module/zfs/dsl_destroy.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) 2012, 2017 by Delphix. All rights reserved.
  24  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  25  * Copyright (c) 2013 by Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2016 Actifio, Inc. All rights reserved.
  27  */
  28
  29 #include <sys/zfs_context.h>
  30 #include <sys/dsl_userhold.h>
  31 #include <sys/dsl_dataset.h>
  32 #include <sys/dsl_synctask.h>
  33 #include <sys/dsl_destroy.h>
  34 #include <sys/dmu_tx.h>
  35 #include <sys/dsl_pool.h>
  36 #include <sys/dsl_dir.h>
  37 #include <sys/dmu_traverse.h>
  38 #include <sys/dsl_scan.h>
  39 #include <sys/dmu_objset.h>
  40 #include <sys/zap.h>
  41 #include <sys/zfeature.h>
  42 #include <sys/zfs_ioctl.h>
  43 #include <sys/dsl_deleg.h>
  44 #include <sys/dmu_impl.h>
  45 #include <sys/zvol.h>
  46 #include <sys/zcp.h>
  47
  48 int
  49 dsl_destroy_snapshot_check_impl(dsl_dataset_t *ds, boolean_t defer)
  50 {
  51         if (!ds->ds_is_snapshot)
  52                 return (SET_ERROR(EINVAL));
  53
  54         if (dsl_dataset_long_held(ds))
  55                 return (SET_ERROR(EBUSY));
  56
  57         /*
  58          * Only allow deferred destroy on pools that support it.
  59          * NOTE: deferred destroy is only supported on snapshots.
  60          */
  61         if (defer) {
  62                 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
  63                     SPA_VERSION_USERREFS)
  64                         return (SET_ERROR(ENOTSUP));
  65                 return (0);
  66         }
  67
  68         /*
  69          * If this snapshot has an elevated user reference count,
  70          * we can't destroy it yet.
  71          */
  72         if (ds->ds_userrefs > 0)
  73                 return (SET_ERROR(EBUSY));
  74
  75         /*
  76          * Can't delete a branch point.
  77          */
  78         if (dsl_dataset_phys(ds)->ds_num_children > 1)
  79                 return (SET_ERROR(EEXIST));
  80
  81         return (0);
  82 }
  83
  84 int
  85 dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx)
  86 {
  87         dsl_destroy_snapshot_arg_t *ddsa = arg;
  88         const char *dsname = ddsa->ddsa_name;
  89         boolean_t defer = ddsa->ddsa_defer;
  90
  91         dsl_pool_t *dp = dmu_tx_pool(tx);
  92         int error = 0;
  93         dsl_dataset_t *ds;
  94
  95         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
  96
  97         /*
  98          * If the snapshot does not exist, silently ignore it, and
  99          * dsl_destroy_snapshot_sync() will be a no-op
 100          * (it's "already destroyed").
 101          */
 102         if (error == ENOENT)
 103                 return (0);
 104
 105         if (error == 0) {
 106                 error = dsl_destroy_snapshot_check_impl(ds, defer);
 107                 dsl_dataset_rele(ds, FTAG);
 108         }
 109
 110         return (error);
 111 }
 112
 113 struct process_old_arg {
 114         dsl_dataset_t *ds;
 115         dsl_dataset_t *ds_prev;
 116         boolean_t after_branch_point;
 117         zio_t *pio;
 118         uint64_t used, comp, uncomp;
 119 };
 120
 121 static int
 122 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
 123 {
 124         struct process_old_arg *poa = arg;
 125         dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
 126
 127         ASSERT(!BP_IS_HOLE(bp));
 128
 129         if (bp->blk_birth <= dsl_dataset_phys(poa->ds)->ds_prev_snap_txg) {
 130                 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
 131                 if (poa->ds_prev && !poa->after_branch_point &&
 132                     bp->blk_birth >
 133                     dsl_dataset_phys(poa->ds_prev)->ds_prev_snap_txg) {
 134                         dsl_dataset_phys(poa->ds_prev)->ds_unique_bytes +=
 135                             bp_get_dsize_sync(dp->dp_spa, bp);
 136                 }
 137         } else {
 138                 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
 139                 poa->comp += BP_GET_PSIZE(bp);
 140                 poa->uncomp += BP_GET_UCSIZE(bp);
 141                 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
 142         }
 143         return (0);
 144 }
 145
 146 static void
 147 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
 148     dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
 149 {
 150         struct process_old_arg poa = { 0 };
 151         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 152         objset_t *mos = dp->dp_meta_objset;
 153         uint64_t deadlist_obj;
 154
 155         ASSERT(ds->ds_deadlist.dl_oldfmt);
 156         ASSERT(ds_next->ds_deadlist.dl_oldfmt);
 157
 158         poa.ds = ds;
 159         poa.ds_prev = ds_prev;
 160         poa.after_branch_point = after_branch_point;
 161         poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
 162         VERIFY0(bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
 163             process_old_cb, &poa, tx));
 164         VERIFY0(zio_wait(poa.pio));
 165         ASSERT3U(poa.used, ==, dsl_dataset_phys(ds)->ds_unique_bytes);
 166
 167         /* change snapused */
 168         dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
 169             -poa.used, -poa.comp, -poa.uncomp, tx);
 170
 171         /* swap next's deadlist to our deadlist */
 172         dsl_deadlist_close(&ds->ds_deadlist);
 173         dsl_deadlist_close(&ds_next->ds_deadlist);
 174         deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
 175         dsl_dataset_phys(ds)->ds_deadlist_obj =
 176             dsl_dataset_phys(ds_next)->ds_deadlist_obj;
 177         dsl_dataset_phys(ds_next)->ds_deadlist_obj = deadlist_obj;
 178         dsl_deadlist_open(&ds->ds_deadlist, mos,
 179             dsl_dataset_phys(ds)->ds_deadlist_obj);
 180         dsl_deadlist_open(&ds_next->ds_deadlist, mos,
 181             dsl_dataset_phys(ds_next)->ds_deadlist_obj);
 182 }
 183
 184 static void
 185 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
 186 {
 187         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 188         zap_cursor_t *zc;
 189         zap_attribute_t *za;
 190
 191         /*
 192          * If it is the old version, dd_clones doesn't exist so we can't
 193          * find the clones, but dsl_deadlist_remove_key() is a no-op so it
 194          * doesn't matter.
