1789Sahrens /* 2789Sahrens * CDDL HEADER START 3789Sahrens * 4789Sahrens * The contents of this file are subject to the terms of the 51544Seschrock * Common Development and Distribution License (the "License"). 61544Seschrock * You may not use this file except in compliance with the License. 7789Sahrens * 8789Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9789Sahrens * or http://www.opensolaris.org/os/licensing. 10789Sahrens * See the License for the specific language governing permissions 11789Sahrens * and limitations under the License. 12789Sahrens * 13789Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14789Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15789Sahrens * If applicable, add the following below this CDDL HEADER, with the 16789Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17789Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18789Sahrens * 19789Sahrens * CDDL HEADER END 20789Sahrens */ 21789Sahrens /* 229008SLin.Ling@Sun.COM * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23789Sahrens * Use is subject to license terms. 24789Sahrens */ 25789Sahrens 26789Sahrens #include <sys/dsl_pool.h> 27789Sahrens #include <sys/dsl_dataset.h> 28789Sahrens #include <sys/dsl_dir.h> 292199Sahrens #include <sys/dsl_synctask.h> 30789Sahrens #include <sys/dmu_tx.h> 31789Sahrens #include <sys/dmu_objset.h> 32789Sahrens #include <sys/arc.h> 33789Sahrens #include <sys/zap.h> 343547Smaybee #include <sys/zio.h> 35789Sahrens #include <sys/zfs_context.h> 36789Sahrens #include <sys/fs/zfs.h> 377046Sahrens #include <sys/zfs_znode.h> 387046Sahrens #include <sys/spa_impl.h> 39789Sahrens 406245Smaybee int zfs_no_write_throttle = 0; 417468SMark.Maybee@Sun.COM int zfs_write_limit_shift = 3; /* 1/8th of physical memory */ 427468SMark.Maybee@Sun.COM int zfs_txg_synctime = 5; /* target secs to sync a txg */ 437468SMark.Maybee@Sun.COM 447468SMark.Maybee@Sun.COM uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */ 457468SMark.Maybee@Sun.COM uint64_t zfs_write_limit_max = 0; /* max data payload per txg */ 467468SMark.Maybee@Sun.COM uint64_t zfs_write_limit_inflated = 0; 476245Smaybee uint64_t zfs_write_limit_override = 0; 486245Smaybee 497468SMark.Maybee@Sun.COM kmutex_t zfs_write_limit_lock; 507468SMark.Maybee@Sun.COM 517468SMark.Maybee@Sun.COM static pgcnt_t old_physmem = 0; 527046Sahrens 531544Seschrock static int 547046Sahrens dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) 55789Sahrens { 56789Sahrens uint64_t obj; 57789Sahrens int err; 58789Sahrens 59789Sahrens err = zap_lookup(dp->dp_meta_objset, 60789Sahrens dp->dp_root_dir->dd_phys->dd_child_dir_zapobj, 617046Sahrens name, sizeof (obj), 1, &obj); 621544Seschrock if (err) 631544Seschrock return (err); 64789Sahrens 657046Sahrens return (dsl_dir_open_obj(dp, obj, name, dp, ddp)); 66789Sahrens } 67789Sahrens 68789Sahrens static dsl_pool_t * 69789Sahrens dsl_pool_open_impl(spa_t *spa, uint64_t txg) 70789Sahrens { 71789Sahrens dsl_pool_t *dp; 72789Sahrens blkptr_t *bp = spa_get_rootblkptr(spa); 73789Sahrens 74789Sahrens dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); 75789Sahrens dp->dp_spa = spa; 76789Sahrens dp->dp_meta_rootbp = *bp; 772856Snd150628 rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL); 786245Smaybee dp->dp_write_limit = zfs_write_limit_min; 79789Sahrens