module/zfs/dsl_dataset.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2013 by Delphix. All rights reserved.
  24  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  25  * Copyright (c) 2014 RackTop Systems.
  26  */
  27
  28 #include <sys/dmu_objset.h>
  29 #include <sys/dsl_dataset.h>
  30 #include <sys/dsl_dir.h>
  31 #include <sys/dsl_prop.h>
  32 #include <sys/dsl_synctask.h>
  33 #include <sys/dmu_traverse.h>
  34 #include <sys/dmu_impl.h>
  35 #include <sys/dmu_tx.h>
  36 #include <sys/arc.h>
  37 #include <sys/zio.h>
  38 #include <sys/zap.h>
  39 #include <sys/zfeature.h>
  40 #include <sys/unique.h>
  41 #include <sys/zfs_context.h>
  42 #include <sys/zfs_ioctl.h>
  43 #include <sys/spa.h>
  44 #include <sys/zfs_znode.h>
  45 #include <sys/zfs_onexit.h>
  46 #include <sys/zvol.h>
  47 #include <sys/dsl_scan.h>
  48 #include <sys/dsl_deadlist.h>
  49 #include <sys/dsl_destroy.h>
  50 #include <sys/dsl_userhold.h>
  51 #include <sys/dsl_bookmark.h>
  52
  53 #define SWITCH64(x, y) \
  54         { \
  55                 uint64_t __tmp = (x); \
  56                 (x) = (y); \
  57                 (y) = __tmp; \
  58         }
  59
  60 #define DS_REF_MAX      (1ULL << 62)
  61
  62 #define DSL_DEADLIST_BLOCKSIZE  SPA_MAXBLOCKSIZE
  63
  64 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
  65 extern inline boolean_t dsl_dataset_is_snapshot(dsl_dataset_t *ds);
  66
  67 /*
  68  * Figure out how much of this delta should be propogated to the dsl_dir
  69  * layer.  If there's a refreservation, that space has already been
  70  * partially accounted for in our ancestors.
  71  */
  72 static int64_t
  73 parent_delta(dsl_dataset_t *ds, int64_t delta)
  74 {
  75         dsl_dataset_phys_t *ds_phys;
  76         uint64_t old_bytes, new_bytes;
  77
  78         if (ds->ds_reserved == 0)
  79                 return (delta);
  80
  81         ds_phys = dsl_dataset_phys(ds);
  82         old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
  83         new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
  84
  85         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
  86         return (new_bytes - old_bytes);
  87 }
  88
  89 void
  90 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
  91 {
  92         int used, compressed, uncompressed;
  93         int64_t delta;
  94
  95         used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
  96         compressed = BP_GET_PSIZE(bp);
  97         uncompressed = BP_GET_UCSIZE(bp);
  98
  99         dprintf_bp(bp, "ds=%p", ds);
 100
 101         ASSERT(dmu_tx_is_syncing(tx));
 102         /* It could have been compressed away to nothing */
 103         if (BP_IS_HOLE(bp))
 104                 return;
 105         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 106         ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 107         if (ds == NULL) {
 108                 dsl_pool_mos_diduse_space(tx->tx_pool,
 109                     used, compressed, uncompressed);
 110                 return;
 111         }
 112
 113         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 114         mutex_enter(&ds->ds_lock);
 115         delta = parent_delta(ds, used);
 116         dsl_dataset_phys(ds)->ds_referenced_bytes += used;
 117         dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
 118         dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
 119         dsl_dataset_phys(ds)->ds_unique_bytes += used;
 120         mutex_exit(&ds->ds_lock);
 121         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 122             compressed, uncompressed, tx);
 123         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 124             DD_USED_REFRSRV, DD_USED_HEAD, tx);
 125 }
 126
 127 int
 128 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 129     boolean_t async)
 130 {
 131         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 132         int compressed = BP_GET_PSIZE(bp);
 133         int uncompressed = BP_GET_UCSIZE(bp);
 134
 135         if (BP_IS_HOLE(bp))
 136                 return (0);
 137
 138         ASSERT(dmu_tx_is_syncing(tx));
 139         ASSERT(bp->blk_birth <= tx->tx_txg);
 140
 141         if (ds == NULL) {
 142                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 143                 dsl_pool_mos_diduse_space(tx->tx_pool,
 144                     -used, -compressed, -uncompressed);
 145                 return (used);
 146         }
 147         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 148
 149         ASSERT(!dsl_dataset_is_snapshot(ds));
 150         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 151
 152         if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
 153                 int64_t delta;
 154
 155                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 156                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 157
 158                 mutex_enter(&ds->ds_lock);
 159                 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
 160                     !DS_UNIQUE_IS_ACCURATE(ds));
 161                 delta = parent_delta(ds, -used);
 162                 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
 163                 mutex_exit(&ds->ds_lock);
 164                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 165                     delta, -compressed, -uncompressed, tx);
 166                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 167                     DD_USED_REFRSRV, DD_USED_HEAD, tx);
 168         } else {
 169                 dprintf_bp(bp, "putting on dead list: %s", "");
 170                 if (async) {
 171                         /*
 172                          * We are here as part of zio's write done callback,
 173                          * which means we're a zio interrupt thread.  We can't
 174                          * call dsl_deadlist_insert() now because it may block
 175                          * waiting for I/O.  Instead, put bp on the deferred
 176                          * queue and let dsl_pool_sync() finish the job.
 177                          */
 178                         bplist_append(&ds->ds_pending_deadlist, bp);
 179                 } else {
 180                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 181                 }
 182                 ASSERT3U(ds->ds_prev->ds_object, ==,
 183                     dsl_dataset_phys(ds)->ds_prev_snap_obj);
 184                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
 185                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 186                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
 187                     ds->ds_object && bp->blk_birth >
 188                     dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
 189                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 190                         mutex_enter(&ds->ds_prev->ds_lock);
 191                         dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
 192                         mutex_exit(&ds->ds_prev->ds_lock);
 193                 }
 194                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 195                         dsl_dir_transfer_space(ds->ds_dir, used,
 196                             DD_USED_HEAD, DD_USED_SNAP, tx);
 197                 }
 198         }
 199         mutex_enter(&ds->ds_lock);
 200         ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
 201         dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
 202         ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
 203         dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
 204         ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
 205         dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
 206         mutex_exit(&ds->ds_lock);
 207
 208         return (used);
 209 }
 210
 211 uint64_t
 212 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
 213 {
 214         uint64_t trysnap = 0;
 215
 216         if (ds == NULL)
 217                 return (0);
 218         /*
 219          * The snapshot creation could fail, but that would cause an
 220          * incorrect FALSE return, which would only result in an
 221          * overestimation of the amount of space that an operation would
 222          * consume, which is OK.
 223          *
 224          * There's also a small window where we could miss a pending
 225          * snapshot, because we could set the sync task in the quiescing
 226          * phase.  So this should only be used as a guess.
 227          */
 228         if (ds->ds_trysnap_txg >
 229             spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
 230                 trysnap = ds->ds_trysnap_txg;
 231         return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
 232 }
 233
 234 boolean_t
 235 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
 236     uint64_t blk_birth)
 237 {
 238         if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
 239             (bp != NULL && BP_IS_HOLE(bp)))
 240                 return (B_FALSE);
 241
 242         ddt_prefetch(dsl_dataset_get_spa(ds), bp);
 243
 244         return (B_TRUE);
 245 }
 246
 247 /* ARGSUSED */
 248 static void
 249 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
 250 {
 251         dsl_dataset_t *ds = dsv;
 252
 253         ASSERT(ds->ds_owner == NULL);
 254
 255         unique_remove(ds->ds_fsid_guid);
 256
 257         if (ds->ds_objset != NULL)
 258                 dmu_objset_evict(ds->ds_objset);
 259
 260         if (ds->ds_prev) {
 261                 dsl_dataset_rele(ds->ds_prev, ds);
 262                 ds->ds_prev = NULL;
 263         }
 264
 265         bplist_destroy(&ds->ds_pending_deadlist);
 266         if (dsl_dataset_phys(ds)->ds_deadlist_obj != 0)
 267                 dsl_deadlist_close(&ds->ds_deadlist);
 268         if (ds->ds_dir)
 269                 dsl_dir_rele(ds->ds_dir, ds);
 270
 271         ASSERT(!list_link_active(&ds->ds_synced_link));
 272
 273         mutex_destroy(&ds->ds_lock);
 274         mutex_destroy(&ds->ds_opening_lock);
 275         mutex_destroy(&ds->ds_sendstream_lock);
 276         refcount_destroy(&ds->ds_longholds);
 277
 278         kmem_free(ds, sizeof (dsl_dataset_t));
 279 }
 280
 281 int
 282 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 283 {
 284         dsl_dataset_phys_t *headphys;
 285         int err;
 286         dmu_buf_t *headdbuf;
 287         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 288         objset_t *mos = dp->dp_meta_objset;
 289
 290         if (ds->ds_snapname[0])
 291                 return (0);
 292         if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
 293                 return (0);
 294
 295         err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
 296             FTAG, &headdbuf);
 297         if (err != 0)
 298                 return (err);
 299         headphys = headdbuf->db_data;
 300         err = zap_value_search(dp->dp_meta_objset,
 301             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 302         dmu_buf_rele(headdbuf, FTAG);
 303         return (err);
 304 }
 305
 306 int
 307 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 308 {
 309         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 310         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 311         matchtype_t mt;
 312         int err;
 313
 314         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 315                 mt = MT_FIRST;
 316         else
 317                 mt = MT_EXACT;
 318
 319         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 320             value, mt, NULL, 0, NULL);
 321         if (err == ENOTSUP && mt == MT_FIRST)
 322                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 323         return (err);
 324 }
 325
 326 int
 327 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
 328     boolean_t adj_cnt)
 329 {
 330         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 331         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 332         matchtype_t mt;
 333         int err;
 334
 335         dsl_dir_snap_cmtime_update(ds->ds_dir);
 336
 337         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 338                 mt = MT_FIRST;
 339         else
 340                 mt = MT_EXACT;
 341
 342         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 343         if (err == ENOTSUP && mt == MT_FIRST)
 344                 err = zap_remove(mos, snapobj, name, tx);
 345
 346         if (err == 0 && adj_cnt)
 347                 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
 348                     DD_FIELD_SNAPSHOT_COUNT, tx);
 349
 350         return (err);
 351 }
 352
 353 int
 354 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 355     dsl_dataset_t **dsp)
 356 {
 357         objset_t *mos = dp->dp_meta_objset;
 358         dmu_buf_t *dbuf;
 359         dsl_dataset_t *ds;
 360         int err;
 361         dmu_object_info_t doi;
 362
 363         ASSERT(dsl_pool_config_held(dp));
 364
 365         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 366         if (err != 0)
 367                 return (err);
 368
 369         /* Make sure dsobj has the correct object type. */
 370         dmu_object_info_from_db(dbuf, &doi);
 371         if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
 372                 dmu_buf_rele(dbuf, tag);
 373                 return (SET_ERROR(EINVAL));
 374         }
 375
 376         ds = dmu_buf_get_user(dbuf);
 377         if (ds == NULL) {
 378                 dsl_dataset_t *winner = NULL;
 379
 380                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 381                 ds->ds_dbuf = dbuf;
 382                 ds->ds_object = dsobj;
 383                 list_link_init(&ds->ds_synced_link);
 384
 385                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 386                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 387                 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
 388                 refcount_create(&ds->ds_longholds);
 389
 390                 bplist_create(&ds->ds_pending_deadlist);
 391                 dsl_deadlist_open(&ds->ds_deadlist,
 392                     mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
 393
 394                 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
 395                     offsetof(dmu_sendarg_t, dsa_link));
 396
 397                 if (err == 0) {
 398                         err = dsl_dir_hold_obj(dp,
 399                             dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds,
 400                             &ds->ds_dir);
 401                 }
 402                 if (err != 0) {
 403                         mutex_destroy(&ds->ds_lock);
 404                         mutex_destroy(&ds->ds_opening_lock);
 405                         mutex_destroy(&ds->ds_sendstream_lock);
 406                         refcount_destroy(&ds->ds_longholds);
 407                         bplist_destroy(&ds->ds_pending_deadlist);
 408                         dsl_deadlist_close(&ds->ds_deadlist);
 409                         kmem_free(ds, sizeof (dsl_dataset_t));
 410                         dmu_buf_rele(dbuf, tag);
 411                         return (err);
 412                 }
 413
 414                 if (!dsl_dataset_is_snapshot(ds)) {
 415                         ds->ds_snapname[0] = '\0';
 416                         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 417                                 err = dsl_dataset_hold_obj(dp,
 418                                     dsl_dataset_phys(ds)->ds_prev_snap_obj,
 419                                     ds, &ds->ds_prev);
 420                         }
 421                         if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 422                                 int zaperr = zap_lookup(mos, ds->ds_object,
 423                                     DS_FIELD_BOOKMARK_NAMES,
 424                                     sizeof (ds->ds_bookmarks), 1,
 425                                     &ds->ds_bookmarks);
 426                                 if (zaperr != ENOENT)
 427                                         VERIFY0(zaperr);
 428                         }
 429                 } else {
