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 /* 22*3f9d6ad7SLin Ling * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23fa9e4066Sahrens */ 24fa9e4066Sahrens 25fa9e4066Sahrens #include <sys/dsl_pool.h> 26fa9e4066Sahrens #include <sys/dsl_dataset.h> 27*3f9d6ad7SLin Ling #include <sys/dsl_prop.h> 28fa9e4066Sahrens #include <sys/dsl_dir.h> 291d452cf5Sahrens #include <sys/dsl_synctask.h> 30*3f9d6ad7SLin Ling #include <sys/dsl_scan.h> 31*3f9d6ad7SLin Ling #include <sys/dnode.h> 32fa9e4066Sahrens #include <sys/dmu_tx.h> 33fa9e4066Sahrens #include <sys/dmu_objset.h> 34fa9e4066Sahrens #include <sys/arc.h> 35fa9e4066Sahrens #include <sys/zap.h> 36c717a561Smaybee #include <sys/zio.h> 37fa9e4066Sahrens #include <sys/zfs_context.h> 38fa9e4066Sahrens #include <sys/fs/zfs.h> 39088f3894Sahrens #include <sys/zfs_znode.h> 40088f3894Sahrens #include <sys/spa_impl.h> 41fa9e4066Sahrens 421ab7f2deSmaybee int zfs_no_write_throttle = 0; 4305715f94SMark Maybee int zfs_write_limit_shift = 3; /* 1/8th of physical memory */ 44bf0ec83aSLin Ling int zfs_txg_synctime_ms = 5000; /* target millisecs to sync a txg */ 4505715f94SMark Maybee 4605715f94SMark Maybee uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */ 4705715f94SMark Maybee uint64_t zfs_write_limit_max = 0; /* max data payload per txg */ 4805715f94SMark Maybee uint64_t zfs_write_limit_inflated = 0; 491ab7f2deSmaybee uint64_t zfs_write_limit_override = 0; 501ab7f2deSmaybee 5105715f94SMark Maybee kmutex_t zfs_write_limit_lock; 5205715f94SMark Maybee 5305715f94SMark Maybee static pgcnt_t old_physmem = 0; 54088f3894Sahrens 55*3f9d6ad7SLin Ling int 56088f3894Sahrens dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) 57fa9e4066Sahrens { 58fa9e4066Sahrens uint64_t obj; 59fa9e4066Sahrens int err; 60fa9e4066Sahrens 61fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, 62fa9e4066Sahrens dp->dp_root_dir->dd_phys->dd_child_dir_zapobj, 63088f3894Sahrens name, sizeof (obj), 1, &obj); 64ea8dc4b6Seschrock if (err) 65ea8dc4b6Seschrock return (err); 66fa9e4066Sahrens 67088f3894Sahrens return (dsl_dir_open_obj(dp, obj, name, dp, ddp)); 68fa9e4066Sahrens } 69fa9e4066Sahrens 70fa9e4066Sahrens static dsl_pool_t * 71fa9e4066Sahrens dsl_pool_open_impl(spa_t *spa, uint64_t txg) 72fa9e4066Sahrens { 73fa9e4066Sahrens dsl_pool_t *dp; 74fa9e4066Sahrens blkptr_t *bp = spa_get_rootblkptr(spa); 75fa9e4066Sahrens 76fa9e4066Sahrens dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); 77fa9e4066Sahrens dp->dp_spa = spa; 78fa9e4066Sahrens dp->dp_meta_rootbp = *bp; 795ad82045Snd rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL); 801ab7f2deSmaybee dp->dp_write_limit = zfs_write_limit_min; 81fa9e4066Sahrens txg_init(dp, txg); 82fa9e4066Sahrens 83fa9e4066Sahrens txg_list_create(&dp->dp_dirty_datasets, 84fa9e4066Sahrens offsetof(dsl_dataset_t, ds_dirty_link)); 85fa9e4066Sahrens txg_list_create(&dp->dp_dirty_dirs, 86fa9e4066Sahrens offsetof(dsl_dir_t, dd_dirty_link)); 871d452cf5Sahrens txg_list_create(&dp->dp_sync_tasks, 881d452cf5Sahrens offsetof(dsl_sync_task_group_t, dstg_node)); 893cb34c60Sahrens list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t), 90fa9e4066Sahrens offsetof(dsl_dataset_t, ds_synced_link)); 91fa9e4066Sahrens 921ab7f2deSmaybee mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); 931ab7f2deSmaybee 949d3574bfSNeil Perrin dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri, 959d3574bfSNeil Perrin 1, 4, 0); 969d3574bfSNeil Perrin 97fa9e4066Sahrens return (dp); 98fa9e4066Sahrens } 99fa9e4066Sahrens 100ea8dc4b6Seschrock int 101ea8dc4b6Seschrock dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) 102fa9e4066Sahrens { 103fa9e4066Sahrens int err; 104fa9e4066Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 105088f3894Sahrens dsl_dir_t *dd; 106088f3894Sahrens dsl_dataset_t *ds; 107fa9e4066Sahrens 108088f3894Sahrens rw_enter(&dp->dp_config_rwlock, RW_WRITER); 109503ad85cSMatthew Ahrens err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, 110503ad85cSMatthew Ahrens &dp->dp_meta_objset); 111ea8dc4b6Seschrock if (err) 112ea8dc4b6Seschrock goto out; 113ea8dc4b6Seschrock 114fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 115fa9e4066Sahrens DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, 116fa9e4066Sahrens &dp->dp_root_dir_obj); 117ea8dc4b6Seschrock if (err) 118ea8dc4b6Seschrock goto out; 119ea8dc4b6Seschrock 120ea8dc4b6Seschrock err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, 121ea8dc4b6Seschrock NULL, dp, &dp->dp_root_dir); 122ea8dc4b6Seschrock if (err) 123ea8dc4b6Seschrock goto out; 124fa9e4066Sahrens 125088f3894Sahrens err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); 126ea8dc4b6Seschrock if (err) 127ea8dc4b6Seschrock goto out; 128ea8dc4b6Seschrock 129088f3894Sahrens if (spa_version(spa) >= SPA_VERSION_ORIGIN) { 130088f3894Sahrens err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); 131088f3894Sahrens if (err) 132088f3894Sahrens goto out; 133088f3894Sahrens err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj, 134088f3894Sahrens FTAG, &ds); 1358f63aa46SLin Ling if (err == 0) { 1368f63aa46SLin Ling err = dsl_dataset_hold_obj(dp, 1378f63aa46SLin Ling ds->ds_phys->ds_prev_snap_obj, dp, 1388f63aa46SLin Ling &dp->dp_origin_snap); 1398f63aa46SLin Ling dsl_dataset_rele(ds, FTAG); 1408f63aa46SLin Ling } 1418f63aa46SLin Ling dsl_dir_close(dd, dp); 142088f3894Sahrens if (err) 143088f3894Sahrens goto out; 144088f3894Sahrens } 145088f3894Sahrens 146ca45db41SChris Kirby err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 147ca45db41SChris Kirby DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, 148ca45db41SChris Kirby &dp->dp_tmp_userrefs_obj); 149ca45db41SChris Kirby if (err == ENOENT) 150ca45db41SChris Kirby err = 0; 151ca45db41SChris Kirby if (err) 152ca45db41SChris Kirby goto out; 153ca45db41SChris Kirby 154*3f9d6ad7SLin Ling err = dsl_scan_init(dp, txg); 155088f3894Sahrens 156ea8dc4b6Seschrock out: 157fa9e4066Sahrens rw_exit(&dp->dp_config_rwlock); 158ea8dc4b6Seschrock if (err) 159ea8dc4b6Seschrock dsl_pool_close(dp); 160ea8dc4b6Seschrock else 161ea8dc4b6Seschrock *dpp = dp; 162fa9e4066Sahrens 163ea8dc4b6Seschrock return (err); 164fa9e4066Sahrens } 165fa9e4066Sahrens 166fa9e4066Sahrens void 167fa9e4066Sahrens dsl_pool_close(dsl_pool_t *dp) 168fa9e4066Sahrens { 169088f3894Sahrens /* drop our references from dsl_pool_open() */ 170088f3894Sahrens 171088f3894Sahrens /* 172088f3894Sahrens * Since we held the origin_snap from "syncing" context (which 173088f3894Sahrens * includes pool-opening context), it actually only got a "ref" 174088f3894Sahrens * and not a hold, so just drop that here. 175088f3894Sahrens */ 176088f3894Sahrens if (dp->dp_origin_snap) 177088f3894Sahrens dsl_dataset_drop_ref(dp->dp_origin_snap, dp); 178ea8dc4b6Seschrock if (dp->dp_mos_dir) 179ea8dc4b6Seschrock dsl_dir_close(dp->dp_mos_dir, dp); 180ea8dc4b6Seschrock if (dp->dp_root_dir) 181ea8dc4b6Seschrock dsl_dir_close(dp->dp_root_dir, dp); 182fa9e4066Sahrens 183fa9e4066Sahrens /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ 184ea8dc4b6Seschrock if (dp->dp_meta_objset) 185503ad85cSMatthew Ahrens dmu_objset_evict(dp->dp_meta_objset); 186fa9e4066Sahrens 187fa9e4066Sahrens txg_list_destroy(&dp->dp_dirty_datasets); 18854a91118SChris Kirby txg_list_destroy(&dp->dp_sync_tasks); 189fa9e4066Sahrens txg_list_destroy(&dp->dp_dirty_dirs); 1903cb34c60Sahrens list_destroy(&dp->dp_synced_datasets); 191fa9e4066Sahrens 192874395d5Smaybee arc_flush(dp->dp_spa); 193fa9e4066Sahrens txg_fini(dp); 194*3f9d6ad7SLin Ling dsl_scan_fini(dp); 1955ad82045Snd rw_destroy(&dp->dp_config_rwlock); 1961ab7f2deSmaybee mutex_destroy(&dp->dp_lock); 1979d3574bfSNeil Perrin taskq_destroy(dp->dp_vnrele_taskq); 19888b7b0f2SMatthew Ahrens if (dp->dp_blkstats) 19988b7b0f2SMatthew Ahrens kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); 200fa9e4066Sahrens kmem_free(dp, sizeof (dsl_pool_t)); 201fa9e4066Sahrens } 202fa9e4066Sahrens 203fa9e4066Sahrens dsl_pool_t * 2040a48a24eStimh dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) 205fa9e4066Sahrens { 206fa9e4066Sahrens int err; 207fa9e4066Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 208fa9e4066Sahrens dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); 209503ad85cSMatthew Ahrens objset_t *os; 210088f3894Sahrens dsl_dataset_t *ds; 211088f3894Sahrens uint64_t dsobj; 212088f3894Sahrens 213088f3894Sahrens /* create and open the MOS (meta-objset) */ 