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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org>
  24  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
  25  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2014 RackTop Systems.
  27  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  28  * Copyright (c) 2014 Integros [integros.com]
  29  * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
  30  */
  31
  32 #include <sys/dmu_objset.h>
  33 #include <sys/dsl_dataset.h>
  34 #include <sys/dsl_dir.h>
  35 #include <sys/dsl_prop.h>
  36 #include <sys/dsl_synctask.h>
  37 #include <sys/dmu_traverse.h>
  38 #include <sys/dmu_impl.h>
  39 #include <sys/dmu_send.h>
  40 #include <sys/dmu_tx.h>
  41 #include <sys/arc.h>
  42 #include <sys/zio.h>
  43 #include <sys/zap.h>
  44 #include <sys/zfeature.h>
  45 #include <sys/unique.h>
  46 #include <sys/zfs_context.h>
  47 #include <sys/zfs_ioctl.h>
  48 #include <sys/spa.h>
  49 #include <sys/zfs_znode.h>
  50 #include <sys/zfs_onexit.h>
  51 #include <sys/zvol.h>
  52 #include <sys/dsl_scan.h>
  53 #include <sys/dsl_deadlist.h>
  54 #include <sys/dsl_destroy.h>
  55 #include <sys/dsl_userhold.h>
  56 #include <sys/dsl_bookmark.h>
  57 #include <sys/dmu_send.h>
  58 #include <sys/zio_checksum.h>
  59 #include <sys/zio_compress.h>
  60 #include <zfs_fletcher.h>
  61
  62 SYSCTL_DECL(_vfs_zfs);
  63
  64 /*
  65  * The SPA supports block sizes up to 16MB.  However, very large blocks
  66  * can have an impact on i/o latency (e.g. tying up a spinning disk for
  67  * ~300ms), and also potentially on the memory allocator.  Therefore,
  68  * we do not allow the recordsize to be set larger than zfs_max_recordsize
  69  * (default 1MB).  Larger blocks can be created by changing this tunable,
  70  * and pools with larger blocks can always be imported and used, regardless
  71  * of this setting.
  72  */
  73 int zfs_max_recordsize = 1 * 1024 * 1024;
  74 SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN,
  75     &zfs_max_recordsize, 0,
  76     "Maximum block size.  Expect dragons when tuning this.");
  77
  78 #define SWITCH64(x, y) \
  79         { \
  80                 uint64_t __tmp = (x); \
  81                 (x) = (y); \
  82                 (y) = __tmp; \
  83         }
  84
  85 #define DS_REF_MAX      (1ULL << 62)
  86
  87 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
  88
  89 extern int spa_asize_inflation;
  90
  91 /*
  92  * Figure out how much of this delta should be propogated to the dsl_dir
  93  * layer.  If there's a refreservation, that space has already been
  94  * partially accounted for in our ancestors.
  95  */
  96 static int64_t
  97 parent_delta(dsl_dataset_t *ds, int64_t delta)
  98 {
  99         dsl_dataset_phys_t *ds_phys;
 100         uint64_t old_bytes, new_bytes;
 101
 102         if (ds->ds_reserved == 0)
 103                 return (delta);
 104
 105         ds_phys = dsl_dataset_phys(ds);
 106         old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
 107         new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
 108
 109         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
 110         return (new_bytes - old_bytes);
 111 }
 112
 113 void
 114 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
 115 {
 116         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 117         int compressed = BP_GET_PSIZE(bp);
 118         int uncompressed = BP_GET_UCSIZE(bp);
 119         int64_t delta;
 120
 121         dprintf_bp(bp, "ds=%p", ds);
 122
 123         ASSERT(dmu_tx_is_syncing(tx));
 124         /* It could have been compressed away to nothing */
 125         if (BP_IS_HOLE(bp))
 126                 return;
 127         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 128         ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 129         if (ds == NULL) {
 130                 dsl_pool_mos_diduse_space(tx->tx_pool,
 131                     used, compressed, uncompressed);
 132                 return;
 133         }
 134
 135         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 136         mutex_enter(&ds->ds_lock);
 137         delta = parent_delta(ds, used);
 138         dsl_dataset_phys(ds)->ds_referenced_bytes += used;
 139         dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
 140         dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
 141         dsl_dataset_phys(ds)->ds_unique_bytes += used;
 142
 143         if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
 144                 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
 145                     B_TRUE;
 146         }
 147
 148         spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
 149         if (f != SPA_FEATURE_NONE)
 150                 ds->ds_feature_activation_needed[f] = B_TRUE;
 151
 152         mutex_exit(&ds->ds_lock);
 153         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 154             compressed, uncompressed, tx);
 155         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 156             DD_USED_REFRSRV, DD_USED_HEAD, NULL);
 157 }
 158
 159 int
 160 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 161     boolean_t async)
 162 {
 163         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 164         int compressed = BP_GET_PSIZE(bp);
 165         int uncompressed = BP_GET_UCSIZE(bp);
 166
 167         if (BP_IS_HOLE(bp))
 168                 return (0);
 169
 170         ASSERT(dmu_tx_is_syncing(tx));
 171         ASSERT(bp->blk_birth <= tx->tx_txg);
 172
 173         if (ds == NULL) {
 174                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 175                 dsl_pool_mos_diduse_space(tx->tx_pool,
 176                     -used, -compressed, -uncompressed);
 177                 return (used);
 178         }
 179         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 180
 181         ASSERT(!ds->ds_is_snapshot);
 182         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 183
 184         if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
 185                 int64_t delta;
 186
 187                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 188                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 189
 190                 mutex_enter(&ds->ds_lock);
 191                 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
 192                     !DS_UNIQUE_IS_ACCURATE(ds));
 193                 delta = parent_delta(ds, -used);
 194                 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
 195                 mutex_exit(&ds->ds_lock);
 196                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 197                     delta, -compressed, -uncompressed, tx);
 198                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 199                     DD_USED_REFRSRV, DD_USED_HEAD, NULL);
 200         } else {
 201                 dprintf_bp(bp, "putting on dead list: %s", "");
 202                 if (async) {
 203                         /*
 204                          * We are here as part of zio's write done callback,
 205                          * which means we're a zio interrupt thread.  We can't
 206                          * call dsl_deadlist_insert() now because it may block
 207                          * waiting for I/O.  Instead, put bp on the deferred
 208                          * queue and let dsl_pool_sync() finish the job.
 209                          */
 210                         bplist_append(&ds->ds_pending_deadlist, bp);
 211                 } else {
 212                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 213                 }
 214                 ASSERT3U(ds->ds_prev->ds_object, ==,
 215                     dsl_dataset_phys(ds)->ds_prev_snap_obj);
 216                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
 217                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 218                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
 219                     ds->ds_object && bp->blk_birth >
 220                     dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
 221                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 222                         mutex_enter(&ds->ds_prev->ds_lock);
 223                         dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
 224                         mutex_exit(&ds->ds_prev->ds_lock);
 225                 }
 226                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 227                         dsl_dir_transfer_space(ds->ds_dir, used,
 228                             DD_USED_HEAD, DD_USED_SNAP, tx);
 229                 }
 230         }
 231         mutex_enter(&ds->ds_lock);
 232         ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
 233         dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
 234         ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
 235         dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
 236         ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
 237         dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
 238         mutex_exit(&ds->ds_lock);
 239
 240         return (used);
 241 }
 242
 243 uint64_t
 244 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
 245 {
 246         uint64_t trysnap = 0;
 247
 248         if (ds == NULL)
 249                 return (0);
 250         /*
 251          * The snapshot creation could fail, but that would cause an
 252          * incorrect FALSE return, which would only result in an
 253          * overestimation of the amount of space that an operation would
 254          * consume, which is OK.
 255          *
 256          * There's also a small window where we could miss a pending
 257          * snapshot, because we could set the sync task in the quiescing
 258          * phase.  So this should only be used as a guess.
 259          */
 260         if (ds->ds_trysnap_txg >
 261             spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
 262                 trysnap = ds->ds_trysnap_txg;
 263         return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
 264 }
 265
 266 boolean_t
 267 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
 268     uint64_t blk_birth)
 269 {
 270         if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
 271             (bp != NULL && BP_IS_HOLE(bp)))
 272                 return (B_FALSE);
 273
 274         ddt_prefetch(dsl_dataset_get_spa(ds), bp);
 275
 276         return (B_TRUE);
 277 }
 278
 279 static void
 280 dsl_dataset_evict(void *dbu)
 281 {
 282         dsl_dataset_t *ds = dbu;
 283
 284         ASSERT(ds->ds_owner == NULL);
 285
 286         ds->ds_dbuf = NULL;
 287
 288         unique_remove(ds->ds_fsid_guid);
 289
 290         if (ds->ds_objset != NULL)
 291                 dmu_objset_evict(ds->ds_objset);
 292
 293         if (ds->ds_prev) {
 294                 dsl_dataset_rele(ds->ds_prev, ds);
 295                 ds->ds_prev = NULL;
 296         }
 297
 298         bplist_destroy(&ds->ds_pending_deadlist);
 299         if (ds->ds_deadlist.dl_os != NULL)
 300                 dsl_deadlist_close(&ds->ds_deadlist);
 301         if (ds->ds_dir)
 302                 dsl_dir_async_rele(ds->ds_dir, ds);
 303
 304         ASSERT(!list_link_active(&ds->ds_synced_link));
 305
 306         list_destroy(&ds->ds_prop_cbs);
 307         if (mutex_owned(&ds->ds_lock))
 308                 mutex_exit(&ds->ds_lock);
 309         mutex_destroy(&ds->ds_lock);
 310         if (mutex_owned(&ds->ds_opening_lock))
 311                 mutex_exit(&ds->ds_opening_lock);
 312         mutex_destroy(&ds->ds_opening_lock);
 313         mutex_destroy(&ds->ds_sendstream_lock);
 314         refcount_destroy(&ds->ds_longholds);
 315
 316         kmem_free(ds, sizeof (dsl_dataset_t));
 317 }
 318
 319 int
 320 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 321 {
 322         dsl_dataset_phys_t *headphys;
 323         int err;
 324         dmu_buf_t *headdbuf;
 325         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 326         objset_t *mos = dp->dp_meta_objset;
 327
 328         if (ds->ds_snapname[0])
 329                 return (0);
 330         if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
 331                 return (0);
 332
 333         err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
 334             FTAG, &headdbuf);
 335         if (err != 0)
 336                 return (err);
 337         headphys = headdbuf->db_data;
 338         err = zap_value_search(dp->dp_meta_objset,
 339             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 340         dmu_buf_rele(headdbuf, FTAG);
 341         return (err);
 342 }
 343
 344 int
 345 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 346 {
 347         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 348         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 349         matchtype_t mt;
 350         int err;
 351
 352         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 353                 mt = MT_FIRST;
 354         else
 355                 mt = MT_EXACT;
 356
 357         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 358             value, mt, NULL, 0, NULL);
 359         if (err == ENOTSUP && mt == MT_FIRST)
 360                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 361         return (err);
 362 }
 363
 364 int
 365 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
 366     boolean_t adj_cnt)
 367 {
 368         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 369         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 370         matchtype_t mt;
 371         int err;
 372
 373         dsl_dir_snap_cmtime_update(ds->ds_dir);
 374
 375         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 376                 mt = MT_FIRST;
 377         else
 378                 mt = MT_EXACT;
 379
 380         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 381         if (err == ENOTSUP && mt == MT_FIRST)
 382                 err = zap_remove(mos, snapobj, name, tx);
 383
 384         if (err == 0 && adj_cnt)
 385                 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
 386                     DD_FIELD_SNAPSHOT_COUNT, tx);
 387
 388         return (err);
 389 }
 390
 391 boolean_t
 392 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
 393 {
 394         dmu_buf_t *dbuf = ds->ds_dbuf;
 395         boolean_t result = B_FALSE;
 396
 397         if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
 398             ds->ds_object, DMU_BONUS_BLKID, tag)) {
 399
 400                 if (ds == dmu_buf_get_user(dbuf))
 401                         result = B_TRUE;
 402                 else
 403                         dmu_buf_rele(dbuf, tag);
 404         }
 405
 406         return (result);
 407 }
 408
 409 int
 410 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 411     dsl_dataset_t **dsp)
 412 {
 413         objset_t *mos = dp->dp_meta_objset;
 414         dmu_buf_t *dbuf;
 415         dsl_dataset_t *ds;
 416         int err;
 417         dmu_object_info_t doi;
 418
 419         ASSERT(dsl_pool_config_held(dp));
 420
 421         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 422         if (err != 0)
 423                 return (err);
 424
 425         /* Make sure dsobj has the correct object type. */
 426         dmu_object_info_from_db(dbuf, &doi);
 427         if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
 428                 dmu_buf_rele(dbuf, tag);
 429                 return (SET_ERROR(EINVAL));
 430         }
 431
 432         ds = dmu_buf_get_user(dbuf);
 433         if (ds == NULL) {
 434                 dsl_dataset_t *winner = NULL;
 435
 436                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 437                 ds->ds_dbuf = dbuf;
 438                 ds->ds_object = dsobj;
 439                 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
 440
 441                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 442                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 443                 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
 444                 refcount_create(&ds->ds_longholds);
 445
 446                 bplist_create(&ds->ds_pending_deadlist);
 447                 dsl_deadlist_open(&ds->ds_deadlist,
 448                     mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
 449
 450                 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
 451                     offsetof(dmu_sendarg_t, dsa_link));
 452
 453                 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
 454                     offsetof(dsl_prop_cb_record_t, cbr_ds_node));
 455
 456                 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 457                         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 458                                 if (!(spa_feature_table[f].fi_flags &
 459                                     ZFEATURE_FLAG_PER_DATASET))
 460                                         continue;
 461                                 err = zap_contains(mos, dsobj,
 462                                     spa_feature_table[f].fi_guid);
 463                                 if (err == 0) {
 464                                         ds->ds_feature_inuse[f] = B_TRUE;
 465                                 } else {
 466                                         ASSERT3U(err, ==, ENOENT);
 467                                         err = 0;
 468                                 }
 469                         }
 470                 }
 471
 472                 err = dsl_dir_hold_obj(dp,
 473                     dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
 474                 if (err != 0) {
 475                         mutex_destroy(&ds->ds_lock);
 476                         mutex_destroy(&ds->ds_opening_lock);
 477                         mutex_destroy(&ds->ds_sendstream_lock);
 478                         refcount_destroy(&ds->ds_longholds);
 479                         bplist_destroy(&ds->ds_pending_deadlist);
 480                         dsl_deadlist_close(&ds->ds_deadlist);
 481                         kmem_free(ds, sizeof (dsl_dataset_t));
 482                         dmu_buf_rele(dbuf, tag);
