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 /* 223f9d6ad7SLin Ling * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23ad135b5dSChristopher Siden * Copyright (c) 2012 by Delphix. All rights reserved. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens #include <sys/dsl_pool.h> 27fa9e4066Sahrens #include <sys/dsl_dataset.h> 283f9d6ad7SLin Ling #include <sys/dsl_prop.h> 29fa9e4066Sahrens #include <sys/dsl_dir.h> 301d452cf5Sahrens #include <sys/dsl_synctask.h> 313f9d6ad7SLin Ling #include <sys/dsl_scan.h> 323f9d6ad7SLin Ling #include <sys/dnode.h> 33fa9e4066Sahrens #include <sys/dmu_tx.h> 34fa9e4066Sahrens #include <sys/dmu_objset.h> 35fa9e4066Sahrens #include <sys/arc.h> 36fa9e4066Sahrens #include <sys/zap.h> 37c717a561Smaybee #include <sys/zio.h> 38fa9e4066Sahrens #include <sys/zfs_context.h> 39fa9e4066Sahrens #include <sys/fs/zfs.h> 40088f3894Sahrens #include <sys/zfs_znode.h> 41088f3894Sahrens #include <sys/spa_impl.h> 42cde58dbcSMatthew Ahrens #include <sys/dsl_deadlist.h> 43ad135b5dSChristopher Siden #include <sys/bptree.h> 44ad135b5dSChristopher Siden #include <sys/zfeature.h> 45ce636f8bSMatthew Ahrens #include <sys/zil_impl.h> 46fa9e4066Sahrens 471ab7f2deSmaybee int zfs_no_write_throttle = 0; 4805715f94SMark Maybee int zfs_write_limit_shift = 3; /* 1/8th of physical memory */ 4944ecc532SGeorge Wilson int zfs_txg_synctime_ms = 1000; /* target millisecs to sync a txg */ 5005715f94SMark Maybee 5105715f94SMark Maybee uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */ 5205715f94SMark Maybee uint64_t zfs_write_limit_max = 0; /* max data payload per txg */ 5305715f94SMark Maybee uint64_t zfs_write_limit_inflated = 0; 541ab7f2deSmaybee uint64_t zfs_write_limit_override = 0; 551ab7f2deSmaybee 5605715f94SMark Maybee kmutex_t zfs_write_limit_lock; 5705715f94SMark Maybee 5805715f94SMark Maybee static pgcnt_t old_physmem = 0; 59088f3894Sahrens 603f9d6ad7SLin Ling int 61088f3894Sahrens dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) 62fa9e4066Sahrens { 63fa9e4066Sahrens uint64_t obj; 64fa9e4066Sahrens int err; 65fa9e4066Sahrens 66fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, 67fa9e4066Sahrens dp->dp_root_dir->dd_phys->dd_child_dir_zapobj, 68088f3894Sahrens name, sizeof (obj), 1, &obj); 69ea8dc4b6Seschrock if (err) 70ea8dc4b6Seschrock return (err); 71fa9e4066Sahrens 72088f3894Sahrens return (dsl_dir_open_obj(dp, obj, name, dp, ddp)); 73fa9e4066Sahrens } 74fa9e4066Sahrens 75fa9e4066Sahrens static dsl_pool_t * 76fa9e4066Sahrens dsl_pool_open_impl(spa_t *spa, uint64_t txg) 77fa9e4066Sahrens { 78fa9e4066Sahrens dsl_pool_t *dp; 79fa9e4066Sahrens blkptr_t *bp = spa_get_rootblkptr(spa); 80fa9e4066Sahrens 81fa9e4066Sahrens dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); 82fa9e4066Sahrens dp->dp_spa = spa; 83fa9e4066Sahrens dp->dp_meta_rootbp = *bp; 845ad82045Snd rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL); 851ab7f2deSmaybee dp->dp_write_limit = zfs_write_limit_min; 86fa9e4066Sahrens txg_init(dp, txg); 87fa9e4066Sahrens 88fa9e4066Sahrens txg_list_create(&dp->dp_dirty_datasets, 89fa9e4066Sahrens offsetof(dsl_dataset_t, ds_dirty_link)); 90ce636f8bSMatthew Ahrens txg_list_create(&dp->dp_dirty_zilogs, 91ce636f8bSMatthew Ahrens offsetof(zilog_t, zl_dirty_link)); 92fa9e4066Sahrens txg_list_create(&dp->dp_dirty_dirs, 93fa9e4066Sahrens offsetof(dsl_dir_t, dd_dirty_link)); 941d452cf5Sahrens txg_list_create(&dp->dp_sync_tasks, 951d452cf5Sahrens offsetof(dsl_sync_task_group_t, dstg_node)); 96fa9e4066Sahrens 971ab7f2deSmaybee mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); 981ab7f2deSmaybee 999d3574bfSNeil Perrin dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri, 1009d3574bfSNeil Perrin 1, 4, 0); 1019d3574bfSNeil Perrin 102fa9e4066Sahrens return (dp); 103fa9e4066Sahrens } 104fa9e4066Sahrens 105ea8dc4b6Seschrock int 106ad135b5dSChristopher Siden dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) 107fa9e4066Sahrens { 108fa9e4066Sahrens int err; 109fa9e4066Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 110ad135b5dSChristopher Siden 111ad135b5dSChristopher Siden err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, 112ad135b5dSChristopher Siden &dp->dp_meta_objset); 113ad135b5dSChristopher Siden if (err != 0) 114ad135b5dSChristopher Siden dsl_pool_close(dp); 115ad135b5dSChristopher Siden else 116ad135b5dSChristopher Siden *dpp = dp; 117ad135b5dSChristopher Siden 118ad135b5dSChristopher Siden return (err); 119ad135b5dSChristopher Siden } 120ad135b5dSChristopher Siden 121ad135b5dSChristopher Siden int 122ad135b5dSChristopher Siden dsl_pool_open(dsl_pool_t *dp) 123ad135b5dSChristopher Siden { 124ad135b5dSChristopher Siden int err; 125088f3894Sahrens dsl_dir_t *dd; 126088f3894Sahrens dsl_dataset_t *ds; 127cde58dbcSMatthew Ahrens uint64_t obj; 128fa9e4066Sahrens 129ad135b5dSChristopher Siden rw_enter(&dp->dp_config_rwlock, RW_WRITER); 130fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 131fa9e4066Sahrens DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, 132fa9e4066Sahrens &dp->dp_root_dir_obj); 133ea8dc4b6Seschrock if (err) 134ea8dc4b6Seschrock goto out; 135ea8dc4b6Seschrock 136ea8dc4b6Seschrock err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, 137ea8dc4b6Seschrock NULL, dp, &dp->dp_root_dir); 138ea8dc4b6Seschrock if (err) 139ea8dc4b6Seschrock goto out; 140fa9e4066Sahrens 141088f3894Sahrens err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); 142ea8dc4b6Seschrock if (err) 143ea8dc4b6Seschrock goto out; 144ea8dc4b6Seschrock 145ad135b5dSChristopher Siden