txg_init(dp, txg); 80789Sahrens 81789Sahrens txg_list_create(&dp->dp_dirty_datasets, 82789Sahrens offsetof(dsl_dataset_t, ds_dirty_link)); 83789Sahrens txg_list_create(&dp->dp_dirty_dirs, 84789Sahrens offsetof(dsl_dir_t, dd_dirty_link)); 852199Sahrens txg_list_create(&dp->dp_sync_tasks, 862199Sahrens offsetof(dsl_sync_task_group_t, dstg_node)); 875367Sahrens list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t), 88789Sahrens offsetof(dsl_dataset_t, ds_synced_link)); 89789Sahrens 906245Smaybee mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); 917046Sahrens mutex_init(&dp->dp_scrub_cancel_lock, NULL, MUTEX_DEFAULT, NULL); 926245Smaybee 939321SNeil.Perrin@Sun.COM dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri, 949321SNeil.Perrin@Sun.COM 1, 4, 0); 959321SNeil.Perrin@Sun.COM 96789Sahrens return (dp); 97789Sahrens } 98789Sahrens 991544Seschrock int 1001544Seschrock dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) 101789Sahrens { 102789Sahrens int err; 103789Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 1047046Sahrens dsl_dir_t *dd; 1057046Sahrens dsl_dataset_t *ds; 106789Sahrens 1077046Sahrens rw_enter(&dp->dp_config_rwlock, RW_WRITER); 10810298SMatthew.Ahrens@Sun.COM err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, 10910298SMatthew.Ahrens@Sun.COM &dp->dp_meta_objset); 1101544Seschrock if (err) 1111544Seschrock goto out; 1121544Seschrock 113789Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 114789Sahrens DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, 115789Sahrens &dp->dp_root_dir_obj); 1161544Seschrock if (err) 1171544Seschrock goto out; 1181544Seschrock 1191544Seschrock err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, 1201544Seschrock NULL, dp, &dp->dp_root_dir); 1211544Seschrock if (err) 1221544Seschrock goto out; 123789Sahrens 1247046Sahrens err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); 1251544Seschrock if (err) 1261544Seschrock goto out; 1271544Seschrock 1287046Sahrens if (spa_version(spa) >= SPA_VERSION_ORIGIN) { 1297046Sahrens err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); 1307046Sahrens if (err) 1317046Sahrens goto out; 1327046Sahrens err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj, 1337046Sahrens FTAG, &ds); 1349008SLin.Ling@Sun.COM if (err == 0) { 1359008SLin.Ling@Sun.COM err = dsl_dataset_hold_obj(dp, 1369008SLin.Ling@Sun.COM ds->ds_phys->ds_prev_snap_obj, dp, 1379008SLin.Ling@Sun.COM &dp->dp_origin_snap); 1389008SLin.Ling@Sun.COM dsl_dataset_rele(ds, FTAG); 1399008SLin.Ling@Sun.COM } 1409008SLin.Ling@Sun.COM dsl_dir_close(dd, dp); 1417046Sahrens if (err) 1427046Sahrens goto out; 1437046Sahrens } 1447046Sahrens 14510342Schris.kirby@sun.com err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 14610342Schris.kirby@sun.com DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, 14710342Schris.kirby@sun.com &dp->dp_tmp_userrefs_obj); 14810342Schris.kirby@sun.com if (err == ENOENT) 14910342Schris.kirby@sun.com err = 0; 15010342Schris.kirby@sun.com if (err) 15110342Schris.kirby@sun.com goto out; 15210342Schris.kirby@sun.com 1537046Sahrens /* get scrub status */ 1547046Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1557046Sahrens DMU_POOL_SCRUB_FUNC, sizeof (uint32_t), 1, 1567046Sahrens &dp->dp_scrub_func); 1577046Sahrens if (err == 0) { 