 195          */
 196         if (dsl_dir_phys(ds->ds_dir)->dd_clones == 0)
 197                 return;
 198
 199         zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
 200         za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
 201
 202         for (zap_cursor_init(zc, mos, dsl_dir_phys(ds->ds_dir)->dd_clones);
 203             zap_cursor_retrieve(zc, za) == 0;
 204             zap_cursor_advance(zc)) {
 205                 dsl_dataset_t *clone;
 206
 207                 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
 208                     za->za_first_integer, FTAG, &clone));
 209                 if (clone->ds_dir->dd_origin_txg > mintxg) {
 210                         dsl_deadlist_remove_key(&clone->ds_deadlist,
 211                             mintxg, tx);
 212                         if (dsl_dataset_remap_deadlist_exists(clone)) {
 213                                 dsl_deadlist_remove_key(
 214                                     &clone->ds_remap_deadlist, mintxg, tx);
 215                         }
 216                         dsl_dataset_remove_clones_key(clone, mintxg, tx);
 217                 }
 218                 dsl_dataset_rele(clone, FTAG);
 219         }
 220         zap_cursor_fini(zc);
 221
 222         kmem_free(za, sizeof (zap_attribute_t));
 223         kmem_free(zc, sizeof (zap_cursor_t));
 224 }
 225
 226 static void
 227 dsl_destroy_snapshot_handle_remaps(dsl_dataset_t *ds, dsl_dataset_t *ds_next,
 228     dmu_tx_t *tx)
 229 {
 230         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 231
 232         /* Move blocks to be obsoleted to pool's obsolete list. */
 233         if (dsl_dataset_remap_deadlist_exists(ds_next)) {
 234                 if (!bpobj_is_open(&dp->dp_obsolete_bpobj))
 235                         dsl_pool_create_obsolete_bpobj(dp, tx);
 236
 237                 dsl_deadlist_move_bpobj(&ds_next->ds_remap_deadlist,
 238                     &dp->dp_obsolete_bpobj,
 239                     dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
 240         }
 241
 242         /* Merge our deadlist into next's and free it. */
 243         if (dsl_dataset_remap_deadlist_exists(ds)) {
 244                 uint64_t remap_deadlist_object =
 245                     dsl_dataset_get_remap_deadlist_object(ds);
 246                 ASSERT(remap_deadlist_object != 0);
 247
 248                 mutex_enter(&ds_next->ds_remap_deadlist_lock);
 249                 if (!dsl_dataset_remap_deadlist_exists(ds_next))
 250                         dsl_dataset_create_remap_deadlist(ds_next, tx);
 251                 mutex_exit(&ds_next->ds_remap_deadlist_lock);
 252
 253                 dsl_deadlist_merge(&ds_next->ds_remap_deadlist,
 254                     remap_deadlist_object, tx);
 255                 dsl_dataset_destroy_remap_deadlist(ds, tx);
 256         }
 257 }
 258
 259 void
 260 dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
 261 {
 262         int after_branch_point = FALSE;
 263         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 264         objset_t *mos = dp->dp_meta_objset;
 265         dsl_dataset_t *ds_prev = NULL;
 266         uint64_t obj;
 267
 268         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
 269         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
 270         ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
 271         rrw_exit(&ds->ds_bp_rwlock, FTAG);
 272         ASSERT(refcount_is_zero(&ds->ds_longholds));
 273
 274         if (defer &&
 275             (ds->ds_userrefs > 0 ||
 276             dsl_dataset_phys(ds)->ds_num_children > 1)) {
 277                 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
 278                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
 279                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_DEFER_DESTROY;
 280                 spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
 281                 return;
 282         }
 283
 284         ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
 285
 286         /* We need to log before removing it from the namespace. */
 287         spa_history_log_internal_ds(ds, "destroy", tx, "");
 288
 289         dsl_scan_ds_destroyed(ds, tx);
 290
 291         obj = ds->ds_object;
 292
 293         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 294                 if (ds->ds_feature_inuse[f]) {
 295                         dsl_dataset_deactivate_feature(obj, f, tx);
 296                         ds->ds_feature_inuse[f] = B_FALSE;
 297                 }
 298         }
 299         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 300                 ASSERT3P(ds->ds_prev, ==, NULL);
 301                 VERIFY0(dsl_dataset_hold_obj(dp,
 302                     dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &ds_prev));
 303                 after_branch_point =
 304                     (dsl_dataset_phys(ds_prev)->ds_next_snap_obj != obj);
 305
 306                 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
 307                 if (after_branch_point &&
 308                     dsl_dataset_phys(ds_prev)->ds_next_clones_obj != 0) {
 309                         dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
 310                         if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
 311                                 VERIFY0(zap_add_int(mos,
 312                                     dsl_dataset_phys(ds_prev)->
 313                                     ds_next_clones_obj,
 314                                     dsl_dataset_phys(ds)->ds_next_snap_obj,
 315                                     tx));
 316                         }
 317                 }
 318                 if (!after_branch_point) {
 319                         dsl_dataset_phys(ds_prev)->ds_next_snap_obj =
