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