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, 2015 by Delphix. All rights reserved.
  25  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2014 RackTop Systems.
  27  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  28  * Copyright (c) 2014 Integros [integros.com]
  29  * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
  30  */
  31
  32 #include <sys/dmu_objset.h>
  33 #include <sys/dsl_dataset.h>
  34 #include <sys/dsl_dir.h>
  35 #include <sys/dsl_prop.h>
  36 #include <sys/dsl_synctask.h>
  37 #include <sys/dmu_traverse.h>
  38 #include <sys/dmu_impl.h>
  39 #include <sys/dmu_send.h>
  40 #include <sys/dmu_tx.h>
  41 #include <sys/arc.h>
  42 #include <sys/zio.h>
  43 #include <sys/zap.h>
  44 #include <sys/zfeature.h>
  45 #include <sys/unique.h>
  46 #include <sys/zfs_context.h>
  47 #include <sys/zfs_ioctl.h>
  48 #include <sys/spa.h>
  49 #include <sys/zfs_znode.h>
  50 #include <sys/zfs_onexit.h>
  51 #include <sys/zvol.h>
  52 #include <sys/dsl_scan.h>
  53 #include <sys/dsl_deadlist.h>
  54 #include <sys/dsl_destroy.h>
  55 #include <sys/dsl_userhold.h>
  56 #include <sys/dsl_bookmark.h>
  57 #include <sys/dmu_send.h>
  58 #include <sys/zio_checksum.h>
  59 #include <sys/zio_compress.h>
  60 #include <zfs_fletcher.h>
  61
  62 SYSCTL_DECL(_vfs_zfs);
  63
  64 /*
  65  * The SPA supports block sizes up to 16MB.  However, very large blocks
  66  * can have an impact on i/o latency (e.g. tying up a spinning disk for
  67  * ~300ms), and also potentially on the memory allocator.  Therefore,
  68  * we do not allow the recordsize to be set larger than zfs_max_recordsize
  69  * (default 1MB).  Larger blocks can be created by changing this tunable,
  70  * and pools with larger blocks can always be imported and used, regardless
  71  * of this setting.
  72  */
  73 int zfs_max_recordsize = 1 * 1024 * 1024;
  74 SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN,
  75     &zfs_max_recordsize, 0,
  76     "Maximum block size.  Expect dragons when tuning this.");
  77
  78 #define SWITCH64(x, y) \
  79         { \
  80                 uint64_t __tmp = (x); \
  81                 (x) = (y); \
  82                 (y) = __tmp; \
  83         }
  84
  85 #define DS_REF_MAX      (1ULL << 62)
  86
  87 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
  88
  89 extern int spa_asize_inflation;
  90
  91 /*
  92  * Figure out how much of this delta should be propogated to the dsl_dir
  93  * layer.  If there's a refreservation, that space has already been
  94  * partially accounted for in our ancestors.
  95  */
  96 static int64_t
  97 parent_delta(dsl_dataset_t *ds, int64_t delta)
  98 {
  99         dsl_dataset_phys_t *ds_phys;
 100         uint64_t old_bytes, new_bytes;
 101
 102         if (ds->ds_reserved == 0)
 103                 return (delta);
 104
 105         ds_phys = dsl_dataset_phys(ds);
 106         old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
 107         new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
 108
 109         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
 110         return (new_bytes - old_bytes);
 111 }
 112
 113 void
 114 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
 115 {
 116         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 117         int compressed = BP_GET_PSIZE(bp);
 118         int uncompressed = BP_GET_UCSIZE(bp);
 119         int64_t delta;
 120
 121         dprintf_bp(bp, "ds=%p", ds);
 122
 123         ASSERT(dmu_tx_is_syncing(tx));
 124         /* It could have been compressed away to nothing */
 125         if (BP_IS_HOLE(bp))
 126                 return;
 127         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 128         ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 129         if (ds == NULL) {
 130                 dsl_pool_mos_diduse_space(tx->tx_pool,
 131                     used, compressed, uncompressed);
 132                 return;
 133         }
 134
 135         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 136         mutex_enter(&ds->ds_lock);
 137         delta = parent_delta(ds, used);
 138         dsl_dataset_phys(ds)->ds_referenced_bytes += used;
 139         dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
 140         dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
 141         dsl_dataset_phys(ds)->ds_unique_bytes += used;
 142
 143         if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
 144                 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
 145                     B_TRUE;
 146         }
 147
 148         spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
 149         if (f != SPA_FEATURE_NONE)
 150                 ds->ds_feature_activation_needed[f] = B_TRUE;
 151
 152         mutex_exit(&ds->ds_lock);
 153         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 154             compressed, uncompressed, tx);
 155         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 156             DD_USED_REFRSRV, DD_USED_HEAD, NULL);
 157 }
 158
 159 int
 160 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 161     boolean_t async)
 162 {
 163         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 164         int compressed = BP_GET_PSIZE(bp);
 165         int uncompressed = BP_GET_UCSIZE(bp);
 166
 167         if (BP_IS_HOLE(bp))
 168                 return (0);
 169
 170         ASSERT(dmu_tx_is_syncing(tx));
 171         ASSERT(bp->blk_birth <= tx->tx_txg);
 172
 173         if (ds == NULL) {
 174                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 175                 dsl_pool_mos_diduse_space(tx->tx_pool,
 176                     -used, -compressed, -uncompressed);
 177                 return (used);
 178         }
 179         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 180
 181         ASSERT(!ds->ds_is_snapshot);
 182         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 183
 184         if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
 185                 int64_t delta;
 186
 187                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 188                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 189
 190                 mutex_enter(&ds->ds_lock);
 191                 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
 192                     !DS_UNIQUE_IS_ACCURATE(ds));
 193                 delta = parent_delta(ds, -used);
 194                 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
 195                 mutex_exit(&ds->ds_lock);
 196                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 197                     delta, -compressed, -uncompressed, tx);
 198                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 199                     DD_USED_REFRSRV, DD_USED_HEAD, NULL);
 200         } else {
 201                 dprintf_bp(bp, "putting on dead list: %s", "");
 202                 if (async) {
 203                         /*
 204                          * We are here as part of zio's write done callback,
 205                          * which means we're a zio interrupt thread.  We can't
 206                          * call dsl_deadlist_insert() now because it may block
 207                          * waiting for I/O.  Instead, put bp on the deferred
 208                          * queue and let dsl_pool_sync() finish the job.
 209                          */
 210                         bplist_append(&ds->ds_pending_deadlist, bp);
 211                 } else {
 212                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 213                 }
 214                 ASSERT3U(ds->ds_prev->ds_object, ==,
 215                     dsl_dataset_phys(ds)->ds_prev_snap_obj);
 216                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
 217                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 218                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
 219                     ds->ds_object && bp->blk_birth >
 220                     dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
 221                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 222                         mutex_enter(&ds->ds_prev->ds_lock);
 223                         dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
 224                         mutex_exit(&ds->ds_prev->ds_lock);
 225                 }
 226                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 227                         dsl_dir_transfer_space(ds->ds_dir, used,
 228                             DD_USED_HEAD, DD_USED_SNAP, tx);
 229                 }
 230         }
 231         mutex_enter(&ds->ds_lock);
 232         ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
 233         dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
 234         ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
 235         dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
 236         ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
 237         dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
 238         mutex_exit(&ds->ds_lock);
 239
 240         return (used);
 241 }
 242
 243 uint64_t
 244 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
 245 {
 246         uint64_t trysnap = 0;
 247
 248         if (ds == NULL)
 249                 return (0);
 250         /*
 251          * The snapshot creation could fail, but that would cause an
 252          * incorrect FALSE return, which would only result in an
 253          * overestimation of the amount of space that an operation would
 254          * consume, which is OK.
 255          *
 256          * There's also a small window where we could miss a pending
 257          * snapshot, because we could set the sync task in the quiescing
 258          * phase.  So this should only be used as a guess.
 259          */
 260         if (ds->ds_trysnap_txg >
 261             spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
 262                 trysnap = ds->ds_trysnap_txg;
 263         return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
 264 }
 265
 266 boolean_t
 267 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
 268     uint64_t blk_birth)
 269 {
 270         if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
 271             (bp != NULL && BP_IS_HOLE(bp)))
 272                 return (B_FALSE);
 273
 274         ddt_prefetch(dsl_dataset_get_spa(ds), bp);
 275
 276         return (B_TRUE);
 277 }
 278
 279 static void
 280 dsl_dataset_evict(void *dbu)
 281 {
 282         dsl_dataset_t *ds = dbu;
 283
 284         ASSERT(ds->ds_owner == NULL);
 285
 286         ds->ds_dbuf = NULL;
 287
 288         unique_remove(ds->ds_fsid_guid);
 289
 290         if (ds->ds_objset != NULL)
 291                 dmu_objset_evict(ds->ds_objset);
 292
 293         if (ds->ds_prev) {
 294                 dsl_dataset_rele(ds->ds_prev, ds);
 295                 ds->ds_prev = NULL;
 296         }
 297
 298         bplist_destroy(&ds->ds_pending_deadlist);
 299         if (ds->ds_deadlist.dl_os != NULL)
 300                 dsl_deadlist_close(&ds->ds_deadlist);
 301         if (ds->ds_dir)
 302                 dsl_dir_async_rele(ds->ds_dir, ds);
 303
 304         ASSERT(!list_link_active(&ds->ds_synced_link));
 305
 306         list_destroy(&ds->ds_prop_cbs);
 307         if (mutex_owned(&ds->ds_lock))
 308                 mutex_exit(&ds->ds_lock);
 309         mutex_destroy(&ds->ds_lock);
 310         if (mutex_owned(&ds->ds_opening_lock))
 311                 mutex_exit(&ds->ds_opening_lock);
 312         mutex_destroy(&ds->ds_opening_lock);
 313         mutex_destroy(&ds->ds_sendstream_lock);
 314         refcount_destroy(&ds->ds_longholds);
 315         rrw_destroy(&ds->ds_bp_rwlock);
 316
 317         kmem_free(ds, sizeof (dsl_dataset_t));
 318 }
 319
 320 int
 321 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 322 {
 323         dsl_dataset_phys_t *headphys;
 324         int err;
 325         dmu_buf_t *headdbuf;
 326         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 327         objset_t *mos = dp->dp_meta_objset;
 328
 329         if (ds->ds_snapname[0])
 330                 return (0);
 331         if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
 332                 return (0);
 333
 334         err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
 335             FTAG, &headdbuf);
 336         if (err != 0)
 337                 return (err);
 338         headphys = headdbuf->db_data;
 339         err = zap_value_search(dp->dp_meta_objset,
 340             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 341         dmu_buf_rele(headdbuf, FTAG);
 342         return (err);
 343 }
 344
 345 int
 346 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 347 {
 348         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 349         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 350         matchtype_t mt;
 351         int err;
 352
 353         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 354                 mt = MT_FIRST;
 355         else
 356                 mt = MT_EXACT;
 357
 358         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 359             value, mt, NULL, 0, NULL);
 360         if (err == ENOTSUP && mt == MT_FIRST)
 361                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 362         return (err);
 363 }
 364
 365 int
 366 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
 367     boolean_t adj_cnt)
 368 {
 369         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 370         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 371         matchtype_t mt;
 372         int err;
 373
 374         dsl_dir_snap_cmtime_update(ds->ds_dir);
 375
 376         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 377                 mt = MT_FIRST;
 378         else
 379                 mt = MT_EXACT;
 380
 381         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 382         if (err == ENOTSUP && mt == MT_FIRST)
 383                 err = zap_remove(mos, snapobj, name, tx);
 384
 385         if (err == 0 && adj_cnt)
 386                 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
 387                     DD_FIELD_SNAPSHOT_COUNT, tx);
 388
 389         return (err);
 390 }
 391
 392 boolean_t
 393 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
 394 {
 395         dmu_buf_t *dbuf = ds->ds_dbuf;
 396         boolean_t result = B_FALSE;
 397
 398         if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
 399             ds->ds_object, DMU_BONUS_BLKID, tag)) {
 400
 401                 if (ds == dmu_buf_get_user(dbuf))
 402                         result = B_TRUE;
 403                 else
 404                         dmu_buf_rele(dbuf, tag);
 405         }
 406
 407         return (result);
 408 }
 409
 410 int
 411 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 412     dsl_dataset_t **dsp)
 413 {
 414         objset_t *mos = dp->dp_meta_objset;
 415         dmu_buf_t *dbuf;
 416         dsl_dataset_t *ds;
 417         int err;
 418         dmu_object_info_t doi;
 419
 420         ASSERT(dsl_pool_config_held(dp));
 421
 422         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 423         if (err != 0)
 424                 return (err);
 425
 426         /* Make sure dsobj has the correct object type. */
 427         dmu_object_info_from_db(dbuf, &doi);
 428         if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
 429                 dmu_buf_rele(dbuf, tag);
 430                 return (SET_ERROR(EINVAL));
 431         }
 432
 433         ds = dmu_buf_get_user(dbuf);
 434         if (ds == NULL) {
 435                 dsl_dataset_t *winner = NULL;
 436
 437                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 438                 ds->ds_dbuf = dbuf;
 439                 ds->ds_object = dsobj;
 440                 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
 441
 442                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 443                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 444                 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
 445                 rrw_init(&ds->ds_bp_rwlock, B_FALSE);
 446                 refcount_create(&ds->ds_longholds);
 447
 448                 bplist_create(&ds->ds_pending_deadlist);
 449                 dsl_deadlist_open(&ds->ds_deadlist,
 450                     mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
 451
 452                 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
 453                     offsetof(dmu_sendarg_t, dsa_link));
 454
 455                 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
 456                     offsetof(dsl_prop_cb_record_t, cbr_ds_node));
 457
 458                 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 459                         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 460                                 if (!(spa_feature_table[f].fi_flags &
 461                                     ZFEATURE_FLAG_PER_DATASET))
 462                                         continue;
 463                                 err = zap_contains(mos, dsobj,
 464                                     spa_feature_table[f].fi_guid);
 465                                 if (err == 0) {
 466                                         ds->ds_feature_inuse[f] = B_TRUE;
 467                                 } else {
 468                                         ASSERT3U(err, ==, ENOENT);
 469                                         err = 0;
 470                                 }
 471                         }
 472                 }
 473
 474                 err = dsl_dir_hold_obj(dp,
 475                     dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
 476                 if (err != 0) {
 477                         mutex_destroy(&ds->ds_lock);
 478                         mutex_destroy(&ds->ds_opening_lock);
 479                         mutex_destroy(&ds->ds_sendstream_lock);
 480                         refcount_destroy(&ds->ds_longholds);
 481                         bplist_destroy(&ds->ds_pending_deadlist);
 482                         dsl_deadlist_close(&ds->ds_deadlist);
 483                         kmem_free(ds, sizeof (dsl_dataset_t));
 484                         dmu_buf_rele(dbuf, tag);
 485                         return (err);