 483                         return (err);
 484                 }
 485
 486                 if (!ds->ds_is_snapshot) {
 487                         ds->ds_snapname[0] = '\0';
 488                         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 489                                 err = dsl_dataset_hold_obj(dp,
 490                                     dsl_dataset_phys(ds)->ds_prev_snap_obj,
 491                                     ds, &ds->ds_prev);
 492                         }
 493                         if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 494                                 int zaperr = zap_lookup(mos, ds->ds_object,
 495                                     DS_FIELD_BOOKMARK_NAMES,
 496                                     sizeof (ds->ds_bookmarks), 1,
 497                                     &ds->ds_bookmarks);
 498                                 if (zaperr != ENOENT)
 499                                         VERIFY0(zaperr);
 500                         }
 501                 } else {
 502                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 503                                 err = dsl_dataset_get_snapname(ds);
 504                         if (err == 0 &&
 505                             dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
 506                                 err = zap_count(
 507                                     ds->ds_dir->dd_pool->dp_meta_objset,
 508                                     dsl_dataset_phys(ds)->ds_userrefs_obj,
 509                                     &ds->ds_userrefs);
 510                         }
 511                 }
 512
 513                 if (err == 0 && !ds->ds_is_snapshot) {
 514                         err = dsl_prop_get_int_ds(ds,
 515                             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
 516                             &ds->ds_reserved);
 517                         if (err == 0) {
 518                                 err = dsl_prop_get_int_ds(ds,
 519                                     zfs_prop_to_name(ZFS_PROP_REFQUOTA),
 520                                     &ds->ds_quota);
 521                         }
 522                 } else {
 523                         ds->ds_reserved = ds->ds_quota = 0;
 524                 }
 525
 526                 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
 527                 if (err == 0)
 528                         winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
 529
 530                 if (err != 0 || winner != NULL) {
 531                         bplist_destroy(&ds->ds_pending_deadlist);
 532                         dsl_deadlist_close(&ds->ds_deadlist);
 533                         if (ds->ds_prev)
 534                                 dsl_dataset_rele(ds->ds_prev, ds);
 535                         dsl_dir_rele(ds->ds_dir, ds);
 536                         mutex_destroy(&ds->ds_lock);
 537                         mutex_destroy(&ds->ds_opening_lock);
 538                         mutex_destroy(&ds->ds_sendstream_lock);
 539                         refcount_destroy(&ds->ds_longholds);
 540                         kmem_free(ds, sizeof (dsl_dataset_t));
 541                         if (err != 0) {
 542                                 dmu_buf_rele(dbuf, tag);
 543                                 return (err);
 544                         }
 545                         ds = winner;
 546                 } else {
 547                         ds->ds_fsid_guid =
 548                             unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
 549                 }
 550         }
 551         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 552         ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
 553         ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
 554             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 555             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 556         *dsp = ds;
 557         return (0);
 558 }
 559
 560 int
 561 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
 562     void *tag, dsl_dataset_t **dsp)
 563 {
 564         dsl_dir_t *dd;
 565         const char *snapname;
 566         uint64_t obj;
 567         int err = 0;
 568         dsl_dataset_t *ds;
 569
 570         err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
 571         if (err != 0)
 572                 return (err);
 573
 574         ASSERT(dsl_pool_config_held(dp));
 575         obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
 576         if (obj != 0)
 577                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
 578         else
 579                 err = SET_ERROR(ENOENT);
 580
 581         /* we may be looking for a snapshot */
 582         if (err == 0 && snapname != NULL) {
 583                 dsl_dataset_t *snap_ds;
 584
 585                 if (*snapname++ != '@') {
 586                         dsl_dataset_rele(ds, tag);
 587                         dsl_dir_rele(dd, FTAG);
 588                         return (SET_ERROR(ENOENT));
 589                 }
 590
 591                 dprintf("looking for snapshot '%s'\n", snapname);
 592                 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
 593                 if (err == 0)
 594                         err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
 595                 dsl_dataset_rele(ds, tag);
 596
 597                 if (err == 0) {
 598                         mutex_enter(&snap_ds->ds_lock);
 599                         if (snap_ds->ds_snapname[0] == 0)
 600                                 (void) strlcpy(snap_ds->ds_snapname, snapname,
 601                                     sizeof (snap_ds->ds_snapname));
 602                         mutex_exit(&snap_ds->ds_lock);
 603                         ds = snap_ds;
 604                 }
 605         }
 606         if (err == 0)
 607                 *dsp = ds;
 608         dsl_dir_rele(dd, FTAG);
 609         return (err);
 610 }
 611
 612 int
 613 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
 614     void *tag, dsl_dataset_t **dsp)
 615 {
 616         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 617         if (err != 0)
 618                 return (err);
 619         if (!dsl_dataset_tryown(*dsp, tag)) {
 620                 dsl_dataset_rele(*dsp, tag);
 621                 *dsp = NULL;
 622                 return (SET_ERROR(EBUSY));
 623         }
 624         return (0);
 625 }
 626
 627 int
 628 dsl_dataset_own(dsl_pool_t *dp, const char *name,
 629     void *tag, dsl_dataset_t **dsp)
 630 {
 631         int err = dsl_dataset_hold(dp, name, tag, dsp);
 632         if (err != 0)
 633                 return (err);
 634         if (!dsl_dataset_tryown(*dsp, tag)) {
 635                 dsl_dataset_rele(*dsp, tag);
 636                 return (SET_ERROR(EBUSY));
 637         }
 638         return (0);
 639 }
 640
 641 /*
 642  * See the comment above dsl_pool_hold() for details.  In summary, a long
 643  * hold is used to prevent destruction of a dataset while the pool hold
 644  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
 645  *
 646  * The dataset and pool must be held when this function is called.  After it
 647  * is called, the pool hold may be released while the dataset is still held
 648  * and accessed.
 649  */
 650 void
 651 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
 652 {
 653         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 654         (void) refcount_add(&ds->ds_longholds, tag);
 655 }
 656
 657 void
 658 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
 659 {
 660         (void) refcount_remove(&ds->ds_longholds, tag);
 661 }
 662
 663 /* Return B_TRUE if there are any long holds on this dataset. */
 664 boolean_t
 665 dsl_dataset_long_held(dsl_dataset_t *ds)
 666 {
 667         return (!refcount_is_zero(&ds->ds_longholds));
 668 }
 669
 670 void
 671 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 672 {
 673         if (ds == NULL) {
 674                 (void) strcpy(name, "mos");
 675         } else {
 676                 dsl_dir_name(ds->ds_dir, name);
 677                 VERIFY0(dsl_dataset_get_snapname(ds));
 678                 if (ds->ds_snapname[0]) {
 679                         VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
 680                             <, ZFS_MAX_DATASET_NAME_LEN);
 681                         /*
 682                          * We use a "recursive" mutex so that we
 683                          * can call dprintf_ds() with ds_lock held.
 684                          */
 685                         if (!MUTEX_HELD(&ds->ds_lock)) {
 686                                 mutex_enter(&ds->ds_lock);
 687                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 688                                     ZFS_MAX_DATASET_NAME_LEN), <,
 689                                     ZFS_MAX_DATASET_NAME_LEN);
 690                                 mutex_exit(&ds->ds_lock);
 691                         } else {
 692                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 693                                     ZFS_MAX_DATASET_NAME_LEN), <,
 694                                     ZFS_MAX_DATASET_NAME_LEN);
 695                         }
 696                 }
 697         }
 698 }
 699
 700 int
 701 dsl_dataset_namelen(dsl_dataset_t *ds)
 702 {
 703         VERIFY0(dsl_dataset_get_snapname(ds));
 704         mutex_enter(&ds->ds_lock);
 705         int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
 706         mutex_exit(&ds->ds_lock);
 707         return (len);
 708 }
 709
 710 void
 711 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 712 {
 713         dmu_buf_rele(ds->ds_dbuf, tag);
 714 }
 715
 716 void
 717 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 718 {
 719         ASSERT3P(ds->ds_owner, ==, tag);
 720         ASSERT(ds->ds_dbuf != NULL);
 721
 722         mutex_enter(&ds->ds_lock);
 723         ds->ds_owner = NULL;
 724         mutex_exit(&ds->ds_lock);
 725         dsl_dataset_long_rele(ds, tag);
 726         dsl_dataset_rele(ds, tag);
 727 }
 728
 729 boolean_t
 730 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
 731 {
 732         boolean_t gotit = FALSE;
 733
 734         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 735         mutex_enter(&ds->ds_lock);
 736         if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
 737                 ds->ds_owner = tag;
 738                 dsl_dataset_long_hold(ds, tag);
 739                 gotit = TRUE;
 740         }
 741         mutex_exit(&ds->ds_lock);
 742         return (gotit);
 743 }
 744
 745 boolean_t
 746 dsl_dataset_has_owner(dsl_dataset_t *ds)
 747 {
 748         boolean_t rv;
 749         mutex_enter(&ds->ds_lock);
 750         rv = (ds->ds_owner != NULL);
 751         mutex_exit(&ds->ds_lock);
 752         return (rv);
 753 }
 754
 755 static void
 756 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 757 {
 758         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 759         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 760         uint64_t zero = 0;
 761
 762         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 763
 764         spa_feature_incr(spa, f, tx);
 765         dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
 766
 767         VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
 768             sizeof (zero), 1, &zero, tx));
 769 }
 770
 771 void
 772 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 773 {
 774         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 775         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 776
 777         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 778
 779         VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
 780         spa_feature_decr(spa, f, tx);
 781 }
 782
 783 uint64_t
 784 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 785     uint64_t flags, dmu_tx_t *tx)
 786 {
 787         dsl_pool_t *dp = dd->dd_pool;
 788         dmu_buf_t *dbuf;
 789         dsl_dataset_phys_t *dsphys;
 790         uint64_t dsobj;
 791         objset_t *mos = dp->dp_meta_objset;
 792
 793         if (origin == NULL)
 794                 origin = dp->dp_origin_snap;
 795
 796         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 797         ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
 798         ASSERT(dmu_tx_is_syncing(tx));
 799         ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
 800
 801         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 802             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 803         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 804         dmu_buf_will_dirty(dbuf, tx);
 805         dsphys = dbuf->db_data;
 806         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 807         dsphys->ds_dir_obj = dd->dd_object;
 808         dsphys->ds_flags = flags;
 809         dsphys->ds_fsid_guid = unique_create();
 810         do {
 811                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 812                     sizeof (dsphys->ds_guid));
 813         } while (dsphys->ds_guid == 0);
 814         dsphys->ds_snapnames_zapobj =
 815             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 816             DMU_OT_NONE, 0, tx);
 817         dsphys->ds_creation_time = gethrestime_sec();
 818         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 819
 820         if (origin == NULL) {
 821                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 822         } else {
 823                 dsl_dataset_t *ohds; /* head of the origin snapshot */
 824
 825                 dsphys->ds_prev_snap_obj = origin->ds_object;
 826                 dsphys->ds_prev_snap_txg =
 827                     dsl_dataset_phys(origin)->ds_creation_txg;
 828                 dsphys->ds_referenced_bytes =
 829                     dsl_dataset_phys(origin)->ds_referenced_bytes;
 830                 dsphys->ds_compressed_bytes =
 831                     dsl_dataset_phys(origin)->ds_compressed_bytes;
 832                 dsphys->ds_uncompressed_bytes =
 833                     dsl_dataset_phys(origin)->ds_uncompressed_bytes;
 834                 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
 835
 836                 /*
 837                  * Inherit flags that describe the dataset's contents
 838                  * (INCONSISTENT) or properties (Case Insensitive).
 839                  */
 840                 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
 841                     (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
 842
 843                 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 844                         if (origin->ds_feature_inuse[f])
 845                                 dsl_dataset_activate_feature(dsobj, f, tx);
 846                 }
 847
 848                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 849                 dsl_dataset_phys(origin)->ds_num_children++;
 850
 851                 VERIFY0(dsl_dataset_hold_obj(dp,
 852                     dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
 853                     FTAG, &ohds));
 854                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 855                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 856                 dsl_dataset_rele(ohds, FTAG);
 857
 858                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 859                         if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
 860                                 dsl_dataset_phys(origin)->ds_next_clones_obj =
 861                                     zap_create(mos,
 862                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 863                         }
 864                         VERIFY0(zap_add_int(mos,
 865                             dsl_dataset_phys(origin)->ds_next_clones_obj,
 866                             dsobj, tx));
 867                 }
 868
 869                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 870                 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
 871                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 872                         if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
 873                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 874                                 dsl_dir_phys(origin->ds_dir)->dd_clones =
 875                                     zap_create(mos,
 876                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 877                         }
 878                         VERIFY0(zap_add_int(mos,
 879                             dsl_dir_phys(origin->ds_dir)->dd_clones,
 880                             dsobj, tx));
 881                 }
 882         }
 883
 884         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 885                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 886
 887         dmu_buf_rele(dbuf, FTAG);
 888
 889         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 890         dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
 891
 892         return (dsobj);
 893 }
 894
 895 static void
 896 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
 897 {
 898         objset_t *os;
 899
 900         VERIFY0(dmu_objset_from_ds(ds, &os));
 901         bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 902         dsl_dataset_dirty(ds, tx);
 903 }
 904
 905 uint64_t
 906 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 907     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 908 {
 909         dsl_pool_t *dp = pdd->dd_pool;
 910         uint64_t dsobj, ddobj;
 911         dsl_dir_t *dd;
 912
 913         ASSERT(dmu_tx_is_syncing(tx));
 914         ASSERT(lastname[0] != '@');
 915
 916         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 917         VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
 918
 919         dsobj = dsl_dataset_create_sync_dd(dd, origin,
 920             flags & ~DS_CREATE_FLAG_NODIRTY, tx);
 921
 922         dsl_deleg_set_create_perms(dd, tx, cr);
 923
 924         /*
 925          * Since we're creating a new node we know it's a leaf, so we can
 926          * initialize the counts if the limit feature is active.