 320                             dsl_dataset_phys(ds)->ds_next_snap_obj;
 321                 }
 322         }
 323
 324         dsl_dataset_t *ds_next;
 325         uint64_t old_unique;
 326         uint64_t used = 0, comp = 0, uncomp = 0;
 327
 328         VERIFY0(dsl_dataset_hold_obj(dp,
 329             dsl_dataset_phys(ds)->ds_next_snap_obj, FTAG, &ds_next));
 330         ASSERT3U(dsl_dataset_phys(ds_next)->ds_prev_snap_obj, ==, obj);
 331
 332         old_unique = dsl_dataset_phys(ds_next)->ds_unique_bytes;
 333
 334         dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
 335         dsl_dataset_phys(ds_next)->ds_prev_snap_obj =
 336             dsl_dataset_phys(ds)->ds_prev_snap_obj;
 337         dsl_dataset_phys(ds_next)->ds_prev_snap_txg =
 338             dsl_dataset_phys(ds)->ds_prev_snap_txg;
 339         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
 340             ds_prev ? dsl_dataset_phys(ds_prev)->ds_creation_txg : 0);
 341
 342         if (ds_next->ds_deadlist.dl_oldfmt) {
 343                 process_old_deadlist(ds, ds_prev, ds_next,
 344                     after_branch_point, tx);
 345         } else {
 346                 /* Adjust prev's unique space. */
 347                 if (ds_prev && !after_branch_point) {
 348                         dsl_deadlist_space_range(&ds_next->ds_deadlist,
 349                             dsl_dataset_phys(ds_prev)->ds_prev_snap_txg,
 350                             dsl_dataset_phys(ds)->ds_prev_snap_txg,
 351                             &used, &comp, &uncomp);
 352                         dsl_dataset_phys(ds_prev)->ds_unique_bytes += used;
 353                 }
 354
 355                 /* Adjust snapused. */
 356                 dsl_deadlist_space_range(&ds_next->ds_deadlist,
 357                     dsl_dataset_phys(ds)->ds_prev_snap_txg, UINT64_MAX,
 358                     &used, &comp, &uncomp);
 359                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
 360                     -used, -comp, -uncomp, tx);
 361
 362                 /* Move blocks to be freed to pool's free list. */
 363                 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
 364                     &dp->dp_free_bpobj, dsl_dataset_phys(ds)->ds_prev_snap_txg,
 365                     tx);
 366                 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
 367                     DD_USED_HEAD, used, comp, uncomp, tx);
 368
 369                 /* Merge our deadlist into next's and free it. */
 370                 dsl_deadlist_merge(&ds_next->ds_deadlist,
 371                     dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
 372         }
 373
 374         dsl_deadlist_close(&ds->ds_deadlist);
 375         dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
 376         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 377         dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
 378
 379         dsl_destroy_snapshot_handle_remaps(ds, ds_next, tx);
 380
 381         /* Collapse range in clone heads */
 382         dsl_dataset_remove_clones_key(ds,
 383             dsl_dataset_phys(ds)->ds_creation_txg, tx);
 384
 385         if (ds_next->ds_is_snapshot) {
 386                 dsl_dataset_t *ds_nextnext;
 387
 388                 /*
 389                  * Update next's unique to include blocks which
 390                  * were previously shared by only this snapshot
 391                  * and it.  Those blocks will be born after the
 392                  * prev snap and before this snap, and will have
 393                  * died after the next snap and before the one
 394                  * after that (ie. be on the snap after next's
 395                  * deadlist).
 396                  */
 397                 VERIFY0(dsl_dataset_hold_obj(dp,
 398                     dsl_dataset_phys(ds_next)->ds_next_snap_obj,
 399                     FTAG, &ds_nextnext));
 400                 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
 401                     dsl_dataset_phys(ds)->ds_prev_snap_txg,
 402                     dsl_dataset_phys(ds)->ds_creation_txg,
 403                     &used, &comp, &uncomp);
 404                 dsl_dataset_phys(ds_next)->ds_unique_bytes += used;
 405                 dsl_dataset_rele(ds_nextnext, FTAG);
 406                 ASSERT3P(ds_next->ds_prev, ==, NULL);
 407
 408                 /* Collapse range in this head. */
 409                 dsl_dataset_t *hds;
 410                 VERIFY0(dsl_dataset_hold_obj(dp,
 411                     dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &hds));
 412                 dsl_deadlist_remove_key(&hds->ds_deadlist,
 413                     dsl_dataset_phys(ds)->ds_creation_txg, tx);
 414                 if (dsl_dataset_remap_deadlist_exists(hds)) {
 415                         dsl_deadlist_remove_key(&hds->ds_remap_deadlist,
 416                             dsl_dataset_phys(ds)->ds_creation_txg, tx);
 417                 }
 418                 dsl_dataset_rele(hds, FTAG);
 419
 420         } else {
 421                 ASSERT3P(ds_next->ds_prev, ==, ds);
 422                 dsl_dataset_rele(ds_next->ds_prev, ds_next);
 423                 ds_next->ds_prev = NULL;
 424                 if (ds_prev) {
 425                         VERIFY0(dsl_dataset_hold_obj(dp,
 426                             dsl_dataset_phys(ds)->ds_prev_snap_obj,
 427                             ds_next, &ds_next->ds_prev));
 428                 }
 429
 430                 dsl_dataset_recalc_head_uniq(ds_next);
 431
 432                 /*
 433                  * Reduce the amount of our unconsumed refreservation
 434                  * being charged to our parent by the amount of
 435                  * new unique data we have gained.
