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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org>
  24  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  25  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2014 RackTop Systems.
  27  */
  28
  29 #include <sys/dmu_objset.h>
  30 #include <sys/dsl_dataset.h>
  31 #include <sys/dsl_dir.h>
  32 #include <sys/dsl_prop.h>
  33 #include <sys/dsl_synctask.h>
  34 #include <sys/dmu_traverse.h>
  35 #include <sys/dmu_impl.h>
  36 #include <sys/dmu_tx.h>
  37 #include <sys/arc.h>
  38 #include <sys/zio.h>
  39 #include <sys/zap.h>
  40 #include <sys/zfeature.h>
  41 #include <sys/unique.h>
  42 #include <sys/zfs_context.h>
  43 #include <sys/zfs_ioctl.h>
  44 #include <sys/spa.h>
  45 #include <sys/zfs_znode.h>
  46 #include <sys/zfs_onexit.h>
  47 #include <sys/zvol.h>
  48 #include <sys/dsl_scan.h>
  49 #include <sys/dsl_deadlist.h>
  50 #include <sys/dsl_destroy.h>
  51 #include <sys/dsl_userhold.h>
  52 #include <sys/dsl_bookmark.h>
  53
  54 SYSCTL_DECL(_vfs_zfs);
  55
  56 /*
  57  * The SPA supports block sizes up to 16MB.  However, very large blocks
  58  * can have an impact on i/o latency (e.g. tying up a spinning disk for
  59  * ~300ms), and also potentially on the memory allocator.  Therefore,
  60  * we do not allow the recordsize to be set larger than zfs_max_recordsize
  61  * (default 1MB).  Larger blocks can be created by changing this tunable,
  62  * and pools with larger blocks can always be imported and used, regardless
  63  * of this setting.
  64  */
  65 int zfs_max_recordsize = 1 * 1024 * 1024;
  66 SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN,
  67     &zfs_max_recordsize, 0,
  68     "Maximum block size.  Expect dragons when tuning this.");
  69
  70 #define SWITCH64(x, y) \
  71         { \
  72                 uint64_t __tmp = (x); \
  73                 (x) = (y); \
  74                 (y) = __tmp; \
  75         }
  76
  77 #define DS_REF_MAX      (1ULL << 62)
  78
  79 /*
  80  * Figure out how much of this delta should be propogated to the dsl_dir
  81  * layer.  If there's a refreservation, that space has already been
  82  * partially accounted for in our ancestors.
  83  */
  84 static int64_t
  85 parent_delta(dsl_dataset_t *ds, int64_t delta)
  86 {
  87         uint64_t old_bytes, new_bytes;
  88
  89         if (ds->ds_reserved == 0)
  90                 return (delta);
  91
  92         old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
  93         new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
  94
  95         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
  96         return (new_bytes - old_bytes);
  97 }
  98
  99 void
 100 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
 101 {
 102         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 103         int compressed = BP_GET_PSIZE(bp);
 104         int uncompressed = BP_GET_UCSIZE(bp);
 105         int64_t delta;
 106
 107         dprintf_bp(bp, "ds=%p", ds);
 108
 109         ASSERT(dmu_tx_is_syncing(tx));
 110         /* It could have been compressed away to nothing */
 111         if (BP_IS_HOLE(bp))
 112                 return;
 113         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 114         ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 115         if (ds == NULL) {
 116                 dsl_pool_mos_diduse_space(tx->tx_pool,
 117                     used, compressed, uncompressed);
 118                 return;
 119         }
 120
 121         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 122         mutex_enter(&ds->ds_lock);
 123         delta = parent_delta(ds, used);
 124         ds->ds_phys->ds_referenced_bytes += used;
 125         ds->ds_phys->ds_compressed_bytes += compressed;
 126         ds->ds_phys->ds_uncompressed_bytes += uncompressed;
 127         ds->ds_phys->ds_unique_bytes += used;
 128         if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE)
 129                 ds->ds_need_large_blocks = B_TRUE;
 130         mutex_exit(&ds->ds_lock);
 131         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 132             compressed, uncompressed, tx);
 133         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 134             DD_USED_REFRSRV, DD_USED_HEAD, tx);
 135 }
 136
 137 int
 138 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 139     boolean_t async)
 140 {
 141         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 142         int compressed = BP_GET_PSIZE(bp);
 143         int uncompressed = BP_GET_UCSIZE(bp);
 144
 145         if (BP_IS_HOLE(bp))
 146                 return (0);
 147
 148         ASSERT(dmu_tx_is_syncing(tx));
 149         ASSERT(bp->blk_birth <= tx->tx_txg);
 150
 151         if (ds == NULL) {
 152                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 153                 dsl_pool_mos_diduse_space(tx->tx_pool,
 154                     -used, -compressed, -uncompressed);
 155                 return (used);
 156         }
 157         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 158
 159         ASSERT(!dsl_dataset_is_snapshot(ds));
 160         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 161
 162         if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
 163                 int64_t delta;
 164
 165                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 166                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 167
 168                 mutex_enter(&ds->ds_lock);
 169                 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
 170                     !DS_UNIQUE_IS_ACCURATE(ds));
 171                 delta = parent_delta(ds, -used);
 172                 ds->ds_phys->ds_unique_bytes -= used;
 173                 mutex_exit(&ds->ds_lock);
 174                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 175                     delta, -compressed, -uncompressed, tx);
 176                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 177                     DD_USED_REFRSRV, DD_USED_HEAD, tx);
 178         } else {
 179                 dprintf_bp(bp, "putting on dead list: %s", "");
 180                 if (async) {
 181                         /*
 182                          * We are here as part of zio's write done callback,
 183                          * which means we're a zio interrupt thread.  We can't
 184                          * call dsl_deadlist_insert() now because it may block
 185                          * waiting for I/O.  Instead, put bp on the deferred
 186                          * queue and let dsl_pool_sync() finish the job.
 187                          */
 188                         bplist_append(&ds->ds_pending_deadlist, bp);
 189                 } else {
 190                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 191                 }
 192                 ASSERT3U(ds->ds_prev->ds_object, ==,
 193                     ds->ds_phys->ds_prev_snap_obj);
 194                 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
 195                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 196                 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
 197                     ds->ds_object && bp->blk_birth >
 198                     ds->ds_prev->ds_phys->ds_prev_snap_txg) {
 199                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 200                         mutex_enter(&ds->ds_prev->ds_lock);
 201                         ds->ds_prev->ds_phys->ds_unique_bytes += used;
 202                         mutex_exit(&ds->ds_prev->ds_lock);
 203                 }
 204                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 205                         dsl_dir_transfer_space(ds->ds_dir, used,
 206                             DD_USED_HEAD, DD_USED_SNAP, tx);
 207                 }
 208         }
 209         mutex_enter(&ds->ds_lock);
 210         ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
 211         ds->ds_phys->ds_referenced_bytes -= used;
 212         ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
 213         ds->ds_phys->ds_compressed_bytes -= compressed;
 214         ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
 215         ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
 216         mutex_exit(&ds->ds_lock);
 217
 218         return (used);
 219 }
 220
 221 uint64_t
 222 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
 223 {
 224         uint64_t trysnap = 0;
 225
 226         if (ds == NULL)
 227                 return (0);
 228         /*
 229          * The snapshot creation could fail, but that would cause an
 230          * incorrect FALSE return, which would only result in an
 231          * overestimation of the amount of space that an operation would
 232          * consume, which is OK.
 233          *
 234          * There's also a small window where we could miss a pending
 235          * snapshot, because we could set the sync task in the quiescing
 236          * phase.  So this should only be used as a guess.
 237          */
 238         if (ds->ds_trysnap_txg >
 239             spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
 240                 trysnap = ds->ds_trysnap_txg;
 241         return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
 242 }
 243
 244 boolean_t
 245 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
 246     uint64_t blk_birth)
 247 {
 248         if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
 249             (bp != NULL && BP_IS_HOLE(bp)))
 250                 return (B_FALSE);
 251
 252         ddt_prefetch(dsl_dataset_get_spa(ds), bp);
 253
 254         return (B_TRUE);
 255 }
 256
 257 /* ARGSUSED */
 258 static void
 259 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
 260 {
 261         dsl_dataset_t *ds = dsv;
 262
 263         ASSERT(ds->ds_owner == NULL);
 264
 265         unique_remove(ds->ds_fsid_guid);
 266
 267         if (ds->ds_objset != NULL)
 268                 dmu_objset_evict(ds->ds_objset);
 269
 270         if (ds->ds_prev) {
 271                 dsl_dataset_rele(ds->ds_prev, ds);
 272                 ds->ds_prev = NULL;
 273         }
 274
 275         bplist_destroy(&ds->ds_pending_deadlist);
 276         if (ds->ds_phys->ds_deadlist_obj != 0)
 277                 dsl_deadlist_close(&ds->ds_deadlist);
 278         if (ds->ds_dir)
 279                 dsl_dir_rele(ds->ds_dir, ds);
 280
 281         ASSERT(!list_link_active(&ds->ds_synced_link));
 282
 283         if (mutex_owned(&ds->ds_lock))
 284                 mutex_exit(&ds->ds_lock);
 285         mutex_destroy(&ds->ds_lock);
 286         if (mutex_owned(&ds->ds_opening_lock))
 287                 mutex_exit(&ds->ds_opening_lock);
 288         mutex_destroy(&ds->ds_opening_lock);
 289         mutex_destroy(&ds->ds_sendstream_lock);
 290         refcount_destroy(&ds->ds_longholds);
 291
 292         kmem_free(ds, sizeof (dsl_dataset_t));
 293 }
 294
 295 int
 296 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 297 {
 298         dsl_dataset_phys_t *headphys;
 299         int err;
 300         dmu_buf_t *headdbuf;
 301         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 302         objset_t *mos = dp->dp_meta_objset;
 303
 304         if (ds->ds_snapname[0])
 305                 return (0);
 306         if (ds->ds_phys->ds_next_snap_obj == 0)
 307                 return (0);
 308
 309         err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
 310             FTAG, &headdbuf);
 311         if (err != 0)
 312                 return (err);
 313         headphys = headdbuf->db_data;
 314         err = zap_value_search(dp->dp_meta_objset,
 315             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 316         dmu_buf_rele(headdbuf, FTAG);
 317         return (err);
 318 }
 319
 320 int
 321 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 322 {
 323         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 324         uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
 325         matchtype_t mt;
 326         int err;
 327
 328         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 329                 mt = MT_FIRST;
 330         else
 331                 mt = MT_EXACT;
 332
 333         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 334             value, mt, NULL, 0, NULL);
 335         if (err == ENOTSUP && mt == MT_FIRST)
 336                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 337         return (err);
 338 }
 339
 340 int
 341 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
 342     boolean_t adj_cnt)
 343 {
 344         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 345         uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
 346         matchtype_t mt;
 347         int err;
 348
 349         dsl_dir_snap_cmtime_update(ds->ds_dir);
 350
 351         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 352                 mt = MT_FIRST;
 353         else
 354                 mt = MT_EXACT;
 355
 356         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 357         if (err == ENOTSUP && mt == MT_FIRST)
 358                 err = zap_remove(mos, snapobj, name, tx);
 359
 360         if (err == 0 && adj_cnt)
 361                 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
 362                     DD_FIELD_SNAPSHOT_COUNT, tx);
 363
 364         return (err);
 365 }
 366
 367 int
 368 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 369     dsl_dataset_t **dsp)
 370 {
 371         objset_t *mos = dp->dp_meta_objset;
 372         dmu_buf_t *dbuf;
 373         dsl_dataset_t *ds;
 374         int err;
 375         dmu_object_info_t doi;
 376
 377         ASSERT(dsl_pool_config_held(dp));
 378
 379         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 380         if (err != 0)
 381                 return (err);
 382
 383         /* Make sure dsobj has the correct object type. */
 384         dmu_object_info_from_db(dbuf, &doi);
 385         if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
 386                 dmu_buf_rele(dbuf, tag);
 387                 return (SET_ERROR(EINVAL));
 388         }
 389
 390         ds = dmu_buf_get_user(dbuf);
 391         if (ds == NULL) {
 392                 dsl_dataset_t *winner = NULL;
 393
 394                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 395                 ds->ds_dbuf = dbuf;
 396                 ds->ds_object = dsobj;
 397                 ds->ds_phys = dbuf->db_data;
 398
 399                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 400                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 401                 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
 402                 refcount_create(&ds->ds_longholds);
 403
 404                 bplist_create(&ds->ds_pending_deadlist);
 405                 dsl_deadlist_open(&ds->ds_deadlist,
 406                     mos, ds->ds_phys->ds_deadlist_obj);
 407
 408                 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
 409                     offsetof(dmu_sendarg_t, dsa_link));
 410
 411                 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 412                         int zaperr = zap_contains(mos, dsobj, DS_FIELD_LARGE_BLOCKS);
 413                         if (zaperr != ENOENT) {
 414                                 VERIFY0(zaperr);
 415                                 ds->ds_large_blocks = B_TRUE;
 416                         }
 417                 }
 418
 419                 if (err == 0) {
 420                         err = dsl_dir_hold_obj(dp,
 421                             ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
 422                 }
 423                 if (err != 0) {
 424                         mutex_destroy(&ds->ds_lock);
 425                         mutex_destroy(&ds->ds_opening_lock);
 426                         mutex_destroy(&ds->ds_sendstream_lock);
 427                         refcount_destroy(&ds->ds_longholds);
 428                         bplist_destroy(&ds->ds_pending_deadlist);
 429                         dsl_deadlist_close(&ds->ds_deadlist);
 430                         kmem_free(ds, sizeof (dsl_dataset_t));
 431                         dmu_buf_rele(dbuf, tag);
 432                         return (err);
 433                 }
 434
 435                 if (!dsl_dataset_is_snapshot(ds)) {