 486                 }
 487
 488                 if (!ds->ds_is_snapshot) {
 489                         ds->ds_snapname[0] = '\0';
 490                         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 491                                 err = dsl_dataset_hold_obj(dp,
 492                                     dsl_dataset_phys(ds)->ds_prev_snap_obj,
 493                                     ds, &ds->ds_prev);
 494                         }
 495                         if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 496                                 int zaperr = zap_lookup(mos, ds->ds_object,
 497                                     DS_FIELD_BOOKMARK_NAMES,
 498                                     sizeof (ds->ds_bookmarks), 1,
 499                                     &ds->ds_bookmarks);
 500                                 if (zaperr != ENOENT)
 501                                         VERIFY0(zaperr);
 502                         }
 503                 } else {
 504                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 505                                 err = dsl_dataset_get_snapname(ds);
 506                         if (err == 0 &&
 507                             dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
 508                                 err = zap_count(
 509                                     ds->ds_dir->dd_pool->dp_meta_objset,
 510                                     dsl_dataset_phys(ds)->ds_userrefs_obj,
 511                                     &ds->ds_userrefs);
 512                         }
 513                 }
 514
 515                 if (err == 0 && !ds->ds_is_snapshot) {
 516                         err = dsl_prop_get_int_ds(ds,
 517                             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
 518                             &ds->ds_reserved);
 519                         if (err == 0) {
 520                                 err = dsl_prop_get_int_ds(ds,
 521                                     zfs_prop_to_name(ZFS_PROP_REFQUOTA),
 522                                     &ds->ds_quota);
 523                         }
 524                 } else {
 525                         ds->ds_reserved = ds->ds_quota = 0;
 526                 }
 527
 528                 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
 529                 if (err == 0)
 530                         winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
 531
 532                 if (err != 0 || winner != NULL) {
 533                         bplist_destroy(&ds->ds_pending_deadlist);
 534                         dsl_deadlist_close(&ds->ds_deadlist);
 535                         if (ds->ds_prev)
 536                                 dsl_dataset_rele(ds->ds_prev, ds);
 537                         dsl_dir_rele(ds->ds_dir, ds);
 538                         mutex_destroy(&ds->ds_lock);
 539                         mutex_destroy(&ds->ds_opening_lock);
 540                         mutex_destroy(&ds->ds_sendstream_lock);
 541                         refcount_destroy(&ds->ds_longholds);
 542                         kmem_free(ds, sizeof (dsl_dataset_t));
 543                         if (err != 0) {
 544                                 dmu_buf_rele(dbuf, tag);
 545                                 return (err);
 546                         }
 547                         ds = winner;
 548                 } else {
 549                         ds->ds_fsid_guid =
 550                             unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
 551                 }
 552         }
 553         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 554         ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
 555         ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
 556             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 557             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 558         *dsp = ds;
 559         return (0);
 560 }
 561
 562 int
 563 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
 564     void *tag, dsl_dataset_t **dsp)
 565 {
 566         dsl_dir_t *dd;
 567         const char *snapname;
 568         uint64_t obj;
 569         int err = 0;
 570         dsl_dataset_t *ds;
 571
 572         err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
 573         if (err != 0)
 574                 return (err);
 575
 576         ASSERT(dsl_pool_config_held(dp));
 577         obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
 578         if (obj != 0)
 579                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
 580         else
 581                 err = SET_ERROR(ENOENT);
 582
 583         /* we may be looking for a snapshot */
 584         if (err == 0 && snapname != NULL) {
 585                 dsl_dataset_t *snap_ds;
 586
 587                 if (*snapname++ != '@') {
 588                         dsl_dataset_rele(ds, tag);
 589                         dsl_dir_rele(dd, FTAG);
 590                         return (SET_ERROR(ENOENT));
 591                 }
 592
 593                 dprintf("looking for snapshot '%s'\n", snapname);
 594                 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
 595                 if (err == 0)
 596                         err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
 597                 dsl_dataset_rele(ds, tag);
 598
 599                 if (err == 0) {
 600                         mutex_enter(&snap_ds->ds_lock);
 601                         if (snap_ds->ds_snapname[0] == 0)
 602                                 (void) strlcpy(snap_ds->ds_snapname, snapname,
 603                                     sizeof (snap_ds->ds_snapname));
 604                         mutex_exit(&snap_ds->ds_lock);
 605                         ds = snap_ds;
 606                 }
 607         }
 608         if (err == 0)
 609                 *dsp = ds;
 610         dsl_dir_rele(dd, FTAG);
 611         return (err);
 612 }
 613
 614 int
 615 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
 616     void *tag, dsl_dataset_t **dsp)
 617 {
 618         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 619         if (err != 0)
 620                 return (err);
 621         if (!dsl_dataset_tryown(*dsp, tag)) {
 622                 dsl_dataset_rele(*dsp, tag);
 623                 *dsp = NULL;
 624                 return (SET_ERROR(EBUSY));
 625         }
 626         return (0);
 627 }
 628
 629 int
 630 dsl_dataset_own(dsl_pool_t *dp, const char *name,
 631     void *tag, dsl_dataset_t **dsp)
 632 {
 633         int err = dsl_dataset_hold(dp, name, tag, dsp);
 634         if (err != 0)
 635                 return (err);
 636         if (!dsl_dataset_tryown(*dsp, tag)) {
 637                 dsl_dataset_rele(*dsp, tag);
 638                 return (SET_ERROR(EBUSY));
 639         }
 640         return (0);
 641 }
 642
 643 /*
 644  * See the comment above dsl_pool_hold() for details.  In summary, a long
 645  * hold is used to prevent destruction of a dataset while the pool hold
 646  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
 647  *
 648  * The dataset and pool must be held when this function is called.  After it
 649  * is called, the pool hold may be released while the dataset is still held
 650  * and accessed.
 651  */
 652 void
 653 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
 654 {
 655         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 656         (void) refcount_add(&ds->ds_longholds, tag);
 657 }
 658
 659 void
 660 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
 661 {
 662         (void) refcount_remove(&ds->ds_longholds, tag);
 663 }
 664
 665 /* Return B_TRUE if there are any long holds on this dataset. */
 666 boolean_t
 667 dsl_dataset_long_held(dsl_dataset_t *ds)
 668 {
 669         return (!refcount_is_zero(&ds->ds_longholds));
 670 }
 671
 672 void
 673 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 674 {
 675         if (ds == NULL) {
 676                 (void) strcpy(name, "mos");
 677         } else {
 678                 dsl_dir_name(ds->ds_dir, name);
 679                 VERIFY0(dsl_dataset_get_snapname(ds));
 680                 if (ds->ds_snapname[0]) {
 681                         VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
 682                             <, ZFS_MAX_DATASET_NAME_LEN);
 683                         /*
 684                          * We use a "recursive" mutex so that we
 685                          * can call dprintf_ds() with ds_lock held.
 686                          */
 687                         if (!MUTEX_HELD(&ds->ds_lock)) {
 688                                 mutex_enter(&ds->ds_lock);
 689                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 690                                     ZFS_MAX_DATASET_NAME_LEN), <,
 691                                     ZFS_MAX_DATASET_NAME_LEN);
 692                                 mutex_exit(&ds->ds_lock);
 693                         } else {
 694                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 695                                     ZFS_MAX_DATASET_NAME_LEN), <,
 696                                     ZFS_MAX_DATASET_NAME_LEN);
 697                         }
 698                 }
 699         }
 700 }
 701
 702 int
 703 dsl_dataset_namelen(dsl_dataset_t *ds)
 704 {
 705         VERIFY0(dsl_dataset_get_snapname(ds));
 706         mutex_enter(&ds->ds_lock);
 707         int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
 708         mutex_exit(&ds->ds_lock);
 709         return (len);
 710 }
 711
 712 void
 713 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 714 {
 715         dmu_buf_rele(ds->ds_dbuf, tag);
 716 }
 717
 718 void
 719 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 720 {
 721         ASSERT3P(ds->ds_owner, ==, tag);
 722         ASSERT(ds->ds_dbuf != NULL);
 723
 724         mutex_enter(&ds->ds_lock);
 725         ds->ds_owner = NULL;
 726         mutex_exit(&ds->ds_lock);
 727         dsl_dataset_long_rele(ds, tag);
 728         dsl_dataset_rele(ds, tag);
 729 }
 730
 731 boolean_t
 732 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
 733 {
 734         boolean_t gotit = FALSE;
 735
 736         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 737         mutex_enter(&ds->ds_lock);
 738         if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
 739                 ds->ds_owner = tag;
 740                 dsl_dataset_long_hold(ds, tag);
 741                 gotit = TRUE;
 742         }
 743         mutex_exit(&ds->ds_lock);
 744         return (gotit);
 745 }
 746
 747 boolean_t
 748 dsl_dataset_has_owner(dsl_dataset_t *ds)
 749 {
 750         boolean_t rv;
 751         mutex_enter(&ds->ds_lock);
 752         rv = (ds->ds_owner != NULL);
 753         mutex_exit(&ds->ds_lock);
 754         return (rv);
 755 }
 756
 757 static void
 758 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 759 {
 760         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 761         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 762         uint64_t zero = 0;
 763
 764         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 765
 766         spa_feature_incr(spa, f, tx);
 767         dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
 768
 769         VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
 770             sizeof (zero), 1, &zero, tx));
 771 }
 772
 773 void
 774 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 775 {
 776         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 777         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 778
 779         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 780
 781         VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
 782         spa_feature_decr(spa, f, tx);
 783 }
 784
 785 uint64_t
 786 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 787     uint64_t flags, dmu_tx_t *tx)
 788 {
 789         dsl_pool_t *dp = dd->dd_pool;
 790         dmu_buf_t *dbuf;
 791         dsl_dataset_phys_t *dsphys;
 792         uint64_t dsobj;
 793         objset_t *mos = dp->dp_meta_objset;
 794
 795         if (origin == NULL)
 796                 origin = dp->dp_origin_snap;
 797
 798         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 799         ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
 800         ASSERT(dmu_tx_is_syncing(tx));
 801         ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
 802
 803         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 804             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 805         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 806         dmu_buf_will_dirty(dbuf, tx);
 807         dsphys = dbuf->db_data;
 808         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 809         dsphys->ds_dir_obj = dd->dd_object;
 810         dsphys->ds_flags = flags;
 811         dsphys->ds_fsid_guid = unique_create();
 812         do {
 813                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 814                     sizeof (dsphys->ds_guid));
 815         } while (dsphys->ds_guid == 0);
 816         dsphys->ds_snapnames_zapobj =
 817             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 818             DMU_OT_NONE, 0, tx);
 819         dsphys->ds_creation_time = gethrestime_sec();
 820         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 821
 822         if (origin == NULL) {
 823                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 824         } else {
 825                 dsl_dataset_t *ohds; /* head of the origin snapshot */
 826
 827                 dsphys->ds_prev_snap_obj = origin->ds_object;
 828                 dsphys->ds_prev_snap_txg =
 829                     dsl_dataset_phys(origin)->ds_creation_txg;
 830                 dsphys->ds_referenced_bytes =
 831                     dsl_dataset_phys(origin)->ds_referenced_bytes;
 832                 dsphys->ds_compressed_bytes =
 833                     dsl_dataset_phys(origin)->ds_compressed_bytes;
 834                 dsphys->ds_uncompressed_bytes =
 835                     dsl_dataset_phys(origin)->ds_uncompressed_bytes;
 836                 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
 837                 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
 838                 rrw_exit(&origin->ds_bp_rwlock, FTAG);
 839
 840                 /*
 841                  * Inherit flags that describe the dataset's contents
 842                  * (INCONSISTENT) or properties (Case Insensitive).
 843                  */
 844                 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
 845                     (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
 846
 847                 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 848                         if (origin->ds_feature_inuse[f])
 849                                 dsl_dataset_activate_feature(dsobj, f, tx);
 850                 }
 851
 852                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 853                 dsl_dataset_phys(origin)->ds_num_children++;
 854
 855                 VERIFY0(dsl_dataset_hold_obj(dp,
 856                     dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
 857                     FTAG, &ohds));
 858                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 859                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 860                 dsl_dataset_rele(ohds, FTAG);
 861
 862                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 863                         if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
 864                                 dsl_dataset_phys(origin)->ds_next_clones_obj =
 865                                     zap_create(mos,
 866                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 867                         }
 868                         VERIFY0(zap_add_int(mos,
 869                             dsl_dataset_phys(origin)->ds_next_clones_obj,
 870                             dsobj, tx));
 871                 }
 872
 873                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 874                 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
 875                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 876                         if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
 877                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 878                                 dsl_dir_phys(origin->ds_dir)->dd_clones =
 879                                     zap_create(mos,
 880                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 881                         }
 882                         VERIFY0(zap_add_int(mos,
 883                             dsl_dir_phys(origin->ds_dir)->dd_clones,
 884                             dsobj, tx));
 885                 }
 886         }
 887
 888         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 889                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 890
 891         dmu_buf_rele(dbuf, FTAG);
 892
 893         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 894         dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
 895
 896         return (dsobj);
 897 }
 898
 899 static void
 900 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
 901 {
 902         objset_t *os;
 903
 904         VERIFY0(dmu_objset_from_ds(ds, &os));
 905         bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 906         dsl_dataset_dirty(ds, tx);
 907 }
 908
 909 uint64_t
 910 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 911     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 912 {
 913         dsl_pool_t *dp = pdd->dd_pool;
 914         uint64_t dsobj, ddobj;
 915         dsl_dir_t *dd;
 916
 917         ASSERT(dmu_tx_is_syncing(tx));
 918         ASSERT(lastname[0] != '@');
 919
 920         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 921         VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
 922
 923         dsobj = dsl_dataset_create_sync_dd(dd, origin,
 924             flags & ~DS_CREATE_FLAG_NODIRTY, tx);
 925
 926         dsl_deleg_set_create_perms(dd, tx, cr);
 927
 928         /*
 929          * Since we're creating a new node we know it's a leaf, so we can
 930          * initialize the counts if the limit feature is active.
