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 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org>
  25  * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
  26  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  27  * Copyright (c) 2014 RackTop Systems.
  28  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  29  * Copyright (c) 2014 Integros [integros.com]
  30  * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
  31  * Copyright 2017 Nexenta Systems, Inc.
  32  */
  33
  34 #include <sys/dmu_objset.h>
  35 #include <sys/dsl_dataset.h>
  36 #include <sys/dsl_dir.h>
  37 #include <sys/dsl_prop.h>
  38 #include <sys/dsl_synctask.h>
  39 #include <sys/dmu_traverse.h>
  40 #include <sys/dmu_impl.h>
  41 #include <sys/dmu_send.h>
  42 #include <sys/dmu_tx.h>
  43 #include <sys/arc.h>
  44 #include <sys/zio.h>
  45 #include <sys/zap.h>
  46 #include <sys/zfeature.h>
  47 #include <sys/unique.h>
  48 #include <sys/zfs_context.h>
  49 #include <sys/zfs_ioctl.h>
  50 #include <sys/spa.h>
  51 #include <sys/zfs_znode.h>
  52 #include <sys/zfs_onexit.h>
  53 #include <sys/zvol.h>
  54 #include <sys/dsl_scan.h>
  55 #include <sys/dsl_deadlist.h>
  56 #include <sys/dsl_destroy.h>
  57 #include <sys/dsl_userhold.h>
  58 #include <sys/dsl_bookmark.h>
  59 #include <sys/dmu_send.h>
  60 #include <sys/zio_checksum.h>
  61 #include <sys/zio_compress.h>
  62 #include <zfs_fletcher.h>
  63
  64 SYSCTL_DECL(_vfs_zfs);
  65
  66 /*
  67  * The SPA supports block sizes up to 16MB.  However, very large blocks
  68  * can have an impact on i/o latency (e.g. tying up a spinning disk for
  69  * ~300ms), and also potentially on the memory allocator.  Therefore,
  70  * we do not allow the recordsize to be set larger than zfs_max_recordsize
  71  * (default 1MB).  Larger blocks can be created by changing this tunable,
  72  * and pools with larger blocks can always be imported and used, regardless
  73  * of this setting.
  74  */
  75 int zfs_max_recordsize = 1 * 1024 * 1024;
  76 SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN,
  77     &zfs_max_recordsize, 0,
  78     "Maximum block size.  Expect dragons when tuning this.");
  79
  80 #define SWITCH64(x, y) \
  81         { \
  82                 uint64_t __tmp = (x); \
  83                 (x) = (y); \
  84                 (y) = __tmp; \
  85         }
  86
  87 #define DS_REF_MAX      (1ULL << 62)
  88
  89 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
  90
  91 extern int spa_asize_inflation;
  92
  93 static zil_header_t zero_zil;
  94
  95 /*
  96  * Figure out how much of this delta should be propogated to the dsl_dir
  97  * layer.  If there's a refreservation, that space has already been
  98  * partially accounted for in our ancestors.
  99  */
 100 static int64_t
 101 parent_delta(dsl_dataset_t *ds, int64_t delta)
 102 {
 103         dsl_dataset_phys_t *ds_phys;
 104         uint64_t old_bytes, new_bytes;
 105
 106         if (ds->ds_reserved == 0)
 107                 return (delta);
 108
 109         ds_phys = dsl_dataset_phys(ds);
 110         old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
 111         new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
 112
 113         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
 114         return (new_bytes - old_bytes);
 115 }
 116
 117 void
 118 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
 119 {
 120         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 121         int compressed = BP_GET_PSIZE(bp);
 122         int uncompressed = BP_GET_UCSIZE(bp);
 123         int64_t delta;
 124
 125         dprintf_bp(bp, "ds=%p", ds);
 126
 127         ASSERT(dmu_tx_is_syncing(tx));
 128         /* It could have been compressed away to nothing */
 129         if (BP_IS_HOLE(bp))
 130                 return;
 131         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 132         ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 133         if (ds == NULL) {
 134                 dsl_pool_mos_diduse_space(tx->tx_pool,
 135                     used, compressed, uncompressed);
 136                 return;
 137         }
 138
 139         ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
 140         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 141         mutex_enter(&ds->ds_lock);
 142         delta = parent_delta(ds, used);
 143         dsl_dataset_phys(ds)->ds_referenced_bytes += used;
 144         dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
 145         dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
 146         dsl_dataset_phys(ds)->ds_unique_bytes += used;
 147
 148         if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
 149                 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
 150                     B_TRUE;
 151         }
 152
 153         spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
 154         if (f != SPA_FEATURE_NONE)
 155                 ds->ds_feature_activation_needed[f] = B_TRUE;
 156
 157         mutex_exit(&ds->ds_lock);
 158         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 159             compressed, uncompressed, tx);
 160         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 161             DD_USED_REFRSRV, DD_USED_HEAD, NULL);
 162 }
 163
 164 int
 165 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 166     boolean_t async)
 167 {
 168         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 169         int compressed = BP_GET_PSIZE(bp);
 170         int uncompressed = BP_GET_UCSIZE(bp);
 171
 172         if (BP_IS_HOLE(bp))
 173                 return (0);
 174
 175         ASSERT(dmu_tx_is_syncing(tx));
 176         ASSERT(bp->blk_birth <= tx->tx_txg);
 177
 178         if (ds == NULL) {
 179                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 180                 dsl_pool_mos_diduse_space(tx->tx_pool,
 181                     -used, -compressed, -uncompressed);
 182                 return (used);
 183         }
 184         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 185
 186         ASSERT(!ds->ds_is_snapshot);
 187         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 188
 189         if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
 190                 int64_t delta;
 191
 192                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 193                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 194
 195                 mutex_enter(&ds->ds_lock);
 196                 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
 197                     !DS_UNIQUE_IS_ACCURATE(ds));
 198                 delta = parent_delta(ds, -used);
 199                 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
 200                 mutex_exit(&ds->ds_lock);
 201                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 202                     delta, -compressed, -uncompressed, tx);
 203                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 204                     DD_USED_REFRSRV, DD_USED_HEAD, NULL);
 205         } else {
 206                 dprintf_bp(bp, "putting on dead list: %s", "");
 207                 if (async) {
 208                         /*
 209                          * We are here as part of zio's write done callback,
 210                          * which means we're a zio interrupt thread.  We can't
 211                          * call dsl_deadlist_insert() now because it may block
 212                          * waiting for I/O.  Instead, put bp on the deferred
 213                          * queue and let dsl_pool_sync() finish the job.
 214                          */
 215                         bplist_append(&ds->ds_pending_deadlist, bp);
 216                 } else {
 217                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 218                 }
 219                 ASSERT3U(ds->ds_prev->ds_object, ==,
 220                     dsl_dataset_phys(ds)->ds_prev_snap_obj);
 221                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
 222                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 223                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
 224                     ds->ds_object && bp->blk_birth >
 225                     dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
 226                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 227                         mutex_enter(&ds->ds_prev->ds_lock);
 228                         dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
 229                         mutex_exit(&ds->ds_prev->ds_lock);
 230                 }
 231                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 232                         dsl_dir_transfer_space(ds->ds_dir, used,
 233                             DD_USED_HEAD, DD_USED_SNAP, tx);
 234                 }
 235         }
 236         mutex_enter(&ds->ds_lock);
 237         ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
 238         dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
 239         ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
 240         dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
 241         ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
 242         dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
 243         mutex_exit(&ds->ds_lock);
 244
 245         return (used);
 246 }
 247
 248 /*
 249  * We have to release the fsid syncronously or we risk that a subsequent
 250  * mount of the same dataset will fail to unique_insert the fsid.  This
 251  * failure would manifest itself as the fsid of this dataset changing
 252  * between mounts which makes NFS clients quite unhappy.
 253  */
 254 static void
 255 dsl_dataset_evict_sync(void *dbu)
 256 {
 257         dsl_dataset_t *ds = dbu;
 258
 259         ASSERT(ds->ds_owner == NULL);
 260
 261         unique_remove(ds->ds_fsid_guid);
 262 }
 263
 264 static void
 265 dsl_dataset_evict_async(void *dbu)
 266 {
 267         dsl_dataset_t *ds = dbu;
 268
 269         ASSERT(ds->ds_owner == NULL);
 270
 271         ds->ds_dbuf = NULL;
 272
 273         if (ds->ds_objset != NULL)
 274                 dmu_objset_evict(ds->ds_objset);
 275
 276         if (ds->ds_prev) {
 277                 dsl_dataset_rele(ds->ds_prev, ds);
 278                 ds->ds_prev = NULL;
 279         }
 280
 281         bplist_destroy(&ds->ds_pending_deadlist);
 282         if (ds->ds_deadlist.dl_os != NULL)
 283                 dsl_deadlist_close(&ds->ds_deadlist);
 284         if (ds->ds_dir)
 285                 dsl_dir_async_rele(ds->ds_dir, ds);
 286
 287         ASSERT(!list_link_active(&ds->ds_synced_link));
 288
 289         list_destroy(&ds->ds_prop_cbs);
 290         if (mutex_owned(&ds->ds_lock))
 291                 mutex_exit(&ds->ds_lock);
 292         mutex_destroy(&ds->ds_lock);
 293         if (mutex_owned(&ds->ds_opening_lock))
 294                 mutex_exit(&ds->ds_opening_lock);
 295         mutex_destroy(&ds->ds_opening_lock);
 296         mutex_destroy(&ds->ds_sendstream_lock);
 297         refcount_destroy(&ds->ds_longholds);
 298         rrw_destroy(&ds->ds_bp_rwlock);
 299
 300         kmem_free(ds, sizeof (dsl_dataset_t));
 301 }
 302
 303 int
 304 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 305 {
 306         dsl_dataset_phys_t *headphys;
 307         int err;
 308         dmu_buf_t *headdbuf;
 309         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 310         objset_t *mos = dp->dp_meta_objset;
 311
 312         if (ds->ds_snapname[0])
 313                 return (0);
 314         if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
 315                 return (0);
 316
 317         err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
 318             FTAG, &headdbuf);
 319         if (err != 0)
 320                 return (err);
 321         headphys = headdbuf->db_data;
 322         err = zap_value_search(dp->dp_meta_objset,
 323             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 324         dmu_buf_rele(headdbuf, FTAG);
 325         return (err);
 326 }
 327
 328 int
 329 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 330 {
 331         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 332         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 333         matchtype_t mt = 0;
 334         int err;
 335
 336         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 337                 mt = MT_NORMALIZE;
 338
 339         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 340             value, mt, NULL, 0, NULL);
 341         if (err == ENOTSUP && (mt & MT_NORMALIZE))
 342                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 343         return (err);
 344 }
 345
 346 int
 347 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
 348     boolean_t adj_cnt)
 349 {
 350         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 351         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 352         matchtype_t mt = 0;
 353         int err;
 354
 355         dsl_dir_snap_cmtime_update(ds->ds_dir);
 356
 357         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 358                 mt = MT_NORMALIZE;
 359
 360         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 361         if (err == ENOTSUP && (mt & MT_NORMALIZE))
 362                 err = zap_remove(mos, snapobj, name, tx);
 363
 364         if (err == 0 && adj_cnt)
 365                 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
 366                     DD_FIELD_SNAPSHOT_COUNT, tx);
 367
 368         return (err);
 369 }
 370
 371 boolean_t
 372 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
 373 {
 374         dmu_buf_t *dbuf = ds->ds_dbuf;
 375         boolean_t result = B_FALSE;
 376
 377         if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
 378             ds->ds_object, DMU_BONUS_BLKID, tag)) {
 379
 380                 if (ds == dmu_buf_get_user(dbuf))
 381                         result = B_TRUE;
 382                 else
 383                         dmu_buf_rele(dbuf, tag);
 384         }
 385
 386         return (result);
 387 }
 388
 389 int
 390 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 391     dsl_dataset_t **dsp)
 392 {
 393         objset_t *mos = dp->dp_meta_objset;
 394         dmu_buf_t *dbuf;
 395         dsl_dataset_t *ds;
 396         int err;
 397         dmu_object_info_t doi;
 398
 399         ASSERT(dsl_pool_config_held(dp));
 400
 401         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 402         if (err != 0)
 403                 return (err);
 404
 405         /* Make sure dsobj has the correct object type. */
 406         dmu_object_info_from_db(dbuf, &doi);
 407         if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
 408                 dmu_buf_rele(dbuf, tag);
 409                 return (SET_ERROR(EINVAL));
 410         }
 411
 412         ds = dmu_buf_get_user(dbuf);
 413         if (ds == NULL) {
 414                 dsl_dataset_t *winner = NULL;
 415
 416                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 417                 ds->ds_dbuf = dbuf;
 418                 ds->ds_object = dsobj;
 419                 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
 420
 421                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 422                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 423                 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
 424                 rrw_init(&ds->ds_bp_rwlock, B_FALSE);
 425                 refcount_create(&ds->ds_longholds);
 426
 427                 bplist_create(&ds->ds_pending_deadlist);
 428                 dsl_deadlist_open(&ds->ds_deadlist,
 429                     mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
 430
 431                 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
 432                     offsetof(dmu_sendarg_t, dsa_link));
 433
 434                 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
 435                     offsetof(dsl_prop_cb_record_t, cbr_ds_node));
 436
 437                 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 438                         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 439                                 if (!(spa_feature_table[f].fi_flags &
 440                                     ZFEATURE_FLAG_PER_DATASET))
 441                                         continue;
 442                                 err = zap_contains(mos, dsobj,
 443                                     spa_feature_table[f].fi_guid);
 444                                 if (err == 0) {
 445                                         ds->ds_feature_inuse[f] = B_TRUE;
 446                                 } else {
 447                                         ASSERT3U(err, ==, ENOENT);
 448                                         err = 0;
 449                                 }
 450                         }
 451                 }
 452
 453                 err = dsl_dir_hold_obj(dp,
 454                     dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
 455                 if (err != 0) {
 456                         mutex_destroy(&ds->ds_lock);
 457                         mutex_destroy(&ds->ds_opening_lock);
 458                         mutex_destroy(&ds->ds_sendstream_lock);
 459                         refcount_destroy(&ds->ds_longholds);
 460                         bplist_destroy(&ds->ds_pending_deadlist);
 461                         dsl_deadlist_close(&ds->ds_deadlist);
 462                         kmem_free(ds, sizeof (dsl_dataset_t));
 463                         dmu_buf_rele(dbuf, tag);
 464                         return (err);
 465                 }
 466
 467                 if (!ds->ds_is_snapshot) {
 468                         ds->ds_snapname[0] = '\0';
 469                         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 470                                 err = dsl_dataset_hold_obj(dp,
 471                                     dsl_dataset_phys(ds)->ds_prev_snap_obj,
 472                                     ds, &ds->ds_prev);
 473                         }
 474                         if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 475                                 int zaperr = zap_lookup(mos, ds->ds_object,
 476                                     DS_FIELD_BOOKMARK_NAMES,
 477                                     sizeof (ds->ds_bookmarks), 1,
 478                                     &ds->ds_bookmarks);
 479                                 if (zaperr != ENOENT)
 480                                         VERIFY0(zaperr);
 481                         }
