1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 225afc78aaSChris Kirby * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23187d6ac0SMatt Ahrens * Copyright (c) 2011 by Delphix. All rights reserved. 24*4e3c9f44SBill Pijewski * Copyright (c) 2012, Joyent, Inc. All rights reserved. 25fa9e4066Sahrens */ 26fa9e4066Sahrens 27fa9e4066Sahrens #include <sys/dmu_objset.h> 28fa9e4066Sahrens #include <sys/dsl_dataset.h> 29fa9e4066Sahrens #include <sys/dsl_dir.h> 3099653d4eSeschrock #include <sys/dsl_prop.h> 311d452cf5Sahrens #include <sys/dsl_synctask.h> 32fa9e4066Sahrens #include <sys/dmu_traverse.h> 33*4e3c9f44SBill Pijewski #include <sys/dmu_impl.h> 34fa9e4066Sahrens #include <sys/dmu_tx.h> 35fa9e4066Sahrens #include <sys/arc.h> 36fa9e4066Sahrens #include <sys/zio.h> 37fa9e4066Sahrens #include <sys/zap.h> 38fa9e4066Sahrens #include <sys/unique.h> 39fa9e4066Sahrens #include <sys/zfs_context.h> 40cdf5b4caSmmusante #include <sys/zfs_ioctl.h> 41ecd6cf80Smarks #include <sys/spa.h> 42088f3894Sahrens #include <sys/zfs_znode.h> 43c99e4bdcSChris Kirby #include <sys/zfs_onexit.h> 44842727c2SChris Kirby #include <sys/zvol.h> 453f9d6ad7SLin Ling #include <sys/dsl_scan.h> 46cde58dbcSMatthew Ahrens #include <sys/dsl_deadlist.h> 47fa9e4066Sahrens 48745cd3c5Smaybee static char *dsl_reaper = "the grim reaper"; 49745cd3c5Smaybee 501d452cf5Sahrens static dsl_checkfunc_t dsl_dataset_destroy_begin_check; 511d452cf5Sahrens static dsl_syncfunc_t dsl_dataset_destroy_begin_sync; 52a9799022Sck static dsl_syncfunc_t dsl_dataset_set_reservation_sync; 53e1930233Sbonwick 54cde58dbcSMatthew Ahrens #define SWITCH64(x, y) \ 55cde58dbcSMatthew Ahrens { \ 56cde58dbcSMatthew Ahrens uint64_t __tmp = (x); \ 57cde58dbcSMatthew Ahrens (x) = (y); \ 58cde58dbcSMatthew Ahrens (y) = __tmp; \ 59cde58dbcSMatthew Ahrens } 60cde58dbcSMatthew Ahrens 6155434c77Sek #define DS_REF_MAX (1ULL << 62) 62fa9e4066Sahrens 63fa9e4066Sahrens #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE 64fa9e4066Sahrens 65745cd3c5Smaybee #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper) 66745cd3c5Smaybee 67fa9e4066Sahrens 68a9799022Sck /* 69a9799022Sck * Figure out how much of this delta should be propogated to the dsl_dir 70a9799022Sck * layer. If there's a refreservation, that space has already been 71a9799022Sck * partially accounted for in our ancestors. 72a9799022Sck */ 73a9799022Sck static int64_t 74a9799022Sck parent_delta(dsl_dataset_t *ds, int64_t delta) 75a9799022Sck { 76a9799022Sck uint64_t old_bytes, new_bytes; 77a9799022Sck 78a9799022Sck if (ds->ds_reserved == 0) 79a9799022Sck return (delta); 80a9799022Sck 81a9799022Sck old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 82a9799022Sck new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 83a9799022Sck 84a9799022Sck ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 85a9799022Sck return (new_bytes - old_bytes); 86a9799022Sck } 87fa9e4066Sahrens 88fa9e4066Sahrens void 89b24ab676SJeff Bonwick dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 90fa9e4066Sahrens { 91b24ab676SJeff Bonwick int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 92fa9e4066Sahrens int compressed = BP_GET_PSIZE(bp); 93fa9e4066Sahrens int uncompressed = BP_GET_UCSIZE(bp); 94a9799022Sck int64_t delta; 95fa9e4066Sahrens 963f9d6ad7SLin Ling dprintf_bp(bp, "ds=%p", ds); 97fa9e4066Sahrens 98fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 99fa9e4066Sahrens /* It could have been compressed away to nothing */ 100fa9e4066Sahrens if (BP_IS_HOLE(bp)) 101fa9e4066Sahrens return; 102fa9e4066Sahrens ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 103fa9e4066Sahrens ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES); 104fa9e4066Sahrens if (ds == NULL) { 105fa9e4066Sahrens /* 106fa9e4066Sahrens * Account for the meta-objset space in its placeholder 107fa9e4066Sahrens * dsl_dir. 108fa9e4066Sahrens */ 109fa9e4066Sahrens ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */ 11074e7dc98SMatthew Ahrens dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, 111fa9e4066Sahrens used, compressed, uncompressed, tx); 112fa9e4066Sahrens dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); 113fa9e4066Sahrens return; 114fa9e4066Sahrens } 115fa9e4066Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 1163f9d6ad7SLin Ling 11702c8f3f0SMatthew Ahrens mutex_enter(&ds->ds_dir->dd_lock); 118fa9e4066Sahrens mutex_enter(&ds->ds_lock); 119a9799022Sck delta = parent_delta(ds, used); 120fa9e4066Sahrens ds->ds_phys->ds_used_bytes += used; 121fa9e4066Sahrens ds->ds_phys->ds_compressed_bytes += compressed; 122fa9e4066Sahrens ds->ds_phys->ds_uncompressed_bytes += uncompressed; 123fa9e4066Sahrens ds->ds_phys->ds_unique_bytes += used; 124fa9e4066Sahrens mutex_exit(&ds->ds_lock); 12574e7dc98SMatthew Ahrens dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 12674e7dc98SMatthew Ahrens compressed, uncompressed, tx); 12774e7dc98SMatthew Ahrens dsl_dir_transfer_space(ds->ds_dir, used - delta, 12874e7dc98SMatthew Ahrens DD_USED_REFRSRV, DD_USED_HEAD, tx); 12902c8f3f0SMatthew Ahrens mutex_exit(&ds->ds_dir->dd_lock); 130fa9e4066Sahrens } 131fa9e4066Sahrens 132cdb0ab79Smaybee int 133b24ab676SJeff Bonwick dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 134b24ab676SJeff Bonwick boolean_t async) 135fa9e4066Sahrens { 136fa9e4066Sahrens if (BP_IS_HOLE(bp)) 137cdb0ab79Smaybee return (0); 138fa9e4066Sahrens 139b24ab676SJeff Bonwick ASSERT(dmu_tx_is_syncing(tx)); 140b24ab676SJeff Bonwick ASSERT(bp->blk_birth <= tx->tx_txg); 141b24ab676SJeff Bonwick 142b24ab676SJeff Bonwick int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 143b24ab676SJeff Bonwick int compressed = BP_GET_PSIZE(bp); 144b24ab676SJeff Bonwick int uncompressed = BP_GET_UCSIZE(bp); 145b24ab676SJeff Bonwick 146fa9e4066Sahrens ASSERT(used > 0); 147fa9e4066Sahrens if (ds == NULL) { 148fa9e4066Sahrens /* 149fa9e4066Sahrens * Account for the meta-objset space in its placeholder 150fa9e4066Sahrens * dataset. 151fa9e4066Sahrens */ 152b24ab676SJeff Bonwick dsl_free(tx->tx_pool, tx->tx_txg, bp); 153fa9e4066Sahrens 15474e7dc98SMatthew Ahrens dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, 155fa9e4066Sahrens -used, -compressed, -uncompressed, tx); 156fa9e4066Sahrens dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); 157cdb0ab79Smaybee return (used); 158fa9e4066Sahrens } 159fa9e4066Sahrens ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 160fa9e4066Sahrens 16174e7dc98SMatthew Ahrens ASSERT(!dsl_dataset_is_snapshot(ds)); 162fa9e4066Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 163fa9e4066Sahrens 164fa9e4066Sahrens if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { 165a9799022Sck int64_t delta; 166c717a561Smaybee 1673f9d6ad7SLin Ling dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 168b24ab676SJeff Bonwick dsl_free(tx->tx_pool, tx->tx_txg, bp); 169fa9e4066Sahrens 17002c8f3f0SMatthew Ahrens mutex_enter(&ds->ds_dir->dd_lock); 171fa9e4066Sahrens mutex_enter(&ds->ds_lock); 172a9799022Sck ASSERT(ds->ds_phys->ds_unique_bytes >= used || 173a9799022Sck !DS_UNIQUE_IS_ACCURATE(ds)); 174a9799022Sck delta = parent_delta(ds, -used); 175fa9e4066Sahrens ds->ds_phys->ds_unique_bytes -= used; 176fa9e4066Sahrens mutex_exit(&ds->ds_lock); 17774e7dc98SMatthew Ahrens dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 178a9799022Sck delta, -compressed, -uncompressed, tx); 17974e7dc98SMatthew Ahrens dsl_dir_transfer_space(ds->ds_dir, -used - delta, 18074e7dc98SMatthew Ahrens DD_USED_REFRSRV, DD_USED_HEAD, tx); 18102c8f3f0SMatthew Ahrens mutex_exit(&ds->ds_dir->dd_lock); 182fa9e4066Sahrens } else { 183fa9e4066Sahrens dprintf_bp(bp, "putting on dead list: %s", ""); 184b24ab676SJeff Bonwick if (async) { 185b24ab676SJeff Bonwick /* 186b24ab676SJeff Bonwick * We are here as part of zio's write done callback, 187b24ab676SJeff Bonwick * which means we're a zio interrupt thread. We can't 188cde58dbcSMatthew Ahrens * call dsl_deadlist_insert() now because it may block 189b24ab676SJeff Bonwick * waiting for I/O. Instead, put bp on the deferred 190b24ab676SJeff Bonwick * queue and let dsl_pool_sync() finish the job. 191b24ab676SJeff Bonwick */ 192cde58dbcSMatthew Ahrens bplist_append(&ds->ds_pending_deadlist, bp); 193b24ab676SJeff Bonwick } else { 194cde58dbcSMatthew Ahrens dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 195b24ab676SJeff Bonwick } 196a4611edeSahrens ASSERT3U(ds->ds_prev->ds_object, ==, 197a4611edeSahrens ds->ds_phys->ds_prev_snap_obj); 198a4611edeSahrens ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); 199fa9e4066Sahrens /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 200a4611edeSahrens if (ds->ds_prev->ds_phys->ds_next_snap_obj == 201a4611edeSahrens ds->ds_object && bp->blk_birth > 202a4611edeSahrens ds->ds_prev->ds_phys->ds_prev_snap_txg) { 203a4611edeSahrens dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 204a4611edeSahrens mutex_enter(&ds->ds_prev->ds_lock); 205a4611edeSahrens ds->ds_prev->ds_phys->ds_unique_bytes += used; 206a4611edeSahrens mutex_exit(&ds->ds_prev->ds_lock); 207fa9e4066Sahrens } 2083f9d6ad7SLin Ling if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 20974e7dc98SMatthew Ahrens dsl_dir_transfer_space(ds->ds_dir, used, 21074e7dc98SMatthew Ahrens DD_USED_HEAD, DD_USED_SNAP, tx); 21174e7dc98SMatthew Ahrens } 212fa9e4066Sahrens } 213fa9e4066Sahrens mutex_enter(&ds->ds_lock); 214fa9e4066Sahrens ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); 215fa9e4066Sahrens ds->ds_phys->ds_used_bytes -= used; 216fa9e4066Sahrens ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); 217fa9e4066Sahrens ds->ds_phys->ds_compressed_bytes -= compressed; 218fa9e4066Sahrens ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); 219fa9e4066Sahrens ds->ds_phys->ds_uncompressed_bytes -= uncompressed; 220fa9e4066Sahrens mutex_exit(&ds->ds_lock); 221cdb0ab79Smaybee 222cdb0ab79Smaybee return (used); 223fa9e4066Sahrens } 224fa9e4066Sahrens 225ea8dc4b6Seschrock uint64_t 226ea8dc4b6Seschrock dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 227fa9e4066Sahrens { 228a2eea2e1Sahrens uint64_t trysnap = 0; 229a2eea2e1Sahrens 230fa9e4066Sahrens if (ds == NULL) 231ea8dc4b6Seschrock return (0); 232fa9e4066Sahrens /* 233fa9e4066Sahrens * The snapshot creation could fail, but that would cause an 234fa9e4066Sahrens * incorrect FALSE return, which would only result in an 235fa9e4066Sahrens * overestimation of the amount of space that an operation would 236fa9e4066Sahrens * consume, which is OK. 237fa9e4066Sahrens * 238fa9e4066Sahrens * There's also a small window where we could miss a pending 239fa9e4066Sahrens * snapshot, because we could set the sync task in the quiescing 240fa9e4066Sahrens * phase. So this should only be used as a guess. 241fa9e4066Sahrens */ 242a2eea2e1Sahrens if (ds->ds_trysnap_txg > 243a2eea2e1Sahrens spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 244a2eea2e1Sahrens trysnap = ds->ds_trysnap_txg; 245a2eea2e1Sahrens return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); 246ea8dc4b6Seschrock } 247ea8dc4b6Seschrock 2483d692628SSanjeev Bagewadi boolean_t 249c7cd2421SGeorge Wilson dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, 250c7cd2421SGeorge Wilson uint64_t blk_birth) 251ea8dc4b6Seschrock { 252c7cd2421SGeorge Wilson if (blk_birth <= dsl_dataset_prev_snap_txg(ds)) 253c7cd2421SGeorge Wilson return (B_FALSE); 254c7cd2421SGeorge Wilson 255837b568bSGeorge Wilson ddt_prefetch(dsl_dataset_get_spa(ds), bp); 256c7cd2421SGeorge Wilson 257c7cd2421SGeorge Wilson return (B_TRUE); 258fa9e4066Sahrens } 259fa9e4066Sahrens 260fa9e4066Sahrens /* ARGSUSED */ 261fa9e4066Sahrens static void 262fa9e4066Sahrens dsl_dataset_evict(dmu_buf_t *db, void *dsv) 263fa9e4066Sahrens { 264fa9e4066Sahrens dsl_dataset_t *ds = dsv; 265fa9e4066Sahrens 266745cd3c5Smaybee ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds)); 267fa9e4066Sahrens 26891ebeef5Sahrens unique_remove(ds->ds_fsid_guid); 269fa9e4066Sahrens 270503ad85cSMatthew Ahrens if (ds->ds_objset != NULL) 271503ad85cSMatthew Ahrens dmu_objset_evict(ds->ds_objset); 272fa9e4066Sahrens 273fa9e4066Sahrens if (ds->ds_prev) { 274745cd3c5Smaybee dsl_dataset_drop_ref(ds->ds_prev, ds); 275fa9e4066Sahrens ds->ds_prev = NULL; 276fa9e4066Sahrens } 277fa9e4066Sahrens 278cde58dbcSMatthew Ahrens bplist_destroy(&ds->ds_pending_deadlist); 279cde58dbcSMatthew Ahrens if (db != NULL) { 280cde58dbcSMatthew Ahrens dsl_deadlist_close(&ds->ds_deadlist); 281cde58dbcSMatthew Ahrens } else { 282cde58dbcSMatthew Ahrens ASSERT(ds->ds_deadlist.dl_dbuf == NULL); 283cde58dbcSMatthew Ahrens ASSERT(!ds->ds_deadlist.dl_oldfmt); 284cde58dbcSMatthew Ahrens } 285745cd3c5Smaybee if (ds->ds_dir) 286745cd3c5Smaybee dsl_dir_close(ds->ds_dir, ds); 287fa9e4066Sahrens 28891ebeef5Sahrens ASSERT(!list_link_active(&ds->ds_synced_link)); 289fa9e4066Sahrens 2905ad82045Snd mutex_destroy(&ds->ds_lock); 291f4b94bdeSMatthew Ahrens mutex_destroy(&ds->ds_recvlock); 29291ebeef5Sahrens mutex_destroy(&ds->ds_opening_lock); 293745cd3c5Smaybee rw_destroy(&ds->ds_rwlock); 294745cd3c5Smaybee cv_destroy(&ds->ds_exclusive_cv); 2955ad82045Snd 296fa9e4066Sahrens kmem_free(ds, sizeof (dsl_dataset_t)); 297fa9e4066Sahrens } 298fa9e4066Sahrens 299ea8dc4b6Seschrock static int 300fa9e4066Sahrens dsl_dataset_get_snapname(dsl_dataset_t *ds) 301fa9e4066Sahrens { 302fa9e4066Sahrens dsl_dataset_phys_t *headphys; 303fa9e4066Sahrens int err; 304fa9e4066Sahrens dmu_buf_t *headdbuf; 305fa9e4066Sahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 306fa9e4066Sahrens objset_t *mos = dp->dp_meta_objset; 307fa9e4066Sahrens 308fa9e4066Sahrens if (ds->ds_snapname[0]) 309ea8dc4b6Seschrock return (0); 310fa9e4066Sahrens if (ds->ds_phys->ds_next_snap_obj == 0) 311ea8dc4b6Seschrock return (0); 312fa9e4066Sahrens 313ea8dc4b6Seschrock err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, 314ea8dc4b6Seschrock FTAG, &headdbuf); 315ea8dc4b6Seschrock if (err) 316ea8dc4b6Seschrock return (err); 317fa9e4066Sahrens headphys = headdbuf->db_data; 318fa9e4066Sahrens err = zap_value_search(dp->dp_meta_objset, 319e7437265Sahrens headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 320ea8dc4b6Seschrock dmu_buf_rele(headdbuf, FTAG); 321ea8dc4b6Seschrock return (err); 322fa9e4066Sahrens } 323fa9e4066Sahrens 324ab04eb8eStimh static int 325745cd3c5Smaybee dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 326ab04eb8eStimh { 327745cd3c5Smaybee objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 328745cd3c5Smaybee uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 329ab04eb8eStimh matchtype_t mt; 330ab04eb8eStimh int err; 331ab04eb8eStimh 332745cd3c5Smaybee if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 333ab04eb8eStimh mt = MT_FIRST; 334ab04eb8eStimh else 335ab04eb8eStimh mt = MT_EXACT; 336ab04eb8eStimh 337745cd3c5Smaybee err = zap_lookup_norm(mos, snapobj, name, 8, 1, 338ab04eb8eStimh value, mt, NULL, 0, NULL); 339ab04eb8eStimh if (err == ENOTSUP && mt == MT_FIRST) 340745cd3c5Smaybee err = zap_lookup(mos, snapobj, name, 8, 1, value); 341ab04eb8eStimh return (err); 342ab04eb8eStimh } 343ab04eb8eStimh 344ab04eb8eStimh static int 345745cd3c5Smaybee dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx) 346ab04eb8eStimh { 347745cd3c5Smaybee objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 348745cd3c5Smaybee uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 349ab04eb8eStimh matchtype_t mt; 350ab04eb8eStimh int err; 351ab04eb8eStimh 35271eb0538SChris Kirby dsl_dir_snap_cmtime_update(ds->ds_dir); 35371eb0538SChris Kirby 354745cd3c5Smaybee if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 355ab04eb8eStimh mt = MT_FIRST; 356ab04eb8eStimh else 357ab04eb8eStimh mt = MT_EXACT; 358ab04eb8eStimh 359745cd3c5Smaybee err = zap_remove_norm(mos, snapobj, name, mt, tx); 360ab04eb8eStimh if (err == ENOTSUP && mt == MT_FIRST) 361745cd3c5Smaybee err = zap_remove(mos, snapobj, name, tx); 362ab04eb8eStimh return (err); 363ab04eb8eStimh } 364ab04eb8eStimh 365745cd3c5Smaybee static int 366745cd3c5Smaybee dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, 367745cd3c5Smaybee dsl_dataset_t **dsp) 368fa9e4066Sahrens { 369fa9e4066Sahrens objset_t *mos = dp->dp_meta_objset; 370fa9e4066Sahrens dmu_buf_t *dbuf; 371fa9e4066Sahrens dsl_dataset_t *ds; 372ea8dc4b6Seschrock int err; 373a7f53a56SChris Kirby dmu_object_info_t doi; 374fa9e4066Sahrens 375fa9e4066Sahrens ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 376fa9e4066Sahrens dsl_pool_sync_context(dp)); 377fa9e4066Sahrens 378ea8dc4b6Seschrock err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 379ea8dc4b6Seschrock if (err) 380ea8dc4b6Seschrock return (err); 381a7f53a56SChris Kirby 382a7f53a56SChris Kirby /* Make sure dsobj has the correct object type. */ 383a7f53a56SChris Kirby dmu_object_info_from_db(dbuf, &doi); 384a7f53a56SChris Kirby if (doi.doi_type != DMU_OT_DSL_DATASET) 385a7f53a56SChris Kirby return (EINVAL); 386a7f53a56SChris Kirby 387fa9e4066Sahrens ds = dmu_buf_get_user(dbuf); 388fa9e4066Sahrens if (ds == NULL) { 389fa9e4066Sahrens dsl_dataset_t *winner; 390fa9e4066Sahrens 391fa9e4066Sahrens ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 392fa9e4066Sahrens ds->ds_dbuf = dbuf; 393fa9e4066Sahrens ds->ds_object = dsobj; 394fa9e4066Sahrens ds->ds_phys = dbuf->db_data; 395fa9e4066Sahrens 3965ad82045Snd mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 397f4b94bdeSMatthew Ahrens mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL); 39891ebeef5Sahrens mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 399*4e3c9f44SBill Pijewski mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 400*4e3c9f44SBill Pijewski 401745cd3c5Smaybee rw_init(&ds->ds_rwlock, 0, 0, 0); 402745cd3c5Smaybee cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL); 4035ad82045Snd 404cde58dbcSMatthew Ahrens bplist_create(&ds->ds_pending_deadlist); 405cde58dbcSMatthew Ahrens dsl_deadlist_open(&ds->ds_deadlist, 406fa9e4066Sahrens mos, ds->ds_phys->ds_deadlist_obj); 407cde58dbcSMatthew Ahrens 408*4e3c9f44SBill Pijewski list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 409*4e3c9f44SBill Pijewski offsetof(dmu_sendarg_t, dsa_link)); 410*4e3c9f44SBill Pijewski 411ea8dc4b6Seschrock if (err == 0) { 412ea8dc4b6Seschrock err = dsl_dir_open_obj(dp, 413ea8dc4b6Seschrock ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); 414ea8dc4b6Seschrock } 415ea8dc4b6Seschrock if (err) { 4165ad82045Snd mutex_destroy(&ds->ds_lock); 417f4b94bdeSMatthew Ahrens mutex_destroy(&ds->ds_recvlock); 41891ebeef5Sahrens mutex_destroy(&ds->ds_opening_lock); 419745cd3c5Smaybee rw_destroy(&ds->ds_rwlock); 420745cd3c5Smaybee cv_destroy(&ds->ds_exclusive_cv); 421cde58dbcSMatthew Ahrens bplist_destroy(&ds->ds_pending_deadlist); 422cde58dbcSMatthew Ahrens dsl_deadlist_close(&ds->ds_deadlist); 423ea8dc4b6Seschrock kmem_free(ds, sizeof (dsl_dataset_t)); 424ea8dc4b6Seschrock dmu_buf_rele(dbuf, tag); 425ea8dc4b6Seschrock return (err); 426ea8dc4b6Seschrock } 427fa9e4066Sahrens 42874e7dc98SMatthew Ahrens if (!dsl_dataset_is_snapshot(ds)) { 429fa9e4066Sahrens ds->ds_snapname[0] = '\0'; 430fa9e4066Sahrens if (ds->ds_phys->ds_prev_snap_obj) { 431745cd3c5Smaybee err = dsl_dataset_get_ref(dp, 432745cd3c5Smaybee ds->ds_phys->ds_prev_snap_obj, 433745cd3c5Smaybee ds, &ds->ds_prev); 434fa9e4066Sahrens } 435842727c2SChris Kirby } else { 436842727c2SChris Kirby if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 437842727c2SChris Kirby err = dsl_dataset_get_snapname(ds); 438842727c2SChris Kirby if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) { 439842727c2SChris Kirby err = zap_count( 440842727c2SChris Kirby ds->ds_dir->dd_pool->dp_meta_objset, 441842727c2SChris Kirby ds->ds_phys->ds_userrefs_obj, 442842727c2SChris Kirby &ds->ds_userrefs); 443842727c2SChris Kirby } 444fa9e4066Sahrens } 445fa9e4066Sahrens 44674e7dc98SMatthew Ahrens if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 44727345066Sck /* 44827345066Sck * In sync context, we're called with either no lock 44927345066Sck * or with the write lock. If we're not syncing, 45027345066Sck * we're always called with the read lock held. 45127345066Sck */ 452cb625fb5Sck boolean_t need_lock = 45327345066Sck !RW_WRITE_HELD(&dp->dp_config_rwlock) && 45427345066Sck dsl_pool_sync_context(dp); 455cb625fb5Sck 456cb625fb5Sck if (need_lock) 457cb625fb5Sck rw_enter(&dp->dp_config_rwlock, RW_READER); 458cb625fb5Sck 459bb0ade09Sahrens err = dsl_prop_get_ds(ds, 460cb625fb5Sck "refreservation", sizeof (uint64_t), 1, 461cb625fb5Sck &ds->ds_reserved, NULL); 462cb625fb5Sck if (err == 0) { 463bb0ade09Sahrens err = dsl_prop_get_ds(ds, 464cb625fb5Sck "refquota", sizeof (uint64_t), 1, 465cb625fb5Sck &ds->ds_quota, NULL); 466cb625fb5Sck } 467cb625fb5Sck 468cb625fb5Sck if (need_lock) 469cb625fb5Sck rw_exit(&dp->dp_config_rwlock); 470cb625fb5Sck } else { 471cb625fb5Sck ds->ds_reserved = ds->ds_quota = 0; 472cb625fb5Sck } 473cb625fb5Sck 474ea8dc4b6Seschrock if (err == 0) { 475ea8dc4b6Seschrock winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, 476ea8dc4b6Seschrock dsl_dataset_evict); 477ea8dc4b6Seschrock } 478ea8dc4b6Seschrock if (err || winner) { 479cde58dbcSMatthew Ahrens bplist_destroy(&ds->ds_pending_deadlist); 480cde58dbcSMatthew Ahrens dsl_deadlist_close(&ds->ds_deadlist); 481745cd3c5Smaybee if (ds->ds_prev) 482745cd3c5Smaybee dsl_dataset_drop_ref(ds->ds_prev, ds); 483fa9e4066Sahrens dsl_dir_close(ds->ds_dir, ds); 4845ad82045Snd mutex_destroy(&ds->ds_lock); 485f4b94bdeSMatthew Ahrens mutex_destroy(&ds->ds_recvlock); 48691ebeef5Sahrens mutex_destroy(&ds->ds_opening_lock); 487745cd3c5Smaybee rw_destroy(&ds->ds_rwlock); 488745cd3c5Smaybee cv_destroy(&ds->ds_exclusive_cv); 489fa9e4066Sahrens kmem_free(ds, sizeof (dsl_dataset_t)); 490ea8dc4b6Seschrock if (err) { 491ea8dc4b6Seschrock dmu_buf_rele(dbuf, tag); 492ea8dc4b6Seschrock return (err); 493ea8dc4b6Seschrock } 494fa9e4066Sahrens ds = winner; 495fa9e4066Sahrens } else { 49691ebeef5Sahrens ds->ds_fsid_guid = 497fa9e4066Sahrens unique_insert(ds->ds_phys->ds_fsid_guid); 498fa9e4066Sahrens } 499fa9e4066Sahrens } 500fa9e4066Sahrens ASSERT3P(ds->ds_dbuf, ==, dbuf); 501fa9e4066Sahrens ASSERT3P(ds->ds_phys, ==, dbuf->db_data); 502088f3894Sahrens ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || 503afc6333aSahrens spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 50484db2a68Sahrens dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 505fa9e4066Sahrens mutex_enter(&ds->ds_lock); 506745cd3c5Smaybee if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) { 507fa9e4066Sahrens mutex_exit(&ds->ds_lock); 508745cd3c5Smaybee dmu_buf_rele(ds->ds_dbuf, tag); 509745cd3c5Smaybee return (ENOENT); 510fa9e4066Sahrens } 511fa9e4066Sahrens mutex_exit(&ds->ds_lock); 512ea8dc4b6Seschrock *dsp = ds; 513ea8dc4b6Seschrock return (0); 514fa9e4066Sahrens } 515fa9e4066Sahrens 516745cd3c5Smaybee static int 517745cd3c5Smaybee dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag) 518745cd3c5Smaybee { 519745cd3c5Smaybee dsl_pool_t *dp = ds->ds_dir->dd_pool; 520745cd3c5Smaybee 521745cd3c5Smaybee /* 522745cd3c5Smaybee * In syncing context we don't want the rwlock lock: there 523745cd3c5Smaybee * may be an existing writer waiting for sync phase to 524745cd3c5Smaybee * finish. We don't need to worry about such writers, since 525745cd3c5Smaybee * sync phase is single-threaded, so the writer can't be 526745cd3c5Smaybee * doing anything while we are active. 527745cd3c5Smaybee */ 528745cd3c5Smaybee if (dsl_pool_sync_context(dp)) { 529745cd3c5Smaybee ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 530745cd3c5Smaybee return (0); 531745cd3c5Smaybee } 532745cd3c5Smaybee 533745cd3c5Smaybee /* 534745cd3c5Smaybee * Normal users will hold the ds_rwlock as a READER until they 535745cd3c5Smaybee * are finished (i.e., call dsl_dataset_rele()). "Owners" will 536745cd3c5Smaybee * drop their READER lock after they set the ds_owner field. 537745cd3c5Smaybee * 538745cd3c5Smaybee * If the dataset is being destroyed, the destroy thread will 539745cd3c5Smaybee * obtain a WRITER lock for exclusive access after it's done its 540745cd3c5Smaybee * open-context work and then change the ds_owner to 541745cd3c5Smaybee * dsl_reaper once destruction is assured. So threads 542745cd3c5Smaybee * may block here temporarily, until the "destructability" of 543745cd3c5Smaybee * the dataset is determined. 