 931          */
 932         if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
 933                 uint64_t cnt = 0;
 934                 objset_t *os = dd->dd_pool->dp_meta_objset;
 935
 936                 dsl_dir_zapify(dd, tx);
 937                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
 938                     sizeof (cnt), 1, &cnt, tx));
 939                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
 940                     sizeof (cnt), 1, &cnt, tx));
 941         }
 942
 943         dsl_dir_rele(dd, FTAG);
 944
 945         /*
 946          * If we are creating a clone, make sure we zero out any stale
 947          * data from the origin snapshots zil header.
 948          */
 949         if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
 950                 dsl_dataset_t *ds;
 951
 952                 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 953                 dsl_dataset_zero_zil(ds, tx);
 954                 dsl_dataset_rele(ds, FTAG);
 955         }
 956
 957         return (dsobj);
 958 }
 959
 960 #ifdef __FreeBSD__
 961 /* FreeBSD ioctl compat begin */
 962 struct destroyarg {
 963         nvlist_t *nvl;
 964         const char *snapname;
 965 };
 966
 967 static int
 968 dsl_check_snap_cb(const char *name, void *arg)
 969 {
 970         struct destroyarg *da = arg;
 971         dsl_dataset_t *ds;
 972         char *dsname;
 973
 974         dsname = kmem_asprintf("%s@%s", name, da->snapname);
 975         fnvlist_add_boolean(da->nvl, dsname);
 976         kmem_free(dsname, strlen(dsname) + 1);
 977
 978         return (0);
 979 }
 980
 981 int
 982 dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
 983     nvlist_t *snaps)
 984 {
 985         struct destroyarg *da;
 986         int err;
 987
 988         da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
 989         da->nvl = snaps;
 990         da->snapname = snapname;
 991         err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
 992             DS_FIND_CHILDREN);
 993         kmem_free(da, sizeof (struct destroyarg));
 994
 995         return (err);
 996 }
 997 /* FreeBSD ioctl compat end */
 998 #endif /* __FreeBSD__ */
 999
1000 /*
1001  * The unique space in the head dataset can be calculated by subtracting
1002  * the space used in the most recent snapshot, that is still being used
1003  * in this file system, from the space currently in use.  To figure out
1004  * the space in the most recent snapshot still in use, we need to take
1005  * the total space used in the snapshot and subtract out the space that
1006  * has been freed up since the snapshot was taken.
1007  */
1008 void
1009 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1010 {
1011         uint64_t mrs_used;
1012         uint64_t dlused, dlcomp, dluncomp;
1013
1014         ASSERT(!ds->ds_is_snapshot);
1015
1016         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1017                 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1018         else
1019                 mrs_used = 0;
1020
1021         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1022
1023         ASSERT3U(dlused, <=, mrs_used);
1024         dsl_dataset_phys(ds)->ds_unique_bytes =
1025             dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1026
1027         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1028             SPA_VERSION_UNIQUE_ACCURATE)
1029                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1030 }
1031
1032 void
1033 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1034     dmu_tx_t *tx)
1035 {
1036         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1037         uint64_t count;
1038         int err;
1039
1040         ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1041         err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1042             obj, tx);
1043         /*
1044          * The err should not be ENOENT, but a bug in a previous version
1045          * of the code could cause upgrade_clones_cb() to not set
1046          * ds_next_snap_obj when it should, leading to a missing entry.
1047          * If we knew that the pool was created after
1048          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1049          * ENOENT.  However, at least we can check that we don't have
1050          * too many entries in the next_clones_obj even after failing to
1051          * remove this one.
1052          */
1053         if (err != ENOENT)
1054                 VERIFY0(err);
1055         ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1056             &count));
1057         ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1058 }
1059
1060
1061 blkptr_t *
1062 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1063 {
1064         return (&dsl_dataset_phys(ds)->ds_bp);
1065 }
1066
1067 spa_t *
1068 dsl_dataset_get_spa(dsl_dataset_t *ds)
1069 {
1070         return (ds->ds_dir->dd_pool->dp_spa);
1071 }
1072
1073 void
1074 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1075 {
1076         dsl_pool_t *dp;
1077
1078         if (ds == NULL) /* this is the meta-objset */
1079                 return;
1080
1081         ASSERT(ds->ds_objset != NULL);
1082
1083         if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1084                 panic("dirtying snapshot!");
1085
1086         dp = ds->ds_dir->dd_pool;
1087
1088         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1089                 /* up the hold count until we can be written out */
1090                 dmu_buf_add_ref(ds->ds_dbuf, ds);
1091         }
1092 }
1093
1094 boolean_t
1095 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1096 {
1097         for (int t = 0; t < TXG_SIZE; t++) {
1098                 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1099                     ds, t))
1100                         return (B_TRUE);
1101         }
1102         return (B_FALSE);
1103 }
1104
1105 static int
1106 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1107 {
1108         uint64_t asize;
1109
1110         if (!dmu_tx_is_syncing(tx))
1111                 return (0);
1112
1113         /*
1114          * If there's an fs-only reservation, any blocks that might become
1115          * owned by the snapshot dataset must be accommodated by space
1116          * outside of the reservation.
1117          */
1118         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1119         asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1120         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1121                 return (SET_ERROR(ENOSPC));
1122
1123         /*
1124          * Propagate any reserved space for this snapshot to other
1125          * snapshot checks in this sync group.
1126          */
1127         if (asize > 0)
1128                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1129
1130         return (0);
1131 }
1132
1133 typedef struct dsl_dataset_snapshot_arg {
1134         nvlist_t *ddsa_snaps;
1135         nvlist_t *ddsa_props;
1136         nvlist_t *ddsa_errors;
1137         cred_t *ddsa_cr;
1138 } dsl_dataset_snapshot_arg_t;
1139
1140 int
1141 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1142     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1143 {
1144         int error;
1145         uint64_t value;
1146
1147         ds->ds_trysnap_txg = tx->tx_txg;
1148
1149         if (!dmu_tx_is_syncing(tx))
1150                 return (0);
1151
1152         /*
1153          * We don't allow multiple snapshots of the same txg.  If there
1154          * is already one, try again.
1155          */
1156         if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1157                 return (SET_ERROR(EAGAIN));
1158
1159         /*
1160          * Check for conflicting snapshot name.
1161          */
1162         error = dsl_dataset_snap_lookup(ds, snapname, &value);
1163         if (error == 0)
1164                 return (SET_ERROR(EEXIST));
1165         if (error != ENOENT)
1166                 return (error);
1167
1168         /*
1169          * We don't allow taking snapshots of inconsistent datasets, such as
1170          * those into which we are currently receiving.  However, if we are
1171          * creating this snapshot as part of a receive, this check will be
1172          * executed atomically with respect to the completion of the receive
1173          * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1174          * case we ignore this, knowing it will be fixed up for us shortly in
1175          * dmu_recv_end_sync().
1176          */
1177         if (!recv && DS_IS_INCONSISTENT(ds))
1178                 return (SET_ERROR(EBUSY));
1179
1180         /*
1181          * Skip the check for temporary snapshots or if we have already checked
1182          * the counts in dsl_dataset_snapshot_check. This means we really only
1183          * check the count here when we're receiving a stream.
1184          */
1185         if (cnt != 0 && cr != NULL) {
1186                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1187                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1188                 if (error != 0)
1189                         return (error);
1190         }
1191
1192         error = dsl_dataset_snapshot_reserve_space(ds, tx);
1193         if (error != 0)
1194                 return (error);
1195
1196         return (0);
1197 }
1198
1199 static int
1200 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1201 {
1202         dsl_dataset_snapshot_arg_t *ddsa = arg;
1203         dsl_pool_t *dp = dmu_tx_pool(tx);
1204         nvpair_t *pair;
1205         int rv = 0;
1206
1207         /*
1208          * Pre-compute how many total new snapshots will be created for each
1209          * level in the tree and below. This is needed for validating the
1210          * snapshot limit when either taking a recursive snapshot or when
1211          * taking multiple snapshots.
1212          *
1213          * The problem is that the counts are not actually adjusted when
1214          * we are checking, only when we finally sync. For a single snapshot,
1215          * this is easy, the count will increase by 1 at each node up the tree,
1216          * but its more complicated for the recursive/multiple snapshot case.
1217          *
1218          * The dsl_fs_ss_limit_check function does recursively check the count
1219          * at each level up the tree but since it is validating each snapshot
1220          * independently we need to be sure that we are validating the complete
1221          * count for the entire set of snapshots. We do this by rolling up the
1222          * counts for each component of the name into an nvlist and then
1223          * checking each of those cases with the aggregated count.
1224          *
1225          * This approach properly handles not only the recursive snapshot
1226          * case (where we get all of those on the ddsa_snaps list) but also
1227          * the sibling case (e.g. snapshot a/b and a/c so that we will also
1228          * validate the limit on 'a' using a count of 2).
1229          *
1230          * We validate the snapshot names in the third loop and only report
1231          * name errors once.
1232          */
1233         if (dmu_tx_is_syncing(tx)) {
1234                 nvlist_t *cnt_track = NULL;
1235                 cnt_track = fnvlist_alloc();
1236
1237                 /* Rollup aggregated counts into the cnt_track list */
1238                 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1239                     pair != NULL;
1240                     pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1241                         char *pdelim;
1242                         uint64_t val;
1243                         char nm[MAXPATHLEN];
1244
1245                         (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1246                         pdelim = strchr(nm, '@');
1247                         if (pdelim == NULL)
1248                                 continue;
1249                         *pdelim = '\0';
1250
1251                         do {
1252                                 if (nvlist_lookup_uint64(cnt_track, nm,
1253                                     &val) == 0) {
1254                                         /* update existing entry */
1255                                         fnvlist_add_uint64(cnt_track, nm,
1256                                             val + 1);
1257                                 } else {
1258                                         /* add to list */
1259                                         fnvlist_add_uint64(cnt_track, nm, 1);
1260                                 }
1261
1262                                 pdelim = strrchr(nm, '/');
1263                                 if (pdelim != NULL)
1264                                         *pdelim = '\0';
1265                         } while (pdelim != NULL);
1266                 }
1267
1268                 /* Check aggregated counts at each level */
1269                 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1270                     pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1271                         int error = 0;
1272                         char *name;
1273                         uint64_t cnt = 0;
1274                         dsl_dataset_t *ds;
1275
1276                         name = nvpair_name(pair);
1277                         cnt = fnvpair_value_uint64(pair);
1278                         ASSERT(cnt > 0);
1279
1280                         error = dsl_dataset_hold(dp, name, FTAG, &ds);
1281                         if (error == 0) {
1282                                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1283                                     ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1284                                     ddsa->ddsa_cr);
1285                                 dsl_dataset_rele(ds, FTAG);
1286                         }
1287
1288                         if (error != 0) {
1289                                 if (ddsa->ddsa_errors != NULL)
1290                                         fnvlist_add_int32(ddsa->ddsa_errors,
1291                                             name, error);
1292                                 rv = error;
1293                                 /* only report one error for this check */
1294                                 break;
1295                         }
1296                 }
1297                 nvlist_free(cnt_track);
1298         }
1299
1300         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1301             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1302                 int error = 0;
1303                 dsl_dataset_t *ds;
1304                 char *name, *atp;
1305                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1306
1307                 name = nvpair_name(pair);
1308                 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1309                         error = SET_ERROR(ENAMETOOLONG);
1310                 if (error == 0) {
1311                         atp = strchr(name, '@');
1312                         if (atp == NULL)
1313                                 error = SET_ERROR(EINVAL);
1314                         if (error == 0)
1315                                 (void) strlcpy(dsname, name, atp - name + 1);
1316                 }
1317                 if (error == 0)
1318                         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1319                 if (error == 0) {
1320                         /* passing 0/NULL skips dsl_fs_ss_limit_check */
1321                         error = dsl_dataset_snapshot_check_impl(ds,
1322                             atp + 1, tx, B_FALSE, 0, NULL);
1323                         dsl_dataset_rele(ds, FTAG);
1324                 }
1325
1326                 if (error != 0) {
1327                         if (ddsa->ddsa_errors != NULL) {
1328                                 fnvlist_add_int32(ddsa->ddsa_errors,
1329                                     name, error);
1330                         }
1331                         rv = error;
1332                 }
1333         }
1334
1335         return (rv);
1336 }
1337
1338 void
1339 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1340     dmu_tx_t *tx)
1341 {
1342         static zil_header_t zero_zil;
1343
1344         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1345         dmu_buf_t *dbuf;
1346         dsl_dataset_phys_t *dsphys;
1347         uint64_t dsobj, crtxg;
1348         objset_t *mos = dp->dp_meta_objset;
1349         objset_t *os;
1350
1351         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1352
1353         /*
1354          * If we are on an old pool, the zil must not be active, in which
1355          * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1356          */
1357         ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1358             dmu_objset_from_ds(ds, &os) != 0 ||
1359             bcmp(&os->os_phys->os_zil_header, &zero_zil,
1360             sizeof (zero_zil)) == 0);
1361
1362         dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1363
1364         /*
1365          * The origin's ds_creation_txg has to be < TXG_INITIAL
1366          */
1367         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1368                 crtxg = 1;
1369         else
1370                 crtxg = tx->tx_txg;
1371
1372         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1373             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1374         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1375         dmu_buf_will_dirty(dbuf, tx);
1376         dsphys = dbuf->db_data;
1377         bzero(dsphys, sizeof (dsl_dataset_phys_t));
1378         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1379         dsphys->ds_fsid_guid = unique_create();
1380         do {
1381                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1382                     sizeof (dsphys->ds_guid));
1383         } while (dsphys->ds_guid == 0);
1384         dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1385         dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1386         dsphys->ds_next_snap_obj = ds->ds_object;
1387         dsphys->ds_num_children = 1;
1388         dsphys->ds_creation_time = gethrestime_sec();
1389         dsphys->ds_creation_txg = crtxg;
1390         dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1391         dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1392         dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1393         dsphys->ds_uncompressed_bytes =
1394             dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1395         dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1396         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1397         dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1398         rrw_exit(&ds->ds_bp_rwlock, FTAG);
1399         dmu_buf_rele(dbuf, FTAG);
1400
1401         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1402                 if (ds->ds_feature_inuse[f])
1403                         dsl_dataset_activate_feature(dsobj, f, tx);
1404         }
1405
1406         ASSERT3U(ds->ds_prev != 0, ==,
1407             dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1408         if (ds->ds_prev) {
1409                 uint64_t next_clones_obj =
1410                     dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1411                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1412                     ds->ds_object ||
1413                     dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1414                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1415                     ds->ds_object) {
1416                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1417                         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1418                             dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1419                         dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1420                 } else if (next_clones_obj != 0) {
1421                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
1422                             dsphys->ds_next_snap_obj, tx);
1423                         VERIFY0(zap_add_int(mos,
1424                             next_clones_obj, dsobj, tx));
1425                 }
1426         }
1427
1428         /*
1429          * If we have a reference-reservation on this dataset, we will
1430          * need to increase the amount of refreservation being charged
1431          * since our unique space is going to zero.