 482                 } else {
 483                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 484                                 err = dsl_dataset_get_snapname(ds);
 485                         if (err == 0 &&
 486                             dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
 487                                 err = zap_count(
 488                                     ds->ds_dir->dd_pool->dp_meta_objset,
 489                                     dsl_dataset_phys(ds)->ds_userrefs_obj,
 490                                     &ds->ds_userrefs);
 491                         }
 492                 }
 493
 494                 if (err == 0 && !ds->ds_is_snapshot) {
 495                         err = dsl_prop_get_int_ds(ds,
 496                             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
 497                             &ds->ds_reserved);
 498                         if (err == 0) {
 499                                 err = dsl_prop_get_int_ds(ds,
 500                                     zfs_prop_to_name(ZFS_PROP_REFQUOTA),
 501                                     &ds->ds_quota);
 502                         }
 503                 } else {
 504                         ds->ds_reserved = ds->ds_quota = 0;
 505                 }
 506
 507                 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
 508                     dsl_dataset_evict_async, &ds->ds_dbuf);
 509                 if (err == 0)
 510                         winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
 511
 512                 if (err != 0 || winner != NULL) {
 513                         bplist_destroy(&ds->ds_pending_deadlist);
 514                         dsl_deadlist_close(&ds->ds_deadlist);
 515                         if (ds->ds_prev)
 516                                 dsl_dataset_rele(ds->ds_prev, ds);
 517                         dsl_dir_rele(ds->ds_dir, ds);
 518                         mutex_destroy(&ds->ds_lock);
 519                         mutex_destroy(&ds->ds_opening_lock);
 520                         mutex_destroy(&ds->ds_sendstream_lock);
 521                         refcount_destroy(&ds->ds_longholds);
 522                         kmem_free(ds, sizeof (dsl_dataset_t));
 523                         if (err != 0) {
 524                                 dmu_buf_rele(dbuf, tag);
 525                                 return (err);
 526                         }
 527                         ds = winner;
 528                 } else {
 529                         ds->ds_fsid_guid =
 530                             unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
 531                         if (ds->ds_fsid_guid !=
 532                             dsl_dataset_phys(ds)->ds_fsid_guid) {
 533                                 zfs_dbgmsg("ds_fsid_guid changed from "
 534                                     "%llx to %llx for pool %s dataset id %llu",
 535                                     (long long)
 536                                     dsl_dataset_phys(ds)->ds_fsid_guid,
 537                                     (long long)ds->ds_fsid_guid,
 538                                     spa_name(dp->dp_spa),
 539                                     dsobj);
 540                         }
 541                 }
 542         }
 543         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 544         ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
 545         ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
 546             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 547             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 548         *dsp = ds;
 549         return (0);
 550 }
 551
 552 int
 553 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
 554     void *tag, dsl_dataset_t **dsp)
 555 {
 556         dsl_dir_t *dd;
 557         const char *snapname;
 558         uint64_t obj;
 559         int err = 0;
 560         dsl_dataset_t *ds;
 561
 562         err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
 563         if (err != 0)
 564                 return (err);
 565
 566         ASSERT(dsl_pool_config_held(dp));
 567         obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
 568         if (obj != 0)
 569                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
 570         else
 571                 err = SET_ERROR(ENOENT);
 572
 573         /* we may be looking for a snapshot */
 574         if (err == 0 && snapname != NULL) {
 575                 dsl_dataset_t *snap_ds;
 576
 577                 if (*snapname++ != '@') {
 578                         dsl_dataset_rele(ds, tag);
 579                         dsl_dir_rele(dd, FTAG);
 580                         return (SET_ERROR(ENOENT));
 581                 }
 582
 583                 dprintf("looking for snapshot '%s'\n", snapname);
 584                 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
 585                 if (err == 0)
 586                         err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
 587                 dsl_dataset_rele(ds, tag);
 588
 589                 if (err == 0) {
 590                         mutex_enter(&snap_ds->ds_lock);
 591                         if (snap_ds->ds_snapname[0] == 0)
 592                                 (void) strlcpy(snap_ds->ds_snapname, snapname,
 593                                     sizeof (snap_ds->ds_snapname));
 594                         mutex_exit(&snap_ds->ds_lock);
 595                         ds = snap_ds;
 596                 }
 597         }
 598         if (err == 0)
 599                 *dsp = ds;
 600         dsl_dir_rele(dd, FTAG);
 601         return (err);
 602 }
 603
 604 int
 605 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
 606     void *tag, dsl_dataset_t **dsp)
 607 {
 608         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 609         if (err != 0)
 610                 return (err);
 611         if (!dsl_dataset_tryown(*dsp, tag)) {
 612                 dsl_dataset_rele(*dsp, tag);
 613                 *dsp = NULL;
 614                 return (SET_ERROR(EBUSY));
 615         }
 616         return (0);
 617 }
 618
 619 int
 620 dsl_dataset_own(dsl_pool_t *dp, const char *name,
 621     void *tag, dsl_dataset_t **dsp)
 622 {
 623         int err = dsl_dataset_hold(dp, name, tag, dsp);
 624         if (err != 0)
 625                 return (err);
 626         if (!dsl_dataset_tryown(*dsp, tag)) {
 627                 dsl_dataset_rele(*dsp, tag);
 628                 return (SET_ERROR(EBUSY));
 629         }
 630         return (0);
 631 }
 632
 633 /*
 634  * See the comment above dsl_pool_hold() for details.  In summary, a long
 635  * hold is used to prevent destruction of a dataset while the pool hold
 636  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
 637  *
 638  * The dataset and pool must be held when this function is called.  After it
 639  * is called, the pool hold may be released while the dataset is still held
 640  * and accessed.
 641  */
 642 void
 643 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
 644 {
 645         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 646         (void) refcount_add(&ds->ds_longholds, tag);
 647 }
 648
 649 void
 650 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
 651 {
 652         (void) refcount_remove(&ds->ds_longholds, tag);
 653 }
 654
 655 /* Return B_TRUE if there are any long holds on this dataset. */
 656 boolean_t
 657 dsl_dataset_long_held(dsl_dataset_t *ds)
 658 {
 659         return (!refcount_is_zero(&ds->ds_longholds));
 660 }
 661
 662 void
 663 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 664 {
 665         if (ds == NULL) {
 666                 (void) strcpy(name, "mos");
 667         } else {
 668                 dsl_dir_name(ds->ds_dir, name);
 669                 VERIFY0(dsl_dataset_get_snapname(ds));
 670                 if (ds->ds_snapname[0]) {
 671                         VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
 672                             <, ZFS_MAX_DATASET_NAME_LEN);
 673                         /*
 674                          * We use a "recursive" mutex so that we
 675                          * can call dprintf_ds() with ds_lock held.
 676                          */
 677                         if (!MUTEX_HELD(&ds->ds_lock)) {
 678                                 mutex_enter(&ds->ds_lock);
 679                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 680                                     ZFS_MAX_DATASET_NAME_LEN), <,
 681                                     ZFS_MAX_DATASET_NAME_LEN);
 682                                 mutex_exit(&ds->ds_lock);
 683                         } else {
 684                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 685                                     ZFS_MAX_DATASET_NAME_LEN), <,
 686                                     ZFS_MAX_DATASET_NAME_LEN);
 687                         }
 688                 }
 689         }
 690 }
 691
 692 int
 693 dsl_dataset_namelen(dsl_dataset_t *ds)
 694 {
 695         VERIFY0(dsl_dataset_get_snapname(ds));
 696         mutex_enter(&ds->ds_lock);
 697         int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
 698         mutex_exit(&ds->ds_lock);
 699         return (len);
 700 }
 701
 702 void
 703 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 704 {
 705         dmu_buf_rele(ds->ds_dbuf, tag);
 706 }
 707
 708 void
 709 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 710 {
 711         ASSERT3P(ds->ds_owner, ==, tag);
 712         ASSERT(ds->ds_dbuf != NULL);
 713
 714         mutex_enter(&ds->ds_lock);
 715         ds->ds_owner = NULL;
 716         mutex_exit(&ds->ds_lock);
 717         dsl_dataset_long_rele(ds, tag);
 718         dsl_dataset_rele(ds, tag);
 719 }
 720
 721 boolean_t
 722 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
 723 {
 724         boolean_t gotit = FALSE;
 725
 726         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 727         mutex_enter(&ds->ds_lock);
 728         if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
 729                 ds->ds_owner = tag;
 730                 dsl_dataset_long_hold(ds, tag);
 731                 gotit = TRUE;
 732         }
 733         mutex_exit(&ds->ds_lock);
 734         return (gotit);
 735 }
 736
 737 boolean_t
 738 dsl_dataset_has_owner(dsl_dataset_t *ds)
 739 {
 740         boolean_t rv;
 741         mutex_enter(&ds->ds_lock);
 742         rv = (ds->ds_owner != NULL);
 743         mutex_exit(&ds->ds_lock);
 744         return (rv);
 745 }
 746
 747 static void
 748 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 749 {
 750         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 751         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 752         uint64_t zero = 0;
 753
 754         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 755
 756         spa_feature_incr(spa, f, tx);
 757         dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
 758
 759         VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
 760             sizeof (zero), 1, &zero, tx));
 761 }
 762
 763 void
 764 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 765 {
 766         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 767         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 768
 769         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 770
 771         VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
 772         spa_feature_decr(spa, f, tx);
 773 }
 774
 775 uint64_t
 776 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 777     uint64_t flags, dmu_tx_t *tx)
 778 {
 779         dsl_pool_t *dp = dd->dd_pool;
 780         dmu_buf_t *dbuf;
 781         dsl_dataset_phys_t *dsphys;
 782         uint64_t dsobj;
 783         objset_t *mos = dp->dp_meta_objset;
 784
 785         if (origin == NULL)
 786                 origin = dp->dp_origin_snap;
 787
 788         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 789         ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
 790         ASSERT(dmu_tx_is_syncing(tx));
 791         ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
 792
 793         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 794             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 795         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 796         dmu_buf_will_dirty(dbuf, tx);
 797         dsphys = dbuf->db_data;
 798         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 799         dsphys->ds_dir_obj = dd->dd_object;
 800         dsphys->ds_flags = flags;
 801         dsphys->ds_fsid_guid = unique_create();
 802         do {
 803                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 804                     sizeof (dsphys->ds_guid));
 805         } while (dsphys->ds_guid == 0);
 806         dsphys->ds_snapnames_zapobj =
 807             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 808             DMU_OT_NONE, 0, tx);
 809         dsphys->ds_creation_time = gethrestime_sec();
 810         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 811
 812         if (origin == NULL) {
 813                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 814         } else {
 815                 dsl_dataset_t *ohds; /* head of the origin snapshot */
 816
 817                 dsphys->ds_prev_snap_obj = origin->ds_object;
 818                 dsphys->ds_prev_snap_txg =
 819                     dsl_dataset_phys(origin)->ds_creation_txg;
 820                 dsphys->ds_referenced_bytes =
 821                     dsl_dataset_phys(origin)->ds_referenced_bytes;
 822                 dsphys->ds_compressed_bytes =
 823                     dsl_dataset_phys(origin)->ds_compressed_bytes;
 824                 dsphys->ds_uncompressed_bytes =
 825                     dsl_dataset_phys(origin)->ds_uncompressed_bytes;
 826                 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
 827                 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
 828                 rrw_exit(&origin->ds_bp_rwlock, FTAG);
 829
 830                 /*
 831                  * Inherit flags that describe the dataset's contents
 832                  * (INCONSISTENT) or properties (Case Insensitive).
 833                  */
 834                 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
 835                     (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
 836
 837                 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 838                         if (origin->ds_feature_inuse[f])
 839                                 dsl_dataset_activate_feature(dsobj, f, tx);
 840                 }
 841
 842                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 843                 dsl_dataset_phys(origin)->ds_num_children++;
 844
 845                 VERIFY0(dsl_dataset_hold_obj(dp,
 846                     dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
 847                     FTAG, &ohds));
 848                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 849                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 850                 dsl_dataset_rele(ohds, FTAG);
 851
 852                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 853                         if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
 854                                 dsl_dataset_phys(origin)->ds_next_clones_obj =
 855                                     zap_create(mos,
 856                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 857                         }
 858                         VERIFY0(zap_add_int(mos,
 859                             dsl_dataset_phys(origin)->ds_next_clones_obj,
 860                             dsobj, tx));
 861                 }
 862
 863                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 864                 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
 865                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 866                         if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
 867                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 868                                 dsl_dir_phys(origin->ds_dir)->dd_clones =
 869                                     zap_create(mos,
 870                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 871                         }
 872                         VERIFY0(zap_add_int(mos,
 873                             dsl_dir_phys(origin->ds_dir)->dd_clones,
 874                             dsobj, tx));
 875                 }
 876         }
 877
 878         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 879                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 880
 881         dmu_buf_rele(dbuf, FTAG);
 882
 883         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 884         dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
 885
 886         return (dsobj);
 887 }
 888
 889 static void
 890 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
 891 {
 892         objset_t *os;
 893
 894         VERIFY0(dmu_objset_from_ds(ds, &os));
 895         if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) {
 896                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
 897                 zio_t *zio;
 898
 899                 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 900
 901                 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
 902                 dsl_dataset_sync(ds, zio, tx);
 903                 VERIFY0(zio_wait(zio));
 904
 905                 /* dsl_dataset_sync_done will drop this reference. */
 906                 dmu_buf_add_ref(ds->ds_dbuf, ds);
 907                 dsl_dataset_sync_done(ds, tx);
 908         }
 909 }
 910
 911 uint64_t
 912 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 913     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 914 {
 915         dsl_pool_t *dp = pdd->dd_pool;
 916         uint64_t dsobj, ddobj;
 917         dsl_dir_t *dd;
 918
 919         ASSERT(dmu_tx_is_syncing(tx));
 920         ASSERT(lastname[0] != '@');
 921
 922         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 923         VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
 924
 925         dsobj = dsl_dataset_create_sync_dd(dd, origin,
 926             flags & ~DS_CREATE_FLAG_NODIRTY, tx);
 927
 928         dsl_deleg_set_create_perms(dd, tx, cr);
 929
 930         /*
 931          * Since we're creating a new node we know it's a leaf, so we can
 932          * initialize the counts if the limit feature is active.