 430                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 431                                 err = dsl_dataset_get_snapname(ds);
 432                         if (err == 0 &&
 433                             dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
 434                                 err = zap_count(
 435                                     ds->ds_dir->dd_pool->dp_meta_objset,
 436                                     dsl_dataset_phys(ds)->ds_userrefs_obj,
 437                                     &ds->ds_userrefs);
 438                         }
 439                 }
 440
 441                 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
 442                         err = dsl_prop_get_int_ds(ds,
 443                             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
 444                             &ds->ds_reserved);
 445                         if (err == 0) {
 446                                 err = dsl_prop_get_int_ds(ds,
 447                                     zfs_prop_to_name(ZFS_PROP_REFQUOTA),
 448                                     &ds->ds_quota);
 449                         }
 450                 } else {
 451                         ds->ds_reserved = ds->ds_quota = 0;
 452                 }
 453
 454                 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
 455                     dsl_dataset_evict)) != NULL) {
 456                         bplist_destroy(&ds->ds_pending_deadlist);
 457                         dsl_deadlist_close(&ds->ds_deadlist);
 458                         if (ds->ds_prev)
 459                                 dsl_dataset_rele(ds->ds_prev, ds);
 460                         dsl_dir_rele(ds->ds_dir, ds);
 461                         mutex_destroy(&ds->ds_lock);
 462                         mutex_destroy(&ds->ds_opening_lock);
 463                         mutex_destroy(&ds->ds_sendstream_lock);
 464                         refcount_destroy(&ds->ds_longholds);
 465                         kmem_free(ds, sizeof (dsl_dataset_t));
 466                         if (err != 0) {
 467                                 dmu_buf_rele(dbuf, tag);
 468                                 return (err);
 469                         }
 470                         ds = winner;
 471                 } else {
 472                         ds->ds_fsid_guid =
 473                             unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
 474                 }
 475         }
 476         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 477         ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
 478         ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
 479             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 480             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 481         *dsp = ds;
 482         return (0);
 483 }
 484
 485 int
 486 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
 487     void *tag, dsl_dataset_t **dsp)
 488 {
 489         dsl_dir_t *dd;
 490         const char *snapname;
 491         uint64_t obj;
 492         int err = 0;
 493
 494         err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
 495         if (err != 0)
 496                 return (err);
 497
 498         ASSERT(dsl_pool_config_held(dp));
 499         obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
 500         if (obj != 0)
 501                 err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
 502         else
 503                 err = SET_ERROR(ENOENT);
 504
 505         /* we may be looking for a snapshot */
 506         if (err == 0 && snapname != NULL) {
 507                 dsl_dataset_t *ds;
 508
 509                 if (*snapname++ != '@') {
 510                         dsl_dataset_rele(*dsp, tag);
 511                         dsl_dir_rele(dd, FTAG);
 512                         return (SET_ERROR(ENOENT));
 513                 }
 514
 515                 dprintf("looking for snapshot '%s'\n", snapname);
 516                 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
 517                 if (err == 0)
 518                         err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
 519                 dsl_dataset_rele(*dsp, tag);
 520
 521                 if (err == 0) {
 522                         mutex_enter(&ds->ds_lock);
 523                         if (ds->ds_snapname[0] == 0)
 524                                 (void) strlcpy(ds->ds_snapname, snapname,
 525                                     sizeof (ds->ds_snapname));
 526                         mutex_exit(&ds->ds_lock);
 527                         *dsp = ds;
 528                 }
 529         }
 530
 531         dsl_dir_rele(dd, FTAG);
 532         return (err);
 533 }
 534
 535 int
 536 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
 537     void *tag, dsl_dataset_t **dsp)
 538 {
 539         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 540         if (err != 0)
 541                 return (err);
 542         if (!dsl_dataset_tryown(*dsp, tag)) {
 543                 dsl_dataset_rele(*dsp, tag);
 544                 *dsp = NULL;
 545                 return (SET_ERROR(EBUSY));
 546         }
 547         return (0);
 548 }
 549
 550 int
 551 dsl_dataset_own(dsl_pool_t *dp, const char *name,
 552     void *tag, dsl_dataset_t **dsp)
 553 {
 554         int err = dsl_dataset_hold(dp, name, tag, dsp);
 555         if (err != 0)
 556                 return (err);
 557         if (!dsl_dataset_tryown(*dsp, tag)) {
 558                 dsl_dataset_rele(*dsp, tag);
 559                 return (SET_ERROR(EBUSY));
 560         }
 561         return (0);
 562 }
 563
 564 /*
 565  * See the comment above dsl_pool_hold() for details.  In summary, a long
 566  * hold is used to prevent destruction of a dataset while the pool hold
 567  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
 568  *
 569  * The dataset and pool must be held when this function is called.  After it
 570  * is called, the pool hold may be released while the dataset is still held
 571  * and accessed.
 572  */
 573 void
 574 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
 575 {
 576         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 577         (void) refcount_add(&ds->ds_longholds, tag);
 578 }
 579
 580 void
 581 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
 582 {
 583         (void) refcount_remove(&ds->ds_longholds, tag);
 584 }
 585
 586 /* Return B_TRUE if there are any long holds on this dataset. */
 587 boolean_t
 588 dsl_dataset_long_held(dsl_dataset_t *ds)
 589 {
 590         return (!refcount_is_zero(&ds->ds_longholds));
 591 }
 592
 593 void
 594 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 595 {
 596         if (ds == NULL) {
 597                 (void) strcpy(name, "mos");
 598         } else {
 599                 dsl_dir_name(ds->ds_dir, name);
 600                 VERIFY0(dsl_dataset_get_snapname(ds));
 601                 if (ds->ds_snapname[0]) {
 602                         (void) strcat(name, "@");
 603                         /*
 604                          * We use a "recursive" mutex so that we
 605                          * can call dprintf_ds() with ds_lock held.
 606                          */
 607                         if (!MUTEX_HELD(&ds->ds_lock)) {
 608                                 mutex_enter(&ds->ds_lock);
 609                                 (void) strcat(name, ds->ds_snapname);
 610                                 mutex_exit(&ds->ds_lock);
 611                         } else {
 612                                 (void) strcat(name, ds->ds_snapname);
 613                         }
 614                 }
 615         }
 616 }
 617
 618 void
 619 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 620 {
 621         dmu_buf_rele(ds->ds_dbuf, tag);
 622 }
 623
 624 void
 625 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 626 {
 627         ASSERT3P(ds->ds_owner, ==, tag);
 628         ASSERT(ds->ds_dbuf != NULL);
 629
 630         mutex_enter(&ds->ds_lock);
 631         ds->ds_owner = NULL;
 632         mutex_exit(&ds->ds_lock);
 633         dsl_dataset_long_rele(ds, tag);
 634         dsl_dataset_rele(ds, tag);
 635 }
 636
 637 boolean_t
 638 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
 639 {
 640         boolean_t gotit = FALSE;
 641
 642         mutex_enter(&ds->ds_lock);
 643         if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
 644                 ds->ds_owner = tag;
 645                 dsl_dataset_long_hold(ds, tag);
 646                 gotit = TRUE;
 647         }
 648         mutex_exit(&ds->ds_lock);
 649         return (gotit);
 650 }
 651
 652 uint64_t
 653 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 654     uint64_t flags, dmu_tx_t *tx)
 655 {
 656         dsl_pool_t *dp = dd->dd_pool;
 657         dmu_buf_t *dbuf;
 658         dsl_dataset_phys_t *dsphys;
 659         uint64_t dsobj;
 660         objset_t *mos = dp->dp_meta_objset;
 661
 662         if (origin == NULL)
 663                 origin = dp->dp_origin_snap;
 664
 665         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 666         ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
 667         ASSERT(dmu_tx_is_syncing(tx));
 668         ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
 669
 670         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 671             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 672         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 673         dmu_buf_will_dirty(dbuf, tx);
 674         dsphys = dbuf->db_data;
 675         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 676         dsphys->ds_dir_obj = dd->dd_object;
 677         dsphys->ds_flags = flags;
 678         dsphys->ds_fsid_guid = unique_create();
 679         (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 680             sizeof (dsphys->ds_guid));
 681         dsphys->ds_snapnames_zapobj =
 682             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 683             DMU_OT_NONE, 0, tx);
 684         dsphys->ds_creation_time = gethrestime_sec();
 685         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 686
 687         if (origin == NULL) {
 688                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 689         } else {
 690                 dsl_dataset_t *ohds; /* head of the origin snapshot */
 691
 692                 dsphys->ds_prev_snap_obj = origin->ds_object;
 693                 dsphys->ds_prev_snap_txg =
 694                     dsl_dataset_phys(origin)->ds_creation_txg;
 695                 dsphys->ds_referenced_bytes =
 696                     dsl_dataset_phys(origin)->ds_referenced_bytes;
 697                 dsphys->ds_compressed_bytes =
 698                     dsl_dataset_phys(origin)->ds_compressed_bytes;
 699                 dsphys->ds_uncompressed_bytes =
 700                     dsl_dataset_phys(origin)->ds_uncompressed_bytes;
 701                 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
 702
 703                 /*
 704                  * Inherit flags that describe the dataset's contents
 705                  * (INCONSISTENT) or properties (Case Insensitive).
 706                  */
 707                 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
 708                     (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
 709
 710                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 711                 dsl_dataset_phys(origin)->ds_num_children++;
 712
 713                 VERIFY0(dsl_dataset_hold_obj(dp,
 714                     dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
 715                     FTAG, &ohds));
 716                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 717                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 718                 dsl_dataset_rele(ohds, FTAG);
 719
 720                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 721                         if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
 722                                 dsl_dataset_phys(origin)->ds_next_clones_obj =
 723                                     zap_create(mos,
 724                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 725                         }
 726                         VERIFY0(zap_add_int(mos,
 727                             dsl_dataset_phys(origin)->ds_next_clones_obj,
 728                             dsobj, tx));
 729                 }
 730
 731                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 732                 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
 733                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 734                         if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
 735                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 736                                 dsl_dir_phys(origin->ds_dir)->dd_clones =
 737                                     zap_create(mos,
 738                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 739                         }
 740                         VERIFY0(zap_add_int(mos,
 741                             dsl_dir_phys(origin->ds_dir)->dd_clones,
 742                             dsobj, tx));
 743                 }
 744         }
 745
 746         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 747                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 748
 749         dmu_buf_rele(dbuf, FTAG);
 750
 751         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 752         dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
 753
 754         return (dsobj);
 755 }
 756
 757 static void
 758 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
 759 {
 760         objset_t *os;
 761
 762         VERIFY0(dmu_objset_from_ds(ds, &os));
 763         bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 764         dsl_dataset_dirty(ds, tx);
 765 }
 766
 767 uint64_t
 768 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 769     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 770 {
 771         dsl_pool_t *dp = pdd->dd_pool;
 772         uint64_t dsobj, ddobj;
 773         dsl_dir_t *dd;
 774
 775         ASSERT(dmu_tx_is_syncing(tx));
 776         ASSERT(lastname[0] != '@');
 777
 778         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 779         VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
 780
 781         dsobj = dsl_dataset_create_sync_dd(dd, origin,
 782             flags & ~DS_CREATE_FLAG_NODIRTY, tx);
 783
 784         dsl_deleg_set_create_perms(dd, tx, cr);
 785
 786         /*
 787          * Since we're creating a new node we know it's a leaf, so we can
 788          * initialize the counts if the limit feature is active.
 789          */
 790         if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
 791                 uint64_t cnt = 0;
 792                 objset_t *os = dd->dd_pool->dp_meta_objset;
 793
 794                 dsl_dir_zapify(dd, tx);
 795                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
 796                     sizeof (cnt), 1, &cnt, tx));
 797                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
 798                     sizeof (cnt), 1, &cnt, tx));
 799         }
 800
 801         dsl_dir_rele(dd, FTAG);
 802
 803         /*
 804          * If we are creating a clone, make sure we zero out any stale
 805          * data from the origin snapshots zil header.
 806          */
 807         if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
 808                 dsl_dataset_t *ds;
 809
 810                 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 811                 dsl_dataset_zero_zil(ds, tx);
 812                 dsl_dataset_rele(ds, FTAG);
 813         }
 814
 815         return (dsobj);
 816 }
 817
 818 /*
 819  * The unique space in the head dataset can be calculated by subtracting
 820  * the space used in the most recent snapshot, that is still being used
 821  * in this file system, from the space currently in use.  To figure out
 822  * the space in the most recent snapshot still in use, we need to take
 823  * the total space used in the snapshot and subtract out the space that
 824  * has been freed up since the snapshot was taken.
