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 https://opensource.org/licenses/CDDL-1.0.
  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, 2018 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/dsl_bookmark.h>
  35 #include <sys/dmu_tx.h>
  36 #include <sys/dsl_pool.h>
  37 #include <sys/dsl_dir.h>
  38 #include <sys/dmu_traverse.h>
  39 #include <sys/dsl_scan.h>
  40 #include <sys/dmu_objset.h>
  41 #include <sys/zap.h>
  42 #include <sys/zfeature.h>
  43 #include <sys/zfs_ioctl.h>
  44 #include <sys/dsl_deleg.h>
  45 #include <sys/dmu_impl.h>
  46 #include <sys/zvol.h>
  47 #include <sys/zcp.h>
  48 #include <sys/dsl_deadlist.h>
  49 #include <sys/zthr.h>
  50 #include <sys/spa_impl.h>
  51
  52 extern int zfs_snapshot_history_enabled;
  53
  54 int
  55 dsl_destroy_snapshot_check_impl(dsl_dataset_t *ds, boolean_t defer)
  56 {
  57         if (!ds->ds_is_snapshot)
  58                 return (SET_ERROR(EINVAL));
  59
  60         if (dsl_dataset_long_held(ds))
  61                 return (SET_ERROR(EBUSY));
  62
  63         /*
  64          * Only allow deferred destroy on pools that support it.
  65          * NOTE: deferred destroy is only supported on snapshots.
  66          */
  67         if (defer) {
  68                 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
  69                     SPA_VERSION_USERREFS)
  70                         return (SET_ERROR(ENOTSUP));
  71                 return (0);
  72         }
  73
  74         /*
  75          * If this snapshot has an elevated user reference count,
  76          * we can't destroy it yet.
  77          */
  78         if (ds->ds_userrefs > 0)
  79                 return (SET_ERROR(EBUSY));
  80
  81         /*
  82          * Can't delete a branch point.
  83          */
  84         if (dsl_dataset_phys(ds)->ds_num_children > 1)
  85                 return (SET_ERROR(EEXIST));
  86
  87         return (0);
  88 }
  89
  90 int
  91 dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx)
  92 {
  93         dsl_destroy_snapshot_arg_t *ddsa = arg;
  94         const char *dsname = ddsa->ddsa_name;
  95         boolean_t defer = ddsa->ddsa_defer;
  96
  97         dsl_pool_t *dp = dmu_tx_pool(tx);
  98         int error = 0;
  99         dsl_dataset_t *ds;
 100
 101         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
 102
 103         /*
 104          * If the snapshot does not exist, silently ignore it, and
 105          * dsl_destroy_snapshot_sync() will be a no-op
 106          * (it's "already destroyed").
 107          */
 108         if (error == ENOENT)
 109                 return (0);
 110
 111         if (error == 0) {
 112                 error = dsl_destroy_snapshot_check_impl(ds, defer);
 113                 dsl_dataset_rele(ds, FTAG);
 114         }
 115
 116         return (error);
 117 }
 118
 119 struct process_old_arg {
 120         dsl_dataset_t *ds;
 121         dsl_dataset_t *ds_prev;
 122         boolean_t after_branch_point;
 123         zio_t *pio;
 124         uint64_t used, comp, uncomp;
 125 };
 126
 127 static int
 128 process_old_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
 129 {
 130         struct process_old_arg *poa = arg;
 131         dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
 132
 133         ASSERT(!BP_IS_HOLE(bp));
 134
 135         if (bp->blk_birth <= dsl_dataset_phys(poa->ds)->ds_prev_snap_txg) {
 136                 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, bp_freed, tx);
 137                 if (poa->ds_prev && !poa->after_branch_point &&
 138                     bp->blk_birth >
 139                     dsl_dataset_phys(poa->ds_prev)->ds_prev_snap_txg) {
 140                         dsl_dataset_phys(poa->ds_prev)->ds_unique_bytes +=
 141                             bp_get_dsize_sync(dp->dp_spa, bp);
 142                 }
 143         } else {
 144                 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
 145                 poa->comp += BP_GET_PSIZE(bp);
 146                 poa->uncomp += BP_GET_UCSIZE(bp);
 147                 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
 148         }
 149         return (0);
 150 }
 151
 152 static void
 153 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
 154     dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
 155 {
 156         struct process_old_arg poa = { 0 };
 157         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 158         objset_t *mos = dp->dp_meta_objset;
 159         uint64_t deadlist_obj;
 160
 161         ASSERT(ds->ds_deadlist.dl_oldfmt);
 162         ASSERT(ds_next->ds_deadlist.dl_oldfmt);
 163
 164         poa.ds = ds;
 165         poa.ds_prev = ds_prev;
 166         poa.after_branch_point = after_branch_point;
 167         poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
 168         VERIFY0(bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
 169             process_old_cb, &poa, tx));
 170         VERIFY0(zio_wait(poa.pio));
 171         ASSERT3U(poa.used, ==, dsl_dataset_phys(ds)->ds_unique_bytes);
 172
 173         /* change snapused */
 174         dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
 175             -poa.used, -poa.comp, -poa.uncomp, tx);
 176
 177         /* swap next's deadlist to our deadlist */
 178         dsl_deadlist_close(&ds->ds_deadlist);
 179         dsl_deadlist_close(&ds_next->ds_deadlist);
 180         deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
 181         dsl_dataset_phys(ds)->ds_deadlist_obj =
 182             dsl_dataset_phys(ds_next)->ds_deadlist_obj;
 183         dsl_dataset_phys(ds_next)->ds_deadlist_obj = deadlist_obj;
 184         dsl_deadlist_open(&ds->ds_deadlist, mos,
 185             dsl_dataset_phys(ds)->ds_deadlist_obj);
 186         dsl_deadlist_open(&ds_next->ds_deadlist, mos,
 187             dsl_dataset_phys(ds_next)->ds_deadlist_obj);
 188 }
 189
 190 typedef struct remaining_clones_key {
 191         dsl_dataset_t *rck_clone;
 192         list_node_t rck_node;
 193 } remaining_clones_key_t;
 194
 195 static remaining_clones_key_t *
 196 rck_alloc(dsl_dataset_t *clone)
 197 {
 198         remaining_clones_key_t *rck = kmem_alloc(sizeof (*rck), KM_SLEEP);
 199         rck->rck_clone = clone;
 200         return (rck);
 201 }
 202
 203 static void
 204 dsl_dir_remove_clones_key_impl(dsl_dir_t *dd, uint64_t mintxg, dmu_tx_t *tx,
 205     list_t *stack, const void *tag)
 206 {
 207         objset_t *mos = dd->dd_pool->dp_meta_objset;
 208
 209         /*
 210          * If it is the old version, dd_clones doesn't exist so we can't
 211          * find the clones, but dsl_deadlist_remove_key() is a no-op so it
 212          * doesn't matter.