544745cd3c5Smaybee */ 545745cd3c5Smaybee ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock)); 546745cd3c5Smaybee mutex_enter(&ds->ds_lock); 547745cd3c5Smaybee while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) { 548745cd3c5Smaybee rw_exit(&dp->dp_config_rwlock); 549745cd3c5Smaybee cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock); 550745cd3c5Smaybee if (DSL_DATASET_IS_DESTROYED(ds)) { 551745cd3c5Smaybee mutex_exit(&ds->ds_lock); 552745cd3c5Smaybee dsl_dataset_drop_ref(ds, tag); 553745cd3c5Smaybee rw_enter(&dp->dp_config_rwlock, RW_READER); 554745cd3c5Smaybee return (ENOENT); 555745cd3c5Smaybee } 556620252bcSChris Kirby /* 557620252bcSChris Kirby * The dp_config_rwlock lives above the ds_lock. And 558620252bcSChris Kirby * we need to check DSL_DATASET_IS_DESTROYED() while 559620252bcSChris Kirby * holding the ds_lock, so we have to drop and reacquire 560620252bcSChris Kirby * the ds_lock here. 561620252bcSChris Kirby */ 562620252bcSChris Kirby mutex_exit(&ds->ds_lock); 563745cd3c5Smaybee rw_enter(&dp->dp_config_rwlock, RW_READER); 564620252bcSChris Kirby mutex_enter(&ds->ds_lock); 565745cd3c5Smaybee } 566745cd3c5Smaybee mutex_exit(&ds->ds_lock); 567745cd3c5Smaybee return (0); 568745cd3c5Smaybee } 569745cd3c5Smaybee 570745cd3c5Smaybee int 571745cd3c5Smaybee dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 572745cd3c5Smaybee dsl_dataset_t **dsp) 573745cd3c5Smaybee { 574745cd3c5Smaybee int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp); 575745cd3c5Smaybee 576745cd3c5Smaybee if (err) 577745cd3c5Smaybee return (err); 578745cd3c5Smaybee return (dsl_dataset_hold_ref(*dsp, tag)); 579745cd3c5Smaybee } 580745cd3c5Smaybee 581745cd3c5Smaybee int 582503ad85cSMatthew Ahrens dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok, 583503ad85cSMatthew Ahrens void *tag, dsl_dataset_t **dsp) 584745cd3c5Smaybee { 585503ad85cSMatthew Ahrens int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 586745cd3c5Smaybee if (err) 587745cd3c5Smaybee return (err); 588503ad85cSMatthew Ahrens if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { 589503ad85cSMatthew Ahrens dsl_dataset_rele(*dsp, tag); 5904f5064b7SMark J Musante *dsp = NULL; 591745cd3c5Smaybee return (EBUSY); 592745cd3c5Smaybee } 593745cd3c5Smaybee return (0); 594745cd3c5Smaybee } 595745cd3c5Smaybee 596fa9e4066Sahrens int 597745cd3c5Smaybee dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp) 598fa9e4066Sahrens { 599fa9e4066Sahrens dsl_dir_t *dd; 600fa9e4066Sahrens dsl_pool_t *dp; 601745cd3c5Smaybee const char *snapname; 602fa9e4066Sahrens uint64_t obj; 603fa9e4066Sahrens int err = 0; 604fa9e4066Sahrens 605745cd3c5Smaybee err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname); 606ea8dc4b6Seschrock if (err) 607ea8dc4b6Seschrock return (err); 608fa9e4066Sahrens 609fa9e4066Sahrens dp = dd->dd_pool; 610fa9e4066Sahrens obj = dd->dd_phys->dd_head_dataset_obj; 611fa9e4066Sahrens rw_enter(&dp->dp_config_rwlock, RW_READER); 612745cd3c5Smaybee if (obj) 613745cd3c5Smaybee err = dsl_dataset_get_ref(dp, obj, tag, dsp); 614745cd3c5Smaybee else 615fa9e4066Sahrens err = ENOENT; 616745cd3c5Smaybee if (err) 617fa9e4066Sahrens goto out; 618fa9e4066Sahrens 619745cd3c5Smaybee err = dsl_dataset_hold_ref(*dsp, tag); 620fa9e4066Sahrens 621745cd3c5Smaybee /* we may be looking for a snapshot */ 622745cd3c5Smaybee if (err == 0 && snapname != NULL) { 623745cd3c5Smaybee dsl_dataset_t *ds = NULL; 624fa9e4066Sahrens 625745cd3c5Smaybee if (*snapname++ != '@') { 626745cd3c5Smaybee dsl_dataset_rele(*dsp, tag); 627fa9e4066Sahrens err = ENOENT; 628fa9e4066Sahrens goto out; 629fa9e4066Sahrens } 630fa9e4066Sahrens 631745cd3c5Smaybee dprintf("looking for snapshot '%s'\n", snapname); 632745cd3c5Smaybee err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 633745cd3c5Smaybee if (err == 0) 634745cd3c5Smaybee err = dsl_dataset_get_ref(dp, obj, tag, &ds); 635745cd3c5Smaybee dsl_dataset_rele(*dsp, tag); 636745cd3c5Smaybee 637745cd3c5Smaybee ASSERT3U((err == 0), ==, (ds != NULL)); 638745cd3c5Smaybee 639745cd3c5Smaybee if (ds) { 640745cd3c5Smaybee mutex_enter(&ds->ds_lock); 641745cd3c5Smaybee if (ds->ds_snapname[0] == 0) 642745cd3c5Smaybee (void) strlcpy(ds->ds_snapname, snapname, 643745cd3c5Smaybee sizeof (ds->ds_snapname)); 644745cd3c5Smaybee mutex_exit(&ds->ds_lock); 645745cd3c5Smaybee err = dsl_dataset_hold_ref(ds, tag); 646745cd3c5Smaybee *dsp = err ? NULL : ds; 647fa9e4066Sahrens } 648fa9e4066Sahrens } 649fa9e4066Sahrens out: 650fa9e4066Sahrens rw_exit(&dp->dp_config_rwlock); 651fa9e4066Sahrens dsl_dir_close(dd, FTAG); 652fa9e4066Sahrens return (err); 653fa9e4066Sahrens } 654fa9e4066Sahrens 655fa9e4066Sahrens int 656503ad85cSMatthew Ahrens dsl_dataset_own(const char *name, boolean_t inconsistentok, 657503ad85cSMatthew Ahrens void *tag, dsl_dataset_t **dsp) 658fa9e4066Sahrens { 659503ad85cSMatthew Ahrens int err = dsl_dataset_hold(name, tag, dsp); 660745cd3c5Smaybee if (err) 661745cd3c5Smaybee return (err); 662503ad85cSMatthew Ahrens if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { 663503ad85cSMatthew Ahrens dsl_dataset_rele(*dsp, tag); 664745cd3c5Smaybee return (EBUSY); 665745cd3c5Smaybee } 666745cd3c5Smaybee return (0); 667fa9e4066Sahrens } 668fa9e4066Sahrens 669fa9e4066Sahrens void 670fa9e4066Sahrens dsl_dataset_name(dsl_dataset_t *ds, char *name) 671fa9e4066Sahrens { 672fa9e4066Sahrens if (ds == NULL) { 673fa9e4066Sahrens (void) strcpy(name, "mos"); 674fa9e4066Sahrens } else { 675fa9e4066Sahrens dsl_dir_name(ds->ds_dir, name); 676ea8dc4b6Seschrock VERIFY(0 == dsl_dataset_get_snapname(ds)); 677fa9e4066Sahrens if (ds->ds_snapname[0]) { 678fa9e4066Sahrens (void) strcat(name, "@"); 679745cd3c5Smaybee /* 680745cd3c5Smaybee * We use a "recursive" mutex so that we 681745cd3c5Smaybee * can call dprintf_ds() with ds_lock held. 682745cd3c5Smaybee */ 683fa9e4066Sahrens if (!MUTEX_HELD(&ds->ds_lock)) { 684fa9e4066Sahrens mutex_enter(&ds->ds_lock); 685fa9e4066Sahrens (void) strcat(name, ds->ds_snapname); 686fa9e4066Sahrens mutex_exit(&ds->ds_lock); 687fa9e4066Sahrens } else { 688fa9e4066Sahrens (void) strcat(name, ds->ds_snapname); 689fa9e4066Sahrens } 690fa9e4066Sahrens } 691fa9e4066Sahrens } 692fa9e4066Sahrens } 693fa9e4066Sahrens 694b7661cccSmmusante static int 695b7661cccSmmusante dsl_dataset_namelen(dsl_dataset_t *ds) 696b7661cccSmmusante { 697b7661cccSmmusante int result; 698b7661cccSmmusante 699b7661cccSmmusante if (ds == NULL) { 700b7661cccSmmusante result = 3; /* "mos" */ 701b7661cccSmmusante } else { 702b7661cccSmmusante result = dsl_dir_namelen(ds->ds_dir); 703b7661cccSmmusante VERIFY(0 == dsl_dataset_get_snapname(ds)); 704b7661cccSmmusante if (ds->ds_snapname[0]) { 705b7661cccSmmusante ++result; /* adding one for the @-sign */ 706b7661cccSmmusante if (!MUTEX_HELD(&ds->ds_lock)) { 707b7661cccSmmusante mutex_enter(&ds->ds_lock); 708b7661cccSmmusante result += strlen(ds->ds_snapname); 709b7661cccSmmusante mutex_exit(&ds->ds_lock); 710b7661cccSmmusante } else { 711b7661cccSmmusante result += strlen(ds->ds_snapname); 712b7661cccSmmusante } 713b7661cccSmmusante } 714b7661cccSmmusante } 715b7661cccSmmusante 716b7661cccSmmusante return (result); 717b7661cccSmmusante } 718b7661cccSmmusante 719088f3894Sahrens void 720745cd3c5Smaybee dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag) 721fa9e4066Sahrens { 722ea8dc4b6Seschrock dmu_buf_rele(ds->ds_dbuf, tag); 723fa9e4066Sahrens } 724fa9e4066Sahrens 7253cb34c60Sahrens void 726745cd3c5Smaybee dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 7273cb34c60Sahrens { 728745cd3c5Smaybee if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) { 729745cd3c5Smaybee rw_exit(&ds->ds_rwlock); 730745cd3c5Smaybee } 731745cd3c5Smaybee dsl_dataset_drop_ref(ds, tag); 732745cd3c5Smaybee } 733745cd3c5Smaybee 734745cd3c5Smaybee void 735503ad85cSMatthew Ahrens dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 736745cd3c5Smaybee { 737503ad85cSMatthew Ahrens ASSERT((ds->ds_owner == tag && ds->ds_dbuf) || 738745cd3c5Smaybee (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL)); 739745cd3c5Smaybee 7403cb34c60Sahrens mutex_enter(&ds->ds_lock); 741745cd3c5Smaybee ds->ds_owner = NULL; 742745cd3c5Smaybee if (RW_WRITE_HELD(&ds->ds_rwlock)) { 743745cd3c5Smaybee rw_exit(&ds->ds_rwlock); 744745cd3c5Smaybee cv_broadcast(&ds->ds_exclusive_cv); 745745cd3c5Smaybee } 7463cb34c60Sahrens mutex_exit(&ds->ds_lock); 747745cd3c5Smaybee if (ds->ds_dbuf) 748503ad85cSMatthew Ahrens dsl_dataset_drop_ref(ds, tag); 749745cd3c5Smaybee else 750cde58dbcSMatthew Ahrens dsl_dataset_evict(NULL, ds); 7513cb34c60Sahrens } 7523cb34c60Sahrens 7533cb34c60Sahrens boolean_t 754503ad85cSMatthew Ahrens dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag) 7553cb34c60Sahrens { 756745cd3c5Smaybee boolean_t gotit = FALSE; 757745cd3c5Smaybee 7583cb34c60Sahrens mutex_enter(&ds->ds_lock); 759745cd3c5Smaybee if (ds->ds_owner == NULL && 760745cd3c5Smaybee (!DS_IS_INCONSISTENT(ds) || inconsistentok)) { 761503ad85cSMatthew Ahrens ds->ds_owner = tag; 762745cd3c5Smaybee if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) 763745cd3c5Smaybee rw_exit(&ds->ds_rwlock); 764745cd3c5Smaybee gotit = TRUE; 7653cb34c60Sahrens } 7663cb34c60Sahrens mutex_exit(&ds->ds_lock); 767745cd3c5Smaybee return (gotit); 768745cd3c5Smaybee } 769745cd3c5Smaybee 770745cd3c5Smaybee void 771745cd3c5Smaybee dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner) 772745cd3c5Smaybee { 773745cd3c5Smaybee ASSERT3P(owner, ==, ds->ds_owner); 774745cd3c5Smaybee if (!RW_WRITE_HELD(&ds->ds_rwlock)) 775745cd3c5Smaybee rw_enter(&ds->ds_rwlock, RW_WRITER); 7763cb34c60Sahrens } 7773cb34c60Sahrens 7781d452cf5Sahrens uint64_t 779088f3894Sahrens dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 780ab04eb8eStimh uint64_t flags, dmu_tx_t *tx) 781fa9e4066Sahrens { 7823cb34c60Sahrens dsl_pool_t *dp = dd->dd_pool; 783fa9e4066Sahrens dmu_buf_t *dbuf; 784fa9e4066Sahrens dsl_dataset_phys_t *dsphys; 7853cb34c60Sahrens uint64_t dsobj; 786fa9e4066Sahrens objset_t *mos = dp->dp_meta_objset; 787fa9e4066Sahrens 788088f3894Sahrens if (origin == NULL) 789088f3894Sahrens origin = dp->dp_origin_snap; 790088f3894Sahrens 7913cb34c60Sahrens ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 7923cb34c60Sahrens ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); 793fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 7943cb34c60Sahrens ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 795fa9e4066Sahrens 7961649cd4bStabriz dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 7971649cd4bStabriz DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 798ea8dc4b6Seschrock VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 799fa9e4066Sahrens dmu_buf_will_dirty(dbuf, tx); 800fa9e4066Sahrens dsphys = dbuf->db_data; 801745cd3c5Smaybee bzero(dsphys, sizeof (dsl_dataset_phys_t)); 802fa9e4066Sahrens dsphys->ds_dir_obj = dd->dd_object; 803ab04eb8eStimh dsphys->ds_flags = flags; 804fa9e4066Sahrens dsphys->ds_fsid_guid = unique_create(); 805fa9e4066Sahrens (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 806fa9e4066Sahrens sizeof (dsphys->ds_guid)); 807fa9e4066Sahrens dsphys->ds_snapnames_zapobj = 808ab04eb8eStimh zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 809ab04eb8eStimh DMU_OT_NONE, 0, tx); 810fa9e4066Sahrens dsphys->ds_creation_time = gethrestime_sec(); 811088f3894Sahrens dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 812a9799022Sck 813cde58dbcSMatthew Ahrens if (origin == NULL) { 814cde58dbcSMatthew Ahrens dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 815cde58dbcSMatthew Ahrens } else { 816cde58dbcSMatthew Ahrens dsl_dataset_t *ohds; 817cde58dbcSMatthew Ahrens 8183cb34c60Sahrens dsphys->ds_prev_snap_obj = origin->ds_object; 819fa9e4066Sahrens dsphys->ds_prev_snap_txg = 8203cb34c60Sahrens origin->ds_phys->ds_creation_txg; 821fa9e4066Sahrens dsphys->ds_used_bytes = 8223cb34c60Sahrens origin->ds_phys->ds_used_bytes; 823fa9e4066Sahrens dsphys->ds_compressed_bytes = 8243cb34c60Sahrens origin->ds_phys->ds_compressed_bytes; 825fa9e4066Sahrens dsphys->ds_uncompressed_bytes = 8263cb34c60Sahrens origin->ds_phys->ds_uncompressed_bytes; 8273cb34c60Sahrens dsphys->ds_bp = origin->ds_phys->ds_bp; 828579ae4d5Stimh dsphys->ds_flags |= origin->ds_phys->ds_flags; 829fa9e4066Sahrens 8303cb34c60Sahrens dmu_buf_will_dirty(origin->ds_dbuf, tx); 8313cb34c60Sahrens origin->ds_phys->ds_num_children++; 832fa9e4066Sahrens 833cde58dbcSMatthew Ahrens VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, 834cde58dbcSMatthew Ahrens origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds)); 835cde58dbcSMatthew Ahrens dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 836cde58dbcSMatthew Ahrens dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 837cde58dbcSMatthew Ahrens dsl_dataset_rele(ohds, FTAG); 838cde58dbcSMatthew Ahrens 839088f3894Sahrens if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 840088f3894Sahrens if (origin->ds_phys->ds_next_clones_obj == 0) { 841088f3894Sahrens origin->ds_phys->ds_next_clones_obj = 842088f3894Sahrens zap_create(mos, 843088f3894Sahrens DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 844088f3894Sahrens } 845088f3894Sahrens VERIFY(0 == zap_add_int(mos, 846088f3894Sahrens origin->ds_phys->ds_next_clones_obj, 847088f3894Sahrens dsobj, tx)); 848088f3894Sahrens } 849088f3894Sahrens 850fa9e4066Sahrens dmu_buf_will_dirty(dd->dd_dbuf, tx); 8513cb34c60Sahrens dd->dd_phys->dd_origin_obj = origin->ds_object; 852cde58dbcSMatthew Ahrens if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 853cde58dbcSMatthew Ahrens if (origin->ds_dir->dd_phys->dd_clones == 0) { 854cde58dbcSMatthew Ahrens dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 855cde58dbcSMatthew Ahrens origin->ds_dir->dd_phys->dd_clones = 856cde58dbcSMatthew Ahrens zap_create(mos, 857cde58dbcSMatthew Ahrens DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 858cde58dbcSMatthew Ahrens } 859cde58dbcSMatthew Ahrens VERIFY3U(0, ==, zap_add_int(mos, 860cde58dbcSMatthew Ahrens origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); 861cde58dbcSMatthew Ahrens } 862fa9e4066Sahrens } 863ab04eb8eStimh 864ab04eb8eStimh if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 865ab04eb8eStimh dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 866ab04eb8eStimh 867ea8dc4b6Seschrock dmu_buf_rele(dbuf, FTAG); 868fa9e4066Sahrens 869fa9e4066Sahrens dmu_buf_will_dirty(dd->dd_dbuf, tx); 870fa9e4066Sahrens dd->dd_phys->dd_head_dataset_obj = dsobj; 8713cb34c60Sahrens 8723cb34c60Sahrens return (dsobj); 8733cb34c60Sahrens } 8743cb34c60Sahrens 8753cb34c60Sahrens uint64_t 876ab04eb8eStimh dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 877ab04eb8eStimh dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 8783cb34c60Sahrens { 8793cb34c60Sahrens dsl_pool_t *dp = pdd->dd_pool; 8803cb34c60Sahrens uint64_t dsobj, ddobj; 8813cb34c60Sahrens dsl_dir_t *dd; 8823cb34c60Sahrens 8833cb34c60Sahrens ASSERT(lastname[0] != '@'); 8843cb34c60Sahrens 885088f3894Sahrens ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 8863cb34c60Sahrens VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd)); 8873cb34c60Sahrens 888088f3894Sahrens dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx); 8893cb34c60Sahrens 8903cb34c60Sahrens dsl_deleg_set_create_perms(dd, tx, cr); 8913cb34c60Sahrens 892fa9e4066Sahrens dsl_dir_close(dd, FTAG); 893fa9e4066Sahrens 894feaa74e4SMark Maybee /* 895feaa74e4SMark Maybee * If we are creating a clone, make sure we zero out any stale 896feaa74e4SMark Maybee * data from the origin snapshots zil header. 897feaa74e4SMark Maybee */ 898feaa74e4SMark Maybee if (origin != NULL) { 899feaa74e4SMark Maybee dsl_dataset_t *ds; 900feaa74e4SMark Maybee objset_t *os; 901feaa74e4SMark Maybee 902feaa74e4SMark Maybee VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 903feaa74e4SMark Maybee VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os)); 904feaa74e4SMark Maybee bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 905feaa74e4SMark Maybee dsl_dataset_dirty(ds, tx); 906feaa74e4SMark Maybee dsl_dataset_rele(ds, FTAG); 907feaa74e4SMark Maybee } 908feaa74e4SMark Maybee 9091d452cf5Sahrens return (dsobj); 910fa9e4066Sahrens } 911fa9e4066Sahrens 9121d452cf5Sahrens /* 91319b94df9SMatthew Ahrens * The snapshots must all be in the same pool. 9141d452cf5Sahrens */ 9151d452cf5Sahrens int 91619b94df9SMatthew Ahrens dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer, char *failed) 9171d452cf5Sahrens { 9181d452cf5Sahrens int err; 9191d452cf5Sahrens dsl_sync_task_t *dst; 9201d452cf5Sahrens spa_t *spa; 92119b94df9SMatthew Ahrens nvpair_t *pair; 92219b94df9SMatthew Ahrens dsl_sync_task_group_t *dstg; 923e5351341SMatthew Ahrens 92419b94df9SMatthew Ahrens pair = nvlist_next_nvpair(snaps, NULL); 92519b94df9SMatthew Ahrens if (pair == NULL) 92619b94df9SMatthew Ahrens return (0); 92719b94df9SMatthew Ahrens 92819b94df9SMatthew Ahrens err = spa_open(nvpair_name(pair), &spa, FTAG); 9291d452cf5Sahrens if (err) 9301d452cf5Sahrens return (err); 93119b94df9SMatthew Ahrens dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 9321d452cf5Sahrens 93319b94df9SMatthew Ahrens for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 93419b94df9SMatthew Ahrens pair = nvlist_next_nvpair(snaps, pair)) { 93519b94df9SMatthew Ahrens dsl_dataset_t *ds; 93619b94df9SMatthew Ahrens int err; 93719b94df9SMatthew Ahrens 93819b94df9SMatthew Ahrens err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds); 93919b94df9SMatthew Ahrens if (err == 0) { 94019b94df9SMatthew Ahrens struct dsl_ds_destroyarg *dsda; 94119b94df9SMatthew Ahrens 94219b94df9SMatthew Ahrens dsl_dataset_make_exclusive(ds, dstg); 94319b94df9SMatthew Ahrens dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), 94419b94df9SMatthew Ahrens KM_SLEEP); 94519b94df9SMatthew Ahrens dsda->ds = ds; 94619b94df9SMatthew Ahrens dsda->defer = defer; 94719b94df9SMatthew Ahrens dsl_sync_task_create(dstg, dsl_dataset_destroy_check, 94819b94df9SMatthew Ahrens dsl_dataset_destroy_sync, dsda, dstg, 0); 94919b94df9SMatthew Ahrens } else if (err == ENOENT) { 95019b94df9SMatthew Ahrens err = 0; 95119b94df9SMatthew Ahrens } else { 95219b94df9SMatthew Ahrens (void) strcpy(failed, nvpair_name(pair)); 95319b94df9SMatthew Ahrens break; 95419b94df9SMatthew Ahrens } 95519b94df9SMatthew Ahrens } 9561d452cf5Sahrens 9571d452cf5Sahrens if (err == 0) 95819b94df9SMatthew Ahrens err = dsl_sync_task_group_wait(dstg); 9591d452cf5Sahrens 96019b94df9SMatthew Ahrens for (dst = list_head(&dstg->dstg_tasks); dst; 96119b94df9SMatthew Ahrens dst = list_next(&dstg->dstg_tasks, dst)) { 962842727c2SChris Kirby struct dsl_ds_destroyarg *dsda = dst->dst_arg1; 963842727c2SChris Kirby dsl_dataset_t *ds = dsda->ds; 964842727c2SChris Kirby 965745cd3c5Smaybee /* 966745cd3c5Smaybee * Return the file system name that triggered the error 967745cd3c5Smaybee */ 9681d452cf5Sahrens if (dst->dst_err) { 96919b94df9SMatthew Ahrens dsl_dataset_name(ds, failed); 970e1930233Sbonwick } 971842727c2SChris Kirby ASSERT3P(dsda->rm_origin, ==, NULL); 97219b94df9SMatthew Ahrens dsl_dataset_disown(ds, dstg); 973842727c2SChris Kirby kmem_free(dsda, sizeof (struct dsl_ds_destroyarg)); 974fa9e4066Sahrens } 975fa9e4066Sahrens 97619b94df9SMatthew Ahrens dsl_sync_task_group_destroy(dstg); 9771d452cf5Sahrens spa_close(spa, FTAG); 978fa9e4066Sahrens return (err); 97919b94df9SMatthew Ahrens 980fa9e4066Sahrens } 981fa9e4066Sahrens 982842727c2SChris Kirby static boolean_t 983842727c2SChris Kirby dsl_dataset_might_destroy_origin(dsl_dataset_t *ds) 984842727c2SChris Kirby { 985842727c2SChris Kirby boolean_t might_destroy = B_FALSE; 986842727c2SChris Kirby 987842727c2SChris Kirby mutex_enter(&ds->ds_lock); 988842727c2SChris Kirby if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 && 989842727c2SChris Kirby DS_IS_DEFER_DESTROY(ds)) 990842727c2SChris Kirby might_destroy = B_TRUE; 991842727c2SChris Kirby mutex_exit(&ds->ds_lock); 992842727c2SChris Kirby 993842727c2SChris Kirby return (might_destroy); 994842727c2SChris Kirby } 995842727c2SChris Kirby 996842727c2SChris Kirby /* 997842727c2SChris Kirby * If we're removing a clone, and these three conditions are true: 998842727c2SChris Kirby * 1) the clone's origin has no other children 999842727c2SChris Kirby * 2) the clone's origin has no user references 1000842727c2SChris Kirby * 3) the clone's origin has been marked for deferred destruction 1001842727c2SChris Kirby * Then, prepare to remove the origin as part of this sync task group. 1002842727c2SChris Kirby */ 1003842727c2SChris Kirby static int 1004842727c2SChris Kirby dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag) 1005842727c2SChris Kirby { 1006842727c2SChris Kirby dsl_dataset_t *ds = dsda->ds; 1007842727c2SChris Kirby dsl_dataset_t *origin = ds->ds_prev; 1008842727c2SChris Kirby 1009842727c2SChris Kirby if (dsl_dataset_might_destroy_origin(origin)) { 1010842727c2SChris Kirby char *name; 1011842727c2SChris Kirby int namelen; 1012842727c2SChris Kirby int error; 1013842727c2SChris Kirby 1014842727c2SChris Kirby namelen = dsl_dataset_namelen(origin) + 1; 1015842727c2SChris Kirby name = kmem_alloc(namelen, KM_SLEEP); 1016842727c2SChris Kirby dsl_dataset_name(origin, name); 1017842727c2SChris Kirby #ifdef _KERNEL 1018842727c2SChris Kirby error = zfs_unmount_snap(name, NULL); 1019842727c2SChris Kirby if (error) { 1020842727c2SChris Kirby kmem_free(name, namelen); 1021842727c2SChris Kirby return (error); 1022842727c2SChris Kirby } 1023842727c2SChris Kirby #endif 1024503ad85cSMatthew Ahrens error = dsl_dataset_own(name, B_TRUE, tag, &origin); 1025842727c2SChris Kirby kmem_free(name, namelen); 1026842727c2SChris Kirby if (error) 1027842727c2SChris Kirby return (error); 1028842727c2SChris Kirby dsda->rm_origin = origin; 1029842727c2SChris Kirby dsl_dataset_make_exclusive(origin, tag); 1030842727c2SChris Kirby } 1031842727c2SChris Kirby 1032842727c2SChris Kirby return (0); 1033842727c2SChris Kirby } 1034842727c2SChris Kirby 10353cb34c60Sahrens /* 1036745cd3c5Smaybee * ds must be opened as OWNER. On return (whether successful or not), 1037745cd3c5Smaybee * ds will be closed and caller can no longer dereference it. 10383cb34c60Sahrens */ 1039fa9e4066Sahrens int 1040842727c2SChris Kirby dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer) 1041fa9e4066Sahrens { 1042fa9e4066Sahrens int err; 10431d452cf5Sahrens dsl_sync_task_group_t *dstg; 10441d452cf5Sahrens objset_t *os; 1045fa9e4066Sahrens dsl_dir_t *dd; 10461d452cf5Sahrens uint64_t obj; 104792241e0bSTom Erickson struct dsl_ds_destroyarg dsda = { 0 }; 104892241e0bSTom Erickson dsl_dataset_t dummy_ds = { 0 }; 1049842727c2SChris Kirby 1050842727c2SChris Kirby dsda.ds = ds; 10511d452cf5Sahrens 10523cb34c60Sahrens if (dsl_dataset_is_snapshot(ds)) { 10531d452cf5Sahrens /* Destroying a snapshot is simpler */ 1054745cd3c5Smaybee dsl_dataset_make_exclusive(ds, tag); 10553baa08fcSek 1056842727c2SChris Kirby dsda.defer = defer; 10571d452cf5Sahrens err = dsl_sync_task_do(ds->ds_dir->dd_pool, 10581d452cf5Sahrens dsl_dataset_destroy_check, dsl_dataset_destroy_sync, 1059842727c2SChris Kirby &dsda, tag, 0); 1060842727c2SChris Kirby ASSERT3P(dsda.rm_origin, ==, NULL); 10613cb34c60Sahrens goto out; 1062922d9a97SChris Kirby } else if (defer) { 1063922d9a97SChris Kirby err = EINVAL; 1064922d9a97SChris Kirby goto out; 10651d452cf5Sahrens } 1066fa9e4066Sahrens 10671d452cf5Sahrens dd = ds->ds_dir; 106892241e0bSTom Erickson dummy_ds.ds_dir = dd; 106992241e0bSTom Erickson dummy_ds.ds_object = ds->ds_object; 1070fa9e4066Sahrens 10711d452cf5Sahrens /* 10721d452cf5Sahrens * Check for errors and mark this ds as inconsistent, in 10731d452cf5Sahrens * case we crash while freeing the objects. 10741d452cf5Sahrens */ 10751d452cf5Sahrens err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check, 10761d452cf5Sahrens dsl_dataset_destroy_begin_sync, ds, NULL, 0); 10773cb34c60Sahrens if (err) 10783cb34c60Sahrens goto out; 10793cb34c60Sahrens 1080503ad85cSMatthew Ahrens err = dmu_objset_from_ds(ds, &os); 10813cb34c60Sahrens if (err) 10823cb34c60Sahrens goto out; 1083fa9e4066Sahrens 10841d452cf5Sahrens /* 10851d452cf5Sahrens * remove the objects in open context, so that we won't 10861d452cf5Sahrens * have too much to do in syncing context. 10871d452cf5Sahrens */ 10886754306eSahrens for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 10896754306eSahrens ds->ds_phys->ds_prev_snap_txg)) { 1090cdb0ab79Smaybee /* 1091cdb0ab79Smaybee * Ignore errors, if there is not enough disk space 1092cdb0ab79Smaybee * we will deal with it in dsl_dataset_destroy_sync(). 1093cdb0ab79Smaybee */ 1094cdb0ab79Smaybee (void) dmu_free_object(os, obj); 10951d452cf5Sahrens } 1096feaa74e4SMark Maybee if (err != ESRCH) 1097feaa74e4SMark Maybee goto out; 10981d452cf5Sahrens 109914843421SMatthew Ahrens /* 1100feaa74e4SMark Maybee * Only the ZIL knows how to free log blocks. 1101feaa74e4SMark Maybee */ 1102feaa74e4SMark Maybee zil_destroy(dmu_objset_zil(os), B_FALSE); 1103feaa74e4SMark Maybee 1104feaa74e4SMark Maybee /* 1105feaa74e4SMark Maybee * Sync out all in-flight IO. 110614843421SMatthew Ahrens */ 110714843421SMatthew Ahrens txg_wait_synced(dd->dd_pool, 0); 110814843421SMatthew Ahrens 110914843421SMatthew Ahrens /* 111014843421SMatthew Ahrens * If we managed to free all the objects in open 111114843421SMatthew Ahrens * context, the user space accounting should be zero. 111214843421SMatthew Ahrens */ 111314843421SMatthew Ahrens if (ds->ds_phys->ds_bp.blk_fill == 0 && 1114503ad85cSMatthew Ahrens dmu_objset_userused_enabled(os)) { 111514843421SMatthew Ahrens uint64_t count; 111614843421SMatthew Ahrens 111714843421SMatthew Ahrens ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 || 111814843421SMatthew Ahrens count == 0); 111914843421SMatthew Ahrens ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 || 112014843421SMatthew Ahrens count == 0); 112114843421SMatthew Ahrens } 112214843421SMatthew Ahrens 112368038c2cSmaybee rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); 112468038c2cSmaybee err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd); 112568038c2cSmaybee rw_exit(&dd->dd_pool->dp_config_rwlock); 112668038c2cSmaybee 112768038c2cSmaybee if (err) 112868038c2cSmaybee goto out; 112968038c2cSmaybee 11301d452cf5Sahrens /* 11311d452cf5Sahrens * Blow away the dsl_dir + head dataset. 11321d452cf5Sahrens */ 1133745cd3c5Smaybee dsl_dataset_make_exclusive(ds, tag); 1134842727c2SChris Kirby /* 1135842727c2SChris Kirby * If we're removing a clone, we might also need to remove its 1136842727c2SChris Kirby * origin. 