 436                         ds->ds_snapname[0] = '\0';
 437                         if (ds->ds_phys->ds_prev_snap_obj != 0) {
 438                                 err = dsl_dataset_hold_obj(dp,
 439                                     ds->ds_phys->ds_prev_snap_obj,
 440                                     ds, &ds->ds_prev);
 441                         }
 442                         if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 443                                 int zaperr = zap_lookup(mos, ds->ds_object,
 444                                     DS_FIELD_BOOKMARK_NAMES,
 445                                     sizeof (ds->ds_bookmarks), 1,
 446                                     &ds->ds_bookmarks);
 447                                 if (zaperr != ENOENT)
 448                                         VERIFY0(zaperr);
 449                         }
 450                 } else {
 451                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 452                                 err = dsl_dataset_get_snapname(ds);
 453                         if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
 454                                 err = zap_count(
 455                                     ds->ds_dir->dd_pool->dp_meta_objset,
 456                                     ds->ds_phys->ds_userrefs_obj,
 457                                     &ds->ds_userrefs);
 458                         }
 459                 }
 460
 461                 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
 462                         err = dsl_prop_get_int_ds(ds,
 463                             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
 464                             &ds->ds_reserved);
 465                         if (err == 0) {
 466                                 err = dsl_prop_get_int_ds(ds,
 467                                     zfs_prop_to_name(ZFS_PROP_REFQUOTA),
 468                                     &ds->ds_quota);
 469                         }
 470                 } else {
 471                         ds->ds_reserved = ds->ds_quota = 0;
 472                 }
 473
 474                 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
 475                     &ds->ds_phys, dsl_dataset_evict)) != NULL) {
 476                         bplist_destroy(&ds->ds_pending_deadlist);
 477                         dsl_deadlist_close(&ds->ds_deadlist);
 478                         if (ds->ds_prev)
 479                                 dsl_dataset_rele(ds->ds_prev, ds);
 480                         dsl_dir_rele(ds->ds_dir, ds);
 481                         mutex_destroy(&ds->ds_lock);
 482                         mutex_destroy(&ds->ds_opening_lock);
 483                         mutex_destroy(&ds->ds_sendstream_lock);
 484                         refcount_destroy(&ds->ds_longholds);
 485                         kmem_free(ds, sizeof (dsl_dataset_t));
 486                         if (err != 0) {
 487                                 dmu_buf_rele(dbuf, tag);
 488                                 return (err);
 489                         }
 490                         ds = winner;
 491                 } else {
 492                         ds->ds_fsid_guid =
 493                             unique_insert(ds->ds_phys->ds_fsid_guid);
 494                 }
 495         }
 496         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 497         ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
 498         ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
 499             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 500             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 501         *dsp = ds;
 502         return (0);
 503 }
 504
 505 int
 506 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
 507     void *tag, dsl_dataset_t **dsp)
 508 {
 509         dsl_dir_t *dd;
 510         const char *snapname;
 511         uint64_t obj;
 512         int err = 0;
 513
 514         err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
 515         if (err != 0)
 516                 return (err);
 517
 518         ASSERT(dsl_pool_config_held(dp));
 519         obj = dd->dd_phys->dd_head_dataset_obj;
 520         if (obj != 0)
 521                 err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
 522         else
 523                 err = SET_ERROR(ENOENT);
 524
 525         /* we may be looking for a snapshot */
 526         if (err == 0 && snapname != NULL) {
 527                 dsl_dataset_t *ds;
 528
 529                 if (*snapname++ != '@') {
 530                         dsl_dataset_rele(*dsp, tag);
 531                         dsl_dir_rele(dd, FTAG);
 532                         return (SET_ERROR(ENOENT));
 533                 }
 534
 535                 dprintf("looking for snapshot '%s'\n", snapname);
 536                 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
 537                 if (err == 0)
 538                         err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
 539                 dsl_dataset_rele(*dsp, tag);
 540
 541                 if (err == 0) {
 542                         mutex_enter(&ds->ds_lock);
 543                         if (ds->ds_snapname[0] == 0)
 544                                 (void) strlcpy(ds->ds_snapname, snapname,
 545                                     sizeof (ds->ds_snapname));
 546                         mutex_exit(&ds->ds_lock);
 547                         *dsp = ds;
 548                 }
 549         }
 550
 551         dsl_dir_rele(dd, FTAG);
 552         return (err);
 553 }
 554
 555 int
 556 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
 557     void *tag, dsl_dataset_t **dsp)
 558 {
 559         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 560         if (err != 0)
 561                 return (err);
 562         if (!dsl_dataset_tryown(*dsp, tag)) {
 563                 dsl_dataset_rele(*dsp, tag);
 564                 *dsp = NULL;
 565                 return (SET_ERROR(EBUSY));
 566         }
 567         return (0);
 568 }
 569
 570 int
 571 dsl_dataset_own(dsl_pool_t *dp, const char *name,
 572     void *tag, dsl_dataset_t **dsp)
 573 {
 574         int err = dsl_dataset_hold(dp, name, tag, dsp);
 575         if (err != 0)
 576                 return (err);
 577         if (!dsl_dataset_tryown(*dsp, tag)) {
 578                 dsl_dataset_rele(*dsp, tag);
 579                 return (SET_ERROR(EBUSY));
 580         }
 581         return (0);
 582 }
 583
 584 /*
 585  * See the comment above dsl_pool_hold() for details.  In summary, a long
 586  * hold is used to prevent destruction of a dataset while the pool hold
 587  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
 588  *
 589  * The dataset and pool must be held when this function is called.  After it
 590  * is called, the pool hold may be released while the dataset is still held
 591  * and accessed.
 592  */
 593 void
 594 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
 595 {
 596         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 597         (void) refcount_add(&ds->ds_longholds, tag);
 598 }
 599
 600 void
 601 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
 602 {
 603         (void) refcount_remove(&ds->ds_longholds, tag);
 604 }
 605
 606 /* Return B_TRUE if there are any long holds on this dataset. */
 607 boolean_t
 608 dsl_dataset_long_held(dsl_dataset_t *ds)
 609 {
 610         return (!refcount_is_zero(&ds->ds_longholds));
 611 }
 612
 613 void
 614 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 615 {
 616         if (ds == NULL) {
 617                 (void) strcpy(name, "mos");
 618         } else {
 619                 dsl_dir_name(ds->ds_dir, name);
 620                 VERIFY0(dsl_dataset_get_snapname(ds));
 621                 if (ds->ds_snapname[0]) {
 622                         (void) strcat(name, "@");
 623                         /*
 624                          * We use a "recursive" mutex so that we
 625                          * can call dprintf_ds() with ds_lock held.
 626                          */
 627                         if (!MUTEX_HELD(&ds->ds_lock)) {
 628                                 mutex_enter(&ds->ds_lock);
 629                                 (void) strcat(name, ds->ds_snapname);
 630                                 mutex_exit(&ds->ds_lock);
 631                         } else {
 632                                 (void) strcat(name, ds->ds_snapname);
 633                         }
 634                 }
 635         }
 636 }
 637
 638 void
 639 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 640 {
 641         dmu_buf_rele(ds->ds_dbuf, tag);
 642 }
 643
 644 void
 645 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 646 {
 647         ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL);
 648
 649         mutex_enter(&ds->ds_lock);
 650         ds->ds_owner = NULL;
 651         mutex_exit(&ds->ds_lock);
 652         dsl_dataset_long_rele(ds, tag);
 653         if (ds->ds_dbuf != NULL)
 654                 dsl_dataset_rele(ds, tag);
 655         else
 656                 dsl_dataset_evict(NULL, ds);
 657 }
 658
 659 boolean_t
 660 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
 661 {
 662         boolean_t gotit = FALSE;
 663
 664         mutex_enter(&ds->ds_lock);
 665         if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
 666                 ds->ds_owner = tag;
 667                 dsl_dataset_long_hold(ds, tag);
 668                 gotit = TRUE;
 669         }
 670         mutex_exit(&ds->ds_lock);
 671         return (gotit);
 672 }
 673
 674 uint64_t
 675 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 676     uint64_t flags, dmu_tx_t *tx)
 677 {
 678         dsl_pool_t *dp = dd->dd_pool;
 679         dmu_buf_t *dbuf;
 680         dsl_dataset_phys_t *dsphys;
 681         uint64_t dsobj;
 682         objset_t *mos = dp->dp_meta_objset;
 683
 684         if (origin == NULL)
 685                 origin = dp->dp_origin_snap;
 686
 687         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 688         ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
 689         ASSERT(dmu_tx_is_syncing(tx));
 690         ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
 691
 692         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 693             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 694         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 695         dmu_buf_will_dirty(dbuf, tx);
 696         dsphys = dbuf->db_data;
 697         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 698         dsphys->ds_dir_obj = dd->dd_object;
 699         dsphys->ds_flags = flags;
 700         dsphys->ds_fsid_guid = unique_create();
 701         do {
 702                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 703                     sizeof (dsphys->ds_guid));
 704         } while (dsphys->ds_guid == 0);
 705         dsphys->ds_snapnames_zapobj =
 706             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 707             DMU_OT_NONE, 0, tx);
 708         dsphys->ds_creation_time = gethrestime_sec();
 709         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 710
 711         if (origin == NULL) {
 712                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 713         } else {
 714                 dsl_dataset_t *ohds; /* head of the origin snapshot */
 715
 716                 dsphys->ds_prev_snap_obj = origin->ds_object;
 717                 dsphys->ds_prev_snap_txg =
 718                     origin->ds_phys->ds_creation_txg;
 719                 dsphys->ds_referenced_bytes =
 720                     origin->ds_phys->ds_referenced_bytes;
 721                 dsphys->ds_compressed_bytes =
 722                     origin->ds_phys->ds_compressed_bytes;
 723                 dsphys->ds_uncompressed_bytes =
 724                     origin->ds_phys->ds_uncompressed_bytes;
 725                 dsphys->ds_bp = origin->ds_phys->ds_bp;
 726
 727                 /*
 728                  * Inherit flags that describe the dataset's contents
 729                  * (INCONSISTENT) or properties (Case Insensitive).
 730                  */
 731                 dsphys->ds_flags |= origin->ds_phys->ds_flags &
 732                     (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
 733
 734                 if (origin->ds_large_blocks)
 735                         dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
 736
 737                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 738                 origin->ds_phys->ds_num_children++;
 739
 740                 VERIFY0(dsl_dataset_hold_obj(dp,
 741                     origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
 742                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 743                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 744                 dsl_dataset_rele(ohds, FTAG);
 745
 746                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 747                         if (origin->ds_phys->ds_next_clones_obj == 0) {
 748                                 origin->ds_phys->ds_next_clones_obj =
 749                                     zap_create(mos,
 750                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 751                         }
 752                         VERIFY0(zap_add_int(mos,
 753                             origin->ds_phys->ds_next_clones_obj, dsobj, tx));
 754                 }
 755
 756                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 757                 dd->dd_phys->dd_origin_obj = origin->ds_object;
 758                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 759                         if (origin->ds_dir->dd_phys->dd_clones == 0) {
 760                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 761                                 origin->ds_dir->dd_phys->dd_clones =
 762                                     zap_create(mos,
 763                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 764                         }
 765                         VERIFY0(zap_add_int(mos,
 766                             origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
 767                 }
 768         }
 769
 770         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 771                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 772
 773         dmu_buf_rele(dbuf, FTAG);
 774
 775         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 776         dd->dd_phys->dd_head_dataset_obj = dsobj;
 777
 778         return (dsobj);
 779 }
 780
 781 static void
 782 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
 783 {
 784         objset_t *os;
 785
 786         VERIFY0(dmu_objset_from_ds(ds, &os));
 787         bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 788         dsl_dataset_dirty(ds, tx);
 789 }
 790
 791 uint64_t
 792 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 793     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 794 {
 795         dsl_pool_t *dp = pdd->dd_pool;
 796         uint64_t dsobj, ddobj;
 797         dsl_dir_t *dd;
 798
 799         ASSERT(dmu_tx_is_syncing(tx));
 800         ASSERT(lastname[0] != '@');
 801
 802         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 803         VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
 804
 805         dsobj = dsl_dataset_create_sync_dd(dd, origin,
 806             flags & ~DS_CREATE_FLAG_NODIRTY, tx);
 807
 808         dsl_deleg_set_create_perms(dd, tx, cr);
 809
 810         /*
 811          * Since we're creating a new node we know it's a leaf, so we can
 812          * initialize the counts if the limit feature is active.
 813          */
 814         if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
 815                 uint64_t cnt = 0;
 816                 objset_t *os = dd->dd_pool->dp_meta_objset;
 817
 818                 dsl_dir_zapify(dd, tx);
 819                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
 820                     sizeof (cnt), 1, &cnt, tx));
 821                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
 822                     sizeof (cnt), 1, &cnt, tx));
 823         }
 824
 825         dsl_dir_rele(dd, FTAG);
 826
 827         /*
 828          * If we are creating a clone, make sure we zero out any stale
 829          * data from the origin snapshots zil header.