if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { 146088f3894Sahrens err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); 147088f3894Sahrens if (err) 148088f3894Sahrens goto out; 149088f3894Sahrens err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj, 150088f3894Sahrens FTAG, &ds); 1518f63aa46SLin Ling if (err == 0) { 1528f63aa46SLin Ling err = dsl_dataset_hold_obj(dp, 1538f63aa46SLin Ling ds->ds_phys->ds_prev_snap_obj, dp, 1548f63aa46SLin Ling &dp->dp_origin_snap); 1558f63aa46SLin Ling dsl_dataset_rele(ds, FTAG); 1568f63aa46SLin Ling } 1578f63aa46SLin Ling dsl_dir_close(dd, dp); 158088f3894Sahrens if (err) 159088f3894Sahrens goto out; 160088f3894Sahrens } 161088f3894Sahrens 162ad135b5dSChristopher Siden if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { 163cde58dbcSMatthew Ahrens err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, 164cde58dbcSMatthew Ahrens &dp->dp_free_dir); 165cde58dbcSMatthew Ahrens if (err) 166cde58dbcSMatthew Ahrens goto out; 167cde58dbcSMatthew Ahrens 168cde58dbcSMatthew Ahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 169cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); 170cde58dbcSMatthew Ahrens if (err) 171cde58dbcSMatthew Ahrens goto out; 172b420f3adSRichard Lowe VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj, 173cde58dbcSMatthew Ahrens dp->dp_meta_objset, obj)); 174cde58dbcSMatthew Ahrens } 175cde58dbcSMatthew Ahrens 176ad135b5dSChristopher Siden if (spa_feature_is_active(dp->dp_spa, 177ad135b5dSChristopher Siden &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) { 178ad135b5dSChristopher Siden err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 179ad135b5dSChristopher Siden DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, 180ad135b5dSChristopher Siden &dp->dp_bptree_obj); 181ad135b5dSChristopher Siden if (err != 0) 182ad135b5dSChristopher Siden goto out; 183ad135b5dSChristopher Siden } 184ad135b5dSChristopher Siden 185f1745736SMatthew Ahrens if (spa_feature_is_active(dp->dp_spa, 186f1745736SMatthew Ahrens &spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ])) { 187f1745736SMatthew Ahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 188f1745736SMatthew Ahrens DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, 189f1745736SMatthew Ahrens &dp->dp_empty_bpobj); 190f1745736SMatthew Ahrens if (err != 0) 191f1745736SMatthew Ahrens goto out; 192f1745736SMatthew Ahrens } 193f1745736SMatthew Ahrens 194ca45db41SChris Kirby err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 195ca45db41SChris Kirby DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, 196ca45db41SChris Kirby &dp->dp_tmp_userrefs_obj); 197ca45db41SChris Kirby if (err == ENOENT) 198ca45db41SChris Kirby err = 0; 199ca45db41SChris Kirby if (err) 200ca45db41SChris Kirby goto out; 201ca45db41SChris Kirby 202ad135b5dSChristopher Siden err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); 203088f3894Sahrens 204ea8dc4b6Seschrock out: 205fa9e4066Sahrens rw_exit(&dp->dp_config_rwlock); 206ea8dc4b6Seschrock return (err); 207fa9e4066Sahrens } 208fa9e4066Sahrens 209fa9e4066Sahrens void 210fa9e4066Sahrens dsl_pool_close(dsl_pool_t *dp) 211fa9e4066Sahrens { 212088f3894Sahrens /* drop our references from dsl_pool_open() */ 213088f3894Sahrens 214088f3894Sahrens /* 215088f3894Sahrens * Since we held the origin_snap from "syncing" context (which 216088f3894Sahrens * includes pool-opening context), it actually only got a "ref" 217088f3894Sahrens * and not a hold, so just drop that here. 218088f3894Sahrens */ 219088f3894Sahrens if (dp->dp_origin_snap) 220088f3894Sahrens dsl_dataset_drop_ref(dp->dp_origin_snap, dp); 221ea8dc4b6Seschrock if (dp->dp_mos_dir) 222ea8dc4b6Seschrock dsl_dir_close(dp->dp_mos_dir, dp); 223cde58dbcSMatthew Ahrens if (dp->dp_free_dir) 224cde58dbcSMatthew Ahrens dsl_dir_close(dp->dp_free_dir, dp); 225ea8dc4b6Seschrock if (dp->dp_root_dir) 226ea8dc4b6Seschrock dsl_dir_close(dp->dp_root_dir, dp); 227fa9e4066Sahrens 228cde58dbcSMatthew Ahrens bpobj_close(&dp->dp_free_bpobj); 229cde58dbcSMatthew Ahrens 230fa9e4066Sahrens /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ 231ea8dc4b6Seschrock if (dp->dp_meta_objset) 232503ad85cSMatthew Ahrens dmu_objset_evict(dp->dp_meta_objset); 233fa9e4066Sahrens 234fa9e4066Sahrens txg_list_destroy(&dp->dp_dirty_datasets); 235ce636f8bSMatthew Ahrens txg_list_destroy(&dp->dp_dirty_zilogs); 23654a91118SChris Kirby txg_list_destroy(&dp->dp_sync_tasks); 237fa9e4066Sahrens txg_list_destroy(&dp->dp_dirty_dirs); 238fa9e4066Sahrens 239874395d5Smaybee arc_flush(dp->dp_spa); 240fa9e4066Sahrens txg_fini(dp); 2413f9d6ad7SLin Ling dsl_scan_fini(dp); 2425ad82045Snd rw_destroy(&dp->dp_config_rwlock); 2431ab7f2deSmaybee mutex_destroy(&dp->dp_lock); 2449d3574bfSNeil Perrin taskq_destroy(dp->dp_vnrele_taskq); 24588b7b0f2SMatthew Ahrens if (dp->dp_blkstats) 24688b7b0f2SMatthew Ahrens kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); 247fa9e4066Sahrens kmem_free(dp, sizeof (dsl_pool_t)); 248fa9e4066Sahrens } 249fa9e4066Sahrens 250fa9e4066Sahrens dsl_pool_t * 2510a48a24eStimh dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) 252fa9e4066Sahrens { 253fa9e4066Sahrens int err; 254fa9e4066Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 255fa9e4066Sahrens dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); 256503ad85cSMatthew Ahrens objset_t *os; 257088f3894Sahrens dsl_dataset_t *ds; 258cde58dbcSMatthew Ahrens uint64_t obj; 259088f3894Sahrens 260088f3894Sahrens /* create and open the MOS (meta-objset) */ 261503ad85cSMatthew Ahrens dp->dp_meta_objset = dmu_objset_create_impl(spa, 262503ad85cSMatthew Ahrens NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); 263fa9e4066Sahrens 264fa9e4066Sahrens /* create the pool directory */ 265fa9e4066Sahrens err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 266fa9e4066Sahrens DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); 267*fb09f5aaSMadhav Suresh ASSERT0(err); 268fa9e4066Sahrens 2693f9d6ad7SLin Ling /* Initialize scan structures */ 270b420f3adSRichard Lowe VERIFY3U(0, ==, dsl_scan_init(dp, txg)); 2713f9d6ad7SLin Ling 272fa9e4066Sahrens /* create and open the root dir */ 273088f3894Sahrens dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); 274ea8dc4b6Seschrock VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj, 275ea8dc4b6Seschrock NULL, dp, &dp->dp_root_dir)); 276fa9e4066Sahrens 277fa9e4066Sahrens /* create and open the meta-objset dir */ 278088f3894Sahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); 279088f3894Sahrens VERIFY(0 == dsl_pool_open_special_dir(dp, 280088f3894Sahrens MOS_DIR_NAME, &dp->dp_mos_dir)); 281088f3894Sahrens 282cde58dbcSMatthew Ahrens if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { 283cde58dbcSMatthew Ahrens /* create and open the free dir */ 284cde58dbcSMatthew Ahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, 285cde58dbcSMatthew Ahrens FREE_DIR_NAME, tx); 286cde58dbcSMatthew Ahrens VERIFY(0 == dsl_pool_open_special_dir(dp, 287cde58dbcSMatthew Ahrens FREE_DIR_NAME, &dp->dp_free_dir)); 288cde58dbcSMatthew Ahrens 289cde58dbcSMatthew Ahrens /* create and open the free_bplist */ 290cde58dbcSMatthew Ahrens obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx); 291cde58dbcSMatthew Ahrens VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 292cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); 293b420f3adSRichard Lowe VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj, 294cde58dbcSMatthew Ahrens dp->dp_meta_objset, obj)); 295cde58dbcSMatthew Ahrens } 296cde58dbcSMatthew Ahrens 297088f3894Sahrens if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) 298088f3894Sahrens dsl_pool_create_origin(dp, tx); 299088f3894Sahrens 300088f3894Sahrens /* create the root dataset */ 301cde58dbcSMatthew Ahrens obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); 302088f3894Sahrens 303088f3894Sahrens /* create the root objset */ 304cde58dbcSMatthew Ahrens VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); 305503ad85cSMatthew Ahrens os = dmu_objset_create_impl(dp->dp_spa, ds, 306088f3894Sahrens dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); 307088f3894Sahrens #ifdef _KERNEL 308503ad85cSMatthew Ahrens zfs_create_fs(os, kcred, zplprops, tx); 309088f3894Sahrens #endif 310088f3894Sahrens dsl_dataset_rele(ds, FTAG); 311fa9e4066Sahrens 312fa9e4066Sahrens dmu_tx_commit(tx); 313fa9e4066Sahrens 314fa9e4066Sahrens return (dp); 315fa9e4066Sahrens } 316fa9e4066Sahrens 317ce636f8bSMatthew Ahrens /* 318ce636f8bSMatthew Ahrens * Account for the meta-objset space in its placeholder dsl_dir. 319ce636f8bSMatthew Ahrens */ 320ce636f8bSMatthew Ahrens void 321ce636f8bSMatthew Ahrens dsl_pool_mos_diduse_space(dsl_pool_t *dp, 322ce636f8bSMatthew Ahrens int64_t used, int64_t comp, int64_t uncomp) 323ce636f8bSMatthew Ahrens { 324ce636f8bSMatthew Ahrens ASSERT3U(comp, ==, uncomp); /* it's all metadata */ 325ce636f8bSMatthew Ahrens mutex_enter(&dp->dp_lock); 326ce636f8bSMatthew Ahrens dp->dp_mos_used_delta += used; 327ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta += comp; 328ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta += uncomp; 329ce636f8bSMatthew Ahrens mutex_exit(&dp->dp_lock); 330ce636f8bSMatthew Ahrens } 331ce636f8bSMatthew Ahrens 332cde58dbcSMatthew Ahrens static int 333cde58dbcSMatthew Ahrens deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 334cde58dbcSMatthew Ahrens { 335cde58dbcSMatthew Ahrens dsl_deadlist_t *dl = arg; 33619b94df9SMatthew Ahrens dsl_pool_t *dp = dmu_objset_pool(dl->dl_os); 33719b94df9SMatthew Ahrens rw_enter(&dp->dp_config_rwlock, RW_READER); 338cde58dbcSMatthew Ahrens dsl_deadlist_insert(dl, bp, tx); 33919b94df9SMatthew Ahrens rw_exit(&dp->dp_config_rwlock); 340cde58dbcSMatthew Ahrens return (0); 341cde58dbcSMatthew Ahrens } 342cde58dbcSMatthew Ahrens 343fa9e4066Sahrens void 344fa9e4066Sahrens dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) 345fa9e4066Sahrens { 346c717a561Smaybee zio_t *zio; 347fa9e4066Sahrens dmu_tx_t *tx; 348c717a561Smaybee dsl_dir_t *dd; 349c717a561Smaybee dsl_dataset_t *ds; 350503ad85cSMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 35105715f94SMark Maybee hrtime_t start, write_time; 35205715f94SMark Maybee uint64_t data_written; 353c717a561Smaybee int err; 354ce636f8bSMatthew Ahrens list_t synced_datasets; 355ce636f8bSMatthew Ahrens 356ce636f8bSMatthew Ahrens list_create(&synced_datasets, sizeof (dsl_dataset_t), 357ce636f8bSMatthew Ahrens offsetof(dsl_dataset_t, ds_synced_link)); 358fa9e4066Sahrens 3593f9d6ad7SLin Ling /* 3603f9d6ad7SLin Ling * We need to copy dp_space_towrite() before doing 3613f9d6ad7SLin Ling * dsl_sync_task_group_sync(), because 3623f9d6ad7SLin Ling * dsl_dataset_snapshot_reserve_space() will increase 3633f9d6ad7SLin Ling * dp_space_towrite but not actually write anything. 