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, 2016 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) 2014 Integros [integros.com]
  27  * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
  28  */
  29 
  30 #include <sys/autosnap.h>
  31 #include <sys/zfs_context.h>
  32 #include <sys/dsl_userhold.h>
  33 #include <sys/dsl_dataset.h>
  34 #include <sys/dsl_synctask.h>
  35 #include <sys/dsl_destroy.h>
  36 #include <sys/dmu_tx.h>
  37 #include <sys/dsl_pool.h>
  38 #include <sys/dsl_dir.h>
  39 #include <sys/dmu_traverse.h>
  40 #include <sys/dsl_scan.h>
  41 #include <sys/dmu_objset.h>
  42 #include <sys/zap.h>
  43 #include <sys/zfeature.h>
  44 #include <sys/zfs_ioctl.h>
  45 #include <sys/dsl_deleg.h>
  46 #include <sys/dmu_impl.h>
  47 #include <sys/wbc.h>
  48 #include <sys/zcp.h>
  49 
  50 int
  51 dsl_destroy_snapshot_check_impl(dsl_dataset_t *ds, boolean_t defer)
  52 {
  53         if (!ds->ds_is_snapshot)
  54                 return (SET_ERROR(EINVAL));
  55 
  56         if (dsl_dataset_long_held(ds))
  57                 return (SET_ERROR(EBUSY));
  58 
  59         /*
  60          * Only allow deferred destroy on pools that support it.
  61          * NOTE: deferred destroy is only supported on snapshots.
  62          */
  63         if (defer) {
  64                 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
  65                     SPA_VERSION_USERREFS)
  66                         return (SET_ERROR(ENOTSUP));
  67                 return (0);
  68         }
  69 
  70         /*
  71          * If this snapshot has an elevated user reference count,
  72          * we can't destroy it yet.
  73          */
  74         if (ds->ds_userrefs > 0)
  75                 return (SET_ERROR(EBUSY));
  76 
  77         /*
  78          * Can't delete a branch point.
  79          */
  80         if (dsl_dataset_phys(ds)->ds_num_children > 1)
  81                 return (SET_ERROR(EEXIST));
  82 
  83         return (0);
  84 }
  85 
  86 int
  87 dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx)
  88 {
  89         dsl_destroy_snapshot_arg_t *ddsa = arg;
  90         const char *dsname = ddsa->ddsa_name;
  91         boolean_t defer = ddsa->ddsa_defer;
  92 
  93         dsl_pool_t *dp = dmu_tx_pool(tx);
  94         int error = 0;
  95         dsl_dataset_t *ds;
  96 
  97         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
  98 
  99         /*
 100          * If the snapshot does not exist, silently ignore it, and
 101          * dsl_destroy_snapshot_sync() will be a no-op
 102          * (it's "already destroyed").
 103          */
 104         if (error == ENOENT)
 105                 return (0);
 106 
 107         if (error == 0) {
 108                 error = dsl_destroy_snapshot_check_impl(ds, defer);
 109                 dsl_dataset_rele(ds, FTAG);
 110         }
 111 
 112         return (error);
 113 }
 114 
 115 struct process_old_arg {
 116         dsl_dataset_t *ds;
 117         dsl_dataset_t *ds_prev;
 118         boolean_t after_branch_point;
 119         zio_t *pio;
 120         uint64_t used, comp, uncomp;
 121 };
 122 
 123 static int
 124 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
 125 {
 126         struct process_old_arg *poa = arg;
 127         dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
 128 
 129         ASSERT(!BP_IS_HOLE(bp));
 130 
 131         if (bp->blk_birth <= dsl_dataset_phys(poa->ds)->ds_prev_snap_txg) {
 132                 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
 133                 if (poa->ds_prev && !poa->after_branch_point &&
 134                     bp->blk_birth >
 135                     dsl_dataset_phys(poa->ds_prev)->ds_prev_snap_txg) {
 136                         dsl_dataset_phys(poa->ds_prev)->ds_unique_bytes +=
 137                             bp_get_dsize_sync(dp->dp_spa, bp);
 138                 }
 139         } else {
 140                 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
 141                 poa->comp += BP_GET_PSIZE(bp);
 142                 poa->uncomp += BP_GET_UCSIZE(bp);
 143                 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
 144         }
 145         return (0);
 146 }
 147 
 148 static void
 149 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
 150     dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
 151 {
 152         struct process_old_arg poa = { 0 };
 153         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 154         objset_t *mos = dp->dp_meta_objset;
 155         uint64_t deadlist_obj;
 156 
 157         ASSERT(ds->ds_deadlist.dl_oldfmt);
 158         ASSERT(ds_next->ds_deadlist.dl_oldfmt);
 159 
 160         poa.ds = ds;
 161         poa.ds_prev = ds_prev;
 162         poa.after_branch_point = after_branch_point;
 163         poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
 164         VERIFY0(bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
 165             process_old_cb, &poa, tx));
 166         VERIFY0(zio_wait(poa.pio));
 167         ASSERT3U(poa.used, ==, dsl_dataset_phys(ds)->ds_unique_bytes);
 168 
 169         /* change snapused */
 170         dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
 171             -poa.used, -poa.comp, -poa.uncomp, tx);
 172 
 173         /* swap next's deadlist to our deadlist */
 174         dsl_deadlist_close(&ds->ds_deadlist);
 175         dsl_deadlist_close(&ds_next->ds_deadlist);
 176         deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
 177         dsl_dataset_phys(ds)->ds_deadlist_obj =
 178             dsl_dataset_phys(ds_next)->ds_deadlist_obj;
 179         dsl_dataset_phys(ds_next)->ds_deadlist_obj = deadlist_obj;
 180         dsl_deadlist_open(&ds->ds_deadlist, mos,
 181             dsl_dataset_phys(ds)->ds_deadlist_obj);
 182         dsl_deadlist_open(&ds_next->ds_deadlist, mos,
 183             dsl_dataset_phys(ds_next)->ds_deadlist_obj);
 184 }
 185 
 186 static void
 187 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
 188 {
 189         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 190         zap_cursor_t *zc;
 191         zap_attribute_t *za;
 192 
 193         /*
 194          * If it is the old version, dd_clones doesn't exist so we can't
 195          * find the clones, but dsl_deadlist_remove_key() is a no-op so it
 196          * doesn't matter.
