsl_dataset_t *ds;
3460         dsl_prop_setarg_t psa;
3461         int err;
3462
3463         dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3464             &reservation);
3465
3466         err = dsl_dataset_hold(dsname, FTAG, &ds);
3467         if (err)
3468                 return (err);
3469
3470         err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3471             dsl_dataset_set_reservation_check,
3472             dsl_dataset_set_reservation_sync, ds, &psa, 0);
3473
3474         dsl_dataset_rele(ds, FTAG);
3475         return (err);
3476 }
3477
3478 typedef struct zfs_hold_cleanup_arg {
3479         dsl_pool_t *dp;
3480         uint64_t dsobj;
3481         char htag[MAXNAMELEN];
3482 } zfs_hold_cleanup_arg_t;
3483
3484 static void
3485 dsl_dataset_user_release_onexit(void *arg)
3486 {
3487         zfs_hold_cleanup_arg_t *ca = arg;
3488
3489         (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3490             B_TRUE);
3491         kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3492 }
3493
3494 void
3495 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3496     minor_t minor)
3497 {
3498         zfs_hold_cleanup_arg_t *ca;
3499
3500         ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3501         ca->dp = ds->ds_dir->dd_pool;
3502         ca->dsobj = ds->ds_object;
3503         (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3504         VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3505             dsl_dataset_user_release_onexit, ca, NULL));
3506 }
3507
3508 /*
3509  * If you add new checks here, you may need to add
3510  * additional checks to the "temporary" case in
3511  * snapshot_check() in dmu_objset.c.
3512  */
3513 static int
3514 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3515 {
3516         dsl_dataset_t *ds = arg1;
3517         struct dsl_ds_holdarg *ha = arg2;
3518         char *htag = ha->htag;
3519         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3520         int error = 0;
3521
3522         if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3523                 return (ENOTSUP);
3524
3525         if (!dsl_dataset_is_snapshot(ds))
3526                 return (EINVAL);
3527
3528         /* tags must be unique */
3529         mutex_enter(&ds->ds_lock);
3530         if (ds->ds_phys->ds_userrefs_obj) {
3531                 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3532                     8, 1, tx);
3533                 if (error == 0)
3534                         error = EEXIST;
3535                 else if (error == ENOENT)
3536                         error = 0;
3537         }
3538         mutex_exit(&ds->ds_lock);
3539
3540         if (error == 0 && ha->temphold &&
3541             strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3542                 error = E2BIG;
3543
3544         return (error);
3545 }
3546
3547 void
3548 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3549 {
3550         dsl_dataset_t *ds = arg1;
3551         struct dsl_ds_holdarg *ha = arg2;
3552         char *htag = ha->htag;
3553         dsl_pool_t *dp = ds->ds_dir->dd_pool;
3554         objset_t *mos = dp->dp_meta_objset;
3555         uint64_t now = gethrestime_sec();
3556         uint64_t zapobj;
3557
3558         mutex_enter(&ds->ds_lock);
3559         if (ds->ds_phys->ds_userrefs_obj == 0) {
3560                 /*
3561                  * This is the first user hold for this dataset.  Create
3562                  * the userrefs zap object.