 830          */
 831         if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
 832                 dsl_dataset_t *ds;
 833
 834                 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 835                 dsl_dataset_zero_zil(ds, tx);
 836                 dsl_dataset_rele(ds, FTAG);
 837         }
 838
 839         return (dsobj);
 840 }
 841
 842 #ifdef __FreeBSD__
 843 /* FreeBSD ioctl compat begin */
 844 struct destroyarg {
 845         nvlist_t *nvl;
 846         const char *snapname;
 847 };
 848
 849 static int
 850 dsl_check_snap_cb(const char *name, void *arg)
 851 {
 852         struct destroyarg *da = arg;
 853         dsl_dataset_t *ds;
 854         char *dsname;
 855
 856         dsname = kmem_asprintf("%s@%s", name, da->snapname);
 857         fnvlist_add_boolean(da->nvl, dsname);
 858         kmem_free(dsname, strlen(dsname) + 1);
 859
 860         return (0);
 861 }
 862
 863 int
 864 dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
 865     nvlist_t *snaps)
 866 {
 867         struct destroyarg *da;
 868         int err;
 869
 870         da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
 871         da->nvl = snaps;
 872         da->snapname = snapname;
 873         err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
 874             DS_FIND_CHILDREN);
 875         kmem_free(da, sizeof (struct destroyarg));
 876
 877         return (err);
 878 }
 879 /* FreeBSD ioctl compat end */
 880 #endif /* __FreeBSD__ */
 881
 882 /*
 883  * The unique space in the head dataset can be calculated by subtracting
 884  * the space used in the most recent snapshot, that is still being used
 885  * in this file system, from the space currently in use.  To figure out
 886  * the space in the most recent snapshot still in use, we need to take
 887  * the total space used in the snapshot and subtract out the space that
 888  * has been freed up since the snapshot was taken.
 889  */
 890 void
 891 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
 892 {
 893         uint64_t mrs_used;
 894         uint64_t dlused, dlcomp, dluncomp;
 895
 896         ASSERT(!dsl_dataset_is_snapshot(ds));
 897
 898         if (ds->ds_phys->ds_prev_snap_obj != 0)
 899                 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
 900         else
 901                 mrs_used = 0;
 902
 903         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
 904
 905         ASSERT3U(dlused, <=, mrs_used);
 906         ds->ds_phys->ds_unique_bytes =
 907             ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
 908
 909         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
 910             SPA_VERSION_UNIQUE_ACCURATE)
 911                 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 912 }
 913
 914 void
 915 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
 916     dmu_tx_t *tx)
 917 {
 918         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 919         uint64_t count;
 920         int err;
 921
 922         ASSERT(ds->ds_phys->ds_num_children >= 2);
 923         err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
 924         /*
 925          * The err should not be ENOENT, but a bug in a previous version
 926          * of the code could cause upgrade_clones_cb() to not set
 927          * ds_next_snap_obj when it should, leading to a missing entry.
 928          * If we knew that the pool was created after
 929          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
 930          * ENOENT.  However, at least we can check that we don't have
 931          * too many entries in the next_clones_obj even after failing to
 932          * remove this one.
 933          */
 934         if (err != ENOENT)
 935                 VERIFY0(err);
 936         ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
 937             &count));
 938         ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
 939 }
 940
 941
 942 blkptr_t *
 943 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
 944 {
 945         return (&ds->ds_phys->ds_bp);
 946 }
 947
 948 void
 949 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
 950 {
 951         ASSERT(dmu_tx_is_syncing(tx));
 952         /* If it's the meta-objset, set dp_meta_rootbp */
 953         if (ds == NULL) {
 954                 tx->tx_pool->dp_meta_rootbp = *bp;
 955         } else {
 956                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
 957                 ds->ds_phys->ds_bp = *bp;
 958         }
 959 }
 960
 961 spa_t *
 962 dsl_dataset_get_spa(dsl_dataset_t *ds)
 963 {
 964         return (ds->ds_dir->dd_pool->dp_spa);
 965 }
 966
 967 void
 968 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
 969 {
 970         dsl_pool_t *dp;
 971
 972         if (ds == NULL) /* this is the meta-objset */
 973                 return;
 974
 975         ASSERT(ds->ds_objset != NULL);
 976
 977         if (ds->ds_phys->ds_next_snap_obj != 0)
 978                 panic("dirtying snapshot!");
 979
 980         dp = ds->ds_dir->dd_pool;
 981
 982         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
 983                 /* up the hold count until we can be written out */
 984                 dmu_buf_add_ref(ds->ds_dbuf, ds);
 985         }
 986 }
 987
 988 boolean_t
 989 dsl_dataset_is_dirty(dsl_dataset_t *ds)
 990 {
 991         for (int t = 0; t < TXG_SIZE; t++) {
 992                 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
 993                     ds, t))
 994                         return (B_TRUE);
 995         }
 996         return (B_FALSE);
 997 }
 998
 999 static int
1000 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1001 {
1002         uint64_t asize;
1003
1004         if (!dmu_tx_is_syncing(tx))
1005                 return (0);
1006
1007         /*
1008          * If there's an fs-only reservation, any blocks that might become
1009          * owned by the snapshot dataset must be accommodated by space
1010          * outside of the reservation.
1011          */
1012         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1013         asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1014         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1015                 return (SET_ERROR(ENOSPC));
1016
1017         /*
1018          * Propagate any reserved space for this snapshot to other
1019          * snapshot checks in this sync group.
1020          */
1021         if (asize > 0)
1022                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1023
1024         return (0);
1025 }
1026
1027 typedef struct dsl_dataset_snapshot_arg {
1028         nvlist_t *ddsa_snaps;
1029         nvlist_t *ddsa_props;
1030         nvlist_t *ddsa_errors;
1031         cred_t *ddsa_cr;
1032 } dsl_dataset_snapshot_arg_t;
1033
1034 int
1035 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1036     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1037 {
1038         int error;
1039         uint64_t value;
1040
1041         ds->ds_trysnap_txg = tx->tx_txg;
1042
1043         if (!dmu_tx_is_syncing(tx))
1044                 return (0);
1045
1046         /*
1047          * We don't allow multiple snapshots of the same txg.  If there
1048          * is already one, try again.
1049          */
1050         if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1051                 return (SET_ERROR(EAGAIN));
1052
1053         /*
1054          * Check for conflicting snapshot name.
1055          */
1056         error = dsl_dataset_snap_lookup(ds, snapname, &value);
1057         if (error == 0)
1058                 return (SET_ERROR(EEXIST));
1059         if (error != ENOENT)
1060                 return (error);
1061
1062         /*
1063          * We don't allow taking snapshots of inconsistent datasets, such as
1064          * those into which we are currently receiving.  However, if we are
1065          * creating this snapshot as part of a receive, this check will be
1066          * executed atomically with respect to the completion of the receive
1067          * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1068          * case we ignore this, knowing it will be fixed up for us shortly in
1069          * dmu_recv_end_sync().
1070          */
1071         if (!recv && DS_IS_INCONSISTENT(ds))
1072                 return (SET_ERROR(EBUSY));
1073
1074         /*
1075          * Skip the check for temporary snapshots or if we have already checked
1076          * the counts in dsl_dataset_snapshot_check. This means we really only
1077          * check the count here when we're receiving a stream.
1078          */
1079         if (cnt != 0 && cr != NULL) {
1080                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1081                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1082                 if (error != 0)
1083                         return (error);
1084         }
1085
1086         error = dsl_dataset_snapshot_reserve_space(ds, tx);
1087         if (error != 0)
1088                 return (error);
1089
1090         return (0);
1091 }
1092
1093 static int
1094 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1095 {
1096         dsl_dataset_snapshot_arg_t *ddsa = arg;
1097         dsl_pool_t *dp = dmu_tx_pool(tx);
1098         nvpair_t *pair;
1099         int rv = 0;
1100
1101         /*
1102          * Pre-compute how many total new snapshots will be created for each
1103          * level in the tree and below. This is needed for validating the
1104          * snapshot limit when either taking a recursive snapshot or when
1105          * taking multiple snapshots.
1106          *
1107          * The problem is that the counts are not actually adjusted when
1108          * we are checking, only when we finally sync. For a single snapshot,
1109          * this is easy, the count will increase by 1 at each node up the tree,
1110          * but its more complicated for the recursive/multiple snapshot case.
1111          *
1112          * The dsl_fs_ss_limit_check function does recursively check the count
1113          * at each level up the tree but since it is validating each snapshot
1114          * independently we need to be sure that we are validating the complete
1115          * count for the entire set of snapshots. We do this by rolling up the
1116          * counts for each component of the name into an nvlist and then
1117          * checking each of those cases with the aggregated count.
1118          *
1119          * This approach properly handles not only the recursive snapshot
1120          * case (where we get all of those on the ddsa_snaps list) but also
1121          * the sibling case (e.g. snapshot a/b and a/c so that we will also
1122          * validate the limit on 'a' using a count of 2).
1123          *
1124          * We validate the snapshot names in the third loop and only report
1125          * name errors once.
1126          */
1127         if (dmu_tx_is_syncing(tx)) {
1128                 nvlist_t *cnt_track = NULL;
1129                 cnt_track = fnvlist_alloc();
1130
1131                 /* Rollup aggregated counts into the cnt_track list */
1132                 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1133                     pair != NULL;
1134                     pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1135                         char *pdelim;
1136                         uint64_t val;
1137                         char nm[MAXPATHLEN];
1138
1139                         (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1140                         pdelim = strchr(nm, '@');
1141                         if (pdelim == NULL)
1142                                 continue;
1143                         *pdelim = '\0';
1144
1145                         do {
1146                                 if (nvlist_lookup_uint64(cnt_track, nm,
1147                                     &val) == 0) {
1148                                         /* update existing entry */
1149                                         fnvlist_add_uint64(cnt_track, nm,
1150                                             val + 1);
1151                                 } else {
1152                                         /* add to list */
1153                                         fnvlist_add_uint64(cnt_track, nm, 1);
1154                                 }
1155
1156                                 pdelim = strrchr(nm, '/');
1157                                 if (pdelim != NULL)
1158                                         *pdelim = '\0';
1159                         } while (pdelim != NULL);
1160                 }
1161
1162                 /* Check aggregated counts at each level */
1163                 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1164                     pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1165                         int error = 0;
1166                         char *name;
1167                         uint64_t cnt = 0;
1168                         dsl_dataset_t *ds;
1169
1170                         name = nvpair_name(pair);
1171                         cnt = fnvpair_value_uint64(pair);
1172                         ASSERT(cnt > 0);
1173
1174                         error = dsl_dataset_hold(dp, name, FTAG, &ds);
1175                         if (error == 0) {
1176                                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1177                                     ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1178                                     ddsa->ddsa_cr);
1179                                 dsl_dataset_rele(ds, FTAG);
1180                         }
1181
1182                         if (error != 0) {
1183                                 if (ddsa->ddsa_errors != NULL)
1184                                         fnvlist_add_int32(ddsa->ddsa_errors,
1185                                             name, error);
1186                                 rv = error;
1187                                 /* only report one error for this check */
1188                                 break;
1189                         }
1190                 }
1191                 nvlist_free(cnt_track);
1192         }
1193
1194         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1195             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1196                 int error = 0;
1197                 dsl_dataset_t *ds;
1198                 char *name, *atp;
1199                 char dsname[MAXNAMELEN];
1200
1201                 name = nvpair_name(pair);
1202                 if (strlen(name) >= MAXNAMELEN)
1203                         error = SET_ERROR(ENAMETOOLONG);
1204                 if (error == 0) {
1205                         atp = strchr(name, '@');
1206                         if (atp == NULL)
1207                                 error = SET_ERROR(EINVAL);
1208                         if (error == 0)
1209                                 (void) strlcpy(dsname, name, atp - name + 1);
1210                 }
1211                 if (error == 0)
1212                         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1213                 if (error == 0) {
1214                         /* passing 0/NULL skips dsl_fs_ss_limit_check */
1215                         error = dsl_dataset_snapshot_check_impl(ds,
1216                             atp + 1, tx, B_FALSE, 0, NULL);
1217                         dsl_dataset_rele(ds, FTAG);
1218                 }
1219
1220                 if (error != 0) {
1221                         if (ddsa->ddsa_errors != NULL) {
1222                                 fnvlist_add_int32(ddsa->ddsa_errors,
1223                                     name, error);
1224                         }
1225                         rv = error;
1226                 }
1227         }
1228
1229         return (rv);
1230 }
1231
1232 void
1233 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1234     dmu_tx_t *tx)
1235 {
1236         static zil_header_t zero_zil;
1237
1238         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1239         dmu_buf_t *dbuf;
1240         dsl_dataset_phys_t *dsphys;
1241         uint64_t dsobj, crtxg;
1242         objset_t *mos = dp->dp_meta_objset;
1243         objset_t *os;
1244
1245         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1246
1247         /*
1248          * If we are on an old pool, the zil must not be active, in which
1249          * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1250          */
1251         ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1252             dmu_objset_from_ds(ds, &os) != 0 ||
1253             bcmp(&os->os_phys->os_zil_header, &zero_zil,
1254             sizeof (zero_zil)) == 0);
1255
1256         dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1257
1258         /*
1259          * The origin's ds_creation_txg has to be < TXG_INITIAL
1260          */
1261         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1262                 crtxg = 1;
1263         else
1264                 crtxg = tx->tx_txg;
1265
1266         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1267             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1268         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1269         dmu_buf_will_dirty(dbuf, tx);
1270         dsphys = dbuf->db_data;
1271         bzero(dsphys, sizeof (dsl_dataset_phys_t));
1272         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1273         dsphys->ds_fsid_guid = unique_create();
1274         do {
1275                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1276                     sizeof (dsphys->ds_guid));
1277         } while (dsphys->ds_guid == 0);
1278         dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1279         dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1280         dsphys->ds_next_snap_obj = ds->ds_object;
1281         dsphys->ds_num_children = 1;
1282         dsphys->ds_creation_time = gethrestime_sec();
1283         dsphys->ds_creation_txg = crtxg;
1284         dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1285         dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
1286         dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1287         dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1288         dsphys->ds_flags = ds->ds_phys->ds_flags;
1289         dsphys->ds_bp = ds->ds_phys->ds_bp;
1290         dmu_buf_rele(dbuf, FTAG);
1291
1292         if (ds->ds_large_blocks)
1293                 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1294
1295         ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1296         if (ds->ds_prev) {
1297                 uint64_t next_clones_obj =
1298                     ds->ds_prev->ds_phys->ds_next_clones_obj;
1299                 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1300                     ds->ds_object ||
1301                     ds->ds_prev->ds_phys->ds_num_children > 1);
1302                 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1303                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1304                         ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1305                             ds->ds_prev->ds_phys->ds_creation_txg);
1306                         ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1307                 } else if (next_clones_obj != 0) {
1308                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
1309                             dsphys->ds_next_snap_obj, tx);
1310                         VERIFY0(zap_add_int(mos,
1311                             next_clones_obj, dsobj, tx));
1312                 }
1313         }
1314
1315         /*
1316          * If we have a reference-reservation on this dataset, we will
1317          * need to increase the amount of refreservation being charged
1318          * since our unique space is going to zero.