1432          */
1433         if (ds->ds_reserved) {
1434                 int64_t delta;
1435                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1436                 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1437                     ds->ds_reserved);
1438                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1439                     delta, 0, 0, tx);
1440         }
1441
1442         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1443         dsl_dataset_phys(ds)->ds_deadlist_obj =
1444             dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1445             dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1446         dsl_deadlist_close(&ds->ds_deadlist);
1447         dsl_deadlist_open(&ds->ds_deadlist, mos,
1448             dsl_dataset_phys(ds)->ds_deadlist_obj);
1449         dsl_deadlist_add_key(&ds->ds_deadlist,
1450             dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1451
1452         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1453         dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1454         dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1455         dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1456         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1457                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1458
1459         VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1460             snapname, 8, 1, &dsobj, tx));
1461
1462         if (ds->ds_prev)
1463                 dsl_dataset_rele(ds->ds_prev, ds);
1464         VERIFY0(dsl_dataset_hold_obj(dp,
1465             dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1466
1467         dsl_scan_ds_snapshotted(ds, tx);
1468
1469         dsl_dir_snap_cmtime_update(ds->ds_dir);
1470
1471         spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1472 }
1473
1474 static void
1475 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1476 {
1477         dsl_dataset_snapshot_arg_t *ddsa = arg;
1478         dsl_pool_t *dp = dmu_tx_pool(tx);
1479         nvpair_t *pair;
1480
1481         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1482             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1483                 dsl_dataset_t *ds;
1484                 char *name, *atp;
1485                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1486
1487                 name = nvpair_name(pair);
1488                 atp = strchr(name, '@');
1489                 (void) strlcpy(dsname, name, atp - name + 1);
1490                 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1491
1492                 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1493                 if (ddsa->ddsa_props != NULL) {
1494                         dsl_props_set_sync_impl(ds->ds_prev,
1495                             ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1496                 }
1497                 dsl_dataset_rele(ds, FTAG);
1498         }
1499 }
1500
1501 /*
1502  * The snapshots must all be in the same pool.
1503  * All-or-nothing: if there are any failures, nothing will be modified.
1504  */
1505 int
1506 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1507 {
1508         dsl_dataset_snapshot_arg_t ddsa;
1509         nvpair_t *pair;
1510         boolean_t needsuspend;
1511         int error;
1512         spa_t *spa;
1513         char *firstname;
1514         nvlist_t *suspended = NULL;
1515
1516         pair = nvlist_next_nvpair(snaps, NULL);
1517         if (pair == NULL)
1518                 return (0);
1519         firstname = nvpair_name(pair);
1520
1521         error = spa_open(firstname, &spa, FTAG);
1522         if (error != 0)
1523                 return (error);
1524         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1525         spa_close(spa, FTAG);
1526
1527         if (needsuspend) {
1528                 suspended = fnvlist_alloc();
1529                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1530                     pair = nvlist_next_nvpair(snaps, pair)) {
1531                         char fsname[ZFS_MAX_DATASET_NAME_LEN];
1532                         char *snapname = nvpair_name(pair);
1533                         char *atp;
1534                         void *cookie;
1535
1536                         atp = strchr(snapname, '@');
1537                         if (atp == NULL) {
1538                                 error = SET_ERROR(EINVAL);
1539                                 break;
1540                         }
1541                         (void) strlcpy(fsname, snapname, atp - snapname + 1);
1542
1543                         error = zil_suspend(fsname, &cookie);
1544                         if (error != 0)
1545                                 break;
1546                         fnvlist_add_uint64(suspended, fsname,
1547                             (uintptr_t)cookie);
1548                 }
1549         }
1550
1551         ddsa.ddsa_snaps = snaps;
1552         ddsa.ddsa_props = props;
1553         ddsa.ddsa_errors = errors;
1554         ddsa.ddsa_cr = CRED();
1555
1556         if (error == 0) {
1557                 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1558                     dsl_dataset_snapshot_sync, &ddsa,
1559                     fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1560         }
1561
1562         if (suspended != NULL) {
1563                 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1564                     pair = nvlist_next_nvpair(suspended, pair)) {
1565                         zil_resume((void *)(uintptr_t)
1566                             fnvpair_value_uint64(pair));
1567                 }
1568                 fnvlist_free(suspended);
1569         }
1570
1571 #ifdef __FreeBSD__
1572 #ifdef _KERNEL
1573         if (error == 0) {
1574                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1575                     pair = nvlist_next_nvpair(snaps, pair)) {
1576                         char *snapname = nvpair_name(pair);
1577                         zvol_create_minors(snapname);
1578                 }
1579         }
1580 #endif
1581 #endif
1582         return (error);
1583 }
1584
1585 typedef struct dsl_dataset_snapshot_tmp_arg {
1586         const char *ddsta_fsname;
1587         const char *ddsta_snapname;
1588         minor_t ddsta_cleanup_minor;
1589         const char *ddsta_htag;
1590 } dsl_dataset_snapshot_tmp_arg_t;
1591
1592 static int
1593 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1594 {
1595         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1596         dsl_pool_t *dp = dmu_tx_pool(tx);
1597         dsl_dataset_t *ds;
1598         int error;
1599
1600         error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1601         if (error != 0)
1602                 return (error);
1603
1604         /* NULL cred means no limit check for tmp snapshot */
1605         error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1606             tx, B_FALSE, 0, NULL);
1607         if (error != 0) {
1608                 dsl_dataset_rele(ds, FTAG);
1609                 return (error);
1610         }
1611
1612         if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1613                 dsl_dataset_rele(ds, FTAG);
1614                 return (SET_ERROR(ENOTSUP));
1615         }
1616         error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1617             B_TRUE, tx);
1618         if (error != 0) {
1619                 dsl_dataset_rele(ds, FTAG);
1620                 return (error);
1621         }
1622
1623         dsl_dataset_rele(ds, FTAG);
1624         return (0);
1625 }
1626
1627 static void
1628 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1629 {
1630         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1631         dsl_pool_t *dp = dmu_tx_pool(tx);
1632         dsl_dataset_t *ds;
1633
1634         VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1635
1636         dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1637         dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1638             ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1639         dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1640
1641         dsl_dataset_rele(ds, FTAG);
1642 }
1643
1644 int
1645 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1646     minor_t cleanup_minor, const char *htag)
1647 {
1648         dsl_dataset_snapshot_tmp_arg_t ddsta;
1649         int error;
1650         spa_t *spa;
1651         boolean_t needsuspend;
1652         void *cookie;
1653
1654         ddsta.ddsta_fsname = fsname;
1655         ddsta.ddsta_snapname = snapname;
1656         ddsta.ddsta_cleanup_minor = cleanup_minor;
1657         ddsta.ddsta_htag = htag;
1658
1659         error = spa_open(fsname, &spa, FTAG);
1660         if (error != 0)
1661                 return (error);
1662         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1663         spa_close(spa, FTAG);
1664
1665         if (needsuspend) {
1666                 error = zil_suspend(fsname, &cookie);
1667                 if (error != 0)
1668                         return (error);
1669         }
1670
1671         error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1672             dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1673
1674         if (needsuspend)
1675                 zil_resume(cookie);
1676         return (error);
1677 }
1678
1679
1680 void
1681 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1682 {
1683         ASSERT(dmu_tx_is_syncing(tx));
1684         ASSERT(ds->ds_objset != NULL);
1685         ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1686
1687         /*
1688          * in case we had to change ds_fsid_guid when we opened it,
1689          * sync it out now.
1690          */
1691         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1692         dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1693
1694         if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1695                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1696                     ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1697                     &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1698                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1699                     ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1700                     &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1701                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1702                     ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1703                     &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1704                 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1705                 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1706                 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1707         }
1708
1709         dmu_objset_sync(ds->ds_objset, zio, tx);
1710
1711         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1712                 if (ds->ds_feature_activation_needed[f]) {
1713                         if (ds->ds_feature_inuse[f])
1714                                 continue;
1715                         dsl_dataset_activate_feature(ds->ds_object, f, tx);
1716                         ds->ds_feature_inuse[f] = B_TRUE;
1717                 }
1718         }
1719 }
1720
1721 static void
1722 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1723 {
1724         uint64_t count = 0;
1725         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1726         zap_cursor_t zc;
1727         zap_attribute_t za;
1728         nvlist_t *propval = fnvlist_alloc();
1729         nvlist_t *val = fnvlist_alloc();
1730
1731         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1732
1733         /*
1734          * There may be missing entries in ds_next_clones_obj
1735          * due to a bug in a previous version of the code.
1736          * Only trust it if it has the right number of entries.
1737          */
1738         if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1739                 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1740                     &count));
1741         }
1742         if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1743                 goto fail;
1744         for (zap_cursor_init(&zc, mos,
1745             dsl_dataset_phys(ds)->ds_next_clones_obj);
1746             zap_cursor_retrieve(&zc, &za) == 0;
1747             zap_cursor_advance(&zc)) {
1748                 dsl_dataset_t *clone;
1749                 char buf[ZFS_MAX_DATASET_NAME_LEN];
1750                 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1751                     za.za_first_integer, FTAG, &clone));
1752                 dsl_dir_name(clone->ds_dir, buf);
1753                 fnvlist_add_boolean(val, buf);
1754                 dsl_dataset_rele(clone, FTAG);
1755         }
1756         zap_cursor_fini(&zc);
1757         fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1758         fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1759 fail:
1760         nvlist_free(val);
1761         nvlist_free(propval);
1762 }
1763
1764 static void
1765 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1766 {
1767         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1768
1769         if (dsl_dataset_has_resume_receive_state(ds)) {
1770                 char *str;
1771                 void *packed;
1772                 uint8_t *compressed;
1773                 uint64_t val;
1774                 nvlist_t *token_nv = fnvlist_alloc();
1775                 size_t packed_size, compressed_size;
1776
1777                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1778                     DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1779                         fnvlist_add_uint64(token_nv, "fromguid", val);
1780                 }
1781                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1782                     DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1783                         fnvlist_add_uint64(token_nv, "object", val);
1784                 }
1785                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1786                     DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1787                         fnvlist_add_uint64(token_nv, "offset", val);
1788                 }
1789                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1790                     DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1791                         fnvlist_add_uint64(token_nv, "bytes", val);
1792                 }
1793                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1794                     DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1795                         fnvlist_add_uint64(token_nv, "toguid", val);
1796                 }
1797                 char buf[256];
1798                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1799                     DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1800                         fnvlist_add_string(token_nv, "toname", buf);
1801                 }
1802                 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1803                     DS_FIELD_RESUME_EMBEDOK) == 0) {
1804                         fnvlist_add_boolean(token_nv, "embedok");
1805                 }
1806                 packed = fnvlist_pack(token_nv, &packed_size);
1807                 fnvlist_free(token_nv);
1808                 compressed = kmem_alloc(packed_size, KM_SLEEP);
1809
1810                 compressed_size = gzip_compress(packed, compressed,
1811                     packed_size, packed_size, 6);
1812
1813                 zio_cksum_t cksum;
1814                 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1815
1816                 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1817                 for (int i = 0; i < compressed_size; i++) {
1818                         (void) sprintf(str + i * 2, "%02x", compressed[i]);
1819                 }
1820                 str[compressed_size * 2] = '\0';
1821                 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1822                     ZFS_SEND_RESUME_TOKEN_VERSION,
1823                     (longlong_t)cksum.zc_word[0],
1824                     (longlong_t)packed_size, str);
1825                 dsl_prop_nvlist_add_string(nv,
1826                     ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1827                 kmem_free(packed, packed_size);
1828                 kmem_free(str, compressed_size * 2 + 1);
1829                 kmem_free(compressed, packed_size);
1830                 strfree(propval);
1831         }
1832 }
1833
1834 void
1835 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1836 {
1837         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1838         uint64_t refd, avail, uobjs, aobjs, ratio;
1839
1840         ASSERT(dsl_pool_config_held(dp));
1841
1842         ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1843             (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1844             dsl_dataset_phys(ds)->ds_compressed_bytes);
1845
1846         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1847         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1848             dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1849
1850         if (ds->ds_is_snapshot) {
1851                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1852                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1853                     dsl_dataset_phys(ds)->ds_unique_bytes);
1854                 get_clones_stat(ds, nv);
1855         } else {
1856                 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1857                         char buf[ZFS_MAX_DATASET_NAME_LEN];
1858                         dsl_dataset_name(ds->ds_prev, buf);
1859                         dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1860                 }
1861
1862                 dsl_dir_stats(ds->ds_dir, nv);
1863         }
1864
1865         dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1866         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1867         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1868
1869         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1870             dsl_dataset_phys(ds)->ds_creation_time);
1871         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1872             dsl_dataset_phys(ds)->ds_creation_txg);
1873         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1874             ds->ds_quota);
1875         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1876             ds->ds_reserved);
1877         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1878             dsl_dataset_phys(ds)->ds_guid);
1879         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1880             dsl_dataset_phys(ds)->ds_unique_bytes);
1881         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1882             ds->ds_object);
1883         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1884             ds->ds_userrefs);
1885         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1886             DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1887
1888         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1889                 uint64_t written, comp, uncomp;
1890                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1891                 dsl_dataset_t *prev;
1892
1893                 int err = dsl_dataset_hold_obj(dp,
1894                     dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1895                 if (err == 0) {
1896                         err = dsl_dataset_space_written(prev, ds, &written,
1897                             &comp, &uncomp);
1898                         dsl_dataset_rele(prev, FTAG);
1899                         if (err == 0) {
1900                                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1901                                     written);
1902                         }
1903                 }
1904         }
1905
1906         if (!dsl_dataset_is_snapshot(ds)) {
1907                 /*
1908                  * A failed "newfs" (e.g. full) resumable receive leaves
1909                  * the stats set on this dataset.  Check here for the prop.