3563                  */
3564                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3565                 zapobj = ds->ds_phys->ds_userrefs_obj =
3566                     zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3567         } else {
3568                 zapobj = ds->ds_phys->ds_userrefs_obj;
3569         }
3570         ds->ds_userrefs++;
3571         mutex_exit(&ds->ds_lock);
3572
3573         VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3574
3575         if (ha->temphold) {
3576                 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3577                     htag, &now, tx));
3578         }
3579
3580         spa_history_log_internal(LOG_DS_USER_HOLD,
3581             dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3582             (int)ha->temphold, ds->ds_object);
3583 }
3584
3585 static int
3586 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3587 {
3588         struct dsl_ds_holdarg *ha = arg;
3589         dsl_dataset_t *ds;
3590         int error;
3591         char *name;
3592
3593         /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3594         name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3595         error = dsl_dataset_hold(name, ha->dstg, &ds);
3596         strfree(name);
3597         if (error == 0) {
3598                 ha->gotone = B_TRUE;
3599                 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3600                     dsl_dataset_user_hold_sync, ds, ha, 0);
3601         } else if (error == ENOENT && ha->recursive) {
3602                 error = 0;
3603         } else {
3604                 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3605         }
3606         return (error);
3607 }
3608
3609 int
3610 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3611     boolean_t temphold)
3612 {
3613         struct dsl_ds_holdarg *ha;
3614         int error;
3615
3616         ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3617         ha->htag = htag;
3618         ha->temphold = temphold;
3619         error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3620             dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3621             ds, ha, 0);
3622         kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3623
3624         return (error);
3625 }
3626
3627 int
3628 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3629     boolean_t recursive, boolean_t temphold, int cleanup_fd)
3630 {
3631         struct dsl_ds_holdarg *ha;
3632         dsl_sync_task_t *dst;
3633         spa_t *spa;
3634         int error;
3635         minor_t minor = 0;
3636
3637         if (cleanup_fd != -1) {
3638                 /* Currently we only support cleanup-on-exit of tempholds. */
3639                 if (!temphold)
3640                         return (EINVAL);
3641                 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3642                 if (error)
3643                         return (error);
3644         }
3645
3646         ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3647
3648         (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3649
3650         error = spa_open(dsname, &spa, FTAG);
3651         if (error) {
3652                 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3653                 if (cleanup_fd != -1)
3654                         zfs_onexit_fd_rele(cleanup_fd);
3655                 return (error);
3656         }
3657
3658         ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3659         ha->htag = htag;
3660         ha->snapname = snapname;
3661         ha->recursive = recursive;
3662         ha->temphold = temphold;
3663
3664         if (recursive) {
3665                 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3666                     ha, DS_FIND_CHILDREN);
3667         } else {
3668                 error = dsl_dataset_user_hold_one(dsname, ha);
3669         }
3670         if (error == 0)
3671                 error = dsl_sync_task_group_wait(ha->dstg);
3672
3673         for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3674             dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3675                 dsl_dataset_t *ds = dst->dst_arg1;
3676
3677                 if (dst->dst_err) {
3678                         dsl_dataset_name(ds, ha->failed);
3679                         *strchr(ha->failed, '@') = '\0';
3680                 } else if (error == 0 && minor != 0 && temphold) {
3681                         /*
3682                          * If this hold is to be released upon process exit,
3683                          * register that action now.
3684                          */
3685                         dsl_register_onexit_hold_cleanup(ds, htag, minor);
3686                 }
3687                 dsl_dataset_rele(ds, ha->dstg);
3688         }
3689
3690         if (error == 0 && recursive && !ha->gotone)
3691                 error = ENOENT;
3692
3693         if (error)
3694                 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3695
3696         dsl_sync_task_group_destroy(ha->dstg);
3697
3698         kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3699         spa_close(spa, FTAG);
3700         if (cleanup_fd != -1)
3701                 zfs_onexit_fd_rele(cleanup_fd);
3702         return (error);
3703 }
3704
3705 struct dsl_ds_releasearg {
3706         dsl_dataset_t *ds;
3707         const char *htag;
3708         boolean_t own;          /* do we own or just hold ds? */
3709 };
3710
3711 static int
3712 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3713     boolean_t *might_destroy)
3714 {
3715         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3716         uint64_t zapobj;
3717         uint64_t tmp;
3718         int error;
3719
3720         *might_destroy = B_FALSE;
3721
3722         mutex_enter(&ds->ds_lock);
3723         zapobj = ds->ds_phys->ds_userrefs_obj;
3724         if (zapobj == 0) {
3725                 /* The tag can't possibly exist */
3726                 mutex_exit(&ds->ds_lock);
3727                 return (ESRCH);
3728         }
3729
3730         /* Make sure the tag exists */
3731         error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3732         if (error) {
3733                 mutex_exit(&ds->ds_lock);
3734                 if (error == ENOENT)
3735                         error = ESRCH;
3736                 return (error);
3737         }
3738
3739         if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3740             DS_IS_DEFER_DESTROY(ds))
3741                 *might_destroy = B_TRUE;
3742
3743         mutex_exit(&ds->ds_lock);
3744         return (0);
3745 }
3746
3747 static int
3748 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3749 {
3750         struct dsl_ds_releasearg *ra = arg1;
3751         dsl_dataset_t *ds = ra->ds;
3752         boolean_t might_destroy;
3753         int error;
3754
3755         if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3756                 return (ENOTSUP);
3757
3758         error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3759         if (error)
3760                 return (error);
3761
3762         if (might_destroy) {
3763                 struct dsl_ds_destroyarg dsda = {0};
3764
3765                 if (dmu_tx_is_syncing(tx)) {
3766                         /*
3767                          * If we're not prepared to remove the snapshot,
3768                          * we can't allow the release to happen right now.