1587046Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1597046Sahrens DMU_POOL_SCRUB_QUEUE, sizeof (uint64_t), 1, 1607046Sahrens &dp->dp_scrub_queue_obj); 1617046Sahrens if (err) 1627046Sahrens goto out; 1637046Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1647046Sahrens DMU_POOL_SCRUB_MIN_TXG, sizeof (uint64_t), 1, 1657046Sahrens &dp->dp_scrub_min_txg); 1667046Sahrens if (err) 1677046Sahrens goto out; 1687046Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1697046Sahrens DMU_POOL_SCRUB_MAX_TXG, sizeof (uint64_t), 1, 1707046Sahrens &dp->dp_scrub_max_txg); 1717046Sahrens if (err) 1727046Sahrens goto out; 1737046Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1747046Sahrens DMU_POOL_SCRUB_BOOKMARK, sizeof (uint64_t), 4, 1757046Sahrens &dp->dp_scrub_bookmark); 1767046Sahrens if (err) 1777046Sahrens goto out; 1787046Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1797046Sahrens DMU_POOL_SCRUB_ERRORS, sizeof (uint64_t), 1, 1807046Sahrens &spa->spa_scrub_errors); 1817046Sahrens if (err) 1827046Sahrens goto out; 1837046Sahrens if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) { 1847046Sahrens /* 1857046Sahrens * A new-type scrub was in progress on an old 1867046Sahrens * pool. Restart from the beginning, since the 1877046Sahrens * old software may have changed the pool in the 1887046Sahrens * meantime. 1897046Sahrens */ 1907046Sahrens dsl_pool_scrub_restart(dp); 1917046Sahrens } 1927046Sahrens } else { 1937046Sahrens /* 1947046Sahrens * It's OK if there is no scrub in progress (and if 1957046Sahrens * there was an I/O error, ignore it). 1967046Sahrens */ 1977046Sahrens err = 0; 1987046Sahrens } 1997046Sahrens 2001544Seschrock out: 201789Sahrens rw_exit(&dp->dp_config_rwlock); 2021544Seschrock if (err) 2031544Seschrock dsl_pool_close(dp); 2041544Seschrock else 2051544Seschrock *dpp = dp; 206789Sahrens 2071544Seschrock return (err); 208789Sahrens } 209789Sahrens 210789Sahrens void 211789Sahrens dsl_pool_close(dsl_pool_t *dp) 212789Sahrens { 2137046Sahrens /* drop our references from dsl_pool_open() */ 2147046Sahrens 2157046Sahrens /* 2167046Sahrens * Since we held the origin_snap from "syncing" context (which 2177046Sahrens * includes pool-opening context), it actually only got a "ref" 2187046Sahrens * and not a hold, so just drop that here. 2197046Sahrens */ 2207046Sahrens if (dp->dp_origin_snap) 2217046Sahrens dsl_dataset_drop_ref(dp->dp_origin_snap, dp); 2221544Seschrock if (dp->dp_mos_dir) 2231544Seschrock dsl_dir_close(dp->dp_mos_dir, dp); 2241544Seschrock if (dp->dp_root_dir) 2251544Seschrock dsl_dir_close(dp->dp_root_dir, dp); 226789Sahrens 227789Sahrens /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ 2281544Seschrock if (dp->dp_meta_objset) 22910298SMatthew.Ahrens@Sun.COM dmu_objset_evict(dp->dp_meta_objset); 230789Sahrens 231789Sahrens txg_list_destroy(&dp->dp_dirty_datasets); 232789Sahrens txg_list_destroy(&dp->dp_dirty_dirs); 2335367Sahrens list_destroy(&dp->dp_synced_datasets); 234789Sahrens 2355642Smaybee arc_flush(dp->dp_spa); 236789Sahrens txg_fini(dp); 2372856Snd150628 rw_destroy(&dp->dp_config_rwlock); 2386245Smaybee mutex_destroy(&dp->dp_lock); 2397046Sahrens mutex_destroy(&dp->dp_scrub_cancel_lock); 2409321SNeil.Perrin@Sun.COM taskq_destroy(dp->dp_vnrele_taskq); 2417837SMatthew.Ahrens@Sun.COM if (dp->dp_blkstats) 