1910                  */
1911                 get_receive_resume_stats(ds, nv);
1912
1913                 /*
1914                  * A failed incremental resumable receive leaves the
1915                  * stats set on our child named "%recv".  Check the child
1916                  * for the prop.
1917                  */
1918                 /* 6 extra bytes for /%recv */
1919                 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1920                 dsl_dataset_t *recv_ds;
1921                 dsl_dataset_name(ds, recvname);
1922                 if (strlcat(recvname, "/", sizeof (recvname)) <
1923                     sizeof (recvname) &&
1924                     strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1925                     sizeof (recvname) &&
1926                     dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1927                         get_receive_resume_stats(recv_ds, nv);
1928                         dsl_dataset_rele(recv_ds, FTAG);
1929                 }
1930         }
1931 }
1932
1933 void
1934 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1935 {
1936         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1937         ASSERT(dsl_pool_config_held(dp));
1938
1939         stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1940         stat->dds_inconsistent =
1941             dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1942         stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1943         stat->dds_origin[0] = '\0';
1944         if (ds->ds_is_snapshot) {
1945                 stat->dds_is_snapshot = B_TRUE;
1946                 stat->dds_num_clones =
1947                     dsl_dataset_phys(ds)->ds_num_children - 1;
1948         } else {
1949                 stat->dds_is_snapshot = B_FALSE;
1950                 stat->dds_num_clones = 0;
1951
1952                 if (dsl_dir_is_clone(ds->ds_dir)) {
1953                         dsl_dataset_t *ods;
1954
1955                         VERIFY0(dsl_dataset_hold_obj(dp,
1956                             dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1957                             FTAG, &ods));
1958                         dsl_dataset_name(ods, stat->dds_origin);
1959                         dsl_dataset_rele(ods, FTAG);
1960                 }
1961         }
1962 }
1963
1964 uint64_t
1965 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1966 {
1967         return (ds->ds_fsid_guid);
1968 }
1969
1970 void
1971 dsl_dataset_space(dsl_dataset_t *ds,
1972     uint64_t *refdbytesp, uint64_t *availbytesp,
1973     uint64_t *usedobjsp, uint64_t *availobjsp)
1974 {
1975         *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1976         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1977         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1978                 *availbytesp +=
1979                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1980         if (ds->ds_quota != 0) {
1981                 /*
1982                  * Adjust available bytes according to refquota
1983                  */
1984                 if (*refdbytesp < ds->ds_quota)
1985                         *availbytesp = MIN(*availbytesp,
1986                             ds->ds_quota - *refdbytesp);
1987                 else
1988                         *availbytesp = 0;
1989         }
1990         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1991         *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1992         rrw_exit(&ds->ds_bp_rwlock, FTAG);
1993         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1994 }
1995
1996 boolean_t
1997 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1998 {
1999         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2000         uint64_t birth;
2001
2002         ASSERT(dsl_pool_config_held(dp));
2003         if (snap == NULL)
2004                 return (B_FALSE);
2005         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2006         birth = dsl_dataset_get_blkptr(ds)->blk_birth;
2007         rrw_exit(&ds->ds_bp_rwlock, FTAG);
2008         if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2009                 objset_t *os, *os_snap;
2010                 /*
2011                  * It may be that only the ZIL differs, because it was
2012                  * reset in the head.  Don't count that as being
2013                  * modified.
2014                  */
2015                 if (dmu_objset_from_ds(ds, &os) != 0)
2016                         return (B_TRUE);
2017                 if (dmu_objset_from_ds(snap, &os_snap) != 0)
2018                         return (B_TRUE);
2019                 return (bcmp(&os->os_phys->os_meta_dnode,
2020                     &os_snap->os_phys->os_meta_dnode,
2021                     sizeof (os->os_phys->os_meta_dnode)) != 0);
2022         }
2023         return (B_FALSE);
2024 }
2025
2026 typedef struct dsl_dataset_rename_snapshot_arg {
2027         const char *ddrsa_fsname;
2028         const char *ddrsa_oldsnapname;
2029         const char *ddrsa_newsnapname;
2030         boolean_t ddrsa_recursive;
2031         dmu_tx_t *ddrsa_tx;
2032 } dsl_dataset_rename_snapshot_arg_t;
2033
2034 /* ARGSUSED */
2035 static int
2036 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2037     dsl_dataset_t *hds, void *arg)
2038 {
2039         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2040         int error;
2041         uint64_t val;
2042
2043         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2044         if (error != 0) {
2045                 /* ignore nonexistent snapshots */
2046                 return (error == ENOENT ? 0 : error);
2047         }
2048
2049         /* new name should not exist */
2050         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2051         if (error == 0)
2052                 error = SET_ERROR(EEXIST);
2053         else if (error == ENOENT)
2054                 error = 0;
2055
2056         /* dataset name + 1 for the "@" + the new snapshot name must fit */
2057         if (dsl_dir_namelen(hds->ds_dir) + 1 +
2058             strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2059                 error = SET_ERROR(ENAMETOOLONG);
2060
2061         return (error);
2062 }
2063
2064 static int
2065 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2066 {
2067         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2068         dsl_pool_t *dp = dmu_tx_pool(tx);
2069         dsl_dataset_t *hds;
2070         int error;
2071
2072         error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2073         if (error != 0)
2074                 return (error);
2075
2076         if (ddrsa->ddrsa_recursive) {
2077                 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2078                     dsl_dataset_rename_snapshot_check_impl, ddrsa,
2079                     DS_FIND_CHILDREN);
2080         } else {
2081                 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2082         }
2083         dsl_dataset_rele(hds, FTAG);
2084         return (error);
2085 }
2086
2087 static int
2088 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2089     dsl_dataset_t *hds, void *arg)
2090 {
2091 #ifdef __FreeBSD__
2092 #ifdef _KERNEL
2093         char *oldname, *newname;
2094 #endif
2095 #endif
2096         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2097         dsl_dataset_t *ds;
2098         uint64_t val;
2099         dmu_tx_t *tx = ddrsa->ddrsa_tx;
2100         int error;
2101
2102         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2103         ASSERT(error == 0 || error == ENOENT);
2104         if (error == ENOENT) {
2105                 /* ignore nonexistent snapshots */
2106                 return (0);
2107         }
2108
2109         VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2110
2111         /* log before we change the name */
2112         spa_history_log_internal_ds(ds, "rename", tx,
2113             "-> @%s", ddrsa->ddrsa_newsnapname);
2114
2115         VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2116             B_FALSE));
2117         mutex_enter(&ds->ds_lock);
2118         (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2119         mutex_exit(&ds->ds_lock);
2120         VERIFY0(zap_add(dp->dp_meta_objset,
2121             dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2122             ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2123
2124 #ifdef __FreeBSD__
2125 #ifdef _KERNEL
2126         oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2127         newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2128         snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2129             ddrsa->ddrsa_oldsnapname);
2130         snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2131             ddrsa->ddrsa_newsnapname);
2132         zfsvfs_update_fromname(oldname, newname);
2133         zvol_rename_minors(oldname, newname);
2134         kmem_free(newname, MAXPATHLEN);
2135         kmem_free(oldname, MAXPATHLEN);
2136 #endif
2137 #endif
2138         dsl_dataset_rele(ds, FTAG);
2139
2140         return (0);
2141 }
2142
2143 static void
2144 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2145 {
2146         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2147         dsl_pool_t *dp = dmu_tx_pool(tx);
2148         dsl_dataset_t *hds;
2149
2150         VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2151         ddrsa->ddrsa_tx = tx;
2152         if (ddrsa->ddrsa_recursive) {
2153                 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2154                     dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2155                     DS_FIND_CHILDREN));
2156         } else {
2157                 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2158         }
2159         dsl_dataset_rele(hds, FTAG);
2160 }
2161
2162 int
2163 dsl_dataset_rename_snapshot(const char *fsname,
2164     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2165 {
2166         dsl_dataset_rename_snapshot_arg_t ddrsa;
2167
2168         ddrsa.ddrsa_fsname = fsname;
2169         ddrsa.ddrsa_oldsnapname = oldsnapname;
2170         ddrsa.ddrsa_newsnapname = newsnapname;
2171         ddrsa.ddrsa_recursive = recursive;
2172
2173         return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2174             dsl_dataset_rename_snapshot_sync, &ddrsa,
2175             1, ZFS_SPACE_CHECK_RESERVED));
2176 }
2177
2178 /*
2179  * If we're doing an ownership handoff, we need to make sure that there is
2180  * only one long hold on the dataset.  We're not allowed to change anything here
2181  * so we don't permanently release the long hold or regular hold here.  We want
2182  * to do this only when syncing to avoid the dataset unexpectedly going away
2183  * when we release the long hold.
2184  */
2185 static int
2186 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2187 {
2188         boolean_t held;
2189
2190         if (!dmu_tx_is_syncing(tx))
2191                 return (0);
2192
2193         if (owner != NULL) {
2194                 VERIFY3P(ds->ds_owner, ==, owner);
2195                 dsl_dataset_long_rele(ds, owner);
2196         }
2197
2198         held = dsl_dataset_long_held(ds);
2199
2200         if (owner != NULL)
2201                 dsl_dataset_long_hold(ds, owner);
2202
2203         if (held)
2204                 return (SET_ERROR(EBUSY));
2205
2206         return (0);
2207 }
2208
2209 typedef struct dsl_dataset_rollback_arg {
2210         const char *ddra_fsname;
2211         void *ddra_owner;
2212         nvlist_t *ddra_result;
2213 } dsl_dataset_rollback_arg_t;
2214
2215 static int
2216 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2217 {
2218         dsl_dataset_rollback_arg_t *ddra = arg;
2219         dsl_pool_t *dp = dmu_tx_pool(tx);
2220         dsl_dataset_t *ds;
2221         int64_t unused_refres_delta;
2222         int error;
2223
2224         error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2225         if (error != 0)
2226                 return (error);
2227
2228         /* must not be a snapshot */
2229         if (ds->ds_is_snapshot) {
2230                 dsl_dataset_rele(ds, FTAG);
2231                 return (SET_ERROR(EINVAL));
2232         }
2233
2234         /* must have a most recent snapshot */
2235         if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2236                 dsl_dataset_rele(ds, FTAG);
2237                 return (SET_ERROR(EINVAL));
2238         }
2239
2240         /* must not have any bookmarks after the most recent snapshot */
2241         nvlist_t *proprequest = fnvlist_alloc();
2242         fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2243         nvlist_t *bookmarks = fnvlist_alloc();
2244         error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2245         fnvlist_free(proprequest);
2246         if (error != 0)
2247                 return (error);
2248         for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2249             pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2250                 nvlist_t *valuenv =
2251                     fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2252                     zfs_prop_to_name(ZFS_PROP_CREATETXG));
2253                 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2254                 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2255                         fnvlist_free(bookmarks);
2256                         dsl_dataset_rele(ds, FTAG);
2257                         return (SET_ERROR(EEXIST));
2258                 }
2259         }
2260         fnvlist_free(bookmarks);
2261
2262         error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2263         if (error != 0) {
2264                 dsl_dataset_rele(ds, FTAG);
2265                 return (error);
2266         }
2267
2268         /*
2269          * Check if the snap we are rolling back to uses more than
2270          * the refquota.
2271          */
2272         if (ds->ds_quota != 0 &&
2273             dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2274                 dsl_dataset_rele(ds, FTAG);
2275                 return (SET_ERROR(EDQUOT));
2276         }
2277
2278         /*
2279          * When we do the clone swap, we will temporarily use more space
2280          * due to the refreservation (the head will no longer have any
2281          * unique space, so the entire amount of the refreservation will need
2282          * to be free).  We will immediately destroy the clone, freeing
2283          * this space, but the freeing happens over many txg's.