 213          */
 214         if (dsl_dir_phys(dd)->dd_clones == 0)
 215                 return;
 216
 217         zap_cursor_t *zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
 218         zap_attribute_t *za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
 219
 220         for (zap_cursor_init(zc, mos, dsl_dir_phys(dd)->dd_clones);
 221             zap_cursor_retrieve(zc, za) == 0;
 222             zap_cursor_advance(zc)) {
 223                 dsl_dataset_t *clone;
 224
 225                 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
 226                     za->za_first_integer, tag, &clone));
 227
 228                 if (clone->ds_dir->dd_origin_txg > mintxg) {
 229                         dsl_deadlist_remove_key(&clone->ds_deadlist,
 230                             mintxg, tx);
 231
 232                         if (dsl_dataset_remap_deadlist_exists(clone)) {
 233                                 dsl_deadlist_remove_key(
 234                                     &clone->ds_remap_deadlist, mintxg, tx);
 235                         }
 236
 237                         list_insert_head(stack, rck_alloc(clone));
 238                 } else {
 239                         dsl_dataset_rele(clone, tag);
 240                 }
 241         }
 242         zap_cursor_fini(zc);
 243
 244         kmem_free(za, sizeof (zap_attribute_t));
 245         kmem_free(zc, sizeof (zap_cursor_t));
 246 }
 247
 248 void
 249 dsl_dir_remove_clones_key(dsl_dir_t *top_dd, uint64_t mintxg, dmu_tx_t *tx)
 250 {
 251         list_t stack;
 252
 253         list_create(&stack, sizeof (remaining_clones_key_t),
 254             offsetof(remaining_clones_key_t, rck_node));
 255
 256         dsl_dir_remove_clones_key_impl(top_dd, mintxg, tx, &stack, FTAG);
 257         for (remaining_clones_key_t *rck = list_remove_head(&stack);
 258             rck != NULL; rck = list_remove_head(&stack)) {
 259                 dsl_dataset_t *clone = rck->rck_clone;
 260                 dsl_dir_t *clone_dir = clone->ds_dir;
 261
 262                 kmem_free(rck, sizeof (*rck));
 263
 264                 dsl_dir_remove_clones_key_impl(clone_dir, mintxg, tx,
 265                     &stack, FTAG);
 266                 dsl_dataset_rele(clone, FTAG);
 267         }
 268
 269         list_destroy(&stack);
 270 }
 271
 272 static void
 273 dsl_destroy_snapshot_handle_remaps(dsl_dataset_t *ds, dsl_dataset_t *ds_next,
 274     dmu_tx_t *tx)
 275 {
 276         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 277
 278         /* Move blocks to be obsoleted to pool's obsolete list. */
 279         if (dsl_dataset_remap_deadlist_exists(ds_next)) {
 280                 if (!bpobj_is_open(&dp->dp_obsolete_bpobj))
 281                         dsl_pool_create_obsolete_bpobj(dp, tx);
 282
 283                 dsl_deadlist_move_bpobj(&ds_next->ds_remap_deadlist,
 284                     &dp->dp_obsolete_bpobj,
 285                     dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
 286         }
 287
 288         /* Merge our deadlist into next's and free it. */
 289         if (dsl_dataset_remap_deadlist_exists(ds)) {
 290                 uint64_t remap_deadlist_object =
 291                     dsl_dataset_get_remap_deadlist_object(ds);
 292                 ASSERT(remap_deadlist_object != 0);
 293
 294                 mutex_enter(&ds_next->ds_remap_deadlist_lock);
 295                 if (!dsl_dataset_remap_deadlist_exists(ds_next))
 296                         dsl_dataset_create_remap_deadlist(ds_next, tx);
 297                 mutex_exit(&ds_next->ds_remap_deadlist_lock);
 298
 299                 dsl_deadlist_merge(&ds_next->ds_remap_deadlist,
 300                     remap_deadlist_object, tx);
 301                 dsl_dataset_destroy_remap_deadlist(ds, tx);
 302         }
 303 }
 304
 305 void
 306 dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
 307 {
 308         int after_branch_point = FALSE;
 309         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 310         objset_t *mos = dp->dp_meta_objset;
 311         dsl_dataset_t *ds_prev = NULL;
 312         uint64_t obj;
 313
 314         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
 315         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
 316         ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
 317         rrw_exit(&ds->ds_bp_rwlock, FTAG);
 318         ASSERT(zfs_refcount_is_zero(&ds->ds_longholds));
 319
 320         if (defer &&
 321             (ds->ds_userrefs > 0 ||
 322             dsl_dataset_phys(ds)->ds_num_children > 1)) {
 323                 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
 324                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
 325                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_DEFER_DESTROY;
 326                 if (zfs_snapshot_history_enabled) {
 327                         spa_history_log_internal_ds(ds, "defer_destroy", tx,
 328                             " ");
 329                 }
 330                 return;
 331         }
 332
 333         ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
 334
 335         if (zfs_snapshot_history_enabled) {
 336                 /* We need to log before removing it from the namespace. */
 337                 spa_history_log_internal_ds(ds, "destroy", tx, " ");
 338         }
 339
 340         dsl_scan_ds_destroyed(ds, tx);
 341
 342         obj = ds->ds_object;
 343
 344         boolean_t book_exists = dsl_bookmark_ds_destroyed(ds, tx);
 345
 346         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 347                 if (dsl_dataset_feature_is_active(ds, f))
 348                         dsl_dataset_deactivate_feature(ds, f, tx);
 349         }
 350         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 351                 ASSERT3P(ds->ds_prev, ==, NULL);
 352                 VERIFY0(dsl_dataset_hold_obj(dp,
 353                     dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &ds_prev));
 354                 after_branch_point =
 355                     (dsl_dataset_phys(ds_prev)->ds_next_snap_obj != obj);
 356
 357                 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
 358                 if (after_branch_point &&
 359                     dsl_dataset_phys(ds_prev)->ds_next_clones_obj != 0) {
 360                         dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
 361                         if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
 362                                 VERIFY0(zap_add_int(mos,
 363                                     dsl_dataset_phys(ds_prev)->
 364                                     ds_next_clones_obj,
