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