1137842727c2SChris Kirby */ 1138842727c2SChris Kirby do { 1139842727c2SChris Kirby dsda.need_prep = B_FALSE; 1140842727c2SChris Kirby if (dsl_dir_is_clone(dd)) { 1141842727c2SChris Kirby err = dsl_dataset_origin_rm_prep(&dsda, tag); 1142842727c2SChris Kirby if (err) { 1143842727c2SChris Kirby dsl_dir_close(dd, FTAG); 1144842727c2SChris Kirby goto out; 1145842727c2SChris Kirby } 1146842727c2SChris Kirby } 1147842727c2SChris Kirby 1148842727c2SChris Kirby dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); 1149842727c2SChris Kirby dsl_sync_task_create(dstg, dsl_dataset_destroy_check, 1150842727c2SChris Kirby dsl_dataset_destroy_sync, &dsda, tag, 0); 1151842727c2SChris Kirby dsl_sync_task_create(dstg, dsl_dir_destroy_check, 115292241e0bSTom Erickson dsl_dir_destroy_sync, &dummy_ds, FTAG, 0); 1153842727c2SChris Kirby err = dsl_sync_task_group_wait(dstg); 1154842727c2SChris Kirby dsl_sync_task_group_destroy(dstg); 1155842727c2SChris Kirby 1156842727c2SChris Kirby /* 1157842727c2SChris Kirby * We could be racing against 'zfs release' or 'zfs destroy -d' 1158842727c2SChris Kirby * on the origin snap, in which case we can get EBUSY if we 1159842727c2SChris Kirby * needed to destroy the origin snap but were not ready to 1160842727c2SChris Kirby * do so. 1161842727c2SChris Kirby */ 1162842727c2SChris Kirby if (dsda.need_prep) { 1163842727c2SChris Kirby ASSERT(err == EBUSY); 1164842727c2SChris Kirby ASSERT(dsl_dir_is_clone(dd)); 1165842727c2SChris Kirby ASSERT(dsda.rm_origin == NULL); 1166842727c2SChris Kirby } 1167842727c2SChris Kirby } while (dsda.need_prep); 1168842727c2SChris Kirby 1169842727c2SChris Kirby if (dsda.rm_origin != NULL) 1170842727c2SChris Kirby dsl_dataset_disown(dsda.rm_origin, tag); 1171842727c2SChris Kirby 1172745cd3c5Smaybee /* if it is successful, dsl_dir_destroy_sync will close the dd */ 11733cb34c60Sahrens if (err) 11741d452cf5Sahrens dsl_dir_close(dd, FTAG); 11753cb34c60Sahrens out: 1176745cd3c5Smaybee dsl_dataset_disown(ds, tag); 1177fa9e4066Sahrens return (err); 1178fa9e4066Sahrens } 1179fa9e4066Sahrens 1180c717a561Smaybee blkptr_t * 1181c717a561Smaybee dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1182fa9e4066Sahrens { 1183c717a561Smaybee return (&ds->ds_phys->ds_bp); 1184fa9e4066Sahrens } 1185fa9e4066Sahrens 1186fa9e4066Sahrens void 1187fa9e4066Sahrens dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 1188fa9e4066Sahrens { 1189fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 1190fa9e4066Sahrens /* If it's the meta-objset, set dp_meta_rootbp */ 1191fa9e4066Sahrens if (ds == NULL) { 1192fa9e4066Sahrens tx->tx_pool->dp_meta_rootbp = *bp; 1193fa9e4066Sahrens } else { 1194fa9e4066Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 1195fa9e4066Sahrens ds->ds_phys->ds_bp = *bp; 1196fa9e4066Sahrens } 1197fa9e4066Sahrens } 1198fa9e4066Sahrens 1199fa9e4066Sahrens spa_t * 1200fa9e4066Sahrens dsl_dataset_get_spa(dsl_dataset_t *ds) 1201fa9e4066Sahrens { 1202fa9e4066Sahrens return (ds->ds_dir->dd_pool->dp_spa); 1203fa9e4066Sahrens } 1204fa9e4066Sahrens 1205fa9e4066Sahrens void 1206fa9e4066Sahrens dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1207fa9e4066Sahrens { 1208fa9e4066Sahrens dsl_pool_t *dp; 1209fa9e4066Sahrens 1210fa9e4066Sahrens if (ds == NULL) /* this is the meta-objset */ 1211fa9e4066Sahrens return; 1212fa9e4066Sahrens 1213503ad85cSMatthew Ahrens ASSERT(ds->ds_objset != NULL); 1214a2eea2e1Sahrens 1215a2eea2e1Sahrens if (ds->ds_phys->ds_next_snap_obj != 0) 1216a2eea2e1Sahrens panic("dirtying snapshot!"); 1217fa9e4066Sahrens 1218fa9e4066Sahrens dp = ds->ds_dir->dd_pool; 1219fa9e4066Sahrens 1220fa9e4066Sahrens if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { 1221fa9e4066Sahrens /* up the hold count until we can be written out */ 1222fa9e4066Sahrens dmu_buf_add_ref(ds->ds_dbuf, ds); 1223fa9e4066Sahrens } 1224fa9e4066Sahrens } 1225fa9e4066Sahrens 1226a9799022Sck /* 1227a9799022Sck * The unique space in the head dataset can be calculated by subtracting 1228a9799022Sck * the space used in the most recent snapshot, that is still being used 1229a9799022Sck * in this file system, from the space currently in use. To figure out 1230a9799022Sck * the space in the most recent snapshot still in use, we need to take 1231a9799022Sck * the total space used in the snapshot and subtract out the space that 1232a9799022Sck * has been freed up since the snapshot was taken. 1233a9799022Sck */ 1234a9799022Sck static void 1235a9799022Sck dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1236a9799022Sck { 1237a9799022Sck uint64_t mrs_used; 1238a9799022Sck uint64_t dlused, dlcomp, dluncomp; 1239a9799022Sck 12403f9d6ad7SLin Ling ASSERT(!dsl_dataset_is_snapshot(ds)); 1241a9799022Sck 1242a9799022Sck if (ds->ds_phys->ds_prev_snap_obj != 0) 1243a9799022Sck mrs_used = ds->ds_prev->ds_phys->ds_used_bytes; 1244a9799022Sck else 1245a9799022Sck mrs_used = 0; 1246a9799022Sck 1247cde58dbcSMatthew Ahrens dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 1248a9799022Sck 1249a9799022Sck ASSERT3U(dlused, <=, mrs_used); 1250a9799022Sck ds->ds_phys->ds_unique_bytes = 1251a9799022Sck ds->ds_phys->ds_used_bytes - (mrs_used - dlused); 1252a9799022Sck 12533f9d6ad7SLin Ling if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1254a9799022Sck SPA_VERSION_UNIQUE_ACCURATE) 1255a9799022Sck ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1256a9799022Sck } 1257a9799022Sck 1258fa9e4066Sahrens struct killarg { 125974e7dc98SMatthew Ahrens dsl_dataset_t *ds; 1260fa9e4066Sahrens dmu_tx_t *tx; 1261fa9e4066Sahrens }; 1262fa9e4066Sahrens 126374e7dc98SMatthew Ahrens /* ARGSUSED */ 1264fa9e4066Sahrens static int 12653f9d6ad7SLin Ling kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf, 1266b24ab676SJeff Bonwick const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1267fa9e4066Sahrens { 1268fa9e4066Sahrens struct killarg *ka = arg; 1269b24ab676SJeff Bonwick dmu_tx_t *tx = ka->tx; 1270fa9e4066Sahrens 127188b7b0f2SMatthew Ahrens if (bp == NULL) 127288b7b0f2SMatthew Ahrens return (0); 1273fa9e4066Sahrens 1274b24ab676SJeff Bonwick if (zb->zb_level == ZB_ZIL_LEVEL) { 1275b24ab676SJeff Bonwick ASSERT(zilog != NULL); 1276ab69d62fSMatthew Ahrens /* 1277ab69d62fSMatthew Ahrens * It's a block in the intent log. It has no 1278ab69d62fSMatthew Ahrens * accounting, so just free it. 1279ab69d62fSMatthew Ahrens */ 1280b24ab676SJeff Bonwick dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp); 1281ab69d62fSMatthew Ahrens } else { 1282b24ab676SJeff Bonwick ASSERT(zilog == NULL); 1283ab69d62fSMatthew Ahrens ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg); 1284b24ab676SJeff Bonwick (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE); 1285ab69d62fSMatthew Ahrens } 128674e7dc98SMatthew Ahrens 1287fa9e4066Sahrens return (0); 1288fa9e4066Sahrens } 1289fa9e4066Sahrens 1290e1930233Sbonwick /* ARGSUSED */ 1291e1930233Sbonwick static int 12921d452cf5Sahrens dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx) 1293e1930233Sbonwick { 12941d452cf5Sahrens dsl_dataset_t *ds = arg1; 12953cb34c60Sahrens objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 12963cb34c60Sahrens uint64_t count; 12973cb34c60Sahrens int err; 1298e1930233Sbonwick 1299e1930233Sbonwick /* 1300e1930233Sbonwick * Can't delete a head dataset if there are snapshots of it. 1301e1930233Sbonwick * (Except if the only snapshots are from the branch we cloned 1302e1930233Sbonwick * from.) 1303e1930233Sbonwick */ 1304e1930233Sbonwick if (ds->ds_prev != NULL && 1305e1930233Sbonwick ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 13064aed9999Svitezslav batrla - Sun Microsystems - Prague Czech Republic return (EBUSY); 1307e1930233Sbonwick 13083cb34c60Sahrens /* 13093cb34c60Sahrens * This is really a dsl_dir thing, but check it here so that 13103cb34c60Sahrens * we'll be less likely to leave this dataset inconsistent & 13113cb34c60Sahrens * nearly destroyed. 13123cb34c60Sahrens */ 13133cb34c60Sahrens err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count); 13143cb34c60Sahrens if (err) 13153cb34c60Sahrens return (err); 13163cb34c60Sahrens if (count != 0) 13173cb34c60Sahrens return (EEXIST); 13183cb34c60Sahrens 1319e1930233Sbonwick return (0); 1320e1930233Sbonwick } 1321e1930233Sbonwick 13221d452cf5Sahrens /* ARGSUSED */ 13231d452cf5Sahrens static void 13243f9d6ad7SLin Ling dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx) 1325fa9e4066Sahrens { 13261d452cf5Sahrens dsl_dataset_t *ds = arg1; 1327ecd6cf80Smarks dsl_pool_t *dp = ds->ds_dir->dd_pool; 1328fa9e4066Sahrens 13291d452cf5Sahrens /* Mark it as inconsistent on-disk, in case we crash */ 13301d452cf5Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 13311d452cf5Sahrens ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; 1332ecd6cf80Smarks 13333f9d6ad7SLin Ling spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx, 13343f9d6ad7SLin Ling "dataset = %llu", ds->ds_object); 13351d452cf5Sahrens } 1336fa9e4066Sahrens 1337842727c2SChris Kirby static int 1338842727c2SChris Kirby dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag, 1339842727c2SChris Kirby dmu_tx_t *tx) 1340842727c2SChris Kirby { 1341842727c2SChris Kirby dsl_dataset_t *ds = dsda->ds; 1342842727c2SChris Kirby dsl_dataset_t *ds_prev = ds->ds_prev; 1343842727c2SChris Kirby 1344842727c2SChris Kirby if (dsl_dataset_might_destroy_origin(ds_prev)) { 1345842727c2SChris Kirby struct dsl_ds_destroyarg ndsda = {0}; 1346842727c2SChris Kirby 1347842727c2SChris Kirby /* 1348842727c2SChris Kirby * If we're not prepared to remove the origin, don't remove 1349842727c2SChris Kirby * the clone either. 1350842727c2SChris Kirby */ 1351842727c2SChris Kirby if (dsda->rm_origin == NULL) { 1352842727c2SChris Kirby dsda->need_prep = B_TRUE; 1353842727c2SChris Kirby return (EBUSY); 1354842727c2SChris Kirby } 1355842727c2SChris Kirby 1356842727c2SChris Kirby ndsda.ds = ds_prev; 1357842727c2SChris Kirby ndsda.is_origin_rm = B_TRUE; 1358842727c2SChris Kirby return (dsl_dataset_destroy_check(&ndsda, tag, tx)); 1359842727c2SChris Kirby } 1360842727c2SChris Kirby 1361842727c2SChris Kirby /* 1362842727c2SChris Kirby * If we're not going to remove the origin after all, 1363842727c2SChris Kirby * undo the open context setup. 1364842727c2SChris Kirby */ 1365842727c2SChris Kirby if (dsda->rm_origin != NULL) { 1366842727c2SChris Kirby dsl_dataset_disown(dsda->rm_origin, tag); 1367842727c2SChris Kirby dsda->rm_origin = NULL; 1368842727c2SChris Kirby } 1369842727c2SChris Kirby 1370842727c2SChris Kirby return (0); 1371842727c2SChris Kirby } 1372842727c2SChris Kirby 137399d5e173STim Haley /* 137499d5e173STim Haley * If you add new checks here, you may need to add 137599d5e173STim Haley * additional checks to the "temporary" case in 137699d5e173STim Haley * snapshot_check() in dmu_objset.c. 137799d5e173STim Haley */ 13781d452cf5Sahrens /* ARGSUSED */ 13793cb34c60Sahrens int 13801d452cf5Sahrens dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) 13811d452cf5Sahrens { 1382842727c2SChris Kirby struct dsl_ds_destroyarg *dsda = arg1; 1383842727c2SChris Kirby dsl_dataset_t *ds = dsda->ds; 1384fa9e4066Sahrens 1385745cd3c5Smaybee /* we have an owner hold, so noone else can destroy us */ 1386745cd3c5Smaybee ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 1387745cd3c5Smaybee 1388842727c2SChris Kirby /* 1389842727c2SChris Kirby * Only allow deferred destroy on pools that support it. 1390842727c2SChris Kirby * NOTE: deferred destroy is only supported on snapshots. 1391842727c2SChris Kirby */ 1392842727c2SChris Kirby if (dsda->defer) { 1393842727c2SChris Kirby if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 1394842727c2SChris Kirby SPA_VERSION_USERREFS) 1395842727c2SChris Kirby return (ENOTSUP); 1396842727c2SChris Kirby ASSERT(dsl_dataset_is_snapshot(ds)); 1397842727c2SChris Kirby return (0); 1398842727c2SChris Kirby } 1399fa9e4066Sahrens 1400fa9e4066Sahrens /* 1401fa9e4066Sahrens * Can't delete a head dataset if there are snapshots of it. 1402fa9e4066Sahrens * (Except if the only snapshots are from the branch we cloned 1403fa9e4066Sahrens * from.) 1404fa9e4066Sahrens */ 1405fa9e4066Sahrens if (ds->ds_prev != NULL && 14061d452cf5Sahrens ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 14074aed9999Svitezslav batrla - Sun Microsystems - Prague Czech Republic return (EBUSY); 1408fa9e4066Sahrens 1409fa9e4066Sahrens /* 1410fa9e4066Sahrens * If we made changes this txg, traverse_dsl_dataset won't find 1411fa9e4066Sahrens * them. Try again. 1412fa9e4066Sahrens */ 14131d452cf5Sahrens if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) 1414fa9e4066Sahrens return (EAGAIN); 14151d452cf5Sahrens 1416842727c2SChris Kirby if (dsl_dataset_is_snapshot(ds)) { 1417842727c2SChris Kirby /* 1418842727c2SChris Kirby * If this snapshot has an elevated user reference count, 1419842727c2SChris Kirby * we can't destroy it yet. 1420842727c2SChris Kirby */ 1421842727c2SChris Kirby if (ds->ds_userrefs > 0 && !dsda->releasing) 1422842727c2SChris Kirby return (EBUSY); 1423842727c2SChris Kirby 1424842727c2SChris Kirby mutex_enter(&ds->ds_lock); 1425842727c2SChris Kirby /* 1426842727c2SChris Kirby * Can't delete a branch point. However, if we're destroying 1427842727c2SChris Kirby * a clone and removing its origin due to it having a user 1428842727c2SChris Kirby * hold count of 0 and having been marked for deferred destroy, 1429842727c2SChris Kirby * it's OK for the origin to have a single clone. 1430842727c2SChris Kirby */ 1431842727c2SChris Kirby if (ds->ds_phys->ds_num_children > 1432842727c2SChris Kirby (dsda->is_origin_rm ? 2 : 1)) { 1433842727c2SChris Kirby mutex_exit(&ds->ds_lock); 1434842727c2SChris Kirby return (EEXIST); 1435842727c2SChris Kirby } 1436842727c2SChris Kirby mutex_exit(&ds->ds_lock); 1437842727c2SChris Kirby } else if (dsl_dir_is_clone(ds->ds_dir)) { 1438842727c2SChris Kirby return (dsl_dataset_origin_check(dsda, arg2, tx)); 1439842727c2SChris Kirby } 1440842727c2SChris Kirby 14411d452cf5Sahrens /* XXX we should do some i/o error checking... */ 14421d452cf5Sahrens return (0); 14431d452cf5Sahrens } 14441d452cf5Sahrens 1445745cd3c5Smaybee struct refsarg { 1446745cd3c5Smaybee kmutex_t lock; 1447745cd3c5Smaybee boolean_t gone; 1448745cd3c5Smaybee kcondvar_t cv; 1449745cd3c5Smaybee }; 1450745cd3c5Smaybee 1451745cd3c5Smaybee /* ARGSUSED */ 1452745cd3c5Smaybee static void 1453745cd3c5Smaybee dsl_dataset_refs_gone(dmu_buf_t *db, void *argv) 1454745cd3c5Smaybee { 1455745cd3c5Smaybee struct refsarg *arg = argv; 1456745cd3c5Smaybee 1457745cd3c5Smaybee mutex_enter(&arg->lock); 1458745cd3c5Smaybee arg->gone = TRUE; 1459745cd3c5Smaybee cv_signal(&arg->cv); 1460745cd3c5Smaybee mutex_exit(&arg->lock); 1461745cd3c5Smaybee } 1462745cd3c5Smaybee 1463745cd3c5Smaybee static void 1464745cd3c5Smaybee dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag) 1465745cd3c5Smaybee { 1466745cd3c5Smaybee struct refsarg arg; 1467745cd3c5Smaybee 1468745cd3c5Smaybee mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL); 1469745cd3c5Smaybee cv_init(&arg.cv, NULL, CV_DEFAULT, NULL); 1470745cd3c5Smaybee arg.gone = FALSE; 1471745cd3c5Smaybee (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys, 1472745cd3c5Smaybee dsl_dataset_refs_gone); 1473745cd3c5Smaybee dmu_buf_rele(ds->ds_dbuf, tag); 1474745cd3c5Smaybee mutex_enter(&arg.lock); 1475745cd3c5Smaybee while (!arg.gone) 1476745cd3c5Smaybee cv_wait(&arg.cv, &arg.lock); 1477745cd3c5Smaybee ASSERT(arg.gone); 1478745cd3c5Smaybee mutex_exit(&arg.lock); 1479745cd3c5Smaybee ds->ds_dbuf = NULL; 1480745cd3c5Smaybee ds->ds_phys = NULL; 1481745cd3c5Smaybee mutex_destroy(&arg.lock); 1482745cd3c5Smaybee cv_destroy(&arg.cv); 1483745cd3c5Smaybee } 1484745cd3c5Smaybee 1485c33e334fSMatthew Ahrens static void 1486c33e334fSMatthew Ahrens remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx) 1487c33e334fSMatthew Ahrens { 1488c33e334fSMatthew Ahrens objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1489c33e334fSMatthew Ahrens uint64_t count; 1490c33e334fSMatthew Ahrens int err; 1491c33e334fSMatthew Ahrens 1492c33e334fSMatthew Ahrens ASSERT(ds->ds_phys->ds_num_children >= 2); 1493c33e334fSMatthew Ahrens err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); 1494c33e334fSMatthew Ahrens /* 1495c33e334fSMatthew Ahrens * The err should not be ENOENT, but a bug in a previous version 1496c33e334fSMatthew Ahrens * of the code could cause upgrade_clones_cb() to not set 1497c33e334fSMatthew Ahrens * ds_next_snap_obj when it should, leading to a missing entry. 1498c33e334fSMatthew Ahrens * If we knew that the pool was created after 1499c33e334fSMatthew Ahrens * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1500c33e334fSMatthew Ahrens * ENOENT. However, at least we can check that we don't have 1501c33e334fSMatthew Ahrens * too many entries in the next_clones_obj even after failing to 1502c33e334fSMatthew Ahrens * remove this one. 1503c33e334fSMatthew Ahrens */ 1504c33e334fSMatthew Ahrens if (err != ENOENT) { 1505c33e334fSMatthew Ahrens VERIFY3U(err, ==, 0); 1506c33e334fSMatthew Ahrens } 1507c33e334fSMatthew Ahrens ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj, 1508c33e334fSMatthew Ahrens &count)); 1509c33e334fSMatthew Ahrens ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); 1510c33e334fSMatthew Ahrens } 1511c33e334fSMatthew Ahrens 1512cde58dbcSMatthew Ahrens static void 1513cde58dbcSMatthew Ahrens dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx) 1514cde58dbcSMatthew Ahrens { 1515cde58dbcSMatthew Ahrens objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1516cde58dbcSMatthew Ahrens zap_cursor_t zc; 1517cde58dbcSMatthew Ahrens zap_attribute_t za; 1518cde58dbcSMatthew Ahrens 1519cde58dbcSMatthew Ahrens /* 1520cde58dbcSMatthew Ahrens * If it is the old version, dd_clones doesn't exist so we can't 1521cde58dbcSMatthew Ahrens * find the clones, but deadlist_remove_key() is a no-op so it 1522cde58dbcSMatthew Ahrens * doesn't matter. 1523cde58dbcSMatthew Ahrens */ 1524cde58dbcSMatthew Ahrens if (ds->ds_dir->dd_phys->dd_clones == 0) 1525cde58dbcSMatthew Ahrens return; 1526cde58dbcSMatthew Ahrens 1527cde58dbcSMatthew Ahrens for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones); 1528cde58dbcSMatthew Ahrens zap_cursor_retrieve(&zc, &za) == 0; 1529cde58dbcSMatthew Ahrens zap_cursor_advance(&zc)) { 1530cde58dbcSMatthew Ahrens dsl_dataset_t *clone; 1531cde58dbcSMatthew Ahrens 1532cde58dbcSMatthew Ahrens VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1533cde58dbcSMatthew Ahrens za.za_first_integer, FTAG, &clone)); 1534cde58dbcSMatthew Ahrens if (clone->ds_dir->dd_origin_txg > mintxg) { 1535cde58dbcSMatthew Ahrens dsl_deadlist_remove_key(&clone->ds_deadlist, 1536cde58dbcSMatthew Ahrens mintxg, tx); 1537cde58dbcSMatthew Ahrens dsl_dataset_remove_clones_key(clone, mintxg, tx); 1538cde58dbcSMatthew Ahrens } 1539cde58dbcSMatthew Ahrens dsl_dataset_rele(clone, FTAG); 1540cde58dbcSMatthew Ahrens } 1541cde58dbcSMatthew Ahrens zap_cursor_fini(&zc); 1542cde58dbcSMatthew Ahrens } 1543cde58dbcSMatthew Ahrens 1544cde58dbcSMatthew Ahrens struct process_old_arg { 1545cde58dbcSMatthew Ahrens dsl_dataset_t *ds; 1546cde58dbcSMatthew Ahrens dsl_dataset_t *ds_prev; 1547cde58dbcSMatthew Ahrens boolean_t after_branch_point; 1548cde58dbcSMatthew Ahrens zio_t *pio; 1549cde58dbcSMatthew Ahrens uint64_t used, comp, uncomp; 1550cde58dbcSMatthew Ahrens }; 1551cde58dbcSMatthew Ahrens 1552cde58dbcSMatthew Ahrens static int 1553cde58dbcSMatthew Ahrens process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 1554cde58dbcSMatthew Ahrens { 1555cde58dbcSMatthew Ahrens struct process_old_arg *poa = arg; 1556cde58dbcSMatthew Ahrens dsl_pool_t *dp = poa->ds->ds_dir->dd_pool; 1557cde58dbcSMatthew Ahrens 1558cde58dbcSMatthew Ahrens if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) { 1559cde58dbcSMatthew Ahrens dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx); 1560cde58dbcSMatthew Ahrens if (poa->ds_prev && !poa->after_branch_point && 1561cde58dbcSMatthew Ahrens bp->blk_birth > 1562cde58dbcSMatthew Ahrens poa->ds_prev->ds_phys->ds_prev_snap_txg) { 1563cde58dbcSMatthew Ahrens poa->ds_prev->ds_phys->ds_unique_bytes += 1564cde58dbcSMatthew Ahrens bp_get_dsize_sync(dp->dp_spa, bp); 1565cde58dbcSMatthew Ahrens } 1566cde58dbcSMatthew Ahrens } else { 1567cde58dbcSMatthew Ahrens poa->used += bp_get_dsize_sync(dp->dp_spa, bp); 1568cde58dbcSMatthew Ahrens poa->comp += BP_GET_PSIZE(bp); 1569cde58dbcSMatthew Ahrens poa->uncomp += BP_GET_UCSIZE(bp); 1570cde58dbcSMatthew Ahrens dsl_free_sync(poa->pio, dp, tx->tx_txg, bp); 1571cde58dbcSMatthew Ahrens } 1572cde58dbcSMatthew Ahrens return (0); 1573cde58dbcSMatthew Ahrens } 1574cde58dbcSMatthew Ahrens 1575cde58dbcSMatthew Ahrens static void 1576cde58dbcSMatthew Ahrens process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev, 1577cde58dbcSMatthew Ahrens dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx) 1578cde58dbcSMatthew Ahrens { 1579cde58dbcSMatthew Ahrens struct process_old_arg poa = { 0 }; 1580cde58dbcSMatthew Ahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 1581cde58dbcSMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 1582cde58dbcSMatthew Ahrens 1583cde58dbcSMatthew Ahrens ASSERT(ds->ds_deadlist.dl_oldfmt); 1584cde58dbcSMatthew Ahrens ASSERT(ds_next->ds_deadlist.dl_oldfmt); 1585cde58dbcSMatthew Ahrens 1586cde58dbcSMatthew Ahrens poa.ds = ds; 1587cde58dbcSMatthew Ahrens poa.ds_prev = ds_prev; 1588cde58dbcSMatthew Ahrens poa.after_branch_point = after_branch_point; 1589cde58dbcSMatthew Ahrens poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1590cde58dbcSMatthew Ahrens VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj, 1591cde58dbcSMatthew Ahrens process_old_cb, &poa, tx)); 1592cde58dbcSMatthew Ahrens VERIFY3U(zio_wait(poa.pio), ==, 0); 1593cde58dbcSMatthew Ahrens ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes); 1594cde58dbcSMatthew Ahrens 1595cde58dbcSMatthew Ahrens /* change snapused */ 1596cde58dbcSMatthew Ahrens dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, 1597cde58dbcSMatthew Ahrens -poa.used, -poa.comp, -poa.uncomp, tx); 1598cde58dbcSMatthew Ahrens 1599cde58dbcSMatthew Ahrens /* swap next's deadlist to our deadlist */ 1600cde58dbcSMatthew Ahrens dsl_deadlist_close(&ds->ds_deadlist); 1601cde58dbcSMatthew Ahrens dsl_deadlist_close(&ds_next->ds_deadlist); 1602cde58dbcSMatthew Ahrens SWITCH64(ds_next->ds_phys->ds_deadlist_obj, 1603cde58dbcSMatthew Ahrens ds->ds_phys->ds_deadlist_obj); 1604cde58dbcSMatthew Ahrens dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); 1605cde58dbcSMatthew Ahrens dsl_deadlist_open(&ds_next->ds_deadlist, mos, 1606cde58dbcSMatthew Ahrens ds_next->ds_phys->ds_deadlist_obj); 1607cde58dbcSMatthew Ahrens } 1608cde58dbcSMatthew Ahrens 16093cb34c60Sahrens void 16103f9d6ad7SLin Ling dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) 16111d452cf5Sahrens { 1612842727c2SChris Kirby struct dsl_ds_destroyarg *dsda = arg1; 1613842727c2SChris Kirby dsl_dataset_t *ds = dsda->ds; 16141d452cf5Sahrens int err; 16151d452cf5Sahrens int after_branch_point = FALSE; 16161d452cf5Sahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 16171d452cf5Sahrens objset_t *mos = dp->dp_meta_objset; 16181d452cf5Sahrens dsl_dataset_t *ds_prev = NULL; 161999d5e173STim Haley boolean_t wont_destroy; 16201d452cf5Sahrens uint64_t obj; 16211d452cf5Sahrens 162299d5e173STim Haley wont_destroy = (dsda->defer && 162399d5e173STim Haley (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1)); 162499d5e173STim Haley 162599d5e173STim Haley ASSERT(ds->ds_owner || wont_destroy); 1626842727c2SChris Kirby ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1); 16271d452cf5Sahrens ASSERT(ds->ds_prev == NULL || 16281d452cf5Sahrens ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); 16291d452cf5Sahrens ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); 16301d452cf5Sahrens 163199d5e173STim Haley if (wont_destroy) { 1632842727c2SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 163399d5e173STim Haley dmu_buf_will_dirty(ds->ds_dbuf, tx); 163499d5e173STim Haley ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY; 163599d5e173STim Haley return; 1636842727c2SChris Kirby } 1637842727c2SChris Kirby 1638745cd3c5Smaybee /* signal any waiters that this dataset is going away */ 1639745cd3c5Smaybee mutex_enter(&ds->ds_lock); 1640745cd3c5Smaybee ds->ds_owner = dsl_reaper; 1641745cd3c5Smaybee cv_broadcast(&ds->ds_exclusive_cv); 1642745cd3c5Smaybee mutex_exit(&ds->ds_lock); 1643745cd3c5Smaybee 1644a9799022Sck /* Remove our reservation */ 1645a9799022Sck if (ds->ds_reserved != 0) { 164692241e0bSTom Erickson dsl_prop_setarg_t psa; 164792241e0bSTom Erickson uint64_t value = 0; 164892241e0bSTom Erickson 164992241e0bSTom Erickson dsl_prop_setarg_init_uint64(&psa, "refreservation", 165092241e0bSTom Erickson (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED), 165192241e0bSTom Erickson &value); 165292241e0bSTom Erickson psa.psa_effective_value = 0; /* predict default value */ 165392241e0bSTom Erickson 16543f9d6ad7SLin Ling dsl_dataset_set_reservation_sync(ds, &psa, tx); 1655a9799022Sck ASSERT3U(ds->ds_reserved, ==, 0); 1656a9799022Sck } 1657a9799022Sck 16581d452cf5Sahrens ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 16591d452cf5Sahrens 16603f9d6ad7SLin Ling dsl_scan_ds_destroyed(ds, tx); 1661088f3894Sahrens 16621d452cf5Sahrens obj = ds->ds_object; 1663fa9e4066Sahrens 1664fa9e4066Sahrens if (ds->ds_phys->ds_prev_snap_obj != 0) { 1665fa9e4066Sahrens if (ds->ds_prev) { 1666fa9e4066Sahrens ds_prev = ds->ds_prev; 1667fa9e4066Sahrens } else { 1668745cd3c5Smaybee VERIFY(0 == dsl_dataset_hold_obj(dp, 1669745cd3c5Smaybee ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev)); 1670fa9e4066Sahrens } 1671fa9e4066Sahrens after_branch_point = 1672fa9e4066Sahrens (ds_prev->ds_phys->ds_next_snap_obj != obj); 1673fa9e4066Sahrens 1674fa9e4066Sahrens dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); 1675088f3894Sahrens if (after_branch_point && 1676088f3894Sahrens ds_prev->ds_phys->ds_next_clones_obj != 0) { 1677c33e334fSMatthew Ahrens remove_from_next_clones(ds_prev, obj, tx); 1678088f3894Sahrens if (ds->ds_phys->ds_next_snap_obj != 0) { 1679088f3894Sahrens VERIFY(0 == zap_add_int(mos, 1680088f3894Sahrens ds_prev->ds_phys->ds_next_clones_obj, 1681088f3894Sahrens ds->ds_phys->ds_next_snap_obj, tx)); 1682088f3894Sahrens } 1683088f3894Sahrens } 1684fa9e4066Sahrens if (after_branch_point && 1685fa9e4066Sahrens ds->ds_phys->ds_next_snap_obj == 0) { 1686fa9e4066Sahrens /* This clone is toast. */ 1687fa9e4066Sahrens ASSERT(ds_prev->ds_phys->ds_num_children > 1); 1688fa9e4066Sahrens ds_prev->ds_phys->ds_num_children--; 1689842727c2SChris Kirby 1690842727c2SChris Kirby /* 1691842727c2SChris Kirby * If the clone's origin has no other clones, no 1692842727c2SChris Kirby * user holds, and has been marked for deferred 1693842727c2SChris Kirby * deletion, then we should have done the necessary 1694842727c2SChris Kirby * destroy setup for it. 1695842727c2SChris Kirby */ 1696842727c2SChris Kirby if (ds_prev->ds_phys->ds_num_children == 1 && 1697842727c2SChris Kirby ds_prev->ds_userrefs == 0 && 1698842727c2SChris Kirby DS_IS_DEFER_DESTROY(ds_prev)) { 1699842727c2SChris Kirby ASSERT3P(dsda->rm_origin, !