1319          */
1320         if (ds->ds_reserved) {
1321                 int64_t delta;
1322                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1323                 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1324                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1325                     delta, 0, 0, tx);
1326         }
1327
1328         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1329         ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
1330             UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
1331         dsl_deadlist_close(&ds->ds_deadlist);
1332         dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1333         dsl_deadlist_add_key(&ds->ds_deadlist,
1334             ds->ds_phys->ds_prev_snap_txg, tx);
1335
1336         ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1337         ds->ds_phys->ds_prev_snap_obj = dsobj;
1338         ds->ds_phys->ds_prev_snap_txg = crtxg;
1339         ds->ds_phys->ds_unique_bytes = 0;
1340         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1341                 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1342
1343         VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1344             snapname, 8, 1, &dsobj, tx));
1345
1346         if (ds->ds_prev)
1347                 dsl_dataset_rele(ds->ds_prev, ds);
1348         VERIFY0(dsl_dataset_hold_obj(dp,
1349             ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1350
1351         dsl_scan_ds_snapshotted(ds, tx);
1352
1353         dsl_dir_snap_cmtime_update(ds->ds_dir);
1354
1355         spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1356 }
1357
1358 static void
1359 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1360 {
1361         dsl_dataset_snapshot_arg_t *ddsa = arg;
1362         dsl_pool_t *dp = dmu_tx_pool(tx);
1363         nvpair_t *pair;
1364
1365         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1366             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1367                 dsl_dataset_t *ds;
1368                 char *name, *atp;
1369                 char dsname[MAXNAMELEN];
1370
1371                 name = nvpair_name(pair);
1372                 atp = strchr(name, '@');
1373                 (void) strlcpy(dsname, name, atp - name + 1);
1374                 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1375
1376                 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1377                 if (ddsa->ddsa_props != NULL) {
1378                         dsl_props_set_sync_impl(ds->ds_prev,
1379                             ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1380                 }
1381                 dsl_dataset_rele(ds, FTAG);
1382         }
1383 }
1384
1385 /*
1386  * The snapshots must all be in the same pool.
1387  * All-or-nothing: if there are any failures, nothing will be modified.
1388  */
1389 int
1390 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1391 {
1392         dsl_dataset_snapshot_arg_t ddsa;
1393         nvpair_t *pair;
1394         boolean_t needsuspend;
1395         int error;
1396         spa_t *spa;
1397         char *firstname;
1398         nvlist_t *suspended = NULL;
1399
1400         pair = nvlist_next_nvpair(snaps, NULL);
1401         if (pair == NULL)
1402                 return (0);
1403         firstname = nvpair_name(pair);
1404
1405         error = spa_open(firstname, &spa, FTAG);
1406         if (error != 0)
1407                 return (error);
1408         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1409         spa_close(spa, FTAG);
1410
1411         if (needsuspend) {
1412                 suspended = fnvlist_alloc();
1413                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1414                     pair = nvlist_next_nvpair(snaps, pair)) {
1415                         char fsname[MAXNAMELEN];
1416                         char *snapname = nvpair_name(pair);
1417                         char *atp;
1418                         void *cookie;
1419
1420                         atp = strchr(snapname, '@');
1421                         if (atp == NULL) {
1422                                 error = SET_ERROR(EINVAL);
1423                                 break;
1424                         }
1425                         (void) strlcpy(fsname, snapname, atp - snapname + 1);
1426
1427                         error = zil_suspend(fsname, &cookie);
1428                         if (error != 0)
1429                                 break;
1430                         fnvlist_add_uint64(suspended, fsname,
1431                             (uintptr_t)cookie);
1432                 }
1433         }
1434
1435         ddsa.ddsa_snaps = snaps;
1436         ddsa.ddsa_props = props;
1437         ddsa.ddsa_errors = errors;
1438         ddsa.ddsa_cr = CRED();
1439
1440         if (error == 0) {
1441                 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1442                     dsl_dataset_snapshot_sync, &ddsa,
1443                     fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1444         }
1445
1446         if (suspended != NULL) {
1447                 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1448                     pair = nvlist_next_nvpair(suspended, pair)) {
1449                         zil_resume((void *)(uintptr_t)
1450                             fnvpair_value_uint64(pair));
1451                 }
1452                 fnvlist_free(suspended);
1453         }
1454
1455 #ifdef __FreeBSD__
1456 #ifdef _KERNEL
1457         if (error == 0) {
1458                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1459                     pair = nvlist_next_nvpair(snaps, pair)) {
1460                         char *snapname = nvpair_name(pair);
1461                         zvol_create_minors(snapname);
1462                 }
1463         }
1464 #endif
1465 #endif
1466         return (error);
1467 }
1468
1469 typedef struct dsl_dataset_snapshot_tmp_arg {
1470         const char *ddsta_fsname;
1471         const char *ddsta_snapname;
1472         minor_t ddsta_cleanup_minor;
1473         const char *ddsta_htag;
1474 } dsl_dataset_snapshot_tmp_arg_t;
1475
1476 static int
1477 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1478 {
1479         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1480         dsl_pool_t *dp = dmu_tx_pool(tx);
1481         dsl_dataset_t *ds;
1482         int error;
1483
1484         error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1485         if (error != 0)
1486                 return (error);
1487
1488         /* NULL cred means no limit check for tmp snapshot */
1489         error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1490             tx, B_FALSE, 0, NULL);
1491         if (error != 0) {
1492                 dsl_dataset_rele(ds, FTAG);
1493                 return (error);
1494         }
1495
1496         if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1497                 dsl_dataset_rele(ds, FTAG);
1498                 return (SET_ERROR(ENOTSUP));
1499         }
1500         error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1501             B_TRUE, tx);
1502         if (error != 0) {
1503                 dsl_dataset_rele(ds, FTAG);
1504                 return (error);
1505         }
1506
1507         dsl_dataset_rele(ds, FTAG);
1508         return (0);
1509 }
1510
1511 static void
1512 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1513 {
1514         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1515         dsl_pool_t *dp = dmu_tx_pool(tx);
1516         dsl_dataset_t *ds;
1517
1518         VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1519
1520         dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1521         dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1522             ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1523         dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1524
1525         dsl_dataset_rele(ds, FTAG);
1526 }
1527
1528 int
1529 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1530     minor_t cleanup_minor, const char *htag)
1531 {
1532         dsl_dataset_snapshot_tmp_arg_t ddsta;
1533         int error;
1534         spa_t *spa;
1535         boolean_t needsuspend;
1536         void *cookie;
1537
1538         ddsta.ddsta_fsname = fsname;
1539         ddsta.ddsta_snapname = snapname;
1540         ddsta.ddsta_cleanup_minor = cleanup_minor;
1541         ddsta.ddsta_htag = htag;
1542
1543         error = spa_open(fsname, &spa, FTAG);
1544         if (error != 0)
1545                 return (error);
1546         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1547         spa_close(spa, FTAG);
1548
1549         if (needsuspend) {
1550                 error = zil_suspend(fsname, &cookie);
1551                 if (error != 0)
1552                         return (error);
1553         }
1554
1555         error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1556             dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1557
1558         if (needsuspend)
1559                 zil_resume(cookie);
1560         return (error);
1561 }
1562
1563
1564 void
1565 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1566 {
1567         ASSERT(dmu_tx_is_syncing(tx));
1568         ASSERT(ds->ds_objset != NULL);
1569         ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1570
1571         /*
1572          * in case we had to change ds_fsid_guid when we opened it,
1573          * sync it out now.
1574          */
1575         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1576         ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1577
1578         dmu_objset_sync(ds->ds_objset, zio, tx);
1579
1580         if (ds->ds_need_large_blocks && !ds->ds_large_blocks) {
1581                 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
1582                 ds->ds_large_blocks = B_TRUE;
1583         }
1584 }
1585
1586 static void
1587 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1588 {
1589         uint64_t count = 0;
1590         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1591         zap_cursor_t zc;
1592         zap_attribute_t za;
1593         nvlist_t *propval = fnvlist_alloc();
1594         nvlist_t *val = fnvlist_alloc();
1595
1596         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1597
1598         /*
1599          * There may be missing entries in ds_next_clones_obj
1600          * due to a bug in a previous version of the code.
1601          * Only trust it if it has the right number of entries.
1602          */
1603         if (ds->ds_phys->ds_next_clones_obj != 0) {
1604                 VERIFY0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1605                     &count));
1606         }
1607         if (count != ds->ds_phys->ds_num_children - 1)
1608                 goto fail;
1609         for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
1610             zap_cursor_retrieve(&zc, &za) == 0;
1611             zap_cursor_advance(&zc)) {
1612                 dsl_dataset_t *clone;
1613                 char buf[ZFS_MAXNAMELEN];
1614                 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1615                     za.za_first_integer, FTAG, &clone));
1616                 dsl_dir_name(clone->ds_dir, buf);
1617                 fnvlist_add_boolean(val, buf);
1618                 dsl_dataset_rele(clone, FTAG);
1619         }
1620         zap_cursor_fini(&zc);
1621         fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1622         fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1623 fail:
1624         nvlist_free(val);
1625         nvlist_free(propval);
1626 }
1627
1628 void
1629 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1630 {
1631         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1632         uint64_t refd, avail, uobjs, aobjs, ratio;
1633
1634         ASSERT(dsl_pool_config_held(dp));
1635
1636         ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1637             (ds->ds_phys->ds_uncompressed_bytes * 100 /
1638             ds->ds_phys->ds_compressed_bytes);
1639
1640         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1641         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1642             ds->ds_phys->ds_uncompressed_bytes);
1643
1644         if (dsl_dataset_is_snapshot(ds)) {
1645                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1646                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1647                     ds->ds_phys->ds_unique_bytes);
1648                 get_clones_stat(ds, nv);
1649         } else {
1650                 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1651                         char buf[MAXNAMELEN];
1652                         dsl_dataset_name(ds->ds_prev, buf);
1653                         dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1654                 }
1655
1656                 dsl_dir_stats(ds->ds_dir, nv);
1657         }
1658
1659         dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1660         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1661         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1662
1663         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1664             ds->ds_phys->ds_creation_time);
1665         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1666             ds->ds_phys->ds_creation_txg);
1667         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1668             ds->ds_quota);
1669         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1670             ds->ds_reserved);
1671         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1672             ds->ds_phys->ds_guid);
1673         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1674             ds->ds_phys->ds_unique_bytes);
1675         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1676             ds->ds_object);
1677         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1678             ds->ds_userrefs);
1679         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1680             DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1681
1682         if (ds->ds_phys->ds_prev_snap_obj != 0) {
1683                 uint64_t written, comp, uncomp;
1684                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1685                 dsl_dataset_t *prev;
1686
1687                 int err = dsl_dataset_hold_obj(dp,
1688                     ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
1689                 if (err == 0) {
1690                         err = dsl_dataset_space_written(prev, ds, &written,
1691                             &comp, &uncomp);
1692                         dsl_dataset_rele(prev, FTAG);
1693                         if (err == 0) {
1694                                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1695                                     written);
1696                         }
1697                 }
1698         }
1699 }
1700
1701 void
1702 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1703 {
1704         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1705         ASSERT(dsl_pool_config_held(dp));
1706
1707         stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1708         stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1709         stat->dds_guid = ds->ds_phys->ds_guid;
1710         stat->dds_origin[0] = '\0';
1711         if (dsl_dataset_is_snapshot(ds)) {
1712                 stat->dds_is_snapshot = B_TRUE;
1713                 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1714         } else {
1715                 stat->dds_is_snapshot = B_FALSE;
1716                 stat->dds_num_clones = 0;
1717
1718                 if (dsl_dir_is_clone(ds->ds_dir)) {
1719                         dsl_dataset_t *ods;
1720
1721                         VERIFY0(dsl_dataset_hold_obj(dp,
1722                             ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1723                         dsl_dataset_name(ods, stat->dds_origin);
1724                         dsl_dataset_rele(ods, FTAG);
1725                 }
1726         }
1727 }
1728
1729 uint64_t
1730 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1731 {
1732         return (ds->ds_fsid_guid);
1733 }
1734
1735 void
1736 dsl_dataset_space(dsl_dataset_t *ds,
1737     uint64_t *refdbytesp, uint64_t *availbytesp,
1738     uint64_t *usedobjsp, uint64_t *availobjsp)
1739 {
1740         *refdbytesp = ds->ds_phys->ds_referenced_bytes;
1741         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1742         if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1743                 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1744         if (ds->ds_quota != 0) {
1745                 /*
1746                  * Adjust available bytes according to refquota
1747                  */
1748                 if (*refdbytesp < ds->ds_quota)
1749                         *availbytesp = MIN(*availbytesp,
1750                             ds->ds_quota - *refdbytesp);
1751                 else
1752                         *availbytesp = 0;
1753         }
1754         *usedobjsp = BP_GET_FILL(&ds->ds_phys->ds_bp);
1755         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1756 }
1757
1758 boolean_t
1759 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1760 {
1761         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1762
1763         ASSERT(dsl_pool_config_held(dp));
1764         if (snap == NULL)
1765                 return (B_FALSE);
1766         if (ds->ds_phys->ds_bp.blk_birth >
1767             snap->ds_phys->ds_creation_txg) {
1768                 objset_t *os, *os_snap;
1769                 /*
1770                  * It may be that only the ZIL differs, because it was
1771                  * reset in the head.  Don't count that as being
1772                  * modified.