 365                                     dsl_dataset_phys(ds)->ds_next_snap_obj,
 366                                     tx));
 367                         }
 368                 }
 369                 if (!after_branch_point) {
 370                         dsl_dataset_phys(ds_prev)->ds_next_snap_obj =
 371                             dsl_dataset_phys(ds)->ds_next_snap_obj;
 372                 }
 373         }
 374
 375         dsl_dataset_t *ds_next;
 376         uint64_t old_unique;
 377         uint64_t used = 0, comp = 0, uncomp = 0;
 378
 379         VERIFY0(dsl_dataset_hold_obj(dp,
 380             dsl_dataset_phys(ds)->ds_next_snap_obj, FTAG, &ds_next));
 381         ASSERT3U(dsl_dataset_phys(ds_next)->ds_prev_snap_obj, ==, obj);
 382
 383         old_unique = dsl_dataset_phys(ds_next)->ds_unique_bytes;
 384
 385         dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
 386         dsl_dataset_phys(ds_next)->ds_prev_snap_obj =
 387             dsl_dataset_phys(ds)->ds_prev_snap_obj;
 388         dsl_dataset_phys(ds_next)->ds_prev_snap_txg =
 389             dsl_dataset_phys(ds)->ds_prev_snap_txg;
 390         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
 391             ds_prev ? dsl_dataset_phys(ds_prev)->ds_creation_txg : 0);
 392
 393         if (ds_next->ds_deadlist.dl_oldfmt) {
 394                 process_old_deadlist(ds, ds_prev, ds_next,
 395                     after_branch_point, tx);
 396         } else {
 397                 /* Adjust prev's unique space. */
 398                 if (ds_prev && !after_branch_point) {
 399                         dsl_deadlist_space_range(&ds_next->ds_deadlist,
 400                             dsl_dataset_phys(ds_prev)->ds_prev_snap_txg,
 401                             dsl_dataset_phys(ds)->ds_prev_snap_txg,
 402                             &used, &comp, &uncomp);
 403                         dsl_dataset_phys(ds_prev)->ds_unique_bytes += used;
 404                 }
 405
 406                 /* Adjust snapused. */
 407                 dsl_deadlist_space_range(&ds_next->ds_deadlist,
 408                     dsl_dataset_phys(ds)->ds_prev_snap_txg, UINT64_MAX,
 409                     &used, &comp, &uncomp);
 410                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
 411                     -used, -comp, -uncomp, tx);
 412
 413                 /* Move blocks to be freed to pool's free list. */
 414                 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
 415                     &dp->dp_free_bpobj, dsl_dataset_phys(ds)->ds_prev_snap_txg,
 416                     tx);
 417                 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
 418                     DD_USED_HEAD, used, comp, uncomp, tx);
 419
 420                 /* Merge our deadlist into next's and free it. */
 421                 dsl_deadlist_merge(&ds_next->ds_deadlist,
 422                     dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
 423
 424                 /*
 425                  * We are done with the deadlist tree (generated/used
 426                  * by dsl_deadlist_move_bpobj() and dsl_deadlist_merge()).
 427                  * Discard it to save memory.
 428                  */
 429                 dsl_deadlist_discard_tree(&ds_next->ds_deadlist);
 430         }
 431
 432         dsl_deadlist_close(&ds->ds_deadlist);
 433         dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
 434         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 435         dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
 436
 437         dsl_destroy_snapshot_handle_remaps(ds, ds_next, tx);
 438
 439         if (!book_exists) {
 440                 /* Collapse range in clone heads */
 441                 dsl_dir_remove_clones_key(ds->ds_dir,
 442                     dsl_dataset_phys(ds)->ds_creation_txg, tx);
 443         }
 444
 445         if (ds_next->ds_is_snapshot) {
 446                 dsl_dataset_t *ds_nextnext;
 447
 448                 /*
 449                  * Update next's unique to include blocks which
 450                  * were previously shared by only this snapshot
 451                  * and it.  Those blocks will be born after the
 452                  * prev snap and before this snap, and will have
 453                  * died after the next snap and before the one
 454                  * after that (ie. be on the snap after next's
 455                  * deadlist).
 456                  */
 457                 VERIFY0(dsl_dataset_hold_obj(dp,
 458                     dsl_dataset_phys(ds_next)->ds_next_snap_obj,
 459                     FTAG, &ds_nextnext));
 460                 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
 461                     dsl_dataset_phys(ds)->ds_prev_snap_txg,
 462                     dsl_dataset_phys(ds)->ds_creation_txg,
 463                     &used, &comp, &uncomp);
 464                 dsl_dataset_phys(ds_next)->ds_unique_bytes += used;
 465                 dsl_dataset_rele(ds_nextnext, FTAG);
 466                 ASSERT3P(ds_next->ds_prev, ==, NULL);
 467
 468                 /* Collapse range in this head. */
 469                 dsl_dataset_t *hds;
 470                 VERIFY0(dsl_dataset_hold_obj(dp,
 471                     dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
 472                     FTAG, &hds));
 473                 if (!book_exists) {
 474                         /* Collapse range in this head. */
 475                         dsl_deadlist_remove_key(&hds->ds_deadlist,
 476                             dsl_dataset_phys(ds)->ds_creation_txg, tx);
 477                 }
 478                 if (dsl_dataset_remap_deadlist_exists(hds)) {
 479                         dsl_deadlist_remove_key(&hds->ds_remap_deadlist,
 480                             dsl_dataset_phys(ds)->ds_creation_txg, tx);
 481                 }
 482                 dsl_dataset_rele(hds, FTAG);
 483
 484         } else {
 485                 ASSERT3P(ds_next->ds_prev, ==, ds);
 486                 dsl_dataset_rele(ds_next->ds_prev, ds_next);
 487                 ds_next->ds_prev = NULL;
 488                 if (ds_prev) {
 489                         VERIFY0(dsl_dataset_hold_obj(dp,
 490                             dsl_dataset_phys(ds)->ds_prev_snap_obj,
 491                             ds_next, &ds_next->ds_prev));
 492                 }
 493
 494                 dsl_dataset_recalc_head_uniq(ds_next);
 495
 496                 /*
 497                  * Reduce the amount of our unconsumed refreservation
 498                  * being charged to our parent by the amount of
 499                  * new unique data we have gained.