 927          */
 928         if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
 929                 uint64_t cnt = 0;
 930                 objset_t *os = dd->dd_pool->dp_meta_objset;
 931
 932                 dsl_dir_zapify(dd, tx);
 933                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
 934                     sizeof (cnt), 1, &cnt, tx));
 935                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
 936                     sizeof (cnt), 1, &cnt, tx));
 937         }
 938
 939         dsl_dir_rele(dd, FTAG);
 940
 941         /*
 942          * If we are creating a clone, make sure we zero out any stale
 943          * data from the origin snapshots zil header.
 944          */
 945         if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
 946                 dsl_dataset_t *ds;
 947
 948                 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 949                 dsl_dataset_zero_zil(ds, tx);
 950                 dsl_dataset_rele(ds, FTAG);
 951         }
 952
 953         return (dsobj);
 954 }
 955
 956 #ifdef __FreeBSD__
 957 /* FreeBSD ioctl compat begin */
 958 struct destroyarg {
 959         nvlist_t *nvl;
 960         const char *snapname;
 961 };
 962
 963 static int
 964 dsl_check_snap_cb(const char *name, void *arg)
 965 {
 966         struct destroyarg *da = arg;
 967         dsl_dataset_t *ds;
 968         char *dsname;
 969
 970         dsname = kmem_asprintf("%s@%s", name, da->snapname);
 971         fnvlist_add_boolean(da->nvl, dsname);
 972         kmem_free(dsname, strlen(dsname) + 1);
 973
 974         return (0);
 975 }
 976
 977 int
 978 dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
 979     nvlist_t *snaps)
 980 {
 981         struct destroyarg *da;
 982         int err;
 983
 984         da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
 985         da->nvl = snaps;
 986         da->snapname = snapname;
 987         err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
 988             DS_FIND_CHILDREN);
 989         kmem_free(da, sizeof (struct destroyarg));
 990
 991         return (err);
 992 }
 993 /* FreeBSD ioctl compat end */
 994 #endif /* __FreeBSD__ */
 995
 996 /*
 997  * The unique space in the head dataset can be calculated by subtracting
 998  * the space used in the most recent snapshot, that is still being used
 999  * in this file system, from the space currently in use.  To figure out
1000  * the space in the most recent snapshot still in use, we need to take
1001  * the total space used in the snapshot and subtract out the space that
1002  * has been freed up since the snapshot was taken.
1003  */
1004 void
1005 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1006 {
1007         uint64_t mrs_used;
1008         uint64_t dlused, dlcomp, dluncomp;
1009
1010         ASSERT(!ds->ds_is_snapshot);
1011
1012         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1013                 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1014         else
1015                 mrs_used = 0;
1016
1017         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1018
1019         ASSERT3U(dlused, <=, mrs_used);
1020         dsl_dataset_phys(ds)->ds_unique_bytes =
1021             dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1022
1023         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1024             SPA_VERSION_UNIQUE_ACCURATE)
1025                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1026 }
1027
1028 void
1029 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1030     dmu_tx_t *tx)
1031 {
1032         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1033         uint64_t count;
1034         int err;
1035
1036         ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1037         err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1038             obj, tx);
1039         /*
1040          * The err should not be ENOENT, but a bug in a previous version
1041          * of the code could cause upgrade_clones_cb() to not set
1042          * ds_next_snap_obj when it should, leading to a missing entry.
1043          * If we knew that the pool was created after
1044          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1045          * ENOENT.  However, at least we can check that we don't have
1046          * too many entries in the next_clones_obj even after failing to
1047          * remove this one.
1048          */
1049         if (err != ENOENT)
1050                 VERIFY0(err);
1051         ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1052             &count));
1053         ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1054 }
1055
1056
1057 blkptr_t *
1058 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1059 {
1060         return (&dsl_dataset_phys(ds)->ds_bp);
1061 }
1062
1063 void
1064 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1065 {
1066         ASSERT(dmu_tx_is_syncing(tx));
1067         /* If it's the meta-objset, set dp_meta_rootbp */
1068         if (ds == NULL) {
1069                 tx->tx_pool->dp_meta_rootbp = *bp;
1070         } else {
1071                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1072                 dsl_dataset_phys(ds)->ds_bp = *bp;
1073         }
1074 }
1075
1076 spa_t *
1077 dsl_dataset_get_spa(dsl_dataset_t *ds)
1078 {
1079         return (ds->ds_dir->dd_pool->dp_spa);
1080 }
1081
1082 void
1083 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1084 {
1085         dsl_pool_t *dp;
1086
1087         if (ds == NULL) /* this is the meta-objset */
1088                 return;
1089
1090         ASSERT(ds->ds_objset != NULL);
1091
1092         if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1093                 panic("dirtying snapshot!");
1094
1095         dp = ds->ds_dir->dd_pool;
1096
1097         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1098                 /* up the hold count until we can be written out */
1099                 dmu_buf_add_ref(ds->ds_dbuf, ds);
1100         }
1101 }
1102
1103 boolean_t
1104 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1105 {
1106         for (int t = 0; t < TXG_SIZE; t++) {
1107                 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1108                     ds, t))
1109                         return (B_TRUE);
1110         }
1111         return (B_FALSE);
1112 }
1113
1114 static int
1115 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1116 {
1117         uint64_t asize;
1118
1119         if (!dmu_tx_is_syncing(tx))
1120                 return (0);
1121
1122         /*
1123          * If there's an fs-only reservation, any blocks that might become
1124          * owned by the snapshot dataset must be accommodated by space
1125          * outside of the reservation.
1126          */
1127         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1128         asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1129         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1130                 return (SET_ERROR(ENOSPC));
1131
1132         /*
1133          * Propagate any reserved space for this snapshot to other
1134          * snapshot checks in this sync group.
1135          */
1136         if (asize > 0)
1137                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1138
1139         return (0);
1140 }
1141
1142 typedef struct dsl_dataset_snapshot_arg {
1143         nvlist_t *ddsa_snaps;
1144         nvlist_t *ddsa_props;
1145         nvlist_t *ddsa_errors;
1146         cred_t *ddsa_cr;
1147 } dsl_dataset_snapshot_arg_t;
1148
1149 int
1150 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1151     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1152 {
1153         int error;
1154         uint64_t value;
1155
1156         ds->ds_trysnap_txg = tx->tx_txg;
1157
1158         if (!dmu_tx_is_syncing(tx))
1159                 return (0);
1160
1161         /*
1162          * We don't allow multiple snapshots of the same txg.  If there
1163          * is already one, try again.
1164          */
1165         if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1166                 return (SET_ERROR(EAGAIN));
1167
1168         /*
1169          * Check for conflicting snapshot name.
1170          */
1171         error = dsl_dataset_snap_lookup(ds, snapname, &value);
1172         if (error == 0)
1173                 return (SET_ERROR(EEXIST));
1174         if (error != ENOENT)
1175                 return (error);
1176
1177         /*
1178          * We don't allow taking snapshots of inconsistent datasets, such as
1179          * those into which we are currently receiving.  However, if we are
1180          * creating this snapshot as part of a receive, this check will be
1181          * executed atomically with respect to the completion of the receive
1182          * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1183          * case we ignore this, knowing it will be fixed up for us shortly in
1184          * dmu_recv_end_sync().
1185          */
1186         if (!recv && DS_IS_INCONSISTENT(ds))
1187                 return (SET_ERROR(EBUSY));
1188
1189         /*
1190          * Skip the check for temporary snapshots or if we have already checked
1191          * the counts in dsl_dataset_snapshot_check. This means we really only
1192          * check the count here when we're receiving a stream.
1193          */
1194         if (cnt != 0 && cr != NULL) {
1195                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1196                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1197                 if (error != 0)
1198                         return (error);
1199         }
1200
1201         error = dsl_dataset_snapshot_reserve_space(ds, tx);
1202         if (error != 0)
1203                 return (error);
1204
1205         return (0);
1206 }
1207
1208 static int
1209 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1210 {
1211         dsl_dataset_snapshot_arg_t *ddsa = arg;
1212         dsl_pool_t *dp = dmu_tx_pool(tx);
1213         nvpair_t *pair;
1214         int rv = 0;
1215
1216         /*
1217          * Pre-compute how many total new snapshots will be created for each
1218          * level in the tree and below. This is needed for validating the
1219          * snapshot limit when either taking a recursive snapshot or when
1220          * taking multiple snapshots.
1221          *
1222          * The problem is that the counts are not actually adjusted when
1223          * we are checking, only when we finally sync. For a single snapshot,
1224          * this is easy, the count will increase by 1 at each node up the tree,
1225          * but its more complicated for the recursive/multiple snapshot case.
1226          *
1227          * The dsl_fs_ss_limit_check function does recursively check the count
1228          * at each level up the tree but since it is validating each snapshot
1229          * independently we need to be sure that we are validating the complete
1230          * count for the entire set of snapshots. We do this by rolling up the
1231          * counts for each component of the name into an nvlist and then
1232          * checking each of those cases with the aggregated count.
1233          *
1234          * This approach properly handles not only the recursive snapshot
1235          * case (where we get all of those on the ddsa_snaps list) but also
1236          * the sibling case (e.g. snapshot a/b and a/c so that we will also
1237          * validate the limit on 'a' using a count of 2).
1238          *
1239          * We validate the snapshot names in the third loop and only report
1240          * name errors once.
1241          */
1242         if (dmu_tx_is_syncing(tx)) {
1243                 nvlist_t *cnt_track = NULL;
1244                 cnt_track = fnvlist_alloc();
1245
1246                 /* Rollup aggregated counts into the cnt_track list */
1247                 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1248                     pair != NULL;
1249                     pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1250                         char *pdelim;
1251                         uint64_t val;
1252                         char nm[MAXPATHLEN];
1253
1254                         (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1255                         pdelim = strchr(nm, '@');
1256                         if (pdelim == NULL)
1257                                 continue;
1258                         *pdelim = '\0';
1259
1260                         do {
1261                                 if (nvlist_lookup_uint64(cnt_track, nm,
1262                                     &val) == 0) {
1263                                         /* update existing entry */
1264                                         fnvlist_add_uint64(cnt_track, nm,
1265                                             val + 1);
1266                                 } else {
1267                                         /* add to list */
1268                                         fnvlist_add_uint64(cnt_track, nm, 1);
1269                                 }
1270
1271                                 pdelim = strrchr(nm, '/');
1272                                 if (pdelim != NULL)
1273                                         *pdelim = '\0';
1274                         } while (pdelim != NULL);
1275                 }
1276
1277                 /* Check aggregated counts at each level */
1278                 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1279                     pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1280                         int error = 0;
1281                         char *name;
1282                         uint64_t cnt = 0;
1283                         dsl_dataset_t *ds;
1284
1285                         name = nvpair_name(pair);
1286                         cnt = fnvpair_value_uint64(pair);
1287                         ASSERT(cnt > 0);
1288
1289                         error = dsl_dataset_hold(dp, name, FTAG, &ds);
1290                         if (error == 0) {
1291                                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1292                                     ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1293                                     ddsa->ddsa_cr);
1294                                 dsl_dataset_rele(ds, FTAG);
1295                         }
1296
1297                         if (error != 0) {
1298                                 if (ddsa->ddsa_errors != NULL)
1299                                         fnvlist_add_int32(ddsa->ddsa_errors,
1300                                             name, error);
1301                                 rv = error;
1302                                 /* only report one error for this check */
1303                                 break;
1304                         }
1305                 }
1306                 nvlist_free(cnt_track);
1307         }
1308
1309         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1310             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1311                 int error = 0;
1312                 dsl_dataset_t *ds;
1313                 char *name, *atp;
1314                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1315
1316                 name = nvpair_name(pair);
1317                 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1318                         error = SET_ERROR(ENAMETOOLONG);
1319                 if (error == 0) {
1320                         atp = strchr(name, '@');
1321                         if (atp == NULL)
1322                                 error = SET_ERROR(EINVAL);
1323                         if (error == 0)
1324                                 (void) strlcpy(dsname, name, atp - name + 1);
1325                 }
1326                 if (error == 0)
1327                         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1328                 if (error == 0) {
1329                         /* passing 0/NULL skips dsl_fs_ss_limit_check */
1330                         error = dsl_dataset_snapshot_check_impl(ds,
1331                             atp + 1, tx, B_FALSE, 0, NULL);
1332                         dsl_dataset_rele(ds, FTAG);
1333                 }
1334
1335                 if (error != 0) {
1336                         if (ddsa->ddsa_errors != NULL) {
1337                                 fnvlist_add_int32(ddsa->ddsa_errors,
1338                                     name, error);
1339                         }
1340                         rv = error;
1341                 }
1342         }
1343
1344         return (rv);
1345 }
1346
1347 void
1348 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1349     dmu_tx_t *tx)
1350 {
1351         static zil_header_t zero_zil;
1352
1353         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1354         dmu_buf_t *dbuf;
1355         dsl_dataset_phys_t *dsphys;
1356         uint64_t dsobj, crtxg;
1357         objset_t *mos = dp->dp_meta_objset;
1358         objset_t *os;
1359
1360         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1361
1362         /*
1363          * If we are on an old pool, the zil must not be active, in which
1364          * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1365          */
1366         ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1367             dmu_objset_from_ds(ds, &os) != 0 ||
1368             bcmp(&os->os_phys->os_zil_header, &zero_zil,
1369             sizeof (zero_zil)) == 0);
1370
1371         dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1372
1373         /*
1374          * The origin's ds_creation_txg has to be < TXG_INITIAL
1375          */
1376         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1377                 crtxg = 1;
1378         else
1379                 crtxg = tx->tx_txg;
1380
1381         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1382             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1383         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1384         dmu_buf_will_dirty(dbuf, tx);
1385         dsphys = dbuf->db_data;
1386         bzero(dsphys, sizeof (dsl_dataset_phys_t));
1387         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1388         dsphys->ds_fsid_guid = unique_create();
1389         do {
1390                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1391                     sizeof (dsphys->ds_guid));
1392         } while (dsphys->ds_guid == 0);
1393         dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1394         dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1395         dsphys->ds_next_snap_obj = ds->ds_object;
1396         dsphys->ds_num_children = 1;
1397         dsphys->ds_creation_time = gethrestime_sec();
1398         dsphys->ds_creation_txg = crtxg;
1399         dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1400         dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1401         dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1402         dsphys->ds_uncompressed_bytes =
1403             dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1404         dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1405         dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1406         dmu_buf_rele(dbuf, FTAG);
1407
1408         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1409                 if (ds->ds_feature_inuse[f])
1410                         dsl_dataset_activate_feature(dsobj, f, tx);
1411         }
1412
1413         ASSERT3U(ds->ds_prev != 0, ==,
1414             dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1415         if (ds->ds_prev) {
1416                 uint64_t next_clones_obj =
1417                     dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1418                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1419                     ds->ds_object ||
1420                     dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1421                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1422                     ds->ds_object) {
1423                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1424                         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1425                             dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1426                         dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1427                 } else if (next_clones_obj != 0) {
1428                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
1429                             dsphys->ds_next_snap_obj, tx);
1430                         VERIFY0(zap_add_int(mos,
1431                             next_clones_obj, dsobj, tx));
1432                 }
1433         }
1434
1435         /*
1436          * If we have a reference-reservation on this dataset, we will
1437          * need to increase the amount of refreservation being charged
1438          * since our unique space is going to zero.