1773                  */
1774                 if (dmu_objset_from_ds(ds, &os) != 0)
1775                         return (B_TRUE);
1776                 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1777                         return (B_TRUE);
1778                 return (bcmp(&os->os_phys->os_meta_dnode,
1779                     &os_snap->os_phys->os_meta_dnode,
1780                     sizeof (os->os_phys->os_meta_dnode)) != 0);
1781         }
1782         return (B_FALSE);
1783 }
1784
1785 typedef struct dsl_dataset_rename_snapshot_arg {
1786         const char *ddrsa_fsname;
1787         const char *ddrsa_oldsnapname;
1788         const char *ddrsa_newsnapname;
1789         boolean_t ddrsa_recursive;
1790         dmu_tx_t *ddrsa_tx;
1791 } dsl_dataset_rename_snapshot_arg_t;
1792
1793 /* ARGSUSED */
1794 static int
1795 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1796     dsl_dataset_t *hds, void *arg)
1797 {
1798         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1799         int error;
1800         uint64_t val;
1801
1802         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1803         if (error != 0) {
1804                 /* ignore nonexistent snapshots */
1805                 return (error == ENOENT ? 0 : error);
1806         }
1807
1808         /* new name should not exist */
1809         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1810         if (error == 0)
1811                 error = SET_ERROR(EEXIST);
1812         else if (error == ENOENT)
1813                 error = 0;
1814
1815         /* dataset name + 1 for the "@" + the new snapshot name must fit */
1816         if (dsl_dir_namelen(hds->ds_dir) + 1 +
1817             strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1818                 error = SET_ERROR(ENAMETOOLONG);
1819
1820         return (error);
1821 }
1822
1823 static int
1824 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1825 {
1826         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1827         dsl_pool_t *dp = dmu_tx_pool(tx);
1828         dsl_dataset_t *hds;
1829         int error;
1830
1831         error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1832         if (error != 0)
1833                 return (error);
1834
1835         if (ddrsa->ddrsa_recursive) {
1836                 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1837                     dsl_dataset_rename_snapshot_check_impl, ddrsa,
1838                     DS_FIND_CHILDREN);
1839         } else {
1840                 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1841         }
1842         dsl_dataset_rele(hds, FTAG);
1843         return (error);
1844 }
1845
1846 static int
1847 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1848     dsl_dataset_t *hds, void *arg)
1849 {
1850 #ifdef __FreeBSD__
1851 #ifdef _KERNEL
1852         char *oldname, *newname;
1853 #endif
1854 #endif
1855         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1856         dsl_dataset_t *ds;
1857         uint64_t val;
1858         dmu_tx_t *tx = ddrsa->ddrsa_tx;
1859         int error;
1860
1861         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1862         ASSERT(error == 0 || error == ENOENT);
1863         if (error == ENOENT) {
1864                 /* ignore nonexistent snapshots */
1865                 return (0);
1866         }
1867
1868         VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1869
1870         /* log before we change the name */
1871         spa_history_log_internal_ds(ds, "rename", tx,
1872             "-> @%s", ddrsa->ddrsa_newsnapname);
1873
1874         VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1875             B_FALSE));
1876         mutex_enter(&ds->ds_lock);
1877         (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1878         mutex_exit(&ds->ds_lock);
1879         VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj,
1880             ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1881
1882 #ifdef __FreeBSD__
1883 #ifdef _KERNEL
1884         oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1885         newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1886         snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
1887             ddrsa->ddrsa_oldsnapname);
1888         snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
1889             ddrsa->ddrsa_newsnapname);
1890         zfsvfs_update_fromname(oldname, newname);
1891         zvol_rename_minors(oldname, newname);
1892         kmem_free(newname, MAXPATHLEN);
1893         kmem_free(oldname, MAXPATHLEN);
1894 #endif
1895 #endif
1896         dsl_dataset_rele(ds, FTAG);
1897
1898         return (0);
1899 }
1900
1901 static void
1902 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1903 {
1904         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1905         dsl_pool_t *dp = dmu_tx_pool(tx);
1906         dsl_dataset_t *hds;
1907
1908         VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1909         ddrsa->ddrsa_tx = tx;
1910         if (ddrsa->ddrsa_recursive) {
1911                 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1912                     dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1913                     DS_FIND_CHILDREN));
1914         } else {
1915                 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1916         }
1917         dsl_dataset_rele(hds, FTAG);
1918 }
1919
1920 int
1921 dsl_dataset_rename_snapshot(const char *fsname,
1922     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1923 {
1924         dsl_dataset_rename_snapshot_arg_t ddrsa;
1925
1926         ddrsa.ddrsa_fsname = fsname;
1927         ddrsa.ddrsa_oldsnapname = oldsnapname;
1928         ddrsa.ddrsa_newsnapname = newsnapname;
1929         ddrsa.ddrsa_recursive = recursive;
1930
1931         return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1932             dsl_dataset_rename_snapshot_sync, &ddrsa,
1933             1, ZFS_SPACE_CHECK_RESERVED));
1934 }
1935
1936 /*
1937  * If we're doing an ownership handoff, we need to make sure that there is
1938  * only one long hold on the dataset.  We're not allowed to change anything here
1939  * so we don't permanently release the long hold or regular hold here.  We want
1940  * to do this only when syncing to avoid the dataset unexpectedly going away
1941  * when we release the long hold.
1942  */
1943 static int
1944 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1945 {
1946         boolean_t held;
1947
1948         if (!dmu_tx_is_syncing(tx))
1949                 return (0);
1950
1951         if (owner != NULL) {
1952                 VERIFY3P(ds->ds_owner, ==, owner);
1953                 dsl_dataset_long_rele(ds, owner);
1954         }
1955
1956         held = dsl_dataset_long_held(ds);
1957
1958         if (owner != NULL)
1959                 dsl_dataset_long_hold(ds, owner);
1960
1961         if (held)
1962                 return (SET_ERROR(EBUSY));
1963
1964         return (0);
1965 }
1966
1967 typedef struct dsl_dataset_rollback_arg {
1968         const char *ddra_fsname;
1969         void *ddra_owner;
1970         nvlist_t *ddra_result;
1971 } dsl_dataset_rollback_arg_t;
1972
1973 static int
1974 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1975 {
1976         dsl_dataset_rollback_arg_t *ddra = arg;
1977         dsl_pool_t *dp = dmu_tx_pool(tx);
1978         dsl_dataset_t *ds;
1979         int64_t unused_refres_delta;
1980         int error;
1981
1982         error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1983         if (error != 0)
1984                 return (error);
1985
1986         /* must not be a snapshot */
1987         if (dsl_dataset_is_snapshot(ds)) {
1988                 dsl_dataset_rele(ds, FTAG);
1989                 return (SET_ERROR(EINVAL));
1990         }
1991
1992         /* must have a most recent snapshot */
1993         if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) {
1994                 dsl_dataset_rele(ds, FTAG);
1995                 return (SET_ERROR(EINVAL));
1996         }
1997
1998         /* must not have any bookmarks after the most recent snapshot */
1999         nvlist_t *proprequest = fnvlist_alloc();
2000         fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2001         nvlist_t *bookmarks = fnvlist_alloc();
2002         error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2003         fnvlist_free(proprequest);
2004         if (error != 0)
2005                 return (error);
2006         for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2007             pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2008                 nvlist_t *valuenv =
2009                     fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2010                     zfs_prop_to_name(ZFS_PROP_CREATETXG));
2011                 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2012                 if (createtxg > ds->ds_phys->ds_prev_snap_txg) {
2013                         fnvlist_free(bookmarks);
2014                         dsl_dataset_rele(ds, FTAG);
2015                         return (SET_ERROR(EEXIST));
2016                 }
2017         }
2018         fnvlist_free(bookmarks);
2019
2020         error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2021         if (error != 0) {
2022                 dsl_dataset_rele(ds, FTAG);
2023                 return (error);
2024         }
2025
2026         /*
2027          * Check if the snap we are rolling back to uses more than
2028          * the refquota.
2029          */
2030         if (ds->ds_quota != 0 &&
2031             ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) {
2032                 dsl_dataset_rele(ds, FTAG);
2033                 return (SET_ERROR(EDQUOT));
2034         }
2035
2036         /*
2037          * When we do the clone swap, we will temporarily use more space
2038          * due to the refreservation (the head will no longer have any
2039          * unique space, so the entire amount of the refreservation will need
2040          * to be free).  We will immediately destroy the clone, freeing
2041          * this space, but the freeing happens over many txg's.
2042          */
2043         unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2044             ds->ds_phys->ds_unique_bytes);
2045
2046         if (unused_refres_delta > 0 &&
2047             unused_refres_delta >
2048             dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2049                 dsl_dataset_rele(ds, FTAG);
2050                 return (SET_ERROR(ENOSPC));
2051         }
2052
2053         dsl_dataset_rele(ds, FTAG);
2054         return (0);
2055 }
2056
2057 static void
2058 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2059 {
2060         dsl_dataset_rollback_arg_t *ddra = arg;
2061         dsl_pool_t *dp = dmu_tx_pool(tx);
2062         dsl_dataset_t *ds, *clone;
2063         uint64_t cloneobj;
2064         char namebuf[ZFS_MAXNAMELEN];
2065
2066         VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2067
2068         dsl_dataset_name(ds->ds_prev, namebuf);
2069         fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2070
2071         cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2072             ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2073
2074         VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2075
2076         dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2077         dsl_dataset_zero_zil(ds, tx);
2078
2079         dsl_destroy_head_sync_impl(clone, tx);
2080
2081         dsl_dataset_rele(clone, FTAG);
2082         dsl_dataset_rele(ds, FTAG);
2083 }
2084
2085 /*
2086  * Rolls back the given filesystem or volume to the most recent snapshot.
2087  * The name of the most recent snapshot will be returned under key "target"
2088  * in the result nvlist.
2089  *
2090  * If owner != NULL:
2091  * - The existing dataset MUST be owned by the specified owner at entry
2092  * - Upon return, dataset will still be held by the same owner, whether we
2093  *   succeed or not.
2094  *
2095  * This mode is required any time the existing filesystem is mounted.  See
2096  * notes above zfs_suspend_fs() for further details.
2097  */
2098 int
2099 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2100 {
2101         dsl_dataset_rollback_arg_t ddra;
2102
2103         ddra.ddra_fsname = fsname;
2104         ddra.ddra_owner = owner;
2105         ddra.ddra_result = result;
2106
2107         return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2108             dsl_dataset_rollback_sync, &ddra,
2109             1, ZFS_SPACE_CHECK_RESERVED));
2110 }
2111
2112 struct promotenode {
2113         list_node_t link;
2114         dsl_dataset_t *ds;
2115 };
2116
2117 typedef struct dsl_dataset_promote_arg {
2118         const char *ddpa_clonename;
2119         dsl_dataset_t *ddpa_clone;
2120         list_t shared_snaps, origin_snaps, clone_snaps;
2121         dsl_dataset_t *origin_origin; /* origin of the origin */
2122         uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2123         char *err_ds;
2124         cred_t *cr;
2125 } dsl_dataset_promote_arg_t;
2126
2127 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2128 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2129     void *tag);
2130 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2131
2132 static int
2133 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2134 {
2135         dsl_dataset_promote_arg_t *ddpa = arg;
2136         dsl_pool_t *dp = dmu_tx_pool(tx);
2137         dsl_dataset_t *hds;
2138         struct promotenode *snap;
2139         dsl_dataset_t *origin_ds;
2140         int err;
2141         uint64_t unused;
2142         uint64_t ss_mv_cnt;
2143
2144         err = promote_hold(ddpa, dp, FTAG);
2145         if (err != 0)
2146                 return (err);
2147
2148         hds = ddpa->ddpa_clone;
2149
2150         if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
2151                 promote_rele(ddpa, FTAG);
2152                 return (SET_ERROR(EXDEV));
2153         }
2154
2155         /*
2156          * Compute and check the amount of space to transfer.  Since this is
2157          * so expensive, don't do the preliminary check.