 197          */
 198         if (dsl_dir_phys(ds->ds_dir)->dd_clones == 0)
 199                 return;
 200         zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
 201         za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
 202 
 203         for (zap_cursor_init(zc, mos, dsl_dir_phys(ds->ds_dir)->dd_clones);
 204             zap_cursor_retrieve(zc, za) == 0;
 205             zap_cursor_advance(zc)) {
 206                 dsl_dataset_t *clone;
 207 
 208                 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
 209                     za->za_first_integer, FTAG, &clone));
 210                 if (clone->ds_dir->dd_origin_txg > mintxg) {
 211                         dsl_deadlist_remove_key(&clone->ds_deadlist,
 212                             mintxg, tx);
 213                         dsl_dataset_remove_clones_key(clone, mintxg, tx);
 214                 }
 215                 dsl_dataset_rele(clone, FTAG);
 216         }
 217         zap_cursor_fini(zc);
 218         kmem_free(zc, sizeof (zap_cursor_t));
 219         kmem_free(za, sizeof (zap_attribute_t));
 220 }
 221 
 222 void
 223 dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
 224 {
 225         int err;
 226         int after_branch_point = FALSE;
 227         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 228         spa_t *spa = dp->dp_spa;
 229         wbc_data_t *wbc_data = spa_get_wbc_data(spa);
 230         objset_t *mos = dp->dp_meta_objset;
 231         dsl_dataset_t *ds_prev = NULL;
 232         uint64_t obj;
 233 
 234         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
 235         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
 236         ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
 237         rrw_exit(&ds->ds_bp_rwlock, FTAG);
 238         ASSERT(refcount_is_zero(&ds->ds_longholds));
 239 
 240         /*
 241          * if an edge snapshot of WBC window is destroyed, the window must be
 242          * aborted
 243          */
 244         mutex_enter(&wbc_data->wbc_lock);
 245         if (dsl_dataset_phys(ds)->ds_creation_txg == wbc_data->wbc_finish_txg)
 246                 wbc_purge_window(spa, tx);
 247         mutex_exit(&wbc_data->wbc_lock);
 248 
 249         if (defer &&
 250             (ds->ds_userrefs > 0 ||
 251             dsl_dataset_phys(ds)->ds_num_children > 1)) {
 252                 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
 253                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
 254                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_DEFER_DESTROY;
 255                 spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
 256                 return;
 257         }
 258 
 259         ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
 260 
 261         /* We need to log before removing it from the namespace. */
 262         spa_history_log_internal_ds(ds, "destroy", tx, "");
 263 
 264         dsl_scan_ds_destroyed(ds, tx);
 265 
 266         obj = ds->ds_object;
 267 
 268         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 269                 if (ds->ds_feature_inuse[f]) {
 270                         dsl_dataset_deactivate_feature(obj, f, tx);
 271                         ds->ds_feature_inuse[f] = B_FALSE;
 272                 }
 273         }
 274         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 275                 ASSERT3P(ds->ds_prev, ==, NULL);
 276                 VERIFY0(dsl_dataset_hold_obj(dp,
 277                     dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &ds_prev));
 278                 after_branch_point =
 279                     (dsl_dataset_phys(ds_prev)->ds_next_snap_obj != obj);
 280 
 281                 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
 282                 if (after_branch_point &&
 283                     dsl_dataset_phys(ds_prev)->ds_next_clones_obj != 0) {
 284                         dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
 285                         if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
 286                                 VERIFY0(zap_add_int(mos,
 287                                     dsl_dataset_phys(ds_prev)->
 288                                     ds_next_clones_obj,
 289                                     dsl_dataset_phys(ds)->ds_next_snap_obj,
 290                                     tx));
 291                         }
 292                 }
 293                 if (!after_branch_point) {
 294                         dsl_dataset_phys(ds_prev)->ds_next_snap_obj =
 295                             dsl_dataset_phys(ds)->ds_next_snap_obj;
 296                 }
 297         }
 298 
 299         dsl_dataset_t *ds_next;
 300         uint64_t old_unique;
 301         uint64_t used = 0, comp = 0, uncomp = 0;
 302 
 303         VERIFY0(dsl_dataset_hold_obj(dp,
 304             dsl_dataset_phys(ds)->ds_next_snap_obj, FTAG, &ds_next));
 305         ASSERT3U(dsl_dataset_phys(ds_next)->ds_prev_snap_obj, ==, obj);
 306 
 307         old_unique = dsl_dataset_phys(ds_next)->ds_unique_bytes;
 308 
 309         dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
 310         dsl_dataset_phys(ds_next)->ds_prev_snap_obj =
 311             dsl_dataset_phys(ds)->ds_prev_snap_obj;
 312         dsl_dataset_phys(ds_next)->ds_prev_snap_txg =
 313             dsl_dataset_phys(ds)->ds_prev_snap_txg;
 314         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
 315             ds_prev ? dsl_dataset_phys(ds_prev)->ds_creation_txg : 0);
 316 
 317         if (ds_next->ds_deadlist.dl_oldfmt) {
 318                 process_old_deadlist(ds, ds_prev, ds_next,
 319                     after_branch_point, tx);
 320         } else {
 321                 /* Adjust prev's unique space. */
 322                 if (ds_prev && !after_branch_point) {
 323                         dsl_deadlist_space_range(&ds_next->ds_deadlist,
 324                             dsl_dataset_phys(ds_prev)->ds_prev_snap_txg,
 325                             dsl_dataset_phys(ds)->ds_prev_snap_txg,
 326                             &used, &comp, &uncomp);
 327                         dsl_dataset_phys(ds_prev)->ds_unique_bytes += used;
 328                 }
 329 
 330                 /* Adjust snapused. */
 331                 dsl_deadlist_space_range(&ds_next->ds_deadlist,
 332                     dsl_dataset_phys(ds)->ds_prev_snap_txg, UINT64_MAX,
 333                     &used, &comp, &uncomp);
 334                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
 335                     -used, -comp, -uncomp, tx);
 336 
 337                 /* Move blocks to be freed to pool's free list. */
 338                 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
 339                     &dp->dp_free_bpobj, dsl_dataset_phys(ds)->ds_prev_snap_txg,
 340                     tx);
 341                 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
 342                     DD_USED_HEAD, used, comp, uncomp, tx);
 343 
 344                 /* Merge our deadlist into next's and free it. */
 345                 dsl_deadlist_merge(&ds_next->ds_deadlist,
 346                     dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
 347         }
 348         dsl_deadlist_close(&ds->ds_deadlist);
 349         dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
 350         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 351         dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
 352 
 353         /* Collapse range in clone heads */
 354         dsl_dataset_remove_clones_key(ds,
 355             dsl_dataset_phys(ds)->ds_creation_txg, tx);
 356 
 357         if (ds_next->ds_is_snapshot) {
 358                 dsl_dataset_t *ds_nextnext;
 359 
 360                 /*
 361                  * Update next's unique to include blocks which
 362                  * were previously shared by only this snapshot
 363                  * and it.  Those blocks will be born after the
 364                  * prev snap and before this snap, and will have
 365                  * died after the next snap and before the one
 366                  * after that (ie. be on the snap after next's
 367                  * deadlist).