=, NULL); 1700842727c2SChris Kirby } else { 1701842727c2SChris Kirby ASSERT3P(dsda->rm_origin, ==, NULL); 1702842727c2SChris Kirby } 1703fa9e4066Sahrens } else if (!after_branch_point) { 1704fa9e4066Sahrens ds_prev->ds_phys->ds_next_snap_obj = 1705fa9e4066Sahrens ds->ds_phys->ds_next_snap_obj; 1706fa9e4066Sahrens } 1707fa9e4066Sahrens } 1708fa9e4066Sahrens 17093f9d6ad7SLin Ling if (dsl_dataset_is_snapshot(ds)) { 1710fa9e4066Sahrens dsl_dataset_t *ds_next; 1711a9799022Sck uint64_t old_unique; 1712cde58dbcSMatthew Ahrens uint64_t used = 0, comp = 0, uncomp = 0; 1713fa9e4066Sahrens 1714745cd3c5Smaybee VERIFY(0 == dsl_dataset_hold_obj(dp, 1715745cd3c5Smaybee ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next)); 1716fa9e4066Sahrens ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); 1717fa9e4066Sahrens 17183f9d6ad7SLin Ling old_unique = ds_next->ds_phys->ds_unique_bytes; 1719a9799022Sck 1720fa9e4066Sahrens dmu_buf_will_dirty(ds_next->ds_dbuf, tx); 1721fa9e4066Sahrens ds_next->ds_phys->ds_prev_snap_obj = 1722fa9e4066Sahrens ds->ds_phys->ds_prev_snap_obj; 1723fa9e4066Sahrens ds_next->ds_phys->ds_prev_snap_txg = 1724fa9e4066Sahrens ds->ds_phys->ds_prev_snap_txg; 1725fa9e4066Sahrens ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1726fa9e4066Sahrens ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); 1727fa9e4066Sahrens 17283113f7ceSGeorge Wilson 1729cde58dbcSMatthew Ahrens if (ds_next->ds_deadlist.dl_oldfmt) { 1730cde58dbcSMatthew Ahrens process_old_deadlist(ds, ds_prev, ds_next, 1731cde58dbcSMatthew Ahrens after_branch_point, tx); 1732cde58dbcSMatthew Ahrens } else { 1733cde58dbcSMatthew Ahrens /* Adjust prev's unique space. */ 1734cde58dbcSMatthew Ahrens if (ds_prev && !after_branch_point) { 1735cde58dbcSMatthew Ahrens dsl_deadlist_space_range(&ds_next->ds_deadlist, 1736cde58dbcSMatthew Ahrens ds_prev->ds_phys->ds_prev_snap_txg, 1737cde58dbcSMatthew Ahrens ds->ds_phys->ds_prev_snap_txg, 1738cde58dbcSMatthew Ahrens &used, &comp, &uncomp); 1739cde58dbcSMatthew Ahrens ds_prev->ds_phys->ds_unique_bytes += used; 1740fa9e4066Sahrens } 174174e7dc98SMatthew Ahrens 1742cde58dbcSMatthew Ahrens /* Adjust snapused. */ 1743cde58dbcSMatthew Ahrens dsl_deadlist_space_range(&ds_next->ds_deadlist, 1744cde58dbcSMatthew Ahrens ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, 1745cde58dbcSMatthew Ahrens &used, &comp, &uncomp); 1746cde58dbcSMatthew Ahrens dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, 1747cde58dbcSMatthew Ahrens -used, -comp, -uncomp, tx); 1748cde58dbcSMatthew Ahrens 1749cde58dbcSMatthew Ahrens /* Move blocks to be freed to pool's free list. */ 1750cde58dbcSMatthew Ahrens dsl_deadlist_move_bpobj(&ds_next->ds_deadlist, 1751cde58dbcSMatthew Ahrens &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg, 1752cde58dbcSMatthew Ahrens tx); 1753cde58dbcSMatthew Ahrens dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, 1754cde58dbcSMatthew Ahrens DD_USED_HEAD, used, comp, uncomp, tx); 1755cde58dbcSMatthew Ahrens dsl_dir_dirty(tx->tx_pool->dp_free_dir, tx); 1756cde58dbcSMatthew Ahrens 1757cde58dbcSMatthew Ahrens /* Merge our deadlist into next's and free it. */ 1758cde58dbcSMatthew Ahrens dsl_deadlist_merge(&ds_next->ds_deadlist, 1759cde58dbcSMatthew Ahrens ds->ds_phys->ds_deadlist_obj, tx); 1760cde58dbcSMatthew Ahrens } 1761cde58dbcSMatthew Ahrens dsl_deadlist_close(&ds->ds_deadlist); 1762cde58dbcSMatthew Ahrens dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); 1763fa9e4066Sahrens 1764cde58dbcSMatthew Ahrens /* Collapse range in clone heads */ 1765cde58dbcSMatthew Ahrens dsl_dataset_remove_clones_key(ds, 1766cde58dbcSMatthew Ahrens ds->ds_phys->ds_creation_txg, tx); 1767fa9e4066Sahrens 17683f9d6ad7SLin Ling if (dsl_dataset_is_snapshot(ds_next)) { 1769cde58dbcSMatthew Ahrens dsl_dataset_t *ds_nextnext; 1770cde58dbcSMatthew Ahrens 1771fa9e4066Sahrens /* 1772fa9e4066Sahrens * Update next's unique to include blocks which 1773fa9e4066Sahrens * were previously shared by only this snapshot 1774fa9e4066Sahrens * and it. Those blocks will be born after the 1775fa9e4066Sahrens * prev snap and before this snap, and will have 1776fa9e4066Sahrens * died after the next snap and before the one 1777fa9e4066Sahrens * after that (ie. be on the snap after next's 1778fa9e4066Sahrens * deadlist). 1779fa9e4066Sahrens */ 1780745cd3c5Smaybee VERIFY(0 == dsl_dataset_hold_obj(dp, 1781745cd3c5Smaybee ds_next->ds_phys->ds_next_snap_obj, 1782cde58dbcSMatthew Ahrens FTAG, &ds_nextnext)); 1783cde58dbcSMatthew Ahrens dsl_deadlist_space_range(&ds_nextnext->ds_deadlist, 178474e7dc98SMatthew Ahrens ds->ds_phys->ds_prev_snap_txg, 1785cde58dbcSMatthew Ahrens ds->ds_phys->ds_creation_txg, 1786cde58dbcSMatthew Ahrens &used, &comp, &uncomp); 1787cde58dbcSMatthew Ahrens ds_next->ds_phys->ds_unique_bytes += used; 1788cde58dbcSMatthew Ahrens dsl_dataset_rele(ds_nextnext, FTAG); 1789fa9e4066Sahrens ASSERT3P(ds_next->ds_prev, ==, NULL); 1790cde58dbcSMatthew Ahrens 1791cde58dbcSMatthew Ahrens /* Collapse range in this head. */ 1792cde58dbcSMatthew Ahrens dsl_dataset_t *hds; 1793cde58dbcSMatthew Ahrens VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, 1794cde58dbcSMatthew Ahrens ds->ds_dir->dd_phys->dd_head_dataset_obj, 1795cde58dbcSMatthew Ahrens FTAG, &hds)); 1796cde58dbcSMatthew Ahrens dsl_deadlist_remove_key(&hds->ds_deadlist, 1797cde58dbcSMatthew Ahrens ds->ds_phys->ds_creation_txg, tx); 1798cde58dbcSMatthew Ahrens dsl_dataset_rele(hds, FTAG); 1799cde58dbcSMatthew Ahrens 1800fa9e4066Sahrens } else { 1801fa9e4066Sahrens ASSERT3P(ds_next->ds_prev, ==, ds); 1802745cd3c5Smaybee dsl_dataset_drop_ref(ds_next->ds_prev, ds_next); 1803745cd3c5Smaybee ds_next->ds_prev = NULL; 1804fa9e4066Sahrens if (ds_prev) { 1805745cd3c5Smaybee VERIFY(0 == dsl_dataset_get_ref(dp, 1806745cd3c5Smaybee ds->ds_phys->ds_prev_snap_obj, 1807745cd3c5Smaybee ds_next, &ds_next->ds_prev)); 1808fa9e4066Sahrens } 1809a9799022Sck 1810a9799022Sck dsl_dataset_recalc_head_uniq(ds_next); 1811a9799022Sck 1812a9799022Sck /* 1813a9799022Sck * Reduce the amount of our unconsmed refreservation 1814a9799022Sck * being charged to our parent by the amount of 1815a9799022Sck * new unique data we have gained. 1816a9799022Sck */ 1817a9799022Sck if (old_unique < ds_next->ds_reserved) { 1818a9799022Sck int64_t mrsdelta; 1819a9799022Sck uint64_t new_unique = 1820a9799022Sck ds_next->ds_phys->ds_unique_bytes; 1821a9799022Sck 1822a9799022Sck ASSERT(old_unique <= new_unique); 1823a9799022Sck mrsdelta = MIN(new_unique - old_unique, 1824a9799022Sck ds_next->ds_reserved - old_unique); 182574e7dc98SMatthew Ahrens dsl_dir_diduse_space(ds->ds_dir, 182674e7dc98SMatthew Ahrens DD_USED_REFRSRV, -mrsdelta, 0, 0, tx); 1827a9799022Sck } 1828fa9e4066Sahrens } 1829745cd3c5Smaybee dsl_dataset_rele(ds_next, FTAG); 1830fa9e4066Sahrens } else { 1831fa9e4066Sahrens /* 1832fa9e4066Sahrens * There's no next snapshot, so this is a head dataset. 1833fa9e4066Sahrens * Destroy the deadlist. Unless it's a clone, the 1834fa9e4066Sahrens * deadlist should be empty. (If it's a clone, it's 1835fa9e4066Sahrens * safe to ignore the deadlist contents.) 1836fa9e4066Sahrens */ 1837fa9e4066Sahrens struct killarg ka; 1838fa9e4066Sahrens 1839cde58dbcSMatthew Ahrens dsl_deadlist_close(&ds->ds_deadlist); 1840cde58dbcSMatthew Ahrens dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); 1841fa9e4066Sahrens ds->ds_phys->ds_deadlist_obj = 0; 1842fa9e4066Sahrens 1843fa9e4066Sahrens /* 1844fa9e4066Sahrens * Free everything that we point to (that's born after 1845fa9e4066Sahrens * the previous snapshot, if we are a clone) 1846fa9e4066Sahrens * 184774e7dc98SMatthew Ahrens * NB: this should be very quick, because we already 184874e7dc98SMatthew Ahrens * freed all the objects in open context. 1849fa9e4066Sahrens */ 185074e7dc98SMatthew Ahrens ka.ds = ds; 1851fa9e4066Sahrens ka.tx = tx; 185288b7b0f2SMatthew Ahrens err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg, 185388b7b0f2SMatthew Ahrens TRAVERSE_POST, kill_blkptr, &ka); 1854fa9e4066Sahrens ASSERT3U(err, ==, 0); 18553e78c5fbSChris Kirby ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || 185674e7dc98SMatthew Ahrens ds->ds_phys->ds_unique_bytes == 0); 1857ca45db41SChris Kirby 1858ca45db41SChris Kirby if (ds->ds_prev != NULL) { 1859cde58dbcSMatthew Ahrens if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 1860cde58dbcSMatthew Ahrens VERIFY3U(0, ==, zap_remove_int(mos, 1861cde58dbcSMatthew Ahrens ds->ds_prev->ds_dir->dd_phys->dd_clones, 1862cde58dbcSMatthew Ahrens ds->ds_object, tx)); 1863cde58dbcSMatthew Ahrens } 1864ca45db41SChris Kirby dsl_dataset_rele(ds->ds_prev, ds); 1865ca45db41SChris Kirby ds->ds_prev = ds_prev = NULL; 1866ca45db41SChris Kirby } 1867fa9e4066Sahrens } 1868fa9e4066Sahrens 18696e0cbcaaSMatthew Ahrens /* 18706e0cbcaaSMatthew Ahrens * This must be done after the dsl_traverse(), because it will 18716e0cbcaaSMatthew Ahrens * re-open the objset. 18726e0cbcaaSMatthew Ahrens */ 18736e0cbcaaSMatthew Ahrens if (ds->ds_objset) { 18746e0cbcaaSMatthew Ahrens dmu_objset_evict(ds->ds_objset); 18756e0cbcaaSMatthew Ahrens ds->ds_objset = NULL; 18766e0cbcaaSMatthew Ahrens } 18776e0cbcaaSMatthew Ahrens 18781d452cf5Sahrens if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { 1879745cd3c5Smaybee /* Erase the link in the dir */ 18801d452cf5Sahrens dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 18811d452cf5Sahrens ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; 1882745cd3c5Smaybee ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0); 1883745cd3c5Smaybee err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); 1884745cd3c5Smaybee ASSERT(err == 0); 1885fa9e4066Sahrens } else { 1886fa9e4066Sahrens /* remove from snapshot namespace */ 1887fa9e4066Sahrens dsl_dataset_t *ds_head; 1888745cd3c5Smaybee ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0); 1889745cd3c5Smaybee VERIFY(0 == dsl_dataset_hold_obj(dp, 1890745cd3c5Smaybee ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head)); 18918660574dSahrens VERIFY(0 == dsl_dataset_get_snapname(ds)); 1892fa9e4066Sahrens #ifdef ZFS_DEBUG 1893fa9e4066Sahrens { 1894fa9e4066Sahrens uint64_t val; 1895ab04eb8eStimh 1896745cd3c5Smaybee err = dsl_dataset_snap_lookup(ds_head, 1897ab04eb8eStimh ds->ds_snapname, &val); 1898fa9e4066Sahrens ASSERT3U(err, ==, 0); 1899fa9e4066Sahrens ASSERT3U(val, ==, obj); 1900fa9e4066Sahrens } 1901fa9e4066Sahrens #endif 1902745cd3c5Smaybee err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx); 1903fa9e4066Sahrens ASSERT(err == 0); 1904745cd3c5Smaybee dsl_dataset_rele(ds_head, FTAG); 1905fa9e4066Sahrens } 1906fa9e4066Sahrens 1907fa9e4066Sahrens if (ds_prev && ds->ds_prev != ds_prev) 1908745cd3c5Smaybee dsl_dataset_rele(ds_prev, FTAG); 1909fa9e4066Sahrens 1910990b4856Slling spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx); 19113f9d6ad7SLin Ling spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx, 19123f9d6ad7SLin Ling "dataset = %llu", ds->ds_object); 1913ecd6cf80Smarks 1914088f3894Sahrens if (ds->ds_phys->ds_next_clones_obj != 0) { 1915088f3894Sahrens uint64_t count; 1916088f3894Sahrens ASSERT(0 == zap_count(mos, 1917088f3894Sahrens ds->ds_phys->ds_next_clones_obj, &count) && count == 0); 1918088f3894Sahrens VERIFY(0 == dmu_object_free(mos, 1919088f3894Sahrens ds->ds_phys->ds_next_clones_obj, tx)); 1920088f3894Sahrens } 192174e7dc98SMatthew Ahrens if (ds->ds_phys->ds_props_obj != 0) 192274e7dc98SMatthew Ahrens VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx)); 1923842727c2SChris Kirby if (ds->ds_phys->ds_userrefs_obj != 0) 1924842727c2SChris Kirby VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx)); 1925745cd3c5Smaybee dsl_dir_close(ds->ds_dir, ds); 1926745cd3c5Smaybee ds->ds_dir = NULL; 1927745cd3c5Smaybee dsl_dataset_drain_refs(ds, tag); 19281d452cf5Sahrens VERIFY(0 == dmu_object_free(mos, obj, tx)); 1929842727c2SChris Kirby 1930842727c2SChris Kirby if (dsda->rm_origin) { 1931842727c2SChris Kirby /* 1932842727c2SChris Kirby * Remove the origin of the clone we just destroyed. 1933842727c2SChris Kirby */ 1934842727c2SChris Kirby struct dsl_ds_destroyarg ndsda = {0}; 1935842727c2SChris Kirby 1936ca45db41SChris Kirby ndsda.ds = dsda->rm_origin; 19373f9d6ad7SLin Ling dsl_dataset_destroy_sync(&ndsda, tag, tx); 1938842727c2SChris Kirby } 1939fa9e4066Sahrens } 1940fa9e4066Sahrens 1941a9799022Sck static int 1942a9799022Sck dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1943a9799022Sck { 1944a9799022Sck uint64_t asize; 1945a9799022Sck 1946a9799022Sck if (!dmu_tx_is_syncing(tx)) 1947a9799022Sck return (0); 1948a9799022Sck 1949a9799022Sck /* 1950a9799022Sck * If there's an fs-only reservation, any blocks that might become 1951a9799022Sck * owned by the snapshot dataset must be accommodated by space 1952a9799022Sck * outside of the reservation. 1953a9799022Sck */ 19543f9d6ad7SLin Ling ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 19553f9d6ad7SLin Ling asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 19566e0cbcaaSMatthew Ahrens if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1957a9799022Sck return (ENOSPC); 1958a9799022Sck 1959a9799022Sck /* 1960a9799022Sck * Propogate any reserved space for this snapshot to other 1961a9799022Sck * snapshot checks in this sync group. 1962a9799022Sck */ 1963a9799022Sck if (asize > 0) 1964a9799022Sck dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1965a9799022Sck 1966a9799022Sck return (0); 1967a9799022Sck } 1968a9799022Sck 1969fa9e4066Sahrens int 19701d452cf5Sahrens dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx) 1971fa9e4066Sahrens { 19723cb34c60Sahrens dsl_dataset_t *ds = arg1; 19731d452cf5Sahrens const char *snapname = arg2; 1974fa9e4066Sahrens int err; 19751d452cf5Sahrens uint64_t value; 1976fa9e4066Sahrens 19771d452cf5Sahrens /* 19781d452cf5Sahrens * We don't allow multiple snapshots of the same txg. If there 19791d452cf5Sahrens * is already one, try again. 19801d452cf5Sahrens */ 19811d452cf5Sahrens if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) 19821d452cf5Sahrens return (EAGAIN); 1983fa9e4066Sahrens 19841d452cf5Sahrens /* 19851d452cf5Sahrens * Check for conflicting name snapshot name. 19861d452cf5Sahrens */ 1987745cd3c5Smaybee err = dsl_dataset_snap_lookup(ds, snapname, &value); 19881d452cf5Sahrens if (err == 0) 1989fa9e4066Sahrens return (EEXIST); 19901d452cf5Sahrens if (err != ENOENT) 19911d452cf5Sahrens return (err); 1992fa9e4066Sahrens 1993b7661cccSmmusante /* 1994b7661cccSmmusante * Check that the dataset's name is not too long. Name consists 1995b7661cccSmmusante * of the dataset's length + 1 for the @-sign + snapshot name's length 1996b7661cccSmmusante */ 1997b7661cccSmmusante if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN) 1998b7661cccSmmusante return (ENAMETOOLONG); 1999b7661cccSmmusante 2000a9799022Sck err = dsl_dataset_snapshot_reserve_space(ds, tx); 2001a9799022Sck if (err) 2002a9799022Sck return (err); 2003a9799022Sck 20041d452cf5Sahrens ds->ds_trysnap_txg = tx->tx_txg; 20051d452cf5Sahrens return (0); 20061d452cf5Sahrens } 2007fa9e4066Sahrens 20081d452cf5Sahrens void 20093f9d6ad7SLin Ling dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx) 20101d452cf5Sahrens { 20113cb34c60Sahrens dsl_dataset_t *ds = arg1; 20121d452cf5Sahrens const char *snapname = arg2; 20131d452cf5Sahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 20141d452cf5Sahrens dmu_buf_t *dbuf; 20151d452cf5Sahrens dsl_dataset_phys_t *dsphys; 2016088f3894Sahrens uint64_t dsobj, crtxg; 20171d452cf5Sahrens objset_t *mos = dp->dp_meta_objset; 20181d452cf5Sahrens int err; 2019fa9e4066Sahrens 20201d452cf5Sahrens ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 2021fa9e4066Sahrens 2022088f3894Sahrens /* 2023088f3894Sahrens * The origin's ds_creation_txg has to be < TXG_INITIAL 2024088f3894Sahrens */ 2025088f3894Sahrens if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 2026088f3894Sahrens crtxg = 1; 2027088f3894Sahrens else 2028088f3894Sahrens crtxg = tx->tx_txg; 2029088f3894Sahrens 20301649cd4bStabriz dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 20311649cd4bStabriz DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 2032ea8dc4b6Seschrock VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 2033fa9e4066Sahrens dmu_buf_will_dirty(dbuf, tx); 2034fa9e4066Sahrens dsphys = dbuf->db_data; 2035745cd3c5Smaybee bzero(dsphys, sizeof (dsl_dataset_phys_t)); 20361d452cf5Sahrens dsphys->ds_dir_obj = ds->ds_dir->dd_object; 2037fa9e4066Sahrens dsphys->ds_fsid_guid = unique_create(); 2038fa9e4066Sahrens (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 2039fa9e4066Sahrens sizeof (dsphys->ds_guid)); 2040fa9e4066Sahrens dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; 2041fa9e4066Sahrens dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; 2042fa9e4066Sahrens dsphys->ds_next_snap_obj = ds->ds_object; 2043fa9e4066Sahrens dsphys->ds_num_children = 1; 2044fa9e4066Sahrens dsphys->ds_creation_time = gethrestime_sec(); 2045088f3894Sahrens dsphys->ds_creation_txg = crtxg; 2046fa9e4066Sahrens dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; 2047fa9e4066Sahrens dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; 2048fa9e4066Sahrens dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; 2049fa9e4066Sahrens dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; 205099653d4eSeschrock dsphys->ds_flags = ds->ds_phys->ds_flags; 2051fa9e4066Sahrens dsphys->ds_bp = ds->ds_phys->ds_bp; 2052ea8dc4b6Seschrock dmu_buf_rele(dbuf, FTAG); 2053fa9e4066Sahrens 20541d452cf5Sahrens ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); 20551d452cf5Sahrens if (ds->ds_prev) { 2056088f3894Sahrens uint64_t next_clones_obj = 2057088f3894Sahrens ds->ds_prev->ds_phys->ds_next_clones_obj; 20581d452cf5Sahrens ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == 2059fa9e4066Sahrens ds->ds_object || 20601d452cf5Sahrens ds->ds_prev->ds_phys->ds_num_children > 1); 20611d452cf5Sahrens if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 20621d452cf5Sahrens dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 2063fa9e4066Sahrens ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 20641d452cf5Sahrens ds->ds_prev->ds_phys->ds_creation_txg); 20651d452cf5Sahrens ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; 2066088f3894Sahrens } else if (next_clones_obj != 0) { 2067c33e334fSMatthew Ahrens remove_from_next_clones(ds->ds_prev, 2068c33e334fSMatthew Ahrens dsphys->ds_next_snap_obj, tx); 2069088f3894Sahrens VERIFY3U(0, ==, zap_add_int(mos, 2070088f3894Sahrens next_clones_obj, dsobj, tx)); 2071fa9e4066Sahrens } 2072fa9e4066Sahrens } 2073fa9e4066Sahrens 2074a9799022Sck /* 2075a9799022Sck * If we have a reference-reservation on this dataset, we will 2076a9799022Sck * need to increase the amount of refreservation being charged 2077a9799022Sck * since our unique space is going to zero. 2078a9799022Sck */ 2079a9799022Sck if (ds->ds_reserved) { 20803f9d6ad7SLin Ling int64_t delta; 20813f9d6ad7SLin Ling ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 20823f9d6ad7SLin Ling delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 208374e7dc98SMatthew Ahrens dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 20843f9d6ad7SLin Ling delta, 0, 0, tx); 2085a9799022Sck } 2086a9799022Sck 2087fa9e4066Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 2088cde58dbcSMatthew Ahrens zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu", 2089cde58dbcSMatthew Ahrens ds->ds_dir->dd_myname, snapname, dsobj, 2090cde58dbcSMatthew Ahrens ds->ds_phys->ds_prev_snap_txg); 2091cde58dbcSMatthew Ahrens ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist, 2092cde58dbcSMatthew Ahrens UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx); 2093cde58dbcSMatthew Ahrens dsl_deadlist_close(&ds->ds_deadlist); 2094cde58dbcSMatthew Ahrens dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); 2095cde58dbcSMatthew Ahrens dsl_deadlist_add_key(&ds->ds_deadlist, 2096cde58dbcSMatthew Ahrens ds->ds_phys->ds_prev_snap_txg, tx); 2097cde58dbcSMatthew Ahrens 2098a4611edeSahrens ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); 2099fa9e4066Sahrens ds->ds_phys->ds_prev_snap_obj = dsobj; 2100088f3894Sahrens ds->ds_phys->ds_prev_snap_txg = crtxg; 2101fa9e4066Sahrens ds->ds_phys->ds_unique_bytes = 0; 2102a9799022Sck if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 2103a9799022Sck ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 2104fa9e4066Sahrens 2105fa9e4066Sahrens err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, 2106fa9e4066Sahrens snapname, 8, 1, &dsobj, tx); 2107fa9e4066Sahrens ASSERT(err == 0); 2108fa9e4066Sahrens 2109fa9e4066Sahrens if (ds->ds_prev) 2110745cd3c5Smaybee dsl_dataset_drop_ref(ds->ds_prev, ds); 2111745cd3c5Smaybee VERIFY(0 == dsl_dataset_get_ref(dp, 2112745cd3c5Smaybee ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 2113ecd6cf80Smarks 21143f9d6ad7SLin Ling dsl_scan_ds_snapshotted(ds, tx); 2115088f3894Sahrens 211671eb0538SChris Kirby dsl_dir_snap_cmtime_update(ds->ds_dir); 211771eb0538SChris Kirby 21183f9d6ad7SLin Ling spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx, 211940feaa91Sahrens "dataset = %llu", dsobj); 2120fa9e4066Sahrens } 2121fa9e4066Sahrens 2122fa9e4066Sahrens void 2123c717a561Smaybee dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 2124fa9e4066Sahrens { 2125fa9e4066Sahrens ASSERT(dmu_tx_is_syncing(tx)); 2126503ad85cSMatthew Ahrens ASSERT(ds->ds_objset != NULL); 2127fa9e4066Sahrens ASSERT(ds->ds_phys->ds_next_snap_obj == 0); 2128fa9e4066Sahrens 212991ebeef5Sahrens /* 213091ebeef5Sahrens * in case we had to change ds_fsid_guid when we opened it, 213191ebeef5Sahrens * sync it out now. 213291ebeef5Sahrens */ 213391ebeef5Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 213491ebeef5Sahrens ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; 213591ebeef5Sahrens 2136fa9e4066Sahrens dsl_dir_dirty(ds->ds_dir, tx); 2137503ad85cSMatthew Ahrens dmu_objset_sync(ds->ds_objset, zio, tx); 2138fa9e4066Sahrens } 2139fa9e4066Sahrens 214019b94df9SMatthew Ahrens static void 214119b94df9SMatthew Ahrens get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 214219b94df9SMatthew Ahrens { 214319b94df9SMatthew Ahrens uint64_t count = 0; 214419b94df9SMatthew Ahrens objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 214519b94df9SMatthew Ahrens zap_cursor_t zc; 214619b94df9SMatthew Ahrens zap_attribute_t za; 214719b94df9SMatthew Ahrens nvlist_t *propval; 214819b94df9SMatthew Ahrens nvlist_t *val; 214919b94df9SMatthew Ahrens 215019b94df9SMatthew Ahrens rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); 215119b94df9SMatthew Ahrens VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 215219b94df9SMatthew Ahrens VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0); 215319b94df9SMatthew Ahrens 215419b94df9SMatthew Ahrens /* 215519b94df9SMatthew Ahrens * There may me missing entries in ds_next_clones_obj 215619b94df9SMatthew Ahrens * due to a bug in a previous version of the code. 215719b94df9SMatthew Ahrens * Only trust it if it has the right number of entries. 