1439          */
1440         if (ds->ds_reserved) {
1441                 int64_t delta;
1442                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1443                 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1444                     ds->ds_reserved);
1445                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1446                     delta, 0, 0, tx);
1447         }
1448
1449         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1450         dsl_dataset_phys(ds)->ds_deadlist_obj =
1451             dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1452             dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1453         dsl_deadlist_close(&ds->ds_deadlist);
1454         dsl_deadlist_open(&ds->ds_deadlist, mos,
1455             dsl_dataset_phys(ds)->ds_deadlist_obj);
1456         dsl_deadlist_add_key(&ds->ds_deadlist,
1457             dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1458
1459         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1460         dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1461         dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1462         dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1463         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1464                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1465
1466         VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1467             snapname, 8, 1, &dsobj, tx));
1468
1469         if (ds->ds_prev)
1470                 dsl_dataset_rele(ds->ds_prev, ds);
1471         VERIFY0(dsl_dataset_hold_obj(dp,
1472             dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1473
1474         dsl_scan_ds_snapshotted(ds, tx);
1475
1476         dsl_dir_snap_cmtime_update(ds->ds_dir);
1477
1478         spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1479 }
1480
1481 static void
1482 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1483 {
1484         dsl_dataset_snapshot_arg_t *ddsa = arg;
1485         dsl_pool_t *dp = dmu_tx_pool(tx);
1486         nvpair_t *pair;
1487
1488         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1489             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1490                 dsl_dataset_t *ds;
1491                 char *name, *atp;
1492                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1493
1494                 name = nvpair_name(pair);
1495                 atp = strchr(name, '@');
1496                 (void) strlcpy(dsname, name, atp - name + 1);
1497                 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1498
1499                 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1500                 if (ddsa->ddsa_props != NULL) {
1501                         dsl_props_set_sync_impl(ds->ds_prev,
1502                             ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1503                 }
1504                 dsl_dataset_rele(ds, FTAG);
1505         }
1506 }
1507
1508 /*
1509  * The snapshots must all be in the same pool.
1510  * All-or-nothing: if there are any failures, nothing will be modified.
1511  */
1512 int
1513 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1514 {
1515         dsl_dataset_snapshot_arg_t ddsa;
1516         nvpair_t *pair;
1517         boolean_t needsuspend;
1518         int error;
1519         spa_t *spa;
1520         char *firstname;
1521         nvlist_t *suspended = NULL;
1522
1523         pair = nvlist_next_nvpair(snaps, NULL);
1524         if (pair == NULL)
1525                 return (0);
1526         firstname = nvpair_name(pair);
1527
1528         error = spa_open(firstname, &spa, FTAG);
1529         if (error != 0)
1530                 return (error);
1531         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1532         spa_close(spa, FTAG);
1533
1534         if (needsuspend) {
1535                 suspended = fnvlist_alloc();
1536                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1537                     pair = nvlist_next_nvpair(snaps, pair)) {
1538                         char fsname[ZFS_MAX_DATASET_NAME_LEN];
1539                         char *snapname = nvpair_name(pair);
1540                         char *atp;
1541                         void *cookie;
1542
1543                         atp = strchr(snapname, '@');
1544                         if (atp == NULL) {
1545                                 error = SET_ERROR(EINVAL);
1546                                 break;
1547                         }
1548                         (void) strlcpy(fsname, snapname, atp - snapname + 1);
1549
1550                         error = zil_suspend(fsname, &cookie);
1551                         if (error != 0)
1552                                 break;
1553                         fnvlist_add_uint64(suspended, fsname,
1554                             (uintptr_t)cookie);
1555                 }
1556         }
1557
1558         ddsa.ddsa_snaps = snaps;
1559         ddsa.ddsa_props = props;
1560         ddsa.ddsa_errors = errors;
1561         ddsa.ddsa_cr = CRED();
1562
1563         if (error == 0) {
1564                 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1565                     dsl_dataset_snapshot_sync, &ddsa,
1566                     fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1567         }
1568
1569         if (suspended != NULL) {
1570                 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1571                     pair = nvlist_next_nvpair(suspended, pair)) {
1572                         zil_resume((void *)(uintptr_t)
1573                             fnvpair_value_uint64(pair));
1574                 }
1575                 fnvlist_free(suspended);
1576         }
1577
1578 #ifdef __FreeBSD__
1579 #ifdef _KERNEL
1580         if (error == 0) {
1581                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1582                     pair = nvlist_next_nvpair(snaps, pair)) {
1583                         char *snapname = nvpair_name(pair);
1584                         zvol_create_minors(snapname);
1585                 }
1586         }
1587 #endif
1588 #endif
1589         return (error);
1590 }
1591
1592 typedef struct dsl_dataset_snapshot_tmp_arg {
1593         const char *ddsta_fsname;
1594         const char *ddsta_snapname;
1595         minor_t ddsta_cleanup_minor;
1596         const char *ddsta_htag;
1597 } dsl_dataset_snapshot_tmp_arg_t;
1598
1599 static int
1600 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1601 {
1602         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1603         dsl_pool_t *dp = dmu_tx_pool(tx);
1604         dsl_dataset_t *ds;
1605         int error;
1606
1607         error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1608         if (error != 0)
1609                 return (error);
1610
1611         /* NULL cred means no limit check for tmp snapshot */
1612         error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1613             tx, B_FALSE, 0, NULL);
1614         if (error != 0) {
1615                 dsl_dataset_rele(ds, FTAG);
1616                 return (error);
1617         }
1618
1619         if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1620                 dsl_dataset_rele(ds, FTAG);
1621                 return (SET_ERROR(ENOTSUP));
1622         }
1623         error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1624             B_TRUE, tx);
1625         if (error != 0) {
1626                 dsl_dataset_rele(ds, FTAG);
1627                 return (error);
1628         }
1629
1630         dsl_dataset_rele(ds, FTAG);
1631         return (0);
1632 }
1633
1634 static void
1635 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1636 {
1637         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1638         dsl_pool_t *dp = dmu_tx_pool(tx);
1639         dsl_dataset_t *ds;
1640
1641         VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1642
1643         dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1644         dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1645             ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1646         dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1647
1648         dsl_dataset_rele(ds, FTAG);
1649 }
1650
1651 int
1652 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1653     minor_t cleanup_minor, const char *htag)
1654 {
1655         dsl_dataset_snapshot_tmp_arg_t ddsta;
1656         int error;
1657         spa_t *spa;
1658         boolean_t needsuspend;
1659         void *cookie;
1660
1661         ddsta.ddsta_fsname = fsname;
1662         ddsta.ddsta_snapname = snapname;
1663         ddsta.ddsta_cleanup_minor = cleanup_minor;
1664         ddsta.ddsta_htag = htag;
1665
1666         error = spa_open(fsname, &spa, FTAG);
1667         if (error != 0)
1668                 return (error);
1669         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1670         spa_close(spa, FTAG);
1671
1672         if (needsuspend) {
1673                 error = zil_suspend(fsname, &cookie);
1674                 if (error != 0)
1675                         return (error);
1676         }
1677
1678         error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1679             dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1680
1681         if (needsuspend)
1682                 zil_resume(cookie);
1683         return (error);
1684 }
1685
1686
1687 void
1688 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1689 {
1690         ASSERT(dmu_tx_is_syncing(tx));
1691         ASSERT(ds->ds_objset != NULL);
1692         ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1693
1694         /*
1695          * in case we had to change ds_fsid_guid when we opened it,
1696          * sync it out now.
1697          */
1698         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1699         dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1700
1701         if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1702                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1703                     ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1704                     &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1705                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1706                     ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1707                     &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1708                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1709                     ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1710                     &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1711                 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1712                 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1713                 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1714         }
1715
1716         dmu_objset_sync(ds->ds_objset, zio, tx);
1717
1718         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1719                 if (ds->ds_feature_activation_needed[f]) {
1720                         if (ds->ds_feature_inuse[f])
1721                                 continue;
1722                         dsl_dataset_activate_feature(ds->ds_object, f, tx);
1723                         ds->ds_feature_inuse[f] = B_TRUE;
1724                 }
1725         }
1726 }
1727
1728 static void
1729 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1730 {
1731         uint64_t count = 0;
1732         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1733         zap_cursor_t zc;
1734         zap_attribute_t za;
1735         nvlist_t *propval = fnvlist_alloc();
1736         nvlist_t *val = fnvlist_alloc();
1737
1738         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1739
1740         /*
1741          * There may be missing entries in ds_next_clones_obj
1742          * due to a bug in a previous version of the code.
1743          * Only trust it if it has the right number of entries.
1744          */
1745         if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1746                 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1747                     &count));
1748         }
1749         if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1750                 goto fail;
1751         for (zap_cursor_init(&zc, mos,
1752             dsl_dataset_phys(ds)->ds_next_clones_obj);
1753             zap_cursor_retrieve(&zc, &za) == 0;
1754             zap_cursor_advance(&zc)) {
1755                 dsl_dataset_t *clone;
1756                 char buf[ZFS_MAX_DATASET_NAME_LEN];
1757                 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1758                     za.za_first_integer, FTAG, &clone));
1759                 dsl_dir_name(clone->ds_dir, buf);
1760                 fnvlist_add_boolean(val, buf);
1761                 dsl_dataset_rele(clone, FTAG);
1762         }
1763         zap_cursor_fini(&zc);
1764         fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1765         fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1766 fail:
1767         nvlist_free(val);
1768         nvlist_free(propval);
1769 }
1770
1771 static void
1772 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1773 {
1774         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1775
1776         if (dsl_dataset_has_resume_receive_state(ds)) {
1777                 char *str;
1778                 void *packed;
1779                 uint8_t *compressed;
1780                 uint64_t val;
1781                 nvlist_t *token_nv = fnvlist_alloc();
1782                 size_t packed_size, compressed_size;
1783
1784                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1785                     DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1786                         fnvlist_add_uint64(token_nv, "fromguid", val);
1787                 }
1788                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1789                     DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1790                         fnvlist_add_uint64(token_nv, "object", val);
1791                 }
1792                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1793                     DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1794                         fnvlist_add_uint64(token_nv, "offset", val);
1795                 }
1796                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1797                     DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1798                         fnvlist_add_uint64(token_nv, "bytes", val);
1799                 }
1800                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1801                     DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1802                         fnvlist_add_uint64(token_nv, "toguid", val);
1803                 }
1804                 char buf[256];
1805                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1806                     DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1807                         fnvlist_add_string(token_nv, "toname", buf);
1808                 }
1809                 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1810                     DS_FIELD_RESUME_EMBEDOK) == 0) {
1811                         fnvlist_add_boolean(token_nv, "embedok");
1812                 }
1813                 packed = fnvlist_pack(token_nv, &packed_size);
1814                 fnvlist_free(token_nv);
1815                 compressed = kmem_alloc(packed_size, KM_SLEEP);
1816
1817                 compressed_size = gzip_compress(packed, compressed,
1818                     packed_size, packed_size, 6);
1819
1820                 zio_cksum_t cksum;
1821                 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1822
1823                 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1824                 for (int i = 0; i < compressed_size; i++) {
1825                         (void) sprintf(str + i * 2, "%02x", compressed[i]);
1826                 }
1827                 str[compressed_size * 2] = '\0';
1828                 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1829                     ZFS_SEND_RESUME_TOKEN_VERSION,
1830                     (longlong_t)cksum.zc_word[0],
1831                     (longlong_t)packed_size, str);
1832                 dsl_prop_nvlist_add_string(nv,
1833                     ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1834                 kmem_free(packed, packed_size);
1835                 kmem_free(str, compressed_size * 2 + 1);
1836                 kmem_free(compressed, packed_size);
1837                 strfree(propval);
1838         }
1839 }
1840
1841 void
1842 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1843 {
1844         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1845         uint64_t refd, avail, uobjs, aobjs, ratio;
1846
1847         ASSERT(dsl_pool_config_held(dp));
1848
1849         ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1850             (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1851             dsl_dataset_phys(ds)->ds_compressed_bytes);
1852
1853         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1854         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1855             dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1856
1857         if (ds->ds_is_snapshot) {
1858                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1859                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1860                     dsl_dataset_phys(ds)->ds_unique_bytes);
1861                 get_clones_stat(ds, nv);
1862         } else {
1863                 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1864                         char buf[ZFS_MAX_DATASET_NAME_LEN];
1865                         dsl_dataset_name(ds->ds_prev, buf);
1866                         dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1867                 }
1868
1869                 dsl_dir_stats(ds->ds_dir, nv);
1870         }
1871
1872         dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1873         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1874         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1875
1876         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1877             dsl_dataset_phys(ds)->ds_creation_time);
1878         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1879             dsl_dataset_phys(ds)->ds_creation_txg);
1880         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1881             ds->ds_quota);
1882         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1883             ds->ds_reserved);
1884         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1885             dsl_dataset_phys(ds)->ds_guid);
1886         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1887             dsl_dataset_phys(ds)->ds_unique_bytes);
1888         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1889             ds->ds_object);
1890         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1891             ds->ds_userrefs);
1892         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1893             DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1894
1895         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1896                 uint64_t written, comp, uncomp;
1897                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1898                 dsl_dataset_t *prev;
1899
1900                 int err = dsl_dataset_hold_obj(dp,
1901                     dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1902                 if (err == 0) {
1903                         err = dsl_dataset_space_written(prev, ds, &written,
1904                             &comp, &uncomp);
1905                         dsl_dataset_rele(prev, FTAG);
1906                         if (err == 0) {
1907                                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1908                                     written);
1909                         }
1910                 }
1911         }
1912
1913         if (!dsl_dataset_is_snapshot(ds)) {
1914                 /*
1915                  * A failed "newfs" (e.g. full) resumable receive leaves
1916                  * the stats set on this dataset.  Check here for the prop.