 825  */
 826 void
 827 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
 828 {
 829         uint64_t mrs_used;
 830         uint64_t dlused, dlcomp, dluncomp;
 831
 832         ASSERT(!dsl_dataset_is_snapshot(ds));
 833
 834         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
 835                 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
 836         else
 837                 mrs_used = 0;
 838
 839         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
 840
 841         ASSERT3U(dlused, <=, mrs_used);
 842         dsl_dataset_phys(ds)->ds_unique_bytes =
 843             dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
 844
 845         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
 846             SPA_VERSION_UNIQUE_ACCURATE)
 847                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 848 }
 849
 850 void
 851 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
 852     dmu_tx_t *tx)
 853 {
 854         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 855         int err;
 856         ASSERTV(uint64_t count);
 857
 858         ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
 859         err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
 860             obj, tx);
 861         /*
 862          * The err should not be ENOENT, but a bug in a previous version
 863          * of the code could cause upgrade_clones_cb() to not set
 864          * ds_next_snap_obj when it should, leading to a missing entry.
 865          * If we knew that the pool was created after
 866          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
 867          * ENOENT.  However, at least we can check that we don't have
 868          * too many entries in the next_clones_obj even after failing to
 869          * remove this one.
 870          */
 871         if (err != ENOENT)
 872                 VERIFY0(err);
 873         ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
 874             &count));
 875         ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
 876 }
 877
 878
 879 blkptr_t *
 880 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
 881 {
 882         return (&dsl_dataset_phys(ds)->ds_bp);
 883 }
 884
 885 void
 886 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
 887 {
 888         ASSERT(dmu_tx_is_syncing(tx));
 889         /* If it's the meta-objset, set dp_meta_rootbp */
 890         if (ds == NULL) {
 891                 tx->tx_pool->dp_meta_rootbp = *bp;
 892         } else {
 893                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
 894                 dsl_dataset_phys(ds)->ds_bp = *bp;
 895         }
 896 }
 897
 898 spa_t *
 899 dsl_dataset_get_spa(dsl_dataset_t *ds)
 900 {
 901         return (ds->ds_dir->dd_pool->dp_spa);
 902 }
 903
 904 void
 905 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
 906 {
 907         dsl_pool_t *dp;
 908
 909         if (ds == NULL) /* this is the meta-objset */
 910                 return;
 911
 912         ASSERT(ds->ds_objset != NULL);
 913
 914         if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
 915                 panic("dirtying snapshot!");
 916
 917         dp = ds->ds_dir->dd_pool;
 918
 919         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
 920                 /* up the hold count until we can be written out */
 921                 dmu_buf_add_ref(ds->ds_dbuf, ds);
 922         }
 923 }
 924
 925 boolean_t
 926 dsl_dataset_is_dirty(dsl_dataset_t *ds)
 927 {
 928         int t;
 929
 930         for (t = 0; t < TXG_SIZE; t++) {
 931                 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
 932                     ds, t))
 933                         return (B_TRUE);
 934         }
 935         return (B_FALSE);
 936 }
 937
 938 static int
 939 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
 940 {
 941         uint64_t asize;
 942
 943         if (!dmu_tx_is_syncing(tx))
 944                 return (0);
 945
 946         /*
 947          * If there's an fs-only reservation, any blocks that might become
 948          * owned by the snapshot dataset must be accommodated by space
 949          * outside of the reservation.
 950          */
 951         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
 952         asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
 953         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
 954                 return (SET_ERROR(ENOSPC));
 955
 956         /*
 957          * Propagate any reserved space for this snapshot to other
 958          * snapshot checks in this sync group.
 959          */
 960         if (asize > 0)
 961                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
 962
 963         return (0);
 964 }
 965
 966 typedef struct dsl_dataset_snapshot_arg {
 967         nvlist_t *ddsa_snaps;
 968         nvlist_t *ddsa_props;
 969         nvlist_t *ddsa_errors;
 970         cred_t *ddsa_cr;
 971 } dsl_dataset_snapshot_arg_t;
 972
 973 int
 974 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
 975     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
 976 {
 977         int error;
 978         uint64_t value;
 979
 980         ds->ds_trysnap_txg = tx->tx_txg;
 981
 982         if (!dmu_tx_is_syncing(tx))
 983                 return (0);
 984
 985         /*
 986          * We don't allow multiple snapshots of the same txg.  If there
 987          * is already one, try again.
 988          */
 989         if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
 990                 return (SET_ERROR(EAGAIN));
 991
 992         /*
 993          * Check for conflicting snapshot name.
 994          */
 995         error = dsl_dataset_snap_lookup(ds, snapname, &value);
 996         if (error == 0)
 997                 return (SET_ERROR(EEXIST));
 998         if (error != ENOENT)
 999                 return (error);
1000
1001         /*
1002          * We don't allow taking snapshots of inconsistent datasets, such as
1003          * those into which we are currently receiving.  However, if we are
1004          * creating this snapshot as part of a receive, this check will be
1005          * executed atomically with respect to the completion of the receive
1006          * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1007          * case we ignore this, knowing it will be fixed up for us shortly in
1008          * dmu_recv_end_sync().
1009          */
1010         if (!recv && DS_IS_INCONSISTENT(ds))
1011                 return (SET_ERROR(EBUSY));
1012
1013         /*
1014          * Skip the check for temporary snapshots or if we have already checked
1015          * the counts in dsl_dataset_snapshot_check. This means we really only
1016          * check the count here when we're receiving a stream.
1017          */
1018         if (cnt != 0 && cr != NULL) {
1019                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1020                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1021                 if (error != 0)
1022                         return (error);
1023         }
1024
1025         error = dsl_dataset_snapshot_reserve_space(ds, tx);
1026         if (error != 0)
1027                 return (error);
1028
1029         return (0);
1030 }
1031
1032 static int
1033 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1034 {
1035         dsl_dataset_snapshot_arg_t *ddsa = arg;
1036         dsl_pool_t *dp = dmu_tx_pool(tx);
1037         nvpair_t *pair;
1038         int rv = 0;
1039
1040         /*
1041          * Pre-compute how many total new snapshots will be created for each
1042          * level in the tree and below. This is needed for validating the
1043          * snapshot limit when either taking a recursive snapshot or when
1044          * taking multiple snapshots.
1045          *
1046          * The problem is that the counts are not actually adjusted when
1047          * we are checking, only when we finally sync. For a single snapshot,
1048          * this is easy, the count will increase by 1 at each node up the tree,
1049          * but its more complicated for the recursive/multiple snapshot case.
1050          *
1051          * The dsl_fs_ss_limit_check function does recursively check the count
1052          * at each level up the tree but since it is validating each snapshot
1053          * independently we need to be sure that we are validating the complete
1054          * count for the entire set of snapshots. We do this by rolling up the
1055          * counts for each component of the name into an nvlist and then
1056          * checking each of those cases with the aggregated count.
1057          *
1058          * This approach properly handles not only the recursive snapshot
1059          * case (where we get all of those on the ddsa_snaps list) but also
1060          * the sibling case (e.g. snapshot a/b and a/c so that we will also
1061          * validate the limit on 'a' using a count of 2).
1062          *
1063          * We validate the snapshot names in the third loop and only report
1064          * name errors once.
1065          */
1066         if (dmu_tx_is_syncing(tx)) {
1067                 char *nm;
1068                 nvlist_t *cnt_track = NULL;
1069                 cnt_track = fnvlist_alloc();
1070
1071                 nm = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1072
1073                 /* Rollup aggregated counts into the cnt_track list */
1074                 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1075                     pair != NULL;
1076                     pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1077                         char *pdelim;
1078                         uint64_t val;
1079
1080                         (void) strlcpy(nm, nvpair_name(pair), MAXPATHLEN);
1081                         pdelim = strchr(nm, '@');
1082                         if (pdelim == NULL)
1083                                 continue;
1084                         *pdelim = '\0';
1085
1086                         do {
1087                                 if (nvlist_lookup_uint64(cnt_track, nm,
1088                                     &val) == 0) {
1089                                         /* update existing entry */
1090                                         fnvlist_add_uint64(cnt_track, nm,
1091                                             val + 1);
1092                                 } else {
1093                                         /* add to list */
1094                                         fnvlist_add_uint64(cnt_track, nm, 1);
1095                                 }
1096
1097                                 pdelim = strrchr(nm, '/');
1098                                 if (pdelim != NULL)
1099                                         *pdelim = '\0';
1100                         } while (pdelim != NULL);
1101                 }
1102
1103                 kmem_free(nm, MAXPATHLEN);
1104
1105                 /* Check aggregated counts at each level */
1106                 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1107                     pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1108                         int error = 0;
1109                         char *name;
1110                         uint64_t cnt = 0;
1111                         dsl_dataset_t *ds;
1112
1113                         name = nvpair_name(pair);
1114                         cnt = fnvpair_value_uint64(pair);
1115                         ASSERT(cnt > 0);
1116
1117                         error = dsl_dataset_hold(dp, name, FTAG, &ds);
1118                         if (error == 0) {
1119                                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1120                                     ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1121                                     ddsa->ddsa_cr);
1122                                 dsl_dataset_rele(ds, FTAG);
1123                         }
1124
1125                         if (error != 0) {
1126                                 if (ddsa->ddsa_errors != NULL)
1127                                         fnvlist_add_int32(ddsa->ddsa_errors,
1128                                             name, error);
1129                                 rv = error;
1130                                 /* only report one error for this check */
1131                                 break;
1132                         }
1133                 }
1134                 nvlist_free(cnt_track);
1135         }
1136
1137         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1138             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1139                 int error = 0;
1140                 dsl_dataset_t *ds;
1141                 char *name, *atp;
1142                 char dsname[MAXNAMELEN];
1143
1144                 name = nvpair_name(pair);
1145                 if (strlen(name) >= MAXNAMELEN)
1146                         error = SET_ERROR(ENAMETOOLONG);
1147                 if (error == 0) {
1148                         atp = strchr(name, '@');
1149                         if (atp == NULL)
1150                                 error = SET_ERROR(EINVAL);
1151                         if (error == 0)
1152                                 (void) strlcpy(dsname, name, atp - name + 1);
1153                 }
1154                 if (error == 0)
1155                         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1156                 if (error == 0) {
1157                         /* passing 0/NULL skips dsl_fs_ss_limit_check */
1158                         error = dsl_dataset_snapshot_check_impl(ds,
1159                             atp + 1, tx, B_FALSE, 0, NULL);
1160                         dsl_dataset_rele(ds, FTAG);
1161                 }
1162
1163                 if (error != 0) {
1164                         if (ddsa->ddsa_errors != NULL) {
1165                                 fnvlist_add_int32(ddsa->ddsa_errors,
1166                                     name, error);
1167                         }
1168                         rv = error;
1169                 }
1170         }
1171
1172         return (rv);
1173 }
1174
1175 void
1176 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1177     dmu_tx_t *tx)
1178 {
1179         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1180         dmu_buf_t *dbuf;
1181         dsl_dataset_phys_t *dsphys;
1182         uint64_t dsobj, crtxg;
1183         objset_t *mos = dp->dp_meta_objset;
1184         ASSERTV(static zil_header_t zero_zil);
1185         ASSERTV(objset_t *os);
1186
1187         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1188
1189         /*
1190          * If we are on an old pool, the zil must not be active, in which
1191          * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1192          */
1193         ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1194             dmu_objset_from_ds(ds, &os) != 0 ||
1195             bcmp(&os->os_phys->os_zil_header, &zero_zil,
1196             sizeof (zero_zil)) == 0);
1197
1198         dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1199
1200         /*
1201          * The origin's ds_creation_txg has to be < TXG_INITIAL
1202          */
1203         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1204                 crtxg = 1;
1205         else
1206                 crtxg = tx->tx_txg;
1207
1208         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1209             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1210         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1211         dmu_buf_will_dirty(dbuf, tx);
1212         dsphys = dbuf->db_data;
1213         bzero(dsphys, sizeof (dsl_dataset_phys_t));
1214         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1215         dsphys->ds_fsid_guid = unique_create();
1216         (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1217             sizeof (dsphys->ds_guid));
1218         dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1219         dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1220         dsphys->ds_next_snap_obj = ds->ds_object;
1221         dsphys->ds_num_children = 1;
1222         dsphys->ds_creation_time = gethrestime_sec();
1223         dsphys->ds_creation_txg = crtxg;
1224         dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1225         dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1226         dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1227         dsphys->ds_uncompressed_bytes =
1228             dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1229         dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1230         dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1231         dmu_buf_rele(dbuf, FTAG);
1232
1233         ASSERT3U(ds->ds_prev != 0, ==,
1234             dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1235         if (ds->ds_prev) {
1236                 uint64_t next_clones_obj =
1237                     dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1238                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1239                     ds->ds_object ||
1240                     dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1241                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1242                     ds->ds_object) {
1243                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1244                         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1245                             dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1246                         dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1247                 } else if (next_clones_obj != 0) {
1248                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
1249                             dsphys->ds_next_snap_obj, tx);
1250                         VERIFY0(zap_add_int(mos,
1251                             next_clones_obj, dsobj, tx));
1252                 }
1253         }
1254
1255         /*
1256          * If we have a reference-reservation on this dataset, we will
1257          * need to increase the amount of refreservation being charged
1258          * since our unique space is going to zero.