 933          */
 934         if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
 935                 uint64_t cnt = 0;
 936                 objset_t *os = dd->dd_pool->dp_meta_objset;
 937
 938                 dsl_dir_zapify(dd, tx);
 939                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
 940                     sizeof (cnt), 1, &cnt, tx));
 941                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
 942                     sizeof (cnt), 1, &cnt, tx));
 943         }
 944
 945         dsl_dir_rele(dd, FTAG);
 946
 947         /*
 948          * If we are creating a clone, make sure we zero out any stale
 949          * data from the origin snapshots zil header.
 950          */
 951         if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
 952                 dsl_dataset_t *ds;
 953
 954                 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 955                 dsl_dataset_zero_zil(ds, tx);
 956                 dsl_dataset_rele(ds, FTAG);
 957         }
 958
 959         return (dsobj);
 960 }
 961
 962 #ifdef __FreeBSD__
 963 /* FreeBSD ioctl compat begin */
 964 struct destroyarg {
 965         nvlist_t *nvl;
 966         const char *snapname;
 967 };
 968
 969 static int
 970 dsl_check_snap_cb(const char *name, void *arg)
 971 {
 972         struct destroyarg *da = arg;
 973         dsl_dataset_t *ds;
 974         char *dsname;
 975
 976         dsname = kmem_asprintf("%s@%s", name, da->snapname);
 977         fnvlist_add_boolean(da->nvl, dsname);
 978         kmem_free(dsname, strlen(dsname) + 1);
 979
 980         return (0);
 981 }
 982
 983 int
 984 dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
 985     nvlist_t *snaps)
 986 {
 987         struct destroyarg *da;
 988         int err;
 989
 990         da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
 991         da->nvl = snaps;
 992         da->snapname = snapname;
 993         err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
 994             DS_FIND_CHILDREN);
 995         kmem_free(da, sizeof (struct destroyarg));
 996
 997         return (err);
 998 }
 999 /* FreeBSD ioctl compat end */
1000 #endif /* __FreeBSD__ */
1001
1002 /*
1003  * The unique space in the head dataset can be calculated by subtracting
1004  * the space used in the most recent snapshot, that is still being used
1005  * in this file system, from the space currently in use.  To figure out
1006  * the space in the most recent snapshot still in use, we need to take
1007  * the total space used in the snapshot and subtract out the space that
1008  * has been freed up since the snapshot was taken.
1009  */
1010 void
1011 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1012 {
1013         uint64_t mrs_used;
1014         uint64_t dlused, dlcomp, dluncomp;
1015
1016         ASSERT(!ds->ds_is_snapshot);
1017
1018         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1019                 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1020         else
1021                 mrs_used = 0;
1022
1023         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1024
1025         ASSERT3U(dlused, <=, mrs_used);
1026         dsl_dataset_phys(ds)->ds_unique_bytes =
1027             dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1028
1029         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1030             SPA_VERSION_UNIQUE_ACCURATE)
1031                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1032 }
1033
1034 void
1035 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1036     dmu_tx_t *tx)
1037 {
1038         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1039         uint64_t count;
1040         int err;
1041
1042         ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1043         err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1044             obj, tx);
1045         /*
1046          * The err should not be ENOENT, but a bug in a previous version
1047          * of the code could cause upgrade_clones_cb() to not set
1048          * ds_next_snap_obj when it should, leading to a missing entry.
1049          * If we knew that the pool was created after
1050          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1051          * ENOENT.  However, at least we can check that we don't have
1052          * too many entries in the next_clones_obj even after failing to
1053          * remove this one.
1054          */
1055         if (err != ENOENT)
1056                 VERIFY0(err);
1057         ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1058             &count));
1059         ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1060 }
1061
1062
1063 blkptr_t *
1064 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1065 {
1066         return (&dsl_dataset_phys(ds)->ds_bp);
1067 }
1068
1069 spa_t *
1070 dsl_dataset_get_spa(dsl_dataset_t *ds)
1071 {
1072         return (ds->ds_dir->dd_pool->dp_spa);
1073 }
1074
1075 void
1076 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1077 {
1078         dsl_pool_t *dp;
1079
1080         if (ds == NULL) /* this is the meta-objset */
1081                 return;
1082
1083         ASSERT(ds->ds_objset != NULL);
1084
1085         if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1086                 panic("dirtying snapshot!");
1087
1088         /* Must not dirty a dataset in the same txg where it got snapshotted. */
1089         ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
1090
1091         dp = ds->ds_dir->dd_pool;
1092         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1093                 /* up the hold count until we can be written out */
1094                 dmu_buf_add_ref(ds->ds_dbuf, ds);
1095         }
1096 }
1097
1098 boolean_t
1099 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1100 {
1101         for (int t = 0; t < TXG_SIZE; t++) {
1102                 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1103                     ds, t))
1104                         return (B_TRUE);
1105         }
1106         return (B_FALSE);
1107 }
1108
1109 static int
1110 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1111 {
1112         uint64_t asize;
1113
1114         if (!dmu_tx_is_syncing(tx))
1115                 return (0);
1116
1117         /*
1118          * If there's an fs-only reservation, any blocks that might become
1119          * owned by the snapshot dataset must be accommodated by space
1120          * outside of the reservation.
1121          */
1122         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1123         asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1124         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1125                 return (SET_ERROR(ENOSPC));
1126
1127         /*
1128          * Propagate any reserved space for this snapshot to other
1129          * snapshot checks in this sync group.
1130          */
1131         if (asize > 0)
1132                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1133
1134         return (0);
1135 }
1136
1137 typedef struct dsl_dataset_snapshot_arg {
1138         nvlist_t *ddsa_snaps;
1139         nvlist_t *ddsa_props;
1140         nvlist_t *ddsa_errors;
1141         cred_t *ddsa_cr;
1142 } dsl_dataset_snapshot_arg_t;
1143
1144 int
1145 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1146     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1147 {
1148         int error;
1149         uint64_t value;
1150
1151         ds->ds_trysnap_txg = tx->tx_txg;
1152
1153         if (!dmu_tx_is_syncing(tx))
1154                 return (0);
1155
1156         /*
1157          * We don't allow multiple snapshots of the same txg.  If there
1158          * is already one, try again.
1159          */
1160         if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1161                 return (SET_ERROR(EAGAIN));
1162
1163         /*
1164          * Check for conflicting snapshot name.
1165          */
1166         error = dsl_dataset_snap_lookup(ds, snapname, &value);
1167         if (error == 0)
1168                 return (SET_ERROR(EEXIST));
1169         if (error != ENOENT)
1170                 return (error);
1171
1172         /*
1173          * We don't allow taking snapshots of inconsistent datasets, such as
1174          * those into which we are currently receiving.  However, if we are
1175          * creating this snapshot as part of a receive, this check will be
1176          * executed atomically with respect to the completion of the receive
1177          * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1178          * case we ignore this, knowing it will be fixed up for us shortly in
1179          * dmu_recv_end_sync().
1180          */
1181         if (!recv && DS_IS_INCONSISTENT(ds))
1182                 return (SET_ERROR(EBUSY));
1183
1184         /*
1185          * Skip the check for temporary snapshots or if we have already checked
1186          * the counts in dsl_dataset_snapshot_check. This means we really only
1187          * check the count here when we're receiving a stream.
1188          */
1189         if (cnt != 0 && cr != NULL) {
1190                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1191                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1192                 if (error != 0)
1193                         return (error);
1194         }
1195
1196         error = dsl_dataset_snapshot_reserve_space(ds, tx);
1197         if (error != 0)
1198                 return (error);
1199
1200         return (0);
1201 }
1202
1203 static int
1204 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1205 {
1206         dsl_dataset_snapshot_arg_t *ddsa = arg;
1207         dsl_pool_t *dp = dmu_tx_pool(tx);
1208         nvpair_t *pair;
1209         int rv = 0;
1210
1211         /*
1212          * Pre-compute how many total new snapshots will be created for each
1213          * level in the tree and below. This is needed for validating the
1214          * snapshot limit when either taking a recursive snapshot or when
1215          * taking multiple snapshots.
1216          *
1217          * The problem is that the counts are not actually adjusted when
1218          * we are checking, only when we finally sync. For a single snapshot,
1219          * this is easy, the count will increase by 1 at each node up the tree,
1220          * but its more complicated for the recursive/multiple snapshot case.
1221          *
1222          * The dsl_fs_ss_limit_check function does recursively check the count
1223          * at each level up the tree but since it is validating each snapshot
1224          * independently we need to be sure that we are validating the complete
1225          * count for the entire set of snapshots. We do this by rolling up the
1226          * counts for each component of the name into an nvlist and then
1227          * checking each of those cases with the aggregated count.
1228          *
1229          * This approach properly handles not only the recursive snapshot
1230          * case (where we get all of those on the ddsa_snaps list) but also
1231          * the sibling case (e.g. snapshot a/b and a/c so that we will also
1232          * validate the limit on 'a' using a count of 2).
1233          *
1234          * We validate the snapshot names in the third loop and only report
1235          * name errors once.
1236          */
1237         if (dmu_tx_is_syncing(tx)) {
1238                 nvlist_t *cnt_track = NULL;
1239                 cnt_track = fnvlist_alloc();
1240
1241                 /* Rollup aggregated counts into the cnt_track list */
1242                 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1243                     pair != NULL;
1244                     pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1245                         char *pdelim;
1246                         uint64_t val;
1247                         char nm[MAXPATHLEN];
1248
1249                         (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1250                         pdelim = strchr(nm, '@');
1251                         if (pdelim == NULL)
1252                                 continue;
1253                         *pdelim = '\0';
1254
1255                         do {
1256                                 if (nvlist_lookup_uint64(cnt_track, nm,
1257                                     &val) == 0) {
1258                                         /* update existing entry */
1259                                         fnvlist_add_uint64(cnt_track, nm,
1260                                             val + 1);
1261                                 } else {
1262                                         /* add to list */
1263                                         fnvlist_add_uint64(cnt_track, nm, 1);
1264                                 }
1265
1266                                 pdelim = strrchr(nm, '/');
1267                                 if (pdelim != NULL)
1268                                         *pdelim = '\0';
1269                         } while (pdelim != NULL);
1270                 }
1271
1272                 /* Check aggregated counts at each level */
1273                 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1274                     pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1275                         int error = 0;
1276                         char *name;
1277                         uint64_t cnt = 0;
1278                         dsl_dataset_t *ds;
1279
1280                         name = nvpair_name(pair);
1281                         cnt = fnvpair_value_uint64(pair);
1282                         ASSERT(cnt > 0);
1283
1284                         error = dsl_dataset_hold(dp, name, FTAG, &ds);
1285                         if (error == 0) {
1286                                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1287                                     ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1288                                     ddsa->ddsa_cr);
1289                                 dsl_dataset_rele(ds, FTAG);
1290                         }
1291
1292                         if (error != 0) {
1293                                 if (ddsa->ddsa_errors != NULL)
1294                                         fnvlist_add_int32(ddsa->ddsa_errors,
1295                                             name, error);
1296                                 rv = error;
1297                                 /* only report one error for this check */
1298                                 break;
1299                         }
1300                 }
1301                 nvlist_free(cnt_track);
1302         }
1303
1304         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1305             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1306                 int error = 0;
1307                 dsl_dataset_t *ds;
1308                 char *name, *atp;
1309                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1310
1311                 name = nvpair_name(pair);
1312                 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1313                         error = SET_ERROR(ENAMETOOLONG);
1314                 if (error == 0) {
1315                         atp = strchr(name, '@');
1316                         if (atp == NULL)
1317                                 error = SET_ERROR(EINVAL);
1318                         if (error == 0)
1319                                 (void) strlcpy(dsname, name, atp - name + 1);
1320                 }
1321                 if (error == 0)
1322                         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1323                 if (error == 0) {
1324                         /* passing 0/NULL skips dsl_fs_ss_limit_check */
1325                         error = dsl_dataset_snapshot_check_impl(ds,
1326                             atp + 1, tx, B_FALSE, 0, NULL);
1327                         dsl_dataset_rele(ds, FTAG);
1328                 }
1329
1330                 if (error != 0) {
1331                         if (ddsa->ddsa_errors != NULL) {
1332                                 fnvlist_add_int32(ddsa->ddsa_errors,
1333                                     name, error);
1334                         }
1335                         rv = error;
1336                 }
1337         }
1338
1339         return (rv);
1340 }
1341
1342 void
1343 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1344     dmu_tx_t *tx)
1345 {
1346         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1347         dmu_buf_t *dbuf;
1348         dsl_dataset_phys_t *dsphys;
1349         uint64_t dsobj, crtxg;
1350         objset_t *mos = dp->dp_meta_objset;
1351         objset_t *os;
1352
1353         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1354
1355         /*
1356          * If we are on an old pool, the zil must not be active, in which
1357          * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1358          */
1359         ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1360             dmu_objset_from_ds(ds, &os) != 0 ||
1361             bcmp(&os->os_phys->os_zil_header, &zero_zil,
1362             sizeof (zero_zil)) == 0);
1363
1364         /* Should not snapshot a dirty dataset. */
1365         ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1366             ds, tx->tx_txg));
1367
1368         dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1369
1370         /*
1371          * The origin's ds_creation_txg has to be < TXG_INITIAL
1372          */
1373         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1374                 crtxg = 1;
1375         else
1376                 crtxg = tx->tx_txg;
1377
1378         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1379             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1380         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1381         dmu_buf_will_dirty(dbuf, tx);
1382         dsphys = dbuf->db_data;
1383         bzero(dsphys, sizeof (dsl_dataset_phys_t));
1384         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1385         dsphys->ds_fsid_guid = unique_create();
1386         do {
1387                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1388                     sizeof (dsphys->ds_guid));
1389         } while (dsphys->ds_guid == 0);
1390         dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1391         dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1392         dsphys->ds_next_snap_obj = ds->ds_object;
1393         dsphys->ds_num_children = 1;
1394         dsphys->ds_creation_time = gethrestime_sec();
1395         dsphys->ds_creation_txg = crtxg;
1396         dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1397         dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1398         dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1399         dsphys->ds_uncompressed_bytes =
1400             dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1401         dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1402         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1403         dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1404         rrw_exit(&ds->ds_bp_rwlock, FTAG);
1405         dmu_buf_rele(dbuf, FTAG);
1406
1407         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1408                 if (ds->ds_feature_inuse[f])
1409                         dsl_dataset_activate_feature(dsobj, f, tx);
1410         }
1411
1412         ASSERT3U(ds->ds_prev != 0, ==,
1413             dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1414         if (ds->ds_prev) {
1415                 uint64_t next_clones_obj =
1416                     dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1417                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1418                     ds->ds_object ||
1419                     dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1420                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1421                     ds->ds_object) {
1422                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1423                         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1424                             dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1425                         dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1426                 } else if (next_clones_obj != 0) {
1427                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
1428                             dsphys->ds_next_snap_obj, tx);
1429                         VERIFY0(zap_add_int(mos,
1430                             next_clones_obj, dsobj, tx));
1431                 }
1432         }
1433
1434         /*
1435          * If we have a reference-reservation on this dataset, we will
1436          * need to increase the amount of refreservation being charged
1437          * since our unique space is going to zero.
