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

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