3643f9d6ad7SLin Ling */ 3653f9d6ad7SLin Ling data_written = dp->dp_space_towrite[txg & TXG_MASK]; 3663f9d6ad7SLin Ling 367fa9e4066Sahrens tx = dmu_tx_create_assigned(dp, txg); 368fa9e4066Sahrens 36905715f94SMark Maybee dp->dp_read_overhead = 0; 3700fd90d51SMark Maybee start = gethrtime(); 37114843421SMatthew Ahrens 372c717a561Smaybee zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 373c717a561Smaybee while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) { 37414843421SMatthew Ahrens /* 37514843421SMatthew Ahrens * We must not sync any non-MOS datasets twice, because 37614843421SMatthew Ahrens * we may have taken a snapshot of them. However, we 37714843421SMatthew Ahrens * may sync newly-created datasets on pass 2. 37814843421SMatthew Ahrens */ 37914843421SMatthew Ahrens ASSERT(!list_link_active(&ds->ds_synced_link)); 380ce636f8bSMatthew Ahrens list_insert_tail(&synced_datasets, ds); 381c717a561Smaybee dsl_dataset_sync(ds, zio, tx); 382c717a561Smaybee } 38305715f94SMark Maybee DTRACE_PROBE(pool_sync__1setup); 384c717a561Smaybee err = zio_wait(zio); 38514843421SMatthew Ahrens 38605715f94SMark Maybee write_time = gethrtime() - start; 387c717a561Smaybee ASSERT(err == 0); 38805715f94SMark Maybee DTRACE_PROBE(pool_sync__2rootzio); 389c717a561Smaybee 390ce636f8bSMatthew Ahrens /* 391ce636f8bSMatthew Ahrens * After the data blocks have been written (ensured by the zio_wait() 392ce636f8bSMatthew Ahrens * above), update the user/group space accounting. 393ce636f8bSMatthew Ahrens */ 394ce636f8bSMatthew Ahrens for (ds = list_head(&synced_datasets); ds; 395ce636f8bSMatthew Ahrens ds = list_next(&synced_datasets, ds)) 3960a586ceaSMark Shellenbaum dmu_objset_do_userquota_updates(ds->ds_objset, tx); 39714843421SMatthew Ahrens 39814843421SMatthew Ahrens /* 39914843421SMatthew Ahrens * Sync the datasets again to push out the changes due to 4003f9d6ad7SLin Ling * userspace updates. This must be done before we process the 401ce636f8bSMatthew Ahrens * sync tasks, so that any snapshots will have the correct 402ce636f8bSMatthew Ahrens * user accounting information (and we won't get confused 403ce636f8bSMatthew Ahrens * about which blocks are part of the snapshot). 40414843421SMatthew Ahrens */ 40514843421SMatthew Ahrens zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 40614843421SMatthew Ahrens while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) { 40714843421SMatthew Ahrens ASSERT(list_link_active(&ds->ds_synced_link)); 40814843421SMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, ds); 40914843421SMatthew Ahrens dsl_dataset_sync(ds, zio, tx); 41014843421SMatthew Ahrens } 41114843421SMatthew Ahrens err = zio_wait(zio); 41214843421SMatthew Ahrens 413b24ab676SJeff Bonwick /* 414ce636f8bSMatthew Ahrens * Now that the datasets have been completely synced, we can 415ce636f8bSMatthew Ahrens * clean up our in-memory structures accumulated while syncing: 416ce636f8bSMatthew Ahrens * 417ce636f8bSMatthew Ahrens * - move dead blocks from the pending deadlist to the on-disk deadlist 418ce636f8bSMatthew Ahrens * - clean up zil records 419ce636f8bSMatthew Ahrens * - release hold from dsl_dataset_dirty() 420b24ab676SJeff Bonwick */ 421ce636f8bSMatthew Ahrens while (ds = list_remove_head(&synced_datasets)) { 422ce636f8bSMatthew Ahrens objset_t *os = ds->ds_objset; 423cde58dbcSMatthew Ahrens bplist_iterate(&ds->ds_pending_deadlist, 424cde58dbcSMatthew Ahrens deadlist_enqueue_cb, &ds->ds_deadlist, tx); 425ce636f8bSMatthew Ahrens ASSERT(!dmu_objset_is_dirty(os, txg)); 426ce636f8bSMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, ds); 427cde58dbcSMatthew Ahrens } 428b24ab676SJeff Bonwick 42905715f94SMark Maybee start = gethrtime(); 430c717a561Smaybee while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) 431c717a561Smaybee dsl_dir_sync(dd, tx); 43205715f94SMark Maybee write_time += gethrtime() - start; 433fa9e4066Sahrens 434ce636f8bSMatthew Ahrens /* 435ce636f8bSMatthew Ahrens * The MOS's space is accounted for in the pool/$MOS 436ce636f8bSMatthew Ahrens * (dp_mos_dir). We can't modify the mos while we're syncing 437ce636f8bSMatthew Ahrens * it, so we remember the deltas and apply them here. 438ce636f8bSMatthew Ahrens */ 439ce636f8bSMatthew Ahrens if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || 440ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta != 0) { 441ce636f8bSMatthew Ahrens dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, 442ce636f8bSMatthew Ahrens dp->dp_mos_used_delta, 443ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta, 444ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta, tx); 445ce636f8bSMatthew Ahrens dp->dp_mos_used_delta = 0; 446ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta = 0; 447ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta = 0; 448ce636f8bSMatthew Ahrens } 449ce636f8bSMatthew Ahrens 45005715f94SMark Maybee start = gethrtime(); 451503ad85cSMatthew Ahrens if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || 452503ad85cSMatthew Ahrens list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { 453c717a561Smaybee zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 454503ad85cSMatthew Ahrens dmu_objset_sync(mos, zio, tx); 455c717a561Smaybee err = zio_wait(zio); 456c717a561Smaybee ASSERT(err == 0); 457fa9e4066Sahrens dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); 458fa9e4066Sahrens spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); 459fa9e4066Sahrens } 46005715f94SMark Maybee write_time += gethrtime() - start; 46105715f94SMark Maybee DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time, 46205715f94SMark Maybee hrtime_t, dp->dp_read_overhead); 46305715f94SMark Maybee write_time -= dp->dp_read_overhead; 464fa9e4066Sahrens 465ce636f8bSMatthew Ahrens /* 466ce636f8bSMatthew Ahrens * If we modify a dataset in the same txg that we want to destroy it, 467ce636f8bSMatthew Ahrens * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. 