1438          */
1439         if (ds->ds_reserved) {
1440                 int64_t delta;
1441                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1442                 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1443                     ds->ds_reserved);
1444                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1445                     delta, 0, 0, tx);
1446         }
1447
1448         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1449         dsl_dataset_phys(ds)->ds_deadlist_obj =
1450             dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1451             dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1452         dsl_deadlist_close(&ds->ds_deadlist);
1453         dsl_deadlist_open(&ds->ds_deadlist, mos,
1454             dsl_dataset_phys(ds)->ds_deadlist_obj);
1455         dsl_deadlist_add_key(&ds->ds_deadlist,
1456             dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1457
1458         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1459         dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1460         dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1461         dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1462         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1463                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1464
1465         VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1466             snapname, 8, 1, &dsobj, tx));
1467
1468         if (ds->ds_prev)
1469                 dsl_dataset_rele(ds->ds_prev, ds);
1470         VERIFY0(dsl_dataset_hold_obj(dp,
1471             dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1472
1473         dsl_scan_ds_snapshotted(ds, tx);
1474
1475         dsl_dir_snap_cmtime_update(ds->ds_dir);
1476
1477         spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1478 }
1479
1480 static void
1481 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1482 {
1483         dsl_dataset_snapshot_arg_t *ddsa = arg;
1484         dsl_pool_t *dp = dmu_tx_pool(tx);
1485         nvpair_t *pair;
1486
1487         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1488             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1489                 dsl_dataset_t *ds;
1490                 char *name, *atp;
1491                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1492
1493                 name = nvpair_name(pair);
1494                 atp = strchr(name, '@');
1495                 (void) strlcpy(dsname, name, atp - name + 1);
1496                 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1497
1498                 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1499                 if (ddsa->ddsa_props != NULL) {
1500                         dsl_props_set_sync_impl(ds->ds_prev,
1501                             ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1502                 }
1503                 dsl_dataset_rele(ds, FTAG);
1504         }
1505 }
1506
1507 /*
1508  * The snapshots must all be in the same pool.
1509  * All-or-nothing: if there are any failures, nothing will be modified.
1510  */
1511 int
1512 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1513 {
1514         dsl_dataset_snapshot_arg_t ddsa;
1515         nvpair_t *pair;
1516         boolean_t needsuspend;
1517         int error;
1518         spa_t *spa;
1519         char *firstname;
1520         nvlist_t *suspended = NULL;
1521
1522         pair = nvlist_next_nvpair(snaps, NULL);
1523         if (pair == NULL)
1524                 return (0);
1525         firstname = nvpair_name(pair);
1526
1527         error = spa_open(firstname, &spa, FTAG);
1528         if (error != 0)
1529                 return (error);
1530         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1531         spa_close(spa, FTAG);
1532
1533         if (needsuspend) {
1534                 suspended = fnvlist_alloc();
1535                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1536                     pair = nvlist_next_nvpair(snaps, pair)) {
1537                         char fsname[ZFS_MAX_DATASET_NAME_LEN];
1538                         char *snapname = nvpair_name(pair);
1539                         char *atp;
1540                         void *cookie;
1541
1542                         atp = strchr(snapname, '@');
1543                         if (atp == NULL) {
1544                                 error = SET_ERROR(EINVAL);
1545                                 break;
1546                         }
1547                         (void) strlcpy(fsname, snapname, atp - snapname + 1);
1548
1549                         error = zil_suspend(fsname, &cookie);
1550                         if (error != 0)
1551                                 break;
1552                         fnvlist_add_uint64(suspended, fsname,
1553                             (uintptr_t)cookie);
1554                 }
1555         }
1556
1557         ddsa.ddsa_snaps = snaps;
1558         ddsa.ddsa_props = props;
1559         ddsa.ddsa_errors = errors;
1560         ddsa.ddsa_cr = CRED();
1561
1562         if (error == 0) {
1563                 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1564                     dsl_dataset_snapshot_sync, &ddsa,
1565                     fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1566         }
1567
1568         if (suspended != NULL) {
1569                 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1570                     pair = nvlist_next_nvpair(suspended, pair)) {
1571                         zil_resume((void *)(uintptr_t)
1572                             fnvpair_value_uint64(pair));
1573                 }
1574                 fnvlist_free(suspended);
1575         }
1576
1577 #ifdef __FreeBSD__
1578 #ifdef _KERNEL
1579         if (error == 0) {
1580                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1581                     pair = nvlist_next_nvpair(snaps, pair)) {
1582                         char *snapname = nvpair_name(pair);
1583                         zvol_create_minors(snapname);
1584                 }
1585         }
1586 #endif
1587 #endif
1588         return (error);
1589 }
1590
1591 typedef struct dsl_dataset_snapshot_tmp_arg {
1592         const char *ddsta_fsname;
1593         const char *ddsta_snapname;
1594         minor_t ddsta_cleanup_minor;
1595         const char *ddsta_htag;
1596 } dsl_dataset_snapshot_tmp_arg_t;
1597
1598 static int
1599 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1600 {
1601         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1602         dsl_pool_t *dp = dmu_tx_pool(tx);
1603         dsl_dataset_t *ds;
1604         int error;
1605
1606         error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1607         if (error != 0)
1608                 return (error);
1609
1610         /* NULL cred means no limit check for tmp snapshot */
1611         error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1612             tx, B_FALSE, 0, NULL);
1613         if (error != 0) {
1614                 dsl_dataset_rele(ds, FTAG);
1615                 return (error);
1616         }
1617
1618         if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1619                 dsl_dataset_rele(ds, FTAG);
1620                 return (SET_ERROR(ENOTSUP));
1621         }
1622         error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1623             B_TRUE, tx);
1624         if (error != 0) {
1625                 dsl_dataset_rele(ds, FTAG);
1626                 return (error);
1627         }
1628
1629         dsl_dataset_rele(ds, FTAG);
1630         return (0);
1631 }
1632
1633 static void
1634 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1635 {
1636         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1637         dsl_pool_t *dp = dmu_tx_pool(tx);
1638         dsl_dataset_t *ds;
1639
1640         VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1641
1642         dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1643         dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1644             ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1645         dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1646
1647         dsl_dataset_rele(ds, FTAG);
1648 }
1649
1650 int
1651 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1652     minor_t cleanup_minor, const char *htag)
1653 {
1654         dsl_dataset_snapshot_tmp_arg_t ddsta;
1655         int error;
1656         spa_t *spa;
1657         boolean_t needsuspend;
1658         void *cookie;
1659
1660         ddsta.ddsta_fsname = fsname;
1661         ddsta.ddsta_snapname = snapname;
1662         ddsta.ddsta_cleanup_minor = cleanup_minor;
1663         ddsta.ddsta_htag = htag;
1664
1665         error = spa_open(fsname, &spa, FTAG);
1666         if (error != 0)
1667                 return (error);
1668         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1669         spa_close(spa, FTAG);
1670
1671         if (needsuspend) {
1672                 error = zil_suspend(fsname, &cookie);
1673                 if (error != 0)
1674                         return (error);
1675         }
1676
1677         error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1678             dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1679
1680         if (needsuspend)
1681                 zil_resume(cookie);
1682         return (error);
1683 }
1684
1685
1686 void
1687 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1688 {
1689         ASSERT(dmu_tx_is_syncing(tx));
1690         ASSERT(ds->ds_objset != NULL);
1691         ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1692
1693         /*
1694          * in case we had to change ds_fsid_guid when we opened it,
1695          * sync it out now.
1696          */
1697         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1698         dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1699
1700         if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1701                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1702                     ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1703                     &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1704                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1705                     ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1706                     &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1707                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1708                     ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1709                     &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1710                 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1711                 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1712                 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1713         }
1714
1715         dmu_objset_sync(ds->ds_objset, zio, tx);
1716
1717         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1718                 if (ds->ds_feature_activation_needed[f]) {
1719                         if (ds->ds_feature_inuse[f])
1720                                 continue;
1721                         dsl_dataset_activate_feature(ds->ds_object, f, tx);
1722                         ds->ds_feature_inuse[f] = B_TRUE;
1723                 }
1724         }
1725 }
1726
1727 static int
1728 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1729 {
1730         dsl_deadlist_t *dl = arg;
1731         dsl_deadlist_insert(dl, bp, tx);
1732         return (0);
1733 }
1734
1735 void
1736 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx)
1737 {
1738         objset_t *os = ds->ds_objset;
1739
1740         bplist_iterate(&ds->ds_pending_deadlist,
1741             deadlist_enqueue_cb, &ds->ds_deadlist, tx);
1742
1743         if (os->os_synced_dnodes != NULL) {
1744                 multilist_destroy(os->os_synced_dnodes);
1745                 os->os_synced_dnodes = NULL;
1746         }
1747
1748         ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx)));
1749
1750         dmu_buf_rele(ds->ds_dbuf, ds);
1751 }
1752
1753 static void
1754 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1755 {
1756         uint64_t count = 0;
1757         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1758         zap_cursor_t zc;
1759         zap_attribute_t za;
1760         nvlist_t *propval = fnvlist_alloc();
1761         nvlist_t *val;
1762
1763         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1764
1765         /*
1766          * We use nvlist_alloc() instead of fnvlist_alloc() because the
1767          * latter would allocate the list with NV_UNIQUE_NAME flag.
1768          * As a result, every time a clone name is appended to the list
1769          * it would be (linearly) searched for for a duplicate name.
1770          * We already know that all clone names must be unique and we
1771          * want avoid the quadratic complexity of double-checking that
1772          * because we can have a large number of clones.
1773          */
1774         VERIFY0(nvlist_alloc(&val, 0, KM_SLEEP));
1775
1776         /*
1777          * There may be missing entries in ds_next_clones_obj
1778          * due to a bug in a previous version of the code.
1779          * Only trust it if it has the right number of entries.
1780          */
1781         if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1782                 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1783                     &count));
1784         }
1785         if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1786                 goto fail;
1787         for (zap_cursor_init(&zc, mos,
1788             dsl_dataset_phys(ds)->ds_next_clones_obj);
1789             zap_cursor_retrieve(&zc, &za) == 0;
1790             zap_cursor_advance(&zc)) {
1791                 dsl_dataset_t *clone;
1792                 char buf[ZFS_MAX_DATASET_NAME_LEN];
1793                 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1794                     za.za_first_integer, FTAG, &clone));
1795                 dsl_dir_name(clone->ds_dir, buf);
1796                 fnvlist_add_boolean(val, buf);
1797                 dsl_dataset_rele(clone, FTAG);
1798         }
1799         zap_cursor_fini(&zc);
1800         fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1801         fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1802 fail:
1803         nvlist_free(val);
1804         nvlist_free(propval);
1805 }
1806
1807 static void
1808 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1809 {
1810         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1811
1812         if (dsl_dataset_has_resume_receive_state(ds)) {
1813                 char *str;
1814                 void *packed;
1815                 uint8_t *compressed;
1816                 uint64_t val;
1817                 nvlist_t *token_nv = fnvlist_alloc();
1818                 size_t packed_size, compressed_size;
1819
1820                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1821                     DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1822                         fnvlist_add_uint64(token_nv, "fromguid", val);
1823                 }
1824                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1825                     DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1826                         fnvlist_add_uint64(token_nv, "object", val);
1827                 }
1828                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1829                     DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1830                         fnvlist_add_uint64(token_nv, "offset", val);
1831                 }
1832                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1833                     DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1834                         fnvlist_add_uint64(token_nv, "bytes", val);
1835                 }
1836                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1837                     DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1838                         fnvlist_add_uint64(token_nv, "toguid", val);
1839                 }
1840                 char buf[256];
1841                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1842                     DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1843                         fnvlist_add_string(token_nv, "toname", buf);
1844                 }
1845                 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1846                     DS_FIELD_RESUME_LARGEBLOCK) == 0) {
1847                         fnvlist_add_boolean(token_nv, "largeblockok");
1848                 }
1849                 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1850                     DS_FIELD_RESUME_EMBEDOK) == 0) {
1851                         fnvlist_add_boolean(token_nv, "embedok");
1852                 }
1853                 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1854                     DS_FIELD_RESUME_COMPRESSOK) == 0) {
1855                         fnvlist_add_boolean(token_nv, "compressok");
1856                 }
1857                 packed = fnvlist_pack(token_nv, &packed_size);
1858                 fnvlist_free(token_nv);
1859                 compressed = kmem_alloc(packed_size, KM_SLEEP);
1860
1861                 compressed_size = gzip_compress(packed, compressed,
1862                     packed_size, packed_size, 6);
1863
1864                 zio_cksum_t cksum;
1865                 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1866
1867                 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1868                 for (int i = 0; i < compressed_size; i++) {
1869                         (void) sprintf(str + i * 2, "%02x", compressed[i]);
1870                 }
1871                 str[compressed_size * 2] = '\0';
1872                 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1873                     ZFS_SEND_RESUME_TOKEN_VERSION,
1874                     (longlong_t)cksum.zc_word[0],
1875                     (longlong_t)packed_size, str);
1876                 dsl_prop_nvlist_add_string(nv,
1877                     ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1878                 kmem_free(packed, packed_size);
1879                 kmem_free(str, compressed_size * 2 + 1);
1880                 kmem_free(compressed, packed_size);
1881                 strfree(propval);
1882         }
1883 }
1884
1885 void
1886 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1887 {
1888         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1889         uint64_t refd, avail, uobjs, aobjs, ratio;
1890
1891         ASSERT(dsl_pool_config_held(dp));
1892
1893         ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1894             (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1895             dsl_dataset_phys(ds)->ds_compressed_bytes);
1896
1897         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1898         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1899             dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1900
1901         if (ds->ds_is_snapshot) {
1902                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1903                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1904                     dsl_dataset_phys(ds)->ds_unique_bytes);
1905                 get_clones_stat(ds, nv);
1906         } else {
1907                 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1908                         char buf[ZFS_MAX_DATASET_NAME_LEN];
1909                         dsl_dataset_name(ds->ds_prev, buf);
1910                         dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1911                 }
1912
1913                 dsl_dir_stats(ds->ds_dir, nv);
1914         }
1915
1916         dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1917         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1918         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1919
1920         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1921             dsl_dataset_phys(ds)->ds_creation_time);
1922         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1923             dsl_dataset_phys(ds)->ds_creation_txg);
1924         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1925             ds->ds_quota);
1926         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1927             ds->ds_reserved);
1928         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1929             dsl_dataset_phys(ds)->ds_guid);
1930         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1931             dsl_dataset_phys(ds)->ds_unique_bytes);
1932         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1933             ds->ds_object);
1934         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1935             ds->ds_userrefs);
1936         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1937             DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1938
1939         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1940                 uint64_t written, comp, uncomp;
1941                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1942                 dsl_dataset_t *prev;
1943
1944                 int err = dsl_dataset_hold_obj(dp,
1945                     dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1946                 if (err == 0) {
1947                         err = dsl_dataset_space_written(prev, ds, &written,
1948                             &comp, &uncomp);
1949                         dsl_dataset_rele(prev, FTAG);
1950                         if (err == 0) {
1951                                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1952                                     written);
1953                         }
1954                 }
1955         }
1956
1957         if (!dsl_dataset_is_snapshot(ds)) {
1958                 /*
1959                  * A failed "newfs" (e.g. full) resumable receive leaves
1960                  * the stats set on this dataset.  Check here for the prop.