 436                  */
 437                 if (old_unique < ds_next->ds_reserved) {
 438                         int64_t mrsdelta;
 439                         uint64_t new_unique =
 440                             dsl_dataset_phys(ds_next)->ds_unique_bytes;
 441
 442                         ASSERT(old_unique <= new_unique);
 443                         mrsdelta = MIN(new_unique - old_unique,
 444                             ds_next->ds_reserved - old_unique);
 445                         dsl_dir_diduse_space(ds->ds_dir,
 446                             DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
 447                 }
 448         }
 449         dsl_dataset_rele(ds_next, FTAG);
 450
 451         /*
 452          * This must be done after the dsl_traverse(), because it will
 453          * re-open the objset.
 454          */
 455         if (ds->ds_objset) {
 456                 dmu_objset_evict(ds->ds_objset);
 457                 ds->ds_objset = NULL;
 458         }
 459
 460         /* remove from snapshot namespace */
 461         dsl_dataset_t *ds_head;
 462         ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0);
 463         VERIFY0(dsl_dataset_hold_obj(dp,
 464             dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &ds_head));
 465         VERIFY0(dsl_dataset_get_snapname(ds));
 466 #ifdef ZFS_DEBUG
 467         {
 468                 uint64_t val;
 469                 int err;
 470
 471                 err = dsl_dataset_snap_lookup(ds_head,
 472                     ds->ds_snapname, &val);
 473                 ASSERT0(err);
 474                 ASSERT3U(val, ==, obj);
 475         }
 476 #endif
 477         VERIFY0(dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx, B_TRUE));
 478         dsl_dataset_rele(ds_head, FTAG);
 479
 480         if (ds_prev != NULL)
 481                 dsl_dataset_rele(ds_prev, FTAG);
 482
 483         spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
 484
 485         if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
 486                 ASSERTV(uint64_t count);
 487                 ASSERT0(zap_count(mos,
 488                     dsl_dataset_phys(ds)->ds_next_clones_obj, &count) &&
 489                     count == 0);
 490                 VERIFY0(dmu_object_free(mos,
 491                     dsl_dataset_phys(ds)->ds_next_clones_obj, tx));
 492         }
 493         if (dsl_dataset_phys(ds)->ds_props_obj != 0)
 494                 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_props_obj,
 495                     tx));
 496         if (dsl_dataset_phys(ds)->ds_userrefs_obj != 0)
 497                 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_userrefs_obj,
 498                     tx));
 499         dsl_dir_rele(ds->ds_dir, ds);
 500         ds->ds_dir = NULL;
 501         dmu_object_free_zapified(mos, obj, tx);
 502 }
 503
 504 void
 505 dsl_destroy_snapshot_sync(void *arg, dmu_tx_t *tx)
 506 {
 507         dsl_destroy_snapshot_arg_t *ddsa = arg;
 508         const char *dsname = ddsa->ddsa_name;
 509         boolean_t defer = ddsa->ddsa_defer;
 510
 511         dsl_pool_t *dp = dmu_tx_pool(tx);
 512         dsl_dataset_t *ds;
 513
 514         int error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
 515         if (error == ENOENT)
 516                 return;
 517         ASSERT0(error);
 518         dsl_destroy_snapshot_sync_impl(ds, defer, tx);
 519         zvol_remove_minors(dp->dp_spa, dsname, B_TRUE);
 520         dsl_dataset_rele(ds, FTAG);
 521 }
 522
 523 /*
 524  * The semantics of this function are described in the comment above
 525  * lzc_destroy_snaps().  To summarize:
 526  *
 527  * The snapshots must all be in the same pool.
 528  *
 529  * Snapshots that don't exist will be silently ignored (considered to be
 530  * "already deleted").
 531  *
 532  * On success, all snaps will be destroyed and this will return 0.
 533  * On failure, no snaps will be destroyed, the errlist will be filled in,
 534  * and this will return an errno.
 535  */
 536 int
 537 dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t defer,
 538     nvlist_t *errlist)
 539 {
 540         if (nvlist_next_nvpair(snaps, NULL) == NULL)
 541                 return (0);
 542
 543         /*
 544          * lzc_destroy_snaps() is documented to take an nvlist whose
 545          * values "don't matter".  We need to convert that nvlist to
 546          * one that we know can be converted to LUA. We also don't
 547          * care about any duplicate entries because the nvlist will
 548          * be converted to a LUA table which should take care of this.