468ce636f8bSMatthew Ahrens * dsl_dir_destroy_check() will fail if there are unexpected holds. 469ce636f8bSMatthew Ahrens * Therefore, we want to sync the MOS (thus syncing the dd_dbuf 470ce636f8bSMatthew Ahrens * and clearing the hold on it) before we process the sync_tasks. 471ce636f8bSMatthew Ahrens * The MOS data dirtied by the sync_tasks will be synced on the next 472ce636f8bSMatthew Ahrens * pass. 473ce636f8bSMatthew Ahrens */ 474ce636f8bSMatthew Ahrens DTRACE_PROBE(pool_sync__3task); 475ce636f8bSMatthew Ahrens if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { 476ce636f8bSMatthew Ahrens dsl_sync_task_group_t *dstg; 477ce636f8bSMatthew Ahrens /* 478ce636f8bSMatthew Ahrens * No more sync tasks should have been added while we 479ce636f8bSMatthew Ahrens * were syncing. 480ce636f8bSMatthew Ahrens */ 481ce636f8bSMatthew Ahrens ASSERT(spa_sync_pass(dp->dp_spa) == 1); 482ce636f8bSMatthew Ahrens while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg)) 483ce636f8bSMatthew Ahrens dsl_sync_task_group_sync(dstg, tx); 484ce636f8bSMatthew Ahrens } 485ce636f8bSMatthew Ahrens 486fa9e4066Sahrens dmu_tx_commit(tx); 48705715f94SMark Maybee 48805715f94SMark Maybee dp->dp_space_towrite[txg & TXG_MASK] = 0; 48905715f94SMark Maybee ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0); 49005715f94SMark Maybee 49105715f94SMark Maybee /* 49205715f94SMark Maybee * If the write limit max has not been explicitly set, set it 49305715f94SMark Maybee * to a fraction of available physical memory (default 1/8th). 49405715f94SMark Maybee * Note that we must inflate the limit because the spa 49505715f94SMark Maybee * inflates write sizes to account for data replication. 49605715f94SMark Maybee * Check this each sync phase to catch changing memory size. 49705715f94SMark Maybee */ 49805715f94SMark Maybee if (physmem != old_physmem && zfs_write_limit_shift) { 49905715f94SMark Maybee mutex_enter(&zfs_write_limit_lock); 50005715f94SMark Maybee old_physmem = physmem; 50105715f94SMark Maybee zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift; 50205715f94SMark Maybee zfs_write_limit_inflated = MAX(zfs_write_limit_min, 50305715f94SMark Maybee spa_get_asize(dp->dp_spa, zfs_write_limit_max)); 50405715f94SMark Maybee mutex_exit(&zfs_write_limit_lock); 50505715f94SMark Maybee } 50605715f94SMark Maybee 50705715f94SMark Maybee /* 50805715f94SMark Maybee * Attempt to keep the sync time consistent by adjusting the 50905715f94SMark Maybee * amount of write traffic allowed into each transaction group. 51005715f94SMark Maybee * Weight the throughput calculation towards the current value: 51105715f94SMark Maybee * thru = 3/4 old_thru + 1/4 new_thru 512fb5dd802SLin Ling * 513fb5dd802SLin Ling * Note: write_time is in nanosecs, so write_time/MICROSEC 514fb5dd802SLin Ling * yields millisecs 51505715f94SMark Maybee */ 51605715f94SMark Maybee ASSERT(zfs_write_limit_min > 0); 517fb5dd802SLin Ling if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) { 518fb5dd802SLin Ling uint64_t throughput = data_written / (write_time / MICROSEC); 519fb5dd802SLin Ling 52005715f94SMark Maybee if (dp->dp_throughput) 52105715f94SMark Maybee dp->dp_throughput = throughput / 4 + 52205715f94SMark Maybee 3 * dp->dp_throughput / 4; 52305715f94SMark Maybee else 52405715f94SMark Maybee dp->dp_throughput = throughput; 52505715f94SMark Maybee dp->dp_write_limit = MIN(zfs_write_limit_inflated, 52605715f94SMark Maybee MAX(zfs_write_limit_min, 527fb5dd802SLin Ling dp->dp_throughput * zfs_txg_synctime_ms)); 52805715f94SMark Maybee } 529fa9e4066Sahrens } 530fa9e4066Sahrens 531fa9e4066Sahrens void 532b24ab676SJeff Bonwick dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) 533fa9e4066Sahrens { 534ce636f8bSMatthew Ahrens zilog_t *zilog; 535fa9e4066Sahrens dsl_dataset_t *ds; 536fa9e4066Sahrens 537ce636f8bSMatthew Ahrens while (zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg)) { 538ce636f8bSMatthew Ahrens ds = dmu_objset_ds(zilog->zl_os); 539ce636f8bSMatthew Ahrens zil_clean(zilog, txg); 540ce636f8bSMatthew Ahrens ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); 541ce636f8bSMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, zilog); 542fa9e4066Sahrens } 543b24ab676SJeff Bonwick ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); 544fa9e4066Sahrens } 545fa9e4066Sahrens 546c717a561Smaybee /* 547c717a561Smaybee * TRUE if the current thread is the tx_sync_thread or if we 548c717a561Smaybee * are being called from SPA context during pool initialization. 549c717a561Smaybee */ 550fa9e4066Sahrens int 551fa9e4066Sahrens dsl_pool_sync_context(dsl_pool_t *dp) 552fa9e4066Sahrens { 553fa9e4066Sahrens return (curthread == dp->dp_tx.tx_sync_thread || 554ad135b5dSChristopher Siden spa_is_initializing(dp->dp_spa)); 555fa9e4066Sahrens } 556fa9e4066Sahrens 557fa9e4066Sahrens uint64_t 558fa9e4066Sahrens dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) 559fa9e4066Sahrens { 560fa9e4066Sahrens uint64_t space, resv; 561fa9e4066Sahrens 562fa9e4066Sahrens /* 56344cd46caSbillm * Reserve about 1.6% (1/64), or at least 32MB, for allocation 564fa9e4066Sahrens * efficiency. 565fa9e4066Sahrens * XXX The intent log is not accounted for, so it must fit 566fa9e4066Sahrens * within this slop. 567fa9e4066Sahrens * 568fa9e4066Sahrens * If we're trying to assess whether it's OK to do a free, 569fa9e4066Sahrens * cut the reservation in half to allow forward progress 570fa9e4066Sahrens * (e.g. make it possible to rm(1) files from a full pool). 