1961                  */
1962                 get_receive_resume_stats(ds, nv);
1963
1964                 /*
1965                  * A failed incremental resumable receive leaves the
1966                  * stats set on our child named "%recv".  Check the child
1967                  * for the prop.
1968                  */
1969                 /* 6 extra bytes for /%recv */
1970                 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1971                 dsl_dataset_t *recv_ds;
1972                 dsl_dataset_name(ds, recvname);
1973                 if (strlcat(recvname, "/", sizeof (recvname)) <
1974                     sizeof (recvname) &&
1975                     strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1976                     sizeof (recvname) &&
1977                     dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1978                         get_receive_resume_stats(recv_ds, nv);
1979                         dsl_dataset_rele(recv_ds, FTAG);
1980                 }
1981         }
1982 }
1983
1984 void
1985 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1986 {
1987         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1988         ASSERT(dsl_pool_config_held(dp));
1989
1990         stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1991         stat->dds_inconsistent =
1992             dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1993         stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1994         stat->dds_origin[0] = '\0';
1995         if (ds->ds_is_snapshot) {
1996                 stat->dds_is_snapshot = B_TRUE;
1997                 stat->dds_num_clones =
1998                     dsl_dataset_phys(ds)->ds_num_children - 1;
1999         } else {
2000                 stat->dds_is_snapshot = B_FALSE;
2001                 stat->dds_num_clones = 0;
2002
2003                 if (dsl_dir_is_clone(ds->ds_dir)) {
2004                         dsl_dataset_t *ods;
2005
2006                         VERIFY0(dsl_dataset_hold_obj(dp,
2007                             dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
2008                             FTAG, &ods));
2009                         dsl_dataset_name(ods, stat->dds_origin);
2010                         dsl_dataset_rele(ods, FTAG);
2011                 }
2012         }
2013 }
2014
2015 uint64_t
2016 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2017 {
2018         return (ds->ds_fsid_guid);
2019 }
2020
2021 void
2022 dsl_dataset_space(dsl_dataset_t *ds,
2023     uint64_t *refdbytesp, uint64_t *availbytesp,
2024     uint64_t *usedobjsp, uint64_t *availobjsp)
2025 {
2026         *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
2027         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2028         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
2029                 *availbytesp +=
2030                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2031         if (ds->ds_quota != 0) {
2032                 /*
2033                  * Adjust available bytes according to refquota
2034                  */
2035                 if (*refdbytesp < ds->ds_quota)
2036                         *availbytesp = MIN(*availbytesp,
2037                             ds->ds_quota - *refdbytesp);
2038                 else
2039                         *availbytesp = 0;
2040         }
2041         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2042         *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
2043         rrw_exit(&ds->ds_bp_rwlock, FTAG);
2044         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2045 }
2046
2047 boolean_t
2048 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2049 {
2050         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2051         uint64_t birth;
2052
2053         ASSERT(dsl_pool_config_held(dp));
2054         if (snap == NULL)
2055                 return (B_FALSE);
2056         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2057         birth = dsl_dataset_get_blkptr(ds)->blk_birth;
2058         rrw_exit(&ds->ds_bp_rwlock, FTAG);
2059         if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2060                 objset_t *os, *os_snap;
2061                 /*
2062                  * It may be that only the ZIL differs, because it was
2063                  * reset in the head.  Don't count that as being
2064                  * modified.
2065                  */
2066                 if (dmu_objset_from_ds(ds, &os) != 0)
2067                         return (B_TRUE);
2068                 if (dmu_objset_from_ds(snap, &os_snap) != 0)
2069                         return (B_TRUE);
2070                 return (bcmp(&os->os_phys->os_meta_dnode,
2071                     &os_snap->os_phys->os_meta_dnode,
2072                     sizeof (os->os_phys->os_meta_dnode)) != 0);
2073         }
2074         return (B_FALSE);
2075 }
2076
2077 typedef struct dsl_dataset_rename_snapshot_arg {
2078         const char *ddrsa_fsname;
2079         const char *ddrsa_oldsnapname;
2080         const char *ddrsa_newsnapname;
2081         boolean_t ddrsa_recursive;
2082         dmu_tx_t *ddrsa_tx;
2083 } dsl_dataset_rename_snapshot_arg_t;
2084
2085 /* ARGSUSED */
2086 static int
2087 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2088     dsl_dataset_t *hds, void *arg)
2089 {
2090         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2091         int error;
2092         uint64_t val;
2093
2094         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2095         if (error != 0) {
2096                 /* ignore nonexistent snapshots */
2097                 return (error == ENOENT ? 0 : error);
2098         }
2099
2100         /* new name should not exist */
2101         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2102         if (error == 0)
2103                 error = SET_ERROR(EEXIST);
2104         else if (error == ENOENT)
2105                 error = 0;
2106
2107         /* dataset name + 1 for the "@" + the new snapshot name must fit */
2108         if (dsl_dir_namelen(hds->ds_dir) + 1 +
2109             strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2110                 error = SET_ERROR(ENAMETOOLONG);
2111
2112         return (error);
2113 }
2114
2115 static int
2116 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2117 {
2118         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2119         dsl_pool_t *dp = dmu_tx_pool(tx);
2120         dsl_dataset_t *hds;
2121         int error;
2122
2123         error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2124         if (error != 0)
2125                 return (error);
2126
2127         if (ddrsa->ddrsa_recursive) {
2128                 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2129                     dsl_dataset_rename_snapshot_check_impl, ddrsa,
2130                     DS_FIND_CHILDREN);
2131         } else {
2132                 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2133         }
2134         dsl_dataset_rele(hds, FTAG);
2135         return (error);
2136 }
2137
2138 static int
2139 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2140     dsl_dataset_t *hds, void *arg)
2141 {
2142 #ifdef __FreeBSD__
2143 #ifdef _KERNEL
2144         char *oldname, *newname;
2145 #endif
2146 #endif
2147         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2148         dsl_dataset_t *ds;
2149         uint64_t val;
2150         dmu_tx_t *tx = ddrsa->ddrsa_tx;
2151         int error;
2152
2153         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2154         ASSERT(error == 0 || error == ENOENT);
2155         if (error == ENOENT) {
2156                 /* ignore nonexistent snapshots */
2157                 return (0);
2158         }
2159
2160         VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2161
2162         /* log before we change the name */
2163         spa_history_log_internal_ds(ds, "rename", tx,
2164             "-> @%s", ddrsa->ddrsa_newsnapname);
2165
2166         VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2167             B_FALSE));
2168         mutex_enter(&ds->ds_lock);
2169         (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2170         mutex_exit(&ds->ds_lock);
2171         VERIFY0(zap_add(dp->dp_meta_objset,
2172             dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2173             ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2174
2175 #ifdef __FreeBSD__
2176 #ifdef _KERNEL
2177         oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2178         newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2179         snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2180             ddrsa->ddrsa_oldsnapname);
2181         snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2182             ddrsa->ddrsa_newsnapname);
2183         zfsvfs_update_fromname(oldname, newname);
2184         zvol_rename_minors(oldname, newname);
2185         kmem_free(newname, MAXPATHLEN);
2186         kmem_free(oldname, MAXPATHLEN);
2187 #endif
2188 #endif
2189         dsl_dataset_rele(ds, FTAG);
2190
2191         return (0);
2192 }
2193
2194 static void
2195 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2196 {
2197         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2198         dsl_pool_t *dp = dmu_tx_pool(tx);
2199         dsl_dataset_t *hds;
2200
2201         VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2202         ddrsa->ddrsa_tx = tx;
2203         if (ddrsa->ddrsa_recursive) {
2204                 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2205                     dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2206                     DS_FIND_CHILDREN));
2207         } else {
2208                 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2209         }
2210         dsl_dataset_rele(hds, FTAG);
2211 }
2212
2213 int
2214 dsl_dataset_rename_snapshot(const char *fsname,
2215     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2216 {
2217         dsl_dataset_rename_snapshot_arg_t ddrsa;
2218
2219         ddrsa.ddrsa_fsname = fsname;
2220         ddrsa.ddrsa_oldsnapname = oldsnapname;
2221         ddrsa.ddrsa_newsnapname = newsnapname;
2222         ddrsa.ddrsa_recursive = recursive;
2223
2224         return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2225             dsl_dataset_rename_snapshot_sync, &ddrsa,
2226             1, ZFS_SPACE_CHECK_RESERVED));
2227 }
2228
2229 /*
2230  * If we're doing an ownership handoff, we need to make sure that there is
2231  * only one long hold on the dataset.  We're not allowed to change anything here
2232  * so we don't permanently release the long hold or regular hold here.  We want
2233  * to do this only when syncing to avoid the dataset unexpectedly going away
2234  * when we release the long hold.
2235  */
2236 static int
2237 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2238 {
2239         boolean_t held;
2240
2241         if (!dmu_tx_is_syncing(tx))
2242                 return (0);
2243
2244         if (owner != NULL) {
2245                 VERIFY3P(ds->ds_owner, ==, owner);
2246                 dsl_dataset_long_rele(ds, owner);
2247         }
2248
2249         held = dsl_dataset_long_held(ds);
2250
2251         if (owner != NULL)
2252                 dsl_dataset_long_hold(ds, owner);
2253
2254         if (held)
2255                 return (SET_ERROR(EBUSY));
2256
2257         return (0);
2258 }
2259
2260 typedef struct dsl_dataset_rollback_arg {
2261         const char *ddra_fsname;
2262         void *ddra_owner;
2263         nvlist_t *ddra_result;
2264 } dsl_dataset_rollback_arg_t;
2265
2266 static int
2267 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2268 {
2269         dsl_dataset_rollback_arg_t *ddra = arg;
2270         dsl_pool_t *dp = dmu_tx_pool(tx);
2271         dsl_dataset_t *ds;
2272         int64_t unused_refres_delta;
2273         int error;
2274
2275         error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2276         if (error != 0)
2277                 return (error);
2278
2279         /* must not be a snapshot */
2280         if (ds->ds_is_snapshot) {
2281                 dsl_dataset_rele(ds, FTAG);
2282                 return (SET_ERROR(EINVAL));
2283         }
2284
2285         /* must have a most recent snapshot */
2286         if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2287                 dsl_dataset_rele(ds, FTAG);
2288                 return (SET_ERROR(EINVAL));
2289         }
2290
2291         /*
2292          * No rollback to a snapshot created in the current txg, because
2293          * the rollback may dirty the dataset and create blocks that are
2294          * not reachable from the rootbp while having a birth txg that
2295          * falls into the snapshot's range.
2296          */
2297         if (dmu_tx_is_syncing(tx) &&
2298             dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
2299                 dsl_dataset_rele(ds, FTAG);
2300                 return (SET_ERROR(EAGAIN));
2301         }
2302
2303         /* must not have any bookmarks after the most recent snapshot */
2304         nvlist_t *proprequest = fnvlist_alloc();
2305         fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2306         nvlist_t *bookmarks = fnvlist_alloc();
2307         error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2308         fnvlist_free(proprequest);
2309         if (error != 0)
2310                 return (error);
2311         for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2312             pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2313                 nvlist_t *valuenv =
2314                     fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2315                     zfs_prop_to_name(ZFS_PROP_CREATETXG));
2316                 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2317                 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2318                         fnvlist_free(bookmarks);
2319                         dsl_dataset_rele(ds, FTAG);
2320                         return (SET_ERROR(EEXIST));
2321                 }
2322         }
2323         fnvlist_free(bookmarks);
2324
2325         error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2326         if (error != 0) {
2327                 dsl_dataset_rele(ds, FTAG);
2328                 return (error);
2329         }
2330
2331         /*
2332          * Check if the snap we are rolling back to uses more than
2333          * the refquota.