214503ad85cSMatthew Ahrens dp->dp_meta_objset = dmu_objset_create_impl(spa, 215503ad85cSMatthew Ahrens NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); 216fa9e4066Sahrens 217fa9e4066Sahrens /* create the pool directory */ 218fa9e4066Sahrens err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 219fa9e4066Sahrens DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); 220fa9e4066Sahrens ASSERT3U(err, ==, 0); 221fa9e4066Sahrens 222*3f9d6ad7SLin Ling /* Initialize scan structures */ 223*3f9d6ad7SLin Ling VERIFY3U(0, ==, dsl_scan_init(dp, txg)); 224*3f9d6ad7SLin Ling 225fa9e4066Sahrens /* create and open the root dir */ 226088f3894Sahrens dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); 227ea8dc4b6Seschrock VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj, 228ea8dc4b6Seschrock NULL, dp, &dp->dp_root_dir)); 229fa9e4066Sahrens 230fa9e4066Sahrens /* create and open the meta-objset dir */ 231088f3894Sahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); 232088f3894Sahrens VERIFY(0 == dsl_pool_open_special_dir(dp, 233088f3894Sahrens MOS_DIR_NAME, &dp->dp_mos_dir)); 234088f3894Sahrens 235088f3894Sahrens if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) 236088f3894Sahrens dsl_pool_create_origin(dp, tx); 237088f3894Sahrens 238088f3894Sahrens /* create the root dataset */ 239088f3894Sahrens dsobj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); 240088f3894Sahrens 241088f3894Sahrens /* create the root objset */ 242088f3894Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 243503ad85cSMatthew Ahrens os = dmu_objset_create_impl(dp->dp_spa, ds, 244088f3894Sahrens dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); 245088f3894Sahrens #ifdef _KERNEL 246503ad85cSMatthew Ahrens zfs_create_fs(os, kcred, zplprops, tx); 247088f3894Sahrens #endif 248088f3894Sahrens dsl_dataset_rele(ds, FTAG); 249fa9e4066Sahrens 250fa9e4066Sahrens dmu_tx_commit(tx); 251fa9e4066Sahrens 252fa9e4066Sahrens return (dp); 253fa9e4066Sahrens } 254fa9e4066Sahrens 255fa9e4066Sahrens void 256fa9e4066Sahrens dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) 257fa9e4066Sahrens { 258c717a561Smaybee zio_t *zio; 259fa9e4066Sahrens dmu_tx_t *tx; 260c717a561Smaybee dsl_dir_t *dd; 261c717a561Smaybee dsl_dataset_t *ds; 262c717a561Smaybee dsl_sync_task_group_t *dstg; 263503ad85cSMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 26405715f94SMark Maybee hrtime_t start, write_time; 26505715f94SMark Maybee uint64_t data_written; 266c717a561Smaybee int err; 267fa9e4066Sahrens 268*3f9d6ad7SLin Ling /* 269*3f9d6ad7SLin Ling * We need to copy dp_space_towrite() before doing 270*3f9d6ad7SLin Ling * dsl_sync_task_group_sync(), because 271*3f9d6ad7SLin Ling * dsl_dataset_snapshot_reserve_space() will increase 272*3f9d6ad7SLin Ling * dp_space_towrite but not actually write anything. 273*3f9d6ad7SLin Ling */ 274*3f9d6ad7SLin Ling data_written = dp->dp_space_towrite[txg & TXG_MASK]; 275*3f9d6ad7SLin Ling 276fa9e4066Sahrens tx = dmu_tx_create_assigned(dp, txg); 277fa9e4066Sahrens 27805715f94SMark Maybee dp->dp_read_overhead = 0; 2790fd90d51SMark Maybee start = gethrtime(); 28014843421SMatthew Ahrens 281c717a561Smaybee zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 282c717a561Smaybee while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) { 28314843421SMatthew Ahrens /* 28414843421SMatthew Ahrens * We must not sync any non-MOS datasets twice, because 28514843421SMatthew Ahrens * we may have taken a snapshot of them. However, we 28614843421SMatthew Ahrens * may sync newly-created datasets on pass 2. 28714843421SMatthew Ahrens */ 28814843421SMatthew Ahrens ASSERT(!list_link_active(&ds->ds_synced_link)); 28914843421SMatthew Ahrens list_insert_tail(&dp->dp_synced_datasets, ds); 290c717a561Smaybee dsl_dataset_sync(ds, zio, tx); 291c717a561Smaybee } 29205715f94SMark Maybee DTRACE_PROBE(pool_sync__1setup); 293c717a561Smaybee err = zio_wait(zio); 29414843421SMatthew Ahrens 29505715f94SMark Maybee write_time = gethrtime() - start; 296c717a561Smaybee ASSERT(err == 0); 29705715f94SMark Maybee DTRACE_PROBE(pool_sync__2rootzio); 298c717a561Smaybee 29914843421SMatthew Ahrens for (ds = list_head(&dp->dp_synced_datasets); ds; 30014843421SMatthew Ahrens ds = list_next(&dp->dp_synced_datasets, ds)) 3010a586ceaSMark Shellenbaum dmu_objset_do_userquota_updates(ds->ds_objset, tx); 30214843421SMatthew Ahrens 30314843421SMatthew Ahrens /* 30414843421SMatthew Ahrens * Sync the datasets again to push out the changes due to 305*3f9d6ad7SLin Ling * userspace updates. This must be done before we process the 30614843421SMatthew Ahrens * sync tasks, because that could cause a snapshot of a dataset 30714843421SMatthew Ahrens * whose ds_bp will be rewritten when we do this 2nd sync. 30814843421SMatthew Ahrens */ 30914843421SMatthew Ahrens zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 31014843421SMatthew Ahrens while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) { 31114843421SMatthew Ahrens ASSERT(list_link_active(&ds->ds_synced_link)); 31214843421SMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, ds); 31314843421SMatthew Ahrens dsl_dataset_sync(ds, zio, tx); 31414843421SMatthew Ahrens } 31514843421SMatthew Ahrens err = zio_wait(zio); 31614843421SMatthew Ahrens 317b24ab676SJeff Bonwick /* 318b24ab676SJeff Bonwick * If anything was added to a deadlist during a zio done callback, 319b24ab676SJeff Bonwick * it had to be put on the deferred queue. Enqueue it for real now. 320b24ab676SJeff Bonwick */ 321b24ab676SJeff Bonwick for (ds = list_head(&dp->dp_synced_datasets); ds; 322b24ab676SJeff Bonwick ds = list_next(&dp->dp_synced_datasets, ds)) 323b24ab676SJeff Bonwick bplist_sync(&ds->ds_deadlist, 324b24ab676SJeff Bonwick bplist_enqueue_cb, &ds->ds_deadlist, tx); 325b24ab676SJeff Bonwick 32614843421SMatthew Ahrens while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg)) { 32714843421SMatthew Ahrens /* 32814843421SMatthew Ahrens * No more sync tasks should have been added while we 32914843421SMatthew Ahrens * were syncing. 33014843421SMatthew Ahrens */ 33114843421SMatthew Ahrens ASSERT(spa_sync_pass(dp->dp_spa) == 1); 332c717a561Smaybee dsl_sync_task_group_sync(dstg, tx); 33314843421SMatthew Ahrens } 33405715f94SMark Maybee DTRACE_PROBE(pool_sync__3task); 33505715f94SMark Maybee 33605715f94SMark Maybee start = gethrtime(); 337c717a561Smaybee while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) 338c717a561Smaybee dsl_dir_sync(dd, tx); 33905715f94SMark Maybee write_time += gethrtime() - start; 340fa9e4066Sahrens 34105715f94SMark Maybee start = gethrtime(); 342503ad85cSMatthew Ahrens if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || 343503ad85cSMatthew Ahrens list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { 344c717a561Smaybee zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 345503ad85cSMatthew Ahrens dmu_objset_sync(mos, zio, tx); 346c717a561Smaybee err = zio_wait(zio); 347c717a561Smaybee ASSERT(err == 0); 348fa9e4066Sahrens dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); 349fa9e4066Sahrens spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); 350fa9e4066Sahrens } 35105715f94SMark Maybee write_time += gethrtime() - start; 35205715f94SMark Maybee DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time, 35305715f94SMark Maybee hrtime_t, dp->dp_read_overhead); 35405715f94SMark Maybee write_time -= dp->dp_read_overhead; 355fa9e4066Sahrens 356fa9e4066Sahrens dmu_tx_commit(tx); 35705715f94SMark Maybee 35805715f94SMark Maybee dp->dp_space_towrite[txg & TXG_MASK] = 0; 35905715f94SMark Maybee ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0); 36005715f94SMark Maybee 36105715f94SMark Maybee /* 36205715f94SMark Maybee * If the write limit max has not been explicitly set, set it 36305715f94SMark Maybee * to a fraction of available physical memory (default 1/8th). 36405715f94SMark Maybee * Note that we must inflate the limit because the spa 36505715f94SMark Maybee * inflates write sizes to account for data replication. 36605715f94SMark Maybee * Check this each sync phase to catch changing memory size. 36705715f94SMark Maybee */ 36805715f94SMark Maybee if (physmem != old_physmem && zfs_write_limit_shift) { 36905715f94SMark Maybee mutex_enter(&zfs_write_limit_lock); 37005715f94SMark Maybee old_physmem = physmem; 37105715f94SMark Maybee zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift; 37205715f94SMark Maybee zfs_write_limit_inflated = MAX(zfs_write_limit_min, 37305715f94SMark Maybee spa_get_asize(dp->dp_spa, zfs_write_limit_max)); 37405715f94SMark Maybee mutex_exit(&zfs_write_limit_lock); 37505715f94SMark Maybee } 37605715f94SMark Maybee 37705715f94SMark Maybee /* 37805715f94SMark Maybee * Attempt to keep the sync time consistent by adjusting the 37905715f94SMark Maybee * amount of write traffic allowed into each transaction group. 