3769                          */
3770                         if (!ra->own)
3771                                 return (EBUSY);
3772                 }
3773                 dsda.ds = ds;
3774                 dsda.releasing = B_TRUE;
3775                 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3776         }
3777
3778         return (0);
3779 }
3780
3781 static void
3782 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3783 {
3784         struct dsl_ds_releasearg *ra = arg1;
3785         dsl_dataset_t *ds = ra->ds;
3786         dsl_pool_t *dp = ds->ds_dir->dd_pool;
3787         objset_t *mos = dp->dp_meta_objset;
3788         uint64_t zapobj;
3789         uint64_t dsobj = ds->ds_object;
3790         uint64_t refs;
3791         int error;
3792
3793         mutex_enter(&ds->ds_lock);
3794         ds->ds_userrefs--;
3795         refs = ds->ds_userrefs;
3796         mutex_exit(&ds->ds_lock);
3797         error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3798         VERIFY(error == 0 || error == ENOENT);
3799         zapobj = ds->ds_phys->ds_userrefs_obj;
3800         VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3801         if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3802             DS_IS_DEFER_DESTROY(ds)) {
3803                 struct dsl_ds_destroyarg dsda = {0};
3804
3805                 ASSERT(ra->own);
3806                 dsda.ds = ds;
3807                 dsda.releasing = B_TRUE;
3808                 /* We already did the destroy_check */
3809                 dsl_dataset_destroy_sync(&dsda, tag, tx);
3810         }
3811
3812         spa_history_log_internal(LOG_DS_USER_RELEASE,
3813             dp->dp_spa, tx, "<%s> %lld dataset = %llu",
3814             ra->htag, (longlong_t)refs, dsobj);
3815 }
3816
3817 static int
3818 dsl_dataset_user_release_one(const char *dsname, void *arg)
3819 {
3820         struct dsl_ds_holdarg *ha = arg;
3821         struct dsl_ds_releasearg *ra;
3822         dsl_dataset_t *ds;
3823         int error;
3824         void *dtag = ha->dstg;
3825         char *name;
3826         boolean_t own = B_FALSE;
3827         boolean_t might_destroy;
3828
3829         /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3830         name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3831         error = dsl_dataset_hold(name, dtag, &ds);
3832         strfree(name);
3833         if (error == ENOENT && ha->recursive)
3834                 return (0);
3835         (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3836         if (error)
3837                 return (error);
3838
3839         ha->gotone = B_TRUE;
3840
3841         ASSERT(dsl_dataset_is_snapshot(ds));
3842
3843         error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3844         if (error) {
3845                 dsl_dataset_rele(ds, dtag);
3846                 return (error);
3847         }
3848
3849         if (might_destroy) {
3850 #ifdef _KERNEL
3851                 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3852                 error = zfs_unmount_snap(name, NULL);
3853                 strfree(name);
3854                 if (error) {
3855                         dsl_dataset_rele(ds, dtag);
3856                         return (error);
3857                 }
3858 #endif
3859                 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3860                         dsl_dataset_rele(ds, dtag);
3861                         return (EBUSY);
3862                 } else {
3863                         own = B_TRUE;
3864                         dsl_dataset_make_exclusive(ds, dtag);
3865                 }
3866         }
3867
3868         ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3869         ra->ds = ds;
3870         ra->htag = ha->htag;
3871         ra->own = own;
3872         dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3873             dsl_dataset_user_release_sync, ra, dtag, 0);
3874
3875         return (0);
3876 }
3877
3878 int
3879 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3880     boolean_t recursive)
3881 {
3882         struct dsl_ds_holdarg *ha;
3883         dsl_sync_task_t *dst;
3884         spa_t *spa;
3885         int error;
3886
3887 top:
3888         ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3889
3890         (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3891
3892         error = spa_open(dsname, &spa, FTAG);
3893         if (error) {
3894                 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3895                 return (error);
3896         }
3897
3898         ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3899         ha->htag = htag;
3900         ha->snapname = snapname;
3901         ha->recursive = recursive;
3902         if (recursive) {
3903                 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3904                     ha, DS_FIND_CHILDREN);
3905         } else {
3906                 error = dsl_dataset_user_release_one(dsname, ha);
3907         }
3908         if (error == 0)
3909                 error = dsl_sync_task_group_wait(ha->dstg);
3910
3911         for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3912             dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3913                 struct dsl_ds_releasearg *ra = dst->dst_arg1;
3914                 dsl_dataset_t *ds = ra->ds;
3915
3916                 if (dst->dst_err)
3917                         dsl_dataset_name(ds, ha->failed);
3918
3919                 if (ra->own)
3920                         dsl_dataset_disown(ds, ha->dstg);
3921                 else
3922                         dsl_dataset_rele(ds, ha->dstg);
3923
3924                 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
3925         }
3926
3927         if (error == 0 && recursive && !ha->gotone)
3928                 error = ENOENT;
3929
3930         if (error && error != EBUSY)
3931                 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3932
3933         dsl_sync_task_group_destroy(ha->dstg);
3934         kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3935         spa_close(spa, FTAG);
3936
3937         /*
3938          * We can get EBUSY if we were racing with deferred destroy and
3939          * dsl_dataset_user_release_check() hadn't done the necessary
3940          * open context setup.  We can also get EBUSY if we're racing
3941          * with destroy and that thread is the ds_owner.  Either way
3942          * the busy condition should be transient, and we should retry
3943          * the release operation.