215819b94df9SMatthew Ahrens */ 215919b94df9SMatthew Ahrens if (ds->ds_phys->ds_next_clones_obj != 0) { 216019b94df9SMatthew Ahrens ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj, 216119b94df9SMatthew Ahrens &count)); 216219b94df9SMatthew Ahrens } 216319b94df9SMatthew Ahrens if (count != ds->ds_phys->ds_num_children - 1) { 216419b94df9SMatthew Ahrens goto fail; 216519b94df9SMatthew Ahrens } 216619b94df9SMatthew Ahrens for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj); 216719b94df9SMatthew Ahrens zap_cursor_retrieve(&zc, &za) == 0; 216819b94df9SMatthew Ahrens zap_cursor_advance(&zc)) { 216919b94df9SMatthew Ahrens dsl_dataset_t *clone; 217019b94df9SMatthew Ahrens char buf[ZFS_MAXNAMELEN]; 217119b94df9SMatthew Ahrens if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 217219b94df9SMatthew Ahrens za.za_first_integer, FTAG, &clone) != 0) { 217319b94df9SMatthew Ahrens goto fail; 217419b94df9SMatthew Ahrens } 217519b94df9SMatthew Ahrens dsl_dir_name(clone->ds_dir, buf); 217619b94df9SMatthew Ahrens VERIFY(nvlist_add_boolean(val, buf) == 0); 217719b94df9SMatthew Ahrens dsl_dataset_rele(clone, FTAG); 217819b94df9SMatthew Ahrens } 217919b94df9SMatthew Ahrens zap_cursor_fini(&zc); 218019b94df9SMatthew Ahrens VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0); 218119b94df9SMatthew Ahrens VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), 218219b94df9SMatthew Ahrens propval) == 0); 218319b94df9SMatthew Ahrens fail: 218419b94df9SMatthew Ahrens nvlist_free(val); 218519b94df9SMatthew Ahrens nvlist_free(propval); 218619b94df9SMatthew Ahrens rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); 218719b94df9SMatthew Ahrens } 218819b94df9SMatthew Ahrens 2189fa9e4066Sahrens void 2190a2eea2e1Sahrens dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 2191fa9e4066Sahrens { 2192187d6ac0SMatt Ahrens uint64_t refd, avail, uobjs, aobjs, ratio; 2193a9799022Sck 2194a2eea2e1Sahrens dsl_dir_stats(ds->ds_dir, nv); 2195fa9e4066Sahrens 2196a9799022Sck dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 2197a9799022Sck dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 2198a9799022Sck dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 2199a9799022Sck 2200a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 2201a2eea2e1Sahrens ds->ds_phys->ds_creation_time); 2202a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2203a2eea2e1Sahrens ds->ds_phys->ds_creation_txg); 2204a9799022Sck dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2205a9799022Sck ds->ds_quota); 2206a9799022Sck dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2207a9799022Sck ds->ds_reserved); 2208c5904d13Seschrock dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2209c5904d13Seschrock ds->ds_phys->ds_guid); 22101d713200SEric Schrock dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 22113f9d6ad7SLin Ling ds->ds_phys->ds_unique_bytes); 22121d713200SEric Schrock dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 22131d713200SEric Schrock ds->ds_object); 221492241e0bSTom Erickson dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 221592241e0bSTom Erickson ds->ds_userrefs); 2216842727c2SChris Kirby dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2217842727c2SChris Kirby DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 2218fa9e4066Sahrens 221919b94df9SMatthew Ahrens if (ds->ds_phys->ds_prev_snap_obj != 0) { 222019b94df9SMatthew Ahrens uint64_t written, comp, uncomp; 222119b94df9SMatthew Ahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 222219b94df9SMatthew Ahrens dsl_dataset_t *prev; 222319b94df9SMatthew Ahrens 222419b94df9SMatthew Ahrens rw_enter(&dp->dp_config_rwlock, RW_READER); 222519b94df9SMatthew Ahrens int err = dsl_dataset_hold_obj(dp, 222619b94df9SMatthew Ahrens ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); 222719b94df9SMatthew Ahrens rw_exit(&dp->dp_config_rwlock); 222819b94df9SMatthew Ahrens if (err == 0) { 222919b94df9SMatthew Ahrens err = dsl_dataset_space_written(prev, ds, &written, 223019b94df9SMatthew Ahrens &comp, &uncomp); 223119b94df9SMatthew Ahrens dsl_dataset_rele(prev, FTAG); 223219b94df9SMatthew Ahrens if (err == 0) { 223319b94df9SMatthew Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 223419b94df9SMatthew Ahrens written); 223519b94df9SMatthew Ahrens } 223619b94df9SMatthew Ahrens } 223719b94df9SMatthew Ahrens } 223819b94df9SMatthew Ahrens 2239187d6ac0SMatt Ahrens ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 : 2240187d6ac0SMatt Ahrens (ds->ds_phys->ds_uncompressed_bytes * 100 / 2241187d6ac0SMatt Ahrens ds->ds_phys->ds_compressed_bytes); 2242187d6ac0SMatt Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 2243187d6ac0SMatt Ahrens 2244fa9e4066Sahrens if (ds->ds_phys->ds_next_snap_obj) { 2245fa9e4066Sahrens /* 2246fa9e4066Sahrens * This is a snapshot; override the dd's space used with 2247a2eea2e1Sahrens * our unique space and compression ratio. 2248fa9e4066Sahrens */ 2249a2eea2e1Sahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2250a2eea2e1Sahrens ds->ds_phys->ds_unique_bytes); 2251187d6ac0SMatt Ahrens dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 225219b94df9SMatthew Ahrens 225319b94df9SMatthew Ahrens get_clones_stat(ds, nv); 2254fa9e4066Sahrens } 2255fa9e4066Sahrens } 2256fa9e4066Sahrens 2257a2eea2e1Sahrens void 2258a2eea2e1Sahrens dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2259a2eea2e1Sahrens { 2260a2eea2e1Sahrens stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; 2261a2eea2e1Sahrens stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; 22623cb34c60Sahrens stat->dds_guid = ds->ds_phys->ds_guid; 2263a2eea2e1Sahrens if (ds->ds_phys->ds_next_snap_obj) { 2264a2eea2e1Sahrens stat->dds_is_snapshot = B_TRUE; 2265a2eea2e1Sahrens stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; 2266ebedde84SEric Taylor } else { 2267ebedde84SEric Taylor stat->dds_is_snapshot = B_FALSE; 2268ebedde84SEric Taylor stat->dds_num_clones = 0; 2269a2eea2e1Sahrens } 2270a2eea2e1Sahrens 2271a2eea2e1Sahrens /* clone origin is really a dsl_dir thing... */ 22724ccbb6e7Sahrens rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); 2273088f3894Sahrens if (dsl_dir_is_clone(ds->ds_dir)) { 2274a2eea2e1Sahrens dsl_dataset_t *ods; 2275a2eea2e1Sahrens 2276745cd3c5Smaybee VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool, 2277745cd3c5Smaybee ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); 22783cb34c60Sahrens dsl_dataset_name(ods, stat->dds_origin); 2279745cd3c5Smaybee dsl_dataset_drop_ref(ods, FTAG); 2280ebedde84SEric Taylor } else { 2281ebedde84SEric Taylor stat->dds_origin[0] = '\0'; 2282a2eea2e1Sahrens } 22834ccbb6e7Sahrens rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); 2284a2eea2e1Sahrens } 2285a2eea2e1Sahrens 2286a2eea2e1Sahrens uint64_t 2287a2eea2e1Sahrens dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2288a2eea2e1Sahrens { 228991ebeef5Sahrens return (ds->ds_fsid_guid); 2290a2eea2e1Sahrens } 2291a2eea2e1Sahrens 2292a2eea2e1Sahrens void 2293a2eea2e1Sahrens dsl_dataset_space(dsl_dataset_t *ds, 2294a2eea2e1Sahrens uint64_t *refdbytesp, uint64_t *availbytesp, 2295a2eea2e1Sahrens uint64_t *usedobjsp, uint64_t *availobjsp) 2296fa9e4066Sahrens { 2297a2eea2e1Sahrens *refdbytesp = ds->ds_phys->ds_used_bytes; 2298a2eea2e1Sahrens *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2299a9799022Sck if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) 2300a9799022Sck *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; 2301a9799022Sck if (ds->ds_quota != 0) { 2302a9799022Sck /* 2303a9799022Sck * Adjust available bytes according to refquota 2304a9799022Sck */ 2305a9799022Sck if (*refdbytesp < ds->ds_quota) 2306a9799022Sck *availbytesp = MIN(*availbytesp, 2307a9799022Sck ds->ds_quota - *refdbytesp); 2308a9799022Sck else 2309a9799022Sck *availbytesp = 0; 2310a9799022Sck } 2311a2eea2e1Sahrens *usedobjsp = ds->ds_phys->ds_bp.blk_fill; 2312a2eea2e1Sahrens *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2313fa9e4066Sahrens } 2314fa9e4066Sahrens 2315f18faf3fSek boolean_t 2316f18faf3fSek dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds) 2317f18faf3fSek { 2318f18faf3fSek dsl_pool_t *dp = ds->ds_dir->dd_pool; 2319f18faf3fSek 2320f18faf3fSek ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 2321f18faf3fSek dsl_pool_sync_context(dp)); 2322f18faf3fSek if (ds->ds_prev == NULL) 2323f18faf3fSek return (B_FALSE); 2324f18faf3fSek if (ds->ds_phys->ds_bp.blk_birth > 23256e0cbcaaSMatthew Ahrens ds->ds_prev->ds_phys->ds_creation_txg) { 23266e0cbcaaSMatthew Ahrens objset_t *os, *os_prev; 23276e0cbcaaSMatthew Ahrens /* 23286e0cbcaaSMatthew Ahrens * It may be that only the ZIL differs, because it was 23296e0cbcaaSMatthew Ahrens * reset in the head. Don't count that as being 23306e0cbcaaSMatthew Ahrens * modified. 23316e0cbcaaSMatthew Ahrens */ 23326e0cbcaaSMatthew Ahrens if (dmu_objset_from_ds(ds, &os) != 0) 23336e0cbcaaSMatthew Ahrens return (B_TRUE); 23346e0cbcaaSMatthew Ahrens if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0) 23356e0cbcaaSMatthew Ahrens return (B_TRUE); 23366e0cbcaaSMatthew Ahrens return (bcmp(&os->os_phys->os_meta_dnode, 23376e0cbcaaSMatthew Ahrens &os_prev->os_phys->os_meta_dnode, 23386e0cbcaaSMatthew Ahrens sizeof (os->os_phys->os_meta_dnode)) != 0); 23396e0cbcaaSMatthew Ahrens } 2340f18faf3fSek return (B_FALSE); 2341f18faf3fSek } 2342f18faf3fSek 23431d452cf5Sahrens /* ARGSUSED */ 2344fa9e4066Sahrens static int 23451d452cf5Sahrens dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) 2346fa9e4066Sahrens { 23471d452cf5Sahrens dsl_dataset_t *ds = arg1; 23481d452cf5Sahrens char *newsnapname = arg2; 23491d452cf5Sahrens dsl_dir_t *dd = ds->ds_dir; 23501d452cf5Sahrens dsl_dataset_t *hds; 2351fa9e4066Sahrens uint64_t val; 23521d452cf5Sahrens int err; 2353fa9e4066Sahrens 2354745cd3c5Smaybee err = dsl_dataset_hold_obj(dd->dd_pool, 2355745cd3c5Smaybee dd->dd_phys->dd_head_dataset_obj, FTAG, &hds); 2356fa9e4066Sahrens if (err) 2357fa9e4066Sahrens return (err); 2358fa9e4066Sahrens 23591d452cf5Sahrens /* new name better not be in use */ 2360745cd3c5Smaybee err = dsl_dataset_snap_lookup(hds, newsnapname, &val); 2361745cd3c5Smaybee dsl_dataset_rele(hds, FTAG); 23621d452cf5Sahrens 23631d452cf5Sahrens if (err == 0) 23641d452cf5Sahrens err = EEXIST; 23651d452cf5Sahrens else if (err == ENOENT) 23661d452cf5Sahrens err = 0; 2367cdf5b4caSmmusante 2368cdf5b4caSmmusante /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2369cdf5b4caSmmusante if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN) 2370cdf5b4caSmmusante err = ENAMETOOLONG; 2371cdf5b4caSmmusante 23721d452cf5Sahrens return (err); 23731d452cf5Sahrens } 2374fa9e4066Sahrens 23751d452cf5Sahrens static void 23763f9d6ad7SLin Ling dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) 23771d452cf5Sahrens { 23781d452cf5Sahrens dsl_dataset_t *ds = arg1; 2379ecd6cf80Smarks const char *newsnapname = arg2; 23801d452cf5Sahrens dsl_dir_t *dd = ds->ds_dir; 23811d452cf5Sahrens objset_t *mos = dd->dd_pool->dp_meta_objset; 23821d452cf5Sahrens dsl_dataset_t *hds; 23831d452cf5Sahrens int err; 2384fa9e4066Sahrens 23851d452cf5Sahrens ASSERT(ds->ds_phys->ds_next_snap_obj != 0); 2386fa9e4066Sahrens 2387745cd3c5Smaybee VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, 2388745cd3c5Smaybee dd->dd_phys->dd_head_dataset_obj, FTAG, &hds)); 2389fa9e4066Sahrens 23901d452cf5Sahrens VERIFY(0 == dsl_dataset_get_snapname(ds)); 2391745cd3c5Smaybee err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx); 2392fa9e4066Sahrens ASSERT3U(err, ==, 0); 23931d452cf5Sahrens mutex_enter(&ds->ds_lock); 23941d452cf5Sahrens (void) strcpy(ds->ds_snapname, newsnapname); 23951d452cf5Sahrens mutex_exit(&ds->ds_lock); 23961d452cf5Sahrens err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, 23971d452cf5Sahrens ds->ds_snapname, 8, 1, &ds->ds_object, tx); 2398fa9e4066Sahrens ASSERT3U(err, ==, 0); 2399fa9e4066Sahrens 24003f9d6ad7SLin Ling spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx, 24013f9d6ad7SLin Ling "dataset = %llu", ds->ds_object); 2402745cd3c5Smaybee dsl_dataset_rele(hds, FTAG); 2403fa9e4066Sahrens } 2404fa9e4066Sahrens 2405f18faf3fSek struct renamesnaparg { 2406cdf5b4caSmmusante dsl_sync_task_group_t *dstg; 2407cdf5b4caSmmusante char failed[MAXPATHLEN]; 2408cdf5b4caSmmusante char *oldsnap; 2409cdf5b4caSmmusante char *newsnap; 2410cdf5b4caSmmusante }; 2411cdf5b4caSmmusante 2412cdf5b4caSmmusante static int 2413fd136879SMatthew Ahrens dsl_snapshot_rename_one(const char *name, void *arg) 2414cdf5b4caSmmusante { 2415f18faf3fSek struct renamesnaparg *ra = arg; 2416cdf5b4caSmmusante dsl_dataset_t *ds = NULL; 2417fd136879SMatthew Ahrens char *snapname; 2418cdf5b4caSmmusante int err; 2419cdf5b4caSmmusante 2420fd136879SMatthew Ahrens snapname = kmem_asprintf("%s@%s", name, ra->oldsnap); 2421fd136879SMatthew Ahrens (void) strlcpy(ra->failed, snapname, sizeof (ra->failed)); 2422ecd6cf80Smarks 2423ecd6cf80Smarks /* 2424ecd6cf80Smarks * For recursive snapshot renames the parent won't be changing 2425ecd6cf80Smarks * so we just pass name for both the to/from argument. 2426ecd6cf80Smarks */ 2427fd136879SMatthew Ahrens err = zfs_secpolicy_rename_perms(snapname, snapname, CRED()); 2428fd136879SMatthew Ahrens if (err != 0) { 2429fd136879SMatthew Ahrens strfree(snapname); 2430fd136879SMatthew Ahrens return (err == ENOENT ? 0 : err); 2431ecd6cf80Smarks } 2432ecd6cf80Smarks 2433745cd3c5Smaybee #ifdef _KERNEL 2434745cd3c5Smaybee /* 2435745cd3c5Smaybee * For all filesystems undergoing rename, we'll need to unmount it. 2436745cd3c5Smaybee */ 2437fd136879SMatthew Ahrens (void) zfs_unmount_snap(snapname, NULL); 2438745cd3c5Smaybee #endif 2439fd136879SMatthew Ahrens err = dsl_dataset_hold(snapname, ra->dstg, &ds); 24403f1f8012SMatthew Ahrens strfree(snapname); 24413f1f8012SMatthew Ahrens if (err != 0) 2442fd136879SMatthew Ahrens return (err == ENOENT ? 0 : err); 2443cdf5b4caSmmusante 2444cdf5b4caSmmusante dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check, 2445cdf5b4caSmmusante dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0); 2446cdf5b4caSmmusante 2447cdf5b4caSmmusante return (0); 2448cdf5b4caSmmusante } 2449cdf5b4caSmmusante 2450cdf5b4caSmmusante static int 2451cdf5b4caSmmusante dsl_recursive_rename(char *oldname, const char *newname) 2452cdf5b4caSmmusante { 2453cdf5b4caSmmusante int err; 2454f18faf3fSek struct renamesnaparg *ra; 2455cdf5b4caSmmusante dsl_sync_task_t *dst; 2456cdf5b4caSmmusante spa_t *spa; 2457cdf5b4caSmmusante char *cp, *fsname = spa_strdup(oldname); 2458fd136879SMatthew Ahrens int len = strlen(oldname) + 1; 2459cdf5b4caSmmusante 2460cdf5b4caSmmusante /* truncate the snapshot name to get the fsname */ 2461cdf5b4caSmmusante cp = strchr(fsname, '@'); 2462cdf5b4caSmmusante *cp = '\0'; 2463cdf5b4caSmmusante 246440feaa91Sahrens err = spa_open(fsname, &spa, FTAG); 2465cdf5b4caSmmusante if (err) { 2466fd136879SMatthew Ahrens kmem_free(fsname, len); 2467cdf5b4caSmmusante return (err); 2468cdf5b4caSmmusante } 2469f18faf3fSek ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP); 2470cdf5b4caSmmusante ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 2471cdf5b4caSmmusante 2472cdf5b4caSmmusante ra->oldsnap = strchr(oldname, '@') + 1; 2473cdf5b4caSmmusante ra->newsnap = strchr(newname, '@') + 1; 2474cdf5b4caSmmusante *ra->failed = '\0'; 2475cdf5b4caSmmusante 2476cdf5b4caSmmusante err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra, 2477cdf5b4caSmmusante DS_FIND_CHILDREN); 2478fd136879SMatthew Ahrens kmem_free(fsname, len); 2479cdf5b4caSmmusante 2480cdf5b4caSmmusante if (err == 0) { 2481cdf5b4caSmmusante err = dsl_sync_task_group_wait(ra->dstg); 2482cdf5b4caSmmusante } 2483cdf5b4caSmmusante 2484cdf5b4caSmmusante for (dst = list_head(&ra->dstg->dstg_tasks); dst; 2485cdf5b4caSmmusante dst = list_next(&ra->dstg->dstg_tasks, dst)) { 2486cdf5b4caSmmusante dsl_dataset_t *ds = dst->dst_arg1; 2487cdf5b4caSmmusante if (dst->dst_err) { 2488cdf5b4caSmmusante dsl_dir_name(ds->ds_dir, ra->failed); 2489fd136879SMatthew Ahrens (void) strlcat(ra->failed, "@", sizeof (ra->failed)); 2490fd136879SMatthew Ahrens (void) strlcat(ra->failed, ra->newsnap, 2491fd136879SMatthew Ahrens sizeof (ra->failed)); 2492cdf5b4caSmmusante } 2493745cd3c5Smaybee dsl_dataset_rele(ds, ra->dstg); 2494cdf5b4caSmmusante } 2495cdf5b4caSmmusante 2496ecd6cf80Smarks if (err) 2497fd136879SMatthew Ahrens (void) strlcpy(oldname, ra->failed, sizeof (ra->failed)); 2498cdf5b4caSmmusante 2499cdf5b4caSmmusante dsl_sync_task_group_destroy(ra->dstg); 2500f18faf3fSek kmem_free(ra, sizeof (struct renamesnaparg)); 2501cdf5b4caSmmusante spa_close(spa, FTAG); 2502cdf5b4caSmmusante return (err); 2503cdf5b4caSmmusante } 2504cdf5b4caSmmusante 25053a5a36beSmmusante static int 2506fd136879SMatthew Ahrens dsl_valid_rename(const char *oldname, void *arg) 25073a5a36beSmmusante { 25083a5a36beSmmusante int delta = *(int *)arg; 25093a5a36beSmmusante 25103a5a36beSmmusante if (strlen(oldname) + delta >= MAXNAMELEN) 25113a5a36beSmmusante return (ENAMETOOLONG); 25123a5a36beSmmusante 25133a5a36beSmmusante return (0); 25143a5a36beSmmusante } 25153a5a36beSmmusante 2516fa9e4066Sahrens #pragma weak dmu_objset_rename = dsl_dataset_rename 2517fa9e4066Sahrens int 2518745cd3c5Smaybee dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive) 2519fa9e4066Sahrens { 2520fa9e4066Sahrens dsl_dir_t *dd; 25211d452cf5Sahrens dsl_dataset_t *ds; 2522fa9e4066Sahrens const char *tail; 2523fa9e4066Sahrens int err; 2524fa9e4066Sahrens 25251d452cf5Sahrens err = dsl_dir_open(oldname, FTAG, &dd, &tail); 2526ea8dc4b6Seschrock if (err) 2527ea8dc4b6Seschrock return (err); 2528370c1af0SSanjeev Bagewadi 2529fa9e4066Sahrens if (tail == NULL) { 25303a5a36beSmmusante int delta = strlen(newname) - strlen(oldname); 25313a5a36beSmmusante 2532088f3894Sahrens /* if we're growing, validate child name lengths */ 25333a5a36beSmmusante if (delta > 0) 25343a5a36beSmmusante err = dmu_objset_find(oldname, dsl_valid_rename, 25353a5a36beSmmusante &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 25363a5a36beSmmusante 2537b91a2f0bSMatthew Ahrens if (err == 0) 25383a5a36beSmmusante err = dsl_dir_rename(dd, newname); 2539fa9e4066Sahrens dsl_dir_close(dd, FTAG); 2540fa9e4066Sahrens return (err); 2541fa9e4066Sahrens } 2542370c1af0SSanjeev Bagewadi 2543fa9e4066Sahrens if (tail[0] != '@') { 2544681d9761SEric Taylor /* the name ended in a nonexistent component */ 2545fa9e4066Sahrens dsl_dir_close(dd, FTAG); 2546fa9e4066Sahrens return (ENOENT); 2547fa9e4066Sahrens } 2548fa9e4066Sahrens 2549fa9e4066Sahrens dsl_dir_close(dd, FTAG); 25501d452cf5Sahrens 25511d452cf5Sahrens /* new name must be snapshot in same filesystem */ 25521d452cf5Sahrens tail = strchr(newname, '@'); 25531d452cf5Sahrens if (tail == NULL) 25541d452cf5Sahrens return (EINVAL); 25551d452cf5Sahrens tail++; 25561d452cf5Sahrens if (strncmp(oldname, newname, tail - newname) != 0) 25571d452cf5Sahrens return (EXDEV); 25581d452cf5Sahrens 2559cdf5b4caSmmusante if (recursive) { 2560cdf5b4caSmmusante err = dsl_recursive_rename(oldname, newname); 2561cdf5b4caSmmusante } else { 2562745cd3c5Smaybee err = dsl_dataset_hold(oldname, FTAG, &ds); 2563cdf5b4caSmmusante if (err) 2564cdf5b4caSmmusante return (err); 25651d452cf5Sahrens 2566cdf5b4caSmmusante err = dsl_sync_task_do(ds->ds_dir->dd_pool, 2567cdf5b4caSmmusante dsl_dataset_snapshot_rename_check, 2568cdf5b4caSmmusante dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); 25691d452cf5Sahrens 2570745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 2571cdf5b4caSmmusante } 25721d452cf5Sahrens 2573fa9e4066Sahrens return (err); 2574fa9e4066Sahrens } 257599653d4eSeschrock 2576088f3894Sahrens struct promotenode { 2577745cd3c5Smaybee list_node_t link; 2578745cd3c5Smaybee dsl_dataset_t *ds; 2579745cd3c5Smaybee }; 2580745cd3c5Smaybee 25811d452cf5Sahrens struct promotearg { 258274e7dc98SMatthew Ahrens list_t shared_snaps, origin_snaps, clone_snaps; 25833f9d6ad7SLin Ling dsl_dataset_t *origin_origin; 258474e7dc98SMatthew Ahrens uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2585681d9761SEric Taylor char *err_ds; 25861d452cf5Sahrens }; 25871d452cf5Sahrens 258874e7dc98SMatthew Ahrens static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 25893f9d6ad7SLin Ling static boolean_t snaplist_unstable(list_t *l); 259074e7dc98SMatthew Ahrens 259199653d4eSeschrock static int 25921d452cf5Sahrens dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) 259399653d4eSeschrock { 25941d452cf5Sahrens dsl_dataset_t *hds = arg1; 25951d452cf5Sahrens struct promotearg *pa = arg2; 259674e7dc98SMatthew Ahrens struct promotenode *snap = list_head(&pa->shared_snaps); 2597745cd3c5Smaybee dsl_dataset_t *origin_ds = snap->ds; 2598745cd3c5Smaybee int err; 2599cde58dbcSMatthew Ahrens uint64_t unused; 26001d452cf5Sahrens 2601088f3894Sahrens /* Check that it is a real clone */ 2602088f3894Sahrens if (!dsl_dir_is_clone(hds->ds_dir)) 260399653d4eSeschrock return (EINVAL); 260499653d4eSeschrock 26051d452cf5Sahrens /* Since this is so expensive, don't do the preliminary check */ 26061d452cf5Sahrens if (!dmu_tx_is_syncing(tx)) 26071d452cf5Sahrens return (0); 26081d452cf5Sahrens 2609745cd3c5Smaybee if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) 2610745cd3c5Smaybee return (EXDEV); 261199653d4eSeschrock 26123cb34c60Sahrens /* compute origin's new unique space */ 261374e7dc98SMatthew Ahrens snap = list_tail(&pa->clone_snaps); 261474e7dc98SMatthew Ahrens ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2615cde58dbcSMatthew Ahrens dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2616cde58dbcSMatthew Ahrens origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2617cde58dbcSMatthew Ahrens &pa->unique, &unused, &unused); 261899653d4eSeschrock 2619745cd3c5Smaybee /* 2620745cd3c5Smaybee * Walk the snapshots that we are moving 2621745cd3c5Smaybee * 262274e7dc98SMatthew Ahrens * Compute space to transfer. Consider the incremental changes 262374e7dc98SMatthew Ahrens * to used for each snapshot: 262474e7dc98SMatthew Ahrens * (my used) = (prev's used) + (blocks born) - (blocks killed) 262574e7dc98SMatthew Ahrens * So each snapshot gave birth to: 262674e7dc98SMatthew Ahrens * (blocks born) = (my used) - (prev's used) + (blocks killed) 2627745cd3c5Smaybee * So a sequence would look like: 262874e7dc98SMatthew Ahrens * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2629745cd3c5Smaybee * Which simplifies to: 263074e7dc98SMatthew Ahrens * uN + kN + kN-1 + ... + k1 + k0 2631745cd3c5Smaybee * Note however, if we stop before we reach the ORIGIN we get: 263274e7dc98SMatthew Ahrens * uN + kN + kN-1 + ... + kM - uM-1 2633745cd3c5Smaybee */ 2634745cd3c5Smaybee pa->used = origin_ds->ds_phys->ds_used_bytes; 2635745cd3c5Smaybee pa->comp = origin_ds->ds_phys->ds_compressed_bytes; 2636745cd3c5Smaybee pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; 263774e7dc98SMatthew Ahrens for (snap = list_head(&pa->shared_snaps); snap; 263874e7dc98SMatthew Ahrens snap = list_next(&pa->shared_snaps, snap)) { 263999653d4eSeschrock uint64_t val, dlused, dlcomp, dluncomp; 2640745cd3c5Smaybee dsl_dataset_t *ds = snap->ds; 264199653d4eSeschrock 264299653d4eSeschrock /* Check that the snapshot name does not conflict */ 264374e7dc98SMatthew Ahrens VERIFY(0 == dsl_dataset_get_snapname(ds)); 2644745cd3c5Smaybee err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2645681d9761SEric Taylor if (err == 0) { 2646681d9761SEric Taylor err = EEXIST; 2647681d9761SEric Taylor goto out; 2648681d9761SEric Taylor } 2649745cd3c5Smaybee if (err != ENOENT) 2650681d9761SEric Taylor goto out; 265199653d4eSeschrock 2652745cd3c5Smaybee /* The very first snapshot does not have a deadlist */ 265374e7dc98SMatthew Ahrens if (ds->ds_phys->ds_prev_snap_obj == 0) 265474e7dc98SMatthew Ahrens continue; 265574e7dc98SMatthew Ahrens 2656cde58dbcSMatthew Ahrens dsl_deadlist_space(&ds->ds_deadlist, 2657cde58dbcSMatthew Ahrens &dlused, &dlcomp, &dluncomp); 265874e7dc98SMatthew Ahrens pa->used += dlused; 265974e7dc98SMatthew Ahrens pa->comp += dlcomp; 266074e7dc98SMatthew Ahrens pa->uncomp += dluncomp; 266174e7dc98SMatthew Ahrens } 2662745cd3c5Smaybee 2663745cd3c5Smaybee /* 2664745cd3c5Smaybee * If we are a clone of a clone then we never reached ORIGIN, 2665745cd3c5Smaybee * so we need to subtract out the clone origin's used space. 2666745cd3c5Smaybee */ 266774e7dc98SMatthew Ahrens if (pa->origin_origin) { 266874e7dc98SMatthew Ahrens pa->used -= pa->origin_origin->ds_phys->ds_used_bytes; 266974e7dc98SMatthew Ahrens pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes; 267074e7dc98SMatthew Ahrens pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes; 267199653d4eSeschrock } 267299653d4eSeschrock 267399653d4eSeschrock /* Check that there is enough space here */ 267474e7dc98SMatthew Ahrens err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 267574e7dc98SMatthew Ahrens pa->used); 267674e7dc98SMatthew Ahrens if (err) 267774e7dc98SMatthew Ahrens return (err); 267874e7dc98SMatthew Ahrens 267974e7dc98SMatthew Ahrens /* 268074e7dc98SMatthew Ahrens * Compute the amounts of space that will be used by snapshots 268174e7dc98SMatthew Ahrens * after the promotion (for both origin and clone). For each, 268274e7dc98SMatthew Ahrens * it is the amount of space that will be on all of their 268374e7dc98SMatthew Ahrens * deadlists (that was not born before their new origin). 