38005715f94SMark Maybee * Weight the throughput calculation towards the current value: 38105715f94SMark Maybee * thru = 3/4 old_thru + 1/4 new_thru 382fb5dd802SLin Ling * 383fb5dd802SLin Ling * Note: write_time is in nanosecs, so write_time/MICROSEC 384fb5dd802SLin Ling * yields millisecs 38505715f94SMark Maybee */ 38605715f94SMark Maybee ASSERT(zfs_write_limit_min > 0); 387fb5dd802SLin Ling if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) { 388fb5dd802SLin Ling uint64_t throughput = data_written / (write_time / MICROSEC); 389fb5dd802SLin Ling 39005715f94SMark Maybee if (dp->dp_throughput) 39105715f94SMark Maybee dp->dp_throughput = throughput / 4 + 39205715f94SMark Maybee 3 * dp->dp_throughput / 4; 39305715f94SMark Maybee else 39405715f94SMark Maybee dp->dp_throughput = throughput; 39505715f94SMark Maybee dp->dp_write_limit = MIN(zfs_write_limit_inflated, 39605715f94SMark Maybee MAX(zfs_write_limit_min, 397fb5dd802SLin Ling dp->dp_throughput * zfs_txg_synctime_ms)); 39805715f94SMark Maybee } 399fa9e4066Sahrens } 400fa9e4066Sahrens 401fa9e4066Sahrens void 402b24ab676SJeff Bonwick dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) 403fa9e4066Sahrens { 404fa9e4066Sahrens dsl_dataset_t *ds; 405b24ab676SJeff Bonwick objset_t *os; 406fa9e4066Sahrens 4073cb34c60Sahrens while (ds = list_head(&dp->dp_synced_datasets)) { 4083cb34c60Sahrens list_remove(&dp->dp_synced_datasets, ds); 409b24ab676SJeff Bonwick os = ds->ds_objset; 410b24ab676SJeff Bonwick zil_clean(os->os_zil); 411b24ab676SJeff Bonwick ASSERT(!dmu_objset_is_dirty(os, txg)); 412af2c4821Smaybee dmu_buf_rele(ds->ds_dbuf, ds); 413fa9e4066Sahrens } 414b24ab676SJeff Bonwick ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); 415fa9e4066Sahrens } 416fa9e4066Sahrens 417c717a561Smaybee /* 418c717a561Smaybee * TRUE if the current thread is the tx_sync_thread or if we 419c717a561Smaybee * are being called from SPA context during pool initialization. 420c717a561Smaybee */ 421fa9e4066Sahrens int 422fa9e4066Sahrens dsl_pool_sync_context(dsl_pool_t *dp) 423fa9e4066Sahrens { 424fa9e4066Sahrens return (curthread == dp->dp_tx.tx_sync_thread || 425c717a561Smaybee spa_get_dsl(dp->dp_spa) == NULL); 426fa9e4066Sahrens } 427fa9e4066Sahrens 428fa9e4066Sahrens uint64_t 429fa9e4066Sahrens dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) 430fa9e4066Sahrens { 431fa9e4066Sahrens uint64_t space, resv; 432fa9e4066Sahrens 433fa9e4066Sahrens /* 43444cd46caSbillm * Reserve about 1.6% (1/64), or at least 32MB, for allocation 435fa9e4066Sahrens * efficiency. 436fa9e4066Sahrens * XXX The intent log is not accounted for, so it must fit 437fa9e4066Sahrens * within this slop. 438fa9e4066Sahrens * 439fa9e4066Sahrens * If we're trying to assess whether it's OK to do a free, 440fa9e4066Sahrens * cut the reservation in half to allow forward progress 441fa9e4066Sahrens * (e.g. make it possible to rm(1) files from a full pool). 442fa9e4066Sahrens */ 443485bbbf5SGeorge Wilson space = spa_get_dspace(dp->dp_spa); 44444cd46caSbillm resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1); 445fa9e4066Sahrens if (netfree) 446fa9e4066Sahrens resv >>= 1; 447fa9e4066Sahrens 448fa9e4066Sahrens return (space - resv); 449fa9e4066Sahrens } 4501ab7f2deSmaybee 4511ab7f2deSmaybee int 4521ab7f2deSmaybee dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx) 4531ab7f2deSmaybee { 4541ab7f2deSmaybee uint64_t reserved = 0; 4551ab7f2deSmaybee uint64_t write_limit = (zfs_write_limit_override ? 