 549          */
 550         nvlist_t *snaps_normalized;
 551         VERIFY0(nvlist_alloc(&snaps_normalized, 0, KM_SLEEP));
 552         for (nvpair_t *pair = nvlist_next_nvpair(snaps, NULL);
 553             pair != NULL; pair = nvlist_next_nvpair(snaps, pair)) {
 554                 fnvlist_add_boolean_value(snaps_normalized,
 555                     nvpair_name(pair), B_TRUE);
 556         }
 557
 558         nvlist_t *arg;
 559         VERIFY0(nvlist_alloc(&arg, 0, KM_SLEEP));
 560         fnvlist_add_nvlist(arg, "snaps", snaps_normalized);
 561         fnvlist_free(snaps_normalized);
 562         fnvlist_add_boolean_value(arg, "defer", defer);
 563
 564         nvlist_t *wrapper;
 565         VERIFY0(nvlist_alloc(&wrapper, 0, KM_SLEEP));
 566         fnvlist_add_nvlist(wrapper, ZCP_ARG_ARGLIST, arg);
 567         fnvlist_free(arg);
 568
 569         const char *program =
 570             "arg = ...\n"
 571             "snaps = arg['snaps']\n"
 572             "defer = arg['defer']\n"
 573             "errors = { }\n"
 574             "has_errors = false\n"
 575             "for snap, v in pairs(snaps) do\n"
 576             "    errno = zfs.check.destroy{snap, defer=defer}\n"
 577             "    zfs.debug('snap: ' .. snap .. ' errno: ' .. errno)\n"
 578             "    if errno == ENOENT then\n"
 579             "        snaps[snap] = nil\n"
 580             "    elseif errno ~= 0 then\n"
 581             "        errors[snap] = errno\n"
 582             "        has_errors = true\n"
 583             "    end\n"
 584             "end\n"
 585             "if has_errors then\n"
 586             "    return errors\n"
 587             "end\n"
 588             "for snap, v in pairs(snaps) do\n"
 589             "    errno = zfs.sync.destroy{snap, defer=defer}\n"
 590             "    assert(errno == 0)\n"
 591             "end\n"
 592             "return { }\n";
 593
 594         nvlist_t *result = fnvlist_alloc();
 595         int error = zcp_eval(nvpair_name(nvlist_next_nvpair(snaps, NULL)),
 596             program,
 597             B_TRUE,
 598             0,
 599             zfs_lua_max_memlimit,
 600             nvlist_next_nvpair(wrapper, NULL), result);
 601         if (error != 0) {
 602                 char *errorstr = NULL;
 603                 (void) nvlist_lookup_string(result, ZCP_RET_ERROR, &errorstr);
 604                 if (errorstr != NULL) {
 605                         zfs_dbgmsg(errorstr);
 606                 }
 607                 return (error);
 608         }
 609         fnvlist_free(wrapper);
 610
 611         /*
 612          * lzc_destroy_snaps() is documented to fill the errlist with
 613          * int32 values, so we need to covert the int64 values that are
 614          * returned from LUA.
 615          */
 616         int rv = 0;
 617         nvlist_t *errlist_raw = fnvlist_lookup_nvlist(result, ZCP_RET_RETURN);
 618         for (nvpair_t *pair = nvlist_next_nvpair(errlist_raw, NULL);
 619             pair != NULL; pair = nvlist_next_nvpair(errlist_raw, pair)) {
 620                 int32_t val = (int32_t)fnvpair_value_int64(pair);
 621                 if (rv == 0)
 622                         rv = val;
 623                 fnvlist_add_int32(errlist, nvpair_name(pair), val);
 624         }
 625         fnvlist_free(result);
 626         return (rv);
 627 }
 628
 629 int
 630 dsl_destroy_snapshot(const char *name, boolean_t defer)
 631 {
 632         int error;
 633         nvlist_t *nvl = fnvlist_alloc();
 634         nvlist_t *errlist = fnvlist_alloc();
 635
 636         fnvlist_add_boolean(nvl, name);
 637         error = dsl_destroy_snapshots_nvl(nvl, defer, errlist);
 638         fnvlist_free(errlist);
 639         fnvlist_free(nvl);
 640         return (error);
 641 }
 642
 643 struct killarg {
 644         dsl_dataset_t *ds;
 645         dmu_tx_t *tx;
 646 };
 647
 648 /* ARGSUSED */
 649 static int
 650 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 651     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
 652 {
 653         struct killarg *ka = arg;
 654         dmu_tx_t *tx = ka->tx;
 655
 656         if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
 657                 return (0);
 658
 659         if (zb->zb_level == ZB_ZIL_LEVEL) {
 660                 ASSERT(zilog != NULL);
 661                 /*
 662                  * It's a block in the intent log.  It has no
 663                  * accounting, so just free it.
 664                  */
 665                 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
 666         } else {
 667                 ASSERT(zilog == NULL);
 668                 ASSERT3U(bp->blk_birth, >,
 669                     dsl_dataset_phys(ka->ds)->ds_prev_snap_txg);
 670                 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
 671         }
 672
 673         return (0);
 674 }
 675
 676 static void
 677 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
 678 {
 679         struct killarg ka;
 680
 681         /*
 682          * Free everything that we point to (that's born after
 683          * the previous snapshot, if we are a clone)
 684          *
 685          * NB: this should be very quick, because we already
 686          * freed all the objects in open context.