268474e7dc98SMatthew Ahrens */ 268574e7dc98SMatthew Ahrens if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 268674e7dc98SMatthew Ahrens uint64_t space; 268774e7dc98SMatthew Ahrens 268874e7dc98SMatthew Ahrens /* 268974e7dc98SMatthew Ahrens * Note, typically this will not be a clone of a clone, 26903f9d6ad7SLin Ling * so dd_origin_txg will be < TXG_INITIAL, so 2691cde58dbcSMatthew Ahrens * these snaplist_space() -> dsl_deadlist_space_range() 269274e7dc98SMatthew Ahrens * calls will be fast because they do not have to 269374e7dc98SMatthew Ahrens * iterate over all bps. 269474e7dc98SMatthew Ahrens */ 269574e7dc98SMatthew Ahrens snap = list_head(&pa->origin_snaps); 269674e7dc98SMatthew Ahrens err = snaplist_space(&pa->shared_snaps, 26973f9d6ad7SLin Ling snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap); 269874e7dc98SMatthew Ahrens if (err) 269974e7dc98SMatthew Ahrens return (err); 270074e7dc98SMatthew Ahrens 270174e7dc98SMatthew Ahrens err = snaplist_space(&pa->clone_snaps, 27023f9d6ad7SLin Ling snap->ds->ds_dir->dd_origin_txg, &space); 270374e7dc98SMatthew Ahrens if (err) 270474e7dc98SMatthew Ahrens return (err); 270574e7dc98SMatthew Ahrens pa->cloneusedsnap += space; 270674e7dc98SMatthew Ahrens } 270774e7dc98SMatthew Ahrens if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 270874e7dc98SMatthew Ahrens err = snaplist_space(&pa->origin_snaps, 270974e7dc98SMatthew Ahrens origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap); 271074e7dc98SMatthew Ahrens if (err) 271174e7dc98SMatthew Ahrens return (err); 2712745cd3c5Smaybee } 27131d452cf5Sahrens 271474e7dc98SMatthew Ahrens return (0); 2715681d9761SEric Taylor out: 2716681d9761SEric Taylor pa->err_ds = snap->ds->ds_snapname; 2717681d9761SEric Taylor return (err); 27181d452cf5Sahrens } 271999653d4eSeschrock 27201d452cf5Sahrens static void 27213f9d6ad7SLin Ling dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) 27221d452cf5Sahrens { 27231d452cf5Sahrens dsl_dataset_t *hds = arg1; 27241d452cf5Sahrens struct promotearg *pa = arg2; 272574e7dc98SMatthew Ahrens struct promotenode *snap = list_head(&pa->shared_snaps); 2726745cd3c5Smaybee dsl_dataset_t *origin_ds = snap->ds; 272774e7dc98SMatthew Ahrens dsl_dataset_t *origin_head; 27281d452cf5Sahrens dsl_dir_t *dd = hds->ds_dir; 27291d452cf5Sahrens dsl_pool_t *dp = hds->ds_dir->dd_pool; 27303cb34c60Sahrens dsl_dir_t *odd = NULL; 2731088f3894Sahrens uint64_t oldnext_obj; 273274e7dc98SMatthew Ahrens int64_t delta; 27331d452cf5Sahrens 27341d452cf5Sahrens ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); 27351d452cf5Sahrens 273674e7dc98SMatthew Ahrens snap = list_head(&pa->origin_snaps); 273774e7dc98SMatthew Ahrens origin_head = snap->ds; 273874e7dc98SMatthew Ahrens 27390b69c2f0Sahrens /* 27403cb34c60Sahrens * We need to explicitly open odd, since origin_ds's dd will be 27410b69c2f0Sahrens * changing. 27420b69c2f0Sahrens */ 27433cb34c60Sahrens VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object, 27443cb34c60Sahrens NULL, FTAG, &odd)); 274599653d4eSeschrock 2746745cd3c5Smaybee /* change origin's next snap */ 2747745cd3c5Smaybee dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2748088f3894Sahrens oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; 274974e7dc98SMatthew Ahrens snap = list_tail(&pa->clone_snaps); 275074e7dc98SMatthew Ahrens ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 275174e7dc98SMatthew Ahrens origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; 2752745cd3c5Smaybee 2753088f3894Sahrens /* change the origin's next clone */ 2754088f3894Sahrens if (origin_ds->ds_phys->ds_next_clones_obj) { 2755c33e334fSMatthew Ahrens remove_from_next_clones(origin_ds, snap->ds->ds_object, tx); 2756088f3894Sahrens VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 2757088f3894Sahrens origin_ds->ds_phys->ds_next_clones_obj, 2758088f3894Sahrens oldnext_obj, tx)); 2759088f3894Sahrens } 2760088f3894Sahrens 2761745cd3c5Smaybee /* change origin */ 2762745cd3c5Smaybee dmu_buf_will_dirty(dd->dd_dbuf, tx); 2763745cd3c5Smaybee ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); 2764745cd3c5Smaybee dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; 27653f9d6ad7SLin Ling dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2766745cd3c5Smaybee dmu_buf_will_dirty(odd->dd_dbuf, tx); 2767745cd3c5Smaybee odd->dd_phys->dd_origin_obj = origin_ds->ds_object; 27683f9d6ad7SLin Ling origin_head->ds_dir->dd_origin_txg = 27693f9d6ad7SLin Ling origin_ds->ds_phys->ds_creation_txg; 2770745cd3c5Smaybee 2771cde58dbcSMatthew Ahrens /* change dd_clone entries */ 2772cde58dbcSMatthew Ahrens if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2773cde58dbcSMatthew Ahrens VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, 2774cde58dbcSMatthew Ahrens odd->dd_phys->dd_clones, hds->ds_object, tx)); 2775cde58dbcSMatthew Ahrens VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 2776cde58dbcSMatthew Ahrens pa->origin_origin->ds_dir->dd_phys->dd_clones, 2777cde58dbcSMatthew Ahrens hds->ds_object, tx)); 2778cde58dbcSMatthew Ahrens 2779cde58dbcSMatthew Ahrens VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, 2780cde58dbcSMatthew Ahrens pa->origin_origin->ds_dir->dd_phys->dd_clones, 2781cde58dbcSMatthew Ahrens origin_head->ds_object, tx)); 2782cde58dbcSMatthew Ahrens if (dd->dd_phys->dd_clones == 0) { 2783cde58dbcSMatthew Ahrens dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset, 2784cde58dbcSMatthew Ahrens DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 2785cde58dbcSMatthew Ahrens } 2786cde58dbcSMatthew Ahrens VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 2787cde58dbcSMatthew Ahrens dd->dd_phys->dd_clones, origin_head->ds_object, tx)); 2788cde58dbcSMatthew Ahrens 2789cde58dbcSMatthew Ahrens } 2790cde58dbcSMatthew Ahrens 279199653d4eSeschrock /* move snapshots to this dir */ 279274e7dc98SMatthew Ahrens for (snap = list_head(&pa->shared_snaps); snap; 279374e7dc98SMatthew Ahrens snap = list_next(&pa->shared_snaps, snap)) { 2794745cd3c5Smaybee dsl_dataset_t *ds = snap->ds; 279599653d4eSeschrock 27963baa08fcSek /* unregister props as dsl_dir is changing */ 2797503ad85cSMatthew Ahrens if (ds->ds_objset) { 2798503ad85cSMatthew Ahrens dmu_objset_evict(ds->ds_objset); 2799503ad85cSMatthew Ahrens ds->ds_objset = NULL; 28003baa08fcSek } 280199653d4eSeschrock /* move snap name entry */ 280274e7dc98SMatthew Ahrens VERIFY(0 == dsl_dataset_get_snapname(ds)); 280374e7dc98SMatthew Ahrens VERIFY(0 == dsl_dataset_snap_remove(origin_head, 2804745cd3c5Smaybee ds->ds_snapname, tx)); 28051d452cf5Sahrens VERIFY(0 == zap_add(dp->dp_meta_objset, 280699653d4eSeschrock hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 280799653d4eSeschrock 8, 1, &ds->ds_object, tx)); 2808cde58dbcSMatthew Ahrens 280999653d4eSeschrock /* change containing dsl_dir */ 281099653d4eSeschrock dmu_buf_will_dirty(ds->ds_dbuf, tx); 28113cb34c60Sahrens ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); 281299653d4eSeschrock ds->ds_phys->ds_dir_obj = dd->dd_object; 28133cb34c60Sahrens ASSERT3P(ds->ds_dir, ==, odd); 281499653d4eSeschrock dsl_dir_close(ds->ds_dir, ds); 28151d452cf5Sahrens VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, 281699653d4eSeschrock NULL, ds, &ds->ds_dir)); 281799653d4eSeschrock 2818cde58dbcSMatthew Ahrens /* move any clone references */ 2819cde58dbcSMatthew Ahrens if (ds->ds_phys->ds_next_clones_obj && 2820cde58dbcSMatthew Ahrens spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2821cde58dbcSMatthew Ahrens zap_cursor_t zc; 2822cde58dbcSMatthew Ahrens zap_attribute_t za; 2823cde58dbcSMatthew Ahrens 2824cde58dbcSMatthew Ahrens for (zap_cursor_init(&zc, dp->dp_meta_objset, 2825cde58dbcSMatthew Ahrens ds->ds_phys->ds_next_clones_obj); 2826cde58dbcSMatthew Ahrens zap_cursor_retrieve(&zc, &za) == 0; 2827cde58dbcSMatthew Ahrens zap_cursor_advance(&zc)) { 2828cde58dbcSMatthew Ahrens dsl_dataset_t *cnds; 2829cde58dbcSMatthew Ahrens uint64_t o; 2830cde58dbcSMatthew Ahrens 2831cde58dbcSMatthew Ahrens if (za.za_first_integer == oldnext_obj) { 2832cde58dbcSMatthew Ahrens /* 2833cde58dbcSMatthew Ahrens * We've already moved the 2834cde58dbcSMatthew Ahrens * origin's reference. 2835cde58dbcSMatthew Ahrens */ 2836cde58dbcSMatthew Ahrens continue; 2837cde58dbcSMatthew Ahrens } 2838cde58dbcSMatthew Ahrens 2839cde58dbcSMatthew Ahrens VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, 2840cde58dbcSMatthew Ahrens za.za_first_integer, FTAG, &cnds)); 2841cde58dbcSMatthew Ahrens o = cnds->ds_dir->dd_phys->dd_head_dataset_obj; 2842cde58dbcSMatthew Ahrens 2843cde58dbcSMatthew Ahrens VERIFY3U(zap_remove_int(dp->dp_meta_objset, 2844cde58dbcSMatthew Ahrens odd->dd_phys->dd_clones, o, tx), ==, 0); 2845cde58dbcSMatthew Ahrens VERIFY3U(zap_add_int(dp->dp_meta_objset, 2846cde58dbcSMatthew Ahrens dd->dd_phys->dd_clones, o, tx), ==, 0); 2847cde58dbcSMatthew Ahrens dsl_dataset_rele(cnds, FTAG); 2848cde58dbcSMatthew Ahrens } 2849cde58dbcSMatthew Ahrens zap_cursor_fini(&zc); 2850cde58dbcSMatthew Ahrens } 2851cde58dbcSMatthew Ahrens 285299653d4eSeschrock ASSERT3U(dsl_prop_numcb(ds), ==, 0); 285374e7dc98SMatthew Ahrens } 285474e7dc98SMatthew Ahrens 285574e7dc98SMatthew Ahrens /* 285674e7dc98SMatthew Ahrens * Change space accounting. 285774e7dc98SMatthew Ahrens * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 285874e7dc98SMatthew Ahrens * both be valid, or both be 0 (resulting in delta == 0). This 285974e7dc98SMatthew Ahrens * is true for each of {clone,origin} independently. 286074e7dc98SMatthew Ahrens */ 286174e7dc98SMatthew Ahrens 286274e7dc98SMatthew Ahrens delta = pa->cloneusedsnap - 286374e7dc98SMatthew Ahrens dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 286474e7dc98SMatthew Ahrens ASSERT3S(delta, >=, 0); 286574e7dc98SMatthew Ahrens ASSERT3U(pa->used, >=, delta); 286674e7dc98SMatthew Ahrens dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 286774e7dc98SMatthew Ahrens dsl_dir_diduse_space(dd, DD_USED_HEAD, 286874e7dc98SMatthew Ahrens pa->used - delta, pa->comp, pa->uncomp, tx); 286974e7dc98SMatthew Ahrens 287074e7dc98SMatthew Ahrens delta = pa->originusedsnap - 287174e7dc98SMatthew Ahrens odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 287274e7dc98SMatthew Ahrens ASSERT3S(delta, <=, 0); 287374e7dc98SMatthew Ahrens ASSERT3U(pa->used, >=, -delta); 287474e7dc98SMatthew Ahrens dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 287574e7dc98SMatthew Ahrens dsl_dir_diduse_space(odd, DD_USED_HEAD, 287674e7dc98SMatthew Ahrens -pa->used - delta, -pa->comp, -pa->uncomp, tx); 287799653d4eSeschrock 28783cb34c60Sahrens origin_ds->ds_phys->ds_unique_bytes = pa->unique; 287999653d4eSeschrock 2880ecd6cf80Smarks /* log history record */ 28813f9d6ad7SLin Ling spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx, 28823f9d6ad7SLin Ling "dataset = %llu", hds->ds_object); 2883ecd6cf80Smarks 28843cb34c60Sahrens dsl_dir_close(odd, FTAG); 288599653d4eSeschrock } 288699653d4eSeschrock 288774e7dc98SMatthew Ahrens static char *snaplist_tag = "snaplist"; 288874e7dc98SMatthew Ahrens /* 288974e7dc98SMatthew Ahrens * Make a list of dsl_dataset_t's for the snapshots between first_obj 289074e7dc98SMatthew Ahrens * (exclusive) and last_obj (inclusive). The list will be in reverse 289174e7dc98SMatthew Ahrens * order (last_obj will be the list_head()). If first_obj == 0, do all 289274e7dc98SMatthew Ahrens * snapshots back to this dataset's origin. 289374e7dc98SMatthew Ahrens */ 289474e7dc98SMatthew Ahrens static int 289574e7dc98SMatthew Ahrens snaplist_make(dsl_pool_t *dp, boolean_t own, 289674e7dc98SMatthew Ahrens uint64_t first_obj, uint64_t last_obj, list_t *l) 289774e7dc98SMatthew Ahrens { 289874e7dc98SMatthew Ahrens uint64_t obj = last_obj; 289974e7dc98SMatthew Ahrens 290074e7dc98SMatthew Ahrens ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock)); 290174e7dc98SMatthew Ahrens 290274e7dc98SMatthew Ahrens list_create(l, sizeof (struct promotenode), 290374e7dc98SMatthew Ahrens offsetof(struct promotenode, link)); 290474e7dc98SMatthew Ahrens 290574e7dc98SMatthew Ahrens while (obj != first_obj) { 290674e7dc98SMatthew Ahrens dsl_dataset_t *ds; 290774e7dc98SMatthew Ahrens struct promotenode *snap; 290874e7dc98SMatthew Ahrens int err; 290974e7dc98SMatthew Ahrens 291074e7dc98SMatthew Ahrens if (own) { 291174e7dc98SMatthew Ahrens err = dsl_dataset_own_obj(dp, obj, 291274e7dc98SMatthew Ahrens 0, snaplist_tag, &ds); 291374e7dc98SMatthew Ahrens if (err == 0) 291474e7dc98SMatthew Ahrens dsl_dataset_make_exclusive(ds, snaplist_tag); 291574e7dc98SMatthew Ahrens } else { 291674e7dc98SMatthew Ahrens err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds); 291774e7dc98SMatthew Ahrens } 291874e7dc98SMatthew Ahrens if (err == ENOENT) { 291974e7dc98SMatthew Ahrens /* lost race with snapshot destroy */ 292074e7dc98SMatthew Ahrens struct promotenode *last = list_tail(l); 292174e7dc98SMatthew Ahrens ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj); 292274e7dc98SMatthew Ahrens obj = last->ds->ds_phys->ds_prev_snap_obj; 292374e7dc98SMatthew Ahrens continue; 292474e7dc98SMatthew Ahrens } else if (err) { 292574e7dc98SMatthew Ahrens return (err); 292674e7dc98SMatthew Ahrens } 292774e7dc98SMatthew Ahrens 292874e7dc98SMatthew Ahrens if (first_obj == 0) 292974e7dc98SMatthew Ahrens first_obj = ds->ds_dir->dd_phys->dd_origin_obj; 293074e7dc98SMatthew Ahrens 293174e7dc98SMatthew Ahrens snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP); 293274e7dc98SMatthew Ahrens snap->ds = ds; 293374e7dc98SMatthew Ahrens list_insert_tail(l, snap); 293474e7dc98SMatthew Ahrens obj = ds->ds_phys->ds_prev_snap_obj; 293574e7dc98SMatthew Ahrens } 293674e7dc98SMatthew Ahrens 293774e7dc98SMatthew Ahrens return (0); 293874e7dc98SMatthew Ahrens } 293974e7dc98SMatthew Ahrens 294074e7dc98SMatthew Ahrens static int 294174e7dc98SMatthew Ahrens snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 294274e7dc98SMatthew Ahrens { 294374e7dc98SMatthew Ahrens struct promotenode *snap; 294474e7dc98SMatthew Ahrens 294574e7dc98SMatthew Ahrens *spacep = 0; 294674e7dc98SMatthew Ahrens for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2947cde58dbcSMatthew Ahrens uint64_t used, comp, uncomp; 2948cde58dbcSMatthew Ahrens dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2949cde58dbcSMatthew Ahrens mintxg, UINT64_MAX, &used, &comp, &uncomp); 295074e7dc98SMatthew Ahrens *spacep += used; 295174e7dc98SMatthew Ahrens } 295274e7dc98SMatthew Ahrens return (0); 295374e7dc98SMatthew Ahrens } 295474e7dc98SMatthew Ahrens 295574e7dc98SMatthew Ahrens static void 295674e7dc98SMatthew Ahrens snaplist_destroy(list_t *l, boolean_t own) 295774e7dc98SMatthew Ahrens { 295874e7dc98SMatthew Ahrens struct promotenode *snap; 295974e7dc98SMatthew Ahrens 29604f5064b7SMark J Musante if (!l || !list_link_active(&l->list_head)) 296174e7dc98SMatthew Ahrens return; 296274e7dc98SMatthew Ahrens 296374e7dc98SMatthew Ahrens while ((snap = list_tail(l)) != NULL) { 296474e7dc98SMatthew Ahrens list_remove(l, snap); 296574e7dc98SMatthew Ahrens if (own) 296674e7dc98SMatthew Ahrens dsl_dataset_disown(snap->ds, snaplist_tag); 296774e7dc98SMatthew Ahrens else 296874e7dc98SMatthew Ahrens dsl_dataset_rele(snap->ds, snaplist_tag); 296974e7dc98SMatthew Ahrens kmem_free(snap, sizeof (struct promotenode)); 297074e7dc98SMatthew Ahrens } 297174e7dc98SMatthew Ahrens list_destroy(l); 297274e7dc98SMatthew Ahrens } 297374e7dc98SMatthew Ahrens 297474e7dc98SMatthew Ahrens /* 297574e7dc98SMatthew Ahrens * Promote a clone. Nomenclature note: 297674e7dc98SMatthew Ahrens * "clone" or "cds": the original clone which is being promoted 297774e7dc98SMatthew Ahrens * "origin" or "ods": the snapshot which is originally clone's origin 297874e7dc98SMatthew Ahrens * "origin head" or "ohds": the dataset which is the head 297974e7dc98SMatthew Ahrens * (filesystem/volume) for the origin 298074e7dc98SMatthew Ahrens * "origin origin": the origin of the origin's filesystem (typically 298174e7dc98SMatthew Ahrens * NULL, indicating that the clone is not a clone of a clone). 298274e7dc98SMatthew Ahrens */ 298399653d4eSeschrock int 2984681d9761SEric Taylor dsl_dataset_promote(const char *name, char *conflsnap) 298599653d4eSeschrock { 298699653d4eSeschrock dsl_dataset_t *ds; 2987745cd3c5Smaybee dsl_dir_t *dd; 2988745cd3c5Smaybee dsl_pool_t *dp; 298999653d4eSeschrock dmu_object_info_t doi; 299074e7dc98SMatthew Ahrens struct promotearg pa = { 0 }; 2991088f3894Sahrens struct promotenode *snap; 2992745cd3c5Smaybee int err; 299399653d4eSeschrock 2994745cd3c5Smaybee err = dsl_dataset_hold(name, FTAG, &ds); 299599653d4eSeschrock if (err) 299699653d4eSeschrock return (err); 2997745cd3c5Smaybee dd = ds->ds_dir; 2998745cd3c5Smaybee dp = dd->dd_pool; 299999653d4eSeschrock 3000745cd3c5Smaybee err = dmu_object_info(dp->dp_meta_objset, 300199653d4eSeschrock ds->ds_phys->ds_snapnames_zapobj, &doi); 300299653d4eSeschrock if (err) { 3003745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 300499653d4eSeschrock return (err); 300599653d4eSeschrock } 300699653d4eSeschrock 300774e7dc98SMatthew Ahrens if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) { 300874e7dc98SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 300974e7dc98SMatthew Ahrens return (EINVAL); 301074e7dc98SMatthew Ahrens } 301174e7dc98SMatthew Ahrens 3012745cd3c5Smaybee /* 3013745cd3c5Smaybee * We are going to inherit all the snapshots taken before our 3014745cd3c5Smaybee * origin (i.e., our new origin will be our parent's origin). 3015745cd3c5Smaybee * Take ownership of them so that we can rename them into our 3016745cd3c5Smaybee * namespace. 3017745cd3c5Smaybee */ 3018745cd3c5Smaybee rw_enter(&dp->dp_config_rwlock, RW_READER); 3019088f3894Sahrens 302074e7dc98SMatthew Ahrens err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj, 302174e7dc98SMatthew Ahrens &pa.shared_snaps); 302274e7dc98SMatthew Ahrens if (err != 0) 302374e7dc98SMatthew Ahrens goto out; 3024088f3894Sahrens 302574e7dc98SMatthew Ahrens err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps); 302674e7dc98SMatthew Ahrens if (err != 0) 302774e7dc98SMatthew Ahrens goto out; 3028088f3894Sahrens 302974e7dc98SMatthew Ahrens snap = list_head(&pa.shared_snaps); 303074e7dc98SMatthew Ahrens ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); 303174e7dc98SMatthew Ahrens err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj, 303274e7dc98SMatthew Ahrens snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps); 303374e7dc98SMatthew Ahrens if (err != 0) 303474e7dc98SMatthew Ahrens goto out; 3035088f3894Sahrens 3036cde58dbcSMatthew Ahrens if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) { 3037cde58dbcSMatthew Ahrens err = dsl_dataset_hold_obj(dp, 303874e7dc98SMatthew Ahrens snap->ds->ds_dir->dd_phys->dd_origin_obj, 3039cde58dbcSMatthew Ahrens FTAG, &pa.origin_origin); 304074e7dc98SMatthew Ahrens if (err != 0) 304174e7dc98SMatthew Ahrens goto out; 304274e7dc98SMatthew Ahrens } 3043745cd3c5Smaybee 304474e7dc98SMatthew Ahrens out: 304574e7dc98SMatthew Ahrens rw_exit(&dp->dp_config_rwlock); 3046745cd3c5Smaybee 304799653d4eSeschrock /* 304899653d4eSeschrock * Add in 128x the snapnames zapobj size, since we will be moving 304999653d4eSeschrock * a bunch of snapnames to the promoted ds, and dirtying their 305099653d4eSeschrock * bonus buffers. 305199653d4eSeschrock */ 305274e7dc98SMatthew Ahrens if (err == 0) { 305374e7dc98SMatthew Ahrens err = dsl_sync_task_do(dp, dsl_dataset_promote_check, 305474e7dc98SMatthew Ahrens dsl_dataset_promote_sync, ds, &pa, 3055b24ab676SJeff Bonwick 2 + 2 * doi.doi_physical_blocks_512); 3056681d9761SEric Taylor if (err && pa.err_ds && conflsnap) 3057681d9761SEric Taylor (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN); 3058745cd3c5Smaybee } 305974e7dc98SMatthew Ahrens 306074e7dc98SMatthew Ahrens snaplist_destroy(&pa.shared_snaps, B_TRUE); 306174e7dc98SMatthew Ahrens snaplist_destroy(&pa.clone_snaps, B_FALSE); 306274e7dc98SMatthew Ahrens snaplist_destroy(&pa.origin_snaps, B_FALSE); 306374e7dc98SMatthew Ahrens if (pa.origin_origin) 3064cde58dbcSMatthew Ahrens dsl_dataset_rele(pa.origin_origin, FTAG); 3065745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 306699653d4eSeschrock return (err); 306799653d4eSeschrock } 3068b1b8ab34Slling 30693cb34c60Sahrens struct cloneswaparg { 30703cb34c60Sahrens dsl_dataset_t *cds; /* clone dataset */ 30713cb34c60Sahrens dsl_dataset_t *ohds; /* origin's head dataset */ 30723cb34c60Sahrens boolean_t force; 3073a9b821a0Sck int64_t unused_refres_delta; /* change in unconsumed refreservation */ 30743cb34c60Sahrens }; 3075f18faf3fSek 3076f18faf3fSek /* ARGSUSED */ 3077f18faf3fSek static int 3078f18faf3fSek dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx) 3079f18faf3fSek { 30803cb34c60Sahrens struct cloneswaparg *csa = arg1; 3081f18faf3fSek 30823cb34c60Sahrens /* they should both be heads */ 30833cb34c60Sahrens if (dsl_dataset_is_snapshot(csa->cds) || 30843cb34c60Sahrens dsl_dataset_is_snapshot(csa->ohds)) 3085f18faf3fSek return (EINVAL); 3086f18faf3fSek 30873cb34c60Sahrens /* the branch point should be just before them */ 30883cb34c60Sahrens if (csa->cds->ds_prev != csa->ohds->ds_prev) 3089f18faf3fSek return (EINVAL); 3090f18faf3fSek 3091ae46e4c7SMatthew Ahrens /* cds should be the clone (unless they are unrelated) */ 3092ae46e4c7SMatthew Ahrens if (csa->cds->ds_prev != NULL && 3093ae46e4c7SMatthew Ahrens csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap && 3094ae46e4c7SMatthew Ahrens csa->ohds->ds_object != 3095ae46e4c7SMatthew Ahrens csa->cds->ds_prev->ds_phys->ds_next_snap_obj) 30963cb34c60Sahrens return (EINVAL); 3097f18faf3fSek 30983cb34c60Sahrens /* the clone should be a child of the origin */ 30993cb34c60Sahrens if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir) 31003cb34c60Sahrens return (EINVAL); 3101f18faf3fSek 31023cb34c60Sahrens /* ohds shouldn't be modified unless 'force' */ 31033cb34c60Sahrens if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds)) 31043cb34c60Sahrens return (ETXTBSY); 3105a9b821a0Sck 3106a9b821a0Sck /* adjust amount of any unconsumed refreservation */ 3107a9b821a0Sck csa->unused_refres_delta = 3108a9b821a0Sck (int64_t)MIN(csa->ohds->ds_reserved, 3109a9b821a0Sck csa->ohds->ds_phys->ds_unique_bytes) - 3110a9b821a0Sck (int64_t)MIN(csa->ohds->ds_reserved, 3111a9b821a0Sck csa->cds->ds_phys->ds_unique_bytes); 3112a9b821a0Sck 3113a9b821a0Sck if (csa->unused_refres_delta > 0 && 3114a9b821a0Sck csa->unused_refres_delta > 3115a9b821a0Sck dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE)) 3116a9b821a0Sck return (ENOSPC); 3117a9b821a0Sck 3118c4cbca4fSChris Kirby if (csa->ohds->ds_quota != 0 && 3119c4cbca4fSChris Kirby csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota) 3120c4cbca4fSChris Kirby return (EDQUOT); 3121c4cbca4fSChris Kirby 31223cb34c60Sahrens return (0); 3123f18faf3fSek } 3124f18faf3fSek 3125f18faf3fSek /* ARGSUSED */ 3126f18faf3fSek static void 31273f9d6ad7SLin Ling dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3128f18faf3fSek { 31293cb34c60Sahrens struct cloneswaparg *csa = arg1; 31303cb34c60Sahrens dsl_pool_t *dp = csa->cds->ds_dir->dd_pool; 3131f18faf3fSek 3132a9b821a0Sck ASSERT(csa->cds->ds_reserved == 0); 3133c4cbca4fSChris Kirby ASSERT(csa->ohds->ds_quota == 0 || 3134c4cbca4fSChris Kirby csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota); 3135a9b821a0Sck 31363cb34c60Sahrens dmu_buf_will_dirty(csa->cds->ds_dbuf, tx); 31373cb34c60Sahrens dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx); 3138f18faf3fSek 3139503ad85cSMatthew Ahrens if (csa->cds->ds_objset != NULL) { 3140503ad85cSMatthew Ahrens dmu_objset_evict(csa->cds->ds_objset); 3141503ad85cSMatthew Ahrens csa->cds->ds_objset = NULL; 31423cb34c60Sahrens } 3143f18faf3fSek 3144503ad85cSMatthew Ahrens if (csa->ohds->ds_objset != NULL) { 3145503ad85cSMatthew Ahrens dmu_objset_evict(csa->ohds->ds_objset); 3146503ad85cSMatthew Ahrens csa->ohds->ds_objset = NULL; 31473cb34c60Sahrens } 3148f18faf3fSek 3149ae46e4c7SMatthew Ahrens /* 3150ae46e4c7SMatthew Ahrens * Reset origin's unique bytes, if it exists. 3151ae46e4c7SMatthew Ahrens */ 3152ae46e4c7SMatthew Ahrens if (csa->cds->ds_prev) { 3153ae46e4c7SMatthew Ahrens dsl_dataset_t *origin = csa->cds->ds_prev; 3154cde58dbcSMatthew Ahrens uint64_t comp, uncomp; 3155cde58dbcSMatthew Ahrens 3156ae46e4c7SMatthew Ahrens dmu_buf_will_dirty(origin->ds_dbuf, tx); 3157cde58dbcSMatthew Ahrens dsl_deadlist_space_range(&csa->cds->ds_deadlist, 3158ae46e4c7SMatthew Ahrens origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, 3159cde58dbcSMatthew Ahrens &origin->ds_phys->ds_unique_bytes, &comp, &uncomp); 3160ae46e4c7SMatthew Ahrens } 3161f18faf3fSek 3162f18faf3fSek /* swap blkptrs */ 3163f18faf3fSek { 3164f18faf3fSek blkptr_t tmp; 31653cb34c60Sahrens tmp = csa->ohds->ds_phys->ds_bp; 31663cb34c60Sahrens csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp; 31673cb34c60Sahrens csa->cds->ds_phys->ds_bp = tmp; 3168f18faf3fSek } 3169f18faf3fSek 3170f18faf3fSek /* set dd_*_bytes */ 3171f18faf3fSek { 3172f18faf3fSek int64_t dused, dcomp, duncomp; 3173f18faf3fSek uint64_t cdl_used, cdl_comp, cdl_uncomp; 3174f18faf3fSek uint64_t odl_used, odl_comp, odl_uncomp; 3175f18faf3fSek 317674e7dc98SMatthew Ahrens ASSERT3U(csa->cds->ds_dir->dd_phys-> 317774e7dc98SMatthew Ahrens dd_used_breakdown[DD_USED_SNAP], ==, 0); 317874e7dc98SMatthew Ahrens 3179cde58dbcSMatthew Ahrens dsl_deadlist_space(&csa->cds->ds_deadlist, 3180cde58dbcSMatthew Ahrens &cdl_used, &cdl_comp, &cdl_uncomp); 3181cde58dbcSMatthew Ahrens dsl_deadlist_space(&csa->ohds->ds_deadlist, 3182cde58dbcSMatthew Ahrens &odl_used, &odl_comp, &odl_uncomp); 318374e7dc98SMatthew Ahrens 31843cb34c60Sahrens dused = csa->cds->ds_phys->ds_used_bytes + cdl_used - 31853cb34c60Sahrens (csa->ohds->ds_phys->ds_used_bytes + odl_used); 31863cb34c60Sahrens dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp - 31873cb34c60Sahrens (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp); 31883cb34c60Sahrens duncomp = csa->cds->ds_phys->ds_uncompressed_bytes + 31893cb34c60Sahrens cdl_uncomp - 31903cb34c60Sahrens (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp); 31913cb34c60Sahrens 319274e7dc98SMatthew Ahrens dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD, 31933cb34c60Sahrens dused, dcomp, duncomp, tx); 319474e7dc98SMatthew Ahrens dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD, 31953cb34c60Sahrens -dused, -dcomp, -duncomp, tx); 319674e7dc98SMatthew Ahrens 319774e7dc98SMatthew Ahrens /* 319874e7dc98SMatthew Ahrens * The difference in the space used by snapshots is the 319974e7dc98SMatthew Ahrens * difference in snapshot space due to the head's 320074e7dc98SMatthew Ahrens * deadlist (since that's the only thing that's 320174e7dc98SMatthew Ahrens * changing that affects the snapused). 