4561ab7f2deSmaybee zfs_write_limit_override : dp->dp_write_limit); 4571ab7f2deSmaybee 4581ab7f2deSmaybee if (zfs_no_write_throttle) { 459c5904d13Seschrock atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], 460c5904d13Seschrock space); 4611ab7f2deSmaybee return (0); 4621ab7f2deSmaybee } 4631ab7f2deSmaybee 4641ab7f2deSmaybee /* 4651ab7f2deSmaybee * Check to see if we have exceeded the maximum allowed IO for 4661ab7f2deSmaybee * this transaction group. We can do this without locks since 4671ab7f2deSmaybee * a little slop here is ok. Note that we do the reserved check 4681ab7f2deSmaybee * with only half the requested reserve: this is because the 4691ab7f2deSmaybee * reserve requests are worst-case, and we really don't want to 4701ab7f2deSmaybee * throttle based off of worst-case estimates. 4711ab7f2deSmaybee */ 4721ab7f2deSmaybee if (write_limit > 0) { 4731ab7f2deSmaybee reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK] 4741ab7f2deSmaybee + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2; 4751ab7f2deSmaybee 4761ab7f2deSmaybee if (reserved && reserved > write_limit) 4771ab7f2deSmaybee return (ERESTART); 4781ab7f2deSmaybee } 4791ab7f2deSmaybee 4801ab7f2deSmaybee atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space); 4811ab7f2deSmaybee 4821ab7f2deSmaybee /* 4831ab7f2deSmaybee * If this transaction group is over 7/8ths capacity, delay 4841ab7f2deSmaybee * the caller 1 clock tick. This will slow down the "fill" 4851ab7f2deSmaybee * rate until the sync process can catch up with us. 4861ab7f2deSmaybee */ 487e8397a2bSgw if (reserved && reserved > (write_limit - (write_limit >> 3))) 4881ab7f2deSmaybee txg_delay(dp, tx->tx_txg, 1); 4891ab7f2deSmaybee 4901ab7f2deSmaybee return (0); 4911ab7f2deSmaybee } 4921ab7f2deSmaybee 4931ab7f2deSmaybee void 4941ab7f2deSmaybee dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 4951ab7f2deSmaybee { 4961ab7f2deSmaybee ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space); 4971ab7f2deSmaybee atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space); 4981ab7f2deSmaybee } 4991ab7f2deSmaybee 5001ab7f2deSmaybee void 5011ab7f2deSmaybee dsl_pool_memory_pressure(dsl_pool_t *dp) 5021ab7f2deSmaybee { 5031ab7f2deSmaybee uint64_t space_inuse = 0; 5041ab7f2deSmaybee int i; 5051ab7f2deSmaybee 5061ab7f2deSmaybee if (dp->dp_write_limit == zfs_write_limit_min) 5071ab7f2deSmaybee return; 5081ab7f2deSmaybee 5091ab7f2deSmaybee for (i = 0; i < TXG_SIZE; i++) { 5101ab7f2deSmaybee space_inuse += dp->dp_space_towrite[i]; 5111ab7f2deSmaybee space_inuse += dp->dp_tempreserved[i]; 5121ab7f2deSmaybee } 5131ab7f2deSmaybee dp->dp_write_limit = MAX(zfs_write_limit_min, 5141ab7f2deSmaybee MIN(dp->dp_write_limit, space_inuse / 4)); 5151ab7f2deSmaybee } 5161ab7f2deSmaybee 5171ab7f2deSmaybee void 5181ab7f2deSmaybee dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 5191ab7f2deSmaybee { 5201ab7f2deSmaybee if (space > 0) { 5211ab7f2deSmaybee mutex_enter(&dp->dp_lock); 5221ab7f2deSmaybee dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space; 5231ab7f2deSmaybee mutex_exit(&dp->dp_lock); 5241ab7f2deSmaybee } 5251ab7f2deSmaybee } 526088f3894Sahrens 527088f3894Sahrens /* ARGSUSED */ 528088f3894Sahrens static int 529088f3894Sahrens upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) 530088f3894Sahrens { 531088f3894Sahrens dmu_tx_t *tx = arg; 532088f3894Sahrens dsl_dataset_t *ds, *prev = NULL; 533088f3894Sahrens int err; 534088f3894Sahrens dsl_pool_t *dp = spa_get_dsl(spa); 535088f3894Sahrens 536088f3894Sahrens err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); 537088f3894Sahrens if (err) 538088f3894Sahrens return (err); 539088f3894Sahrens 540088f3894Sahrens while (ds->ds_phys->ds_prev_snap_obj != 0) { 541088f3894Sahrens err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, 542088f3894Sahrens FTAG, &prev); 543088f3894Sahrens if (err) { 544088f3894Sahrens dsl_dataset_rele(ds, FTAG); 545088f3894Sahrens return (err); 546088f3894Sahrens } 547088f3894Sahrens 548088f3894Sahrens if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) 549088f3894Sahrens break; 550088f3894Sahrens dsl_dataset_rele(ds, FTAG); 551088f3894Sahrens ds = prev; 552088f3894Sahrens prev = NULL; 553088f3894Sahrens } 554088f3894Sahrens 555088f3894Sahrens if (prev == NULL) { 556088f3894Sahrens prev = dp->dp_origin_snap; 557088f3894Sahrens 558088f3894Sahrens /* 559088f3894Sahrens * The $ORIGIN can't have any data, or the accounting 560088f3894Sahrens * will be wrong. 