2284          */
2285         unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2286             dsl_dataset_phys(ds)->ds_unique_bytes);
2287
2288         if (unused_refres_delta > 0 &&
2289             unused_refres_delta >
2290             dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2291                 dsl_dataset_rele(ds, FTAG);
2292                 return (SET_ERROR(ENOSPC));
2293         }
2294
2295         dsl_dataset_rele(ds, FTAG);
2296         return (0);
2297 }
2298
2299 static void
2300 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2301 {
2302         dsl_dataset_rollback_arg_t *ddra = arg;
2303         dsl_pool_t *dp = dmu_tx_pool(tx);
2304         dsl_dataset_t *ds, *clone;
2305         uint64_t cloneobj;
2306         char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2307
2308         VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2309
2310         dsl_dataset_name(ds->ds_prev, namebuf);
2311         fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2312
2313         cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2314             ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2315
2316         VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2317
2318         dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2319         dsl_dataset_zero_zil(ds, tx);
2320
2321         dsl_destroy_head_sync_impl(clone, tx);
2322
2323         dsl_dataset_rele(clone, FTAG);
2324         dsl_dataset_rele(ds, FTAG);
2325 }
2326
2327 /*
2328  * Rolls back the given filesystem or volume to the most recent snapshot.
2329  * The name of the most recent snapshot will be returned under key "target"
2330  * in the result nvlist.
2331  *
2332  * If owner != NULL:
2333  * - The existing dataset MUST be owned by the specified owner at entry
2334  * - Upon return, dataset will still be held by the same owner, whether we
2335  *   succeed or not.
2336  *
2337  * This mode is required any time the existing filesystem is mounted.  See
2338  * notes above zfs_suspend_fs() for further details.
2339  */
2340 int
2341 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2342 {
2343         dsl_dataset_rollback_arg_t ddra;
2344
2345         ddra.ddra_fsname = fsname;
2346         ddra.ddra_owner = owner;
2347         ddra.ddra_result = result;
2348
2349         return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2350             dsl_dataset_rollback_sync, &ddra,
2351             1, ZFS_SPACE_CHECK_RESERVED));
2352 }
2353
2354 struct promotenode {
2355         list_node_t link;
2356         dsl_dataset_t *ds;
2357 };
2358
2359 typedef struct dsl_dataset_promote_arg {
2360         const char *ddpa_clonename;
2361         dsl_dataset_t *ddpa_clone;
2362         list_t shared_snaps, origin_snaps, clone_snaps;
2363         dsl_dataset_t *origin_origin; /* origin of the origin */
2364         uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2365         char *err_ds;
2366         cred_t *cr;
2367 } dsl_dataset_promote_arg_t;
2368
2369 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2370 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2371     void *tag);
2372 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2373
2374 static int
2375 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2376 {
2377         dsl_dataset_promote_arg_t *ddpa = arg;
2378         dsl_pool_t *dp = dmu_tx_pool(tx);
2379         dsl_dataset_t *hds;
2380         struct promotenode *snap;
2381         dsl_dataset_t *origin_ds;
2382         int err;
2383         uint64_t unused;
2384         uint64_t ss_mv_cnt;
2385         size_t max_snap_len;
2386
2387         err = promote_hold(ddpa, dp, FTAG);
2388         if (err != 0)
2389                 return (err);
2390
2391         hds = ddpa->ddpa_clone;
2392         max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2393
2394         if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2395                 promote_rele(ddpa, FTAG);
2396                 return (SET_ERROR(EXDEV));
2397         }
2398
2399         /*
2400          * Compute and check the amount of space to transfer.  Since this is
2401          * so expensive, don't do the preliminary check.
2402          */
2403         if (!dmu_tx_is_syncing(tx)) {
2404                 promote_rele(ddpa, FTAG);
2405                 return (0);
2406         }
2407
2408         snap = list_head(&ddpa->shared_snaps);
2409         origin_ds = snap->ds;
2410
2411         /* compute origin's new unique space */
2412         snap = list_tail(&ddpa->clone_snaps);
2413         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2414             origin_ds->ds_object);
2415         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2416             dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2417             &ddpa->unique, &unused, &unused);
2418
2419         /*
2420          * Walk the snapshots that we are moving
2421          *
2422          * Compute space to transfer.  Consider the incremental changes
2423          * to used by each snapshot:
2424          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2425          * So each snapshot gave birth to:
2426          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2427          * So a sequence would look like:
2428          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2429          * Which simplifies to:
2430          * uN + kN + kN-1 + ... + k1 + k0
2431          * Note however, if we stop before we reach the ORIGIN we get:
2432          * uN + kN + kN-1 + ... + kM - uM-1
2433          */
2434         ss_mv_cnt = 0;
2435         ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2436         ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2437         ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2438         for (snap = list_head(&ddpa->shared_snaps); snap;
2439             snap = list_next(&ddpa->shared_snaps, snap)) {
2440                 uint64_t val, dlused, dlcomp, dluncomp;
2441                 dsl_dataset_t *ds = snap->ds;
2442
2443                 ss_mv_cnt++;
2444
2445                 /*
2446                  * If there are long holds, we won't be able to evict
2447                  * the objset.
2448                  */
2449                 if (dsl_dataset_long_held(ds)) {
2450                         err = SET_ERROR(EBUSY);
2451                         goto out;
2452                 }
2453
2454                 /* Check that the snapshot name does not conflict */
2455                 VERIFY0(dsl_dataset_get_snapname(ds));
2456                 if (strlen(ds->ds_snapname) >= max_snap_len) {
2457                         err = SET_ERROR(ENAMETOOLONG);
2458                         goto out;
2459                 }
2460                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2461                 if (err == 0) {
2462                         (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2463                         err = SET_ERROR(EEXIST);
2464                         goto out;
2465                 }
2466                 if (err != ENOENT)
2467                         goto out;
2468
2469                 /* The very first snapshot does not have a deadlist */
2470                 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2471                         continue;
2472
2473                 dsl_deadlist_space(&ds->ds_deadlist,
2474                     &dlused, &dlcomp, &dluncomp);
2475                 ddpa->used += dlused;
2476                 ddpa->comp += dlcomp;
2477                 ddpa->uncomp += dluncomp;
2478         }
2479
2480         /*
2481          * If we are a clone of a clone then we never reached ORIGIN,
2482          * so we need to subtract out the clone origin's used space.
2483          */
2484         if (ddpa->origin_origin) {
2485                 ddpa->used -=
2486                     dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2487                 ddpa->comp -=
2488                     dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2489                 ddpa->uncomp -=
2490                     dsl_dataset_phys(ddpa->origin_origin)->
2491                     ds_uncompressed_bytes;
2492         }
2493
2494         /* Check that there is enough space and limit headroom here */
2495         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2496             0, ss_mv_cnt, ddpa->used, ddpa->cr);
2497         if (err != 0)
2498                 goto out;
2499
2500         /*
2501          * Compute the amounts of space that will be used by snapshots
2502          * after the promotion (for both origin and clone).  For each,
2503          * it is the amount of space that will be on all of their
2504          * deadlists (that was not born before their new origin).
2505          */
2506         if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2507                 uint64_t space;
2508
2509                 /*
2510                  * Note, typically this will not be a clone of a clone,
2511                  * so dd_origin_txg will be < TXG_INITIAL, so
2512                  * these snaplist_space() -> dsl_deadlist_space_range()
2513                  * calls will be fast because they do not have to
2514                  * iterate over all bps.
2515                  */
2516                 snap = list_head(&ddpa->origin_snaps);
2517                 err = snaplist_space(&ddpa->shared_snaps,
2518                     snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2519                 if (err != 0)
2520                         goto out;
2521
2522                 err = snaplist_space(&ddpa->clone_snaps,
2523                     snap->ds->ds_dir->dd_origin_txg, &space);
2524                 if (err != 0)
2525                         goto out;
2526                 ddpa->cloneusedsnap += space;
2527         }
2528         if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2529             DD_FLAG_USED_BREAKDOWN) {
2530                 err = snaplist_space(&ddpa->origin_snaps,
2531                     dsl_dataset_phys(origin_ds)->ds_creation_txg,
2532                     &ddpa->originusedsnap);
2533                 if (err != 0)
2534                         goto out;
2535         }
2536
2537 out:
2538         promote_rele(ddpa, FTAG);
2539         return (err);
2540 }
2541
2542 static void
2543 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2544 {
2545         dsl_dataset_promote_arg_t *ddpa = arg;
2546         dsl_pool_t *dp = dmu_tx_pool(tx);
2547         dsl_dataset_t *hds;
2548         struct promotenode *snap;
2549         dsl_dataset_t *origin_ds;
2550         dsl_dataset_t *origin_head;
2551         dsl_dir_t *dd;
2552         dsl_dir_t *odd = NULL;
2553         uint64_t oldnext_obj;
2554         int64_t delta;
2555 #if defined(__FreeBSD__) && defined(_KERNEL)
2556         char *oldname, *newname;
2557 #endif
2558
2559         VERIFY0(promote_hold(ddpa, dp, FTAG));
2560         hds = ddpa->ddpa_clone;
2561
2562         ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2563
2564         snap = list_head(&ddpa->shared_snaps);
2565         origin_ds = snap->ds;
2566         dd = hds->ds_dir;
2567
2568         snap = list_head(&ddpa->origin_snaps);
2569         origin_head = snap->ds;
2570
2571         /*
2572          * We need to explicitly open odd, since origin_ds's dd will be
2573          * changing.
2574          */
2575         VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2576             NULL, FTAG, &odd));
2577
2578         /* change origin's next snap */
2579         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2580         oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2581         snap = list_tail(&ddpa->clone_snaps);
2582         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2583             origin_ds->ds_object);
2584         dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2585
2586         /* change the origin's next clone */
2587         if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2588                 dsl_dataset_remove_from_next_clones(origin_ds,
2589                     snap->ds->ds_object, tx);
2590                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2591                     dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2592                     oldnext_obj, tx));
2593         }
2594
2595         /* change origin */
2596         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2597         ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2598         dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2599         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2600         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2601         dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2602         origin_head->ds_dir->dd_origin_txg =
2603             dsl_dataset_phys(origin_ds)->ds_creation_txg;
2604
2605         /* change dd_clone entries */
2606         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2607                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2608                     dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2609                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2610                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2611                     hds->ds_object, tx));
2612
2613                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2614                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2615                     origin_head->ds_object, tx));
2616                 if (dsl_dir_phys(dd)->dd_clones == 0) {
2617                         dsl_dir_phys(dd)->dd_clones =
2618                             zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2619                             DMU_OT_NONE, 0, tx);
2620                 }
2621                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2622                     dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2623         }
2624
2625 #if defined(__FreeBSD__) && defined(_KERNEL)
2626         /* Take the spa_namespace_lock early so zvol renames don't deadlock. */
2627         mutex_enter(&spa_namespace_lock);
2628
2629         oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2630         newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2631 #endif
2632
2633         /* move snapshots to this dir */
2634         for (snap = list_head(&ddpa->shared_snaps); snap;
2635             snap = list_next(&ddpa->shared_snaps, snap)) {
2636                 dsl_dataset_t *ds = snap->ds;
2637
2638                 /*
2639                  * Property callbacks are registered to a particular
2640                  * dsl_dir.  Since ours is changing, evict the objset
2641                  * so that they will be unregistered from the old dsl_dir.
2642                  */
2643                 if (ds->ds_objset) {
2644                         dmu_objset_evict(ds->ds_objset);
2645                         ds->ds_objset = NULL;
2646                 }
2647
2648                 /* move snap name entry */
2649                 VERIFY0(dsl_dataset_get_snapname(ds));
2650                 VERIFY0(dsl_dataset_snap_remove(origin_head,
2651                     ds->ds_snapname, tx, B_TRUE));
2652                 VERIFY0(zap_add(dp->dp_meta_objset,
2653                     dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2654                     8, 1, &ds->ds_object, tx));
2655                 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2656                     DD_FIELD_SNAPSHOT_COUNT, tx);
2657
2658                 /* change containing dsl_dir */
2659                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2660                 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2661                 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2662                 ASSERT3P(ds->ds_dir, ==, odd);
2663                 dsl_dir_rele(ds->ds_dir, ds);
2664                 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2665                     NULL, ds, &ds->ds_dir));
2666
2667 #if defined(__FreeBSD__) && defined(_KERNEL)
2668                 dsl_dataset_name(ds, newname);
2669                 zfsvfs_update_fromname(oldname, newname);
2670                 zvol_rename_minors(oldname, newname);
2671 #endif
2672
2673                 /* move any clone references */
2674                 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2675                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2676                         zap_cursor_t zc;
2677                         zap_attribute_t za;
2678
2679                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
2680                             dsl_dataset_phys(ds)->ds_next_clones_obj);
2681                             zap_cursor_retrieve(&zc, &za) == 0;
2682                             zap_cursor_advance(&zc)) {
2683                                 dsl_dataset_t *cnds;
2684                                 uint64_t o;
2685
2686                                 if (za.za_first_integer == oldnext_obj) {
2687                                         /*
2688                                          * We've already moved the
2689                                          * origin's reference.
2690                                          */
2691                                         continue;
2692                                 }
2693
2694                                 VERIFY0(dsl_dataset_hold_obj(dp,
2695                                     za.za_first_integer, FTAG, &cnds));
2696                                 o = dsl_dir_phys(cnds->ds_dir)->
2697                                     dd_head_dataset_obj;
2698
2699                                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2700                                     dsl_dir_phys(odd)->dd_clones, o, tx));
2701                                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2702                                     dsl_dir_phys(dd)->dd_clones, o, tx));
2703                                 dsl_dataset_rele(cnds, FTAG);
2704                         }
2705                         zap_cursor_fini(&zc);
2706                 }
2707
2708                 ASSERT(!dsl_prop_hascb(ds));
2709         }
2710
2711 #if defined(__FreeBSD__) && defined(_KERNEL)
2712         mutex_exit(&spa_namespace_lock);
2713
2714         kmem_free(newname, MAXPATHLEN);
2715         kmem_free(oldname, MAXPATHLEN);
2716 #endif
2717         /*
2718          * Change space accounting.
2719          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2720          * both be valid, or both be 0 (resulting in delta == 0).  This
2721          * is true for each of {clone,origin} independently.