 500                  */
 501                 if (old_unique < ds_next->ds_reserved) {
 502                         int64_t mrsdelta;
 503                         uint64_t new_unique =
 504                             dsl_dataset_phys(ds_next)->ds_unique_bytes;
 505
 506                         ASSERT(old_unique <= new_unique);
 507                         mrsdelta = MIN(new_unique - old_unique,
 508                             ds_next->ds_reserved - old_unique);
 509                         dsl_dir_diduse_space(ds->ds_dir,
 510                             DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
 511                 }
 512         }
 513         dsl_dataset_rele(ds_next, FTAG);
 514
 515         /*
 516          * This must be done after the dsl_traverse(), because it will
 517          * re-open the objset.
 518          */
 519         if (ds->ds_objset) {
 520                 dmu_objset_evict(ds->ds_objset);
 521                 ds->ds_objset = NULL;
 522         }
 523
 524         /* remove from snapshot namespace */
 525         dsl_dataset_t *ds_head;
 526         ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0);
 527         VERIFY0(dsl_dataset_hold_obj(dp,
 528             dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &ds_head));
 529         VERIFY0(dsl_dataset_get_snapname(ds));
 530 #ifdef ZFS_DEBUG
 531         {
 532                 uint64_t val;
 533                 int err;
 534
 535                 err = dsl_dataset_snap_lookup(ds_head,
 536                     ds->ds_snapname, &val);
 537                 ASSERT0(err);
 538                 ASSERT3U(val, ==, obj);
 539         }
 540 #endif
 541         VERIFY0(dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx, B_TRUE));
 542         dsl_dataset_rele(ds_head, FTAG);
 543
 544         if (ds_prev != NULL)
 545                 dsl_dataset_rele(ds_prev, FTAG);
 546
 547         spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
 548
 549         if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
 550                 uint64_t count __maybe_unused;
 551                 ASSERT0(zap_count(mos,
 552                     dsl_dataset_phys(ds)->ds_next_clones_obj, &count) &&
 553                     count == 0);
 554                 VERIFY0(dmu_object_free(mos,
 555                     dsl_dataset_phys(ds)->ds_next_clones_obj, tx));
 556         }
 557         if (dsl_dataset_phys(ds)->ds_props_obj != 0)
 558                 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_props_obj,
 559                     tx));
 560         if (dsl_dataset_phys(ds)->ds_userrefs_obj != 0)
 561                 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_userrefs_obj,
 562                     tx));
 563         dsl_dir_rele(ds->ds_dir, ds);
 564         ds->ds_dir = NULL;
 565         dmu_object_free_zapified(mos, obj, tx);
 566 }
 567
 568 void
 569 dsl_destroy_snapshot_sync(void *arg, dmu_tx_t *tx)
 570 {
 571         dsl_destroy_snapshot_arg_t *ddsa = arg;
 572         const char *dsname = ddsa->ddsa_name;
 573         boolean_t defer = ddsa->ddsa_defer;
 574
 575         dsl_pool_t *dp = dmu_tx_pool(tx);
 576         dsl_dataset_t *ds;
 577
 578         int error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
 579         if (error == ENOENT)
 580                 return;
 581         ASSERT0(error);
 582         dsl_destroy_snapshot_sync_impl(ds, defer, tx);
 583         zvol_remove_minors(dp->dp_spa, dsname, B_TRUE);
 584         dsl_dataset_rele(ds, FTAG);
 585 }
 586
 587 /*
 588  * The semantics of this function are described in the comment above
 589  * lzc_destroy_snaps().  To summarize:
 590  *
 591  * The snapshots must all be in the same pool.
 592  *
 593  * Snapshots that don't exist will be silently ignored (considered to be
 594  * "already deleted").
 595  *
 596  * On success, all snaps will be destroyed and this will return 0.
 597  * On failure, no snaps will be destroyed, the errlist will be filled in,
 598  * and this will return an errno.
 599  */
 600 int
 601 dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t defer,
 602     nvlist_t *errlist)
 603 {
 604         if (nvlist_next_nvpair(snaps, NULL) == NULL)
 605                 return (0);
 606
 607         /*
 608          * lzc_destroy_snaps() is documented to take an nvlist whose
 609          * values "don't matter".  We need to convert that nvlist to
 610          * one that we know can be converted to LUA.
 611          */
 612         nvlist_t *snaps_normalized = fnvlist_alloc();
 613         for (nvpair_t *pair = nvlist_next_nvpair(snaps, NULL);
 614             pair != NULL; pair = nvlist_next_nvpair(snaps, pair)) {
 615                 fnvlist_add_boolean_value(snaps_normalized,
 616                     nvpair_name(pair), B_TRUE);
 617         }
 618
 619         nvlist_t *arg = fnvlist_alloc();
 620         fnvlist_add_nvlist(arg, "snaps", snaps_normalized);
 621         fnvlist_free(snaps_normalized);
 622         fnvlist_add_boolean_value(arg, "defer", defer);
 623
 624         nvlist_t *wrapper = fnvlist_alloc();
 625         fnvlist_add_nvlist(wrapper, ZCP_ARG_ARGLIST, arg);
 626         fnvlist_free(arg);
 627
 628         const char *program =
 629             "arg = ...\n"
 630             "snaps = arg['snaps']\n"
 631             "defer = arg['defer']\n"
 632             "errors = { }\n"
 633             "has_errors = false\n"
 634             "for snap, v in pairs(snaps) do\n"
 635             "    errno = zfs.check.destroy{snap, defer=defer}\n"
 636             "    zfs.debug('snap: ' .. snap .. ' errno: ' .. errno)\n"
 637             "    if errno == ENOENT then\n"
 638             "        snaps[snap] = nil\n"
 639             "    elseif errno ~= 0 then\n"
 640             "        errors[snap] = errno\n"
 641             "        has_errors = true\n"
 642             "    end\n"
 643             "end\n"
 644             "if has_errors then\n"
 645             "    return errors\n"
 646             "end\n"
 647             "for snap, v in pairs(snaps) do\n"
 648             "    errno = zfs.sync.destroy{snap, defer=defer}\n"
 649             "    assert(errno == 0)\n"
 650             "end\n"
 651             "return { }\n";
 652
 653         nvlist_t *result = fnvlist_alloc();
 654         int error = zcp_eval(nvpair_name(nvlist_next_nvpair(snaps, NULL)),
 655             program,
 656             B_TRUE,
 657             0,
 658             zfs_lua_max_memlimit,
 659             fnvlist_lookup_nvpair(wrapper, ZCP_ARG_ARGLIST), result);
 660         if (error != 0) {
 661                 char *errorstr = NULL;
 662                 (void) nvlist_lookup_string(result, ZCP_RET_ERROR, &errorstr);
 663                 if (errorstr != NULL) {
 664                         zfs_dbgmsg("%s", errorstr);
 665                 }
 666                 fnvlist_free(wrapper);
 667                 fnvlist_free(result);
 668                 return (error);
 669         }
 670         fnvlist_free(wrapper);
 671
 672         /*
 673          * lzc_destroy_snaps() is documented to fill the errlist with
 674          * int32 values, so we need to convert the int64 values that are
 675          * returned from LUA.