1259          */
1260         if (ds->ds_reserved) {
1261                 int64_t delta;
1262                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1263                 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1264                     ds->ds_reserved);
1265                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1266                     delta, 0, 0, tx);
1267         }
1268
1269         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1270         dsl_dataset_phys(ds)->ds_deadlist_obj =
1271             dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1272             dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1273         dsl_deadlist_close(&ds->ds_deadlist);
1274         dsl_deadlist_open(&ds->ds_deadlist, mos,
1275             dsl_dataset_phys(ds)->ds_deadlist_obj);
1276         dsl_deadlist_add_key(&ds->ds_deadlist,
1277             dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1278
1279         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1280         dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1281         dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1282         dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1283         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1284                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1285
1286         VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1287             snapname, 8, 1, &dsobj, tx));
1288
1289         if (ds->ds_prev)
1290                 dsl_dataset_rele(ds->ds_prev, ds);
1291         VERIFY0(dsl_dataset_hold_obj(dp,
1292             dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1293
1294         dsl_scan_ds_snapshotted(ds, tx);
1295
1296         dsl_dir_snap_cmtime_update(ds->ds_dir);
1297
1298         spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1299 }
1300
1301 static void
1302 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1303 {
1304         dsl_dataset_snapshot_arg_t *ddsa = arg;
1305         dsl_pool_t *dp = dmu_tx_pool(tx);
1306         nvpair_t *pair;
1307
1308         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1309             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1310                 dsl_dataset_t *ds;
1311                 char *name, *atp;
1312                 char dsname[MAXNAMELEN];
1313
1314                 name = nvpair_name(pair);
1315                 atp = strchr(name, '@');
1316                 (void) strlcpy(dsname, name, atp - name + 1);
1317                 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1318
1319                 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1320                 if (ddsa->ddsa_props != NULL) {
1321                         dsl_props_set_sync_impl(ds->ds_prev,
1322                             ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1323                 }
1324                 dsl_dataset_rele(ds, FTAG);
1325         }
1326 }
1327
1328 /*
1329  * The snapshots must all be in the same pool.
1330  * All-or-nothing: if there are any failures, nothing will be modified.
1331  */
1332 int
1333 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1334 {
1335         dsl_dataset_snapshot_arg_t ddsa;
1336         nvpair_t *pair;
1337         boolean_t needsuspend;
1338         int error;
1339         spa_t *spa;
1340         char *firstname;
1341         nvlist_t *suspended = NULL;
1342
1343         pair = nvlist_next_nvpair(snaps, NULL);
1344         if (pair == NULL)
1345                 return (0);
1346         firstname = nvpair_name(pair);
1347
1348         error = spa_open(firstname, &spa, FTAG);
1349         if (error != 0)
1350                 return (error);
1351         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1352         spa_close(spa, FTAG);
1353
1354         if (needsuspend) {
1355                 suspended = fnvlist_alloc();
1356                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1357                     pair = nvlist_next_nvpair(snaps, pair)) {
1358                         char fsname[MAXNAMELEN];
1359                         char *snapname = nvpair_name(pair);
1360                         char *atp;
1361                         void *cookie;
1362
1363                         atp = strchr(snapname, '@');
1364                         if (atp == NULL) {
1365                                 error = SET_ERROR(EINVAL);
1366                                 break;
1367                         }
1368                         (void) strlcpy(fsname, snapname, atp - snapname + 1);
1369
1370                         error = zil_suspend(fsname, &cookie);
1371                         if (error != 0)
1372                                 break;
1373                         fnvlist_add_uint64(suspended, fsname,
1374                             (uintptr_t)cookie);
1375                 }
1376         }
1377
1378         ddsa.ddsa_snaps = snaps;
1379         ddsa.ddsa_props = props;
1380         ddsa.ddsa_errors = errors;
1381         ddsa.ddsa_cr = CRED();
1382
1383         if (error == 0) {
1384                 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1385                     dsl_dataset_snapshot_sync, &ddsa,
1386                     fnvlist_num_pairs(snaps) * 3);
1387         }
1388
1389         if (suspended != NULL) {
1390                 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1391                     pair = nvlist_next_nvpair(suspended, pair)) {
1392                         zil_resume((void *)(uintptr_t)
1393                             fnvpair_value_uint64(pair));
1394                 }
1395                 fnvlist_free(suspended);
1396         }
1397
1398 #ifdef _KERNEL
1399         if (error == 0) {
1400                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1401                     pair = nvlist_next_nvpair(snaps, pair)) {
1402                         char *snapname = nvpair_name(pair);
1403                         zvol_create_minors(snapname);
1404                 }
1405         }
1406 #endif
1407
1408         return (error);
1409 }
1410
1411 typedef struct dsl_dataset_snapshot_tmp_arg {
1412         const char *ddsta_fsname;
1413         const char *ddsta_snapname;
1414         minor_t ddsta_cleanup_minor;
1415         const char *ddsta_htag;
1416 } dsl_dataset_snapshot_tmp_arg_t;
1417
1418 static int
1419 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1420 {
1421         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1422         dsl_pool_t *dp = dmu_tx_pool(tx);
1423         dsl_dataset_t *ds;
1424         int error;
1425
1426         error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1427         if (error != 0)
1428                 return (error);
1429
1430         /* NULL cred means no limit check for tmp snapshot */
1431         error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1432             tx, B_FALSE, 0, NULL);
1433         if (error != 0) {
1434                 dsl_dataset_rele(ds, FTAG);
1435                 return (error);
1436         }
1437
1438         if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1439                 dsl_dataset_rele(ds, FTAG);
1440                 return (SET_ERROR(ENOTSUP));
1441         }
1442         error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1443             B_TRUE, tx);
1444         if (error != 0) {
1445                 dsl_dataset_rele(ds, FTAG);
1446                 return (error);
1447         }
1448
1449         dsl_dataset_rele(ds, FTAG);
1450         return (0);
1451 }
1452
1453 static void
1454 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1455 {
1456         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1457         dsl_pool_t *dp = dmu_tx_pool(tx);
1458         dsl_dataset_t *ds;
1459
1460         VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1461
1462         dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1463         dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1464             ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1465         dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1466
1467         dsl_dataset_rele(ds, FTAG);
1468 }
1469
1470 int
1471 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1472     minor_t cleanup_minor, const char *htag)
1473 {
1474         dsl_dataset_snapshot_tmp_arg_t ddsta;
1475         int error;
1476         spa_t *spa;
1477         boolean_t needsuspend;
1478         void *cookie;
1479
1480         ddsta.ddsta_fsname = fsname;
1481         ddsta.ddsta_snapname = snapname;
1482         ddsta.ddsta_cleanup_minor = cleanup_minor;
1483         ddsta.ddsta_htag = htag;
1484
1485         error = spa_open(fsname, &spa, FTAG);
1486         if (error != 0)
1487                 return (error);
1488         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1489         spa_close(spa, FTAG);
1490
1491         if (needsuspend) {
1492                 error = zil_suspend(fsname, &cookie);
1493                 if (error != 0)
1494                         return (error);
1495         }
1496
1497         error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1498             dsl_dataset_snapshot_tmp_sync, &ddsta, 3);
1499
1500         if (needsuspend)
1501                 zil_resume(cookie);
1502         return (error);
1503 }
1504
1505
1506 void
1507 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1508 {
1509         ASSERT(dmu_tx_is_syncing(tx));
1510         ASSERT(ds->ds_objset != NULL);
1511         ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1512
1513         /*
1514          * in case we had to change ds_fsid_guid when we opened it,
1515          * sync it out now.
1516          */
1517         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1518         dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1519
1520         dmu_objset_sync(ds->ds_objset, zio, tx);
1521 }
1522
1523 static void
1524 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1525 {
1526         uint64_t count = 0;
1527         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1528         zap_cursor_t zc;
1529         zap_attribute_t za;
1530         nvlist_t *propval = fnvlist_alloc();
1531         nvlist_t *val = fnvlist_alloc();
1532
1533         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1534
1535         /*
1536          * There may be missing entries in ds_next_clones_obj
1537          * due to a bug in a previous version of the code.
1538          * Only trust it if it has the right number of entries.
1539          */
1540         if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1541                 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1542                     &count));
1543         }
1544         if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1545                 goto fail;
1546         for (zap_cursor_init(&zc, mos,
1547             dsl_dataset_phys(ds)->ds_next_clones_obj);
1548             zap_cursor_retrieve(&zc, &za) == 0;
1549             zap_cursor_advance(&zc)) {
1550                 dsl_dataset_t *clone;
1551                 char buf[ZFS_MAXNAMELEN];
1552                 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1553                     za.za_first_integer, FTAG, &clone));
1554                 dsl_dir_name(clone->ds_dir, buf);
1555                 fnvlist_add_boolean(val, buf);
1556                 dsl_dataset_rele(clone, FTAG);
1557         }
1558         zap_cursor_fini(&zc);
1559         fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1560         fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1561 fail:
1562         nvlist_free(val);
1563         nvlist_free(propval);
1564 }
1565
1566 void
1567 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1568 {
1569         uint64_t refd, avail, uobjs, aobjs, ratio;
1570         ASSERTV(dsl_pool_t *dp = ds->ds_dir->dd_pool);
1571
1572         ASSERT(dsl_pool_config_held(dp));
1573
1574         ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1575             (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1576             dsl_dataset_phys(ds)->ds_compressed_bytes);
1577
1578         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1579         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1580             dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1581
1582         if (dsl_dataset_is_snapshot(ds)) {
1583                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1584                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1585                     dsl_dataset_phys(ds)->ds_unique_bytes);
1586                 get_clones_stat(ds, nv);
1587         } else {
1588                 dsl_dir_stats(ds->ds_dir, nv);
1589         }
1590
1591         dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1592         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1593         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1594
1595         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1596             dsl_dataset_phys(ds)->ds_creation_time);
1597         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1598             dsl_dataset_phys(ds)->ds_creation_txg);
1599         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1600             ds->ds_quota);
1601         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1602             ds->ds_reserved);
1603         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1604             dsl_dataset_phys(ds)->ds_guid);
1605         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1606             dsl_dataset_phys(ds)->ds_unique_bytes);
1607         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1608             ds->ds_object);
1609         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1610             ds->ds_userrefs);
1611         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1612             DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1613
1614         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1615                 uint64_t written, comp, uncomp;
1616                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1617                 dsl_dataset_t *prev;
1618                 int err;
1619
1620                 err = dsl_dataset_hold_obj(dp,
1621                     dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1622                 if (err == 0) {
1623                         err = dsl_dataset_space_written(prev, ds, &written,
1624                             &comp, &uncomp);
1625                         dsl_dataset_rele(prev, FTAG);
1626                         if (err == 0) {
1627                                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1628                                     written);
1629                         }
1630                 }
1631         }
1632
1633 }
1634
1635 void
1636 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1637 {
1638         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1639         ASSERT(dsl_pool_config_held(dp));
1640
1641         stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1642         stat->dds_inconsistent =
1643             dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1644         stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1645         stat->dds_origin[0] = '\0';
1646         if (dsl_dataset_is_snapshot(ds)) {
1647                 stat->dds_is_snapshot = B_TRUE;
1648                 stat->dds_num_clones =
1649                     dsl_dataset_phys(ds)->ds_num_children - 1;
1650         } else {
1651                 stat->dds_is_snapshot = B_FALSE;
1652                 stat->dds_num_clones = 0;
1653
1654                 if (dsl_dir_is_clone(ds->ds_dir)) {
1655                         dsl_dataset_t *ods;
1656
1657                         VERIFY0(dsl_dataset_hold_obj(dp,
1658                             dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1659                             FTAG, &ods));
1660                         dsl_dataset_name(ods, stat->dds_origin);
1661                         dsl_dataset_rele(ods, FTAG);
1662                 }
1663         }
1664 }
1665
1666 uint64_t
1667 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1668 {
1669         return (ds->ds_fsid_guid);
1670 }
1671
1672 void
1673 dsl_dataset_space(dsl_dataset_t *ds,
1674     uint64_t *refdbytesp, uint64_t *availbytesp,
1675     uint64_t *usedobjsp, uint64_t *availobjsp)
1676 {
1677         *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1678         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1679         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1680                 *availbytesp +=
1681                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1682         if (ds->ds_quota != 0) {
1683                 /*
1684                  * Adjust available bytes according to refquota
1685                  */
1686                 if (*refdbytesp < ds->ds_quota)
1687                         *availbytesp = MIN(*availbytesp,
1688                             ds->ds_quota - *refdbytesp);
1689                 else
1690                         *availbytesp = 0;
1691         }
1692         *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1693         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1694 }
1695
1696 boolean_t
1697 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1698 {
1699         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1700         if (snap == NULL)
1701                 return (B_FALSE);
1702         if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1703             dsl_dataset_phys(snap)->ds_creation_txg) {
1704                 objset_t *os, *os_snap;
1705                 /*
1706                  * It may be that only the ZIL differs, because it was
1707                  * reset in the head.  Don't count that as being
1708                  * modified.