3944          */
3945         if (error == EBUSY)
3946                 goto top;
3947
3948         return (error);
3949 }
3950
3951 /*
3952  * Called at spa_load time (with retry == B_FALSE) to release a stale
3953  * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
3954  */
3955 int
3956 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
3957     boolean_t retry)
3958 {
3959         dsl_dataset_t *ds;
3960         char *snap;
3961         char *name;
3962         int namelen;
3963         int error;
3964
3965         do {
3966                 rw_enter(&dp->dp_config_rwlock, RW_READER);
3967                 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
3968                 rw_exit(&dp->dp_config_rwlock);
3969                 if (error)
3970                         return (error);
3971                 namelen = dsl_dataset_namelen(ds)+1;
3972                 name = kmem_alloc(namelen, KM_SLEEP);
3973                 dsl_dataset_name(ds, name);
3974                 dsl_dataset_rele(ds, FTAG);
3975
3976                 snap = strchr(name, '@');
3977                 *snap = '\0';
3978                 ++snap;
3979                 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
3980                 kmem_free(name, namelen);
3981
3982                 /*
3983                  * The object can't have been destroyed because we have a hold,
3984                  * but it might have been renamed, resulting in ENOENT.  Retry
3985                  * if we've been requested to do so.
3986                  *
3987                  * It would be nice if we could use the dsobj all the way
3988                  * through and avoid ENOENT entirely.  But we might need to
3989                  * unmount the snapshot, and there's currently no way to lookup
3990                  * a vfsp using a ZFS object id.
3991                  */
3992         } while ((error == ENOENT) && retry);
3993
3994         return (error);
3995 }
3996
3997 int
3998 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
3999 {
4000         dsl_dataset_t *ds;
4001         int err;
4002
4003         err = dsl_dataset_hold(dsname, FTAG, &ds);
4004         if (err)
4005                 return (err);
4006
4007         VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4008         if (ds->ds_phys->ds_userrefs_obj != 0) {
4009                 zap_attribute_t *za;
4010                 zap_cursor_t zc;
4011
4012                 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4013                 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4014                     ds->ds_phys->ds_userrefs_obj);
4015                     zap_cursor_retrieve(&zc, za) == 0;
4016                     zap_cursor_advance(&zc)) {
4017                         VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4018                             za->za_first_integer));
4019                 }
4020                 zap_cursor_fini(&zc);
4021                 kmem_free(za, sizeof (zap_attribute_t));
4022         }
4023         dsl_dataset_rele(ds, FTAG);
4024         return (0);
4025 }
4026
4027 /*
4028  * Note, this fuction is used as the callback for dmu_objset_find().  We
4029  * always return 0 so that we will continue to find and process
4030  * inconsistent datasets, even if we encounter an error trying to
4031  * process one of them.
4032  */
4033 /* ARGSUSED */
4034 int
4035 dsl_destroy_inconsistent(const char *dsname, void *arg)
4036 {
4037         dsl_dataset_t *ds;
4038
4039         if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4040                 if (DS_IS_INCONSISTENT(ds))
4041                         (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4042                 else
4043                         dsl_dataset_disown(ds, FTAG);
4044         }
4045         return (0);
4046 }