2722          */
2723
2724         delta = ddpa->cloneusedsnap -
2725             dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2726         ASSERT3S(delta, >=, 0);
2727         ASSERT3U(ddpa->used, >=, delta);
2728         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2729         dsl_dir_diduse_space(dd, DD_USED_HEAD,
2730             ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2731
2732         delta = ddpa->originusedsnap -
2733             dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2734         ASSERT3S(delta, <=, 0);
2735         ASSERT3U(ddpa->used, >=, -delta);
2736         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2737         dsl_dir_diduse_space(odd, DD_USED_HEAD,
2738             -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2739
2740         dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2741
2742         /* log history record */
2743         spa_history_log_internal_ds(hds, "promote", tx, "");
2744
2745         dsl_dir_rele(odd, FTAG);
2746         promote_rele(ddpa, FTAG);
2747 }
2748
2749 /*
2750  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2751  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2752  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2753  * snapshots back to this dataset's origin.
2754  */
2755 static int
2756 snaplist_make(dsl_pool_t *dp,
2757     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2758 {
2759         uint64_t obj = last_obj;
2760
2761         list_create(l, sizeof (struct promotenode),
2762             offsetof(struct promotenode, link));
2763
2764         while (obj != first_obj) {
2765                 dsl_dataset_t *ds;
2766                 struct promotenode *snap;
2767                 int err;
2768
2769                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2770                 ASSERT(err != ENOENT);
2771                 if (err != 0)
2772                         return (err);
2773
2774                 if (first_obj == 0)
2775                         first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2776
2777                 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2778                 snap->ds = ds;
2779                 list_insert_tail(l, snap);
2780                 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2781         }
2782
2783         return (0);
2784 }
2785
2786 static int
2787 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2788 {
2789         struct promotenode *snap;
2790
2791         *spacep = 0;
2792         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2793                 uint64_t used, comp, uncomp;
2794                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2795                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
2796                 *spacep += used;
2797         }
2798         return (0);
2799 }
2800
2801 static void
2802 snaplist_destroy(list_t *l, void *tag)
2803 {
2804         struct promotenode *snap;
2805
2806         if (l == NULL || !list_link_active(&l->list_head))
2807                 return;
2808
2809         while ((snap = list_tail(l)) != NULL) {
2810                 list_remove(l, snap);
2811                 dsl_dataset_rele(snap->ds, tag);
2812                 kmem_free(snap, sizeof (*snap));
2813         }
2814         list_destroy(l);
2815 }
2816
2817 static int
2818 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2819 {
2820         int error;
2821         dsl_dir_t *dd;
2822         struct promotenode *snap;
2823
2824         error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2825             &ddpa->ddpa_clone);
2826         if (error != 0)
2827                 return (error);
2828         dd = ddpa->ddpa_clone->ds_dir;
2829
2830         if (ddpa->ddpa_clone->ds_is_snapshot ||
2831             !dsl_dir_is_clone(dd)) {
2832                 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2833                 return (SET_ERROR(EINVAL));
2834         }
2835
2836         error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2837             &ddpa->shared_snaps, tag);
2838         if (error != 0)
2839                 goto out;
2840
2841         error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2842             &ddpa->clone_snaps, tag);
2843         if (error != 0)
2844                 goto out;
2845
2846         snap = list_head(&ddpa->shared_snaps);
2847         ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2848         error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2849             dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2850             &ddpa->origin_snaps, tag);
2851         if (error != 0)
2852                 goto out;
2853
2854         if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2855                 error = dsl_dataset_hold_obj(dp,
2856                     dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2857                     tag, &ddpa->origin_origin);
2858                 if (error != 0)
2859                         goto out;
2860         }
2861 out:
2862         if (error != 0)
2863                 promote_rele(ddpa, tag);
2864         return (error);
2865 }
2866
2867 static void
2868 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2869 {
2870         snaplist_destroy(&ddpa->shared_snaps, tag);
2871         snaplist_destroy(&ddpa->clone_snaps, tag);
2872         snaplist_destroy(&ddpa->origin_snaps, tag);
2873         if (ddpa->origin_origin != NULL)
2874                 dsl_dataset_rele(ddpa->origin_origin, tag);
2875         dsl_dataset_rele(ddpa->ddpa_clone, tag);
2876 }
2877
2878 /*
2879  * Promote a clone.
2880  *
2881  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2882  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2883  */
2884 int
2885 dsl_dataset_promote(const char *name, char *conflsnap)
2886 {
2887         dsl_dataset_promote_arg_t ddpa = { 0 };
2888         uint64_t numsnaps;
2889         int error;
2890         objset_t *os;
2891
2892         /*
2893          * We will modify space proportional to the number of
2894          * snapshots.  Compute numsnaps.
2895          */
2896         error = dmu_objset_hold(name, FTAG, &os);
2897         if (error != 0)
2898                 return (error);
2899         error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2900             dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2901             &numsnaps);
2902         dmu_objset_rele(os, FTAG);
2903         if (error != 0)
2904                 return (error);
2905
2906         ddpa.ddpa_clonename = name;
2907         ddpa.err_ds = conflsnap;
2908         ddpa.cr = CRED();
2909
2910         return (dsl_sync_task(name, dsl_dataset_promote_check,
2911             dsl_dataset_promote_sync, &ddpa,
2912             2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2913 }
2914
2915 int
2916 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2917     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2918 {
2919         /*
2920          * "slack" factor for received datasets with refquota set on them.
2921          * See the bottom of this function for details on its use.
2922          */
2923         uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
2924         int64_t unused_refres_delta;
2925
2926         /* they should both be heads */
2927         if (clone->ds_is_snapshot ||
2928             origin_head->ds_is_snapshot)
2929                 return (SET_ERROR(EINVAL));
2930
2931         /* if we are not forcing, the branch point should be just before them */
2932         if (!force && clone->ds_prev != origin_head->ds_prev)
2933                 return (SET_ERROR(EINVAL));
2934
2935         /* clone should be the clone (unless they are unrelated) */
2936         if (clone->ds_prev != NULL &&
2937             clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2938             origin_head->ds_dir != clone->ds_prev->ds_dir)
2939                 return (SET_ERROR(EINVAL));
2940
2941         /* the clone should be a child of the origin */
2942         if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2943                 return (SET_ERROR(EINVAL));
2944
2945         /* origin_head shouldn't be modified unless 'force' */
2946         if (!force &&
2947             dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2948                 return (SET_ERROR(ETXTBSY));
2949
2950         /* origin_head should have no long holds (e.g. is not mounted) */
2951         if (dsl_dataset_handoff_check(origin_head, owner, tx))
2952                 return (SET_ERROR(EBUSY));
2953
2954         /* check amount of any unconsumed refreservation */
2955         unused_refres_delta =
2956             (int64_t)MIN(origin_head->ds_reserved,
2957             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2958             (int64_t)MIN(origin_head->ds_reserved,
2959             dsl_dataset_phys(clone)->ds_unique_bytes);
2960
2961         if (unused_refres_delta > 0 &&
2962             unused_refres_delta >
2963             dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2964                 return (SET_ERROR(ENOSPC));
2965
2966         /*
2967          * The clone can't be too much over the head's refquota.
2968          *
2969          * To ensure that the entire refquota can be used, we allow one
2970          * transaction to exceed the the refquota.  Therefore, this check
2971          * needs to also allow for the space referenced to be more than the
2972          * refquota.  The maximum amount of space that one transaction can use
2973          * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
2974          * overage ensures that we are able to receive a filesystem that
2975          * exceeds the refquota on the source system.
2976          *
2977          * So that overage is the refquota_slack we use below.
2978          */
2979         if (origin_head->ds_quota != 0 &&
2980             dsl_dataset_phys(clone)->ds_referenced_bytes >
2981             origin_head->ds_quota + refquota_slack)
2982                 return (SET_ERROR(EDQUOT));
2983
2984         return (0);
2985 }
2986
2987 void
2988 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2989     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2990 {
2991         dsl_pool_t *dp = dmu_tx_pool(tx);
2992         int64_t unused_refres_delta;
2993
2994         ASSERT(clone->ds_reserved == 0);
2995         /*
2996          * NOTE: On DEBUG kernels there could be a race between this and
2997          * the check function if spa_asize_inflation is adjusted...
2998          */
2999         ASSERT(origin_head->ds_quota == 0 ||
3000             dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
3001             DMU_MAX_ACCESS * spa_asize_inflation);
3002         ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
3003
3004         /*
3005          * Swap per-dataset feature flags.
3006          */
3007         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
3008                 if (!(spa_feature_table[f].fi_flags &
3009                     ZFEATURE_FLAG_PER_DATASET)) {
3010                         ASSERT(!clone->ds_feature_inuse[f]);
3011                         ASSERT(!origin_head->ds_feature_inuse[f]);
3012                         continue;
3013                 }
3014
3015                 boolean_t clone_inuse = clone->ds_feature_inuse[f];
3016                 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3017
3018                 if (clone_inuse) {
3019                         dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3020                         clone->ds_feature_inuse[f] = B_FALSE;
3021                 }
3022                 if (origin_head_inuse) {
3023                         dsl_dataset_deactivate_feature(origin_head->ds_object,
3024                             f, tx);
3025                         origin_head->ds_feature_inuse[f] = B_FALSE;
3026                 }
3027                 if (clone_inuse) {
3028                         dsl_dataset_activate_feature(origin_head->ds_object,
3029                             f, tx);
3030                         origin_head->ds_feature_inuse[f] = B_TRUE;
3031                 }
3032                 if (origin_head_inuse) {
3033                         dsl_dataset_activate_feature(clone->ds_object, f, tx);
3034                         clone->ds_feature_inuse[f] = B_TRUE;
3035                 }
3036         }
3037
3038         dmu_buf_will_dirty(clone->ds_dbuf, tx);
3039         dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3040
3041         if (clone->ds_objset != NULL) {
3042                 dmu_objset_evict(clone->ds_objset);
3043                 clone->ds_objset = NULL;
3044         }
3045
3046         if (origin_head->ds_objset != NULL) {
3047                 dmu_objset_evict(origin_head->ds_objset);
3048                 origin_head->ds_objset = NULL;
3049         }
3050
3051         unused_refres_delta =
3052             (int64_t)MIN(origin_head->ds_reserved,
3053             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3054             (int64_t)MIN(origin_head->ds_reserved,
3055             dsl_dataset_phys(clone)->ds_unique_bytes);
3056
3057         /*
3058          * Reset origin's unique bytes, if it exists.
3059          */
3060         if (clone->ds_prev) {
3061                 dsl_dataset_t *origin = clone->ds_prev;
3062                 uint64_t comp, uncomp;
3063
3064                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3065                 dsl_deadlist_space_range(&clone->ds_deadlist,
3066                     dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3067                     &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3068         }
3069
3070         /* swap blkptrs */
3071         {
3072                 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
3073                 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
3074                 blkptr_t tmp;
3075                 tmp = dsl_dataset_phys(origin_head)->ds_bp;
3076                 dsl_dataset_phys(origin_head)->ds_bp =
3077                     dsl_dataset_phys(clone)->ds_bp;
3078                 dsl_dataset_phys(clone)->ds_bp = tmp;
3079                 rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
3080                 rrw_exit(&clone->ds_bp_rwlock, FTAG);
3081         }
3082
3083         /* set dd_*_bytes */
3084         {
3085                 int64_t dused, dcomp, duncomp;
3086                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3087                 uint64_t odl_used, odl_comp, odl_uncomp;
3088
3089                 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3090                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
3091
3092                 dsl_deadlist_space(&clone->ds_deadlist,
3093                     &cdl_used, &cdl_comp, &cdl_uncomp);
3094                 dsl_deadlist_space(&origin_head->ds_deadlist,
3095                     &odl_used, &odl_comp, &odl_uncomp);
3096
3097                 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3098                     cdl_used -
3099                     (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3100                     odl_used);
3101                 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3102                     cdl_comp -
3103                     (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3104                     odl_comp);
3105                 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3106                     cdl_uncomp -
3107                     (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3108                     odl_uncomp);
3109
3110                 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3111                     dused, dcomp, duncomp, tx);
3112                 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3113                     -dused, -dcomp, -duncomp, tx);
3114
3115                 /*
3116                  * The difference in the space used by snapshots is the
3117                  * difference in snapshot space due to the head's
3118                  * deadlist (since that's the only thing that's
3119                  * changing that affects the snapused).
3120                  */
3121                 dsl_deadlist_space_range(&clone->ds_deadlist,
3122                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3123                     &cdl_used, &cdl_comp, &cdl_uncomp);
3124                 dsl_deadlist_space_range(&origin_head->ds_deadlist,
3125                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3126                     &odl_used, &odl_comp, &odl_uncomp);
3127                 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3128                     DD_USED_HEAD, DD_USED_SNAP, NULL);
3129         }
3130
3131         /* swap ds_*_bytes */
3132         SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3133             dsl_dataset_phys(clone)->ds_referenced_bytes);
3134         SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3135             dsl_dataset_phys(clone)->ds_compressed_bytes);
3136         SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3137             dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3138         SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3139             dsl_dataset_phys(clone)->ds_unique_bytes);
3140
3141         /* apply any parent delta for change in unconsumed refreservation */
3142         dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3143             unused_refres_delta, 0, 0, tx);
3144
3145         /*
3146          * Swap deadlists.
3147          */
3148         dsl_deadlist_close(&clone->ds_deadlist);
3149         dsl_deadlist_close(&origin_head->ds_deadlist);
3150         SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3151             dsl_dataset_phys(clone)->ds_deadlist_obj);
3152         dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3153             dsl_dataset_phys(clone)->ds_deadlist_obj);
3154         dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3155             dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3156
3157         dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3158
3159         spa_history_log_internal_ds(clone, "clone swap", tx,
3160             "parent=%s", origin_head->ds_dir->dd_myname);
3161 }
3162
3163 /*
3164  * Given a pool name and a dataset object number in that pool,
3165  * return the name of that dataset.