571fa9e4066Sahrens */ 572485bbbf5SGeorge Wilson space = spa_get_dspace(dp->dp_spa); 57344cd46caSbillm resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1); 574fa9e4066Sahrens if (netfree) 575fa9e4066Sahrens resv >>= 1; 576fa9e4066Sahrens 577fa9e4066Sahrens return (space - resv); 578fa9e4066Sahrens } 5791ab7f2deSmaybee 5801ab7f2deSmaybee int 5811ab7f2deSmaybee dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx) 5821ab7f2deSmaybee { 5831ab7f2deSmaybee uint64_t reserved = 0; 5841ab7f2deSmaybee uint64_t write_limit = (zfs_write_limit_override ? 5851ab7f2deSmaybee zfs_write_limit_override : dp->dp_write_limit); 5861ab7f2deSmaybee 5871ab7f2deSmaybee if (zfs_no_write_throttle) { 588c5904d13Seschrock atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], 589c5904d13Seschrock space); 5901ab7f2deSmaybee return (0); 5911ab7f2deSmaybee } 5921ab7f2deSmaybee 5931ab7f2deSmaybee /* 5941ab7f2deSmaybee * Check to see if we have exceeded the maximum allowed IO for 5951ab7f2deSmaybee * this transaction group. We can do this without locks since 5961ab7f2deSmaybee * a little slop here is ok. Note that we do the reserved check 5971ab7f2deSmaybee * with only half the requested reserve: this is because the 5981ab7f2deSmaybee * reserve requests are worst-case, and we really don't want to 5991ab7f2deSmaybee * throttle based off of worst-case estimates. 6001ab7f2deSmaybee */ 6011ab7f2deSmaybee if (write_limit > 0) { 6021ab7f2deSmaybee reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK] 6031ab7f2deSmaybee + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2; 6041ab7f2deSmaybee 6051ab7f2deSmaybee if (reserved && reserved > write_limit) 6061ab7f2deSmaybee return (ERESTART); 6071ab7f2deSmaybee } 6081ab7f2deSmaybee 6091ab7f2deSmaybee atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space); 6101ab7f2deSmaybee 6111ab7f2deSmaybee /* 6121ab7f2deSmaybee * If this transaction group is over 7/8ths capacity, delay 6131ab7f2deSmaybee * the caller 1 clock tick. This will slow down the "fill" 6141ab7f2deSmaybee * rate until the sync process can catch up with us. 6151ab7f2deSmaybee */ 616e8397a2bSgw if (reserved && reserved > (write_limit - (write_limit >> 3))) 6171ab7f2deSmaybee txg_delay(dp, tx->tx_txg, 1); 6181ab7f2deSmaybee 6191ab7f2deSmaybee return (0); 6201ab7f2deSmaybee } 6211ab7f2deSmaybee 6221ab7f2deSmaybee void 6231ab7f2deSmaybee dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 6241ab7f2deSmaybee { 6251ab7f2deSmaybee ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space); 6261ab7f2deSmaybee atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space); 6271ab7f2deSmaybee } 6281ab7f2deSmaybee 6291ab7f2deSmaybee void 6301ab7f2deSmaybee dsl_pool_memory_pressure(dsl_pool_t *dp) 6311ab7f2deSmaybee { 6321ab7f2deSmaybee uint64_t space_inuse = 0; 6331ab7f2deSmaybee int i; 6341ab7f2deSmaybee 6351ab7f2deSmaybee if (dp->dp_write_limit == zfs_write_limit_min) 6361ab7f2deSmaybee return; 6371ab7f2deSmaybee 6381ab7f2deSmaybee for (i = 0; i < TXG_SIZE; i++) { 6391ab7f2deSmaybee space_inuse += dp->dp_space_towrite[i]; 6401ab7f2deSmaybee space_inuse += dp->dp_tempreserved[i]; 6411ab7f2deSmaybee } 6421ab7f2deSmaybee dp->dp_write_limit = MAX(zfs_write_limit_min, 6431ab7f2deSmaybee MIN(dp->dp_write_limit, space_inuse / 4)); 6441ab7f2deSmaybee } 6451ab7f2deSmaybee 6461ab7f2deSmaybee void 6471ab7f2deSmaybee dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 6481ab7f2deSmaybee { 6491ab7f2deSmaybee if (space > 0) { 6501ab7f2deSmaybee mutex_enter(&dp->dp_lock); 6511ab7f2deSmaybee dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space; 6521ab7f2deSmaybee mutex_exit(&dp->dp_lock); 6531ab7f2deSmaybee } 6541ab7f2deSmaybee } 655088f3894Sahrens 656088f3894Sahrens /* ARGSUSED */ 657088f3894Sahrens static int 658088f3894Sahrens upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) 659088f3894Sahrens { 660088f3894Sahrens dmu_tx_t *tx = arg; 661088f3894Sahrens dsl_dataset_t *ds, *prev = NULL; 662088f3894Sahrens int err; 663088f3894Sahrens dsl_pool_t *dp = spa_get_dsl(spa); 664088f3894Sahrens 665088f3894Sahrens err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); 666088f3894Sahrens if (err) 667088f3894Sahrens return (err); 668088f3894Sahrens 669088f3894Sahrens while (ds->ds_phys->ds_prev_snap_obj != 0) { 670088f3894Sahrens err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, 671088f3894Sahrens FTAG, &prev); 672088f3894Sahrens if (err) { 673088f3894Sahrens dsl_dataset_rele(ds, FTAG); 674088f3894Sahrens return (err); 675088f3894Sahrens } 676088f3894Sahrens 677088f3894Sahrens if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) 678088f3894Sahrens break; 679088f3894Sahrens dsl_dataset_rele(ds, FTAG); 680088f3894Sahrens ds = prev; 681088f3894Sahrens prev = NULL; 682088f3894Sahrens } 683088f3894Sahrens 684088f3894Sahrens if (prev == NULL) { 685088f3894Sahrens prev = dp->dp_origin_snap; 686088f3894Sahrens 687088f3894Sahrens /* 688088f3894Sahrens * The $ORIGIN can't have any data, or the accounting 689088f3894Sahrens * will be wrong. 