2158          */
2159         if (!dmu_tx_is_syncing(tx)) {
2160                 promote_rele(ddpa, FTAG);
2161                 return (0);
2162         }
2163
2164         snap = list_head(&ddpa->shared_snaps);
2165         origin_ds = snap->ds;
2166
2167         /* compute origin's new unique space */
2168         snap = list_tail(&ddpa->clone_snaps);
2169         ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2170         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2171             origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2172             &ddpa->unique, &unused, &unused);
2173
2174         /*
2175          * Walk the snapshots that we are moving
2176          *
2177          * Compute space to transfer.  Consider the incremental changes
2178          * to used by each snapshot:
2179          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2180          * So each snapshot gave birth to:
2181          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2182          * So a sequence would look like:
2183          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2184          * Which simplifies to:
2185          * uN + kN + kN-1 + ... + k1 + k0
2186          * Note however, if we stop before we reach the ORIGIN we get:
2187          * uN + kN + kN-1 + ... + kM - uM-1
2188          */
2189         ss_mv_cnt = 0;
2190         ddpa->used = origin_ds->ds_phys->ds_referenced_bytes;
2191         ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2192         ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2193         for (snap = list_head(&ddpa->shared_snaps); snap;
2194             snap = list_next(&ddpa->shared_snaps, snap)) {
2195                 uint64_t val, dlused, dlcomp, dluncomp;
2196                 dsl_dataset_t *ds = snap->ds;
2197
2198                 ss_mv_cnt++;
2199
2200                 /*
2201                  * If there are long holds, we won't be able to evict
2202                  * the objset.
2203                  */
2204                 if (dsl_dataset_long_held(ds)) {
2205                         err = SET_ERROR(EBUSY);
2206                         goto out;
2207                 }
2208
2209                 /* Check that the snapshot name does not conflict */
2210                 VERIFY0(dsl_dataset_get_snapname(ds));
2211                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2212                 if (err == 0) {
2213                         (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2214                         err = SET_ERROR(EEXIST);
2215                         goto out;
2216                 }
2217                 if (err != ENOENT)
2218                         goto out;
2219
2220                 /* The very first snapshot does not have a deadlist */
2221                 if (ds->ds_phys->ds_prev_snap_obj == 0)
2222                         continue;
2223
2224                 dsl_deadlist_space(&ds->ds_deadlist,
2225                     &dlused, &dlcomp, &dluncomp);
2226                 ddpa->used += dlused;
2227                 ddpa->comp += dlcomp;
2228                 ddpa->uncomp += dluncomp;
2229         }
2230
2231         /*
2232          * If we are a clone of a clone then we never reached ORIGIN,
2233          * so we need to subtract out the clone origin's used space.
2234          */
2235         if (ddpa->origin_origin) {
2236                 ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes;
2237                 ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes;
2238                 ddpa->uncomp -=
2239                     ddpa->origin_origin->ds_phys->ds_uncompressed_bytes;
2240         }
2241
2242         /* Check that there is enough space and limit headroom here */
2243         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2244             0, ss_mv_cnt, ddpa->used, ddpa->cr);
2245         if (err != 0)
2246                 goto out;
2247
2248         /*
2249          * Compute the amounts of space that will be used by snapshots
2250          * after the promotion (for both origin and clone).  For each,
2251          * it is the amount of space that will be on all of their
2252          * deadlists (that was not born before their new origin).
2253          */
2254         if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2255                 uint64_t space;
2256
2257                 /*
2258                  * Note, typically this will not be a clone of a clone,
2259                  * so dd_origin_txg will be < TXG_INITIAL, so
2260                  * these snaplist_space() -> dsl_deadlist_space_range()
2261                  * calls will be fast because they do not have to
2262                  * iterate over all bps.
2263                  */
2264                 snap = list_head(&ddpa->origin_snaps);
2265                 err = snaplist_space(&ddpa->shared_snaps,
2266                     snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2267                 if (err != 0)
2268                         goto out;
2269
2270                 err = snaplist_space(&ddpa->clone_snaps,
2271                     snap->ds->ds_dir->dd_origin_txg, &space);
2272                 if (err != 0)
2273                         goto out;
2274                 ddpa->cloneusedsnap += space;
2275         }
2276         if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2277                 err = snaplist_space(&ddpa->origin_snaps,
2278                     origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap);
2279                 if (err != 0)
2280                         goto out;
2281         }
2282
2283 out:
2284         promote_rele(ddpa, FTAG);
2285         return (err);
2286 }
2287
2288 static void
2289 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2290 {
2291         dsl_dataset_promote_arg_t *ddpa = arg;
2292         dsl_pool_t *dp = dmu_tx_pool(tx);
2293         dsl_dataset_t *hds;
2294         struct promotenode *snap;
2295         dsl_dataset_t *origin_ds;
2296         dsl_dataset_t *origin_head;
2297         dsl_dir_t *dd;
2298         dsl_dir_t *odd = NULL;
2299         uint64_t oldnext_obj;
2300         int64_t delta;
2301 #if defined(__FreeBSD__) && defined(_KERNEL)
2302         char *oldname, *newname;
2303 #endif
2304
2305         VERIFY0(promote_hold(ddpa, dp, FTAG));
2306         hds = ddpa->ddpa_clone;
2307
2308         ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE);
2309
2310         snap = list_head(&ddpa->shared_snaps);
2311         origin_ds = snap->ds;
2312         dd = hds->ds_dir;
2313
2314         snap = list_head(&ddpa->origin_snaps);
2315         origin_head = snap->ds;
2316
2317         /*
2318          * We need to explicitly open odd, since origin_ds's dd will be
2319          * changing.
2320          */
2321         VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2322             NULL, FTAG, &odd));
2323
2324         /* change origin's next snap */
2325         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2326         oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2327         snap = list_tail(&ddpa->clone_snaps);
2328         ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2329         origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2330
2331         /* change the origin's next clone */
2332         if (origin_ds->ds_phys->ds_next_clones_obj) {
2333                 dsl_dataset_remove_from_next_clones(origin_ds,
2334                     snap->ds->ds_object, tx);
2335                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2336                     origin_ds->ds_phys->ds_next_clones_obj,
2337                     oldnext_obj, tx));
2338         }
2339
2340         /* change origin */
2341         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2342         ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2343         dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2344         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2345         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2346         odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2347         origin_head->ds_dir->dd_origin_txg =
2348             origin_ds->ds_phys->ds_creation_txg;
2349
2350         /* change dd_clone entries */
2351         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2352                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2353                     odd->dd_phys->dd_clones, hds->ds_object, tx));
2354                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2355                     ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2356                     hds->ds_object, tx));
2357
2358                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2359                     ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2360                     origin_head->ds_object, tx));
2361                 if (dd->dd_phys->dd_clones == 0) {
2362                         dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2363                             DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2364                 }
2365                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2366                     dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2367         }
2368
2369 #if defined(__FreeBSD__) && defined(_KERNEL)
2370         /* Take the spa_namespace_lock early so zvol renames don't deadlock. */
2371         mutex_enter(&spa_namespace_lock);
2372
2373         oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2374         newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2375 #endif
2376
2377         /* move snapshots to this dir */
2378         for (snap = list_head(&ddpa->shared_snaps); snap;
2379             snap = list_next(&ddpa->shared_snaps, snap)) {
2380                 dsl_dataset_t *ds = snap->ds;
2381
2382                 /*
2383                  * Property callbacks are registered to a particular
2384                  * dsl_dir.  Since ours is changing, evict the objset
2385                  * so that they will be unregistered from the old dsl_dir.
2386                  */
2387                 if (ds->ds_objset) {
2388                         dmu_objset_evict(ds->ds_objset);
2389                         ds->ds_objset = NULL;
2390                 }
2391
2392                 /* move snap name entry */
2393                 VERIFY0(dsl_dataset_get_snapname(ds));
2394                 VERIFY0(dsl_dataset_snap_remove(origin_head,
2395                     ds->ds_snapname, tx, B_TRUE));
2396                 VERIFY0(zap_add(dp->dp_meta_objset,
2397                     hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2398                     8, 1, &ds->ds_object, tx));
2399                 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2400                     DD_FIELD_SNAPSHOT_COUNT, tx);
2401
2402                 /* change containing dsl_dir */
2403                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2404                 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2405                 ds->ds_phys->ds_dir_obj = dd->dd_object;
2406                 ASSERT3P(ds->ds_dir, ==, odd);
2407                 dsl_dir_rele(ds->ds_dir, ds);
2408                 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2409                     NULL, ds, &ds->ds_dir));
2410
2411 #if defined(__FreeBSD__) && defined(_KERNEL)
2412                 dsl_dataset_name(ds, newname);
2413                 zfsvfs_update_fromname(oldname, newname);
2414                 zvol_rename_minors(oldname, newname);
2415 #endif
2416
2417                 /* move any clone references */
2418                 if (ds->ds_phys->ds_next_clones_obj &&
2419                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2420                         zap_cursor_t zc;
2421                         zap_attribute_t za;
2422
2423                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
2424                             ds->ds_phys->ds_next_clones_obj);
2425                             zap_cursor_retrieve(&zc, &za) == 0;
2426                             zap_cursor_advance(&zc)) {
2427                                 dsl_dataset_t *cnds;
2428                                 uint64_t o;
2429
2430                                 if (za.za_first_integer == oldnext_obj) {
2431                                         /*
2432                                          * We've already moved the
2433                                          * origin's reference.
2434                                          */
2435                                         continue;
2436                                 }
2437
2438                                 VERIFY0(dsl_dataset_hold_obj(dp,
2439                                     za.za_first_integer, FTAG, &cnds));
2440                                 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2441
2442                                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2443                                     odd->dd_phys->dd_clones, o, tx));
2444                                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2445                                     dd->dd_phys->dd_clones, o, tx));
2446                                 dsl_dataset_rele(cnds, FTAG);
2447                         }
2448                         zap_cursor_fini(&zc);
2449                 }
2450
2451                 ASSERT(!dsl_prop_hascb(ds));
2452         }
2453
2454 #if defined(__FreeBSD__) && defined(_KERNEL)
2455         mutex_exit(&spa_namespace_lock);
2456
2457         kmem_free(newname, MAXPATHLEN);
2458         kmem_free(oldname, MAXPATHLEN);
2459 #endif
2460         /*
2461          * Change space accounting.
2462          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2463          * both be valid, or both be 0 (resulting in delta == 0).  This
2464          * is true for each of {clone,origin} independently.
2465          */
2466
2467         delta = ddpa->cloneusedsnap -
2468             dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2469         ASSERT3S(delta, >=, 0);
2470         ASSERT3U(ddpa->used, >=, delta);
2471         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2472         dsl_dir_diduse_space(dd, DD_USED_HEAD,
2473             ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2474
2475         delta = ddpa->originusedsnap -
2476             odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2477         ASSERT3S(delta, <=, 0);
2478         ASSERT3U(ddpa->used, >=, -delta);
2479         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2480         dsl_dir_diduse_space(odd, DD_USED_HEAD,
2481             -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2482
2483         origin_ds->ds_phys->ds_unique_bytes = ddpa->unique;
2484
2485         /* log history record */
2486         spa_history_log_internal_ds(hds, "promote", tx, "");
2487
2488         dsl_dir_rele(odd, FTAG);
2489         promote_rele(ddpa, FTAG);
2490 }
2491
2492 /*
2493  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2494  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2495  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2496  * snapshots back to this dataset's origin.