320274e7dc98SMatthew Ahrens */ 3203cde58dbcSMatthew Ahrens dsl_deadlist_space_range(&csa->cds->ds_deadlist, 3204cde58dbcSMatthew Ahrens csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX, 3205cde58dbcSMatthew Ahrens &cdl_used, &cdl_comp, &cdl_uncomp); 3206cde58dbcSMatthew Ahrens dsl_deadlist_space_range(&csa->ohds->ds_deadlist, 3207cde58dbcSMatthew Ahrens csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX, 3208cde58dbcSMatthew Ahrens &odl_used, &odl_comp, &odl_uncomp); 320974e7dc98SMatthew Ahrens dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used, 321074e7dc98SMatthew Ahrens DD_USED_HEAD, DD_USED_SNAP, tx); 32113cb34c60Sahrens } 32123cb34c60Sahrens 3213f18faf3fSek /* swap ds_*_bytes */ 32143cb34c60Sahrens SWITCH64(csa->ohds->ds_phys->ds_used_bytes, 32153cb34c60Sahrens csa->cds->ds_phys->ds_used_bytes); 32163cb34c60Sahrens SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes, 32173cb34c60Sahrens csa->cds->ds_phys->ds_compressed_bytes); 32183cb34c60Sahrens SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes, 32193cb34c60Sahrens csa->cds->ds_phys->ds_uncompressed_bytes); 3220a9b821a0Sck SWITCH64(csa->ohds->ds_phys->ds_unique_bytes, 3221a9b821a0Sck csa->cds->ds_phys->ds_unique_bytes); 3222a9b821a0Sck 3223a9b821a0Sck /* apply any parent delta for change in unconsumed refreservation */ 322474e7dc98SMatthew Ahrens dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV, 322574e7dc98SMatthew Ahrens csa->unused_refres_delta, 0, 0, tx); 3226f18faf3fSek 3227cde58dbcSMatthew Ahrens /* 3228cde58dbcSMatthew Ahrens * Swap deadlists. 3229cde58dbcSMatthew Ahrens */ 3230cde58dbcSMatthew Ahrens dsl_deadlist_close(&csa->cds->ds_deadlist); 3231cde58dbcSMatthew Ahrens dsl_deadlist_close(&csa->ohds->ds_deadlist); 32323cb34c60Sahrens SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj, 32333cb34c60Sahrens csa->cds->ds_phys->ds_deadlist_obj); 3234cde58dbcSMatthew Ahrens dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset, 3235cde58dbcSMatthew Ahrens csa->cds->ds_phys->ds_deadlist_obj); 3236cde58dbcSMatthew Ahrens dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset, 3237cde58dbcSMatthew Ahrens csa->ohds->ds_phys->ds_deadlist_obj); 323888b7b0f2SMatthew Ahrens 32393f9d6ad7SLin Ling dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx); 3240f18faf3fSek } 3241f18faf3fSek 3242f18faf3fSek /* 3243ae46e4c7SMatthew Ahrens * Swap 'clone' with its origin head datasets. Used at the end of "zfs 3244ae46e4c7SMatthew Ahrens * recv" into an existing fs to swizzle the file system to the new 3245ae46e4c7SMatthew Ahrens * version, and by "zfs rollback". Can also be used to swap two 3246ae46e4c7SMatthew Ahrens * independent head datasets if neither has any snapshots. 3247f18faf3fSek */ 3248f18faf3fSek int 32493cb34c60Sahrens dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, 32503cb34c60Sahrens boolean_t force) 3251f18faf3fSek { 32523cb34c60Sahrens struct cloneswaparg csa; 3253745cd3c5Smaybee int error; 3254f18faf3fSek 3255745cd3c5Smaybee ASSERT(clone->ds_owner); 3256745cd3c5Smaybee ASSERT(origin_head->ds_owner); 3257745cd3c5Smaybee retry: 325877972028SChris Kirby /* 325977972028SChris Kirby * Need exclusive access for the swap. If we're swapping these 326077972028SChris Kirby * datasets back after an error, we already hold the locks. 326177972028SChris Kirby */ 326277972028SChris Kirby if (!RW_WRITE_HELD(&clone->ds_rwlock)) 326377972028SChris Kirby rw_enter(&clone->ds_rwlock, RW_WRITER); 326477972028SChris Kirby if (!RW_WRITE_HELD(&origin_head->ds_rwlock) && 326577972028SChris Kirby !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) { 3266745cd3c5Smaybee rw_exit(&clone->ds_rwlock); 3267745cd3c5Smaybee rw_enter(&origin_head->ds_rwlock, RW_WRITER); 3268745cd3c5Smaybee if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) { 3269745cd3c5Smaybee rw_exit(&origin_head->ds_rwlock); 3270745cd3c5Smaybee goto retry; 3271745cd3c5Smaybee } 3272745cd3c5Smaybee } 32733cb34c60Sahrens csa.cds = clone; 32743cb34c60Sahrens csa.ohds = origin_head; 32753cb34c60Sahrens csa.force = force; 3276745cd3c5Smaybee error = dsl_sync_task_do(clone->ds_dir->dd_pool, 3277f18faf3fSek dsl_dataset_clone_swap_check, 3278745cd3c5Smaybee dsl_dataset_clone_swap_sync, &csa, NULL, 9); 3279745cd3c5Smaybee return (error); 3280f18faf3fSek } 3281f18faf3fSek 3282b1b8ab34Slling /* 3283b1b8ab34Slling * Given a pool name and a dataset object number in that pool, 3284b1b8ab34Slling * return the name of that dataset. 3285b1b8ab34Slling */ 3286b1b8ab34Slling int 3287b1b8ab34Slling dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 3288b1b8ab34Slling { 3289b1b8ab34Slling spa_t *spa; 3290b1b8ab34Slling dsl_pool_t *dp; 3291745cd3c5Smaybee dsl_dataset_t *ds; 3292b1b8ab34Slling int error; 3293b1b8ab34Slling 3294b1b8ab34Slling if ((error = spa_open(pname, &spa, FTAG)) != 0) 3295b1b8ab34Slling return (error); 3296b1b8ab34Slling dp = spa_get_dsl(spa); 3297b1b8ab34Slling rw_enter(&dp->dp_config_rwlock, RW_READER); 3298745cd3c5Smaybee if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) { 3299745cd3c5Smaybee dsl_dataset_name(ds, buf); 3300745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 3301b1b8ab34Slling } 3302b1b8ab34Slling rw_exit(&dp->dp_config_rwlock); 3303b1b8ab34Slling spa_close(spa, FTAG); 3304b1b8ab34Slling 3305745cd3c5Smaybee return (error); 3306b1b8ab34Slling } 3307a9799022Sck 3308a9799022Sck int 3309a9799022Sck dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 3310745cd3c5Smaybee uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 3311a9799022Sck { 3312a9799022Sck int error = 0; 3313a9799022Sck 3314a9799022Sck ASSERT3S(asize, >, 0); 3315a9799022Sck 33169082849eSck /* 33179082849eSck * *ref_rsrv is the portion of asize that will come from any 33189082849eSck * unconsumed refreservation space. 33199082849eSck */ 33209082849eSck *ref_rsrv = 0; 33219082849eSck 3322a9799022Sck mutex_enter(&ds->ds_lock); 3323a9799022Sck /* 3324a9799022Sck * Make a space adjustment for reserved bytes. 3325a9799022Sck */ 3326a9799022Sck if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { 3327a9799022Sck ASSERT3U(*used, >=, 3328a9799022Sck ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 3329a9799022Sck *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 33309082849eSck *ref_rsrv = 33319082849eSck asize - MIN(asize, parent_delta(ds, asize + inflight)); 3332a9799022Sck } 3333a9799022Sck 3334a9799022Sck if (!check_quota || ds->ds_quota == 0) { 3335a9799022Sck mutex_exit(&ds->ds_lock); 3336a9799022Sck return (0); 3337a9799022Sck } 3338a9799022Sck /* 3339a9799022Sck * If they are requesting more space, and our current estimate 3340a9799022Sck * is over quota, they get to try again unless the actual 3341a9799022Sck * on-disk is over quota and there are no pending changes (which 3342a9799022Sck * may free up space for us). 3343a9799022Sck */ 3344a9799022Sck if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) { 3345a9799022Sck if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota) 3346a9799022Sck error = ERESTART; 3347a9799022Sck else 3348a9799022Sck error = EDQUOT; 3349a9799022Sck } 3350a9799022Sck mutex_exit(&ds->ds_lock); 3351a9799022Sck 3352a9799022Sck return (error); 3353a9799022Sck } 3354a9799022Sck 3355a9799022Sck /* ARGSUSED */ 3356a9799022Sck static int 3357a9799022Sck dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) 3358a9799022Sck { 3359a9799022Sck dsl_dataset_t *ds = arg1; 336092241e0bSTom Erickson dsl_prop_setarg_t *psa = arg2; 336192241e0bSTom Erickson int err; 3362a9799022Sck 3363a9799022Sck if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA) 3364a9799022Sck return (ENOTSUP); 3365a9799022Sck 336692241e0bSTom Erickson if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) 336792241e0bSTom Erickson return (err); 336892241e0bSTom Erickson 336992241e0bSTom Erickson if (psa->psa_effective_value == 0) 3370a9799022Sck return (0); 3371a9799022Sck 337292241e0bSTom Erickson if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes || 337392241e0bSTom Erickson psa->psa_effective_value < ds->ds_reserved) 3374a9799022Sck return (ENOSPC); 3375a9799022Sck 3376a9799022Sck return (0); 3377a9799022Sck } 3378a9799022Sck 33793f9d6ad7SLin Ling extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *); 338092241e0bSTom Erickson 3381a9799022Sck void 33823f9d6ad7SLin Ling dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3383a9799022Sck { 3384a9799022Sck dsl_dataset_t *ds = arg1; 338592241e0bSTom Erickson dsl_prop_setarg_t *psa = arg2; 338692241e0bSTom Erickson uint64_t effective_value = psa->psa_effective_value; 3387a9799022Sck 33883f9d6ad7SLin Ling dsl_prop_set_sync(ds, psa, tx); 338992241e0bSTom Erickson DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); 3390a9799022Sck 339192241e0bSTom Erickson if (ds->ds_quota != effective_value) { 339292241e0bSTom Erickson dmu_buf_will_dirty(ds->ds_dbuf, tx); 339392241e0bSTom Erickson ds->ds_quota = effective_value; 3394a9799022Sck 33953f9d6ad7SLin Ling spa_history_log_internal(LOG_DS_REFQUOTA, 33963f9d6ad7SLin Ling ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ", 339792241e0bSTom Erickson (longlong_t)ds->ds_quota, ds->ds_object); 339892241e0bSTom Erickson } 3399a9799022Sck } 3400a9799022Sck 3401a9799022Sck int 340292241e0bSTom Erickson dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota) 3403a9799022Sck { 3404a9799022Sck dsl_dataset_t *ds; 340592241e0bSTom Erickson dsl_prop_setarg_t psa; 3406a9799022Sck int err; 3407a9799022Sck 340892241e0bSTom Erickson dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a); 340992241e0bSTom Erickson 3410745cd3c5Smaybee err = dsl_dataset_hold(dsname, FTAG, &ds); 3411a9799022Sck if (err) 3412a9799022Sck return (err); 3413a9799022Sck 341492241e0bSTom Erickson /* 341592241e0bSTom Erickson * If someone removes a file, then tries to set the quota, we 341692241e0bSTom Erickson * want to make sure the file freeing takes effect. 341792241e0bSTom Erickson */ 341892241e0bSTom Erickson txg_wait_open(ds->ds_dir->dd_pool, 0); 341992241e0bSTom Erickson 342092241e0bSTom Erickson err = dsl_sync_task_do(ds->ds_dir->dd_pool, 342192241e0bSTom Erickson dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync, 342292241e0bSTom Erickson ds, &psa, 0); 3423a9799022Sck 3424745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 3425a9799022Sck return (err); 3426a9799022Sck } 3427a9799022Sck 3428a9799022Sck static int 3429a9799022Sck dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) 3430a9799022Sck { 3431a9799022Sck dsl_dataset_t *ds = arg1; 343292241e0bSTom Erickson dsl_prop_setarg_t *psa = arg2; 343392241e0bSTom Erickson uint64_t effective_value; 3434a9799022Sck uint64_t unique; 343592241e0bSTom Erickson int err; 3436a9799022Sck 3437a9799022Sck if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 3438a9799022Sck SPA_VERSION_REFRESERVATION) 3439a9799022Sck return (ENOTSUP); 3440a9799022Sck 3441a9799022Sck if (dsl_dataset_is_snapshot(ds)) 3442a9799022Sck return (EINVAL); 3443a9799022Sck 344492241e0bSTom Erickson if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) 344592241e0bSTom Erickson return (err); 344692241e0bSTom Erickson 344792241e0bSTom Erickson effective_value = psa->psa_effective_value; 344892241e0bSTom Erickson 3449a9799022Sck /* 3450a9799022Sck * If we are doing the preliminary check in open context, the 3451a9799022Sck * space estimates may be inaccurate. 3452a9799022Sck */ 3453a9799022Sck if (!dmu_tx_is_syncing(tx)) 3454a9799022Sck return (0); 3455a9799022Sck 3456a9799022Sck mutex_enter(&ds->ds_lock); 34573f9d6ad7SLin Ling if (!DS_UNIQUE_IS_ACCURATE(ds)) 34583f9d6ad7SLin Ling dsl_dataset_recalc_head_uniq(ds); 34593f9d6ad7SLin Ling unique = ds->ds_phys->ds_unique_bytes; 3460a9799022Sck mutex_exit(&ds->ds_lock); 3461a9799022Sck 346292241e0bSTom Erickson if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) { 346392241e0bSTom Erickson uint64_t delta = MAX(unique, effective_value) - 3464379c004dSEric Schrock MAX(unique, ds->ds_reserved); 3465379c004dSEric Schrock 3466379c004dSEric Schrock if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 3467379c004dSEric Schrock return (ENOSPC); 3468379c004dSEric Schrock if (ds->ds_quota > 0 && 346992241e0bSTom Erickson effective_value > ds->ds_quota) 3470379c004dSEric Schrock return (ENOSPC); 3471379c004dSEric Schrock } 3472a9799022Sck 3473a9799022Sck return (0); 3474a9799022Sck } 3475a9799022Sck 3476a9799022Sck static void 34773f9d6ad7SLin Ling dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3478a9799022Sck { 3479a9799022Sck dsl_dataset_t *ds = arg1; 348092241e0bSTom Erickson dsl_prop_setarg_t *psa = arg2; 348192241e0bSTom Erickson uint64_t effective_value = psa->psa_effective_value; 348202c8f3f0SMatthew Ahrens uint64_t unique; 348302c8f3f0SMatthew Ahrens int64_t delta; 348402c8f3f0SMatthew Ahrens 34853f9d6ad7SLin Ling dsl_prop_set_sync(ds, psa, tx); 348692241e0bSTom Erickson DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); 348792241e0bSTom Erickson 348802c8f3f0SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 348902c8f3f0SMatthew Ahrens 349002c8f3f0SMatthew Ahrens mutex_enter(&ds->ds_dir->dd_lock); 349102c8f3f0SMatthew Ahrens mutex_enter(&ds->ds_lock); 34923f9d6ad7SLin Ling ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 34933f9d6ad7SLin Ling unique = ds->ds_phys->ds_unique_bytes; 349492241e0bSTom Erickson delta = MAX(0, (int64_t)(effective_value - unique)) - 349502c8f3f0SMatthew Ahrens MAX(0, (int64_t)(ds->ds_reserved - unique)); 349692241e0bSTom Erickson ds->ds_reserved = effective_value; 349702c8f3f0SMatthew Ahrens mutex_exit(&ds->ds_lock); 349802c8f3f0SMatthew Ahrens 349902c8f3f0SMatthew Ahrens dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 350002c8f3f0SMatthew Ahrens mutex_exit(&ds->ds_dir->dd_lock); 3501a9799022Sck 35023f9d6ad7SLin Ling spa_history_log_internal(LOG_DS_REFRESERV, 35033f9d6ad7SLin Ling ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu", 350492241e0bSTom Erickson (longlong_t)effective_value, ds->ds_object); 3505a9799022Sck } 3506a9799022Sck 3507a9799022Sck int 350892241e0bSTom Erickson dsl_dataset_set_reservation(const char *dsname, zprop_source_t source, 350992241e0bSTom Erickson uint64_t reservation) 3510a9799022Sck { 3511a9799022Sck dsl_dataset_t *ds; 351292241e0bSTom Erickson dsl_prop_setarg_t psa; 3513a9799022Sck int err; 3514a9799022Sck 351592241e0bSTom Erickson dsl_prop_setarg_init_uint64(&psa, "refreservation", source, 351692241e0bSTom Erickson &reservation); 351792241e0bSTom Erickson 3518745cd3c5Smaybee err = dsl_dataset_hold(dsname, FTAG, &ds); 3519a9799022Sck if (err) 3520a9799022Sck return (err); 3521a9799022Sck 3522a9799022Sck err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3523a9799022Sck dsl_dataset_set_reservation_check, 352492241e0bSTom Erickson dsl_dataset_set_reservation_sync, ds, &psa, 0); 352592241e0bSTom Erickson 3526745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 3527a9799022Sck return (err); 3528a9799022Sck } 3529842727c2SChris Kirby 3530c99e4bdcSChris Kirby typedef struct zfs_hold_cleanup_arg { 3531a7f53a56SChris Kirby dsl_pool_t *dp; 3532a7f53a56SChris Kirby uint64_t dsobj; 3533c99e4bdcSChris Kirby char htag[MAXNAMELEN]; 3534c99e4bdcSChris Kirby } zfs_hold_cleanup_arg_t; 3535c99e4bdcSChris Kirby 3536c99e4bdcSChris Kirby static void 3537c99e4bdcSChris Kirby dsl_dataset_user_release_onexit(void *arg) 3538c99e4bdcSChris Kirby { 3539c99e4bdcSChris Kirby zfs_hold_cleanup_arg_t *ca = arg; 3540c99e4bdcSChris Kirby 3541a7f53a56SChris Kirby (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag, 3542a7f53a56SChris Kirby B_TRUE); 3543c99e4bdcSChris Kirby kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t)); 3544c99e4bdcSChris Kirby } 3545c99e4bdcSChris Kirby 3546a7f53a56SChris Kirby void 3547a7f53a56SChris Kirby dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag, 3548a7f53a56SChris Kirby minor_t minor) 3549a7f53a56SChris Kirby { 3550a7f53a56SChris Kirby zfs_hold_cleanup_arg_t *ca; 3551a7f53a56SChris Kirby 3552a7f53a56SChris Kirby ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP); 3553a7f53a56SChris Kirby ca->dp = ds->ds_dir->dd_pool; 3554a7f53a56SChris Kirby ca->dsobj = ds->ds_object; 3555a7f53a56SChris Kirby (void) strlcpy(ca->htag, htag, sizeof (ca->htag)); 3556a7f53a56SChris Kirby VERIFY3U(0, ==, zfs_onexit_add_cb(minor, 3557a7f53a56SChris Kirby dsl_dataset_user_release_onexit, ca, NULL)); 3558a7f53a56SChris Kirby } 3559a7f53a56SChris Kirby 356015508ac0SChris Kirby /* 356199d5e173STim Haley * If you add new checks here, you may need to add 356299d5e173STim Haley * additional checks to the "temporary" case in 356399d5e173STim Haley * snapshot_check() in dmu_objset.c. 356415508ac0SChris Kirby */ 3565842727c2SChris Kirby static int 3566842727c2SChris Kirby dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx) 3567842727c2SChris Kirby { 3568842727c2SChris Kirby dsl_dataset_t *ds = arg1; 3569ca45db41SChris Kirby struct dsl_ds_holdarg *ha = arg2; 3570ca45db41SChris Kirby char *htag = ha->htag; 3571842727c2SChris Kirby objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3572842727c2SChris Kirby int error = 0; 3573842727c2SChris Kirby 3574842727c2SChris Kirby if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) 3575842727c2SChris Kirby return (ENOTSUP); 3576842727c2SChris Kirby 3577842727c2SChris Kirby if (!dsl_dataset_is_snapshot(ds)) 3578842727c2SChris Kirby return (EINVAL); 3579842727c2SChris Kirby 3580842727c2SChris Kirby /* tags must be unique */ 3581842727c2SChris Kirby mutex_enter(&ds->ds_lock); 3582842727c2SChris Kirby if (ds->ds_phys->ds_userrefs_obj) { 3583842727c2SChris Kirby error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag, 3584842727c2SChris Kirby 8, 1, tx); 3585842727c2SChris Kirby if (error == 0) 3586842727c2SChris Kirby error = EEXIST; 3587842727c2SChris Kirby else if (error == ENOENT) 3588842727c2SChris Kirby error = 0; 3589842727c2SChris Kirby } 3590842727c2SChris Kirby mutex_exit(&ds->ds_lock); 3591842727c2SChris Kirby 359215508ac0SChris Kirby if (error == 0 && ha->temphold && 359315508ac0SChris Kirby strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN) 359415508ac0SChris Kirby error = E2BIG; 359515508ac0SChris Kirby 3596842727c2SChris Kirby return (error); 3597842727c2SChris Kirby } 3598842727c2SChris Kirby 359999d5e173STim Haley void 36003f9d6ad7SLin Ling dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3601842727c2SChris Kirby { 3602842727c2SChris Kirby dsl_dataset_t *ds = arg1; 3603ca45db41SChris Kirby struct dsl_ds_holdarg *ha = arg2; 3604ca45db41SChris Kirby char *htag = ha->htag; 3605ca45db41SChris Kirby dsl_pool_t *dp = ds->ds_dir->dd_pool; 3606ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 360715508ac0SChris Kirby uint64_t now = gethrestime_sec(); 3608842727c2SChris Kirby uint64_t zapobj; 3609842727c2SChris Kirby 3610842727c2SChris Kirby mutex_enter(&ds->ds_lock); 3611842727c2SChris Kirby if (ds->ds_phys->ds_userrefs_obj == 0) { 3612842727c2SChris Kirby /* 3613842727c2SChris Kirby * This is the first user hold for this dataset. Create 3614842727c2SChris Kirby * the userrefs zap object. 3615842727c2SChris Kirby */ 3616842727c2SChris Kirby dmu_buf_will_dirty(ds->ds_dbuf, tx); 3617842727c2SChris Kirby zapobj = ds->ds_phys->ds_userrefs_obj = 3618842727c2SChris Kirby zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx); 3619842727c2SChris Kirby } else { 3620842727c2SChris Kirby zapobj = ds->ds_phys->ds_userrefs_obj; 3621842727c2SChris Kirby } 3622842727c2SChris Kirby ds->ds_userrefs++; 3623842727c2SChris Kirby mutex_exit(&ds->ds_lock); 3624842727c2SChris Kirby 3625842727c2SChris Kirby VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx)); 3626842727c2SChris Kirby 3627ca45db41SChris Kirby if (ha->temphold) { 3628ca45db41SChris Kirby VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object, 3629ca45db41SChris Kirby htag, &now, tx)); 3630ca45db41SChris Kirby } 3631ca45db41SChris Kirby 36323f9d6ad7SLin Ling spa_history_log_internal(LOG_DS_USER_HOLD, 36333f9d6ad7SLin Ling dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag, 3634ca45db41SChris Kirby (int)ha->temphold, ds->ds_object); 3635842727c2SChris Kirby } 3636842727c2SChris Kirby 3637842727c2SChris Kirby static int 3638fd136879SMatthew Ahrens dsl_dataset_user_hold_one(const char *dsname, void *arg) 3639842727c2SChris Kirby { 3640842727c2SChris Kirby struct dsl_ds_holdarg *ha = arg; 3641842727c2SChris Kirby dsl_dataset_t *ds; 3642842727c2SChris Kirby int error; 3643842727c2SChris Kirby char *name; 3644842727c2SChris Kirby 3645842727c2SChris Kirby /* alloc a buffer to hold dsname@snapname plus terminating NULL */ 3646ae46e4c7SMatthew Ahrens name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3647842727c2SChris Kirby error = dsl_dataset_hold(name, ha->dstg, &ds); 3648ae46e4c7SMatthew Ahrens strfree(name); 3649842727c2SChris Kirby if (error == 0) { 3650d7747cbcSChris Kirby ha->gotone = B_TRUE; 3651842727c2SChris Kirby dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check, 3652ca45db41SChris Kirby dsl_dataset_user_hold_sync, ds, ha, 0); 3653842727c2SChris Kirby } else if (error == ENOENT && ha->recursive) { 3654842727c2SChris Kirby error = 0; 3655842727c2SChris Kirby } else { 3656fd136879SMatthew Ahrens (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3657842727c2SChris Kirby } 3658842727c2SChris Kirby return (error); 3659842727c2SChris Kirby } 3660842727c2SChris Kirby 3661a7f53a56SChris Kirby int 3662a7f53a56SChris Kirby dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag, 3663a7f53a56SChris Kirby boolean_t temphold) 3664a7f53a56SChris Kirby { 3665a7f53a56SChris Kirby struct dsl_ds_holdarg *ha; 3666a7f53a56SChris Kirby int error; 3667a7f53a56SChris Kirby 3668a7f53a56SChris Kirby ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3669a7f53a56SChris Kirby ha->htag = htag; 3670a7f53a56SChris Kirby ha->temphold = temphold; 3671a7f53a56SChris Kirby error = dsl_sync_task_do(ds->ds_dir->dd_pool, 3672a7f53a56SChris Kirby dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync, 3673a7f53a56SChris Kirby ds, ha, 0); 3674a7f53a56SChris Kirby kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3675a7f53a56SChris Kirby 3676a7f53a56SChris Kirby return (error); 3677a7f53a56SChris Kirby } 3678a7f53a56SChris Kirby 3679842727c2SChris Kirby int 3680842727c2SChris Kirby dsl_dataset_user_hold(char *dsname, char *snapname, char *htag, 3681c99e4bdcSChris Kirby boolean_t recursive, boolean_t temphold, int cleanup_fd) 3682842727c2SChris Kirby { 3683842727c2SChris Kirby struct dsl_ds_holdarg *ha; 3684842727c2SChris Kirby dsl_sync_task_t *dst; 3685842727c2SChris Kirby spa_t *spa; 3686842727c2SChris Kirby int error; 3687a7f53a56SChris Kirby minor_t minor = 0; 3688a7f53a56SChris Kirby 3689a7f53a56SChris Kirby if (cleanup_fd != -1) { 3690a7f53a56SChris Kirby /* Currently we only support cleanup-on-exit of tempholds. */ 3691a7f53a56SChris Kirby if (!temphold) 3692a7f53a56SChris Kirby return (EINVAL); 3693a7f53a56SChris Kirby error = zfs_onexit_fd_hold(cleanup_fd, &minor); 3694a7f53a56SChris Kirby if (error) 3695a7f53a56SChris Kirby return (error); 3696a7f53a56SChris Kirby } 3697842727c2SChris Kirby 3698842727c2SChris Kirby ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3699842727c2SChris Kirby 3700842727c2SChris Kirby (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3701842727c2SChris Kirby 3702842727c2SChris Kirby error = spa_open(dsname, &spa, FTAG); 3703842727c2SChris Kirby if (error) { 3704842727c2SChris Kirby kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3705a7f53a56SChris Kirby if (cleanup_fd != -1) 3706a7f53a56SChris Kirby zfs_onexit_fd_rele(cleanup_fd); 3707842727c2SChris Kirby return (error); 3708842727c2SChris Kirby } 3709842727c2SChris Kirby 3710842727c2SChris Kirby ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 3711842727c2SChris Kirby ha->htag = htag; 3712842727c2SChris Kirby ha->snapname = snapname; 3713842727c2SChris Kirby ha->recursive = recursive; 3714ca45db41SChris Kirby ha->temphold = temphold; 3715c99e4bdcSChris Kirby 3716842727c2SChris Kirby if (recursive) { 3717842727c2SChris Kirby error = dmu_objset_find(dsname, dsl_dataset_user_hold_one, 3718842727c2SChris Kirby ha, DS_FIND_CHILDREN); 3719842727c2SChris Kirby } else { 3720842727c2SChris Kirby error = dsl_dataset_user_hold_one(dsname, ha); 3721842727c2SChris Kirby } 3722842727c2SChris Kirby if (error == 0) 3723842727c2SChris Kirby error = dsl_sync_task_group_wait(ha->dstg); 3724842727c2SChris Kirby 3725842727c2SChris Kirby for (dst = list_head(&ha->dstg->dstg_tasks); dst; 3726842727c2SChris Kirby dst = list_next(&ha->dstg->dstg_tasks, dst)) { 3727842727c2SChris Kirby dsl_dataset_t *ds = dst->dst_arg1; 3728842727c2SChris Kirby 3729842727c2SChris Kirby if (dst->dst_err) { 3730842727c2SChris Kirby dsl_dataset_name(ds, ha->failed); 3731842727c2SChris Kirby *strchr(ha->failed, '@') = '\0'; 3732a7f53a56SChris Kirby } else if (error == 0 && minor != 0 && temphold) { 3733a7f53a56SChris Kirby /* 3734a7f53a56SChris Kirby * If this hold is to be released upon process exit, 3735a7f53a56SChris Kirby * register that action now. 3736a7f53a56SChris Kirby */ 3737a7f53a56SChris Kirby dsl_register_onexit_hold_cleanup(ds, htag, minor); 3738842727c2SChris Kirby } 3739842727c2SChris Kirby dsl_dataset_rele(ds, ha->dstg); 3740842727c2SChris Kirby } 3741842727c2SChris Kirby 3742d7747cbcSChris Kirby if (error == 0 && recursive && !ha->gotone) 3743d7747cbcSChris Kirby error = ENOENT; 3744d7747cbcSChris Kirby 3745842727c2SChris Kirby if (error) 3746fd136879SMatthew Ahrens (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); 3747842727c2SChris Kirby 3748842727c2SChris Kirby dsl_sync_task_group_destroy(ha->dstg); 3749c99e4bdcSChris Kirby 3750842727c2SChris Kirby kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3751842727c2SChris Kirby spa_close(spa, FTAG); 3752a7f53a56SChris Kirby if (cleanup_fd != -1) 3753a7f53a56SChris Kirby zfs_onexit_fd_rele(cleanup_fd); 3754842727c2SChris Kirby return (error); 3755842727c2SChris Kirby } 3756842727c2SChris Kirby 3757842727c2SChris Kirby struct dsl_ds_releasearg { 3758842727c2SChris Kirby dsl_dataset_t *ds; 3759842727c2SChris Kirby const char *htag; 3760842727c2SChris Kirby boolean_t own; /* do we own or just hold ds? */ 3761842727c2SChris Kirby }; 3762842727c2SChris Kirby 3763842727c2SChris Kirby static int 3764842727c2SChris Kirby dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag, 3765842727c2SChris Kirby boolean_t *might_destroy) 3766842727c2SChris Kirby { 3767842727c2SChris Kirby objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3768842727c2SChris Kirby uint64_t zapobj; 3769842727c2SChris Kirby uint64_t tmp; 3770842727c2SChris Kirby int error; 3771842727c2SChris Kirby 3772842727c2SChris Kirby *might_destroy = B_FALSE; 3773842727c2SChris Kirby 3774842727c2SChris Kirby mutex_enter(&ds->ds_lock); 3775842727c2SChris Kirby zapobj = ds->ds_phys->ds_userrefs_obj; 3776842727c2SChris Kirby if (zapobj == 0) { 3777842727c2SChris Kirby /* The tag can't possibly exist */ 3778842727c2SChris Kirby mutex_exit(&ds->ds_lock); 3779842727c2SChris Kirby return (ESRCH); 3780842727c2SChris Kirby } 3781842727c2SChris Kirby 3782842727c2SChris Kirby /* Make sure the tag exists */ 3783842727c2SChris Kirby error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp); 3784842727c2SChris Kirby if (error) { 3785842727c2SChris Kirby mutex_exit(&ds->ds_lock); 3786842727c2SChris Kirby if (error == ENOENT) 3787842727c2SChris Kirby error = ESRCH; 3788842727c2SChris Kirby return (error); 3789842727c2SChris Kirby } 3790842727c2SChris Kirby 3791842727c2SChris Kirby if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 && 3792842727c2SChris Kirby DS_IS_DEFER_DESTROY(ds)) 3793842727c2SChris Kirby *might_destroy = B_TRUE; 3794842727c2SChris Kirby 3795842727c2SChris Kirby mutex_exit(&ds->ds_lock); 3796842727c2SChris Kirby return (0); 3797842727c2SChris Kirby } 3798842727c2SChris Kirby 3799842727c2SChris Kirby static int 3800842727c2SChris Kirby dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx) 3801842727c2SChris Kirby { 3802842727c2SChris Kirby struct dsl_ds_releasearg *ra = arg1; 3803842727c2SChris Kirby dsl_dataset_t *ds = ra->ds; 3804842727c2SChris Kirby boolean_t might_destroy; 3805842727c2SChris Kirby int error; 3806842727c2SChris Kirby 3807842727c2SChris Kirby if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) 3808842727c2SChris Kirby return (ENOTSUP); 3809842727c2SChris Kirby 3810842727c2SChris Kirby error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy); 3811842727c2SChris Kirby if (error) 3812842727c2SChris Kirby return (error); 3813842727c2SChris Kirby 3814842727c2SChris Kirby if (might_destroy) { 3815842727c2SChris Kirby struct dsl_ds_destroyarg dsda = {0}; 3816842727c2SChris Kirby 3817842727c2SChris Kirby if (dmu_tx_is_syncing(tx)) { 3818842727c2SChris Kirby /* 3819842727c2SChris Kirby * If we're not prepared to remove the snapshot, 3820842727c2SChris Kirby * we can't allow the release to happen right now. 3821842727c2SChris Kirby */ 3822842727c2SChris Kirby if (!ra->own) 3823842727c2SChris Kirby return (EBUSY); 3824842727c2SChris Kirby } 3825842727c2SChris Kirby dsda.ds = ds; 3826842727c2SChris Kirby dsda.releasing = B_TRUE; 3827842727c2SChris Kirby return (dsl_dataset_destroy_check(&dsda, tag, tx)); 3828842727c2SChris Kirby } 3829842727c2SChris Kirby 3830842727c2SChris Kirby return (0); 3831842727c2SChris Kirby } 3832842727c2SChris Kirby 3833842727c2SChris Kirby static void 38343f9d6ad7SLin Ling dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx) 3835842727c2SChris Kirby { 3836842727c2SChris Kirby struct dsl_ds_releasearg *ra = arg1; 3837842727c2SChris Kirby dsl_dataset_t *ds = ra->ds; 3838ca45db41SChris Kirby dsl_pool_t *dp = ds->ds_dir->dd_pool; 3839ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 3840842727c2SChris Kirby uint64_t zapobj; 3841842727c2SChris Kirby uint64_t dsobj = ds->ds_object; 3842842727c2SChris Kirby uint64_t refs; 3843ca45db41SChris Kirby int error; 3844842727c2SChris Kirby 3845842727c2SChris Kirby mutex_enter(&ds->ds_lock); 3846842727c2SChris Kirby ds->ds_userrefs--; 3847842727c2SChris Kirby refs = ds->ds_userrefs; 3848842727c2SChris Kirby mutex_exit(&ds->ds_lock); 3849ca45db41SChris Kirby error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx); 3850ca45db41SChris Kirby VERIFY(error == 0 || error == ENOENT); 3851842727c2SChris Kirby zapobj = ds->ds_phys->ds_userrefs_obj; 3852842727c2SChris Kirby VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx)); 3853842727c2SChris Kirby if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 && 3854842727c2SChris Kirby DS_IS_DEFER_DESTROY(ds)) { 3855842727c2SChris Kirby struct dsl_ds_destroyarg dsda = {0}; 3856842727c2SChris Kirby 3857842727c2SChris Kirby ASSERT(ra->own); 3858842727c2SChris Kirby dsda.ds = ds; 3859842727c2SChris Kirby dsda.releasing = B_TRUE; 3860842727c2SChris Kirby /* We already did the destroy_check */ 38613f9d6ad7SLin Ling dsl_dataset_destroy_sync(&dsda, tag, tx); 3862842727c2SChris Kirby } 3863842727c2SChris Kirby 38643f9d6ad7SLin Ling spa_history_log_internal(LOG_DS_USER_RELEASE, 38653f9d6ad7SLin Ling dp->dp_spa, tx, "<%s> %lld dataset = %llu", 3866842727c2SChris Kirby ra->htag, (longlong_t)refs, dsobj); 3867842727c2SChris Kirby } 3868842727c2SChris Kirby 3869842727c2SChris Kirby static int 3870fd136879SMatthew Ahrens dsl_dataset_user_release_one(const char *dsname, void *arg) 3871842727c2SChris Kirby { 3872842727c2SChris Kirby struct dsl_ds_holdarg *ha = arg; 3873842727c2SChris Kirby struct dsl_ds_releasearg *ra; 3874842727c2SChris Kirby dsl_dataset_t *ds; 3875842727c2SChris Kirby int error; 3876842727c2SChris Kirby void *dtag = ha->dstg; 3877842727c2SChris Kirby char *name; 3878842727c2SChris Kirby boolean_t own = B_FALSE; 3879842727c2SChris Kirby boolean_t might_destroy; 3880842727c2SChris Kirby 3881842727c2SChris Kirby /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */ 3882ae46e4c7SMatthew Ahrens name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3883842727c2SChris Kirby error = dsl_dataset_hold(name, dtag, &ds); 3884ae46e4c7SMatthew Ahrens strfree(name); 3885842727c2SChris Kirby if (error == ENOENT && ha->recursive) 3886842727c2SChris Kirby return (0); 3887fd136879SMatthew Ahrens (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3888842727c2SChris Kirby if (error) 3889842727c2SChris Kirby return (error); 3890842727c2SChris Kirby 3891d7747cbcSChris Kirby ha->gotone = B_TRUE; 3892d7747cbcSChris Kirby 3893842727c2SChris Kirby ASSERT(dsl_dataset_is_snapshot(ds)); 3894842727c2SChris Kirby 3895842727c2SChris Kirby error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy); 3896842727c2SChris Kirby if (error) { 3897842727c2SChris Kirby dsl_dataset_rele(ds, dtag); 3898842727c2SChris Kirby return (error); 3899842727c2SChris Kirby } 3900842727c2SChris Kirby 3901842727c2SChris Kirby if (might_destroy) { 3902842727c2SChris Kirby #ifdef _KERNEL 39035afc78aaSChris Kirby name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3904842727c2SChris Kirby error = zfs_unmount_snap(name, NULL); 39055afc78aaSChris Kirby strfree(name); 3906842727c2SChris Kirby if (error) { 3907842727c2SChris Kirby dsl_dataset_rele(ds, dtag); 3908842727c2SChris Kirby return (error); 3909842727c2SChris Kirby } 3910842727c2SChris Kirby #endif 3911503ad85cSMatthew Ahrens if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) { 3912842727c2SChris Kirby dsl_dataset_rele(ds, dtag); 3913842727c2SChris Kirby return (EBUSY); 3914842727c2SChris Kirby } else { 3915842727c2SChris Kirby own = B_TRUE; 3916842727c2SChris Kirby dsl_dataset_make_exclusive(ds, dtag); 3917842727c2SChris Kirby } 3918842727c2SChris Kirby } 3919842727c2SChris Kirby 3920842727c2SChris Kirby ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP); 3921842727c2SChris Kirby ra->ds = ds; 3922842727c2SChris Kirby ra->htag = ha->htag; 3923842727c2SChris Kirby ra->own = own; 3924842727c2SChris Kirby dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check, 3925842727c2SChris Kirby dsl_dataset_user_release_sync, ra, dtag, 0); 3926842727c2SChris Kirby 3927842727c2SChris Kirby return (0); 3928842727c2SChris Kirby } 3929842727c2SChris Kirby 3930842727c2SChris Kirby int 3931842727c2SChris Kirby dsl_dataset_user_release(char *dsname, char *snapname, char *htag, 3932842727c2SChris Kirby boolean_t recursive) 3933842727c2SChris Kirby { 3934842727c2SChris Kirby struct dsl_ds_holdarg *ha; 3935842727c2SChris Kirby dsl_sync_task_t *dst; 3936842727c2SChris Kirby spa_t *spa; 3937842727c2SChris Kirby int error; 3938842727c2SChris Kirby 3939620252bcSChris Kirby top: 3940842727c2SChris Kirby ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3941842727c2SChris Kirby 3942842727c2SChris Kirby (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3943842727c2SChris Kirby 3944842727c2SChris Kirby error = spa_open(dsname, &spa, FTAG); 3945842727c2SChris Kirby if (error) { 3946842727c2SChris Kirby kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3947842727c2SChris Kirby return (error); 3948842727c2SChris Kirby } 3949842727c2SChris Kirby 3950842727c2SChris Kirby ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 3951842727c2SChris Kirby ha->htag = htag; 3952842727c2SChris Kirby ha->snapname = snapname; 3953842727c2SChris Kirby ha->recursive = recursive; 3954842727c2SChris Kirby if (recursive) { 3955842727c2SChris Kirby error = dmu_objset_find(dsname, dsl_dataset_user_release_one, 3956842727c2SChris Kirby ha, DS_FIND_CHILDREN); 3957842727c2SChris Kirby } else { 3958842727c2SChris Kirby error = dsl_dataset_user_release_one(dsname, ha); 3959842727c2SChris Kirby } 3960842727c2SChris Kirby if (error == 0) 3961842727c2SChris Kirby error = dsl_sync_task_group_wait(ha->dstg); 3962842727c2SChris Kirby 3963842727c2SChris Kirby for (dst = list_head(&ha->dstg->dstg_tasks); dst; 3964842727c2SChris Kirby dst = list_next(&ha->dstg->dstg_tasks, dst)) { 3965842727c2SChris Kirby struct dsl_ds_releasearg *ra = dst->dst_arg1; 3966842727c2SChris Kirby dsl_dataset_t *ds = ra->ds; 3967842727c2SChris Kirby 3968842727c2SChris Kirby if (dst->dst_err) 3969842727c2SChris Kirby dsl_dataset_name(ds, ha->failed); 3970842727c2SChris Kirby 3971842727c2SChris Kirby if (ra->own) 3972842727c2SChris Kirby dsl_dataset_disown(ds, ha->dstg); 3973842727c2SChris Kirby else 3974842727c2SChris Kirby dsl_dataset_rele(ds, ha->dstg); 3975842727c2SChris Kirby 3976842727c2SChris Kirby kmem_free(ra, sizeof (struct dsl_ds_releasearg)); 3977842727c2SChris Kirby } 3978842727c2SChris Kirby 3979d7747cbcSChris Kirby if (error == 0 && recursive && !ha->gotone) 3980d7747cbcSChris Kirby error = ENOENT; 3981d7747cbcSChris Kirby 3982620252bcSChris Kirby if (error && error != EBUSY) 3983fd136879SMatthew Ahrens (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); 3984842727c2SChris Kirby 3985842727c2SChris Kirby dsl_sync_task_group_destroy(ha->dstg); 3986842727c2SChris Kirby kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3987842727c2SChris Kirby spa_close(spa, FTAG); 3988620252bcSChris Kirby 3989620252bcSChris Kirby /* 3990620252bcSChris Kirby * We can get EBUSY if we were racing with deferred destroy and 3991620252bcSChris Kirby * dsl_dataset_user_release_check() hadn't done the necessary 3992620252bcSChris Kirby * open context setup. We can also get EBUSY if we're racing 3993620252bcSChris Kirby * with destroy and that thread is the ds_owner. Either way 3994620252bcSChris Kirby * the busy condition should be transient, and we should retry 3995620252bcSChris Kirby * the release operation. 3996620252bcSChris Kirby */ 3997620252bcSChris Kirby if (error == EBUSY) 3998620252bcSChris Kirby goto top; 3999620252bcSChris Kirby 4000842727c2SChris Kirby return (error); 4001842727c2SChris Kirby } 4002842727c2SChris Kirby 4003ca45db41SChris Kirby /* 4004a7f53a56SChris Kirby * Called at spa_load time (with retry == B_FALSE) to release a stale 4005a7f53a56SChris Kirby * temporary user hold. Also called by the onexit code (with retry == B_TRUE). 4006ca45db41SChris Kirby */ 4007ca45db41SChris Kirby int 4008a7f53a56SChris Kirby dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag, 4009a7f53a56SChris Kirby boolean_t retry) 4010ca45db41SChris Kirby { 4011ca45db41SChris Kirby dsl_dataset_t *ds; 4012ca45db41SChris Kirby char *snap; 4013ca45db41SChris Kirby char *name; 4014ca45db41SChris Kirby int namelen; 4015ca45db41SChris Kirby int error; 4016ca45db41SChris Kirby 4017a7f53a56SChris Kirby do { 4018a7f53a56SChris Kirby rw_enter(&dp->dp_config_rwlock, RW_READER); 4019a7f53a56SChris Kirby error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); 4020a7f53a56SChris Kirby rw_exit(&dp->dp_config_rwlock); 4021a7f53a56SChris Kirby if (error) 4022a7f53a56SChris Kirby return (error); 4023a7f53a56SChris Kirby namelen = dsl_dataset_namelen(ds)+1; 4024a7f53a56SChris Kirby name = kmem_alloc(namelen, KM_SLEEP); 4025a7f53a56SChris Kirby dsl_dataset_name(ds, name); 4026a7f53a56SChris Kirby dsl_dataset_rele(ds, FTAG); 4027ca45db41SChris Kirby 4028a7f53a56SChris Kirby snap = strchr(name, '@'); 4029a7f53a56SChris Kirby *snap = '\0'; 4030a7f53a56SChris Kirby ++snap; 4031a7f53a56SChris Kirby error = dsl_dataset_user_release(name, snap, htag, B_FALSE); 4032a7f53a56SChris Kirby kmem_free(name, namelen); 4033a7f53a56SChris Kirby 4034a7f53a56SChris Kirby /* 4035a7f53a56SChris Kirby * The object can't have been destroyed because we have a hold, 4036a7f53a56SChris Kirby * but it might have been renamed, resulting in ENOENT. Retry 4037a7f53a56SChris Kirby * if we've been requested to do so. 4038a7f53a56SChris Kirby * 4039a7f53a56SChris Kirby * It would be nice if we could use the dsobj all the way 4040a7f53a56SChris Kirby * through and avoid ENOENT entirely. But we might need to 4041a7f53a56SChris Kirby * unmount the snapshot, and there's currently no way to lookup 4042a7f53a56SChris Kirby * a vfsp using a ZFS object id. 4043a7f53a56SChris Kirby */ 4044a7f53a56SChris Kirby } while ((error == ENOENT) && retry); 4045a7f53a56SChris Kirby 4046a7f53a56SChris Kirby return (error); 4047ca45db41SChris Kirby } 4048ca45db41SChris Kirby 4049842727c2SChris Kirby int 4050842727c2SChris Kirby dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp) 4051842727c2SChris Kirby { 4052842727c2SChris Kirby dsl_dataset_t *ds; 4053842727c2SChris Kirby int err; 4054842727c2SChris Kirby 4055842727c2SChris Kirby err = dsl_dataset_hold(dsname, FTAG, &ds); 4056842727c2SChris Kirby if (err) 4057842727c2SChris Kirby return (err); 4058842727c2SChris Kirby 4059842727c2SChris Kirby VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP)); 4060842727c2SChris Kirby if (ds->ds_phys->ds_userrefs_obj != 0) { 4061842727c2SChris Kirby zap_attribute_t *za; 4062842727c2SChris Kirby zap_cursor_t zc; 4063842727c2SChris Kirby 4064842727c2SChris Kirby za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 4065842727c2SChris Kirby for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset, 4066842727c2SChris Kirby ds->ds_phys->ds_userrefs_obj); 4067842727c2SChris Kirby zap_cursor_retrieve(&zc, za) == 0; 4068842727c2SChris Kirby zap_cursor_advance(&zc)) { 4069842727c2SChris Kirby VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name, 4070842727c2SChris Kirby za->za_first_integer)); 4071842727c2SChris Kirby } 4072842727c2SChris Kirby zap_cursor_fini(&zc); 4073842727c2SChris Kirby kmem_free(za, sizeof (zap_attribute_t)); 4074842727c2SChris Kirby } 4075842727c2SChris Kirby dsl_dataset_rele(ds, FTAG); 4076842727c2SChris Kirby return (0); 4077842727c2SChris Kirby } 4078503ad85cSMatthew Ahrens 4079503ad85cSMatthew Ahrens /* 408019b94df9SMatthew Ahrens * Note, this function is used as the callback for dmu_objset_find(). We 4081503ad85cSMatthew Ahrens * always return 0 so that we will continue to find and process 4082503ad85cSMatthew Ahrens * inconsistent datasets, even if we encounter an error trying to 4083503ad85cSMatthew Ahrens * process one of them. 4084503ad85cSMatthew Ahrens */ 4085503ad85cSMatthew Ahrens /* ARGSUSED */ 4086503ad85cSMatthew Ahrens int 4087fd136879SMatthew Ahrens dsl_destroy_inconsistent(const char *dsname, void *arg) 4088503ad85cSMatthew Ahrens { 4089503ad85cSMatthew Ahrens dsl_dataset_t *ds; 4090503ad85cSMatthew Ahrens 4091503ad85cSMatthew Ahrens if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) { 4092503ad85cSMatthew Ahrens if (DS_IS_INCONSISTENT(ds)) 4093503ad85cSMatthew Ahrens (void) dsl_dataset_destroy(ds, FTAG, B_FALSE); 4094503ad85cSMatthew Ahrens else 4095503ad85cSMatthew Ahrens dsl_dataset_disown(ds, FTAG); 4096503ad85cSMatthew Ahrens } 4097503ad85cSMatthew Ahrens return (0); 4098503ad85cSMatthew Ahrens } 409919b94df9SMatthew Ahrens 410019b94df9SMatthew Ahrens /* 410119b94df9SMatthew Ahrens * Return (in *usedp) the amount of space written in new that is not 410219b94df9SMatthew Ahrens * present in oldsnap. New may be a snapshot or the head. Old must be 410319b94df9SMatthew Ahrens * a snapshot before new, in new's filesystem (or its origin). If not then 410419b94df9SMatthew Ahrens * fail and return EINVAL. 410519b94df9SMatthew Ahrens * 410619b94df9SMatthew Ahrens * The written space is calculated by considering two components: First, we 410719b94df9SMatthew Ahrens * ignore any freed space, and calculate the written as new's used space 410819b94df9SMatthew Ahrens * minus old's used space. Next, we add in the amount of space that was freed 410919b94df9SMatthew Ahrens * between the two snapshots, thus reducing new's used space relative to old's. 411019b94df9SMatthew Ahrens * Specifically, this is the space that was born before old->ds_creation_txg, 411119b94df9SMatthew Ahrens * and freed before new (ie. on new's deadlist or a previous deadlist). 411219b94df9SMatthew Ahrens * 411319b94df9SMatthew Ahrens * space freed [---------------------] 411419b94df9SMatthew Ahrens * snapshots ---O-------O--------O-------O------ 411519b94df9SMatthew Ahrens * oldsnap new 411619b94df9SMatthew Ahrens */ 411719b94df9SMatthew Ahrens int 411819b94df9SMatthew Ahrens dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 411919b94df9SMatthew Ahrens uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 412019b94df9SMatthew Ahrens { 412119b94df9SMatthew Ahrens int err = 0; 412219b94df9SMatthew Ahrens uint64_t snapobj; 412319b94df9SMatthew Ahrens dsl_pool_t *dp = new->ds_dir->dd_pool; 412419b94df9SMatthew Ahrens 412519b94df9SMatthew Ahrens *usedp = 0; 412619b94df9SMatthew Ahrens *usedp += new->ds_phys->ds_used_bytes; 412719b94df9SMatthew Ahrens *usedp -= oldsnap->ds_phys->ds_used_bytes; 412819b94df9SMatthew Ahrens 412919b94df9SMatthew Ahrens *compp = 0; 413019b94df9SMatthew Ahrens *compp += new->ds_phys->ds_compressed_bytes; 413119b94df9SMatthew Ahrens *compp -= oldsnap->ds_phys->ds_compressed_bytes; 413219b94df9SMatthew Ahrens 413319b94df9SMatthew Ahrens *uncompp = 0; 413419b94df9SMatthew Ahrens *uncompp += new->ds_phys->ds_uncompressed_bytes; 413519b94df9SMatthew Ahrens *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes; 413619b94df9SMatthew Ahrens 413719b94df9SMatthew Ahrens rw_enter(&dp->dp_config_rwlock, RW_READER); 413819b94df9SMatthew Ahrens snapobj = new->ds_object; 413919b94df9SMatthew Ahrens while (snapobj != oldsnap->ds_object) { 414019b94df9SMatthew Ahrens dsl_dataset_t *snap; 414119b94df9SMatthew Ahrens uint64_t used, comp, uncomp; 414219b94df9SMatthew Ahrens 414319b94df9SMatthew Ahrens err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 414419b94df9SMatthew Ahrens if (err != 0) 414519b94df9SMatthew Ahrens break; 414619b94df9SMatthew Ahrens 414719b94df9SMatthew Ahrens if (snap->ds_phys->ds_prev_snap_txg == 414819b94df9SMatthew Ahrens oldsnap->ds_phys->ds_creation_txg) { 414919b94df9SMatthew Ahrens /* 415019b94df9SMatthew Ahrens * The blocks in the deadlist can not be born after 415119b94df9SMatthew Ahrens * ds_prev_snap_txg, so get the whole deadlist space, 415219b94df9SMatthew Ahrens * which is more efficient (especially for old-format 415319b94df9SMatthew Ahrens * deadlists). Unfortunately the deadlist code 415419b94df9SMatthew Ahrens * doesn't have enough information to make this 415519b94df9SMatthew Ahrens * optimization itself. 415619b94df9SMatthew Ahrens */ 415719b94df9SMatthew Ahrens dsl_deadlist_space(&snap->ds_deadlist, 415819b94df9SMatthew Ahrens &used, &comp, &uncomp); 415919b94df9SMatthew Ahrens } else { 416019b94df9SMatthew Ahrens dsl_deadlist_space_range(&snap->ds_deadlist, 416119b94df9SMatthew Ahrens 0, oldsnap->ds_phys->ds_creation_txg, 416219b94df9SMatthew Ahrens &used, &comp, &uncomp); 416319b94df9SMatthew Ahrens } 416419b94df9SMatthew Ahrens *usedp += used; 416519b94df9SMatthew Ahrens *compp += comp; 416619b94df9SMatthew Ahrens *uncompp += uncomp; 416719b94df9SMatthew Ahrens 416819b94df9SMatthew Ahrens /* 416919b94df9SMatthew Ahrens * If we get to the beginning of the chain of snapshots 417019b94df9SMatthew Ahrens * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 417119b94df9SMatthew Ahrens * was not a snapshot of/before new. 417219b94df9SMatthew Ahrens */ 417319b94df9SMatthew Ahrens snapobj = snap->ds_phys->ds_prev_snap_obj; 417419b94df9SMatthew Ahrens dsl_dataset_rele(snap, FTAG); 417519b94df9SMatthew Ahrens if (snapobj == 0) { 417619b94df9SMatthew Ahrens err = EINVAL; 417719b94df9SMatthew Ahrens break; 417819b94df9SMatthew Ahrens } 417919b94df9SMatthew Ahrens 418019b94df9SMatthew Ahrens } 418119b94df9SMatthew Ahrens rw_exit(&dp->dp_config_rwlock); 418219b94df9SMatthew Ahrens return (err); 418319b94df9SMatthew Ahrens } 418419b94df9SMatthew Ahrens 418519b94df9SMatthew Ahrens /* 418619b94df9SMatthew Ahrens * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 418719b94df9SMatthew Ahrens * lastsnap, and all snapshots in between are deleted. 418819b94df9SMatthew Ahrens * 418919b94df9SMatthew Ahrens * blocks that would be freed [---------------------------] 419019b94df9SMatthew Ahrens * snapshots ---O-------O--------O-------O--------O 419119b94df9SMatthew Ahrens * firstsnap lastsnap 419219b94df9SMatthew Ahrens * 419319b94df9SMatthew Ahrens * This is the set of blocks that were born after the snap before firstsnap, 419419b94df9SMatthew Ahrens * (birth > firstsnap->prev_snap_txg) and died before the snap after the 419519b94df9SMatthew Ahrens * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 419619b94df9SMatthew Ahrens * We calculate this by iterating over the relevant deadlists (from the snap 419719b94df9SMatthew Ahrens * after lastsnap, backward to the snap after firstsnap), summing up the 419819b94df9SMatthew Ahrens * space on the deadlist that was born after the snap before firstsnap. 419919b94df9SMatthew Ahrens */ 420019b94df9SMatthew Ahrens int 420119b94df9SMatthew Ahrens dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 420219b94df9SMatthew Ahrens dsl_dataset_t *lastsnap, 420319b94df9SMatthew Ahrens uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 420419b94df9SMatthew Ahrens { 420519b94df9SMatthew Ahrens int err = 0; 420619b94df9SMatthew Ahrens uint64_t snapobj; 420719b94df9SMatthew Ahrens dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 420819b94df9SMatthew Ahrens 420919b94df9SMatthew Ahrens ASSERT(dsl_dataset_is_snapshot(firstsnap)); 421019b94df9SMatthew Ahrens ASSERT(dsl_dataset_is_snapshot(lastsnap)); 421119b94df9SMatthew Ahrens 421219b94df9SMatthew Ahrens /* 421319b94df9SMatthew Ahrens * Check that the snapshots are in the same dsl_dir, and firstsnap 421419b94df9SMatthew Ahrens * is before lastsnap. 421519b94df9SMatthew Ahrens */ 421619b94df9SMatthew Ahrens if (firstsnap->ds_dir != lastsnap->ds_dir || 421719b94df9SMatthew Ahrens firstsnap->ds_phys->ds_creation_txg > 421819b94df9SMatthew Ahrens lastsnap->ds_phys->ds_creation_txg) 421919b94df9SMatthew Ahrens return (EINVAL); 422019b94df9SMatthew Ahrens 422119b94df9SMatthew Ahrens *usedp = *compp = *uncompp = 0; 422219b94df9SMatthew Ahrens 422319b94df9SMatthew Ahrens rw_enter(&dp->dp_config_rwlock, RW_READER); 422419b94df9SMatthew Ahrens snapobj = lastsnap->ds_phys->ds_next_snap_obj; 422519b94df9SMatthew Ahrens while (snapobj != firstsnap->ds_object) { 422619b94df9SMatthew Ahrens dsl_dataset_t *ds; 422719b94df9SMatthew Ahrens uint64_t used, comp, uncomp; 422819b94df9SMatthew Ahrens 422919b94df9SMatthew Ahrens err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 423019b94df9SMatthew Ahrens if (err != 0) 423119b94df9SMatthew Ahrens break; 423219b94df9SMatthew Ahrens 423319b94df9SMatthew Ahrens dsl_deadlist_space_range(&ds->ds_deadlist, 423419b94df9SMatthew Ahrens firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX, 423519b94df9SMatthew Ahrens &used, &comp, &uncomp); 423619b94df9SMatthew Ahrens *usedp += used; 423719b94df9SMatthew Ahrens *compp += comp; 423819b94df9SMatthew Ahrens *uncompp += uncomp; 423919b94df9SMatthew Ahrens 424019b94df9SMatthew Ahrens snapobj = ds->ds_phys->ds_prev_snap_obj; 424119b94df9SMatthew Ahrens ASSERT3U(snapobj, !=, 0); 424219b94df9SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 424319b94df9SMatthew Ahrens } 424419b94df9SMatthew Ahrens rw_exit(&dp->dp_config_rwlock); 424519b94df9SMatthew Ahrens return (err); 424619b94df9SMatthew Ahrens } 4247