 676          */
 677         int rv = 0;
 678         nvlist_t *errlist_raw = fnvlist_lookup_nvlist(result, ZCP_RET_RETURN);
 679         for (nvpair_t *pair = nvlist_next_nvpair(errlist_raw, NULL);
 680             pair != NULL; pair = nvlist_next_nvpair(errlist_raw, pair)) {
 681                 int32_t val = (int32_t)fnvpair_value_int64(pair);
 682                 if (rv == 0)
 683                         rv = val;
 684                 fnvlist_add_int32(errlist, nvpair_name(pair), val);
 685         }
 686         fnvlist_free(result);
 687         return (rv);
 688 }
 689
 690 int
 691 dsl_destroy_snapshot(const char *name, boolean_t defer)
 692 {
 693         int error;
 694         nvlist_t *nvl = fnvlist_alloc();
 695         nvlist_t *errlist = fnvlist_alloc();
 696
 697         fnvlist_add_boolean(nvl, name);
 698         error = dsl_destroy_snapshots_nvl(nvl, defer, errlist);
 699         fnvlist_free(errlist);
 700         fnvlist_free(nvl);
 701         return (error);
 702 }
 703
 704 struct killarg {
 705         dsl_dataset_t *ds;
 706         dmu_tx_t *tx;
 707 };
 708
 709 static int
 710 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 711     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
 712 {
 713         (void) spa, (void) dnp;
 714         struct killarg *ka = arg;
 715         dmu_tx_t *tx = ka->tx;
 716
 717         if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
 718             BP_IS_EMBEDDED(bp))
 719                 return (0);
 720
 721         if (zb->zb_level == ZB_ZIL_LEVEL) {
 722                 ASSERT(zilog != NULL);
 723                 /*
 724                  * It's a block in the intent log.  It has no
 725                  * accounting, so just free it.
 726                  */
 727                 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
 728         } else {
 729                 ASSERT(zilog == NULL);
 730                 ASSERT3U(bp->blk_birth, >,
 731                     dsl_dataset_phys(ka->ds)->ds_prev_snap_txg);
 732                 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
 733         }
 734
 735         return (0);
 736 }
 737
 738 static void
 739 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
 740 {
 741         struct killarg ka;
 742
 743         spa_history_log_internal_ds(ds, "destroy", tx,
 744             "(synchronous, mintxg=%llu)",
 745             (long long)dsl_dataset_phys(ds)->ds_prev_snap_txg);
 746
 747         /*
 748          * Free everything that we point to (that's born after
 749          * the previous snapshot, if we are a clone)
 750          *
 751          * NB: this should be very quick, because we already
 752          * freed all the objects in open context.
 753          */
 754         ka.ds = ds;
 755         ka.tx = tx;
 756         VERIFY0(traverse_dataset(ds,
 757             dsl_dataset_phys(ds)->ds_prev_snap_txg, TRAVERSE_POST |
 758             TRAVERSE_NO_DECRYPT, kill_blkptr, &ka));
 759         ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
 760             dsl_dataset_phys(ds)->ds_unique_bytes == 0);
 761 }
 762
 763 int
 764 dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
 765 {
 766         int error;
 767         uint64_t count;
 768         objset_t *mos;
 769
 770         ASSERT(!ds->ds_is_snapshot);
 771         if (ds->ds_is_snapshot)
 772                 return (SET_ERROR(EINVAL));
 773
 774         if (zfs_refcount_count(&ds->ds_longholds) != expected_holds)
 775                 return (SET_ERROR(EBUSY));
 776
 777         ASSERT0(ds->ds_dir->dd_activity_waiters);
 778
 779         mos = ds->ds_dir->dd_pool->dp_meta_objset;
 780
 781         /*
 782          * Can't delete a head dataset if there are snapshots of it.
 783          * (Except if the only snapshots are from the branch we cloned
 784          * from.)
 785          */
 786         if (ds->ds_prev != NULL &&
 787             dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object)
 788                 return (SET_ERROR(EBUSY));
 789
 790         /*
 791          * Can't delete if there are children of this fs.
 792          */
 793         error = zap_count(mos,
 794             dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, &count);
 795         if (error != 0)
 796                 return (error);
 797         if (count != 0)
 798                 return (SET_ERROR(EEXIST));
 799
 800         if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev) &&
 801             dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
 802             ds->ds_prev->ds_userrefs == 0) {
 803                 /* We need to remove the origin snapshot as well. */
 804                 if (!zfs_refcount_is_zero(&ds->ds_prev->ds_longholds))
 805                         return (SET_ERROR(EBUSY));
 806         }
 807         return (0);
 808 }
 809
 810 int
 811 dsl_destroy_head_check(void *arg, dmu_tx_t *tx)
 812 {
 813         dsl_destroy_head_arg_t *ddha = arg;
 814         dsl_pool_t *dp = dmu_tx_pool(tx);
 815         dsl_dataset_t *ds;
 816         int error;
 817
 818         error = dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds);
 819         if (error != 0)
 820                 return (error);
 821
 822         error = dsl_destroy_head_check_impl(ds, 0);
 823         dsl_dataset_rele(ds, FTAG);
 824         return (error);
 825 }
 826
 827 static void
 828 dsl_dir_destroy_sync(uint64_t ddobj, dmu_tx_t *tx)
 829 {
 830         dsl_dir_t *dd;
 831         dsl_pool_t *dp = dmu_tx_pool(tx);
 832         objset_t *mos = dp->dp_meta_objset;
 833         dd_used_t t;
 834
 835         ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx)->dp_config_rwlock));
 836
 837         VERIFY0(dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd));
 838
 839         ASSERT0(dsl_dir_phys(dd)->dd_head_dataset_obj);
 840
 841         /* Decrement the filesystem count for all parent filesystems. */
 842         if (dd->dd_parent != NULL)
 843                 dsl_fs_ss_count_adjust(dd->dd_parent, -1,
 844                     DD_FIELD_FILESYSTEM_COUNT, tx);
 845
 846         /*
 847          * Remove our reservation. The impl() routine avoids setting the
 848          * actual property, which would require the (already destroyed) ds.