1917                  */
1918                 get_receive_resume_stats(ds, nv);
1919
1920                 /*
1921                  * A failed incremental resumable receive leaves the
1922                  * stats set on our child named "%recv".  Check the child
1923                  * for the prop.
1924                  */
1925                 /* 6 extra bytes for /%recv */
1926                 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1927                 dsl_dataset_t *recv_ds;
1928                 dsl_dataset_name(ds, recvname);
1929                 if (strlcat(recvname, "/", sizeof (recvname)) <
1930                     sizeof (recvname) &&
1931                     strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1932                     sizeof (recvname) &&
1933                     dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1934                         get_receive_resume_stats(recv_ds, nv);
1935                         dsl_dataset_rele(recv_ds, FTAG);
1936                 }
1937         }
1938 }
1939
1940 void
1941 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1942 {
1943         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1944         ASSERT(dsl_pool_config_held(dp));
1945
1946         stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1947         stat->dds_inconsistent =
1948             dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1949         stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1950         stat->dds_origin[0] = '\0';
1951         if (ds->ds_is_snapshot) {
1952                 stat->dds_is_snapshot = B_TRUE;
1953                 stat->dds_num_clones =
1954                     dsl_dataset_phys(ds)->ds_num_children - 1;
1955         } else {
1956                 stat->dds_is_snapshot = B_FALSE;
1957                 stat->dds_num_clones = 0;
1958
1959                 if (dsl_dir_is_clone(ds->ds_dir)) {
1960                         dsl_dataset_t *ods;
1961
1962                         VERIFY0(dsl_dataset_hold_obj(dp,
1963                             dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1964                             FTAG, &ods));
1965                         dsl_dataset_name(ods, stat->dds_origin);
1966                         dsl_dataset_rele(ods, FTAG);
1967                 }
1968         }
1969 }
1970
1971 uint64_t
1972 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1973 {
1974         return (ds->ds_fsid_guid);
1975 }
1976
1977 void
1978 dsl_dataset_space(dsl_dataset_t *ds,
1979     uint64_t *refdbytesp, uint64_t *availbytesp,
1980     uint64_t *usedobjsp, uint64_t *availobjsp)
1981 {
1982         *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1983         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1984         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1985                 *availbytesp +=
1986                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1987         if (ds->ds_quota != 0) {
1988                 /*
1989                  * Adjust available bytes according to refquota
1990                  */
1991                 if (*refdbytesp < ds->ds_quota)
1992                         *availbytesp = MIN(*availbytesp,
1993                             ds->ds_quota - *refdbytesp);
1994                 else
1995                         *availbytesp = 0;
1996         }
1997         *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1998         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1999 }
2000
2001 boolean_t
2002 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2003 {
2004         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2005
2006         ASSERT(dsl_pool_config_held(dp));
2007         if (snap == NULL)
2008                 return (B_FALSE);
2009         if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
2010             dsl_dataset_phys(snap)->ds_creation_txg) {
2011                 objset_t *os, *os_snap;
2012                 /*
2013                  * It may be that only the ZIL differs, because it was
2014                  * reset in the head.  Don't count that as being
2015                  * modified.
2016                  */
2017                 if (dmu_objset_from_ds(ds, &os) != 0)
2018                         return (B_TRUE);
2019                 if (dmu_objset_from_ds(snap, &os_snap) != 0)
2020                         return (B_TRUE);
2021                 return (bcmp(&os->os_phys->os_meta_dnode,
2022                     &os_snap->os_phys->os_meta_dnode,
2023                     sizeof (os->os_phys->os_meta_dnode)) != 0);
2024         }
2025         return (B_FALSE);
2026 }
2027
2028 typedef struct dsl_dataset_rename_snapshot_arg {
2029         const char *ddrsa_fsname;
2030         const char *ddrsa_oldsnapname;
2031         const char *ddrsa_newsnapname;
2032         boolean_t ddrsa_recursive;
2033         dmu_tx_t *ddrsa_tx;
2034 } dsl_dataset_rename_snapshot_arg_t;
2035
2036 /* ARGSUSED */
2037 static int
2038 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2039     dsl_dataset_t *hds, void *arg)
2040 {
2041         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2042         int error;
2043         uint64_t val;
2044
2045         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2046         if (error != 0) {
2047                 /* ignore nonexistent snapshots */
2048                 return (error == ENOENT ? 0 : error);
2049         }
2050
2051         /* new name should not exist */
2052         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2053         if (error == 0)
2054                 error = SET_ERROR(EEXIST);
2055         else if (error == ENOENT)
2056                 error = 0;
2057
2058         /* dataset name + 1 for the "@" + the new snapshot name must fit */
2059         if (dsl_dir_namelen(hds->ds_dir) + 1 +
2060             strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2061                 error = SET_ERROR(ENAMETOOLONG);
2062
2063         return (error);
2064 }
2065
2066 static int
2067 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2068 {
2069         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2070         dsl_pool_t *dp = dmu_tx_pool(tx);
2071         dsl_dataset_t *hds;
2072         int error;
2073
2074         error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2075         if (error != 0)
2076                 return (error);
2077
2078         if (ddrsa->ddrsa_recursive) {
2079                 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2080                     dsl_dataset_rename_snapshot_check_impl, ddrsa,
2081                     DS_FIND_CHILDREN);
2082         } else {
2083                 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2084         }
2085         dsl_dataset_rele(hds, FTAG);
2086         return (error);
2087 }
2088
2089 static int
2090 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2091     dsl_dataset_t *hds, void *arg)
2092 {
2093 #ifdef __FreeBSD__
2094 #ifdef _KERNEL
2095         char *oldname, *newname;
2096 #endif
2097 #endif
2098         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2099         dsl_dataset_t *ds;
2100         uint64_t val;
2101         dmu_tx_t *tx = ddrsa->ddrsa_tx;
2102         int error;
2103
2104         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2105         ASSERT(error == 0 || error == ENOENT);
2106         if (error == ENOENT) {
2107                 /* ignore nonexistent snapshots */
2108                 return (0);
2109         }
2110
2111         VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2112
2113         /* log before we change the name */
2114         spa_history_log_internal_ds(ds, "rename", tx,
2115             "-> @%s", ddrsa->ddrsa_newsnapname);
2116
2117         VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2118             B_FALSE));
2119         mutex_enter(&ds->ds_lock);
2120         (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2121         mutex_exit(&ds->ds_lock);
2122         VERIFY0(zap_add(dp->dp_meta_objset,
2123             dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2124             ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2125
2126 #ifdef __FreeBSD__
2127 #ifdef _KERNEL
2128         oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2129         newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2130         snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2131             ddrsa->ddrsa_oldsnapname);
2132         snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2133             ddrsa->ddrsa_newsnapname);
2134         zfsvfs_update_fromname(oldname, newname);
2135         zvol_rename_minors(oldname, newname);
2136         kmem_free(newname, MAXPATHLEN);
2137         kmem_free(oldname, MAXPATHLEN);
2138 #endif
2139 #endif
2140         dsl_dataset_rele(ds, FTAG);
2141
2142         return (0);
2143 }
2144
2145 static void
2146 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2147 {
2148         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2149         dsl_pool_t *dp = dmu_tx_pool(tx);
2150         dsl_dataset_t *hds;
2151
2152         VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2153         ddrsa->ddrsa_tx = tx;
2154         if (ddrsa->ddrsa_recursive) {
2155                 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2156                     dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2157                     DS_FIND_CHILDREN));
2158         } else {
2159                 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2160         }
2161         dsl_dataset_rele(hds, FTAG);
2162 }
2163
2164 int
2165 dsl_dataset_rename_snapshot(const char *fsname,
2166     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2167 {
2168         dsl_dataset_rename_snapshot_arg_t ddrsa;
2169
2170         ddrsa.ddrsa_fsname = fsname;
2171         ddrsa.ddrsa_oldsnapname = oldsnapname;
2172         ddrsa.ddrsa_newsnapname = newsnapname;
2173         ddrsa.ddrsa_recursive = recursive;
2174
2175         return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2176             dsl_dataset_rename_snapshot_sync, &ddrsa,
2177             1, ZFS_SPACE_CHECK_RESERVED));
2178 }
2179
2180 /*
2181  * If we're doing an ownership handoff, we need to make sure that there is
2182  * only one long hold on the dataset.  We're not allowed to change anything here
2183  * so we don't permanently release the long hold or regular hold here.  We want
2184  * to do this only when syncing to avoid the dataset unexpectedly going away
2185  * when we release the long hold.
2186  */
2187 static int
2188 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2189 {
2190         boolean_t held;
2191
2192         if (!dmu_tx_is_syncing(tx))
2193                 return (0);
2194
2195         if (owner != NULL) {
2196                 VERIFY3P(ds->ds_owner, ==, owner);
2197                 dsl_dataset_long_rele(ds, owner);
2198         }
2199
2200         held = dsl_dataset_long_held(ds);
2201
2202         if (owner != NULL)
2203                 dsl_dataset_long_hold(ds, owner);
2204
2205         if (held)
2206                 return (SET_ERROR(EBUSY));
2207
2208         return (0);
2209 }
2210
2211 typedef struct dsl_dataset_rollback_arg {
2212         const char *ddra_fsname;
2213         void *ddra_owner;
2214         nvlist_t *ddra_result;
2215 } dsl_dataset_rollback_arg_t;
2216
2217 static int
2218 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2219 {
2220         dsl_dataset_rollback_arg_t *ddra = arg;
2221         dsl_pool_t *dp = dmu_tx_pool(tx);
2222         dsl_dataset_t *ds;
2223         int64_t unused_refres_delta;
2224         int error;
2225
2226         error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2227         if (error != 0)
2228                 return (error);
2229
2230         /* must not be a snapshot */
2231         if (ds->ds_is_snapshot) {
2232                 dsl_dataset_rele(ds, FTAG);
2233                 return (SET_ERROR(EINVAL));
2234         }
2235
2236         /* must have a most recent snapshot */
2237         if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2238                 dsl_dataset_rele(ds, FTAG);
2239                 return (SET_ERROR(EINVAL));
2240         }
2241
2242         /* must not have any bookmarks after the most recent snapshot */
2243         nvlist_t *proprequest = fnvlist_alloc();
2244         fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2245         nvlist_t *bookmarks = fnvlist_alloc();
2246         error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2247         fnvlist_free(proprequest);
2248         if (error != 0)
2249                 return (error);
2250         for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2251             pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2252                 nvlist_t *valuenv =
2253                     fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2254                     zfs_prop_to_name(ZFS_PROP_CREATETXG));
2255                 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2256                 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2257                         fnvlist_free(bookmarks);
2258                         dsl_dataset_rele(ds, FTAG);
2259                         return (SET_ERROR(EEXIST));
2260                 }
2261         }
2262         fnvlist_free(bookmarks);
2263
2264         error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2265         if (error != 0) {
2266                 dsl_dataset_rele(ds, FTAG);
2267                 return (error);
2268         }
2269
2270         /*
2271          * Check if the snap we are rolling back to uses more than
2272          * the refquota.
2273          */
2274         if (ds->ds_quota != 0 &&
2275             dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2276                 dsl_dataset_rele(ds, FTAG);
2277                 return (SET_ERROR(EDQUOT));
2278         }
2279
2280         /*
2281          * When we do the clone swap, we will temporarily use more space
2282          * due to the refreservation (the head will no longer have any
2283          * unique space, so the entire amount of the refreservation will need
2284          * to be free).  We will immediately destroy the clone, freeing
2285          * this space, but the freeing happens over many txg's.