3166  */
3167 int
3168 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3169 {
3170         dsl_pool_t *dp;
3171         dsl_dataset_t *ds;
3172         int error;
3173
3174         error = dsl_pool_hold(pname, FTAG, &dp);
3175         if (error != 0)
3176                 return (error);
3177
3178         error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3179         if (error == 0) {
3180                 dsl_dataset_name(ds, buf);
3181                 dsl_dataset_rele(ds, FTAG);
3182         }
3183         dsl_pool_rele(dp, FTAG);
3184
3185         return (error);
3186 }
3187
3188 int
3189 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3190     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3191 {
3192         int error = 0;
3193
3194         ASSERT3S(asize, >, 0);
3195
3196         /*
3197          * *ref_rsrv is the portion of asize that will come from any
3198          * unconsumed refreservation space.
3199          */
3200         *ref_rsrv = 0;
3201
3202         mutex_enter(&ds->ds_lock);
3203         /*
3204          * Make a space adjustment for reserved bytes.
3205          */
3206         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3207                 ASSERT3U(*used, >=,
3208                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3209                 *used -=
3210                     (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3211                 *ref_rsrv =
3212                     asize - MIN(asize, parent_delta(ds, asize + inflight));
3213         }
3214
3215         if (!check_quota || ds->ds_quota == 0) {
3216                 mutex_exit(&ds->ds_lock);
3217                 return (0);
3218         }
3219         /*
3220          * If they are requesting more space, and our current estimate
3221          * is over quota, they get to try again unless the actual
3222          * on-disk is over quota and there are no pending changes (which
3223          * may free up space for us).
3224          */
3225         if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3226             ds->ds_quota) {
3227                 if (inflight > 0 ||
3228                     dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3229                         error = SET_ERROR(ERESTART);
3230                 else
3231                         error = SET_ERROR(EDQUOT);
3232         }
3233         mutex_exit(&ds->ds_lock);
3234
3235         return (error);
3236 }
3237
3238 typedef struct dsl_dataset_set_qr_arg {
3239         const char *ddsqra_name;
3240         zprop_source_t ddsqra_source;
3241         uint64_t ddsqra_value;
3242 } dsl_dataset_set_qr_arg_t;
3243
3244
3245 /* ARGSUSED */
3246 static int
3247 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3248 {
3249         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3250         dsl_pool_t *dp = dmu_tx_pool(tx);
3251         dsl_dataset_t *ds;
3252         int error;
3253         uint64_t newval;
3254
3255         if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3256                 return (SET_ERROR(ENOTSUP));
3257
3258         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3259         if (error != 0)
3260                 return (error);
3261
3262         if (ds->ds_is_snapshot) {
3263                 dsl_dataset_rele(ds, FTAG);
3264                 return (SET_ERROR(EINVAL));
3265         }
3266
3267         error = dsl_prop_predict(ds->ds_dir,
3268             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3269             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3270         if (error != 0) {
3271                 dsl_dataset_rele(ds, FTAG);
3272                 return (error);
3273         }
3274
3275         if (newval == 0) {
3276                 dsl_dataset_rele(ds, FTAG);
3277                 return (0);
3278         }
3279
3280         if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3281             newval < ds->ds_reserved) {
3282                 dsl_dataset_rele(ds, FTAG);
3283                 return (SET_ERROR(ENOSPC));
3284         }
3285
3286         dsl_dataset_rele(ds, FTAG);
3287         return (0);
3288 }
3289
3290 static void
3291 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3292 {
3293         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3294         dsl_pool_t *dp = dmu_tx_pool(tx);
3295         dsl_dataset_t *ds;
3296         uint64_t newval;
3297
3298         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3299
3300         dsl_prop_set_sync_impl(ds,
3301             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3302             ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3303             &ddsqra->ddsqra_value, tx);
3304
3305         VERIFY0(dsl_prop_get_int_ds(ds,
3306             zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3307
3308         if (ds->ds_quota != newval) {
3309                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3310                 ds->ds_quota = newval;
3311         }
3312         dsl_dataset_rele(ds, FTAG);
3313 }
3314
3315 int
3316 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3317     uint64_t refquota)
3318 {
3319         dsl_dataset_set_qr_arg_t ddsqra;
3320
3321         ddsqra.ddsqra_name = dsname;
3322         ddsqra.ddsqra_source = source;
3323         ddsqra.ddsqra_value = refquota;
3324
3325         return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3326             dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3327 }
3328
3329 static int
3330 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3331 {
3332         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3333         dsl_pool_t *dp = dmu_tx_pool(tx);
3334         dsl_dataset_t *ds;
3335         int error;
3336         uint64_t newval, unique;
3337
3338         if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3339                 return (SET_ERROR(ENOTSUP));
3340
3341         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3342         if (error != 0)
3343                 return (error);
3344
3345         if (ds->ds_is_snapshot) {
3346                 dsl_dataset_rele(ds, FTAG);
3347                 return (SET_ERROR(EINVAL));
3348         }
3349
3350         error = dsl_prop_predict(ds->ds_dir,
3351             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3352             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3353         if (error != 0) {
3354                 dsl_dataset_rele(ds, FTAG);
3355                 return (error);
3356         }
3357
3358         /*
3359          * If we are doing the preliminary check in open context, the
3360          * space estimates may be inaccurate.
3361          */
3362         if (!dmu_tx_is_syncing(tx)) {
3363                 dsl_dataset_rele(ds, FTAG);
3364                 return (0);
3365         }
3366
3367         mutex_enter(&ds->ds_lock);
3368         if (!DS_UNIQUE_IS_ACCURATE(ds))
3369                 dsl_dataset_recalc_head_uniq(ds);
3370         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3371         mutex_exit(&ds->ds_lock);
3372
3373         if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3374                 uint64_t delta = MAX(unique, newval) -
3375                     MAX(unique, ds->ds_reserved);
3376
3377                 if (delta >
3378                     dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3379                     (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3380                         dsl_dataset_rele(ds, FTAG);
3381                         return (SET_ERROR(ENOSPC));
3382                 }
3383         }
3384
3385         dsl_dataset_rele(ds, FTAG);
3386         return (0);
3387 }
3388
3389 void
3390 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3391     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3392 {
3393         uint64_t newval;
3394         uint64_t unique;
3395         int64_t delta;
3396
3397         dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3398             source, sizeof (value), 1, &value, tx);
3399
3400         VERIFY0(dsl_prop_get_int_ds(ds,
3401             zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3402
3403         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3404         mutex_enter(&ds->ds_dir->dd_lock);
3405         mutex_enter(&ds->ds_lock);
3406         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3407         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3408         delta = MAX(0, (int64_t)(newval - unique)) -
3409             MAX(0, (int64_t)(ds->ds_reserved - unique));
3410         ds->ds_reserved = newval;
3411         mutex_exit(&ds->ds_lock);
3412
3413         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3414         mutex_exit(&ds->ds_dir->dd_lock);
3415 }
3416
3417 static void
3418 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3419 {
3420         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3421         dsl_pool_t *dp = dmu_tx_pool(tx);
3422         dsl_dataset_t *ds;
3423
3424         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3425         dsl_dataset_set_refreservation_sync_impl(ds,
3426             ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3427         dsl_dataset_rele(ds, FTAG);
3428 }
3429
3430 int
3431 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3432     uint64_t refreservation)
3433 {
3434         dsl_dataset_set_qr_arg_t ddsqra;
3435
3436         ddsqra.ddsqra_name = dsname;
3437         ddsqra.ddsqra_source = source;
3438         ddsqra.ddsqra_value = refreservation;
3439
3440         return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3441             dsl_dataset_set_refreservation_sync, &ddsqra,
3442             0, ZFS_SPACE_CHECK_NONE));
3443 }
3444
3445 /*
3446  * Return (in *usedp) the amount of space written in new that is not
3447  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3448  * a snapshot before new, in new's filesystem (or its origin).  If not then
3449  * fail and return EINVAL.
3450  *
3451  * The written space is calculated by considering two components:  First, we
3452  * ignore any freed space, and calculate the written as new's used space
3453  * minus old's used space.  Next, we add in the amount of space that was freed
3454  * between the two snapshots, thus reducing new's used space relative to old's.
3455  * Specifically, this is the space that was born before old->ds_creation_txg,
3456  * and freed before new (ie. on new's deadlist or a previous deadlist).
3457  *
3458  * space freed                         [---------------------]
3459  * snapshots                       ---O-------O--------O-------O------
3460  *                                         oldsnap            new
3461  */
3462 int
3463 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3464     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3465 {
3466         int err = 0;
3467         uint64_t snapobj;
3468         dsl_pool_t *dp = new->ds_dir->dd_pool;
3469
3470         ASSERT(dsl_pool_config_held(dp));
3471
3472         *usedp = 0;
3473         *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3474         *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3475
3476         *compp = 0;
3477         *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3478         *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3479
3480         *uncompp = 0;
3481         *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3482         *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3483
3484         snapobj = new->ds_object;
3485         while (snapobj != oldsnap->ds_object) {
3486                 dsl_dataset_t *snap;
3487                 uint64_t used, comp, uncomp;
3488
3489                 if (snapobj == new->ds_object) {
3490                         snap = new;
3491                 } else {
3492                         err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3493                         if (err != 0)
3494                                 break;
3495                 }
3496
3497                 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3498                     dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3499                         /*
3500                          * The blocks in the deadlist can not be born after
3501                          * ds_prev_snap_txg, so get the whole deadlist space,
3502                          * which is more efficient (especially for old-format
3503                          * deadlists).  Unfortunately the deadlist code
3504                          * doesn't have enough information to make this
3505                          * optimization itself.
3506                          */
3507                         dsl_deadlist_space(&snap->ds_deadlist,
3508                             &used, &comp, &uncomp);
3509                 } else {
3510                         dsl_deadlist_space_range(&snap->ds_deadlist,
3511                             0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3512                             &used, &comp, &uncomp);
3513                 }
3514                 *usedp += used;
3515                 *compp += comp;
3516                 *uncompp += uncomp;
3517
3518                 /*
3519                  * If we get to the beginning of the chain of snapshots
3520                  * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3521                  * was not a snapshot of/before new.
3522                  */
3523                 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3524                 if (snap != new)
3525                         dsl_dataset_rele(snap, FTAG);
3526                 if (snapobj == 0) {
3527                         err = SET_ERROR(EINVAL);
3528                         break;
3529                 }
3530
3531         }
3532         return (err);
3533 }
3534
3535 /*
3536  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3537  * lastsnap, and all snapshots in between are deleted.
3538  *
3539  * blocks that would be freed            [---------------------------]
3540  * snapshots                       ---O-------O--------O-------O--------O
3541  *                                        firstsnap        lastsnap
3542  *
3543  * This is the set of blocks that were born after the snap before firstsnap,
3544  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3545  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3546  * We calculate this by iterating over the relevant deadlists (from the snap
3547  * after lastsnap, backward to the snap after firstsnap), summing up the
3548  * space on the deadlist that was born after the snap before firstsnap.
3549  */
3550 int
3551 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3552     dsl_dataset_t *lastsnap,
3553     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3554 {
3555         int err = 0;
3556         uint64_t snapobj;
3557         dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3558
3559         ASSERT(firstsnap->ds_is_snapshot);
3560         ASSERT(lastsnap->ds_is_snapshot);
3561
3562         /*
3563          * Check that the snapshots are in the same dsl_dir, and firstsnap
3564          * is before lastsnap.
3565          */
3566         if (firstsnap->ds_dir != lastsnap->ds_dir ||
3567             dsl_dataset_phys(firstsnap)->ds_creation_txg >
3568             dsl_dataset_phys(lastsnap)->ds_creation_txg)
3569                 return (SET_ERROR(EINVAL));
3570
3571         *usedp = *compp = *uncompp = 0;
3572
3573         snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3574         while (snapobj != firstsnap->ds_object) {
3575                 dsl_dataset_t *ds;
3576                 uint64_t used, comp, uncomp;
3577
3578                 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3579                 if (err != 0)
3580                         break;
3581
3582                 dsl_deadlist_space_range(&ds->ds_deadlist,
3583                     dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3584                     &used, &comp, &uncomp);
3585                 *usedp += used;
3586                 *compp += comp;
3587                 *uncompp += uncomp;
3588
3589                 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3590                 ASSERT3U(snapobj, !=, 0);
3591                 dsl_dataset_rele(ds, FTAG);
3592         }
3593         return (err);
3594 }
3595
3596 /*
3597  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3598  * For example, they could both be snapshots of the same filesystem, and
3599  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3600  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3601  * filesystem.  Or 'earlier' could be the origin's origin.
3602  *
3603  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3604  */
3605 boolean_t
3606 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3607     uint64_t earlier_txg)
3608 {
3609         dsl_pool_t *dp = later->ds_dir->dd_pool;
3610         int error;
3611         boolean_t ret;
3612
3613         ASSERT(dsl_pool_config_held(dp));
3614         ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3615
3616         if (earlier_txg == 0)
3617                 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3618
3619         if (later->ds_is_snapshot &&
3620             earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3621                 return (B_FALSE);
3622
3623         if (later->ds_dir == earlier->ds_dir)
3624                 return (B_TRUE);
3625         if (!dsl_dir_is_clone(later->ds_dir))
3626                 return (B_FALSE);
3627
3628         if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3629                 return (B_TRUE);
3630         dsl_dataset_t *origin;
3631         error = dsl_dataset_hold_obj(dp,
3632             dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3633         if (error != 0)
3634                 return (B_FALSE);
3635         ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3636         dsl_dataset_rele(origin, FTAG);
3637         return (ret);
3638 }
3639
3640 void
3641 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3642 {
3643         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3644         dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3645 }
3646
3647 boolean_t
3648 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3649 {
3650         dmu_object_info_t doi;
3651
3652         dmu_object_info_from_db(ds->ds_dbuf, &doi);
3653         return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3654 }
3655
3656 boolean_t
3657 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3658 {
3659         return (dsl_dataset_is_zapified(ds) &&
3660             zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3661             ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3662 }