1709                  */
1710                 if (dmu_objset_from_ds(ds, &os) != 0)
1711                         return (B_TRUE);
1712                 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1713                         return (B_TRUE);
1714                 return (bcmp(&os->os_phys->os_meta_dnode,
1715                     &os_snap->os_phys->os_meta_dnode,
1716                     sizeof (os->os_phys->os_meta_dnode)) != 0);
1717         }
1718         return (B_FALSE);
1719 }
1720
1721 typedef struct dsl_dataset_rename_snapshot_arg {
1722         const char *ddrsa_fsname;
1723         const char *ddrsa_oldsnapname;
1724         const char *ddrsa_newsnapname;
1725         boolean_t ddrsa_recursive;
1726         dmu_tx_t *ddrsa_tx;
1727 } dsl_dataset_rename_snapshot_arg_t;
1728
1729 /* ARGSUSED */
1730 static int
1731 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1732     dsl_dataset_t *hds, void *arg)
1733 {
1734         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1735         int error;
1736         uint64_t val;
1737
1738         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1739         if (error != 0) {
1740                 /* ignore nonexistent snapshots */
1741                 return (error == ENOENT ? 0 : error);
1742         }
1743
1744         /* new name should not exist */
1745         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1746         if (error == 0)
1747                 error = SET_ERROR(EEXIST);
1748         else if (error == ENOENT)
1749                 error = 0;
1750
1751         /* dataset name + 1 for the "@" + the new snapshot name must fit */
1752         if (dsl_dir_namelen(hds->ds_dir) + 1 +
1753             strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1754                 error = SET_ERROR(ENAMETOOLONG);
1755
1756         return (error);
1757 }
1758
1759 static int
1760 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1761 {
1762         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1763         dsl_pool_t *dp = dmu_tx_pool(tx);
1764         dsl_dataset_t *hds;
1765         int error;
1766
1767         error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1768         if (error != 0)
1769                 return (error);
1770
1771         if (ddrsa->ddrsa_recursive) {
1772                 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1773                     dsl_dataset_rename_snapshot_check_impl, ddrsa,
1774                     DS_FIND_CHILDREN);
1775         } else {
1776                 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1777         }
1778         dsl_dataset_rele(hds, FTAG);
1779         return (error);
1780 }
1781
1782 static int
1783 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1784     dsl_dataset_t *hds, void *arg)
1785 {
1786         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1787         dsl_dataset_t *ds;
1788         uint64_t val;
1789         dmu_tx_t *tx = ddrsa->ddrsa_tx;
1790         int error;
1791
1792         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1793         ASSERT(error == 0 || error == ENOENT);
1794         if (error == ENOENT) {
1795                 /* ignore nonexistent snapshots */
1796                 return (0);
1797         }
1798
1799         VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1800
1801         /* log before we change the name */
1802         spa_history_log_internal_ds(ds, "rename", tx,
1803             "-> @%s", ddrsa->ddrsa_newsnapname);
1804
1805         VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1806             B_FALSE));
1807         mutex_enter(&ds->ds_lock);
1808         (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1809         mutex_exit(&ds->ds_lock);
1810         VERIFY0(zap_add(dp->dp_meta_objset,
1811             dsl_dataset_phys(hds)->ds_snapnames_zapobj,
1812             ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1813
1814         dsl_dataset_rele(ds, FTAG);
1815         return (0);
1816 }
1817
1818 static void
1819 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1820 {
1821         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1822         dsl_pool_t *dp = dmu_tx_pool(tx);
1823         dsl_dataset_t *hds;
1824
1825         VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1826         ddrsa->ddrsa_tx = tx;
1827         if (ddrsa->ddrsa_recursive) {
1828                 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1829                     dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1830                     DS_FIND_CHILDREN));
1831         } else {
1832                 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1833         }
1834         dsl_dataset_rele(hds, FTAG);
1835 }
1836
1837 int
1838 dsl_dataset_rename_snapshot(const char *fsname,
1839     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1840 {
1841 #ifdef _KERNEL
1842         char *oldname, *newname;
1843 #endif
1844         int error;
1845
1846         dsl_dataset_rename_snapshot_arg_t ddrsa;
1847
1848         ddrsa.ddrsa_fsname = fsname;
1849         ddrsa.ddrsa_oldsnapname = oldsnapname;
1850         ddrsa.ddrsa_newsnapname = newsnapname;
1851         ddrsa.ddrsa_recursive = recursive;
1852
1853         error = dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1854             dsl_dataset_rename_snapshot_sync, &ddrsa, 1);
1855
1856         if (error)
1857             return (SET_ERROR(error));
1858
1859 #ifdef _KERNEL
1860         oldname = kmem_asprintf("%s@%s", fsname, oldsnapname);
1861         newname = kmem_asprintf("%s@%s", fsname, newsnapname);
1862         zvol_rename_minors(oldname, newname);
1863         strfree(newname);
1864         strfree(oldname);
1865 #endif
1866
1867         return (0);
1868 }
1869
1870 /*
1871  * If we're doing an ownership handoff, we need to make sure that there is
1872  * only one long hold on the dataset.  We're not allowed to change anything here
1873  * so we don't permanently release the long hold or regular hold here.  We want
1874  * to do this only when syncing to avoid the dataset unexpectedly going away
1875  * when we release the long hold.
1876  */
1877 static int
1878 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1879 {
1880         boolean_t held;
1881
1882         if (!dmu_tx_is_syncing(tx))
1883                 return (0);
1884
1885         if (owner != NULL) {
1886                 VERIFY3P(ds->ds_owner, ==, owner);
1887                 dsl_dataset_long_rele(ds, owner);
1888         }
1889
1890         held = dsl_dataset_long_held(ds);
1891
1892         if (owner != NULL)
1893                 dsl_dataset_long_hold(ds, owner);
1894
1895         if (held)
1896                 return (SET_ERROR(EBUSY));
1897
1898         return (0);
1899 }
1900
1901 typedef struct dsl_dataset_rollback_arg {
1902         const char *ddra_fsname;
1903         void *ddra_owner;
1904         nvlist_t *ddra_result;
1905 } dsl_dataset_rollback_arg_t;
1906
1907 static int
1908 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1909 {
1910         dsl_dataset_rollback_arg_t *ddra = arg;
1911         dsl_pool_t *dp = dmu_tx_pool(tx);
1912         dsl_dataset_t *ds;
1913         int64_t unused_refres_delta;
1914         int error;
1915         nvpair_t *pair;
1916         nvlist_t *proprequest, *bookmarks;
1917
1918         error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1919         if (error != 0)
1920                 return (error);
1921
1922         /* must not be a snapshot */
1923         if (dsl_dataset_is_snapshot(ds)) {
1924                 dsl_dataset_rele(ds, FTAG);
1925                 return (SET_ERROR(EINVAL));
1926         }
1927
1928         /* must have a most recent snapshot */
1929         if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
1930                 dsl_dataset_rele(ds, FTAG);
1931                 return (SET_ERROR(EINVAL));
1932         }
1933
1934         /* must not have any bookmarks after the most recent snapshot */
1935         proprequest = fnvlist_alloc();
1936         fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
1937         bookmarks = fnvlist_alloc();
1938         error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
1939         fnvlist_free(proprequest);
1940         if (error != 0)
1941                 return (error);
1942         for (pair = nvlist_next_nvpair(bookmarks, NULL);
1943             pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
1944                 nvlist_t *valuenv =
1945                     fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
1946                     zfs_prop_to_name(ZFS_PROP_CREATETXG));
1947                 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
1948                 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
1949                         fnvlist_free(bookmarks);
1950                         dsl_dataset_rele(ds, FTAG);
1951                         return (SET_ERROR(EEXIST));
1952                 }
1953         }
1954         fnvlist_free(bookmarks);
1955
1956         error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
1957         if (error != 0) {
1958                 dsl_dataset_rele(ds, FTAG);
1959                 return (error);
1960         }
1961
1962         /*
1963          * Check if the snap we are rolling back to uses more than
1964          * the refquota.
1965          */
1966         if (ds->ds_quota != 0 &&
1967             dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
1968                 dsl_dataset_rele(ds, FTAG);
1969                 return (SET_ERROR(EDQUOT));
1970         }
1971
1972         /*
1973          * When we do the clone swap, we will temporarily use more space
1974          * due to the refreservation (the head will no longer have any
1975          * unique space, so the entire amount of the refreservation will need
1976          * to be free).  We will immediately destroy the clone, freeing
1977          * this space, but the freeing happens over many txg's.
1978          */
1979         unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1980             dsl_dataset_phys(ds)->ds_unique_bytes);
1981
1982         if (unused_refres_delta > 0 &&
1983             unused_refres_delta >
1984             dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1985                 dsl_dataset_rele(ds, FTAG);
1986                 return (SET_ERROR(ENOSPC));
1987         }
1988
1989         dsl_dataset_rele(ds, FTAG);
1990         return (0);
1991 }
1992
1993 static void
1994 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1995 {
1996         dsl_dataset_rollback_arg_t *ddra = arg;
1997         dsl_pool_t *dp = dmu_tx_pool(tx);
1998         dsl_dataset_t *ds, *clone;
1999         uint64_t cloneobj;
2000         char namebuf[ZFS_MAXNAMELEN];
2001
2002         VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2003
2004         dsl_dataset_name(ds->ds_prev, namebuf);
2005         fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2006
2007         cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2008             ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2009
2010         VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2011
2012         dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2013         dsl_dataset_zero_zil(ds, tx);
2014
2015         dsl_destroy_head_sync_impl(clone, tx);
2016
2017         dsl_dataset_rele(clone, FTAG);
2018         dsl_dataset_rele(ds, FTAG);
2019 }
2020
2021 /*
2022  * Rolls back the given filesystem or volume to the most recent snapshot.
2023  * The name of the most recent snapshot will be returned under key "target"
2024  * in the result nvlist.
2025  *
2026  * If owner != NULL:
2027  * - The existing dataset MUST be owned by the specified owner at entry
2028  * - Upon return, dataset will still be held by the same owner, whether we
2029  *   succeed or not.
2030  *
2031  * This mode is required any time the existing filesystem is mounted.  See
2032  * notes above zfs_suspend_fs() for further details.
2033  */
2034 int
2035 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2036 {
2037         dsl_dataset_rollback_arg_t ddra;
2038
2039         ddra.ddra_fsname = fsname;
2040         ddra.ddra_owner = owner;
2041         ddra.ddra_result = result;
2042
2043         return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2044             dsl_dataset_rollback_sync, &ddra, 1));
2045 }
2046
2047 struct promotenode {
2048         list_node_t link;
2049         dsl_dataset_t *ds;
2050 };
2051
2052 typedef struct dsl_dataset_promote_arg {
2053         const char *ddpa_clonename;
2054         dsl_dataset_t *ddpa_clone;
2055         list_t shared_snaps, origin_snaps, clone_snaps;
2056         dsl_dataset_t *origin_origin; /* origin of the origin */
2057         uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2058         char *err_ds;
2059         cred_t *cr;
2060 } dsl_dataset_promote_arg_t;
2061
2062 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2063 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2064     void *tag);
2065 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2066
2067 static int
2068 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2069 {
2070         dsl_dataset_promote_arg_t *ddpa = arg;
2071         dsl_pool_t *dp = dmu_tx_pool(tx);
2072         dsl_dataset_t *hds;
2073         struct promotenode *snap;
2074         dsl_dataset_t *origin_ds;
2075         int err;
2076         uint64_t unused;
2077         uint64_t ss_mv_cnt;
2078
2079         err = promote_hold(ddpa, dp, FTAG);
2080         if (err != 0)
2081                 return (err);
2082
2083         hds = ddpa->ddpa_clone;
2084
2085         if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2086                 promote_rele(ddpa, FTAG);
2087                 return (SET_ERROR(EXDEV));
2088         }
2089
2090         /*
2091          * Compute and check the amount of space to transfer.  Since this is
2092          * so expensive, don't do the preliminary check.