2497  */
2498 static int
2499 snaplist_make(dsl_pool_t *dp,
2500     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2501 {
2502         uint64_t obj = last_obj;
2503
2504         list_create(l, sizeof (struct promotenode),
2505             offsetof(struct promotenode, link));
2506
2507         while (obj != first_obj) {
2508                 dsl_dataset_t *ds;
2509                 struct promotenode *snap;
2510                 int err;
2511
2512                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2513                 ASSERT(err != ENOENT);
2514                 if (err != 0)
2515                         return (err);
2516
2517                 if (first_obj == 0)
2518                         first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2519
2520                 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2521                 snap->ds = ds;
2522                 list_insert_tail(l, snap);
2523                 obj = ds->ds_phys->ds_prev_snap_obj;
2524         }
2525
2526         return (0);
2527 }
2528
2529 static int
2530 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2531 {
2532         struct promotenode *snap;
2533
2534         *spacep = 0;
2535         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2536                 uint64_t used, comp, uncomp;
2537                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2538                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
2539                 *spacep += used;
2540         }
2541         return (0);
2542 }
2543
2544 static void
2545 snaplist_destroy(list_t *l, void *tag)
2546 {
2547         struct promotenode *snap;
2548
2549         if (l == NULL || !list_link_active(&l->list_head))
2550                 return;
2551
2552         while ((snap = list_tail(l)) != NULL) {
2553                 list_remove(l, snap);
2554                 dsl_dataset_rele(snap->ds, tag);
2555                 kmem_free(snap, sizeof (*snap));
2556         }
2557         list_destroy(l);
2558 }
2559
2560 static int
2561 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2562 {
2563         int error;
2564         dsl_dir_t *dd;
2565         struct promotenode *snap;
2566
2567         error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2568             &ddpa->ddpa_clone);
2569         if (error != 0)
2570                 return (error);
2571         dd = ddpa->ddpa_clone->ds_dir;
2572
2573         if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2574             !dsl_dir_is_clone(dd)) {
2575                 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2576                 return (SET_ERROR(EINVAL));
2577         }
2578
2579         error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj,
2580             &ddpa->shared_snaps, tag);
2581         if (error != 0)
2582                 goto out;
2583
2584         error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2585             &ddpa->clone_snaps, tag);
2586         if (error != 0)
2587                 goto out;
2588
2589         snap = list_head(&ddpa->shared_snaps);
2590         ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2591         error = snaplist_make(dp, dd->dd_phys->dd_origin_obj,
2592             snap->ds->ds_dir->dd_phys->dd_head_dataset_obj,
2593             &ddpa->origin_snaps, tag);
2594         if (error != 0)
2595                 goto out;
2596
2597         if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2598                 error = dsl_dataset_hold_obj(dp,
2599                     snap->ds->ds_dir->dd_phys->dd_origin_obj,
2600                     tag, &ddpa->origin_origin);
2601                 if (error != 0)
2602                         goto out;
2603         }
2604 out:
2605         if (error != 0)
2606                 promote_rele(ddpa, tag);
2607         return (error);
2608 }
2609
2610 static void
2611 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2612 {
2613         snaplist_destroy(&ddpa->shared_snaps, tag);
2614         snaplist_destroy(&ddpa->clone_snaps, tag);
2615         snaplist_destroy(&ddpa->origin_snaps, tag);
2616         if (ddpa->origin_origin != NULL)
2617                 dsl_dataset_rele(ddpa->origin_origin, tag);
2618         dsl_dataset_rele(ddpa->ddpa_clone, tag);
2619 }
2620
2621 /*
2622  * Promote a clone.
2623  *
2624  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2625  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2626  */
2627 int
2628 dsl_dataset_promote(const char *name, char *conflsnap)
2629 {
2630         dsl_dataset_promote_arg_t ddpa = { 0 };
2631         uint64_t numsnaps;
2632         int error;
2633         objset_t *os;
2634
2635         /*
2636          * We will modify space proportional to the number of
2637          * snapshots.  Compute numsnaps.
2638          */
2639         error = dmu_objset_hold(name, FTAG, &os);
2640         if (error != 0)
2641                 return (error);
2642         error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2643             dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps);
2644         dmu_objset_rele(os, FTAG);
2645         if (error != 0)
2646                 return (error);
2647
2648         ddpa.ddpa_clonename = name;
2649         ddpa.err_ds = conflsnap;
2650         ddpa.cr = CRED();
2651
2652         return (dsl_sync_task(name, dsl_dataset_promote_check,
2653             dsl_dataset_promote_sync, &ddpa,
2654             2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2655 }
2656
2657 int
2658 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2659     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2660 {
2661         int64_t unused_refres_delta;
2662
2663         /* they should both be heads */
2664         if (dsl_dataset_is_snapshot(clone) ||
2665             dsl_dataset_is_snapshot(origin_head))
2666                 return (SET_ERROR(EINVAL));
2667
2668         /* if we are not forcing, the branch point should be just before them */
2669         if (!force && clone->ds_prev != origin_head->ds_prev)
2670                 return (SET_ERROR(EINVAL));
2671
2672         /* clone should be the clone (unless they are unrelated) */
2673         if (clone->ds_prev != NULL &&
2674             clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2675             origin_head->ds_dir != clone->ds_prev->ds_dir)
2676                 return (SET_ERROR(EINVAL));
2677
2678         /* the clone should be a child of the origin */
2679         if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2680                 return (SET_ERROR(EINVAL));
2681
2682         /* origin_head shouldn't be modified unless 'force' */
2683         if (!force &&
2684             dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2685                 return (SET_ERROR(ETXTBSY));
2686
2687         /* origin_head should have no long holds (e.g. is not mounted) */
2688         if (dsl_dataset_handoff_check(origin_head, owner, tx))
2689                 return (SET_ERROR(EBUSY));
2690
2691         /* check amount of any unconsumed refreservation */
2692         unused_refres_delta =
2693             (int64_t)MIN(origin_head->ds_reserved,
2694             origin_head->ds_phys->ds_unique_bytes) -
2695             (int64_t)MIN(origin_head->ds_reserved,
2696             clone->ds_phys->ds_unique_bytes);
2697
2698         if (unused_refres_delta > 0 &&
2699             unused_refres_delta >
2700             dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2701                 return (SET_ERROR(ENOSPC));
2702
2703         /* clone can't be over the head's refquota */
2704         if (origin_head->ds_quota != 0 &&
2705             clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota)
2706                 return (SET_ERROR(EDQUOT));
2707
2708         return (0);
2709 }
2710
2711 void
2712 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2713     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2714 {
2715         dsl_pool_t *dp = dmu_tx_pool(tx);
2716         int64_t unused_refres_delta;
2717
2718         ASSERT(clone->ds_reserved == 0);
2719         ASSERT(origin_head->ds_quota == 0 ||
2720             clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota);
2721         ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2722
2723         dmu_buf_will_dirty(clone->ds_dbuf, tx);
2724         dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2725
2726         if (clone->ds_objset != NULL) {
2727                 dmu_objset_evict(clone->ds_objset);
2728                 clone->ds_objset = NULL;
2729         }
2730
2731         if (origin_head->ds_objset != NULL) {
2732                 dmu_objset_evict(origin_head->ds_objset);
2733                 origin_head->ds_objset = NULL;
2734         }
2735
2736         unused_refres_delta =
2737             (int64_t)MIN(origin_head->ds_reserved,
2738             origin_head->ds_phys->ds_unique_bytes) -
2739             (int64_t)MIN(origin_head->ds_reserved,
2740             clone->ds_phys->ds_unique_bytes);
2741
2742         /*
2743          * Reset origin's unique bytes, if it exists.
2744          */
2745         if (clone->ds_prev) {
2746                 dsl_dataset_t *origin = clone->ds_prev;
2747                 uint64_t comp, uncomp;
2748
2749                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2750                 dsl_deadlist_space_range(&clone->ds_deadlist,
2751                     origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2752                     &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
2753         }
2754
2755         /* swap blkptrs */
2756         {
2757                 blkptr_t tmp;
2758                 tmp = origin_head->ds_phys->ds_bp;
2759                 origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp;
2760                 clone->ds_phys->ds_bp = tmp;
2761         }
2762
2763         /* set dd_*_bytes */
2764         {
2765                 int64_t dused, dcomp, duncomp;
2766                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2767                 uint64_t odl_used, odl_comp, odl_uncomp;
2768
2769                 ASSERT3U(clone->ds_dir->dd_phys->
2770                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
2771
2772                 dsl_deadlist_space(&clone->ds_deadlist,
2773                     &cdl_used, &cdl_comp, &cdl_uncomp);
2774                 dsl_deadlist_space(&origin_head->ds_deadlist,
2775                     &odl_used, &odl_comp, &odl_uncomp);
2776
2777                 dused = clone->ds_phys->ds_referenced_bytes + cdl_used -
2778                     (origin_head->ds_phys->ds_referenced_bytes + odl_used);
2779                 dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp -
2780                     (origin_head->ds_phys->ds_compressed_bytes + odl_comp);
2781                 duncomp = clone->ds_phys->ds_uncompressed_bytes +
2782                     cdl_uncomp -
2783                     (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2784
2785                 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2786                     dused, dcomp, duncomp, tx);
2787                 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2788                     -dused, -dcomp, -duncomp, tx);
2789
2790                 /*
2791                  * The difference in the space used by snapshots is the
2792                  * difference in snapshot space due to the head's
2793                  * deadlist (since that's the only thing that's
2794                  * changing that affects the snapused).
2795                  */
2796                 dsl_deadlist_space_range(&clone->ds_deadlist,
2797                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2798                     &cdl_used, &cdl_comp, &cdl_uncomp);
2799                 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2800                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2801                     &odl_used, &odl_comp, &odl_uncomp);
2802                 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2803                     DD_USED_HEAD, DD_USED_SNAP, tx);
2804         }
2805
2806         /* swap ds_*_bytes */
2807         SWITCH64(origin_head->ds_phys->ds_referenced_bytes,
2808             clone->ds_phys->ds_referenced_bytes);
2809         SWITCH64(origin_head->ds_phys->ds_compressed_bytes,
2810             clone->ds_phys->ds_compressed_bytes);
2811         SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes,
2812             clone->ds_phys->ds_uncompressed_bytes);
2813         SWITCH64(origin_head->ds_phys->ds_unique_bytes,
2814             clone->ds_phys->ds_unique_bytes);
2815
2816         /* apply any parent delta for change in unconsumed refreservation */
2817         dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2818             unused_refres_delta, 0, 0, tx);
2819
2820         /*
2821          * Swap deadlists.
2822          */
2823         dsl_deadlist_close(&clone->ds_deadlist);
2824         dsl_deadlist_close(&origin_head->ds_deadlist);
2825         SWITCH64(origin_head->ds_phys->ds_deadlist_obj,
2826             clone->ds_phys->ds_deadlist_obj);
2827         dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2828             clone->ds_phys->ds_deadlist_obj);
2829         dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2830             origin_head->ds_phys->ds_deadlist_obj);
2831
2832         dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2833
2834         spa_history_log_internal_ds(clone, "clone swap", tx,
2835             "parent=%s", origin_head->ds_dir->dd_myname);
2836 }
2837
2838 /*
2839  * Given a pool name and a dataset object number in that pool,
2840  * return the name of that dataset.
2841  */
2842 int
2843 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2844 {
2845         dsl_pool_t *dp;
2846         dsl_dataset_t *ds;
2847         int error;
2848
2849         error = dsl_pool_hold(pname, FTAG, &dp);
2850         if (error != 0)
2851                 return (error);
2852
2853         error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2854         if (error == 0) {
2855                 dsl_dataset_name(ds, buf);
2856                 dsl_dataset_rele(ds, FTAG);
2857         }
2858         dsl_pool_rele(dp, FTAG);
2859
2860         return (error);
2861 }
2862
2863 int
2864 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2865     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2866 {
2867         int error = 0;
2868
2869         ASSERT3S(asize, >, 0);
2870
2871         /*
2872          * *ref_rsrv is the portion of asize that will come from any
2873          * unconsumed refreservation space.
2874          */
2875         *ref_rsrv = 0;
2876
2877         mutex_enter(&ds->ds_lock);
2878         /*
2879          * Make a space adjustment for reserved bytes.
2880          */
2881         if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2882                 ASSERT3U(*used, >=,
2883                     ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2884                 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2885                 *ref_rsrv =
2886                     asize - MIN(asize, parent_delta(ds, asize + inflight));
2887         }
2888
2889         if (!check_quota || ds->ds_quota == 0) {
2890                 mutex_exit(&ds->ds_lock);
2891                 return (0);
2892         }
2893         /*
2894          * If they are requesting more space, and our current estimate
2895          * is over quota, they get to try again unless the actual
2896          * on-disk is over quota and there are no pending changes (which
2897          * may free up space for us).