690088f3894Sahrens */ 691088f3894Sahrens ASSERT(prev->ds_phys->ds_bp.blk_birth == 0); 692088f3894Sahrens 693088f3894Sahrens /* The origin doesn't get attached to itself */ 694088f3894Sahrens if (ds->ds_object == prev->ds_object) { 695088f3894Sahrens dsl_dataset_rele(ds, FTAG); 696088f3894Sahrens return (0); 697088f3894Sahrens } 698088f3894Sahrens 699088f3894Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 700088f3894Sahrens ds->ds_phys->ds_prev_snap_obj = prev->ds_object; 701088f3894Sahrens ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg; 702088f3894Sahrens 703088f3894Sahrens dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 704088f3894Sahrens ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object; 705088f3894Sahrens 706088f3894Sahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 707088f3894Sahrens prev->ds_phys->ds_num_children++; 708088f3894Sahrens 709088f3894Sahrens if (ds->ds_phys->ds_next_snap_obj == 0) { 710088f3894Sahrens ASSERT(ds->ds_prev == NULL); 711088f3894Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, 712088f3894Sahrens ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 713088f3894Sahrens } 714088f3894Sahrens } 715088f3894Sahrens 716088f3894Sahrens ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object); 717088f3894Sahrens ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object); 718088f3894Sahrens 719088f3894Sahrens if (prev->ds_phys->ds_next_clones_obj == 0) { 720c33e334fSMatthew Ahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 721088f3894Sahrens prev->ds_phys->ds_next_clones_obj = 722088f3894Sahrens zap_create(dp->dp_meta_objset, 723088f3894Sahrens DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 724088f3894Sahrens } 725088f3894Sahrens VERIFY(0 == zap_add_int(dp->dp_meta_objset, 726088f3894Sahrens prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx)); 727088f3894Sahrens 728088f3894Sahrens dsl_dataset_rele(ds, FTAG); 729088f3894Sahrens if (prev != dp->dp_origin_snap) 730088f3894Sahrens dsl_dataset_rele(prev, FTAG); 731088f3894Sahrens return (0); 732088f3894Sahrens } 733088f3894Sahrens 734088f3894Sahrens void 735088f3894Sahrens dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) 736088f3894Sahrens { 737088f3894Sahrens ASSERT(dmu_tx_is_syncing(tx)); 738088f3894Sahrens ASSERT(dp->dp_origin_snap != NULL); 739088f3894Sahrens 740b420f3adSRichard Lowe VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb, 741c33e334fSMatthew Ahrens tx, DS_FIND_CHILDREN)); 742088f3894Sahrens } 743088f3894Sahrens 744cde58dbcSMatthew Ahrens /* ARGSUSED */ 745cde58dbcSMatthew Ahrens static int 746cde58dbcSMatthew Ahrens upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) 747cde58dbcSMatthew Ahrens { 748cde58dbcSMatthew Ahrens dmu_tx_t *tx = arg; 749cde58dbcSMatthew Ahrens dsl_dataset_t *ds; 750cde58dbcSMatthew Ahrens dsl_pool_t *dp = spa_get_dsl(spa); 751cde58dbcSMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 752cde58dbcSMatthew Ahrens 753b420f3adSRichard Lowe VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 754cde58dbcSMatthew Ahrens 755cde58dbcSMatthew Ahrens if (ds->ds_dir->dd_phys->dd_origin_obj) { 756cde58dbcSMatthew Ahrens dsl_dataset_t *origin; 757cde58dbcSMatthew Ahrens 758b420f3adSRichard Lowe VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, 759cde58dbcSMatthew Ahrens ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin)); 760cde58dbcSMatthew Ahrens 761cde58dbcSMatthew Ahrens if (origin->ds_dir->dd_phys->dd_clones == 0) { 762cde58dbcSMatthew Ahrens dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 763cde58dbcSMatthew Ahrens origin->ds_dir->dd_phys->dd_clones = zap_create(mos, 764cde58dbcSMatthew Ahrens DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 765cde58dbcSMatthew Ahrens } 766cde58dbcSMatthew Ahrens 767b420f3adSRichard Lowe VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 768cde58dbcSMatthew Ahrens origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); 769cde58dbcSMatthew Ahrens 770cde58dbcSMatthew Ahrens dsl_dataset_rele(origin, FTAG); 771cde58dbcSMatthew Ahrens } 772cde58dbcSMatthew Ahrens 773cde58dbcSMatthew Ahrens dsl_dataset_rele(ds, FTAG); 774cde58dbcSMatthew Ahrens return (0); 775cde58dbcSMatthew Ahrens } 776cde58dbcSMatthew Ahrens 777cde58dbcSMatthew Ahrens void 778cde58dbcSMatthew Ahrens dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) 779cde58dbcSMatthew Ahrens { 780cde58dbcSMatthew Ahrens ASSERT(dmu_tx_is_syncing(tx)); 781cde58dbcSMatthew Ahrens uint64_t obj; 782cde58dbcSMatthew Ahrens 783cde58dbcSMatthew Ahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); 784cde58dbcSMatthew Ahrens VERIFY(0 == dsl_pool_open_special_dir(dp, 785cde58dbcSMatthew Ahrens FREE_DIR_NAME, &dp->dp_free_dir)); 786cde58dbcSMatthew Ahrens 787cde58dbcSMatthew Ahrens /* 788cde58dbcSMatthew Ahrens * We can't use bpobj_alloc(), because spa_version() still 789cde58dbcSMatthew Ahrens * returns the old version, and we need a new-version bpobj with 790cde58dbcSMatthew Ahrens * subobj support. So call dmu_object_alloc() directly. 791cde58dbcSMatthew Ahrens */ 792cde58dbcSMatthew Ahrens obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, 793cde58dbcSMatthew Ahrens SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); 794b420f3adSRichard Lowe VERIFY3U(0, ==, zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 795cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); 796b420f3adSRichard Lowe VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj, 797cde58dbcSMatthew Ahrens dp->dp_meta_objset, obj)); 798cde58dbcSMatthew Ahrens 799b420f3adSRichard Lowe VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, 800cde58dbcSMatthew Ahrens upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN)); 801cde58dbcSMatthew Ahrens } 802cde58dbcSMatthew Ahrens 803088f3894Sahrens void 804088f3894Sahrens dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) 805088f3894Sahrens { 806088f3894Sahrens uint64_t dsobj; 807088f3894Sahrens dsl_dataset_t *ds; 808088f3894Sahrens 809088f3894Sahrens ASSERT(dmu_tx_is_syncing(tx)); 810088f3894Sahrens ASSERT(dp->dp_origin_snap == NULL); 