2427837SMatthew.Ahrens@Sun.COM kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); 243789Sahrens kmem_free(dp, sizeof (dsl_pool_t)); 244789Sahrens } 245789Sahrens 246789Sahrens dsl_pool_t * 2477184Stimh dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) 248789Sahrens { 249789Sahrens int err; 250789Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 251789Sahrens dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); 25210298SMatthew.Ahrens@Sun.COM objset_t *os; 2537046Sahrens dsl_dataset_t *ds; 2547046Sahrens uint64_t dsobj; 2557046Sahrens 2567046Sahrens /* create and open the MOS (meta-objset) */ 25710298SMatthew.Ahrens@Sun.COM dp->dp_meta_objset = dmu_objset_create_impl(spa, 25810298SMatthew.Ahrens@Sun.COM NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); 259789Sahrens 260789Sahrens /* create the pool directory */ 261789Sahrens err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 262789Sahrens DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); 263789Sahrens ASSERT3U(err, ==, 0); 264789Sahrens 265789Sahrens /* create and open the root dir */ 2667046Sahrens dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); 2671544Seschrock VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj, 2681544Seschrock NULL, dp, &dp->dp_root_dir)); 269789Sahrens 270789Sahrens /* create and open the meta-objset dir */ 2717046Sahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); 2727046Sahrens VERIFY(0 == dsl_pool_open_special_dir(dp, 2737046Sahrens MOS_DIR_NAME, &dp->dp_mos_dir)); 2747046Sahrens 2757046Sahrens if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) 2767046Sahrens dsl_pool_create_origin(dp, tx); 2777046Sahrens 2787046Sahrens /* create the root dataset */ 2797046Sahrens dsobj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); 2807046Sahrens 2817046Sahrens /* create the root objset */ 2827046Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 28310298SMatthew.Ahrens@Sun.COM os = dmu_objset_create_impl(dp->dp_spa, ds, 2847046Sahrens dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); 2857046Sahrens #ifdef _KERNEL 28610298SMatthew.Ahrens@Sun.COM zfs_create_fs(os, kcred, zplprops, tx); 2877046Sahrens #endif 2887046Sahrens dsl_dataset_rele(ds, FTAG); 289789Sahrens 290789Sahrens dmu_tx_commit(tx); 291789Sahrens 292789Sahrens return (dp); 293789Sahrens } 294789Sahrens 295789Sahrens void 296789Sahrens dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) 297789Sahrens { 2983547Smaybee zio_t *zio; 299789Sahrens dmu_tx_t *tx; 3003547Smaybee dsl_dir_t *dd; 3013547Smaybee dsl_dataset_t *ds; 3023547Smaybee dsl_sync_task_group_t *dstg; 30310298SMatthew.Ahrens@Sun.COM objset_t *mos = dp->dp_meta_objset; 3047468SMark.Maybee@Sun.COM hrtime_t start, write_time; 3057468SMark.Maybee@Sun.COM uint64_t data_written; 3063547Smaybee int err; 307789Sahrens 308789Sahrens tx = dmu_tx_create_assigned(dp, txg); 309789Sahrens 3107468SMark.Maybee@Sun.COM dp->dp_read_overhead = 0; 3119366SMark.Maybee@Sun.COM start = gethrtime(); 3129396SMatthew.Ahrens@Sun.COM 3133547Smaybee zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 3143547Smaybee while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) { 3159396SMatthew.Ahrens@Sun.COM /* 3169396SMatthew.Ahrens@Sun.COM * We must not sync any non-MOS datasets twice, because 3179396SMatthew.Ahrens@Sun.COM * we may have taken a snapshot of them. However, we 3189396SMatthew.Ahrens@Sun.COM * may sync newly-created datasets on pass 2. 