2286          */
2287         unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2288             dsl_dataset_phys(ds)->ds_unique_bytes);
2289
2290         if (unused_refres_delta > 0 &&
2291             unused_refres_delta >
2292             dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2293                 dsl_dataset_rele(ds, FTAG);
2294                 return (SET_ERROR(ENOSPC));
2295         }
2296
2297         dsl_dataset_rele(ds, FTAG);
2298         return (0);
2299 }
2300
2301 static void
2302 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2303 {
2304         dsl_dataset_rollback_arg_t *ddra = arg;
2305         dsl_pool_t *dp = dmu_tx_pool(tx);
2306         dsl_dataset_t *ds, *clone;
2307         uint64_t cloneobj;
2308         char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2309
2310         VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2311
2312         dsl_dataset_name(ds->ds_prev, namebuf);
2313         fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2314
2315         cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2316             ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2317
2318         VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2319
2320         dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2321         dsl_dataset_zero_zil(ds, tx);
2322
2323         dsl_destroy_head_sync_impl(clone, tx);
2324
2325         dsl_dataset_rele(clone, FTAG);
2326         dsl_dataset_rele(ds, FTAG);
2327 }
2328
2329 /*
2330  * Rolls back the given filesystem or volume to the most recent snapshot.
2331  * The name of the most recent snapshot will be returned under key "target"
2332  * in the result nvlist.
2333  *
2334  * If owner != NULL:
2335  * - The existing dataset MUST be owned by the specified owner at entry
2336  * - Upon return, dataset will still be held by the same owner, whether we
2337  *   succeed or not.
2338  *
2339  * This mode is required any time the existing filesystem is mounted.  See
2340  * notes above zfs_suspend_fs() for further details.
2341  */
2342 int
2343 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2344 {
2345         dsl_dataset_rollback_arg_t ddra;
2346
2347         ddra.ddra_fsname = fsname;
2348         ddra.ddra_owner = owner;
2349         ddra.ddra_result = result;
2350
2351         return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2352             dsl_dataset_rollback_sync, &ddra,
2353             1, ZFS_SPACE_CHECK_RESERVED));
2354 }
2355
2356 struct promotenode {
2357         list_node_t link;
2358         dsl_dataset_t *ds;
2359 };
2360
2361 typedef struct dsl_dataset_promote_arg {
2362         const char *ddpa_clonename;
2363         dsl_dataset_t *ddpa_clone;
2364         list_t shared_snaps, origin_snaps, clone_snaps;
2365         dsl_dataset_t *origin_origin; /* origin of the origin */
2366         uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2367         char *err_ds;
2368         cred_t *cr;
2369 } dsl_dataset_promote_arg_t;
2370
2371 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2372 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2373     void *tag);
2374 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2375
2376 static int
2377 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2378 {
2379         dsl_dataset_promote_arg_t *ddpa = arg;
2380         dsl_pool_t *dp = dmu_tx_pool(tx);
2381         dsl_dataset_t *hds;
2382         struct promotenode *snap;
2383         dsl_dataset_t *origin_ds;
2384         int err;
2385         uint64_t unused;
2386         uint64_t ss_mv_cnt;
2387         size_t max_snap_len;
2388
2389         err = promote_hold(ddpa, dp, FTAG);
2390         if (err != 0)
2391                 return (err);
2392
2393         hds = ddpa->ddpa_clone;
2394         max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2395
2396         if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2397                 promote_rele(ddpa, FTAG);
2398                 return (SET_ERROR(EXDEV));
2399         }
2400
2401         /*
2402          * Compute and check the amount of space to transfer.  Since this is
2403          * so expensive, don't do the preliminary check.
2404          */
2405         if (!dmu_tx_is_syncing(tx)) {
2406                 promote_rele(ddpa, FTAG);
2407                 return (0);
2408         }
2409
2410         snap = list_head(&ddpa->shared_snaps);
2411         origin_ds = snap->ds;
2412
2413         /* compute origin's new unique space */
2414         snap = list_tail(&ddpa->clone_snaps);
2415         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2416             origin_ds->ds_object);
2417         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2418             dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2419             &ddpa->unique, &unused, &unused);
2420
2421         /*
2422          * Walk the snapshots that we are moving
2423          *
2424          * Compute space to transfer.  Consider the incremental changes
2425          * to used by each snapshot:
2426          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2427          * So each snapshot gave birth to:
2428          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2429          * So a sequence would look like:
2430          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2431          * Which simplifies to:
2432          * uN + kN + kN-1 + ... + k1 + k0
2433          * Note however, if we stop before we reach the ORIGIN we get:
2434          * uN + kN + kN-1 + ... + kM - uM-1
2435          */
2436         ss_mv_cnt = 0;
2437         ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2438         ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2439         ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2440         for (snap = list_head(&ddpa->shared_snaps); snap;
2441             snap = list_next(&ddpa->shared_snaps, snap)) {
2442                 uint64_t val, dlused, dlcomp, dluncomp;
2443                 dsl_dataset_t *ds = snap->ds;
2444
2445                 ss_mv_cnt++;
2446
2447                 /*
2448                  * If there are long holds, we won't be able to evict
2449                  * the objset.
2450                  */
2451                 if (dsl_dataset_long_held(ds)) {
2452                         err = SET_ERROR(EBUSY);
2453                         goto out;
2454                 }
2455
2456                 /* Check that the snapshot name does not conflict */
2457                 VERIFY0(dsl_dataset_get_snapname(ds));
2458                 if (strlen(ds->ds_snapname) >= max_snap_len) {
2459                         err = SET_ERROR(ENAMETOOLONG);
2460                         goto out;
2461                 }
2462                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2463                 if (err == 0) {
2464                         (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2465                         err = SET_ERROR(EEXIST);
2466                         goto out;
2467                 }
2468                 if (err != ENOENT)
2469                         goto out;
2470
2471                 /* The very first snapshot does not have a deadlist */
2472                 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2473                         continue;
2474
2475                 dsl_deadlist_space(&ds->ds_deadlist,
2476                     &dlused, &dlcomp, &dluncomp);
2477                 ddpa->used += dlused;
2478                 ddpa->comp += dlcomp;
2479                 ddpa->uncomp += dluncomp;
2480         }
2481
2482         /*
2483          * If we are a clone of a clone then we never reached ORIGIN,
2484          * so we need to subtract out the clone origin's used space.
2485          */
2486         if (ddpa->origin_origin) {
2487                 ddpa->used -=
2488                     dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2489                 ddpa->comp -=
2490                     dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2491                 ddpa->uncomp -=
2492                     dsl_dataset_phys(ddpa->origin_origin)->
2493                     ds_uncompressed_bytes;
2494         }
2495
2496         /* Check that there is enough space and limit headroom here */
2497         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2498             0, ss_mv_cnt, ddpa->used, ddpa->cr);
2499         if (err != 0)
2500                 goto out;
2501
2502         /*
2503          * Compute the amounts of space that will be used by snapshots
2504          * after the promotion (for both origin and clone).  For each,
2505          * it is the amount of space that will be on all of their
2506          * deadlists (that was not born before their new origin).
2507          */
2508         if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2509                 uint64_t space;
2510
2511                 /*
2512                  * Note, typically this will not be a clone of a clone,
2513                  * so dd_origin_txg will be < TXG_INITIAL, so
2514                  * these snaplist_space() -> dsl_deadlist_space_range()
2515                  * calls will be fast because they do not have to
2516                  * iterate over all bps.
2517                  */
2518                 snap = list_head(&ddpa->origin_snaps);
2519                 err = snaplist_space(&ddpa->shared_snaps,
2520                     snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2521                 if (err != 0)
2522                         goto out;
2523
2524                 err = snaplist_space(&ddpa->clone_snaps,
2525                     snap->ds->ds_dir->dd_origin_txg, &space);
2526                 if (err != 0)
2527                         goto out;
2528                 ddpa->cloneusedsnap += space;
2529         }
2530         if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2531             DD_FLAG_USED_BREAKDOWN) {
2532                 err = snaplist_space(&ddpa->origin_snaps,
2533                     dsl_dataset_phys(origin_ds)->ds_creation_txg,
2534                     &ddpa->originusedsnap);
2535                 if (err != 0)
2536                         goto out;
2537         }
2538
2539 out:
2540         promote_rele(ddpa, FTAG);
2541         return (err);
2542 }
2543
2544 static void
2545 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2546 {
2547         dsl_dataset_promote_arg_t *ddpa = arg;
2548         dsl_pool_t *dp = dmu_tx_pool(tx);
2549         dsl_dataset_t *hds;
2550         struct promotenode *snap;
2551         dsl_dataset_t *origin_ds;
2552         dsl_dataset_t *origin_head;
2553         dsl_dir_t *dd;
2554         dsl_dir_t *odd = NULL;
2555         uint64_t oldnext_obj;
2556         int64_t delta;
2557 #if defined(__FreeBSD__) && defined(_KERNEL)
2558         char *oldname, *newname;
2559 #endif
2560
2561         VERIFY0(promote_hold(ddpa, dp, FTAG));
2562         hds = ddpa->ddpa_clone;
2563
2564         ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2565
2566         snap = list_head(&ddpa->shared_snaps);
2567         origin_ds = snap->ds;
2568         dd = hds->ds_dir;
2569
2570         snap = list_head(&ddpa->origin_snaps);
2571         origin_head = snap->ds;
2572
2573         /*
2574          * We need to explicitly open odd, since origin_ds's dd will be
2575          * changing.
2576          */
2577         VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2578             NULL, FTAG, &odd));
2579
2580         /* change origin's next snap */
2581         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2582         oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2583         snap = list_tail(&ddpa->clone_snaps);
2584         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2585             origin_ds->ds_object);
2586         dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2587
2588         /* change the origin's next clone */
2589         if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2590                 dsl_dataset_remove_from_next_clones(origin_ds,
2591                     snap->ds->ds_object, tx);
2592                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2593                     dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2594                     oldnext_obj, tx));
2595         }
2596
2597         /* change origin */
2598         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2599         ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2600         dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2601         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2602         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2603         dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2604         origin_head->ds_dir->dd_origin_txg =
2605             dsl_dataset_phys(origin_ds)->ds_creation_txg;
2606
2607         /* change dd_clone entries */
2608         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2609                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2610                     dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2611                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2612                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2613                     hds->ds_object, tx));
2614
2615                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2616                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2617                     origin_head->ds_object, tx));
2618                 if (dsl_dir_phys(dd)->dd_clones == 0) {
2619                         dsl_dir_phys(dd)->dd_clones =
2620                             zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2621                             DMU_OT_NONE, 0, tx);
2622                 }
2623                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2624                     dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2625         }
2626
2627 #if defined(__FreeBSD__) && defined(_KERNEL)
2628         /* Take the spa_namespace_lock early so zvol renames don't deadlock. */
2629         mutex_enter(&spa_namespace_lock);
2630
2631         oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2632         newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2633 #endif
2634
2635         /* move snapshots to this dir */
2636         for (snap = list_head(&ddpa->shared_snaps); snap;
2637             snap = list_next(&ddpa->shared_snaps, snap)) {
2638                 dsl_dataset_t *ds = snap->ds;
2639
2640                 /*
2641                  * Property callbacks are registered to a particular
2642                  * dsl_dir.  Since ours is changing, evict the objset
2643                  * so that they will be unregistered from the old dsl_dir.
2644                  */
2645                 if (ds->ds_objset) {
2646                         dmu_objset_evict(ds->ds_objset);
2647                         ds->ds_objset = NULL;
2648                 }
2649
2650                 /* move snap name entry */
2651                 VERIFY0(dsl_dataset_get_snapname(ds));
2652                 VERIFY0(dsl_dataset_snap_remove(origin_head,
2653                     ds->ds_snapname, tx, B_TRUE));
2654                 VERIFY0(zap_add(dp->dp_meta_objset,
2655                     dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2656                     8, 1, &ds->ds_object, tx));
2657                 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2658                     DD_FIELD_SNAPSHOT_COUNT, tx);
2659
2660                 /* change containing dsl_dir */
2661                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2662                 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2663                 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2664                 ASSERT3P(ds->ds_dir, ==, odd);
2665                 dsl_dir_rele(ds->ds_dir, ds);
2666                 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2667                     NULL, ds, &ds->ds_dir));
2668
2669 #if defined(__FreeBSD__) && defined(_KERNEL)
2670                 dsl_dataset_name(ds, newname);
2671                 zfsvfs_update_fromname(oldname, newname);
2672                 zvol_rename_minors(oldname, newname);
2673 #endif
2674
2675                 /* move any clone references */
2676                 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2677                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2678                         zap_cursor_t zc;
2679                         zap_attribute_t za;
2680
2681                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
2682                             dsl_dataset_phys(ds)->ds_next_clones_obj);
2683                             zap_cursor_retrieve(&zc, &za) == 0;
2684                             zap_cursor_advance(&zc)) {
2685                                 dsl_dataset_t *cnds;
2686                                 uint64_t o;
2687
2688                                 if (za.za_first_integer == oldnext_obj) {
2689                                         /*
2690                                          * We've already moved the
2691                                          * origin's reference.
2692                                          */
2693                                         continue;
2694                                 }
2695
2696                                 VERIFY0(dsl_dataset_hold_obj(dp,
2697                                     za.za_first_integer, FTAG, &cnds));
2698                                 o = dsl_dir_phys(cnds->ds_dir)->
2699                                     dd_head_dataset_obj;
2700
2701                                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2702                                     dsl_dir_phys(odd)->dd_clones, o, tx));
2703                                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2704                                     dsl_dir_phys(dd)->dd_clones, o, tx));
2705                                 dsl_dataset_rele(cnds, FTAG);
2706                         }
2707                         zap_cursor_fini(&zc);
2708                 }
2709
2710                 ASSERT(!dsl_prop_hascb(ds));
2711         }
2712
2713 #if defined(__FreeBSD__) && defined(_KERNEL)
2714         mutex_exit(&spa_namespace_lock);
2715
2716         kmem_free(newname, MAXPATHLEN);
2717         kmem_free(oldname, MAXPATHLEN);
2718 #endif
2719         /*
2720          * Change space accounting.
2721          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2722          * both be valid, or both be 0 (resulting in delta == 0).  This
2723          * is true for each of {clone,origin} independently.