 368                  */
 369                 VERIFY0(dsl_dataset_hold_obj(dp,
 370                     dsl_dataset_phys(ds_next)->ds_next_snap_obj,
 371                     FTAG, &ds_nextnext));
 372                 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
 373                     dsl_dataset_phys(ds)->ds_prev_snap_txg,
 374                     dsl_dataset_phys(ds)->ds_creation_txg,
 375                     &used, &comp, &uncomp);
 376                 dsl_dataset_phys(ds_next)->ds_unique_bytes += used;
 377                 dsl_dataset_rele(ds_nextnext, FTAG);
 378                 ASSERT3P(ds_next->ds_prev, ==, NULL);
 379 
 380                 /* Collapse range in this head. */
 381                 dsl_dataset_t *hds;
 382                 VERIFY0(dsl_dataset_hold_obj(dp,
 383                     dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &hds));
 384                 dsl_deadlist_remove_key(&hds->ds_deadlist,
 385                     dsl_dataset_phys(ds)->ds_creation_txg, tx);
 386                 dsl_dataset_rele(hds, FTAG);
 387 
 388         } else {
 389                 ASSERT3P(ds_next->ds_prev, ==, ds);
 390                 dsl_dataset_rele(ds_next->ds_prev, ds_next);
 391                 ds_next->ds_prev = NULL;
 392                 if (ds_prev) {
 393                         VERIFY0(dsl_dataset_hold_obj(dp,
 394                             dsl_dataset_phys(ds)->ds_prev_snap_obj,
 395                             ds_next, &ds_next->ds_prev));
 396                 }
 397 
 398                 dsl_dataset_recalc_head_uniq(ds_next);
 399 
 400                 /*
 401                  * Reduce the amount of our unconsumed refreservation
 402                  * being charged to our parent by the amount of
 403                  * new unique data we have gained.
 404                  */
 405                 if (old_unique < ds_next->ds_reserved) {
 406                         int64_t mrsdelta;
 407                         uint64_t new_unique =
 408                             dsl_dataset_phys(ds_next)->ds_unique_bytes;
 409 
 410                         ASSERT(old_unique <= new_unique);
 411                         mrsdelta = MIN(new_unique - old_unique,
 412                             ds_next->ds_reserved - old_unique);
 413                         dsl_dir_diduse_space(ds->ds_dir,
 414                             DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
 415                 }
 416         }
 417         dsl_dataset_rele(ds_next, FTAG);
 418 
 419         /*
 420          * This must be done after the dsl_traverse(), because it will
 421          * re-open the objset.
 422          */
 423         if (ds->ds_objset) {
 424                 dmu_objset_evict(ds->ds_objset);
 425                 ds->ds_objset = NULL;
 426         }
 427 
 428         /* remove from snapshot namespace */
 429         dsl_dataset_t *ds_head;
 430         ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0);
 431         VERIFY0(dsl_dataset_hold_obj(dp,
 432             dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &ds_head));
 433         VERIFY0(dsl_dataset_get_snapname(ds));
 434 #ifdef ZFS_DEBUG
 435         {
 436                 uint64_t val;
 437 
 438                 err = dsl_dataset_snap_lookup(ds_head,
 439                     ds->ds_snapname, &val);
 440                 ASSERT0(err);
 441                 ASSERT3U(val, ==, obj);
 442         }
 443 #endif
 444         VERIFY0(dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx, B_TRUE));
 445         dsl_dataset_rele(ds_head, FTAG);
 446 
 447         if (ds_prev != NULL)
 448                 dsl_dataset_rele(ds_prev, FTAG);
 449 
 450         spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
 451 
 452         if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
 453                 uint64_t count;
 454                 ASSERT0(zap_count(mos,
 455                     dsl_dataset_phys(ds)->ds_next_clones_obj, &count) &&
 456                     count == 0);
 457                 VERIFY0(dmu_object_free(mos,
 458                     dsl_dataset_phys(ds)->ds_next_clones_obj, tx));
 459         }
 460         if (dsl_dataset_phys(ds)->ds_props_obj != 0)
 461                 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_props_obj,
 462                     tx));
 463         if (dsl_dataset_phys(ds)->ds_userrefs_obj != 0)
 464                 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_userrefs_obj,
 465                     tx));
 466         dsl_dir_rele(ds->ds_dir, ds);
 467         ds->ds_dir = NULL;
 468         dmu_object_free_zapified(mos, obj, tx);
 469 }
 470 
 471 void
 472 dsl_destroy_snapshot_sync(void *arg, dmu_tx_t *tx)
 473 {
 474         dsl_destroy_snapshot_arg_t *ddsa = arg;
 475         const char *dsname = ddsa->ddsa_name;
 476         boolean_t defer = ddsa->ddsa_defer;
 477 
 478         dsl_pool_t *dp = dmu_tx_pool(tx);
 479         dsl_dataset_t *ds;
 480 
 481         int error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
 482         if (error == ENOENT)
 483                 return;
 484         ASSERT0(error);
 485 
 486         if (autosnap_check_name(strchr(dsname, '@')))
 487                 autosnap_exempt_snapshot(dp->dp_spa, dsname);
 488 
 489         dsl_destroy_snapshot_sync_impl(ds, defer, tx);
 490         dsl_dataset_rele(ds, FTAG);
 491 }
 492 
 493 /*
 494  * The semantics of this function are described in the comment above
 495  * lzc_destroy_snaps().  To summarize:
 496  *
 497  * The snapshots must all be in the same pool.