3199396SMatthew.Ahrens@Sun.COM */ 3209396SMatthew.Ahrens@Sun.COM ASSERT(!list_link_active(&ds->ds_synced_link)); 3219396SMatthew.Ahrens@Sun.COM list_insert_tail(&dp->dp_synced_datasets, ds); 3223547Smaybee dsl_dataset_sync(ds, zio, tx); 3233547Smaybee } 3247468SMark.Maybee@Sun.COM DTRACE_PROBE(pool_sync__1setup); 3259396SMatthew.Ahrens@Sun.COM err = zio_wait(zio); 3267468SMark.Maybee@Sun.COM 3277468SMark.Maybee@Sun.COM write_time = gethrtime() - start; 3283547Smaybee ASSERT(err == 0); 3297468SMark.Maybee@Sun.COM DTRACE_PROBE(pool_sync__2rootzio); 330789Sahrens 3319396SMatthew.Ahrens@Sun.COM for (ds = list_head(&dp->dp_synced_datasets); ds; 3329396SMatthew.Ahrens@Sun.COM ds = list_next(&dp->dp_synced_datasets, ds)) 33310298SMatthew.Ahrens@Sun.COM dmu_objset_do_userquota_callbacks(ds->ds_objset, tx); 3349396SMatthew.Ahrens@Sun.COM 3359396SMatthew.Ahrens@Sun.COM /* 3369396SMatthew.Ahrens@Sun.COM * Sync the datasets again to push out the changes due to 3379396SMatthew.Ahrens@Sun.COM * userquota updates. This must be done before we process the 3389396SMatthew.Ahrens@Sun.COM * sync tasks, because that could cause a snapshot of a dataset 3399396SMatthew.Ahrens@Sun.COM * whose ds_bp will be rewritten when we do this 2nd sync. 3409396SMatthew.Ahrens@Sun.COM */ 3419396SMatthew.Ahrens@Sun.COM zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 3429396SMatthew.Ahrens@Sun.COM while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) { 3439396SMatthew.Ahrens@Sun.COM ASSERT(list_link_active(&ds->ds_synced_link)); 3449396SMatthew.Ahrens@Sun.COM dmu_buf_rele(ds->ds_dbuf, ds); 3459396SMatthew.Ahrens@Sun.COM dsl_dataset_sync(ds, zio, tx); 3469396SMatthew.Ahrens@Sun.COM } 3479396SMatthew.Ahrens@Sun.COM err = zio_wait(zio); 3489396SMatthew.Ahrens@Sun.COM 34910922SJeff.Bonwick@Sun.COM /* 35010922SJeff.Bonwick@Sun.COM * If anything was added to a deadlist during a zio done callback, 35110922SJeff.Bonwick@Sun.COM * it had to be put on the deferred queue. Enqueue it for real now. 35210922SJeff.Bonwick@Sun.COM */ 35310922SJeff.Bonwick@Sun.COM for (ds = list_head(&dp->dp_synced_datasets); ds; 35410922SJeff.Bonwick@Sun.COM ds = list_next(&dp->dp_synced_datasets, ds)) 35510922SJeff.Bonwick@Sun.COM bplist_sync(&ds->ds_deadlist, 35610922SJeff.Bonwick@Sun.COM bplist_enqueue_cb, &ds->ds_deadlist, tx); 35710922SJeff.Bonwick@Sun.COM 3589396SMatthew.Ahrens@Sun.COM while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg)) { 3599396SMatthew.Ahrens@Sun.COM /* 3609396SMatthew.Ahrens@Sun.COM * No more sync tasks should have been added while we 3619396SMatthew.Ahrens@Sun.COM * were syncing. 3629396SMatthew.Ahrens@Sun.COM */ 3639396SMatthew.Ahrens@Sun.COM ASSERT(spa_sync_pass(dp->dp_spa) == 1); 3643547Smaybee dsl_sync_task_group_sync(dstg, tx); 3659396SMatthew.Ahrens@Sun.COM } 3667468SMark.Maybee@Sun.COM DTRACE_PROBE(pool_sync__3task); 3677468SMark.Maybee@Sun.COM 3687468SMark.Maybee@Sun.COM start = gethrtime(); 3693547Smaybee while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) 3703547Smaybee dsl_dir_sync(dd, tx); 3717468SMark.Maybee@Sun.COM write_time += gethrtime() - start; 372789Sahrens 3737046Sahrens if (spa_sync_pass(dp->dp_spa) == 1) 3747046Sahrens