 687          */
 688         ka.ds = ds;
 689         ka.tx = tx;
 690         VERIFY0(traverse_dataset(ds,
 691             dsl_dataset_phys(ds)->ds_prev_snap_txg, TRAVERSE_POST |
 692             TRAVERSE_NO_DECRYPT, kill_blkptr, &ka));
 693         ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
 694             dsl_dataset_phys(ds)->ds_unique_bytes == 0);
 695 }
 696
 697 int
 698 dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
 699 {
 700         int error;
 701         uint64_t count;
 702         objset_t *mos;
 703
 704         ASSERT(!ds->ds_is_snapshot);
 705         if (ds->ds_is_snapshot)
 706                 return (SET_ERROR(EINVAL));
 707
 708         if (refcount_count(&ds->ds_longholds) != expected_holds)
 709                 return (SET_ERROR(EBUSY));
 710
 711         mos = ds->ds_dir->dd_pool->dp_meta_objset;
 712
 713         /*
 714          * Can't delete a head dataset if there are snapshots of it.
 715          * (Except if the only snapshots are from the branch we cloned
 716          * from.)
 717          */
 718         if (ds->ds_prev != NULL &&
 719             dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object)
 720                 return (SET_ERROR(EBUSY));
 721
 722         /*
 723          * Can't delete if there are children of this fs.
 724          */
 725         error = zap_count(mos,
 726             dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, &count);
 727         if (error != 0)
 728                 return (error);
 729         if (count != 0)
 730                 return (SET_ERROR(EEXIST));
 731
 732         if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev) &&
 733             dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
 734             ds->ds_prev->ds_userrefs == 0) {
 735                 /* We need to remove the origin snapshot as well. */
 736                 if (!refcount_is_zero(&ds->ds_prev->ds_longholds))
 737                         return (SET_ERROR(EBUSY));
 738         }
 739         return (0);
 740 }
 741
 742 int
 743 dsl_destroy_head_check(void *arg, dmu_tx_t *tx)
 744 {
 745         dsl_destroy_head_arg_t *ddha = arg;
 746         dsl_pool_t *dp = dmu_tx_pool(tx);
 747         dsl_dataset_t *ds;
 748         int error;
 749
 750         error = dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds);
 751         if (error != 0)
 752                 return (error);
 753
 754         error = dsl_destroy_head_check_impl(ds, 0);
 755         dsl_dataset_rele(ds, FTAG);
 756         return (error);
 757 }
 758
 759 static void
 760 dsl_dir_destroy_sync(uint64_t ddobj, dmu_tx_t *tx)
 761 {
 762         dsl_dir_t *dd;
 763         dsl_pool_t *dp = dmu_tx_pool(tx);
 764         objset_t *mos = dp->dp_meta_objset;
 765         dd_used_t t;
 766
 767         ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx)->dp_config_rwlock));
 768
 769         VERIFY0(dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd));
 770
 771         ASSERT0(dsl_dir_phys(dd)->dd_head_dataset_obj);
 772
 773         /*
 774          * Decrement the filesystem count for all parent filesystems.
 775          *
 776          * When we receive an incremental stream into a filesystem that already
 777          * exists, a temporary clone is created.  We never count this temporary
 778          * clone, whose name begins with a '%'.
 779          */
 780         if (dd->dd_myname[0] != '%' && dd->dd_parent != NULL)
 781                 dsl_fs_ss_count_adjust(dd->dd_parent, -1,
 782                     DD_FIELD_FILESYSTEM_COUNT, tx);
 783
 784         /*
 785          * Remove our reservation. The impl() routine avoids setting the
 786          * actual property, which would require the (already destroyed) ds.
 787          */
 788         dsl_dir_set_reservation_sync_impl(dd, 0, tx);
 789
 790         ASSERT0(dsl_dir_phys(dd)->dd_used_bytes);
 791         ASSERT0(dsl_dir_phys(dd)->dd_reserved);
 792         for (t = 0; t < DD_USED_NUM; t++)
 793                 ASSERT0(dsl_dir_phys(dd)->dd_used_breakdown[t]);
 794
 795         if (dd->dd_crypto_obj != 0) {
 796                 dsl_crypto_key_destroy_sync(dd->dd_crypto_obj, tx);
 797                 (void) spa_keystore_unload_wkey_impl(dp->dp_spa, dd->dd_object);
 798         }
 799
 800         VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_child_dir_zapobj, tx));
 801         VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_props_zapobj, tx));
 802         VERIFY0(dsl_deleg_destroy(mos, dsl_dir_phys(dd)->dd_deleg_zapobj, tx));
 803         VERIFY0(zap_remove(mos,
 804             dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
 805             dd->dd_myname, tx));
 806
 807         dsl_dir_rele(dd, FTAG);
 808         dmu_object_free_zapified(mos, ddobj, tx);
 809 }
 810
 811 void
 812 dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
 813 {
 814         dsl_pool_t *dp = dmu_tx_pool(tx);
 815         objset_t *mos = dp->dp_meta_objset;
 816         uint64_t obj, ddobj, prevobj = 0;
 817         boolean_t rmorigin;
 818
 819         ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
 820         ASSERT(ds->ds_prev == NULL ||
 821             dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj != ds->ds_object);
 822         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
 823         ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
 824         rrw_exit(&ds->ds_bp_rwlock, FTAG);
 825         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
 826
 827         /* We need to log before removing it from the namespace. */
 828         spa_history_log_internal_ds(ds, "destroy", tx, "");
 829
 830         rmorigin = (dsl_dir_is_clone(ds->ds_dir) &&
 831             DS_IS_DEFER_DESTROY(ds->ds_prev) &&
 832             dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
 833             ds->ds_prev->ds_userrefs == 0);
 834
 835         /* Remove our reservation. */
 836         if (ds->ds_reserved != 0) {
 837                 dsl_dataset_set_refreservation_sync_impl(ds,
 838                     (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
 839                     0, tx);
 840                 ASSERT0(ds->ds_reserved);
 841         }
 842
 843         obj = ds->ds_object;
 844
 845         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 846                 if (ds->ds_feature_inuse[f]) {
 847                         dsl_dataset_deactivate_feature(obj, f, tx);
 848                         ds->ds_feature_inuse[f] = B_FALSE;
 849                 }
 850         }
 851
 852         dsl_scan_ds_destroyed(ds, tx);
 853
 854         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 855                 /* This is a clone */
 856                 ASSERT(ds->ds_prev != NULL);
 857                 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj, !=,
 858                     obj);
 859                 ASSERT0(dsl_dataset_phys(ds)->ds_next_snap_obj);
 860
 861                 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 862                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj != 0) {
 863                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
 864                             obj, tx);
 865                 }
 866
 867                 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_num_children, >, 1);
 868                 dsl_dataset_phys(ds->ds_prev)->ds_num_children--;
 869         }
 870
 871         /*
 872          * Destroy the deadlist.  Unless it's a clone, the
 873          * deadlist should be empty since the dataset has no snapshots.