2334          */
2335         if (ds->ds_quota != 0 &&
2336             dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2337                 dsl_dataset_rele(ds, FTAG);
2338                 return (SET_ERROR(EDQUOT));
2339         }
2340
2341         /*
2342          * When we do the clone swap, we will temporarily use more space
2343          * due to the refreservation (the head will no longer have any
2344          * unique space, so the entire amount of the refreservation will need
2345          * to be free).  We will immediately destroy the clone, freeing
2346          * this space, but the freeing happens over many txg's.
2347          */
2348         unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2349             dsl_dataset_phys(ds)->ds_unique_bytes);
2350
2351         if (unused_refres_delta > 0 &&
2352             unused_refres_delta >
2353             dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2354                 dsl_dataset_rele(ds, FTAG);
2355                 return (SET_ERROR(ENOSPC));
2356         }
2357
2358         dsl_dataset_rele(ds, FTAG);
2359         return (0);
2360 }
2361
2362 static void
2363 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2364 {
2365         dsl_dataset_rollback_arg_t *ddra = arg;
2366         dsl_pool_t *dp = dmu_tx_pool(tx);
2367         dsl_dataset_t *ds, *clone;
2368         uint64_t cloneobj;
2369         char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2370
2371         VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2372
2373         dsl_dataset_name(ds->ds_prev, namebuf);
2374         fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2375
2376         cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2377             ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2378
2379         VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2380
2381         dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2382         dsl_dataset_zero_zil(ds, tx);
2383
2384         dsl_destroy_head_sync_impl(clone, tx);
2385
2386         dsl_dataset_rele(clone, FTAG);
2387         dsl_dataset_rele(ds, FTAG);
2388 }
2389
2390 /*
2391  * Rolls back the given filesystem or volume to the most recent snapshot.
2392  * The name of the most recent snapshot will be returned under key "target"
2393  * in the result nvlist.
2394  *
2395  * If owner != NULL:
2396  * - The existing dataset MUST be owned by the specified owner at entry
2397  * - Upon return, dataset will still be held by the same owner, whether we
2398  *   succeed or not.
2399  *
2400  * This mode is required any time the existing filesystem is mounted.  See
2401  * notes above zfs_suspend_fs() for further details.
2402  */
2403 int
2404 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2405 {
2406         dsl_dataset_rollback_arg_t ddra;
2407
2408         ddra.ddra_fsname = fsname;
2409         ddra.ddra_owner = owner;
2410         ddra.ddra_result = result;
2411
2412         return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2413             dsl_dataset_rollback_sync, &ddra,
2414             1, ZFS_SPACE_CHECK_RESERVED));
2415 }
2416
2417 struct promotenode {
2418         list_node_t link;
2419         dsl_dataset_t *ds;
2420 };
2421
2422 typedef struct dsl_dataset_promote_arg {
2423         const char *ddpa_clonename;
2424         dsl_dataset_t *ddpa_clone;
2425         list_t shared_snaps, origin_snaps, clone_snaps;
2426         dsl_dataset_t *origin_origin; /* origin of the origin */
2427         uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2428         char *err_ds;
2429         cred_t *cr;
2430 } dsl_dataset_promote_arg_t;
2431
2432 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2433 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2434     void *tag);
2435 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2436
2437 static int
2438 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2439 {
2440         dsl_dataset_promote_arg_t *ddpa = arg;
2441         dsl_pool_t *dp = dmu_tx_pool(tx);
2442         dsl_dataset_t *hds;
2443         struct promotenode *snap;
2444         dsl_dataset_t *origin_ds;
2445         int err;
2446         uint64_t unused;
2447         uint64_t ss_mv_cnt;
2448         size_t max_snap_len;
2449
2450         err = promote_hold(ddpa, dp, FTAG);
2451         if (err != 0)
2452                 return (err);
2453
2454         hds = ddpa->ddpa_clone;
2455         max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2456
2457         if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2458                 promote_rele(ddpa, FTAG);
2459                 return (SET_ERROR(EXDEV));
2460         }
2461
2462         /*
2463          * Compute and check the amount of space to transfer.  Since this is
2464          * so expensive, don't do the preliminary check.
2465          */
2466         if (!dmu_tx_is_syncing(tx)) {
2467                 promote_rele(ddpa, FTAG);
2468                 return (0);
2469         }
2470
2471         snap = list_head(&ddpa->shared_snaps);
2472         origin_ds = snap->ds;
2473
2474         /* compute origin's new unique space */
2475         snap = list_tail(&ddpa->clone_snaps);
2476         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2477             origin_ds->ds_object);
2478         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2479             dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2480             &ddpa->unique, &unused, &unused);
2481
2482         /*
2483          * Walk the snapshots that we are moving
2484          *
2485          * Compute space to transfer.  Consider the incremental changes
2486          * to used by each snapshot:
2487          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2488          * So each snapshot gave birth to:
2489          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2490          * So a sequence would look like:
2491          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2492          * Which simplifies to:
2493          * uN + kN + kN-1 + ... + k1 + k0
2494          * Note however, if we stop before we reach the ORIGIN we get:
2495          * uN + kN + kN-1 + ... + kM - uM-1
2496          */
2497         ss_mv_cnt = 0;
2498         ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2499         ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2500         ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2501         for (snap = list_head(&ddpa->shared_snaps); snap;
2502             snap = list_next(&ddpa->shared_snaps, snap)) {
2503                 uint64_t val, dlused, dlcomp, dluncomp;
2504                 dsl_dataset_t *ds = snap->ds;
2505
2506                 ss_mv_cnt++;
2507
2508                 /*
2509                  * If there are long holds, we won't be able to evict
2510                  * the objset.
2511                  */
2512                 if (dsl_dataset_long_held(ds)) {
2513                         err = SET_ERROR(EBUSY);
2514                         goto out;
2515                 }
2516
2517                 /* Check that the snapshot name does not conflict */
2518                 VERIFY0(dsl_dataset_get_snapname(ds));
2519                 if (strlen(ds->ds_snapname) >= max_snap_len) {
2520                         err = SET_ERROR(ENAMETOOLONG);
2521                         goto out;
2522                 }
2523                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2524                 if (err == 0) {
2525                         (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2526                         err = SET_ERROR(EEXIST);
2527                         goto out;
2528                 }
2529                 if (err != ENOENT)
2530                         goto out;
2531
2532                 /* The very first snapshot does not have a deadlist */
2533                 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2534                         continue;
2535
2536                 dsl_deadlist_space(&ds->ds_deadlist,
2537                     &dlused, &dlcomp, &dluncomp);
2538                 ddpa->used += dlused;
2539                 ddpa->comp += dlcomp;
2540                 ddpa->uncomp += dluncomp;
2541         }
2542
2543         /*
2544          * If we are a clone of a clone then we never reached ORIGIN,
2545          * so we need to subtract out the clone origin's used space.
2546          */
2547         if (ddpa->origin_origin) {
2548                 ddpa->used -=
2549                     dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2550                 ddpa->comp -=
2551                     dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2552                 ddpa->uncomp -=
2553                     dsl_dataset_phys(ddpa->origin_origin)->
2554                     ds_uncompressed_bytes;
2555         }
2556
2557         /* Check that there is enough space and limit headroom here */
2558         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2559             0, ss_mv_cnt, ddpa->used, ddpa->cr);
2560         if (err != 0)
2561                 goto out;
2562
2563         /*
2564          * Compute the amounts of space that will be used by snapshots
2565          * after the promotion (for both origin and clone).  For each,
2566          * it is the amount of space that will be on all of their
2567          * deadlists (that was not born before their new origin).
2568          */
2569         if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2570                 uint64_t space;
2571
2572                 /*
2573                  * Note, typically this will not be a clone of a clone,
2574                  * so dd_origin_txg will be < TXG_INITIAL, so
2575                  * these snaplist_space() -> dsl_deadlist_space_range()
2576                  * calls will be fast because they do not have to
2577                  * iterate over all bps.
2578                  */
2579                 snap = list_head(&ddpa->origin_snaps);
2580                 err = snaplist_space(&ddpa->shared_snaps,
2581                     snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2582                 if (err != 0)
2583                         goto out;
2584
2585                 err = snaplist_space(&ddpa->clone_snaps,
2586                     snap->ds->ds_dir->dd_origin_txg, &space);
2587                 if (err != 0)
2588                         goto out;
2589                 ddpa->cloneusedsnap += space;
2590         }
2591         if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2592             DD_FLAG_USED_BREAKDOWN) {
2593                 err = snaplist_space(&ddpa->origin_snaps,
2594                     dsl_dataset_phys(origin_ds)->ds_creation_txg,
2595                     &ddpa->originusedsnap);
2596                 if (err != 0)
2597                         goto out;
2598         }
2599
2600 out:
2601         promote_rele(ddpa, FTAG);
2602         return (err);
2603 }
2604
2605 static void
2606 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2607 {
2608         dsl_dataset_promote_arg_t *ddpa = arg;
2609         dsl_pool_t *dp = dmu_tx_pool(tx);
2610         dsl_dataset_t *hds;
2611         struct promotenode *snap;
2612         dsl_dataset_t *origin_ds;
2613         dsl_dataset_t *origin_head;
2614         dsl_dir_t *dd;
2615         dsl_dir_t *odd = NULL;
2616         uint64_t oldnext_obj;
2617         int64_t delta;
2618 #if defined(__FreeBSD__) && defined(_KERNEL)
2619         char *oldname, *newname;
2620 #endif
2621
2622         VERIFY0(promote_hold(ddpa, dp, FTAG));
2623         hds = ddpa->ddpa_clone;
2624
2625         ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2626
2627         snap = list_head(&ddpa->shared_snaps);
2628         origin_ds = snap->ds;
2629         dd = hds->ds_dir;
2630
2631         snap = list_head(&ddpa->origin_snaps);
2632         origin_head = snap->ds;
2633
2634         /*
2635          * We need to explicitly open odd, since origin_ds's dd will be
2636          * changing.
2637          */
2638         VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2639             NULL, FTAG, &odd));
2640
2641         /* change origin's next snap */
2642         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2643         oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2644         snap = list_tail(&ddpa->clone_snaps);
2645         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2646             origin_ds->ds_object);
2647         dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2648
2649         /* change the origin's next clone */
2650         if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2651                 dsl_dataset_remove_from_next_clones(origin_ds,
2652                     snap->ds->ds_object, tx);
2653                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2654                     dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2655                     oldnext_obj, tx));
2656         }
2657
2658         /* change origin */
2659         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2660         ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2661         dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2662         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2663         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2664         dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2665         origin_head->ds_dir->dd_origin_txg =
2666             dsl_dataset_phys(origin_ds)->ds_creation_txg;
2667
2668         /* change dd_clone entries */
2669         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2670                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2671                     dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2672                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2673                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2674                     hds->ds_object, tx));
2675
2676                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2677                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2678                     origin_head->ds_object, tx));
2679                 if (dsl_dir_phys(dd)->dd_clones == 0) {
2680                         dsl_dir_phys(dd)->dd_clones =
2681                             zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2682                             DMU_OT_NONE, 0, tx);
2683                 }
2684                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2685                     dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2686         }
2687
2688 #if defined(__FreeBSD__) && defined(_KERNEL)
2689         /* Take the spa_namespace_lock early so zvol renames don't deadlock. */
2690         mutex_enter(&spa_namespace_lock);
2691
2692         oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2693         newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2694 #endif
2695
2696         /* move snapshots to this dir */
2697         for (snap = list_head(&ddpa->shared_snaps); snap;
2698             snap = list_next(&ddpa->shared_snaps, snap)) {
2699                 dsl_dataset_t *ds = snap->ds;
2700
2701                 /*
2702                  * Property callbacks are registered to a particular
2703                  * dsl_dir.  Since ours is changing, evict the objset
2704                  * so that they will be unregistered from the old dsl_dir.
2705                  */
2706                 if (ds->ds_objset) {
2707                         dmu_objset_evict(ds->ds_objset);
2708                         ds->ds_objset = NULL;
2709                 }
2710
2711                 /* move snap name entry */
2712                 VERIFY0(dsl_dataset_get_snapname(ds));
2713                 VERIFY0(dsl_dataset_snap_remove(origin_head,
2714                     ds->ds_snapname, tx, B_TRUE));
2715                 VERIFY0(zap_add(dp->dp_meta_objset,
2716                     dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2717                     8, 1, &ds->ds_object, tx));
2718                 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2719                     DD_FIELD_SNAPSHOT_COUNT, tx);
2720
2721                 /* change containing dsl_dir */
2722                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2723                 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2724                 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2725                 ASSERT3P(ds->ds_dir, ==, odd);
2726                 dsl_dir_rele(ds->ds_dir, ds);
2727                 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2728                     NULL, ds, &ds->ds_dir));
2729
2730 #if defined(__FreeBSD__) && defined(_KERNEL)
2731                 dsl_dataset_name(ds, newname);
2732                 zfsvfs_update_fromname(oldname, newname);
2733                 zvol_rename_minors(oldname, newname);
2734 #endif
2735
2736                 /* move any clone references */
2737                 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2738                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2739                         zap_cursor_t zc;
2740                         zap_attribute_t za;
2741
2742                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
2743                             dsl_dataset_phys(ds)->ds_next_clones_obj);
2744                             zap_cursor_retrieve(&zc, &za) == 0;
2745                             zap_cursor_advance(&zc)) {
2746                                 dsl_dataset_t *cnds;
2747                                 uint64_t o;
2748
2749                                 if (za.za_first_integer == oldnext_obj) {
2750                                         /*
2751                                          * We've already moved the
2752                                          * origin's reference.
2753                                          */
2754                                         continue;
2755                                 }
2756
2757                                 VERIFY0(dsl_dataset_hold_obj(dp,
2758                                     za.za_first_integer, FTAG, &cnds));
2759                                 o = dsl_dir_phys(cnds->ds_dir)->
2760                                     dd_head_dataset_obj;
2761
2762                                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2763                                     dsl_dir_phys(odd)->dd_clones, o, tx));
2764                                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2765                                     dsl_dir_phys(dd)->dd_clones, o, tx));
2766                                 dsl_dataset_rele(cnds, FTAG);
2767                         }
2768                         zap_cursor_fini(&zc);
2769                 }
2770
2771                 ASSERT(!dsl_prop_hascb(ds));
2772         }
2773
2774 #if defined(__FreeBSD__) && defined(_KERNEL)
2775         mutex_exit(&spa_namespace_lock);
2776
2777         kmem_free(newname, MAXPATHLEN);
2778         kmem_free(oldname, MAXPATHLEN);
2779 #endif
2780         /*
2781          * Change space accounting.