 498  *
 499  * Snapshots that don't exist will be silently ignored (considered to be
 500  * "already deleted").
 501  *
 502  * On success, all snaps will be destroyed and this will return 0.
 503  * On failure, no snaps will be destroyed, the errlist will be filled in,
 504  * and this will return an errno.
 505  */
 506 int
 507 dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t defer,
 508     nvlist_t *errlist)
 509 {
 510         if (nvlist_next_nvpair(snaps, NULL) == NULL)
 511                 return (0);
 512 
 513         /*
 514          * lzc_destroy_snaps() is documented to take an nvlist whose
 515          * values "don't matter".  We need to convert that nvlist to
 516          * one that we know can be converted to LUA. We also don't
 517          * care about any duplicate entries because the nvlist will
 518          * be converted to a LUA table which should take care of this.
 519          */
 520         nvlist_t *snaps_normalized;
 521         VERIFY0(nvlist_alloc(&snaps_normalized, 0, KM_SLEEP));
 522         for (nvpair_t *pair = nvlist_next_nvpair(snaps, NULL);
 523             pair != NULL; pair = nvlist_next_nvpair(snaps, pair)) {
 524                 fnvlist_add_boolean_value(snaps_normalized,
 525                     nvpair_name(pair), B_TRUE);
 526         }
 527 
 528         nvlist_t *arg;
 529         VERIFY0(nvlist_alloc(&arg, 0, KM_SLEEP));
 530         fnvlist_add_nvlist(arg, "snaps", snaps_normalized);
 531         fnvlist_free(snaps_normalized);
 532         fnvlist_add_boolean_value(arg, "defer", defer);
 533 
 534         nvlist_t *wrapper;
 535         VERIFY0(nvlist_alloc(&wrapper, 0, KM_SLEEP));
 536         fnvlist_add_nvlist(wrapper, ZCP_ARG_ARGLIST, arg);
 537         fnvlist_free(arg);
 538 
 539         const char *program =
 540             "arg = ...\n"
 541             "snaps = arg['snaps']\n"
 542             "defer = arg['defer']\n"
 543             "errors = { }\n"
 544             "has_errors = false\n"
 545             "for snap, v in pairs(snaps) do\n"
 546             "    errno = zfs.check.destroy{snap, defer=defer}\n"
 547             "    zfs.debug('snap: ' .. snap .. ' errno: ' .. errno)\n"
 548             "    if errno == ENOENT then\n"
 549             "        snaps[snap] = nil\n"
 550             "    elseif errno ~= 0 then\n"
 551             "        errors[snap] = errno\n"
 552             "        has_errors = true\n"
 553             "    end\n"
 554             "end\n"
 555             "if has_errors then\n"
 556             "    return errors\n"
 557             "end\n"
 558             "for snap, v in pairs(snaps) do\n"
 559             "    errno = zfs.sync.destroy{snap, defer=defer}\n"
 560             "    assert(errno == 0)\n"
 561             "end\n"
 562             "return { }\n";
 563 
 564         nvlist_t *result = fnvlist_alloc();
 565         int error = zcp_eval(nvpair_name(nvlist_next_nvpair(snaps, NULL)),
 566             program,
 567             B_TRUE,
 568             0,
 569             zfs_lua_max_memlimit,
 570             nvlist_next_nvpair(wrapper, NULL), result);
 571         if (error != 0) {
 572                 char *errorstr = NULL;
 573                 (void) nvlist_lookup_string(result, ZCP_RET_ERROR, &errorstr);
 574                 if (errorstr != NULL) {
 575                         zfs_dbgmsg(errorstr);
 576                 }
 577                 return (error);
 578         }
 579         fnvlist_free(wrapper);
 580 
 581         /*
 582          * lzc_destroy_snaps() is documented to fill the errlist with
 583          * int32 values, so we need to covert the int64 values that are
 584          * returned from LUA.
 585          */
 586         int rv = 0;
 587         nvlist_t *errlist_raw = fnvlist_lookup_nvlist(result, ZCP_RET_RETURN);
 588         for (nvpair_t *pair = nvlist_next_nvpair(errlist_raw, NULL);
 589             pair != NULL; pair = nvlist_next_nvpair(errlist_raw, pair)) {
 590                 int32_t val = (int32_t)fnvpair_value_int64(pair);
 591                 if (rv == 0)
 592                         rv = val;
 593                 fnvlist_add_int32(errlist, nvpair_name(pair), val);
 594         }
 595         fnvlist_free(result);
 596         return (rv);
 597 }
 598 
 599 int
 600 dsl_destroy_snapshot(const char *name, boolean_t defer)
 601 {
 602         int error;
 603         nvlist_t *nvl = fnvlist_alloc();
 604         nvlist_t *errlist = fnvlist_alloc();
 605 
 606         fnvlist_add_boolean(nvl, name);
 607         error = dsl_destroy_snapshots_nvl(nvl, defer, errlist);
 608         fnvlist_free(errlist);
 609         fnvlist_free(nvl);
 610         return (error);
 611 }
 612 
 613 struct killarg {
 614         dsl_dataset_t *ds;
 615         dmu_tx_t *tx;
 616 };
 617 
 618 /* ARGSUSED */
 619 static int
 620 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 621     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
 622 {
 623         struct killarg *ka = arg;
 624         dmu_tx_t *tx = ka->tx;
 625 
 626         if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
 627                 return (0);
 628 
 629         if (zb->zb_level == ZB_ZIL_LEVEL) {
 630                 ASSERT(zilog != NULL);
 631                 /*
 632                  * It's a block in the intent log.  It has no
 633                  * accounting, so just free it.
 634                  */
 635                 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
 636         } else {
 637                 ASSERT(zilog == NULL);
 638                 ASSERT3U(bp->blk_birth, >,
 639                     dsl_dataset_phys(ka->ds)->ds_prev_snap_txg);
 640                 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
 641         }
 642 
 643         return (0);
 644 }
 645 
 646 static void
 647 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
 648 {
 649         struct killarg ka;
 650 
 651         /*
 652          * Free everything that we point to (that's born after
 653          * the previous snapshot, if we are a clone)
 654          *
 655          * NB: this should be very quick, because we already
 656          * freed all the objects in open context.