 874          * (If it's a clone, it's safe to ignore the deadlist contents
 875          * since they are still referenced by the origin snapshot.)
 876          */
 877         dsl_deadlist_close(&ds->ds_deadlist);
 878         dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
 879         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 880         dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
 881
 882         if (dsl_dataset_remap_deadlist_exists(ds))
 883                 dsl_dataset_destroy_remap_deadlist(ds, tx);
 884
 885         objset_t *os;
 886         VERIFY0(dmu_objset_from_ds(ds, &os));
 887
 888         if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) {
 889                 old_synchronous_dataset_destroy(ds, tx);
 890         } else {
 891                 /*
 892                  * Move the bptree into the pool's list of trees to
 893                  * clean up and update space accounting information.
 894                  */
 895                 uint64_t used, comp, uncomp;
 896
 897                 zil_destroy_sync(dmu_objset_zil(os), tx);
 898
 899                 if (!spa_feature_is_active(dp->dp_spa,
 900                     SPA_FEATURE_ASYNC_DESTROY)) {
 901                         dsl_scan_t *scn = dp->dp_scan;
 902                         spa_feature_incr(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY,
 903                             tx);
 904                         dp->dp_bptree_obj = bptree_alloc(mos, tx);
 905                         VERIFY0(zap_add(mos,
 906                             DMU_POOL_DIRECTORY_OBJECT,
 907                             DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
 908                             &dp->dp_bptree_obj, tx));
 909                         ASSERT(!scn->scn_async_destroying);
 910                         scn->scn_async_destroying = B_TRUE;
 911                 }
 912
 913                 used = dsl_dir_phys(ds->ds_dir)->dd_used_bytes;
 914                 comp = dsl_dir_phys(ds->ds_dir)->dd_compressed_bytes;
 915                 uncomp = dsl_dir_phys(ds->ds_dir)->dd_uncompressed_bytes;
 916
 917                 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
 918                     dsl_dataset_phys(ds)->ds_unique_bytes == used);
 919
 920                 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
 921                 bptree_add(mos, dp->dp_bptree_obj,
 922                     &dsl_dataset_phys(ds)->ds_bp,
 923                     dsl_dataset_phys(ds)->ds_prev_snap_txg,
 924                     used, comp, uncomp, tx);
 925                 rrw_exit(&ds->ds_bp_rwlock, FTAG);
 926                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 927                     -used, -comp, -uncomp, tx);
 928                 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
 929                     used, comp, uncomp, tx);
 930         }
 931
 932         if (ds->ds_prev != NULL) {
 933                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 934                         VERIFY0(zap_remove_int(mos,
 935                             dsl_dir_phys(ds->ds_prev->ds_dir)->dd_clones,
 936                             ds->ds_object, tx));
 937                 }
 938                 prevobj = ds->ds_prev->ds_object;
 939                 dsl_dataset_rele(ds->ds_prev, ds);
 940                 ds->ds_prev = NULL;
 941         }
 942
 943         /*
 944          * This must be done after the dsl_traverse(), because it will
 945          * re-open the objset.