561088f3894Sahrens */ 562088f3894Sahrens ASSERT(prev->ds_phys->ds_bp.blk_birth == 0); 563088f3894Sahrens 564088f3894Sahrens /* The origin doesn't get attached to itself */ 565088f3894Sahrens if (ds->ds_object == prev->ds_object) { 566088f3894Sahrens dsl_dataset_rele(ds, FTAG); 567088f3894Sahrens return (0); 568088f3894Sahrens } 569088f3894Sahrens 570088f3894Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 571088f3894Sahrens ds->ds_phys->ds_prev_snap_obj = prev->ds_object; 572088f3894Sahrens ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg; 573088f3894Sahrens 574088f3894Sahrens dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 575088f3894Sahrens ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object; 576088f3894Sahrens 577088f3894Sahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 578088f3894Sahrens prev->ds_phys->ds_num_children++; 579088f3894Sahrens 580088f3894Sahrens if (ds->ds_phys->ds_next_snap_obj == 0) { 581088f3894Sahrens ASSERT(ds->ds_prev == NULL); 582088f3894Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, 583088f3894Sahrens ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 584088f3894Sahrens } 585088f3894Sahrens } 586088f3894Sahrens 587088f3894Sahrens ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object); 588088f3894Sahrens ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object); 589088f3894Sahrens 590088f3894Sahrens if (prev->ds_phys->ds_next_clones_obj == 0) { 591c33e334fSMatthew Ahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 592088f3894Sahrens prev->ds_phys->ds_next_clones_obj = 593088f3894Sahrens zap_create(dp->dp_meta_objset, 594088f3894Sahrens DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 595088f3894Sahrens } 596088f3894Sahrens VERIFY(0 == zap_add_int(dp->dp_meta_objset, 597088f3894Sahrens prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx)); 598088f3894Sahrens 599088f3894Sahrens dsl_dataset_rele(ds, FTAG); 600088f3894Sahrens if (prev != dp->dp_origin_snap) 601088f3894Sahrens dsl_dataset_rele(prev, FTAG); 602088f3894Sahrens return (0); 603088f3894Sahrens } 604088f3894Sahrens 605088f3894Sahrens void 606088f3894Sahrens dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) 607088f3894Sahrens { 608088f3894Sahrens ASSERT(dmu_tx_is_syncing(tx)); 609088f3894Sahrens ASSERT(dp->dp_origin_snap != NULL); 610088f3894Sahrens 611c33e334fSMatthew Ahrens VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb, 612c33e334fSMatthew Ahrens tx, DS_FIND_CHILDREN)); 613088f3894Sahrens } 614088f3894Sahrens 615088f3894Sahrens void 616088f3894Sahrens dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) 617088f3894Sahrens { 618088f3894Sahrens uint64_t dsobj; 619088f3894Sahrens dsl_dataset_t *ds; 620088f3894Sahrens 621088f3894Sahrens ASSERT(dmu_tx_is_syncing(tx)); 622088f3894Sahrens ASSERT(dp->dp_origin_snap == NULL); 623088f3894Sahrens 624088f3894Sahrens /* create the origin dir, ds, & snap-ds */ 625088f3894Sahrens rw_enter(&dp->dp_config_rwlock, RW_WRITER); 626088f3894Sahrens dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, 627088f3894Sahrens NULL, 0, kcred, tx); 628088f3894Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 629*3f9d6ad7SLin Ling dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx); 630088f3894Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, 631088f3894Sahrens dp, &dp->dp_origin_snap)); 632088f3894Sahrens dsl_dataset_rele(ds, FTAG); 633088f3894Sahrens rw_exit(&dp->dp_config_rwlock); 634088f3894Sahrens } 6359d3574bfSNeil Perrin 6369d3574bfSNeil Perrin taskq_t * 6379d3574bfSNeil Perrin dsl_pool_vnrele_taskq(dsl_pool_t *dp) 6389d3574bfSNeil Perrin { 6399d3574bfSNeil Perrin return (dp->dp_vnrele_taskq); 6409d3574bfSNeil Perrin } 641ca45db41SChris Kirby 642ca45db41SChris Kirby /* 643ca45db41SChris Kirby * Walk through the pool-wide zap object of temporary snapshot user holds 644ca45db41SChris Kirby * and release them. 645ca45db41SChris Kirby */ 646ca45db41SChris Kirby void 647ca45db41SChris Kirby dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) 648ca45db41SChris Kirby { 649ca45db41SChris Kirby zap_attribute_t za; 650ca45db41SChris Kirby zap_cursor_t zc; 651ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 652ca45db41SChris Kirby uint64_t zapobj = dp->dp_tmp_userrefs_obj; 653ca45db41SChris Kirby 654ca45db41SChris Kirby