 849          */
 850         dsl_dir_set_reservation_sync_impl(dd, 0, tx);
 851
 852         ASSERT0(dsl_dir_phys(dd)->dd_used_bytes);
 853         ASSERT0(dsl_dir_phys(dd)->dd_reserved);
 854         for (t = 0; t < DD_USED_NUM; t++)
 855                 ASSERT0(dsl_dir_phys(dd)->dd_used_breakdown[t]);
 856
 857         if (dd->dd_crypto_obj != 0) {
 858                 dsl_crypto_key_destroy_sync(dd->dd_crypto_obj, tx);
 859                 (void) spa_keystore_unload_wkey_impl(dp->dp_spa, dd->dd_object);
 860         }
 861
 862         VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_child_dir_zapobj, tx));
 863         VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_props_zapobj, tx));
 864         if (dsl_dir_phys(dd)->dd_clones != 0)
 865                 VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_clones, tx));
 866         VERIFY0(dsl_deleg_destroy(mos, dsl_dir_phys(dd)->dd_deleg_zapobj, tx));
 867         VERIFY0(zap_remove(mos,
 868             dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
 869             dd->dd_myname, tx));
 870
 871         dsl_dir_rele(dd, FTAG);
 872         dmu_object_free_zapified(mos, ddobj, tx);
 873 }
 874
 875 static void
 876 dsl_clone_destroy_assert(dsl_dir_t *dd)
 877 {
 878         uint64_t used, comp, uncomp;
 879
 880         ASSERT(dsl_dir_is_clone(dd));
 881         dsl_deadlist_space(&dd->dd_livelist, &used, &comp, &uncomp);
 882
 883         ASSERT3U(dsl_dir_phys(dd)->dd_used_bytes, ==, used);
 884         ASSERT3U(dsl_dir_phys(dd)->dd_compressed_bytes, ==, comp);
 885         /*
 886          * Greater than because we do not track embedded block pointers in
 887          * the livelist
 888          */
 889         ASSERT3U(dsl_dir_phys(dd)->dd_uncompressed_bytes, >=, uncomp);
 890
 891         ASSERT(list_is_empty(&dd->dd_pending_allocs.bpl_list));
 892         ASSERT(list_is_empty(&dd->dd_pending_frees.bpl_list));
 893 }
 894
 895 /*
 896  * Start the delete process for a clone. Free its zil, verify the space usage
 897  * and queue the blkptrs for deletion by adding the livelist to the pool-wide
 898  * delete queue.
 899  */
 900 static void
 901 dsl_async_clone_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
 902 {
 903         uint64_t zap_obj, to_delete, used, comp, uncomp;
 904         objset_t *os;
 905         dsl_dir_t *dd = ds->ds_dir;
 906         dsl_pool_t *dp = dmu_tx_pool(tx);
 907         objset_t *mos = dp->dp_meta_objset;
 908         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 909         VERIFY0(dmu_objset_from_ds(ds, &os));
 910
 911         uint64_t mintxg = 0;
 912         dsl_deadlist_entry_t *dle = dsl_deadlist_first(&dd->dd_livelist);
 913         if (dle != NULL)
 914                 mintxg = dle->dle_mintxg;
 915
 916         spa_history_log_internal_ds(ds, "destroy", tx,
 917             "(livelist, mintxg=%llu)", (long long)mintxg);
 918
 919         /* Check that the clone is in a correct state to be deleted */
 920         dsl_clone_destroy_assert(dd);
 921
 922         /* Destroy the zil */
 923         zil_destroy_sync(dmu_objset_zil(os), tx);
 924
 925         VERIFY0(zap_lookup(mos, dd->dd_object,
 926             DD_FIELD_LIVELIST, sizeof (uint64_t), 1, &to_delete));
 927         /* Initialize deleted_clones entry to track livelists to cleanup */
 928         int error = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
 929             DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
 930         if (error == ENOENT) {
 931                 zap_obj = zap_create(mos, DMU_OTN_ZAP_METADATA,
 932                     DMU_OT_NONE, 0, tx);
 933                 VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
 934                     DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1,
 935                     &(zap_obj), tx));
 936                 spa->spa_livelists_to_delete = zap_obj;
 937         } else if (error != 0) {
 938                 zfs_panic_recover("zfs: error %d was returned while looking "
 939                     "up DMU_POOL_DELETED_CLONES in the zap", error);
 940                 return;
 941         }
 942         VERIFY0(zap_add_int(mos, zap_obj, to_delete, tx));
 943
 944         /* Clone is no longer using space, now tracked by dp_free_dir */
 945         dsl_deadlist_space(&dd->dd_livelist, &used, &comp, &uncomp);
 946         dsl_dir_diduse_space(dd, DD_USED_HEAD,
 947             -used, -comp, -dsl_dir_phys(dd)->dd_uncompressed_bytes,
 948             tx);
 949         dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
 950             used, comp, uncomp, tx);
 951         dsl_dir_remove_livelist(dd, tx, B_FALSE);
 952         zthr_wakeup(spa->spa_livelist_delete_zthr);
 953 }
 954
 955 /*
 956  * Move the bptree into the pool's list of trees to clean up, update space
 957  * accounting information and destroy the zil.