 946          */
 947         if (ds->ds_objset) {
 948                 dmu_objset_evict(ds->ds_objset);
 949                 ds->ds_objset = NULL;
 950         }
 951
 952         /* Erase the link in the dir */
 953         dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
 954         dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj = 0;
 955         ddobj = ds->ds_dir->dd_object;
 956         ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0);
 957         VERIFY0(zap_destroy(mos,
 958             dsl_dataset_phys(ds)->ds_snapnames_zapobj, tx));
 959
 960         if (ds->ds_bookmarks != 0) {
 961                 VERIFY0(zap_destroy(mos, ds->ds_bookmarks, tx));
 962                 spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
 963         }
 964
 965         spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
 966
 967         ASSERT0(dsl_dataset_phys(ds)->ds_next_clones_obj);
 968         ASSERT0(dsl_dataset_phys(ds)->ds_props_obj);
 969         ASSERT0(dsl_dataset_phys(ds)->ds_userrefs_obj);
 970         dsl_dir_rele(ds->ds_dir, ds);
 971         ds->ds_dir = NULL;
 972         dmu_object_free_zapified(mos, obj, tx);
 973
 974         dsl_dir_destroy_sync(ddobj, tx);
 975
 976         if (rmorigin) {
 977                 dsl_dataset_t *prev;
 978                 VERIFY0(dsl_dataset_hold_obj(dp, prevobj, FTAG, &prev));
 979                 dsl_destroy_snapshot_sync_impl(prev, B_FALSE, tx);
 980                 dsl_dataset_rele(prev, FTAG);
 981         }
 982 }
 983
 984 void
 985 dsl_destroy_head_sync(void *arg, dmu_tx_t *tx)
 986 {
 987         dsl_destroy_head_arg_t *ddha = arg;
 988         dsl_pool_t *dp = dmu_tx_pool(tx);
 989         dsl_dataset_t *ds;
 990
 991         VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
 992         dsl_destroy_head_sync_impl(ds, tx);
 993         zvol_remove_minors(dp->dp_spa, ddha->ddha_name, B_TRUE);
 994         dsl_dataset_rele(ds, FTAG);
 995 }
 996
 997 static void
 998 dsl_destroy_head_begin_sync(void *arg, dmu_tx_t *tx)
 999 {
1000         dsl_destroy_head_arg_t *ddha = arg;
1001         dsl_pool_t *dp = dmu_tx_pool(tx);
1002         dsl_dataset_t *ds;
1003
1004         VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
1005
1006         /* Mark it as inconsistent on-disk, in case we crash */
1007         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1008         dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT;
1009
1010         spa_history_log_internal_ds(ds, "destroy begin", tx, "");
1011         dsl_dataset_rele(ds, FTAG);
1012 }
1013
1014 int
1015 dsl_destroy_head(const char *name)
1016 {
1017         dsl_destroy_head_arg_t ddha;
1018         int error;
1019         spa_t *spa;
1020         boolean_t isenabled;
1021
1022 #ifdef _KERNEL
1023         zfs_destroy_unmount_origin(name);
1024 #endif
1025
1026         error = spa_open(name, &spa, FTAG);
1027         if (error != 0)
1028                 return (error);
1029         isenabled = spa_feature_is_enabled(spa, SPA_FEATURE_ASYNC_DESTROY);
1030         spa_close(spa, FTAG);
1031
1032         ddha.ddha_name = name;
1033
1034         if (!isenabled) {
1035                 objset_t *os;
1036
1037                 error = dsl_sync_task(name, dsl_destroy_head_check,
1038                     dsl_destroy_head_begin_sync, &ddha,
1039                     0, ZFS_SPACE_CHECK_DESTROY);
1040                 if (error != 0)
1041                         return (error);
1042
1043                 /*
1044                  * Head deletion is processed in one txg on old pools;
1045                  * remove the objects from open context so that the txg sync
1046                  * is not too long.
1047                  */
1048                 error = dmu_objset_own(name, DMU_OST_ANY, B_FALSE, B_FALSE,
1049                     FTAG, &os);
1050                 if (error == 0) {
1051                         uint64_t prev_snap_txg =
1052                             dsl_dataset_phys(dmu_objset_ds(os))->
1053                             ds_prev_snap_txg;
1054                         for (uint64_t obj = 0; error == 0;
1055                             error = dmu_object_next(os, &obj, FALSE,
1056                             prev_snap_txg))
1057                                 (void) dmu_free_long_object(os, obj);
1058                         /* sync out all frees */
1059                         txg_wait_synced(dmu_objset_pool(os), 0);
1060                         dmu_objset_disown(os, B_FALSE, FTAG);
1061                 }
1062         }
1063
1064         return (dsl_sync_task(name, dsl_destroy_head_check,
1065             dsl_destroy_head_sync, &ddha, 0, ZFS_SPACE_CHECK_DESTROY));
1066 }
1067
1068 /*
1069  * Note, this function is used as the callback for dmu_objset_find().  We
1070  * always return 0 so that we will continue to find and process
1071  * inconsistent datasets, even if we encounter an error trying to
1072  * process one of them.
1073  */
1074 /* ARGSUSED */
1075 int
1076 dsl_destroy_inconsistent(const char *dsname, void *arg)
1077 {
1078         objset_t *os;
1079
1080         if (dmu_objset_hold(dsname, FTAG, &os) == 0) {
1081                 boolean_t need_destroy = DS_IS_INCONSISTENT(dmu_objset_ds(os));
1082
1083                 /*
1084                  * If the dataset is inconsistent because a resumable receive
1085                  * has failed, then do not destroy it.
1086                  */
1087                 if (dsl_dataset_has_resume_receive_state(dmu_objset_ds(os)))
1088                         need_destroy = B_FALSE;
1089
1090                 dmu_objset_rele(os, FTAG);
1091                 if (need_destroy)
1092                         (void) dsl_destroy_head(dsname);
1093         }
1094         return (0);
1095 }
1096
1097
1098 #if defined(_KERNEL)
1099 EXPORT_SYMBOL(dsl_destroy_head);
1100 EXPORT_SYMBOL(dsl_destroy_head_sync_impl);
1101 EXPORT_SYMBOL(dsl_dataset_user_hold_check_one);
1102 EXPORT_SYMBOL(dsl_destroy_snapshot_sync_impl);
1103 EXPORT_SYMBOL(dsl_destroy_inconsistent);
1104 EXPORT_SYMBOL(dsl_dataset_user_release_tmp);
1105 EXPORT_SYMBOL(dsl_destroy_head_check_impl);
1106 #endif