 657          */
 658         ka.ds = ds;
 659         ka.tx = tx;
 660         VERIFY0(traverse_dataset(ds,
 661             dsl_dataset_phys(ds)->ds_prev_snap_txg, TRAVERSE_POST,
 662             kill_blkptr, &ka));
 663         ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
 664             dsl_dataset_phys(ds)->ds_unique_bytes == 0);
 665 }
 666 
 667 int
 668 dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
 669 {
 670         int error;
 671         uint64_t count;
 672         objset_t *mos;
 673 
 674         ASSERT(!ds->ds_is_snapshot);
 675         if (ds->ds_is_snapshot)
 676                 return (SET_ERROR(EINVAL));
 677 
 678         if (refcount_count(&ds->ds_longholds) != expected_holds)
 679                 return (SET_ERROR(EBUSY));
 680 
 681         mos = ds->ds_dir->dd_pool->dp_meta_objset;
 682 
 683         /*
 684          * Can't delete a head dataset if there are snapshots of it.
 685          * (Except if the only snapshots are from the branch we cloned
 686          * from.)
 687          */
 688         if (ds->ds_prev != NULL &&
 689             dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object)
 690                 return (SET_ERROR(EBUSY));
 691 
 692         /*
 693          * Can't delete if there are children of this fs.
 694          */
 695         error = zap_count(mos,
 696             dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, &count);
 697         if (error != 0)
 698                 return (error);
 699         if (count != 0)
 700                 return (SET_ERROR(EEXIST));
 701 
 702         if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev) &&
 703             dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
 704             ds->ds_prev->ds_userrefs == 0) {
 705                 /* We need to remove the origin snapshot as well. */
 706                 if (!refcount_is_zero(&ds->ds_prev->ds_longholds))
 707                         return (SET_ERROR(EBUSY));
 708         }
 709         return (0);
 710 }
 711 
 712 int
 713 dsl_destroy_head_check(void *arg, dmu_tx_t *tx)
 714 {
 715         dsl_destroy_head_arg_t *ddha = arg;
 716         dsl_pool_t *dp = dmu_tx_pool(tx);
 717         dsl_dataset_t *ds;
 718         int error;
 719 
 720         error = dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds);
 721         if (error != 0)
 722                 return (error);
 723 
 724         error = dsl_destroy_head_check_impl(ds, 0);
 725         dsl_dataset_rele(ds, FTAG);
 726         return (error);
 727 }
 728 
 729 static void
 730 dsl_dir_destroy_sync(uint64_t ddobj, dmu_tx_t *tx)
 731 {
 732         dsl_dir_t *dd;
 733         dsl_pool_t *dp = dmu_tx_pool(tx);
 734         objset_t *mos = dp->dp_meta_objset;
 735         dd_used_t t;
 736 
 737         ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx)->dp_config_rwlock));
 738 
 739         VERIFY0(dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd));
 740 
 741         ASSERT0(dsl_dir_phys(dd)->dd_head_dataset_obj);
 742 
 743         /*
 744          * Decrement the filesystem count for all parent filesystems.
 745          *
 746          * When we receive an incremental stream into a filesystem that already
 747          * exists, a temporary clone is created.  We never count this temporary
 748          * clone, whose name begins with a '%'.
 749          */
 750         if (dd->dd_myname[0] != '%' && dd->dd_parent != NULL)
 751                 dsl_fs_ss_count_adjust(dd->dd_parent, -1,
 752                     DD_FIELD_FILESYSTEM_COUNT, tx);
 753 
 754         /*
 755          * Remove our reservation. The impl() routine avoids setting the
 756          * actual property, which would require the (already destroyed) ds.
 757          */
 758         dsl_dir_set_reservation_sync_impl(dd, 0, tx);
 759 
 760         ASSERT0(dsl_dir_phys(dd)->dd_used_bytes);
 761         ASSERT0(dsl_dir_phys(dd)->dd_reserved);
 762         for (t = 0; t < DD_USED_NUM; t++)
 763                 ASSERT0(dsl_dir_phys(dd)->dd_used_breakdown[t]);
 764 
 765         VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_child_dir_zapobj, tx));
 766         VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_props_zapobj, tx));
 767         VERIFY0(dsl_deleg_destroy(mos, dsl_dir_phys(dd)->dd_deleg_zapobj, tx));
 768         VERIFY0(zap_remove(mos,
 769             dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
 770             dd->dd_myname, tx));
 771 
 772         dsl_dir_rele(dd, FTAG);
 773         dmu_object_free_zapified(mos, ddobj, tx);
 774 }
 775 
 776 void
 777 dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
 778 {
 779         dsl_pool_t *dp = dmu_tx_pool(tx);
 780         objset_t *mos = dp->dp_meta_objset;
 781         uint64_t obj, ddobj, prevobj = 0;
 782         boolean_t rmorigin;
 783 
 784         ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
 785         ASSERT(ds->ds_prev == NULL ||
 786             dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj != ds->ds_object);
 787         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
 788         ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
 789         rrw_exit(&ds->ds_bp_rwlock, FTAG);
 790         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
 791 
 792         /* We need to log before removing it from the namespace. */
 793         spa_history_log_internal_ds(ds, "destroy", tx, "");
 794 
 795         rmorigin = (dsl_dir_is_clone(ds->ds_dir) &&
 796             DS_IS_DEFER_DESTROY(ds->ds_prev) &&
 797             dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
 798             ds->ds_prev->ds_userrefs == 0);
 799 
 800         /* Remove our reservation. */
 801         if (ds->ds_reserved != 0) {
 802                 dsl_dataset_set_refreservation_sync_impl(ds,
 803                     (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
 804                     0, tx);
 805                 ASSERT0(ds->ds_reserved);
 806         }
 807 
 808         obj = ds->ds_object;
 809 
 810         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 811                 if (ds->ds_feature_inuse[f]) {
 812                         dsl_dataset_deactivate_feature(obj, f, tx);
 813                         ds->ds_feature_inuse[f] = B_FALSE;
 814                 }
 815         }
 816 
 817         dsl_scan_ds_destroyed(ds, tx);
 818 
 819         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 820                 /* This is a clone */
 821                 ASSERT(ds->ds_prev != NULL);
 822                 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj, !=,
 823                     obj);
 824                 ASSERT0(dsl_dataset_phys(ds)->ds_next_snap_obj);
 825 
 826                 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 827                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj != 0) {
 828                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
 829                             obj, tx);
 830                 }
 831 
 832                 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_num_children, >, 1);
 833                 dsl_dataset_phys(ds->ds_prev)->ds_num_children--;
 834         }
 835 
 836         /*
 837          * Destroy the deadlist.  Unless it's a clone, the
 838          * deadlist should be empty.  (If it's a clone, it's
 839          * safe to ignore the deadlist contents.)