 958  */
 959 static void
 960 dsl_async_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
 961 {
 962         uint64_t used, comp, uncomp;
 963         objset_t *os;
 964
 965         VERIFY0(dmu_objset_from_ds(ds, &os));
 966         dsl_pool_t *dp = dmu_tx_pool(tx);
 967         objset_t *mos = dp->dp_meta_objset;
 968
 969         spa_history_log_internal_ds(ds, "destroy", tx,
 970             "(bptree, mintxg=%llu)",
 971             (long long)dsl_dataset_phys(ds)->ds_prev_snap_txg);
 972
 973         zil_destroy_sync(dmu_objset_zil(os), tx);
 974
 975         if (!spa_feature_is_active(dp->dp_spa,
 976             SPA_FEATURE_ASYNC_DESTROY)) {
 977                 dsl_scan_t *scn = dp->dp_scan;
 978                 spa_feature_incr(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY,
 979                     tx);
 980                 dp->dp_bptree_obj = bptree_alloc(mos, tx);
 981                 VERIFY0(zap_add(mos,
 982                     DMU_POOL_DIRECTORY_OBJECT,
 983                     DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
 984                     &dp->dp_bptree_obj, tx));
 985                 ASSERT(!scn->scn_async_destroying);
 986                 scn->scn_async_destroying = B_TRUE;
 987         }
 988
 989         used = dsl_dir_phys(ds->ds_dir)->dd_used_bytes;
 990         comp = dsl_dir_phys(ds->ds_dir)->dd_compressed_bytes;
 991         uncomp = dsl_dir_phys(ds->ds_dir)->dd_uncompressed_bytes;
 992
 993         ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
 994             dsl_dataset_phys(ds)->ds_unique_bytes == used);
 995
 996         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
 997         bptree_add(mos, dp->dp_bptree_obj,
 998             &dsl_dataset_phys(ds)->ds_bp,
 999             dsl_dataset_phys(ds)->ds_prev_snap_txg,
1000             used, comp, uncomp, tx);
1001         rrw_exit(&ds->ds_bp_rwlock, FTAG);
1002         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
1003             -used, -comp, -uncomp, tx);
1004         dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1005             used, comp, uncomp, tx);
1006 }
1007
1008 void
1009 dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
1010 {
1011         dsl_pool_t *dp = dmu_tx_pool(tx);
1012         objset_t *mos = dp->dp_meta_objset;
1013         uint64_t obj, ddobj, prevobj = 0;
1014         boolean_t rmorigin;
1015
1016         ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
1017         ASSERT(ds->ds_prev == NULL ||
1018             dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj != ds->ds_object);
1019         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1020         ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
1021         rrw_exit(&ds->ds_bp_rwlock, FTAG);
1022         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1023
1024         dsl_dir_cancel_waiters(ds->ds_dir);
1025
1026         rmorigin = (dsl_dir_is_clone(ds->ds_dir) &&
1027             DS_IS_DEFER_DESTROY(ds->ds_prev) &&
1028             dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
1029             ds->ds_prev->ds_userrefs == 0);
1030
1031         /* Remove our reservation. */
1032         if (ds->ds_reserved != 0) {
1033                 dsl_dataset_set_refreservation_sync_impl(ds,
1034                     (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1035                     0, tx);
1036                 ASSERT0(ds->ds_reserved);
1037         }
1038
1039         obj = ds->ds_object;
1040
1041         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1042                 if (dsl_dataset_feature_is_active(ds, f))
1043                         dsl_dataset_deactivate_feature(ds, f, tx);
1044         }
1045
1046         dsl_scan_ds_destroyed(ds, tx);
1047
1048         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1049                 /* This is a clone */
1050                 ASSERT(ds->ds_prev != NULL);
1051                 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj, !=,
1052                     obj);
1053                 ASSERT0(dsl_dataset_phys(ds)->ds_next_snap_obj);
1054
1055                 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1056                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj != 0) {
1057                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
1058                             obj, tx);
1059                 }
1060
1061                 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_num_children, >, 1);
1062                 dsl_dataset_phys(ds->ds_prev)->ds_num_children--;
1063         }
1064
1065         /*
1066          * Destroy the deadlist. Unless it's a clone, the
1067          * deadlist should be empty since the dataset has no snapshots.
1068          * (If it's a clone, it's safe to ignore the deadlist contents
1069          * since they are still referenced by the origin snapshot.)
1070          */
1071         dsl_deadlist_close(&ds->ds_deadlist);
1072         dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
1073         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1074         dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
1075
1076         if (dsl_dataset_remap_deadlist_exists(ds))
1077                 dsl_dataset_destroy_remap_deadlist(ds, tx);
1078
1079         /*
1080          * Each destroy is responsible for both destroying (enqueuing
1081          * to be destroyed) the blkptrs comprising the dataset as well as
1082          * those belonging to the zil.
1083          */
1084         if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist)) {
1085                 dsl_async_clone_destroy(ds, tx);
1086         } else if (spa_feature_is_enabled(dp->dp_spa,
1087             SPA_FEATURE_ASYNC_DESTROY)) {
1088                 dsl_async_dataset_destroy(ds, tx);
1089         } else {
1090                 old_synchronous_dataset_destroy(ds, tx);
1091         }
1092
1093         if (ds->ds_prev != NULL) {
1094                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1095                         VERIFY0(zap_remove_int(mos,
1096                             dsl_dir_phys(ds->ds_prev->ds_dir)->dd_clones,
1097                             ds->ds_object, tx));
1098                 }
1099                 prevobj = ds->ds_prev->ds_object;
1100                 dsl_dataset_rele(ds->ds_prev, ds);
1101                 ds->ds_prev = NULL;
1102         }
1103
1104         /*
1105          * This must be done after the dsl_traverse(), because it will
1106          * re-open the objset.