2093          */
2094         if (!dmu_tx_is_syncing(tx)) {
2095                 promote_rele(ddpa, FTAG);
2096                 return (0);
2097         }
2098
2099         snap = list_head(&ddpa->shared_snaps);
2100         origin_ds = snap->ds;
2101
2102         /* compute origin's new unique space */
2103         snap = list_tail(&ddpa->clone_snaps);
2104         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2105             origin_ds->ds_object);
2106         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2107             dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2108             &ddpa->unique, &unused, &unused);
2109
2110         /*
2111          * Walk the snapshots that we are moving
2112          *
2113          * Compute space to transfer.  Consider the incremental changes
2114          * to used by each snapshot:
2115          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2116          * So each snapshot gave birth to:
2117          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2118          * So a sequence would look like:
2119          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2120          * Which simplifies to:
2121          * uN + kN + kN-1 + ... + k1 + k0
2122          * Note however, if we stop before we reach the ORIGIN we get:
2123          * uN + kN + kN-1 + ... + kM - uM-1
2124          */
2125         ss_mv_cnt = 0;
2126         ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2127         ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2128         ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2129         for (snap = list_head(&ddpa->shared_snaps); snap;
2130             snap = list_next(&ddpa->shared_snaps, snap)) {
2131                 uint64_t val, dlused, dlcomp, dluncomp;
2132                 dsl_dataset_t *ds = snap->ds;
2133
2134                 ss_mv_cnt++;
2135
2136                 /*
2137                  * If there are long holds, we won't be able to evict
2138                  * the objset.
2139                  */
2140                 if (dsl_dataset_long_held(ds)) {
2141                         err = SET_ERROR(EBUSY);
2142                         goto out;
2143                 }
2144
2145                 /* Check that the snapshot name does not conflict */
2146                 VERIFY0(dsl_dataset_get_snapname(ds));
2147                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2148                 if (err == 0) {
2149                         (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2150                         err = SET_ERROR(EEXIST);
2151                         goto out;
2152                 }
2153                 if (err != ENOENT)
2154                         goto out;
2155
2156                 /* The very first snapshot does not have a deadlist */
2157                 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2158                         continue;
2159
2160                 dsl_deadlist_space(&ds->ds_deadlist,
2161                     &dlused, &dlcomp, &dluncomp);
2162                 ddpa->used += dlused;
2163                 ddpa->comp += dlcomp;
2164                 ddpa->uncomp += dluncomp;
2165         }
2166
2167         /*
2168          * If we are a clone of a clone then we never reached ORIGIN,
2169          * so we need to subtract out the clone origin's used space.
2170          */
2171         if (ddpa->origin_origin) {
2172                 ddpa->used -=
2173                     dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2174                 ddpa->comp -=
2175                     dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2176                 ddpa->uncomp -=
2177                     dsl_dataset_phys(ddpa->origin_origin)->
2178                     ds_uncompressed_bytes;
2179         }
2180
2181         /* Check that there is enough space and limit headroom here */
2182         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2183             0, ss_mv_cnt, ddpa->used, ddpa->cr);
2184         if (err != 0)
2185                 goto out;
2186
2187         /*
2188          * Compute the amounts of space that will be used by snapshots
2189          * after the promotion (for both origin and clone).  For each,
2190          * it is the amount of space that will be on all of their
2191          * deadlists (that was not born before their new origin).
2192          */
2193         if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2194                 uint64_t space;
2195
2196                 /*
2197                  * Note, typically this will not be a clone of a clone,
2198                  * so dd_origin_txg will be < TXG_INITIAL, so
2199                  * these snaplist_space() -> dsl_deadlist_space_range()
2200                  * calls will be fast because they do not have to
2201                  * iterate over all bps.
2202                  */
2203                 snap = list_head(&ddpa->origin_snaps);
2204                 err = snaplist_space(&ddpa->shared_snaps,
2205                     snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2206                 if (err != 0)
2207                         goto out;
2208
2209                 err = snaplist_space(&ddpa->clone_snaps,
2210                     snap->ds->ds_dir->dd_origin_txg, &space);
2211                 if (err != 0)
2212                         goto out;
2213                 ddpa->cloneusedsnap += space;
2214         }
2215         if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2216             DD_FLAG_USED_BREAKDOWN) {
2217                 err = snaplist_space(&ddpa->origin_snaps,
2218                     dsl_dataset_phys(origin_ds)->ds_creation_txg,
2219                     &ddpa->originusedsnap);
2220                 if (err != 0)
2221                         goto out;
2222         }
2223
2224 out:
2225         promote_rele(ddpa, FTAG);
2226         return (err);
2227 }
2228
2229 static void
2230 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2231 {
2232         dsl_dataset_promote_arg_t *ddpa = arg;
2233         dsl_pool_t *dp = dmu_tx_pool(tx);
2234         dsl_dataset_t *hds;
2235         struct promotenode *snap;
2236         dsl_dataset_t *origin_ds;
2237         dsl_dataset_t *origin_head;
2238         dsl_dir_t *dd;
2239         dsl_dir_t *odd = NULL;
2240         uint64_t oldnext_obj;
2241         int64_t delta;
2242
2243         VERIFY0(promote_hold(ddpa, dp, FTAG));
2244         hds = ddpa->ddpa_clone;
2245
2246         ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2247
2248         snap = list_head(&ddpa->shared_snaps);
2249         origin_ds = snap->ds;
2250         dd = hds->ds_dir;
2251
2252         snap = list_head(&ddpa->origin_snaps);
2253         origin_head = snap->ds;
2254
2255         /*
2256          * We need to explicitly open odd, since origin_ds's dd will be
2257          * changing.
2258          */
2259         VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2260             NULL, FTAG, &odd));
2261
2262         /* change origin's next snap */
2263         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2264         oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2265         snap = list_tail(&ddpa->clone_snaps);
2266         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2267             origin_ds->ds_object);
2268         dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2269
2270         /* change the origin's next clone */
2271         if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2272                 dsl_dataset_remove_from_next_clones(origin_ds,
2273                     snap->ds->ds_object, tx);
2274                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2275                     dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2276                     oldnext_obj, tx));
2277         }
2278
2279         /* change origin */
2280         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2281         ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2282         dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2283         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2284         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2285         dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2286         origin_head->ds_dir->dd_origin_txg =
2287             dsl_dataset_phys(origin_ds)->ds_creation_txg;
2288
2289         /* change dd_clone entries */
2290         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2291                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2292                     dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2293                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2294                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2295                     hds->ds_object, tx));
2296
2297                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2298                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2299                     origin_head->ds_object, tx));
2300                 if (dsl_dir_phys(dd)->dd_clones == 0) {
2301                         dsl_dir_phys(dd)->dd_clones =
2302                             zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2303                             DMU_OT_NONE, 0, tx);
2304                 }
2305                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2306                     dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2307         }
2308
2309         /* move snapshots to this dir */
2310         for (snap = list_head(&ddpa->shared_snaps); snap;
2311             snap = list_next(&ddpa->shared_snaps, snap)) {
2312                 dsl_dataset_t *ds = snap->ds;
2313
2314                 /*
2315                  * Property callbacks are registered to a particular
2316                  * dsl_dir.  Since ours is changing, evict the objset
2317                  * so that they will be unregistered from the old dsl_dir.
2318                  */
2319                 if (ds->ds_objset) {
2320                         dmu_objset_evict(ds->ds_objset);
2321                         ds->ds_objset = NULL;
2322                 }
2323
2324                 /* move snap name entry */
2325                 VERIFY0(dsl_dataset_get_snapname(ds));
2326                 VERIFY0(dsl_dataset_snap_remove(origin_head,
2327                     ds->ds_snapname, tx, B_TRUE));
2328                 VERIFY0(zap_add(dp->dp_meta_objset,
2329                     dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2330                     8, 1, &ds->ds_object, tx));
2331                 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2332                     DD_FIELD_SNAPSHOT_COUNT, tx);
2333
2334                 /* change containing dsl_dir */
2335                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2336                 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2337                 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2338                 ASSERT3P(ds->ds_dir, ==, odd);
2339                 dsl_dir_rele(ds->ds_dir, ds);
2340                 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2341                     NULL, ds, &ds->ds_dir));
2342
2343                 /* move any clone references */
2344                 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2345                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2346                         zap_cursor_t zc;
2347                         zap_attribute_t za;
2348
2349                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
2350                             dsl_dataset_phys(ds)->ds_next_clones_obj);
2351                             zap_cursor_retrieve(&zc, &za) == 0;
2352                             zap_cursor_advance(&zc)) {
2353                                 dsl_dataset_t *cnds;
2354                                 uint64_t o;
2355
2356                                 if (za.za_first_integer == oldnext_obj) {
2357                                         /*
2358                                          * We've already moved the
2359                                          * origin's reference.
2360                                          */
2361                                         continue;
2362                                 }
2363
2364                                 VERIFY0(dsl_dataset_hold_obj(dp,
2365                                     za.za_first_integer, FTAG, &cnds));
2366                                 o = dsl_dir_phys(cnds->ds_dir)->
2367                                     dd_head_dataset_obj;
2368
2369                                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2370                                     dsl_dir_phys(odd)->dd_clones, o, tx));
2371                                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2372                                     dsl_dir_phys(dd)->dd_clones, o, tx));
2373                                 dsl_dataset_rele(cnds, FTAG);
2374                         }
2375                         zap_cursor_fini(&zc);
2376                 }
2377
2378                 ASSERT(!dsl_prop_hascb(ds));
2379         }
2380
2381         /*
2382          * Change space accounting.
2383          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2384          * both be valid, or both be 0 (resulting in delta == 0).  This
2385          * is true for each of {clone,origin} independently.
2386          */
2387
2388         delta = ddpa->cloneusedsnap -
2389             dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2390         ASSERT3S(delta, >=, 0);
2391         ASSERT3U(ddpa->used, >=, delta);
2392         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2393         dsl_dir_diduse_space(dd, DD_USED_HEAD,
2394             ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2395
2396         delta = ddpa->originusedsnap -
2397             dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2398         ASSERT3S(delta, <=, 0);
2399         ASSERT3U(ddpa->used, >=, -delta);
2400         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2401         dsl_dir_diduse_space(odd, DD_USED_HEAD,
2402             -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2403
2404         dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2405
2406         /* log history record */
2407         spa_history_log_internal_ds(hds, "promote", tx, "");
2408
2409         dsl_dir_rele(odd, FTAG);
2410         promote_rele(ddpa, FTAG);
2411 }
2412
2413 /*
2414  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2415  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2416  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2417  * snapshots back to this dataset's origin.
2418  */
2419 static int
2420 snaplist_make(dsl_pool_t *dp,
2421     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2422 {
2423         uint64_t obj = last_obj;
2424
2425         list_create(l, sizeof (struct promotenode),
2426             offsetof(struct promotenode, link));
2427
2428         while (obj != first_obj) {
2429                 dsl_dataset_t *ds;
2430                 struct promotenode *snap;
2431                 int err;
2432
2433                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2434                 ASSERT(err != ENOENT);
2435                 if (err != 0)
2436                         return (err);
2437
2438                 if (first_obj == 0)
2439                         first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2440
2441                 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2442                 snap->ds = ds;
2443                 list_insert_tail(l, snap);
2444                 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2445         }
2446
2447         return (0);
2448 }
2449
2450 static int
2451 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2452 {
2453         struct promotenode *snap;
2454
2455         *spacep = 0;
2456         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2457                 uint64_t used, comp, uncomp;
2458                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2459                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
2460                 *spacep += used;
2461         }
2462         return (0);
2463 }
2464
2465 static void
2466 snaplist_destroy(list_t *l, void *tag)
2467 {
2468         struct promotenode *snap;
2469
2470         if (l == NULL || !list_link_active(&l->list_head))
2471                 return;
2472
2473         while ((snap = list_tail(l)) != NULL) {
2474                 list_remove(l, snap);
2475                 dsl_dataset_rele(snap->ds, tag);
2476                 kmem_free(snap, sizeof (*snap));
2477         }
2478         list_destroy(l);
2479 }
2480
2481 static int
2482 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2483 {
2484         int error;
2485         dsl_dir_t *dd;
2486         struct promotenode *snap;
2487
2488         error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2489             &ddpa->ddpa_clone);
2490         if (error != 0)
2491                 return (error);
2492         dd = ddpa->ddpa_clone->ds_dir;
2493
2494         if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2495             !dsl_dir_is_clone(dd)) {
2496                 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2497                 return (SET_ERROR(EINVAL));
2498         }
2499
2500         error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2501             &ddpa->shared_snaps, tag);
2502         if (error != 0)
2503                 goto out;
2504
2505         error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2506             &ddpa->clone_snaps, tag);
2507         if (error != 0)
2508                 goto out;
2509
2510         snap = list_head(&ddpa->shared_snaps);
2511         ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2512         error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2513             dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2514             &ddpa->origin_snaps, tag);
2515         if (error != 0)
2516                 goto out;
2517
2518         if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2519                 error = dsl_dataset_hold_obj(dp,
2520                     dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2521                     tag, &ddpa->origin_origin);
2522                 if (error != 0)
2523                         goto out;
2524         }
2525 out:
2526         if (error != 0)
2527                 promote_rele(ddpa, tag);
2528         return (error);
2529 }
2530
2531 static void
2532 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2533 {
2534         snaplist_destroy(&ddpa->shared_snaps, tag);
2535         snaplist_destroy(&ddpa->clone_snaps, tag);
2536         snaplist_destroy(&ddpa->origin_snaps, tag);
2537         if (ddpa->origin_origin != NULL)
2538                 dsl_dataset_rele(ddpa->origin_origin, tag);
2539         dsl_dataset_rele(ddpa->ddpa_clone, tag);
2540 }
2541
2542 /*
2543  * Promote a clone.