 840          */
 841         dsl_deadlist_close(&ds->ds_deadlist);
 842         dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
 843         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 844         dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
 845 
 846         objset_t *os;
 847         VERIFY0(dmu_objset_from_ds(ds, &os));
 848 
 849         if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_WBC)) {
 850                 wbc_process_objset(spa_get_wbc_data(dp->dp_spa), os, B_TRUE);
 851 
 852                 /*
 853                  * If WBC was activated for this dataset and it is a root
 854                  * of WBC-ed tree of datasets then need to decrement WBC
 855                  * feature flag refcounter, to be sure that 'feature@wbc'
 856                  * shows correct information about the status of WBC
 857                  */
 858                 if (os->os_wbc_root_ds_obj != 0 &&
 859                     ds->ds_object == os->os_wbc_root_ds_obj)
 860                         spa_feature_decr(os->os_spa, SPA_FEATURE_WBC, tx);
 861         }
 862 
 863         if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) {
 864                 old_synchronous_dataset_destroy(ds, tx);
 865         } else {
 866                 /*
 867                  * Move the bptree into the pool's list of trees to
 868                  * clean up and update space accounting information.
 869                  */
 870                 uint64_t used, comp, uncomp;
 871 
 872                 zil_destroy_sync(dmu_objset_zil(os), tx);
 873 
 874                 if (!spa_feature_is_active(dp->dp_spa,
 875                     SPA_FEATURE_ASYNC_DESTROY)) {
 876                         dsl_scan_t *scn = dp->dp_scan;
 877                         spa_feature_incr(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY,
 878                             tx);
 879                         dp->dp_bptree_obj = bptree_alloc(mos, tx);
 880                         VERIFY0(zap_add(mos,
 881                             DMU_POOL_DIRECTORY_OBJECT,
 882                             DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
 883                             &dp->dp_bptree_obj, tx));
 884                         ASSERT(!scn->scn_async_destroying);
 885                         scn->scn_async_destroying = B_TRUE;
 886                 }
 887 
 888                 used = dsl_dir_phys(ds->ds_dir)->dd_used_bytes;
 889                 comp = dsl_dir_phys(ds->ds_dir)->dd_compressed_bytes;
 890                 uncomp = dsl_dir_phys(ds->ds_dir)->dd_uncompressed_bytes;
 891 
 892                 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
 893                     dsl_dataset_phys(ds)->ds_unique_bytes == used);
 894 
 895                 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
 896                 bptree_add(mos, dp->dp_bptree_obj,
 897                     &dsl_dataset_phys(ds)->ds_bp,
 898                     dsl_dataset_phys(ds)->ds_prev_snap_txg,
 899                     used, comp, uncomp, tx);
 900                 rrw_exit(&ds->ds_bp_rwlock, FTAG);
 901                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 902                     -used, -comp, -uncomp, tx);
 903                 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
 904                     used, comp, uncomp, tx);
 905         }
 906 
 907         if (ds->ds_prev != NULL) {
 908                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 909                         VERIFY0(zap_remove_int(mos,
 910                             dsl_dir_phys(ds->ds_prev->ds_dir)->dd_clones,
 911                             ds->ds_object, tx));
 912                 }
 913                 prevobj = ds->ds_prev->ds_object;
 914                 dsl_dataset_rele(ds->ds_prev, ds);
 915                 ds->ds_prev = NULL;
 916         }
 917 
 918         /*
 919          * This must be done after the dsl_traverse(), because it will
 920          * re-open the objset.
 921          */
 922         if (ds->ds_objset) {
 923                 dmu_objset_evict(ds->ds_objset);
 924                 ds->ds_objset = NULL;
 925         }
 926 
 927         /* Erase the link in the dir */
 928         dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
 929         dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj = 0;
 930         ddobj = ds->ds_dir->dd_object;
 931         ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0);
 932         VERIFY0(zap_destroy(mos,
 933             dsl_dataset_phys(ds)->ds_snapnames_zapobj, tx));
 934 
 935         if (ds->ds_bookmarks != 0) {
 936                 VERIFY0(zap_destroy(mos, ds->ds_bookmarks, tx));
 937                 spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
 938         }
 939 
 940         spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
 941 
 942         ASSERT0(dsl_dataset_phys(ds)->ds_next_clones_obj);
 943         ASSERT0(dsl_dataset_phys(ds)->ds_props_obj);