if (zapobj == 0) 655ca45db41SChris Kirby return; 656ca45db41SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 657ca45db41SChris Kirby 658ca45db41SChris Kirby for (zap_cursor_init(&zc, mos, zapobj); 659ca45db41SChris Kirby zap_cursor_retrieve(&zc, &za) == 0; 660ca45db41SChris Kirby zap_cursor_advance(&zc)) { 661ca45db41SChris Kirby char *htag; 662ca45db41SChris Kirby uint64_t dsobj; 663ca45db41SChris Kirby 664ca45db41SChris Kirby htag = strchr(za.za_name, '-'); 665ca45db41SChris Kirby *htag = '\0'; 666ca45db41SChris Kirby ++htag; 667ca45db41SChris Kirby dsobj = strtonum(za.za_name, NULL); 668ca45db41SChris Kirby (void) dsl_dataset_user_release_tmp(dp, dsobj, htag); 669ca45db41SChris Kirby } 670ca45db41SChris Kirby zap_cursor_fini(&zc); 671ca45db41SChris Kirby } 672ca45db41SChris Kirby 673ca45db41SChris Kirby /* 674ca45db41SChris Kirby * Create the pool-wide zap object for storing temporary snapshot holds. 675ca45db41SChris Kirby */ 676ca45db41SChris Kirby void 677ca45db41SChris Kirby dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) 678ca45db41SChris Kirby { 679ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 680ca45db41SChris Kirby 681ca45db41SChris Kirby ASSERT(dp->dp_tmp_userrefs_obj == 0); 682ca45db41SChris Kirby ASSERT(dmu_tx_is_syncing(tx)); 683ca45db41SChris Kirby 684ca45db41SChris Kirby dp->dp_tmp_userrefs_obj = zap_create(mos, DMU_OT_USERREFS, 685ca45db41SChris Kirby DMU_OT_NONE, 0, tx); 686ca45db41SChris Kirby 687ca45db41SChris Kirby VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, 688ca45db41SChris Kirby sizeof (uint64_t), 1, &dp->dp_tmp_userrefs_obj, tx) == 0); 689ca45db41SChris Kirby } 690ca45db41SChris Kirby 691ca45db41SChris Kirby static int 692ca45db41SChris Kirby dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, 69315508ac0SChris Kirby const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding) 694ca45db41SChris Kirby { 695ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 696ca45db41SChris Kirby uint64_t zapobj = dp->dp_tmp_userrefs_obj; 697ca45db41SChris Kirby char *name; 698ca45db41SChris Kirby int error; 699ca45db41SChris Kirby 700ca45db41SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 701ca45db41SChris Kirby ASSERT(dmu_tx_is_syncing(tx)); 702ca45db41SChris Kirby 703ca45db41SChris Kirby /* 704ca45db41SChris Kirby * If the pool was created prior to SPA_VERSION_USERREFS, the 705ca45db41SChris Kirby * zap object for temporary holds might not exist yet. 706ca45db41SChris Kirby */ 707ca45db41SChris Kirby if (zapobj == 0) { 708ca45db41SChris Kirby if (holding) { 709ca45db41SChris Kirby dsl_pool_user_hold_create_obj(dp, tx); 710ca45db41SChris Kirby zapobj = dp->dp_tmp_userrefs_obj; 711ca45db41SChris Kirby } else { 712ca45db41SChris Kirby return (ENOENT); 713ca45db41SChris Kirby } 714ca45db41SChris Kirby } 715ca45db41SChris Kirby 716ca45db41SChris Kirby name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); 717ca45db41SChris Kirby if (holding) 71815508ac0SChris Kirby error = zap_add(mos, zapobj, name, 8, 1, now, tx); 719ca45db41SChris Kirby else 720ca45db41SChris Kirby error = zap_remove(mos, zapobj, name, tx); 721ca45db41SChris Kirby strfree(name); 722ca45db41SChris Kirby 723ca45db41SChris Kirby return (error); 724ca45db41SChris Kirby } 725ca45db41SChris Kirby 726ca45db41SChris Kirby /* 727ca45db41SChris Kirby * Add a temporary hold for the given dataset object and tag. 728ca45db41SChris Kirby */ 729ca45db41SChris Kirby int 730ca45db41SChris Kirby dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 73115508ac0SChris Kirby uint64_t *now, dmu_tx_t *tx) 732ca45db41SChris Kirby { 73315508ac0SChris Kirby return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); 734ca45db41SChris Kirby } 735ca45db41SChris Kirby 736ca45db41SChris Kirby /* 737ca45db41SChris Kirby * Release a temporary hold for the given dataset object and tag. 738ca45db41SChris Kirby */ 739ca45db41SChris Kirby int 740ca45db41SChris Kirby dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 741ca45db41SChris Kirby dmu_tx_t *tx) 742ca45db41SChris Kirby { 743ca45db41SChris Kirby return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL, 744ca45db41SChris Kirby tx, B_FALSE)); 745ca45db41SChris Kirby } 746