2782          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2783          * both be valid, or both be 0 (resulting in delta == 0).  This
2784          * is true for each of {clone,origin} independently.
2785          */
2786
2787         delta = ddpa->cloneusedsnap -
2788             dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2789         ASSERT3S(delta, >=, 0);
2790         ASSERT3U(ddpa->used, >=, delta);
2791         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2792         dsl_dir_diduse_space(dd, DD_USED_HEAD,
2793             ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2794
2795         delta = ddpa->originusedsnap -
2796             dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2797         ASSERT3S(delta, <=, 0);
2798         ASSERT3U(ddpa->used, >=, -delta);
2799         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2800         dsl_dir_diduse_space(odd, DD_USED_HEAD,
2801             -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2802
2803         dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2804
2805         /* log history record */
2806         spa_history_log_internal_ds(hds, "promote", tx, "");
2807
2808         dsl_dir_rele(odd, FTAG);
2809         promote_rele(ddpa, FTAG);
2810 }
2811
2812 /*
2813  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2814  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2815  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2816  * snapshots back to this dataset's origin.
2817  */
2818 static int
2819 snaplist_make(dsl_pool_t *dp,
2820     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2821 {
2822         uint64_t obj = last_obj;
2823
2824         list_create(l, sizeof (struct promotenode),
2825             offsetof(struct promotenode, link));
2826
2827         while (obj != first_obj) {
2828                 dsl_dataset_t *ds;
2829                 struct promotenode *snap;
2830                 int err;
2831
2832                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2833                 ASSERT(err != ENOENT);
2834                 if (err != 0)
2835                         return (err);
2836
2837                 if (first_obj == 0)
2838                         first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2839
2840                 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2841                 snap->ds = ds;
2842                 list_insert_tail(l, snap);
2843                 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2844         }
2845
2846         return (0);
2847 }
2848
2849 static int
2850 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2851 {
2852         struct promotenode *snap;
2853
2854         *spacep = 0;
2855         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2856                 uint64_t used, comp, uncomp;
2857                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2858                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
2859                 *spacep += used;
2860         }
2861         return (0);
2862 }
2863
2864 static void
2865 snaplist_destroy(list_t *l, void *tag)
2866 {
2867         struct promotenode *snap;
2868
2869         if (l == NULL || !list_link_active(&l->list_head))
2870                 return;
2871
2872         while ((snap = list_tail(l)) != NULL) {
2873                 list_remove(l, snap);
2874                 dsl_dataset_rele(snap->ds, tag);
2875                 kmem_free(snap, sizeof (*snap));
2876         }
2877         list_destroy(l);
2878 }
2879
2880 static int
2881 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2882 {
2883         int error;
2884         dsl_dir_t *dd;
2885         struct promotenode *snap;
2886
2887         error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2888             &ddpa->ddpa_clone);
2889         if (error != 0)
2890                 return (error);
2891         dd = ddpa->ddpa_clone->ds_dir;
2892
2893         if (ddpa->ddpa_clone->ds_is_snapshot ||
2894             !dsl_dir_is_clone(dd)) {
2895                 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2896                 return (SET_ERROR(EINVAL));
2897         }
2898
2899         error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2900             &ddpa->shared_snaps, tag);
2901         if (error != 0)
2902                 goto out;
2903
2904         error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2905             &ddpa->clone_snaps, tag);
2906         if (error != 0)
2907                 goto out;
2908
2909         snap = list_head(&ddpa->shared_snaps);
2910         ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2911         error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2912             dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2913             &ddpa->origin_snaps, tag);
2914         if (error != 0)
2915                 goto out;
2916
2917         if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2918                 error = dsl_dataset_hold_obj(dp,
2919                     dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2920                     tag, &ddpa->origin_origin);
2921                 if (error != 0)
2922                         goto out;
2923         }
2924 out:
2925         if (error != 0)
2926                 promote_rele(ddpa, tag);
2927         return (error);
2928 }
2929
2930 static void
2931 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2932 {
2933         snaplist_destroy(&ddpa->shared_snaps, tag);
2934         snaplist_destroy(&ddpa->clone_snaps, tag);
2935         snaplist_destroy(&ddpa->origin_snaps, tag);
2936         if (ddpa->origin_origin != NULL)
2937                 dsl_dataset_rele(ddpa->origin_origin, tag);
2938         dsl_dataset_rele(ddpa->ddpa_clone, tag);
2939 }
2940
2941 /*
2942  * Promote a clone.
2943  *
2944  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2945  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2946  */
2947 int
2948 dsl_dataset_promote(const char *name, char *conflsnap)
2949 {
2950         dsl_dataset_promote_arg_t ddpa = { 0 };
2951         uint64_t numsnaps;
2952         int error;
2953         objset_t *os;
2954
2955         /*
2956          * We will modify space proportional to the number of
2957          * snapshots.  Compute numsnaps.
2958          */
2959         error = dmu_objset_hold(name, FTAG, &os);
2960         if (error != 0)
2961                 return (error);
2962         error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2963             dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2964             &numsnaps);
2965         dmu_objset_rele(os, FTAG);
2966         if (error != 0)
2967                 return (error);
2968
2969         ddpa.ddpa_clonename = name;
2970         ddpa.err_ds = conflsnap;
2971         ddpa.cr = CRED();
2972
2973         return (dsl_sync_task(name, dsl_dataset_promote_check,
2974             dsl_dataset_promote_sync, &ddpa,
2975             2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2976 }
2977
2978 int
2979 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2980     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2981 {
2982         /*
2983          * "slack" factor for received datasets with refquota set on them.
2984          * See the bottom of this function for details on its use.
2985          */
2986         uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
2987         int64_t unused_refres_delta;
2988
2989         /* they should both be heads */
2990         if (clone->ds_is_snapshot ||
2991             origin_head->ds_is_snapshot)
2992                 return (SET_ERROR(EINVAL));
2993
2994         /* if we are not forcing, the branch point should be just before them */
2995         if (!force && clone->ds_prev != origin_head->ds_prev)
2996                 return (SET_ERROR(EINVAL));
2997
2998         /* clone should be the clone (unless they are unrelated) */
2999         if (clone->ds_prev != NULL &&
3000             clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3001             origin_head->ds_dir != clone->ds_prev->ds_dir)
3002                 return (SET_ERROR(EINVAL));
3003
3004         /* the clone should be a child of the origin */
3005         if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3006                 return (SET_ERROR(EINVAL));
3007
3008         /* origin_head shouldn't be modified unless 'force' */
3009         if (!force &&
3010             dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3011                 return (SET_ERROR(ETXTBSY));
3012
3013         /* origin_head should have no long holds (e.g. is not mounted) */
3014         if (dsl_dataset_handoff_check(origin_head, owner, tx))
3015                 return (SET_ERROR(EBUSY));
3016
3017         /* check amount of any unconsumed refreservation */
3018         unused_refres_delta =
3019             (int64_t)MIN(origin_head->ds_reserved,
3020             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3021             (int64_t)MIN(origin_head->ds_reserved,
3022             dsl_dataset_phys(clone)->ds_unique_bytes);
3023
3024         if (unused_refres_delta > 0 &&
3025             unused_refres_delta >
3026             dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3027                 return (SET_ERROR(ENOSPC));
3028
3029         /*
3030          * The clone can't be too much over the head's refquota.
3031          *
3032          * To ensure that the entire refquota can be used, we allow one
3033          * transaction to exceed the the refquota.  Therefore, this check
3034          * needs to also allow for the space referenced to be more than the
3035          * refquota.  The maximum amount of space that one transaction can use
3036          * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
3037          * overage ensures that we are able to receive a filesystem that
3038          * exceeds the refquota on the source system.
3039          *
3040          * So that overage is the refquota_slack we use below.
3041          */
3042         if (origin_head->ds_quota != 0 &&
3043             dsl_dataset_phys(clone)->ds_referenced_bytes >
3044             origin_head->ds_quota + refquota_slack)
3045                 return (SET_ERROR(EDQUOT));
3046
3047         return (0);
3048 }
3049
3050 void
3051 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
3052     dsl_dataset_t *origin_head, dmu_tx_t *tx)
3053 {
3054         dsl_pool_t *dp = dmu_tx_pool(tx);
3055         int64_t unused_refres_delta;
3056
3057         ASSERT(clone->ds_reserved == 0);
3058         /*
3059          * NOTE: On DEBUG kernels there could be a race between this and
3060          * the check function if spa_asize_inflation is adjusted...
3061          */
3062         ASSERT(origin_head->ds_quota == 0 ||
3063             dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
3064             DMU_MAX_ACCESS * spa_asize_inflation);
3065         ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
3066
3067         /*
3068          * Swap per-dataset feature flags.
3069          */
3070         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
3071                 if (!(spa_feature_table[f].fi_flags &
3072                     ZFEATURE_FLAG_PER_DATASET)) {
3073                         ASSERT(!clone->ds_feature_inuse[f]);
3074                         ASSERT(!origin_head->ds_feature_inuse[f]);
3075                         continue;
3076                 }
3077
3078                 boolean_t clone_inuse = clone->ds_feature_inuse[f];
3079                 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3080
3081                 if (clone_inuse) {
3082                         dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3083                         clone->ds_feature_inuse[f] = B_FALSE;
3084                 }
3085                 if (origin_head_inuse) {
3086                         dsl_dataset_deactivate_feature(origin_head->ds_object,
3087                             f, tx);
3088                         origin_head->ds_feature_inuse[f] = B_FALSE;
3089                 }
3090                 if (clone_inuse) {
3091                         dsl_dataset_activate_feature(origin_head->ds_object,
3092                             f, tx);
3093                         origin_head->ds_feature_inuse[f] = B_TRUE;
3094                 }
3095                 if (origin_head_inuse) {
3096                         dsl_dataset_activate_feature(clone->ds_object, f, tx);
3097                         clone->ds_feature_inuse[f] = B_TRUE;
3098                 }
3099         }
3100
3101         dmu_buf_will_dirty(clone->ds_dbuf, tx);
3102         dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3103
3104         if (clone->ds_objset != NULL) {
3105                 dmu_objset_evict(clone->ds_objset);
3106                 clone->ds_objset = NULL;
3107         }
3108
3109         if (origin_head->ds_objset != NULL) {
3110                 dmu_objset_evict(origin_head->ds_objset);
3111                 origin_head->ds_objset = NULL;
3112         }
3113
3114         unused_refres_delta =
3115             (int64_t)MIN(origin_head->ds_reserved,
3116             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3117             (int64_t)MIN(origin_head->ds_reserved,
3118             dsl_dataset_phys(clone)->ds_unique_bytes);
3119
3120         /*
3121          * Reset origin's unique bytes, if it exists.
3122          */
3123         if (clone->ds_prev) {
3124                 dsl_dataset_t *origin = clone->ds_prev;
3125                 uint64_t comp, uncomp;
3126
3127                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3128                 dsl_deadlist_space_range(&clone->ds_deadlist,
3129                     dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3130                     &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3131         }
3132
3133         /* swap blkptrs */
3134         {
3135                 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
3136                 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
3137                 blkptr_t tmp;
3138                 tmp = dsl_dataset_phys(origin_head)->ds_bp;
3139                 dsl_dataset_phys(origin_head)->ds_bp =
3140                     dsl_dataset_phys(clone)->ds_bp;
3141                 dsl_dataset_phys(clone)->ds_bp = tmp;
3142                 rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
3143                 rrw_exit(&clone->ds_bp_rwlock, FTAG);
3144         }
3145
3146         /* set dd_*_bytes */
3147         {
3148                 int64_t dused, dcomp, duncomp;
3149                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3150                 uint64_t odl_used, odl_comp, odl_uncomp;
3151
3152                 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3153                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
3154
3155                 dsl_deadlist_space(&clone->ds_deadlist,
3156                     &cdl_used, &cdl_comp, &cdl_uncomp);
3157                 dsl_deadlist_space(&origin_head->ds_deadlist,
3158                     &odl_used, &odl_comp, &odl_uncomp);
3159
3160                 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3161                     cdl_used -
3162                     (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3163                     odl_used);
3164                 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3165                     cdl_comp -
3166                     (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3167                     odl_comp);
3168                 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3169                     cdl_uncomp -
3170                     (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3171                     odl_uncomp);
3172
3173                 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3174                     dused, dcomp, duncomp, tx);
3175                 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3176                     -dused, -dcomp, -duncomp, tx);
3177
3178                 /*
3179                  * The difference in the space used by snapshots is the
3180                  * difference in snapshot space due to the head's
3181                  * deadlist (since that's the only thing that's
3182                  * changing that affects the snapused).
3183                  */
3184                 dsl_deadlist_space_range(&clone->ds_deadlist,
3185                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3186                     &cdl_used, &cdl_comp, &cdl_uncomp);
3187                 dsl_deadlist_space_range(&origin_head->ds_deadlist,
3188                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3189                     &odl_used, &odl_comp, &odl_uncomp);
3190                 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3191                     DD_USED_HEAD, DD_USED_SNAP, NULL);
3192         }
3193
3194         /* swap ds_*_bytes */
3195         SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3196             dsl_dataset_phys(clone)->ds_referenced_bytes);
3197         SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3198             dsl_dataset_phys(clone)->ds_compressed_bytes);
3199         SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3200             dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3201         SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3202             dsl_dataset_phys(clone)->ds_unique_bytes);
3203
3204         /* apply any parent delta for change in unconsumed refreservation */
3205         dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3206             unused_refres_delta, 0, 0, tx);
3207
3208         /*
3209          * Swap deadlists.
3210          */
3211         dsl_deadlist_close(&clone->ds_deadlist);
3212         dsl_deadlist_close(&origin_head->ds_deadlist);
3213         SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3214             dsl_dataset_phys(clone)->ds_deadlist_obj);
3215         dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3216             dsl_dataset_phys(clone)->ds_deadlist_obj);
3217         dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3218             dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3219
3220         dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3221
3222         spa_history_log_internal_ds(clone, "clone swap", tx,
3223             "parent=%s", origin_head->ds_dir->dd_myname);
3224 }
3225
3226 /*
3227  * Given a pool name and a dataset object number in that pool,
3228  * return the name of that dataset.