 944         ASSERT0(dsl_dataset_phys(ds)->ds_userrefs_obj);
 945         dsl_dir_rele(ds->ds_dir, ds);
 946         ds->ds_dir = NULL;
 947         dmu_object_free_zapified(mos, obj, tx);
 948 
 949         dsl_dir_destroy_sync(ddobj, tx);
 950 
 951         if (rmorigin) {
 952                 dsl_dataset_t *prev;
 953                 VERIFY0(dsl_dataset_hold_obj(dp, prevobj, FTAG, &prev));
 954                 dsl_destroy_snapshot_sync_impl(prev, B_FALSE, tx);
 955                 dsl_dataset_rele(prev, FTAG);
 956         }
 957 }
 958 
 959 void
 960 dsl_destroy_head_sync(void *arg, dmu_tx_t *tx)
 961 {
 962         dsl_destroy_head_arg_t *ddha = arg;
 963         dsl_pool_t *dp = dmu_tx_pool(tx);
 964         dsl_dataset_t *ds;
 965 
 966         VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
 967         dsl_destroy_head_sync_impl(ds, tx);
 968         dsl_dataset_rele(ds, FTAG);
 969 }
 970 
 971 static void
 972 dsl_destroy_head_begin_sync(void *arg, dmu_tx_t *tx)
 973 {
 974         dsl_destroy_head_arg_t *ddha = arg;
 975         dsl_pool_t *dp = dmu_tx_pool(tx);
 976         dsl_dataset_t *ds;
 977 
 978         VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
 979 
 980         /* Mark it as inconsistent on-disk, in case we crash */
 981         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 982         dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT;
 983 
 984         spa_history_log_internal_ds(ds, "destroy begin", tx, "");
 985         dsl_dataset_rele(ds, FTAG);
 986 }
 987 
 988 int
 989 dsl_destroy_head(const char *name)
 990 {
 991         dsl_destroy_head_arg_t ddha;
 992         int error;
 993         spa_t *spa;
 994         boolean_t isenabled;
 995 
 996 #ifdef _KERNEL
 997         zfs_destroy_unmount_origin(name);
 998 #endif
 999 
1000         error = spa_open(name, &spa, FTAG);
1001         if (error != 0)
1002                 return (error);
1003         isenabled = spa_feature_is_enabled(spa, SPA_FEATURE_ASYNC_DESTROY);
1004 
1005         spa_close(spa, FTAG);
1006 
1007         ddha.ddha_name = name;
1008 
1009         if (!isenabled) {
1010                 objset_t *os;
1011 
1012                 error = dsl_sync_task(name, dsl_destroy_head_check,
1013                     dsl_destroy_head_begin_sync, &ddha,
1014                     0, ZFS_SPACE_CHECK_NONE);
1015                 if (error != 0)
1016                         return (error);
1017 
1018                 /*
1019                  * Head deletion is processed in one txg on old pools;
1020                  * remove the objects from open context so that the txg sync
1021                  * is not too long.
1022                  */
1023                 error = dmu_objset_own(name, DMU_OST_ANY, B_FALSE, FTAG, &os);
1024                 if (error == 0) {
1025                         uint64_t prev_snap_txg =
1026                             dsl_dataset_phys(dmu_objset_ds(os))->
1027                             ds_prev_snap_txg;
1028                         for (uint64_t obj = 0; error == 0;
1029                             error = dmu_object_next(os, &obj, FALSE,
1030                             prev_snap_txg))
1031                                 (void) dmu_free_long_object(os, obj);
1032                         /* sync out all frees */
1033                         txg_wait_synced(dmu_objset_pool(os), 0);
1034                         dmu_objset_disown(os, FTAG);
1035                 }
1036         }
1037 
1038         return (dsl_sync_task(name, dsl_destroy_head_check,
1039             dsl_destroy_head_sync, &ddha, 0, ZFS_SPACE_CHECK_NONE));
1040 }
1041 
1042 typedef struct {
1043         kmutex_t        lock;
1044         list_t list;
1045 } dsl_inconsistent_walker_cb_t;
1046 
1047 typedef struct {
1048         char name[ZFS_MAX_DATASET_NAME_LEN];
1049         list_node_t node;
1050 } dsl_inconsistent_node_t;
1051 
1052 /* ARGSUSED */
1053 static int
1054 dsl_collect_inconsistent_datasets_cb(dsl_pool_t *dp,
1055     dsl_dataset_t *ds, void *arg)
1056 {
1057         dsl_inconsistent_node_t *ds_node;
1058         dsl_inconsistent_walker_cb_t *walker =
1059             (dsl_inconsistent_walker_cb_t *)arg;
1060 
1061         if (!DS_IS_INCONSISTENT(ds))
1062                 return (0);
1063 
1064         /*
1065          * If the dataset is inconsistent because a resumable receive
1066          * has failed, then do not destroy it.
1067          */
1068         if (dsl_dataset_has_resume_receive_state(ds))
1069                 return (0);
1070 
1071         ds_node = kmem_alloc(sizeof (dsl_inconsistent_node_t), KM_SLEEP);
1072         dsl_dataset_name(ds, ds_node->name);
1073 
1074         mutex_enter(&walker->lock);
1075         list_insert_tail(&walker->list, ds_node);
1076         mutex_exit(&walker->lock);
1077 
1078         return (0);
1079 }
1080 
1081 /*
1082  * Walk in parallel over the entire pool and gather inconsistent
1083  * datasets namely, those that don't have resume token and destroy them.