1107          */
1108         if (ds->ds_objset) {
1109                 dmu_objset_evict(ds->ds_objset);
1110                 ds->ds_objset = NULL;
1111         }
1112
1113         /* Erase the link in the dir */
1114         dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1115         dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj = 0;
1116         ddobj = ds->ds_dir->dd_object;
1117         ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0);
1118         VERIFY0(zap_destroy(mos,
1119             dsl_dataset_phys(ds)->ds_snapnames_zapobj, tx));
1120
1121         if (ds->ds_bookmarks_obj != 0) {
1122                 void *cookie = NULL;
1123                 dsl_bookmark_node_t *dbn;
1124
1125                 while ((dbn = avl_destroy_nodes(&ds->ds_bookmarks, &cookie)) !=
1126                     NULL) {
1127                         if (dbn->dbn_phys.zbm_redaction_obj != 0) {
1128                                 VERIFY0(dmu_object_free(mos,
1129                                     dbn->dbn_phys.zbm_redaction_obj, tx));
1130                                 spa_feature_decr(dmu_objset_spa(mos),
1131                                     SPA_FEATURE_REDACTION_BOOKMARKS, tx);
1132                         }
1133                         if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN) {
1134                                 spa_feature_decr(dmu_objset_spa(mos),
1135                                     SPA_FEATURE_BOOKMARK_WRITTEN, tx);
1136                         }
1137                         spa_strfree(dbn->dbn_name);
1138                         mutex_destroy(&dbn->dbn_lock);
1139                         kmem_free(dbn, sizeof (*dbn));
1140                 }
1141                 avl_destroy(&ds->ds_bookmarks);
1142                 VERIFY0(zap_destroy(mos, ds->ds_bookmarks_obj, tx));
1143                 spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
1144         }
1145
1146         spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1147
1148         ASSERT0(dsl_dataset_phys(ds)->ds_next_clones_obj);
1149         ASSERT0(dsl_dataset_phys(ds)->ds_props_obj);
1150         ASSERT0(dsl_dataset_phys(ds)->ds_userrefs_obj);
1151         dsl_dir_rele(ds->ds_dir, ds);
1152         ds->ds_dir = NULL;
1153         dmu_object_free_zapified(mos, obj, tx);
1154
1155         dsl_dir_destroy_sync(ddobj, tx);
1156
1157         if (rmorigin) {
1158                 dsl_dataset_t *prev;
1159                 VERIFY0(dsl_dataset_hold_obj(dp, prevobj, FTAG, &prev));
1160                 dsl_destroy_snapshot_sync_impl(prev, B_FALSE, tx);
1161                 dsl_dataset_rele(prev, FTAG);
1162         }
1163         /* Delete errlog. */
1164         if (spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_HEAD_ERRLOG))
1165                 spa_delete_dataset_errlog(dp->dp_spa, ds->ds_object, tx);
1166 }
1167
1168 void
1169 dsl_destroy_head_sync(void *arg, dmu_tx_t *tx)
1170 {
1171         dsl_destroy_head_arg_t *ddha = arg;
1172         dsl_pool_t *dp = dmu_tx_pool(tx);
1173         dsl_dataset_t *ds;
1174
1175         VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
1176         dsl_destroy_head_sync_impl(ds, tx);
1177         zvol_remove_minors(dp->dp_spa, ddha->ddha_name, B_TRUE);
1178         dsl_dataset_rele(ds, FTAG);
1179 }
1180
1181 static void
1182 dsl_destroy_head_begin_sync(void *arg, dmu_tx_t *tx)
1183 {
1184         dsl_destroy_head_arg_t *ddha = arg;
1185         dsl_pool_t *dp = dmu_tx_pool(tx);
1186         dsl_dataset_t *ds;
1187
1188         VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
1189
1190         /* Mark it as inconsistent on-disk, in case we crash */
1191         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1192         dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT;
1193
1194         spa_history_log_internal_ds(ds, "destroy begin", tx, " ");
1195         dsl_dataset_rele(ds, FTAG);
1196 }
1197
1198 int
1199 dsl_destroy_head(const char *name)
1200 {
1201         dsl_destroy_head_arg_t ddha;
1202         int error;
1203         spa_t *spa;
1204         boolean_t isenabled;
1205
1206 #ifdef _KERNEL
1207         zfs_destroy_unmount_origin(name);
1208 #endif
1209
1210         error = spa_open(name, &spa, FTAG);
1211         if (error != 0)
1212                 return (error);
1213         isenabled = spa_feature_is_enabled(spa, SPA_FEATURE_ASYNC_DESTROY);
1214         spa_close(spa, FTAG);
1215
1216         ddha.ddha_name = name;
1217
1218         if (!isenabled) {
1219                 objset_t *os;
1220
1221                 error = dsl_sync_task(name, dsl_destroy_head_check,
1222                     dsl_destroy_head_begin_sync, &ddha,
1223                     0, ZFS_SPACE_CHECK_DESTROY);
1224                 if (error != 0)
1225                         return (error);
1226
1227                 /*
1228                  * Head deletion is processed in one txg on old pools;
1229                  * remove the objects from open context so that the txg sync
1230                  * is not too long. This optimization can only work for
1231                  * encrypted datasets if the wrapping key is loaded.
1232                  */
1233                 error = dmu_objset_own(name, DMU_OST_ANY, B_FALSE, B_TRUE,
1234                     FTAG, &os);
1235                 if (error == 0) {
1236                         uint64_t prev_snap_txg =
1237                             dsl_dataset_phys(dmu_objset_ds(os))->
1238                             ds_prev_snap_txg;
1239                         for (uint64_t obj = 0; error == 0;
1240                             error = dmu_object_next(os, &obj, FALSE,
1241                             prev_snap_txg))
1242                                 (void) dmu_free_long_object(os, obj);
1243                         /* sync out all frees */
1244                         txg_wait_synced(dmu_objset_pool(os), 0);
1245                         dmu_objset_disown(os, B_TRUE, FTAG);
1246                 }
1247         }
1248
1249         return (dsl_sync_task(name, dsl_destroy_head_check,
1250             dsl_destroy_head_sync, &ddha, 0, ZFS_SPACE_CHECK_DESTROY));
1251 }
1252
1253 /*
1254  * Note, this function is used as the callback for dmu_objset_find().  We
1255  * always return 0 so that we will continue to find and process
1256  * inconsistent datasets, even if we encounter an error trying to
1257  * process one of them.
1258  */
1259 int
1260 dsl_destroy_inconsistent(const char *dsname, void *arg)
1261 {
1262         (void) arg;
1263         objset_t *os;
1264
1265         if (dmu_objset_hold(dsname, FTAG, &os) == 0) {
1266                 boolean_t need_destroy = DS_IS_INCONSISTENT(dmu_objset_ds(os));
1267
1268                 /*
1269                  * If the dataset is inconsistent because a resumable receive
1270                  * has failed, then do not destroy it.
1271                  */
1272                 if (dsl_dataset_has_resume_receive_state(dmu_objset_ds(os)))
1273                         need_destroy = B_FALSE;
1274
1275                 dmu_objset_rele(os, FTAG);
1276                 if (need_destroy)
1277                         (void) dsl_destroy_head(dsname);
1278         }
1279         return (0);
1280 }
1281
1282
1283 #if defined(_KERNEL)
1284 EXPORT_SYMBOL(dsl_destroy_head);
1285 EXPORT_SYMBOL(dsl_destroy_head_sync_impl);
1286 EXPORT_SYMBOL(dsl_dataset_user_hold_check_one);
1287 EXPORT_SYMBOL(dsl_destroy_snapshot_sync_impl);
1288 EXPORT_SYMBOL(dsl_destroy_inconsistent);
1289 EXPORT_SYMBOL(dsl_dataset_user_release_tmp);
1290 EXPORT_SYMBOL(dsl_destroy_head_check_impl);
1291 #endif