2898          */
2899         if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
2900                 if (inflight > 0 ||
2901                     ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
2902                         error = SET_ERROR(ERESTART);
2903                 else
2904                         error = SET_ERROR(EDQUOT);
2905         }
2906         mutex_exit(&ds->ds_lock);
2907
2908         return (error);
2909 }
2910
2911 typedef struct dsl_dataset_set_qr_arg {
2912         const char *ddsqra_name;
2913         zprop_source_t ddsqra_source;
2914         uint64_t ddsqra_value;
2915 } dsl_dataset_set_qr_arg_t;
2916
2917
2918 /* ARGSUSED */
2919 static int
2920 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2921 {
2922         dsl_dataset_set_qr_arg_t *ddsqra = arg;
2923         dsl_pool_t *dp = dmu_tx_pool(tx);
2924         dsl_dataset_t *ds;
2925         int error;
2926         uint64_t newval;
2927
2928         if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2929                 return (SET_ERROR(ENOTSUP));
2930
2931         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2932         if (error != 0)
2933                 return (error);
2934
2935         if (dsl_dataset_is_snapshot(ds)) {
2936                 dsl_dataset_rele(ds, FTAG);
2937                 return (SET_ERROR(EINVAL));
2938         }
2939
2940         error = dsl_prop_predict(ds->ds_dir,
2941             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2942             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2943         if (error != 0) {
2944                 dsl_dataset_rele(ds, FTAG);
2945                 return (error);
2946         }
2947
2948         if (newval == 0) {
2949                 dsl_dataset_rele(ds, FTAG);
2950                 return (0);
2951         }
2952
2953         if (newval < ds->ds_phys->ds_referenced_bytes ||
2954             newval < ds->ds_reserved) {
2955                 dsl_dataset_rele(ds, FTAG);
2956                 return (SET_ERROR(ENOSPC));
2957         }
2958
2959         dsl_dataset_rele(ds, FTAG);
2960         return (0);
2961 }
2962
2963 static void
2964 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2965 {
2966         dsl_dataset_set_qr_arg_t *ddsqra = arg;
2967         dsl_pool_t *dp = dmu_tx_pool(tx);
2968         dsl_dataset_t *ds;
2969         uint64_t newval;
2970
2971         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2972
2973         dsl_prop_set_sync_impl(ds,
2974             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2975             ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2976             &ddsqra->ddsqra_value, tx);
2977
2978         VERIFY0(dsl_prop_get_int_ds(ds,
2979             zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2980
2981         if (ds->ds_quota != newval) {
2982                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2983                 ds->ds_quota = newval;
2984         }
2985         dsl_dataset_rele(ds, FTAG);
2986 }
2987
2988 int
2989 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2990     uint64_t refquota)
2991 {
2992         dsl_dataset_set_qr_arg_t ddsqra;
2993
2994         ddsqra.ddsqra_name = dsname;
2995         ddsqra.ddsqra_source = source;
2996         ddsqra.ddsqra_value = refquota;
2997
2998         return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2999             dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3000 }
3001
3002 static int
3003 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3004 {
3005         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3006         dsl_pool_t *dp = dmu_tx_pool(tx);
3007         dsl_dataset_t *ds;
3008         int error;
3009         uint64_t newval, unique;
3010
3011         if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3012                 return (SET_ERROR(ENOTSUP));
3013
3014         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3015         if (error != 0)
3016                 return (error);
3017
3018         if (dsl_dataset_is_snapshot(ds)) {
3019                 dsl_dataset_rele(ds, FTAG);
3020                 return (SET_ERROR(EINVAL));
3021         }
3022
3023         error = dsl_prop_predict(ds->ds_dir,
3024             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3025             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3026         if (error != 0) {
3027                 dsl_dataset_rele(ds, FTAG);
3028                 return (error);
3029         }
3030
3031         /*
3032          * If we are doing the preliminary check in open context, the
3033          * space estimates may be inaccurate.
3034          */
3035         if (!dmu_tx_is_syncing(tx)) {
3036                 dsl_dataset_rele(ds, FTAG);
3037                 return (0);
3038         }
3039
3040         mutex_enter(&ds->ds_lock);
3041         if (!DS_UNIQUE_IS_ACCURATE(ds))
3042                 dsl_dataset_recalc_head_uniq(ds);
3043         unique = ds->ds_phys->ds_unique_bytes;
3044         mutex_exit(&ds->ds_lock);
3045
3046         if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3047                 uint64_t delta = MAX(unique, newval) -
3048                     MAX(unique, ds->ds_reserved);
3049
3050                 if (delta >
3051                     dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3052                     (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3053                         dsl_dataset_rele(ds, FTAG);
3054                         return (SET_ERROR(ENOSPC));
3055                 }
3056         }
3057
3058         dsl_dataset_rele(ds, FTAG);
3059         return (0);
3060 }
3061
3062 void
3063 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3064     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3065 {
3066         uint64_t newval;
3067         uint64_t unique;
3068         int64_t delta;
3069
3070         dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3071             source, sizeof (value), 1, &value, tx);
3072
3073         VERIFY0(dsl_prop_get_int_ds(ds,
3074             zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3075
3076         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3077         mutex_enter(&ds->ds_dir->dd_lock);
3078         mutex_enter(&ds->ds_lock);
3079         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3080         unique = ds->ds_phys->ds_unique_bytes;
3081         delta = MAX(0, (int64_t)(newval - unique)) -
3082             MAX(0, (int64_t)(ds->ds_reserved - unique));
3083         ds->ds_reserved = newval;
3084         mutex_exit(&ds->ds_lock);
3085
3086         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3087         mutex_exit(&ds->ds_dir->dd_lock);
3088 }
3089
3090 static void
3091 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3092 {
3093         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3094         dsl_pool_t *dp = dmu_tx_pool(tx);
3095         dsl_dataset_t *ds;
3096
3097         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3098         dsl_dataset_set_refreservation_sync_impl(ds,
3099             ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3100         dsl_dataset_rele(ds, FTAG);
3101 }
3102
3103 int
3104 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3105     uint64_t refreservation)
3106 {
3107         dsl_dataset_set_qr_arg_t ddsqra;
3108
3109         ddsqra.ddsqra_name = dsname;
3110         ddsqra.ddsqra_source = source;
3111         ddsqra.ddsqra_value = refreservation;
3112
3113         return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3114             dsl_dataset_set_refreservation_sync, &ddsqra,
3115             0, ZFS_SPACE_CHECK_NONE));
3116 }
3117
3118 /*
3119  * Return (in *usedp) the amount of space written in new that is not
3120  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3121  * a snapshot before new, in new's filesystem (or its origin).  If not then
3122  * fail and return EINVAL.
3123  *
3124  * The written space is calculated by considering two components:  First, we
3125  * ignore any freed space, and calculate the written as new's used space
3126  * minus old's used space.  Next, we add in the amount of space that was freed
3127  * between the two snapshots, thus reducing new's used space relative to old's.
3128  * Specifically, this is the space that was born before old->ds_creation_txg,
3129  * and freed before new (ie. on new's deadlist or a previous deadlist).
3130  *
3131  * space freed                         [---------------------]
3132  * snapshots                       ---O-------O--------O-------O------
3133  *                                         oldsnap            new
3134  */
3135 int
3136 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3137     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3138 {
3139         int err = 0;
3140         uint64_t snapobj;
3141         dsl_pool_t *dp = new->ds_dir->dd_pool;
3142
3143         ASSERT(dsl_pool_config_held(dp));
3144
3145         *usedp = 0;
3146         *usedp += new->ds_phys->ds_referenced_bytes;
3147         *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
3148
3149         *compp = 0;
3150         *compp += new->ds_phys->ds_compressed_bytes;
3151         *compp -= oldsnap->ds_phys->ds_compressed_bytes;
3152
3153         *uncompp = 0;
3154         *uncompp += new->ds_phys->ds_uncompressed_bytes;
3155         *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
3156
3157         snapobj = new->ds_object;
3158         while (snapobj != oldsnap->ds_object) {
3159                 dsl_dataset_t *snap;
3160                 uint64_t used, comp, uncomp;
3161
3162                 if (snapobj == new->ds_object) {
3163                         snap = new;
3164                 } else {
3165                         err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3166                         if (err != 0)
3167                                 break;
3168                 }
3169
3170                 if (snap->ds_phys->ds_prev_snap_txg ==
3171                     oldsnap->ds_phys->ds_creation_txg) {
3172                         /*
3173                          * The blocks in the deadlist can not be born after
3174                          * ds_prev_snap_txg, so get the whole deadlist space,
3175                          * which is more efficient (especially for old-format
3176                          * deadlists).  Unfortunately the deadlist code
3177                          * doesn't have enough information to make this
3178                          * optimization itself.
3179                          */
3180                         dsl_deadlist_space(&snap->ds_deadlist,
3181                             &used, &comp, &uncomp);
3182                 } else {
3183                         dsl_deadlist_space_range(&snap->ds_deadlist,
3184                             0, oldsnap->ds_phys->ds_creation_txg,
3185                             &used, &comp, &uncomp);
3186                 }
3187                 *usedp += used;
3188                 *compp += comp;
3189                 *uncompp += uncomp;
3190
3191                 /*
3192                  * If we get to the beginning of the chain of snapshots
3193                  * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3194                  * was not a snapshot of/before new.
3195                  */
3196                 snapobj = snap->ds_phys->ds_prev_snap_obj;
3197                 if (snap != new)
3198                         dsl_dataset_rele(snap, FTAG);
3199                 if (snapobj == 0) {
3200                         err = SET_ERROR(EINVAL);
3201                         break;
3202                 }
3203
3204         }
3205         return (err);
3206 }
3207
3208 /*
3209  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3210  * lastsnap, and all snapshots in between are deleted.
3211  *
3212  * blocks that would be freed            [---------------------------]
3213  * snapshots                       ---O-------O--------O-------O--------O
3214  *                                        firstsnap        lastsnap
3215  *
3216  * This is the set of blocks that were born after the snap before firstsnap,
3217  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3218  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3219  * We calculate this by iterating over the relevant deadlists (from the snap
3220  * after lastsnap, backward to the snap after firstsnap), summing up the
3221  * space on the deadlist that was born after the snap before firstsnap.
3222  */
3223 int
3224 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3225     dsl_dataset_t *lastsnap,
3226     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3227 {
3228         int err = 0;
3229         uint64_t snapobj;
3230         dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3231
3232         ASSERT(dsl_dataset_is_snapshot(firstsnap));
3233         ASSERT(dsl_dataset_is_snapshot(lastsnap));
3234
3235         /*
3236          * Check that the snapshots are in the same dsl_dir, and firstsnap
3237          * is before lastsnap.
3238          */
3239         if (firstsnap->ds_dir != lastsnap->ds_dir ||
3240             firstsnap->ds_phys->ds_creation_txg >
3241             lastsnap->ds_phys->ds_creation_txg)
3242                 return (SET_ERROR(EINVAL));
3243
3244         *usedp = *compp = *uncompp = 0;
3245
3246         snapobj = lastsnap->ds_phys->ds_next_snap_obj;
3247         while (snapobj != firstsnap->ds_object) {
3248                 dsl_dataset_t *ds;
3249                 uint64_t used, comp, uncomp;
3250
3251                 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3252                 if (err != 0)
3253                         break;
3254
3255                 dsl_deadlist_space_range(&ds->ds_deadlist,
3256                     firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3257                     &used, &comp, &uncomp);
3258                 *usedp += used;
3259                 *compp += comp;
3260                 *uncompp += uncomp;
3261
3262                 snapobj = ds->ds_phys->ds_prev_snap_obj;
3263                 ASSERT3U(snapobj, !=, 0);
3264                 dsl_dataset_rele(ds, FTAG);
3265         }
3266         return (err);
3267 }
3268
3269 static int
3270 dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx)
3271 {
3272         const char *dsname = arg;
3273         dsl_dataset_t *ds;
3274         dsl_pool_t *dp = dmu_tx_pool(tx);
3275         int error = 0;
3276
3277         if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
3278                 return (SET_ERROR(ENOTSUP));
3279
3280         ASSERT(spa_feature_is_enabled(dp->dp_spa,
3281             SPA_FEATURE_EXTENSIBLE_DATASET));
3282
3283         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
3284         if (error != 0)
3285                 return (error);
3286
3287         if (ds->ds_large_blocks)
3288                 error = EALREADY;
3289         dsl_dataset_rele(ds, FTAG);
3290
3291         return (error);
3292 }
3293
3294 void
3295 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx)
3296 {
3297         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
3298         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
3299         uint64_t zero = 0;
3300
3301         spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx);
3302         dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
3303
3304         VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS,
3305             sizeof (zero), 1, &zero, tx));
3306 }
3307
3308 static void
3309 dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx)
3310 {
3311         const char *dsname = arg;
3312         dsl_dataset_t *ds;
3313
3314         VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds));
3315
3316         dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
3317         ASSERT(!ds->ds_large_blocks);
3318         ds->ds_large_blocks = B_TRUE;
3319         dsl_dataset_rele(ds, FTAG);
3320 }
3321
3322 int
3323 dsl_dataset_activate_large_blocks(const char *dsname)
3324 {
3325         int error;
3326
3327         error = dsl_sync_task(dsname,
3328             dsl_dataset_activate_large_blocks_check,
3329             dsl_dataset_activate_large_blocks_sync, (void *)dsname,
3330             1, ZFS_SPACE_CHECK_RESERVED);
3331
3332         /*
3333          * EALREADY indicates that this dataset already supports large blocks.
3334          */
3335         if (error == EALREADY)
3336                 error = 0;
3337         return (error);
3338 }
3339
3340 /*
3341  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3342  * For example, they could both be snapshots of the same filesystem, and
3343  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3344  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3345  * filesystem.  Or 'earlier' could be the origin's origin.
3346  *
3347  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3348  */
3349 boolean_t
3350 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3351         uint64_t earlier_txg)
3352 {
3353         dsl_pool_t *dp = later->ds_dir->dd_pool;
3354         int error;
3355         boolean_t ret;
3356
3357         ASSERT(dsl_pool_config_held(dp));
3358         ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
3359
3360         if (earlier_txg == 0)
3361                 earlier_txg = earlier->ds_phys->ds_creation_txg;
3362
3363         if (dsl_dataset_is_snapshot(later) &&
3364             earlier_txg >= later->ds_phys->ds_creation_txg)
3365                 return (B_FALSE);
3366
3367         if (later->ds_dir == earlier->ds_dir)
3368                 return (B_TRUE);
3369         if (!dsl_dir_is_clone(later->ds_dir))
3370                 return (B_FALSE);
3371
3372         if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object)
3373                 return (B_TRUE);
3374         dsl_dataset_t *origin;
3375         error = dsl_dataset_hold_obj(dp,
3376             later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
3377         if (error != 0)
3378                 return (B_FALSE);
3379         ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3380         dsl_dataset_rele(origin, FTAG);
3381         return (ret);
3382 }
3383
3384
3385 void
3386 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3387 {
3388         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3389         dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3390 }