2724          */
2725
2726         delta = ddpa->cloneusedsnap -
2727             dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2728         ASSERT3S(delta, >=, 0);
2729         ASSERT3U(ddpa->used, >=, delta);
2730         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2731         dsl_dir_diduse_space(dd, DD_USED_HEAD,
2732             ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2733
2734         delta = ddpa->originusedsnap -
2735             dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2736         ASSERT3S(delta, <=, 0);
2737         ASSERT3U(ddpa->used, >=, -delta);
2738         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2739         dsl_dir_diduse_space(odd, DD_USED_HEAD,
2740             -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2741
2742         dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2743
2744         /* log history record */
2745         spa_history_log_internal_ds(hds, "promote", tx, "");
2746
2747         dsl_dir_rele(odd, FTAG);
2748         promote_rele(ddpa, FTAG);
2749 }
2750
2751 /*
2752  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2753  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2754  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2755  * snapshots back to this dataset's origin.
2756  */
2757 static int
2758 snaplist_make(dsl_pool_t *dp,
2759     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2760 {
2761         uint64_t obj = last_obj;
2762
2763         list_create(l, sizeof (struct promotenode),
2764             offsetof(struct promotenode, link));
2765
2766         while (obj != first_obj) {
2767                 dsl_dataset_t *ds;
2768                 struct promotenode *snap;
2769                 int err;
2770
2771                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2772                 ASSERT(err != ENOENT);
2773                 if (err != 0)
2774                         return (err);
2775
2776                 if (first_obj == 0)
2777                         first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2778
2779                 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2780                 snap->ds = ds;
2781                 list_insert_tail(l, snap);
2782                 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2783         }
2784
2785         return (0);
2786 }
2787
2788 static int
2789 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2790 {
2791         struct promotenode *snap;
2792
2793         *spacep = 0;
2794         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2795                 uint64_t used, comp, uncomp;
2796                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2797                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
2798                 *spacep += used;
2799         }
2800         return (0);
2801 }
2802
2803 static void
2804 snaplist_destroy(list_t *l, void *tag)
2805 {
2806         struct promotenode *snap;
2807
2808         if (l == NULL || !list_link_active(&l->list_head))
2809                 return;
2810
2811         while ((snap = list_tail(l)) != NULL) {
2812                 list_remove(l, snap);
2813                 dsl_dataset_rele(snap->ds, tag);
2814                 kmem_free(snap, sizeof (*snap));
2815         }
2816         list_destroy(l);
2817 }
2818
2819 static int
2820 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2821 {
2822         int error;
2823         dsl_dir_t *dd;
2824         struct promotenode *snap;
2825
2826         error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2827             &ddpa->ddpa_clone);
2828         if (error != 0)
2829                 return (error);
2830         dd = ddpa->ddpa_clone->ds_dir;
2831
2832         if (ddpa->ddpa_clone->ds_is_snapshot ||
2833             !dsl_dir_is_clone(dd)) {
2834                 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2835                 return (SET_ERROR(EINVAL));
2836         }
2837
2838         error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2839             &ddpa->shared_snaps, tag);
2840         if (error != 0)
2841                 goto out;
2842
2843         error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2844             &ddpa->clone_snaps, tag);
2845         if (error != 0)
2846                 goto out;
2847
2848         snap = list_head(&ddpa->shared_snaps);
2849         ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2850         error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2851             dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2852             &ddpa->origin_snaps, tag);
2853         if (error != 0)
2854                 goto out;
2855
2856         if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2857                 error = dsl_dataset_hold_obj(dp,
2858                     dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2859                     tag, &ddpa->origin_origin);
2860                 if (error != 0)
2861                         goto out;
2862         }
2863 out:
2864         if (error != 0)
2865                 promote_rele(ddpa, tag);
2866         return (error);
2867 }
2868
2869 static void
2870 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2871 {
2872         snaplist_destroy(&ddpa->shared_snaps, tag);
2873         snaplist_destroy(&ddpa->clone_snaps, tag);
2874         snaplist_destroy(&ddpa->origin_snaps, tag);
2875         if (ddpa->origin_origin != NULL)
2876                 dsl_dataset_rele(ddpa->origin_origin, tag);
2877         dsl_dataset_rele(ddpa->ddpa_clone, tag);
2878 }
2879
2880 /*
2881  * Promote a clone.
2882  *
2883  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2884  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2885  */
2886 int
2887 dsl_dataset_promote(const char *name, char *conflsnap)
2888 {
2889         dsl_dataset_promote_arg_t ddpa = { 0 };
2890         uint64_t numsnaps;
2891         int error;
2892         objset_t *os;
2893
2894         /*
2895          * We will modify space proportional to the number of
2896          * snapshots.  Compute numsnaps.
2897          */
2898         error = dmu_objset_hold(name, FTAG, &os);
2899         if (error != 0)
2900                 return (error);
2901         error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2902             dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2903             &numsnaps);
2904         dmu_objset_rele(os, FTAG);
2905         if (error != 0)
2906                 return (error);
2907
2908         ddpa.ddpa_clonename = name;
2909         ddpa.err_ds = conflsnap;
2910         ddpa.cr = CRED();
2911
2912         return (dsl_sync_task(name, dsl_dataset_promote_check,
2913             dsl_dataset_promote_sync, &ddpa,
2914             2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2915 }
2916
2917 int
2918 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2919     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2920 {
2921         /*
2922          * "slack" factor for received datasets with refquota set on them.
2923          * See the bottom of this function for details on its use.
2924          */
2925         uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
2926         int64_t unused_refres_delta;
2927
2928         /* they should both be heads */
2929         if (clone->ds_is_snapshot ||
2930             origin_head->ds_is_snapshot)
2931                 return (SET_ERROR(EINVAL));
2932
2933         /* if we are not forcing, the branch point should be just before them */
2934         if (!force && clone->ds_prev != origin_head->ds_prev)
2935                 return (SET_ERROR(EINVAL));
2936
2937         /* clone should be the clone (unless they are unrelated) */
2938         if (clone->ds_prev != NULL &&
2939             clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2940             origin_head->ds_dir != clone->ds_prev->ds_dir)
2941                 return (SET_ERROR(EINVAL));
2942
2943         /* the clone should be a child of the origin */
2944         if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2945                 return (SET_ERROR(EINVAL));
2946
2947         /* origin_head shouldn't be modified unless 'force' */
2948         if (!force &&
2949             dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2950                 return (SET_ERROR(ETXTBSY));
2951
2952         /* origin_head should have no long holds (e.g. is not mounted) */
2953         if (dsl_dataset_handoff_check(origin_head, owner, tx))
2954                 return (SET_ERROR(EBUSY));
2955
2956         /* check amount of any unconsumed refreservation */
2957         unused_refres_delta =
2958             (int64_t)MIN(origin_head->ds_reserved,
2959             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2960             (int64_t)MIN(origin_head->ds_reserved,
2961             dsl_dataset_phys(clone)->ds_unique_bytes);
2962
2963         if (unused_refres_delta > 0 &&
2964             unused_refres_delta >
2965             dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2966                 return (SET_ERROR(ENOSPC));
2967
2968         /*
2969          * The clone can't be too much over the head's refquota.
2970          *
2971          * To ensure that the entire refquota can be used, we allow one
2972          * transaction to exceed the the refquota.  Therefore, this check
2973          * needs to also allow for the space referenced to be more than the
2974          * refquota.  The maximum amount of space that one transaction can use
2975          * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
2976          * overage ensures that we are able to receive a filesystem that
2977          * exceeds the refquota on the source system.
2978          *
2979          * So that overage is the refquota_slack we use below.
2980          */
2981         if (origin_head->ds_quota != 0 &&
2982             dsl_dataset_phys(clone)->ds_referenced_bytes >
2983             origin_head->ds_quota + refquota_slack)
2984                 return (SET_ERROR(EDQUOT));
2985
2986         return (0);
2987 }
2988
2989 void
2990 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2991     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2992 {
2993         dsl_pool_t *dp = dmu_tx_pool(tx);
2994         int64_t unused_refres_delta;
2995
2996         ASSERT(clone->ds_reserved == 0);
2997         /*
2998          * NOTE: On DEBUG kernels there could be a race between this and
2999          * the check function if spa_asize_inflation is adjusted...
3000          */
3001         ASSERT(origin_head->ds_quota == 0 ||
3002             dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
3003             DMU_MAX_ACCESS * spa_asize_inflation);
3004         ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
3005
3006         /*
3007          * Swap per-dataset feature flags.
3008          */
3009         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
3010                 if (!(spa_feature_table[f].fi_flags &
3011                     ZFEATURE_FLAG_PER_DATASET)) {
3012                         ASSERT(!clone->ds_feature_inuse[f]);
3013                         ASSERT(!origin_head->ds_feature_inuse[f]);
3014                         continue;
3015                 }
3016
3017                 boolean_t clone_inuse = clone->ds_feature_inuse[f];
3018                 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3019
3020                 if (clone_inuse) {
3021                         dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3022                         clone->ds_feature_inuse[f] = B_FALSE;
3023                 }
3024                 if (origin_head_inuse) {
3025                         dsl_dataset_deactivate_feature(origin_head->ds_object,
3026                             f, tx);
3027                         origin_head->ds_feature_inuse[f] = B_FALSE;
3028                 }
3029                 if (clone_inuse) {
3030                         dsl_dataset_activate_feature(origin_head->ds_object,
3031                             f, tx);
3032                         origin_head->ds_feature_inuse[f] = B_TRUE;
3033                 }
3034                 if (origin_head_inuse) {
3035                         dsl_dataset_activate_feature(clone->ds_object, f, tx);
3036                         clone->ds_feature_inuse[f] = B_TRUE;
3037                 }
3038         }
3039
3040         dmu_buf_will_dirty(clone->ds_dbuf, tx);
3041         dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3042
3043         if (clone->ds_objset != NULL) {
3044                 dmu_objset_evict(clone->ds_objset);
3045                 clone->ds_objset = NULL;
3046         }
3047
3048         if (origin_head->ds_objset != NULL) {
3049                 dmu_objset_evict(origin_head->ds_objset);
3050                 origin_head->ds_objset = NULL;
3051         }
3052
3053         unused_refres_delta =
3054             (int64_t)MIN(origin_head->ds_reserved,
3055             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3056             (int64_t)MIN(origin_head->ds_reserved,
3057             dsl_dataset_phys(clone)->ds_unique_bytes);
3058
3059         /*
3060          * Reset origin's unique bytes, if it exists.
3061          */
3062         if (clone->ds_prev) {
3063                 dsl_dataset_t *origin = clone->ds_prev;
3064                 uint64_t comp, uncomp;
3065
3066                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3067                 dsl_deadlist_space_range(&clone->ds_deadlist,
3068                     dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3069                     &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3070         }
3071
3072         /* swap blkptrs */
3073         {
3074                 blkptr_t tmp;
3075                 tmp = dsl_dataset_phys(origin_head)->ds_bp;
3076                 dsl_dataset_phys(origin_head)->ds_bp =
3077                     dsl_dataset_phys(clone)->ds_bp;
3078                 dsl_dataset_phys(clone)->ds_bp = tmp;
3079         }
3080
3081         /* set dd_*_bytes */
3082         {
3083                 int64_t dused, dcomp, duncomp;
3084                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3085                 uint64_t odl_used, odl_comp, odl_uncomp;
3086
3087                 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3088                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
3089
3090                 dsl_deadlist_space(&clone->ds_deadlist,
3091                     &cdl_used, &cdl_comp, &cdl_uncomp);
3092                 dsl_deadlist_space(&origin_head->ds_deadlist,
3093                     &odl_used, &odl_comp, &odl_uncomp);
3094
3095                 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3096                     cdl_used -
3097                     (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3098                     odl_used);
3099                 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3100                     cdl_comp -
3101                     (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3102                     odl_comp);
3103                 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3104                     cdl_uncomp -
3105                     (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3106                     odl_uncomp);
3107
3108                 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3109                     dused, dcomp, duncomp, tx);
3110                 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3111                     -dused, -dcomp, -duncomp, tx);
3112
3113                 /*
3114                  * The difference in the space used by snapshots is the
3115                  * difference in snapshot space due to the head's
3116                  * deadlist (since that's the only thing that's
3117                  * changing that affects the snapused).
3118                  */
3119                 dsl_deadlist_space_range(&clone->ds_deadlist,
3120                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3121                     &cdl_used, &cdl_comp, &cdl_uncomp);
3122                 dsl_deadlist_space_range(&origin_head->ds_deadlist,
3123                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3124                     &odl_used, &odl_comp, &odl_uncomp);
3125                 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3126                     DD_USED_HEAD, DD_USED_SNAP, NULL);
3127         }
3128
3129         /* swap ds_*_bytes */
3130         SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3131             dsl_dataset_phys(clone)->ds_referenced_bytes);
3132         SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3133             dsl_dataset_phys(clone)->ds_compressed_bytes);
3134         SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3135             dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3136         SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3137             dsl_dataset_phys(clone)->ds_unique_bytes);
3138
3139         /* apply any parent delta for change in unconsumed refreservation */
3140         dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3141             unused_refres_delta, 0, 0, tx);
3142
3143         /*
3144          * Swap deadlists.
3145          */
3146         dsl_deadlist_close(&clone->ds_deadlist);
3147         dsl_deadlist_close(&origin_head->ds_deadlist);
3148         SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3149             dsl_dataset_phys(clone)->ds_deadlist_obj);
3150         dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3151             dsl_dataset_phys(clone)->ds_deadlist_obj);
3152         dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3153             dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3154
3155         dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3156
3157         spa_history_log_internal_ds(clone, "clone swap", tx,
3158             "parent=%s", origin_head->ds_dir->dd_myname);
3159 }
3160
3161 /*
3162  * Given a pool name and a dataset object number in that pool,
3163  * return the name of that dataset.