1084  */
1085 void
1086 dsl_destroy_inconsistent(dsl_pool_t *dp)
1087 {
1088         dsl_inconsistent_walker_cb_t walker;
1089         dsl_inconsistent_node_t *ds_node;
1090 
1091         mutex_init(&walker.lock, NULL, MUTEX_DEFAULT, NULL);
1092         list_create(&walker.list, sizeof (dsl_inconsistent_node_t),
1093             offsetof(dsl_inconsistent_node_t, node));
1094 
1095         VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1096                 dsl_collect_inconsistent_datasets_cb,
1097             &walker, DS_FIND_CHILDREN));
1098 
1099         while ((ds_node = list_remove_head(&walker.list)) != NULL) {
1100                 (void) dsl_destroy_head(ds_node->name);
1101                 kmem_free(ds_node, sizeof (dsl_inconsistent_node_t));
1102         }
1103 
1104         list_destroy(&walker.list);
1105         mutex_destroy(&walker.lock);
1106 }
1107 
1108 typedef struct {
1109         const char *from_ds;
1110         boolean_t defer;
1111 } dmu_destroy_atomically_arg_t;
1112 
1113 static int
1114 dsl_destroy_atomically_sync(void *arg, dmu_tx_t *tx)
1115 {
1116         dmu_destroy_atomically_arg_t *ddaa = arg;
1117         boolean_t defer = ddaa->defer;
1118         dsl_pool_t *dp = dmu_tx_pool(tx);
1119         zfs_ds_collector_entry_t *tail;
1120         list_t namestack;
1121         int err = 0;
1122 
1123         /* do not perfrom checks in ioctl */
1124         if (!dmu_tx_is_syncing(tx))
1125                 return (0);
1126 
1127         ASSERT(dsl_pool_config_held(dp));
1128 
1129         if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY))
1130                 return (SET_ERROR(ENOTSUP));
1131 
1132         /* It is possible than autosnap watches the DS */
1133         if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_WBC)) {
1134                 objset_t *os = NULL;
1135                 dsl_dataset_t *ds = NULL;
1136 
1137                 err = dsl_dataset_hold(dp, ddaa->from_ds, FTAG, &ds);
1138                 if (err != 0)
1139                         return (err);
1140 
1141                 err = dmu_objset_from_ds(ds, &os);
1142                 if (err != 0) {
1143                         dsl_dataset_rele(ds, FTAG);
1144                         return (err);
1145                 }
1146 
1147                 if (!dmu_objset_is_snapshot(os)) {
1148                         wbc_process_objset(spa_get_wbc_data(dp->dp_spa),
1149                             os, B_TRUE);
1150                 }
1151 
1152                 dsl_dataset_rele(ds, FTAG);
1153         }
1154 
1155         /* initialize the stack of datasets */
1156         list_create(&namestack, sizeof (zfs_ds_collector_entry_t),
1157             offsetof(zfs_ds_collector_entry_t, node));
1158         tail = dsl_dataset_collector_cache_alloc();
1159 
1160         /* push the head */
1161         tail->cookie = 0;
1162         tail->cookie_is_snap = B_FALSE;
1163         (void) strcpy(tail->name, ddaa->from_ds);
1164         list_insert_tail(&namestack, tail);
1165 
1166         /* the head is processed at the very end and after all is done */
1167         while (err == 0 && ((tail = list_tail(&namestack)) != NULL)) {
1168                 zfs_ds_collector_entry_t *el;
1169                 objset_t *os;
1170                 dsl_dataset_t *ds;
1171                 char *p;
1172 
1173                 /* init new entry */
1174                 el = dsl_dataset_collector_cache_alloc();
1175                 el->cookie = 0;
1176                 el->cookie_is_snap = B_FALSE;
1177                 (void) strcpy(el->name, tail->name);
1178                 p = el->name + strlen(el->name);
1179 
1180                 /* hold the current dataset to traverse its children */
1181                 err = dsl_dataset_hold(dp, tail->name, FTAG, &ds);
1182                 if (err != 0) {
1183                         dsl_dataset_collector_cache_free(el);
1184                         break;
1185                 }
1186 
1187                 err  = dmu_objset_from_ds(ds, &os);
1188                 if (err != 0) {
1189                         dsl_dataset_rele(ds, FTAG);
1190                         dsl_dataset_collector_cache_free(el);
1191                         break;
1192                 }
1193 
1194                 if (dmu_objset_is_snapshot(os)) {
1195                         /* traverse clones for snapshots */
1196                         err = dmu_clone_list_next(os, MAXNAMELEN,
1197                             el->name, NULL, &tail->cookie);
1198                 } else {
1199                         /* for filesystems traverse fs first, then snaps */
1200                         if (!tail->cookie_is_snap) {
1201                                 *p++ = '/';
1202                                 do {
1203                                         *p = '\0';
1204                                         err = dmu_dir_list_next(os,
1205                                             MAXNAMELEN - (p - el->name),
1206                                             p, NULL, &tail->cookie);
1207                                 } while (err == 0 &&
1208                                     dataset_name_hidden(el->name));
1209 
1210                                 /* no more fs, move to snapshots */
1211                                 if (err == ENOENT) {
1212                                         *(--p) = '\0';
1213                                         tail->cookie_is_snap = 1;
1214                                         tail->cookie = 0;
1215                                         err = 0;
1216                                 }
1217                         }
1218 
1219                         if (err == 0 && tail->cookie_is_snap) {
1220                                 *p++ = '@';
1221                                 *p = '\0';
1222                                 err = dmu_snapshot_list_next(os,
1223                                     MAXNAMELEN - (p - el->name),
1224                                     p, NULL, &tail->cookie, NULL);
1225                         }
1226                 }
1227 
1228                 if (err == 0) {
1229                         /* a children found, add it and continue */
1230                         list_insert_tail(&namestack, el);
1231                         dsl_dataset_rele(ds, FTAG);
1232                         continue;
1233                 }
1234 
1235                 dsl_dataset_collector_cache_free(el);
1236 
1237                 if (err != ENOENT) {
1238                         dsl_dataset_rele(ds, FTAG);
1239                         break;
1240                 }
1241 
1242                 /*
1243                  * There are no more children of the dataset, pop it from stack
1244                  * and destroy it
1245                  */
1246 
1247                 err = 0;
1248 
1249                 list_remove(&namestack, tail);
1250 
1251                 if (dmu_objset_is_snapshot(os)) {
1252                         err = dsl_destroy_snapshot_check_impl(ds, defer);
1253                         if (err == 0)
1254                                 dsl_destroy_snapshot_sync_impl(ds, defer, tx);
1255                 } else if (strchr(tail->name, '/') != NULL) {
1256                         err = dsl_destroy_head_check_impl(ds, 0);
1257                         if (err == 0)
1258                                 dsl_destroy_head_sync_impl(ds, tx);
1259                 }
1260 
1261                 dsl_dataset_rele(ds, FTAG);
1262                 dsl_dataset_collector_cache_free(tail);
1263         }
1264 
1265         if (err != 0) {
1266                 while ((tail = list_remove_tail(&namestack)) != NULL)
1267                         dsl_dataset_collector_cache_free(tail);
1268         }
1269 
1270         ASSERT(list_head(&namestack) == NULL);
1271 
1272         list_destroy(&namestack);
1273 
1274         return (err);
1275 }
1276 
1277 /*ARGSUSED*/
1278 void
1279 dsl_destroy_atomically_sync_dummy(void *arg, dmu_tx_t *tx)
1280 {
1281 }
1282 
1283 int
1284 dsl_destroy_atomically(const char *name, boolean_t defer)
1285 {
1286         dmu_destroy_atomically_arg_t ddaa;
1287 
1288         ddaa.from_ds = name;
1289         ddaa.defer = defer;
1290 
1291         return (dsl_sync_task(name, dsl_destroy_atomically_sync,
1292             dsl_destroy_atomically_sync_dummy, &ddaa, 0, ZFS_SPACE_CHECK_NONE));
1293 }