811088f3894Sahrens 812088f3894Sahrens /* create the origin dir, ds, & snap-ds */ 813088f3894Sahrens rw_enter(&dp->dp_config_rwlock, RW_WRITER); 814088f3894Sahrens dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, 815088f3894Sahrens NULL, 0, kcred, tx); 816088f3894Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 8173f9d6ad7SLin Ling dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx); 818088f3894Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, 819088f3894Sahrens dp, &dp->dp_origin_snap)); 820088f3894Sahrens dsl_dataset_rele(ds, FTAG); 821088f3894Sahrens rw_exit(&dp->dp_config_rwlock); 822088f3894Sahrens } 8239d3574bfSNeil Perrin 8249d3574bfSNeil Perrin taskq_t * 8259d3574bfSNeil Perrin dsl_pool_vnrele_taskq(dsl_pool_t *dp) 8269d3574bfSNeil Perrin { 8279d3574bfSNeil Perrin return (dp->dp_vnrele_taskq); 8289d3574bfSNeil Perrin } 829ca45db41SChris Kirby 830ca45db41SChris Kirby /* 831ca45db41SChris Kirby * Walk through the pool-wide zap object of temporary snapshot user holds 832ca45db41SChris Kirby * and release them. 833ca45db41SChris Kirby */ 834ca45db41SChris Kirby void 835ca45db41SChris Kirby dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) 836ca45db41SChris Kirby { 837ca45db41SChris Kirby zap_attribute_t za; 838ca45db41SChris Kirby zap_cursor_t zc; 839ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 840ca45db41SChris Kirby uint64_t zapobj = dp->dp_tmp_userrefs_obj; 841ca45db41SChris Kirby 842ca45db41SChris Kirby if (zapobj == 0) 843ca45db41SChris Kirby return; 844ca45db41SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 845ca45db41SChris Kirby 846ca45db41SChris Kirby for (zap_cursor_init(&zc, mos, zapobj); 847ca45db41SChris Kirby zap_cursor_retrieve(&zc, &za) == 0; 848ca45db41SChris Kirby zap_cursor_advance(&zc)) { 849ca45db41SChris Kirby char *htag; 850ca45db41SChris Kirby uint64_t dsobj; 851ca45db41SChris Kirby 852ca45db41SChris Kirby htag = strchr(za.za_name, '-'); 853ca45db41SChris Kirby *htag = '\0'; 854ca45db41SChris Kirby ++htag; 855ca45db41SChris Kirby dsobj = strtonum(za.za_name, NULL); 856a7f53a56SChris Kirby (void) dsl_dataset_user_release_tmp(dp, dsobj, htag, B_FALSE); 857ca45db41SChris Kirby } 858ca45db41SChris Kirby zap_cursor_fini(&zc); 859ca45db41SChris Kirby } 860ca45db41SChris Kirby 861ca45db41SChris Kirby /* 862ca45db41SChris Kirby * Create the pool-wide zap object for storing temporary snapshot holds. 863ca45db41SChris Kirby */ 864ca45db41SChris Kirby void 865ca45db41SChris Kirby dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) 866ca45db41SChris Kirby { 867ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 868ca45db41SChris Kirby 869ca45db41SChris Kirby ASSERT(dp->dp_tmp_userrefs_obj == 0); 870ca45db41SChris Kirby ASSERT(dmu_tx_is_syncing(tx)); 871ca45db41SChris Kirby 872ad135b5dSChristopher Siden dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, 873ad135b5dSChristopher Siden DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); 874ca45db41SChris Kirby } 875ca45db41SChris Kirby 876ca45db41SChris Kirby static int 877ca45db41SChris Kirby dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, 87815508ac0SChris Kirby const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding) 879ca45db41SChris Kirby { 880ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 881ca45db41SChris Kirby uint64_t zapobj = dp->dp_tmp_userrefs_obj; 882ca45db41SChris Kirby char *name; 883ca45db41SChris Kirby int error; 884ca45db41SChris Kirby 885ca45db41SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 886ca45db41SChris Kirby ASSERT(dmu_tx_is_syncing(tx)); 887ca45db41SChris Kirby 888ca45db41SChris Kirby /* 889ca45db41SChris Kirby * If the pool was created prior to SPA_VERSION_USERREFS, the 890ca45db41SChris Kirby * zap object for temporary holds might not exist yet. 891ca45db41SChris Kirby */ 892ca45db41SChris Kirby if (zapobj == 0) { 893ca45db41SChris Kirby if (holding) { 894ca45db41SChris Kirby dsl_pool_user_hold_create_obj(dp, tx); 895ca45db41SChris Kirby zapobj = dp->dp_tmp_userrefs_obj; 896ca45db41SChris Kirby } else { 897ca45db41SChris Kirby return (ENOENT); 898ca45db41SChris Kirby } 899ca45db41SChris Kirby } 900ca45db41SChris Kirby 901ca45db41SChris Kirby name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); 902ca45db41SChris Kirby if (holding) 90315508ac0SChris Kirby error = zap_add(mos, zapobj, name, 8, 1, now, tx); 904ca45db41SChris Kirby else 905ca45db41SChris Kirby error = zap_remove(mos, zapobj, name, tx); 906ca45db41SChris Kirby strfree(name); 907ca45db41SChris Kirby 908ca45db41SChris Kirby return (error); 909ca45db41SChris Kirby } 910ca45db41SChris Kirby 911ca45db41SChris Kirby /* 912ca45db41SChris Kirby * Add a temporary hold for the given dataset object and tag. 913ca45db41SChris Kirby */ 914ca45db41SChris Kirby int 915ca45db41SChris Kirby dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 91615508ac0SChris Kirby uint64_t *now, dmu_tx_t *tx) 917ca45db41SChris Kirby { 91815508ac0SChris Kirby return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); 919ca45db41SChris Kirby } 920ca45db41SChris Kirby 921ca45db41SChris Kirby /* 922ca45db41SChris Kirby * Release a temporary hold for the given dataset object and tag. 923ca45db41SChris Kirby */ 924ca45db41SChris Kirby int 925ca45db41SChris Kirby dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 926ca45db41SChris Kirby dmu_tx_t *tx) 927ca45db41SChris Kirby { 928ca45db41SChris Kirby return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL, 929ca45db41SChris Kirby tx, B_FALSE)); 930ca45db41SChris Kirby } 931