3229  */
3230 int
3231 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3232 {
3233         dsl_pool_t *dp;
3234         dsl_dataset_t *ds;
3235         int error;
3236
3237         error = dsl_pool_hold(pname, FTAG, &dp);
3238         if (error != 0)
3239                 return (error);
3240
3241         error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3242         if (error == 0) {
3243                 dsl_dataset_name(ds, buf);
3244                 dsl_dataset_rele(ds, FTAG);
3245         }
3246         dsl_pool_rele(dp, FTAG);
3247
3248         return (error);
3249 }
3250
3251 int
3252 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3253     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3254 {
3255         int error = 0;
3256
3257         ASSERT3S(asize, >, 0);
3258
3259         /*
3260          * *ref_rsrv is the portion of asize that will come from any
3261          * unconsumed refreservation space.
3262          */
3263         *ref_rsrv = 0;
3264
3265         mutex_enter(&ds->ds_lock);
3266         /*
3267          * Make a space adjustment for reserved bytes.
3268          */
3269         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3270                 ASSERT3U(*used, >=,
3271                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3272                 *used -=
3273                     (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3274                 *ref_rsrv =
3275                     asize - MIN(asize, parent_delta(ds, asize + inflight));
3276         }
3277
3278         if (!check_quota || ds->ds_quota == 0) {
3279                 mutex_exit(&ds->ds_lock);
3280                 return (0);
3281         }
3282         /*
3283          * If they are requesting more space, and our current estimate
3284          * is over quota, they get to try again unless the actual
3285          * on-disk is over quota and there are no pending changes (which
3286          * may free up space for us).
3287          */
3288         if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3289             ds->ds_quota) {
3290                 if (inflight > 0 ||
3291                     dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3292                         error = SET_ERROR(ERESTART);
3293                 else
3294                         error = SET_ERROR(EDQUOT);
3295         }
3296         mutex_exit(&ds->ds_lock);
3297
3298         return (error);
3299 }
3300
3301 typedef struct dsl_dataset_set_qr_arg {
3302         const char *ddsqra_name;
3303         zprop_source_t ddsqra_source;
3304         uint64_t ddsqra_value;
3305 } dsl_dataset_set_qr_arg_t;
3306
3307
3308 /* ARGSUSED */
3309 static int
3310 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3311 {
3312         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3313         dsl_pool_t *dp = dmu_tx_pool(tx);
3314         dsl_dataset_t *ds;
3315         int error;
3316         uint64_t newval;
3317
3318         if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3319                 return (SET_ERROR(ENOTSUP));
3320
3321         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3322         if (error != 0)
3323                 return (error);
3324
3325         if (ds->ds_is_snapshot) {
3326                 dsl_dataset_rele(ds, FTAG);
3327                 return (SET_ERROR(EINVAL));
3328         }
3329
3330         error = dsl_prop_predict(ds->ds_dir,
3331             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3332             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3333         if (error != 0) {
3334                 dsl_dataset_rele(ds, FTAG);
3335                 return (error);
3336         }
3337
3338         if (newval == 0) {
3339                 dsl_dataset_rele(ds, FTAG);
3340                 return (0);
3341         }
3342
3343         if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3344             newval < ds->ds_reserved) {
3345                 dsl_dataset_rele(ds, FTAG);
3346                 return (SET_ERROR(ENOSPC));
3347         }
3348
3349         dsl_dataset_rele(ds, FTAG);
3350         return (0);
3351 }
3352
3353 static void
3354 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3355 {
3356         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3357         dsl_pool_t *dp = dmu_tx_pool(tx);
3358         dsl_dataset_t *ds;
3359         uint64_t newval;
3360
3361         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3362
3363         dsl_prop_set_sync_impl(ds,
3364             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3365             ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3366             &ddsqra->ddsqra_value, tx);
3367
3368         VERIFY0(dsl_prop_get_int_ds(ds,
3369             zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3370
3371         if (ds->ds_quota != newval) {
3372                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3373                 ds->ds_quota = newval;
3374         }
3375         dsl_dataset_rele(ds, FTAG);
3376 }
3377
3378 int
3379 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3380     uint64_t refquota)
3381 {
3382         dsl_dataset_set_qr_arg_t ddsqra;
3383
3384         ddsqra.ddsqra_name = dsname;
3385         ddsqra.ddsqra_source = source;
3386         ddsqra.ddsqra_value = refquota;
3387
3388         return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3389             dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3390 }
3391
3392 static int
3393 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3394 {
3395         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3396         dsl_pool_t *dp = dmu_tx_pool(tx);
3397         dsl_dataset_t *ds;
3398         int error;
3399         uint64_t newval, unique;
3400
3401         if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3402                 return (SET_ERROR(ENOTSUP));
3403
3404         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3405         if (error != 0)
3406                 return (error);
3407
3408         if (ds->ds_is_snapshot) {
3409                 dsl_dataset_rele(ds, FTAG);
3410                 return (SET_ERROR(EINVAL));
3411         }
3412
3413         error = dsl_prop_predict(ds->ds_dir,
3414             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3415             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3416         if (error != 0) {
3417                 dsl_dataset_rele(ds, FTAG);
3418                 return (error);
3419         }
3420
3421         /*
3422          * If we are doing the preliminary check in open context, the
3423          * space estimates may be inaccurate.
3424          */
3425         if (!dmu_tx_is_syncing(tx)) {
3426                 dsl_dataset_rele(ds, FTAG);
3427                 return (0);
3428         }
3429
3430         mutex_enter(&ds->ds_lock);
3431         if (!DS_UNIQUE_IS_ACCURATE(ds))
3432                 dsl_dataset_recalc_head_uniq(ds);
3433         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3434         mutex_exit(&ds->ds_lock);
3435
3436         if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3437                 uint64_t delta = MAX(unique, newval) -
3438                     MAX(unique, ds->ds_reserved);
3439
3440                 if (delta >
3441                     dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3442                     (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3443                         dsl_dataset_rele(ds, FTAG);
3444                         return (SET_ERROR(ENOSPC));
3445                 }
3446         }
3447
3448         dsl_dataset_rele(ds, FTAG);
3449         return (0);
3450 }
3451
3452 void
3453 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3454     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3455 {
3456         uint64_t newval;
3457         uint64_t unique;
3458         int64_t delta;
3459
3460         dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3461             source, sizeof (value), 1, &value, tx);
3462
3463         VERIFY0(dsl_prop_get_int_ds(ds,
3464             zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3465
3466         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3467         mutex_enter(&ds->ds_dir->dd_lock);
3468         mutex_enter(&ds->ds_lock);
3469         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3470         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3471         delta = MAX(0, (int64_t)(newval - unique)) -
3472             MAX(0, (int64_t)(ds->ds_reserved - unique));
3473         ds->ds_reserved = newval;
3474         mutex_exit(&ds->ds_lock);
3475
3476         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3477         mutex_exit(&ds->ds_dir->dd_lock);
3478 }
3479
3480 static void
3481 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3482 {
3483         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3484         dsl_pool_t *dp = dmu_tx_pool(tx);
3485         dsl_dataset_t *ds;
3486
3487         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3488         dsl_dataset_set_refreservation_sync_impl(ds,
3489             ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3490         dsl_dataset_rele(ds, FTAG);
3491 }
3492
3493 int
3494 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3495     uint64_t refreservation)
3496 {
3497         dsl_dataset_set_qr_arg_t ddsqra;
3498
3499         ddsqra.ddsqra_name = dsname;
3500         ddsqra.ddsqra_source = source;
3501         ddsqra.ddsqra_value = refreservation;
3502
3503         return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3504             dsl_dataset_set_refreservation_sync, &ddsqra,
3505             0, ZFS_SPACE_CHECK_NONE));
3506 }
3507
3508 /*
3509  * Return (in *usedp) the amount of space written in new that is not
3510  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3511  * a snapshot before new, in new's filesystem (or its origin).  If not then
3512  * fail and return EINVAL.
3513  *
3514  * The written space is calculated by considering two components:  First, we
3515  * ignore any freed space, and calculate the written as new's used space
3516  * minus old's used space.  Next, we add in the amount of space that was freed
3517  * between the two snapshots, thus reducing new's used space relative to old's.
3518  * Specifically, this is the space that was born before old->ds_creation_txg,
3519  * and freed before new (ie. on new's deadlist or a previous deadlist).
3520  *
3521  * space freed                         [---------------------]
3522  * snapshots                       ---O-------O--------O-------O------
3523  *                                         oldsnap            new
3524  */
3525 int
3526 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3527     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3528 {
3529         int err = 0;
3530         uint64_t snapobj;
3531         dsl_pool_t *dp = new->ds_dir->dd_pool;
3532
3533         ASSERT(dsl_pool_config_held(dp));
3534
3535         *usedp = 0;
3536         *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3537         *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3538
3539         *compp = 0;
3540         *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3541         *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3542
3543         *uncompp = 0;
3544         *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3545         *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3546
3547         snapobj = new->ds_object;
3548         while (snapobj != oldsnap->ds_object) {
3549                 dsl_dataset_t *snap;
3550                 uint64_t used, comp, uncomp;
3551
3552                 if (snapobj == new->ds_object) {
3553                         snap = new;
3554                 } else {
3555                         err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3556                         if (err != 0)
3557                                 break;
3558                 }
3559
3560                 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3561                     dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3562                         /*
3563                          * The blocks in the deadlist can not be born after
3564                          * ds_prev_snap_txg, so get the whole deadlist space,
3565                          * which is more efficient (especially for old-format
3566                          * deadlists).  Unfortunately the deadlist code
3567                          * doesn't have enough information to make this
3568                          * optimization itself.
3569                          */
3570                         dsl_deadlist_space(&snap->ds_deadlist,
3571                             &used, &comp, &uncomp);
3572                 } else {
3573                         dsl_deadlist_space_range(&snap->ds_deadlist,
3574                             0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3575                             &used, &comp, &uncomp);
3576                 }
3577                 *usedp += used;
3578                 *compp += comp;
3579                 *uncompp += uncomp;
3580
3581                 /*
3582                  * If we get to the beginning of the chain of snapshots
3583                  * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3584                  * was not a snapshot of/before new.
3585                  */
3586                 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3587                 if (snap != new)
3588                         dsl_dataset_rele(snap, FTAG);
3589                 if (snapobj == 0) {
3590                         err = SET_ERROR(EINVAL);
3591                         break;
3592                 }
3593
3594         }
3595         return (err);
3596 }
3597
3598 /*
3599  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3600  * lastsnap, and all snapshots in between are deleted.
3601  *
3602  * blocks that would be freed            [---------------------------]
3603  * snapshots                       ---O-------O--------O-------O--------O
3604  *                                        firstsnap        lastsnap
3605  *
3606  * This is the set of blocks that were born after the snap before firstsnap,
3607  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3608  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3609  * We calculate this by iterating over the relevant deadlists (from the snap
3610  * after lastsnap, backward to the snap after firstsnap), summing up the
3611  * space on the deadlist that was born after the snap before firstsnap.
3612  */
3613 int
3614 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3615     dsl_dataset_t *lastsnap,
3616     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3617 {
3618         int err = 0;
3619         uint64_t snapobj;
3620         dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3621
3622         ASSERT(firstsnap->ds_is_snapshot);
3623         ASSERT(lastsnap->ds_is_snapshot);
3624
3625         /*
3626          * Check that the snapshots are in the same dsl_dir, and firstsnap
3627          * is before lastsnap.
3628          */
3629         if (firstsnap->ds_dir != lastsnap->ds_dir ||
3630             dsl_dataset_phys(firstsnap)->ds_creation_txg >
3631             dsl_dataset_phys(lastsnap)->ds_creation_txg)
3632                 return (SET_ERROR(EINVAL));
3633
3634         *usedp = *compp = *uncompp = 0;
3635
3636         snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3637         while (snapobj != firstsnap->ds_object) {
3638                 dsl_dataset_t *ds;
3639                 uint64_t used, comp, uncomp;
3640
3641                 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3642                 if (err != 0)
3643                         break;
3644
3645                 dsl_deadlist_space_range(&ds->ds_deadlist,
3646                     dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3647                     &used, &comp, &uncomp);
3648                 *usedp += used;
3649                 *compp += comp;
3650                 *uncompp += uncomp;
3651
3652                 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3653                 ASSERT3U(snapobj, !=, 0);
3654                 dsl_dataset_rele(ds, FTAG);
3655         }
3656         return (err);
3657 }
3658
3659 /*
3660  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3661  * For example, they could both be snapshots of the same filesystem, and
3662  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3663  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3664  * filesystem.  Or 'earlier' could be the origin's origin.
3665  *
3666  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3667  */
3668 boolean_t
3669 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3670     uint64_t earlier_txg)
3671 {
3672         dsl_pool_t *dp = later->ds_dir->dd_pool;
3673         int error;
3674         boolean_t ret;
3675
3676         ASSERT(dsl_pool_config_held(dp));
3677         ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3678
3679         if (earlier_txg == 0)
3680                 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3681
3682         if (later->ds_is_snapshot &&
3683             earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3684                 return (B_FALSE);
3685
3686         if (later->ds_dir == earlier->ds_dir)
3687                 return (B_TRUE);
3688         if (!dsl_dir_is_clone(later->ds_dir))
3689                 return (B_FALSE);
3690
3691         if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3692                 return (B_TRUE);
3693         dsl_dataset_t *origin;
3694         error = dsl_dataset_hold_obj(dp,
3695             dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3696         if (error != 0)
3697                 return (B_FALSE);
3698         ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3699         dsl_dataset_rele(origin, FTAG);
3700         return (ret);
3701 }
3702
3703 void
3704 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3705 {
3706         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3707         dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3708 }
3709
3710 boolean_t
3711 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3712 {
3713         dmu_object_info_t doi;
3714
3715         dmu_object_info_from_db(ds->ds_dbuf, &doi);
3716         return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3717 }
3718
3719 boolean_t
3720 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3721 {
3722         return (dsl_dataset_is_zapified(ds) &&
3723             zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3724             ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3725 }