3164  */
3165 int
3166 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3167 {
3168         dsl_pool_t *dp;
3169         dsl_dataset_t *ds;
3170         int error;
3171
3172         error = dsl_pool_hold(pname, FTAG, &dp);
3173         if (error != 0)
3174                 return (error);
3175
3176         error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3177         if (error == 0) {
3178                 dsl_dataset_name(ds, buf);
3179                 dsl_dataset_rele(ds, FTAG);
3180         }
3181         dsl_pool_rele(dp, FTAG);
3182
3183         return (error);
3184 }
3185
3186 int
3187 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3188     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3189 {
3190         int error = 0;
3191
3192         ASSERT3S(asize, >, 0);
3193
3194         /*
3195          * *ref_rsrv is the portion of asize that will come from any
3196          * unconsumed refreservation space.
3197          */
3198         *ref_rsrv = 0;
3199
3200         mutex_enter(&ds->ds_lock);
3201         /*
3202          * Make a space adjustment for reserved bytes.
3203          */
3204         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3205                 ASSERT3U(*used, >=,
3206                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3207                 *used -=
3208                     (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3209                 *ref_rsrv =
3210                     asize - MIN(asize, parent_delta(ds, asize + inflight));
3211         }
3212
3213         if (!check_quota || ds->ds_quota == 0) {
3214                 mutex_exit(&ds->ds_lock);
3215                 return (0);
3216         }
3217         /*
3218          * If they are requesting more space, and our current estimate
3219          * is over quota, they get to try again unless the actual
3220          * on-disk is over quota and there are no pending changes (which
3221          * may free up space for us).
3222          */
3223         if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3224             ds->ds_quota) {
3225                 if (inflight > 0 ||
3226                     dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3227                         error = SET_ERROR(ERESTART);
3228                 else
3229                         error = SET_ERROR(EDQUOT);
3230         }
3231         mutex_exit(&ds->ds_lock);
3232
3233         return (error);
3234 }
3235
3236 typedef struct dsl_dataset_set_qr_arg {
3237         const char *ddsqra_name;
3238         zprop_source_t ddsqra_source;
3239         uint64_t ddsqra_value;
3240 } dsl_dataset_set_qr_arg_t;
3241
3242
3243 /* ARGSUSED */
3244 static int
3245 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3246 {
3247         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3248         dsl_pool_t *dp = dmu_tx_pool(tx);
3249         dsl_dataset_t *ds;
3250         int error;
3251         uint64_t newval;
3252
3253         if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3254                 return (SET_ERROR(ENOTSUP));
3255
3256         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3257         if (error != 0)
3258                 return (error);
3259
3260         if (ds->ds_is_snapshot) {
3261                 dsl_dataset_rele(ds, FTAG);
3262                 return (SET_ERROR(EINVAL));
3263         }
3264
3265         error = dsl_prop_predict(ds->ds_dir,
3266             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3267             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3268         if (error != 0) {
3269                 dsl_dataset_rele(ds, FTAG);
3270                 return (error);
3271         }
3272
3273         if (newval == 0) {
3274                 dsl_dataset_rele(ds, FTAG);
3275                 return (0);
3276         }
3277
3278         if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3279             newval < ds->ds_reserved) {
3280                 dsl_dataset_rele(ds, FTAG);
3281                 return (SET_ERROR(ENOSPC));
3282         }
3283
3284         dsl_dataset_rele(ds, FTAG);
3285         return (0);
3286 }
3287
3288 static void
3289 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3290 {
3291         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3292         dsl_pool_t *dp = dmu_tx_pool(tx);
3293         dsl_dataset_t *ds;
3294         uint64_t newval;
3295
3296         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3297
3298         dsl_prop_set_sync_impl(ds,
3299             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3300             ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3301             &ddsqra->ddsqra_value, tx);
3302
3303         VERIFY0(dsl_prop_get_int_ds(ds,
3304             zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3305
3306         if (ds->ds_quota != newval) {
3307                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3308                 ds->ds_quota = newval;
3309         }
3310         dsl_dataset_rele(ds, FTAG);
3311 }
3312
3313 int
3314 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3315     uint64_t refquota)
3316 {
3317         dsl_dataset_set_qr_arg_t ddsqra;
3318
3319         ddsqra.ddsqra_name = dsname;
3320         ddsqra.ddsqra_source = source;
3321         ddsqra.ddsqra_value = refquota;
3322
3323         return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3324             dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3325 }
3326
3327 static int
3328 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3329 {
3330         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3331         dsl_pool_t *dp = dmu_tx_pool(tx);
3332         dsl_dataset_t *ds;
3333         int error;
3334         uint64_t newval, unique;
3335
3336         if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3337                 return (SET_ERROR(ENOTSUP));
3338
3339         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3340         if (error != 0)
3341                 return (error);
3342
3343         if (ds->ds_is_snapshot) {
3344                 dsl_dataset_rele(ds, FTAG);
3345                 return (SET_ERROR(EINVAL));
3346         }
3347
3348         error = dsl_prop_predict(ds->ds_dir,
3349             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3350             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3351         if (error != 0) {
3352                 dsl_dataset_rele(ds, FTAG);
3353                 return (error);
3354         }
3355
3356         /*
3357          * If we are doing the preliminary check in open context, the
3358          * space estimates may be inaccurate.
3359          */
3360         if (!dmu_tx_is_syncing(tx)) {
3361                 dsl_dataset_rele(ds, FTAG);
3362                 return (0);
3363         }
3364
3365         mutex_enter(&ds->ds_lock);
3366         if (!DS_UNIQUE_IS_ACCURATE(ds))
3367                 dsl_dataset_recalc_head_uniq(ds);
3368         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3369         mutex_exit(&ds->ds_lock);
3370
3371         if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3372                 uint64_t delta = MAX(unique, newval) -
3373                     MAX(unique, ds->ds_reserved);
3374
3375                 if (delta >
3376                     dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3377                     (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3378                         dsl_dataset_rele(ds, FTAG);
3379                         return (SET_ERROR(ENOSPC));
3380                 }
3381         }
3382
3383         dsl_dataset_rele(ds, FTAG);
3384         return (0);
3385 }
3386
3387 void
3388 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3389     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3390 {
3391         uint64_t newval;
3392         uint64_t unique;
3393         int64_t delta;
3394
3395         dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3396             source, sizeof (value), 1, &value, tx);
3397
3398         VERIFY0(dsl_prop_get_int_ds(ds,
3399             zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3400
3401         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3402         mutex_enter(&ds->ds_dir->dd_lock);
3403         mutex_enter(&ds->ds_lock);
3404         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3405         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3406         delta = MAX(0, (int64_t)(newval - unique)) -
3407             MAX(0, (int64_t)(ds->ds_reserved - unique));
3408         ds->ds_reserved = newval;
3409         mutex_exit(&ds->ds_lock);
3410
3411         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3412         mutex_exit(&ds->ds_dir->dd_lock);
3413 }
3414
3415 static void
3416 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3417 {
3418         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3419         dsl_pool_t *dp = dmu_tx_pool(tx);
3420         dsl_dataset_t *ds;
3421
3422         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3423         dsl_dataset_set_refreservation_sync_impl(ds,
3424             ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3425         dsl_dataset_rele(ds, FTAG);
3426 }
3427
3428 int
3429 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3430     uint64_t refreservation)
3431 {
3432         dsl_dataset_set_qr_arg_t ddsqra;
3433
3434         ddsqra.ddsqra_name = dsname;
3435         ddsqra.ddsqra_source = source;
3436         ddsqra.ddsqra_value = refreservation;
3437
3438         return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3439             dsl_dataset_set_refreservation_sync, &ddsqra,
3440             0, ZFS_SPACE_CHECK_NONE));
3441 }
3442
3443 /*
3444  * Return (in *usedp) the amount of space written in new that is not
3445  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3446  * a snapshot before new, in new's filesystem (or its origin).  If not then
3447  * fail and return EINVAL.
3448  *
3449  * The written space is calculated by considering two components:  First, we
3450  * ignore any freed space, and calculate the written as new's used space
3451  * minus old's used space.  Next, we add in the amount of space that was freed
3452  * between the two snapshots, thus reducing new's used space relative to old's.
3453  * Specifically, this is the space that was born before old->ds_creation_txg,
3454  * and freed before new (ie. on new's deadlist or a previous deadlist).
3455  *
3456  * space freed                         [---------------------]
3457  * snapshots                       ---O-------O--------O-------O------
3458  *                                         oldsnap            new
3459  */
3460 int
3461 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3462     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3463 {
3464         int err = 0;
3465         uint64_t snapobj;
3466         dsl_pool_t *dp = new->ds_dir->dd_pool;
3467
3468         ASSERT(dsl_pool_config_held(dp));
3469
3470         *usedp = 0;
3471         *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3472         *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3473
3474         *compp = 0;
3475         *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3476         *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3477
3478         *uncompp = 0;
3479         *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3480         *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3481
3482         snapobj = new->ds_object;
3483         while (snapobj != oldsnap->ds_object) {
3484                 dsl_dataset_t *snap;
3485                 uint64_t used, comp, uncomp;
3486
3487                 if (snapobj == new->ds_object) {
3488                         snap = new;
3489                 } else {
3490                         err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3491                         if (err != 0)
3492                                 break;
3493                 }
3494
3495                 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3496                     dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3497                         /*
3498                          * The blocks in the deadlist can not be born after
3499                          * ds_prev_snap_txg, so get the whole deadlist space,
3500                          * which is more efficient (especially for old-format
3501                          * deadlists).  Unfortunately the deadlist code
3502                          * doesn't have enough information to make this
3503                          * optimization itself.
3504                          */
3505                         dsl_deadlist_space(&snap->ds_deadlist,
3506                             &used, &comp, &uncomp);
3507                 } else {
3508                         dsl_deadlist_space_range(&snap->ds_deadlist,
3509                             0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3510                             &used, &comp, &uncomp);
3511                 }
3512                 *usedp += used;
3513                 *compp += comp;
3514                 *uncompp += uncomp;
3515
3516                 /*
3517                  * If we get to the beginning of the chain of snapshots
3518                  * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3519                  * was not a snapshot of/before new.
3520                  */
3521                 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3522                 if (snap != new)
3523                         dsl_dataset_rele(snap, FTAG);
3524                 if (snapobj == 0) {
3525                         err = SET_ERROR(EINVAL);
3526                         break;
3527                 }
3528
3529         }
3530         return (err);
3531 }
3532
3533 /*
3534  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3535  * lastsnap, and all snapshots in between are deleted.
3536  *
3537  * blocks that would be freed            [---------------------------]
3538  * snapshots                       ---O-------O--------O-------O--------O
3539  *                                        firstsnap        lastsnap
3540  *
3541  * This is the set of blocks that were born after the snap before firstsnap,
3542  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3543  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3544  * We calculate this by iterating over the relevant deadlists (from the snap
3545  * after lastsnap, backward to the snap after firstsnap), summing up the
3546  * space on the deadlist that was born after the snap before firstsnap.
3547  */
3548 int
3549 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3550     dsl_dataset_t *lastsnap,
3551     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3552 {
3553         int err = 0;
3554         uint64_t snapobj;
3555         dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3556
3557         ASSERT(firstsnap->ds_is_snapshot);
3558         ASSERT(lastsnap->ds_is_snapshot);
3559
3560         /*
3561          * Check that the snapshots are in the same dsl_dir, and firstsnap
3562          * is before lastsnap.
3563          */
3564         if (firstsnap->ds_dir != lastsnap->ds_dir ||
3565             dsl_dataset_phys(firstsnap)->ds_creation_txg >
3566             dsl_dataset_phys(lastsnap)->ds_creation_txg)
3567                 return (SET_ERROR(EINVAL));
3568
3569         *usedp = *compp = *uncompp = 0;
3570
3571         snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3572         while (snapobj != firstsnap->ds_object) {
3573                 dsl_dataset_t *ds;
3574                 uint64_t used, comp, uncomp;
3575
3576                 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3577                 if (err != 0)
3578                         break;
3579
3580                 dsl_deadlist_space_range(&ds->ds_deadlist,
3581                     dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3582                     &used, &comp, &uncomp);
3583                 *usedp += used;
3584                 *compp += comp;
3585                 *uncompp += uncomp;
3586
3587                 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3588                 ASSERT3U(snapobj, !=, 0);
3589                 dsl_dataset_rele(ds, FTAG);
3590         }
3591         return (err);
3592 }
3593
3594 /*
3595  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3596  * For example, they could both be snapshots of the same filesystem, and
3597  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3598  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3599  * filesystem.  Or 'earlier' could be the origin's origin.
3600  *
3601  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3602  */
3603 boolean_t
3604 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3605     uint64_t earlier_txg)
3606 {
3607         dsl_pool_t *dp = later->ds_dir->dd_pool;
3608         int error;
3609         boolean_t ret;
3610
3611         ASSERT(dsl_pool_config_held(dp));
3612         ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3613
3614         if (earlier_txg == 0)
3615                 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3616
3617         if (later->ds_is_snapshot &&
3618             earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3619                 return (B_FALSE);
3620
3621         if (later->ds_dir == earlier->ds_dir)
3622                 return (B_TRUE);
3623         if (!dsl_dir_is_clone(later->ds_dir))
3624                 return (B_FALSE);
3625
3626         if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3627                 return (B_TRUE);
3628         dsl_dataset_t *origin;
3629         error = dsl_dataset_hold_obj(dp,
3630             dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3631         if (error != 0)
3632                 return (B_FALSE);
3633         ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3634         dsl_dataset_rele(origin, FTAG);
3635         return (ret);
3636 }
3637
3638 void
3639 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3640 {
3641         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3642         dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3643 }
3644
3645 boolean_t
3646 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3647 {
3648         dmu_object_info_t doi;
3649
3650         dmu_object_info_from_db(ds->ds_dbuf, &doi);
3651         return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3652 }
3653
3654 boolean_t
3655 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3656 {
3657         return (dsl_dataset_is_zapified(ds) &&
3658             zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3659             ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3660 }