2544  *
2545  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2546  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2547  */
2548 int
2549 dsl_dataset_promote(const char *name, char *conflsnap)
2550 {
2551         dsl_dataset_promote_arg_t ddpa = { 0 };
2552         uint64_t numsnaps;
2553         int error;
2554         objset_t *os;
2555
2556         /*
2557          * We will modify space proportional to the number of
2558          * snapshots.  Compute numsnaps.
2559          */
2560         error = dmu_objset_hold(name, FTAG, &os);
2561         if (error != 0)
2562                 return (error);
2563         error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2564             dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2565             &numsnaps);
2566         dmu_objset_rele(os, FTAG);
2567         if (error != 0)
2568                 return (error);
2569
2570         ddpa.ddpa_clonename = name;
2571         ddpa.err_ds = conflsnap;
2572         ddpa.cr = CRED();
2573
2574         return (dsl_sync_task(name, dsl_dataset_promote_check,
2575             dsl_dataset_promote_sync, &ddpa, 2 + numsnaps));
2576 }
2577
2578 int
2579 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2580     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2581 {
2582         int64_t unused_refres_delta;
2583
2584         /* they should both be heads */
2585         if (dsl_dataset_is_snapshot(clone) ||
2586             dsl_dataset_is_snapshot(origin_head))
2587                 return (SET_ERROR(EINVAL));
2588
2589         /* if we are not forcing, the branch point should be just before them */
2590         if (!force && clone->ds_prev != origin_head->ds_prev)
2591                 return (SET_ERROR(EINVAL));
2592
2593         /* clone should be the clone (unless they are unrelated) */
2594         if (clone->ds_prev != NULL &&
2595             clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2596             origin_head->ds_dir != clone->ds_prev->ds_dir)
2597                 return (SET_ERROR(EINVAL));
2598
2599         /* the clone should be a child of the origin */
2600         if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2601                 return (SET_ERROR(EINVAL));
2602
2603         /* origin_head shouldn't be modified unless 'force' */
2604         if (!force &&
2605             dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2606                 return (SET_ERROR(ETXTBSY));
2607
2608         /* origin_head should have no long holds (e.g. is not mounted) */
2609         if (dsl_dataset_handoff_check(origin_head, owner, tx))
2610                 return (SET_ERROR(EBUSY));
2611
2612         /* check amount of any unconsumed refreservation */
2613         unused_refres_delta =
2614             (int64_t)MIN(origin_head->ds_reserved,
2615             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2616             (int64_t)MIN(origin_head->ds_reserved,
2617             dsl_dataset_phys(clone)->ds_unique_bytes);
2618
2619         if (unused_refres_delta > 0 &&
2620             unused_refres_delta >
2621             dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2622                 return (SET_ERROR(ENOSPC));
2623
2624         /* clone can't be over the head's refquota */
2625         if (origin_head->ds_quota != 0 &&
2626             dsl_dataset_phys(clone)->ds_referenced_bytes >
2627             origin_head->ds_quota)
2628                 return (SET_ERROR(EDQUOT));
2629
2630         return (0);
2631 }
2632
2633 void
2634 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2635     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2636 {
2637         dsl_pool_t *dp = dmu_tx_pool(tx);
2638         int64_t unused_refres_delta;
2639
2640         ASSERT(clone->ds_reserved == 0);
2641         ASSERT(origin_head->ds_quota == 0 ||
2642             dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2643         ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2644
2645         dmu_buf_will_dirty(clone->ds_dbuf, tx);
2646         dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2647
2648         if (clone->ds_objset != NULL) {
2649                 dmu_objset_evict(clone->ds_objset);
2650                 clone->ds_objset = NULL;
2651         }
2652
2653         if (origin_head->ds_objset != NULL) {
2654                 dmu_objset_evict(origin_head->ds_objset);
2655                 origin_head->ds_objset = NULL;
2656         }
2657
2658         unused_refres_delta =
2659             (int64_t)MIN(origin_head->ds_reserved,
2660             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2661             (int64_t)MIN(origin_head->ds_reserved,
2662             dsl_dataset_phys(clone)->ds_unique_bytes);
2663
2664         /*
2665          * Reset origin's unique bytes, if it exists.
2666          */
2667         if (clone->ds_prev) {
2668                 dsl_dataset_t *origin = clone->ds_prev;
2669                 uint64_t comp, uncomp;
2670
2671                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2672                 dsl_deadlist_space_range(&clone->ds_deadlist,
2673                     dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2674                     &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2675         }
2676
2677         /* swap blkptrs */
2678         {
2679                 blkptr_t tmp;
2680                 tmp = dsl_dataset_phys(origin_head)->ds_bp;
2681                 dsl_dataset_phys(origin_head)->ds_bp =
2682                     dsl_dataset_phys(clone)->ds_bp;
2683                 dsl_dataset_phys(clone)->ds_bp = tmp;
2684         }
2685
2686         /* set dd_*_bytes */
2687         {
2688                 int64_t dused, dcomp, duncomp;
2689                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2690                 uint64_t odl_used, odl_comp, odl_uncomp;
2691
2692                 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2693                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
2694
2695                 dsl_deadlist_space(&clone->ds_deadlist,
2696                     &cdl_used, &cdl_comp, &cdl_uncomp);
2697                 dsl_deadlist_space(&origin_head->ds_deadlist,
2698                     &odl_used, &odl_comp, &odl_uncomp);
2699
2700                 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2701                     cdl_used -
2702                     (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2703                     odl_used);
2704                 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2705                     cdl_comp -
2706                     (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2707                     odl_comp);
2708                 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2709                     cdl_uncomp -
2710                     (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2711                     odl_uncomp);
2712
2713                 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2714                     dused, dcomp, duncomp, tx);
2715                 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2716                     -dused, -dcomp, -duncomp, tx);
2717
2718                 /*
2719                  * The difference in the space used by snapshots is the
2720                  * difference in snapshot space due to the head's
2721                  * deadlist (since that's the only thing that's
2722                  * changing that affects the snapused).
2723                  */
2724                 dsl_deadlist_space_range(&clone->ds_deadlist,
2725                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2726                     &cdl_used, &cdl_comp, &cdl_uncomp);
2727                 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2728                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2729                     &odl_used, &odl_comp, &odl_uncomp);
2730                 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2731                     DD_USED_HEAD, DD_USED_SNAP, tx);
2732         }
2733
2734         /* swap ds_*_bytes */
2735         SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2736             dsl_dataset_phys(clone)->ds_referenced_bytes);
2737         SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2738             dsl_dataset_phys(clone)->ds_compressed_bytes);
2739         SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2740             dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2741         SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2742             dsl_dataset_phys(clone)->ds_unique_bytes);
2743
2744         /* apply any parent delta for change in unconsumed refreservation */
2745         dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2746             unused_refres_delta, 0, 0, tx);
2747
2748         /*
2749          * Swap deadlists.
2750          */
2751         dsl_deadlist_close(&clone->ds_deadlist);
2752         dsl_deadlist_close(&origin_head->ds_deadlist);
2753         SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2754             dsl_dataset_phys(clone)->ds_deadlist_obj);
2755         dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2756             dsl_dataset_phys(clone)->ds_deadlist_obj);
2757         dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2758             dsl_dataset_phys(origin_head)->ds_deadlist_obj);
2759
2760         dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2761
2762         spa_history_log_internal_ds(clone, "clone swap", tx,
2763             "parent=%s", origin_head->ds_dir->dd_myname);
2764 }
2765
2766 /*
2767  * Given a pool name and a dataset object number in that pool,
2768  * return the name of that dataset.
2769  */
2770 int
2771 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2772 {
2773         dsl_pool_t *dp;
2774         dsl_dataset_t *ds;
2775         int error;
2776
2777         error = dsl_pool_hold(pname, FTAG, &dp);
2778         if (error != 0)
2779                 return (error);
2780
2781         error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2782         if (error == 0) {
2783                 dsl_dataset_name(ds, buf);
2784                 dsl_dataset_rele(ds, FTAG);
2785         }
2786         dsl_pool_rele(dp, FTAG);
2787
2788         return (error);
2789 }
2790
2791 int
2792 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2793     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2794 {
2795         int error = 0;
2796
2797         ASSERT3S(asize, >, 0);
2798
2799         /*
2800          * *ref_rsrv is the portion of asize that will come from any
2801          * unconsumed refreservation space.
2802          */
2803         *ref_rsrv = 0;
2804
2805         mutex_enter(&ds->ds_lock);
2806         /*
2807          * Make a space adjustment for reserved bytes.
2808          */
2809         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2810                 ASSERT3U(*used, >=,
2811                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2812                 *used -=
2813                     (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2814                 *ref_rsrv =
2815                     asize - MIN(asize, parent_delta(ds, asize + inflight));
2816         }
2817
2818         if (!check_quota || ds->ds_quota == 0) {
2819                 mutex_exit(&ds->ds_lock);
2820                 return (0);
2821         }
2822         /*
2823          * If they are requesting more space, and our current estimate
2824          * is over quota, they get to try again unless the actual
2825          * on-disk is over quota and there are no pending changes (which
2826          * may free up space for us).
2827          */
2828         if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
2829             ds->ds_quota) {
2830                 if (inflight > 0 ||
2831                     dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
2832                         error = SET_ERROR(ERESTART);
2833                 else
2834                         error = SET_ERROR(EDQUOT);
2835         }
2836         mutex_exit(&ds->ds_lock);
2837
2838         return (error);
2839 }
2840
2841 typedef struct dsl_dataset_set_qr_arg {
2842         const char *ddsqra_name;
2843         zprop_source_t ddsqra_source;
2844         uint64_t ddsqra_value;
2845 } dsl_dataset_set_qr_arg_t;
2846
2847
2848 /* ARGSUSED */
2849 static int
2850 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2851 {
2852         dsl_dataset_set_qr_arg_t *ddsqra = arg;
2853         dsl_pool_t *dp = dmu_tx_pool(tx);
2854         dsl_dataset_t *ds;
2855         int error;
2856         uint64_t newval;
2857
2858         if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2859                 return (SET_ERROR(ENOTSUP));
2860
2861         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2862         if (error != 0)
2863                 return (error);
2864
2865         if (dsl_dataset_is_snapshot(ds)) {
2866                 dsl_dataset_rele(ds, FTAG);
2867                 return (SET_ERROR(EINVAL));
2868         }
2869
2870         error = dsl_prop_predict(ds->ds_dir,
2871             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2872             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2873         if (error != 0) {
2874                 dsl_dataset_rele(ds, FTAG);
2875                 return (error);
2876         }
2877
2878         if (newval == 0) {
2879                 dsl_dataset_rele(ds, FTAG);
2880                 return (0);
2881         }
2882
2883         if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
2884             newval < ds->ds_reserved) {
2885                 dsl_dataset_rele(ds, FTAG);
2886                 return (SET_ERROR(ENOSPC));
2887         }
2888
2889         dsl_dataset_rele(ds, FTAG);
2890         return (0);
2891 }
2892
2893 static void
2894 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2895 {
2896         dsl_dataset_set_qr_arg_t *ddsqra = arg;
2897         dsl_pool_t *dp = dmu_tx_pool(tx);
2898         dsl_dataset_t *ds;
2899         uint64_t newval;
2900
2901         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2902
2903         dsl_prop_set_sync_impl(ds,
2904             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2905             ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2906             &ddsqra->ddsqra_value, tx);
2907
2908         VERIFY0(dsl_prop_get_int_ds(ds,
2909             zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2910
2911         if (ds->ds_quota != newval) {
2912                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2913                 ds->ds_quota = newval;
2914         }
2915         dsl_dataset_rele(ds, FTAG);
2916 }
2917
2918 int
2919 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2920     uint64_t refquota)
2921 {
2922         dsl_dataset_set_qr_arg_t ddsqra;
2923
2924         ddsqra.ddsqra_name = dsname;
2925         ddsqra.ddsqra_source = source;
2926         ddsqra.ddsqra_value = refquota;
2927
2928         return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2929             dsl_dataset_set_refquota_sync, &ddsqra, 0));
2930 }
2931
2932 static int
2933 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2934 {
2935         dsl_dataset_set_qr_arg_t *ddsqra = arg;
2936         dsl_pool_t *dp = dmu_tx_pool(tx);
2937         dsl_dataset_t *ds;
2938         int error;
2939         uint64_t newval, unique;
2940
2941         if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2942                 return (SET_ERROR(ENOTSUP));
2943
2944         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2945         if (error != 0)
2946                 return (error);
2947
2948         if (dsl_dataset_is_snapshot(ds)) {
2949                 dsl_dataset_rele(ds, FTAG);
2950                 return (SET_ERROR(EINVAL));
2951         }
2952
2953         error = dsl_prop_predict(ds->ds_dir,
2954             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2955             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2956         if (error != 0) {
2957                 dsl_dataset_rele(ds, FTAG);
2958                 return (error);
2959         }
2960
2961         /*
2962          * If we are doing the preliminary check in open context, the
2963          * space estimates may be inaccurate.
2964          */
2965         if (!dmu_tx_is_syncing(tx)) {
2966                 dsl_dataset_rele(ds, FTAG);
2967                 return (0);
2968         }
2969
2970         mutex_enter(&ds->ds_lock);
2971         if (!DS_UNIQUE_IS_ACCURATE(ds))
2972                 dsl_dataset_recalc_head_uniq(ds);
2973         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
2974         mutex_exit(&ds->ds_lock);
2975
2976         if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2977                 uint64_t delta = MAX(unique, newval) -
2978                     MAX(unique, ds->ds_reserved);
2979
2980                 if (delta >
2981                     dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2982                     (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2983                         dsl_dataset_rele(ds, FTAG);
2984                         return (SET_ERROR(ENOSPC));
2985                 }
2986         }
2987
2988         dsl_dataset_rele(ds, FTAG);
2989         return (0);
2990 }
2991
2992 void
2993 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2994     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2995 {
2996         uint64_t newval;
2997         uint64_t unique;
2998         int64_t delta;
2999
3000         dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3001             source, sizeof (value), 1, &value, tx);
3002
3003         VERIFY0(dsl_prop_get_int_ds(ds,
3004             zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3005
3006         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3007         mutex_enter(&ds->ds_dir->dd_lock);
3008         mutex_enter(&ds->ds_lock);
3009         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3010         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3011         delta = MAX(0, (int64_t)(newval - unique)) -
3012             MAX(0, (int64_t)(ds->ds_reserved - unique));
3013         ds->ds_reserved = newval;
3014         mutex_exit(&ds->ds_lock);
3015
3016         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3017         mutex_exit(&ds->ds_dir->dd_lock);
3018 }
3019
3020 static void
3021 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3022 {
3023         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3024         dsl_pool_t *dp = dmu_tx_pool(tx);
3025         dsl_dataset_t *ds;
3026
3027         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3028         dsl_dataset_set_refreservation_sync_impl(ds,
3029             ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3030         dsl_dataset_rele(ds, FTAG);
3031 }
3032
3033 int
3034 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3035     uint64_t refreservation)
3036 {
3037         dsl_dataset_set_qr_arg_t ddsqra;
3038
3039         ddsqra.ddsqra_name = dsname;
3040         ddsqra.ddsqra_source = source;
3041         ddsqra.ddsqra_value = refreservation;
3042
3043         return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3044             dsl_dataset_set_refreservation_sync, &ddsqra, 0));
3045 }
3046
3047 /*
3048  * Return (in *usedp) the amount of space written in new that is not
3049  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3050  * a snapshot before new, in new's filesystem (or its origin).  If not then
3051  * fail and return EINVAL.
3052  *
3053  * The written space is calculated by considering two components:  First, we
3054  * ignore any freed space, and calculate the written as new's used space
3055  * minus old's used space.  Next, we add in the amount of space that was freed
3056  * between the two snapshots, thus reducing new's used space relative to old's.
3057  * Specifically, this is the space that was born before old->ds_creation_txg,
3058  * and freed before new (ie. on new's deadlist or a previous deadlist).
3059  *
3060  * space freed                         [---------------------]
3061  * snapshots                       ---O-------O--------O-------O------
3062  *                                         oldsnap            new
3063  */
3064 int
3065 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3066     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3067 {
3068         int err = 0;
3069         uint64_t snapobj;
3070         dsl_pool_t *dp = new->ds_dir->dd_pool;
3071
3072         ASSERT(dsl_pool_config_held(dp));
3073
3074         *usedp = 0;
3075         *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3076         *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3077
3078         *compp = 0;
3079         *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3080         *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3081
3082         *uncompp = 0;
3083         *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3084         *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3085
3086         snapobj = new->ds_object;
3087         while (snapobj != oldsnap->ds_object) {
3088                 dsl_dataset_t *snap;
3089                 uint64_t used, comp, uncomp;
3090
3091                 if (snapobj == new->ds_object) {
3092                         snap = new;
3093                 } else {
3094                         err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3095                         if (err != 0)
3096                                 break;
3097                 }
3098
3099                 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3100                     dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3101                         /*
3102                          * The blocks in the deadlist can not be born after
3103                          * ds_prev_snap_txg, so get the whole deadlist space,
3104                          * which is more efficient (especially for old-format
3105                          * deadlists).  Unfortunately the deadlist code
3106                          * doesn't have enough information to make this
3107                          * optimization itself.
3108                          */
3109                         dsl_deadlist_space(&snap->ds_deadlist,
3110                             &used, &comp, &uncomp);
3111                 } else {
3112                         dsl_deadlist_space_range(&snap->ds_deadlist,
3113                             0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3114                             &used, &comp, &uncomp);
3115                 }
3116                 *usedp += used;
3117                 *compp += comp;
3118                 *uncompp += uncomp;
3119
3120                 /*
3121                  * If we get to the beginning of the chain of snapshots
3122                  * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3123                  * was not a snapshot of/before new.
3124                  */
3125                 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3126                 if (snap != new)
3127                         dsl_dataset_rele(snap, FTAG);
3128                 if (snapobj == 0) {
3129                         err = SET_ERROR(EINVAL);
3130                         break;
3131                 }
3132
3133         }
3134         return (err);
3135 }
3136
3137 /*
3138  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3139  * lastsnap, and all snapshots in between are deleted.
3140  *
3141  * blocks that would be freed            [---------------------------]
3142  * snapshots                       ---O-------O--------O-------O--------O
3143  *                                        firstsnap        lastsnap
3144  *
3145  * This is the set of blocks that were born after the snap before firstsnap,
3146  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3147  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3148  * We calculate this by iterating over the relevant deadlists (from the snap
3149  * after lastsnap, backward to the snap after firstsnap), summing up the
3150  * space on the deadlist that was born after the snap before firstsnap.
3151  */
3152 int
3153 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3154     dsl_dataset_t *lastsnap,
3155     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3156 {
3157         int err = 0;
3158         uint64_t snapobj;
3159         dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3160
3161         ASSERT(dsl_dataset_is_snapshot(firstsnap));
3162         ASSERT(dsl_dataset_is_snapshot(lastsnap));
3163
3164         /*
3165          * Check that the snapshots are in the same dsl_dir, and firstsnap
3166          * is before lastsnap.
3167          */
3168         if (firstsnap->ds_dir != lastsnap->ds_dir ||
3169             dsl_dataset_phys(firstsnap)->ds_creation_txg >
3170             dsl_dataset_phys(lastsnap)->ds_creation_txg)
3171                 return (SET_ERROR(EINVAL));
3172
3173         *usedp = *compp = *uncompp = 0;
3174
3175         snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3176         while (snapobj != firstsnap->ds_object) {
3177                 dsl_dataset_t *ds;
3178                 uint64_t used, comp, uncomp;
3179
3180                 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3181                 if (err != 0)
3182                         break;
3183
3184                 dsl_deadlist_space_range(&ds->ds_deadlist,
3185                     dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3186                     &used, &comp, &uncomp);
3187                 *usedp += used;
3188                 *compp += comp;
3189                 *uncompp += uncomp;
3190
3191                 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3192                 ASSERT3U(snapobj, !=, 0);
3193                 dsl_dataset_rele(ds, FTAG);
3194         }
3195         return (err);
3196 }
3197
3198 /*
3199  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3200  * For example, they could both be snapshots of the same filesystem, and
3201  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3202  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3203  * filesystem.  Or 'earlier' could be the origin's origin.
3204  *
3205  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3206  */
3207 boolean_t
3208 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3209         uint64_t earlier_txg)
3210 {
3211         dsl_pool_t *dp = later->ds_dir->dd_pool;
3212         int error;
3213         boolean_t ret;
3214         dsl_dataset_t *origin;
3215
3216         ASSERT(dsl_pool_config_held(dp));
3217         ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
3218
3219         if (earlier_txg == 0)
3220                 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3221
3222         if (dsl_dataset_is_snapshot(later) &&
3223             earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3224                 return (B_FALSE);
3225
3226         if (later->ds_dir == earlier->ds_dir)
3227                 return (B_TRUE);
3228         if (!dsl_dir_is_clone(later->ds_dir))
3229                 return (B_FALSE);
3230
3231         if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3232                 return (B_TRUE);
3233         error = dsl_dataset_hold_obj(dp,
3234             dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3235         if (error != 0)
3236                 return (B_FALSE);
3237         ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3238         dsl_dataset_rele(origin, FTAG);
3239         return (ret);
3240 }
3241
3242
3243 void
3244 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3245 {
3246         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3247         dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3248 }
3249
3250 #if defined(_KERNEL) && defined(HAVE_SPL)
3251 EXPORT_SYMBOL(dsl_dataset_hold);
3252 EXPORT_SYMBOL(dsl_dataset_hold_obj);
3253 EXPORT_SYMBOL(dsl_dataset_own);
3254 EXPORT_SYMBOL(dsl_dataset_own_obj);
3255 EXPORT_SYMBOL(dsl_dataset_name);
3256 EXPORT_SYMBOL(dsl_dataset_rele);
3257 EXPORT_SYMBOL(dsl_dataset_disown);
3258 EXPORT_SYMBOL(dsl_dataset_tryown);
3259 EXPORT_SYMBOL(dsl_dataset_create_sync);
3260 EXPORT_SYMBOL(dsl_dataset_create_sync_dd);
3261 EXPORT_SYMBOL(dsl_dataset_snapshot_check);
3262 EXPORT_SYMBOL(dsl_dataset_snapshot_sync);
3263 EXPORT_SYMBOL(dsl_dataset_promote);
3264 EXPORT_SYMBOL(dsl_dataset_user_hold);
3265 EXPORT_SYMBOL(dsl_dataset_user_release);
3266 EXPORT_SYMBOL(dsl_dataset_get_holds);
3267 EXPORT_SYMBOL(dsl_dataset_get_blkptr);
3268 EXPORT_SYMBOL(dsl_dataset_set_blkptr);
3269 EXPORT_SYMBOL(dsl_dataset_get_spa);
3270 EXPORT_SYMBOL(dsl_dataset_modified_since_snap);
3271 EXPORT_SYMBOL(dsl_dataset_space_written);
3272 EXPORT_SYMBOL(dsl_dataset_space_wouldfree);
3273 EXPORT_SYMBOL(dsl_dataset_sync);
3274 EXPORT_SYMBOL(dsl_dataset_block_born);
3275 EXPORT_SYMBOL(dsl_dataset_block_kill);
3276 EXPORT_SYMBOL(dsl_dataset_block_freeable);
3277 EXPORT_SYMBOL(dsl_dataset_prev_snap_txg);
3278 EXPORT_SYMBOL(dsl_dataset_dirty);
3279 EXPORT_SYMBOL(dsl_dataset_stats);
3280 EXPORT_SYMBOL(dsl_dataset_fast_stat);
3281 EXPORT_SYMBOL(dsl_dataset_space);
3282 EXPORT_SYMBOL(dsl_dataset_fsid_guid);
3283 EXPORT_SYMBOL(dsl_dsobj_to_dsname);
3284 EXPORT_SYMBOL(dsl_dataset_check_quota);
3285 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl);
3286 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl);
3287 #endif