dsl_pool_scrub_sync(dp, tx); 3757046Sahrens 3767468SMark.Maybee@Sun.COM start = gethrtime(); 37710298SMatthew.Ahrens@Sun.COM if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || 37810298SMatthew.Ahrens@Sun.COM list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { 3793547Smaybee zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 38010298SMatthew.Ahrens@Sun.COM dmu_objset_sync(mos, zio, tx); 3813547Smaybee err = zio_wait(zio); 3823547Smaybee ASSERT(err == 0); 383789Sahrens dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); 384789Sahrens spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); 385789Sahrens } 3867468SMark.Maybee@Sun.COM write_time += gethrtime() - start; 3877468SMark.Maybee@Sun.COM DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time, 3887468SMark.Maybee@Sun.COM hrtime_t, dp->dp_read_overhead); 3897468SMark.Maybee@Sun.COM write_time -= dp->dp_read_overhead; 390789Sahrens 391789Sahrens dmu_tx_commit(tx); 3927468SMark.Maybee@Sun.COM 3937468SMark.Maybee@Sun.COM data_written = dp->dp_space_towrite[txg & TXG_MASK]; 3947468SMark.Maybee@Sun.COM dp->dp_space_towrite[txg & TXG_MASK] = 0; 3957468SMark.Maybee@Sun.COM ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0); 3967468SMark.Maybee@Sun.COM 3977468SMark.Maybee@Sun.COM /* 3987468SMark.Maybee@Sun.COM * If the write limit max has not been explicitly set, set it 3997468SMark.Maybee@Sun.COM * to a fraction of available physical memory (default 1/8th). 4007468SMark.Maybee@Sun.COM * Note that we must inflate the limit because the spa 4017468SMark.Maybee@Sun.COM * inflates write sizes to account for data replication. 4027468SMark.Maybee@Sun.COM * Check this each sync phase to catch changing memory size. 4037468SMark.Maybee@Sun.COM */ 4047468SMark.Maybee@Sun.COM if (physmem != old_physmem && zfs_write_limit_shift) { 4057468SMark.Maybee@Sun.COM mutex_enter(&zfs_write_limit_lock); 4067468SMark.Maybee@Sun.COM old_physmem = physmem; 4077468SMark.Maybee@Sun.COM zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift; 4087468SMark.Maybee@Sun.COM zfs_write_limit_inflated = MAX(zfs_write_limit_min, 4097468SMark.Maybee@Sun.COM spa_get_asize(dp->dp_spa, zfs_write_limit_max)); 4107468SMark.Maybee@Sun.COM mutex_exit(&zfs_write_limit_lock); 4117468SMark.Maybee@Sun.COM } 4127468SMark.Maybee@Sun.COM 4137468SMark.Maybee@Sun.COM /* 4147468SMark.Maybee@Sun.COM * Attempt to keep the sync time consistent by adjusting the 4157468SMark.Maybee@Sun.COM * amount of write traffic allowed into each transaction group. 4167468SMark.Maybee@Sun.COM * Weight the throughput calculation towards the current value: 4177468SMark.Maybee@Sun.COM * thru = 3/4 old_thru + 1/4 new_thru 4187468SMark.Maybee@Sun.COM */ 4197468SMark.Maybee@Sun.COM ASSERT(zfs_write_limit_min > 0); 4207468SMark.Maybee@Sun.COM if (data_written > zfs_write_limit_min / 8 && write_time > 0) { 4217468SMark.Maybee@Sun.COM uint64_t throughput = (data_written * NANOSEC) / write_time; 4227468SMark.Maybee@Sun.COM if (dp->dp_throughput) 4237468SMark.Maybee@Sun.COM dp->dp_throughput = throughput / 4 + 4247468SMark.Maybee@Sun.COM 3 * dp->dp_throughput / 4; 4257468SMark.Maybee@Sun.COM else 4267468SMark.Maybee@Sun.COM dp->dp_throughput = throughput; 4277468SMark.Maybee@Sun.COM dp->dp_write_limit = MIN(zfs_write_limit_inflated, 4287468SMark.Maybee@Sun.COM MAX(zfs_write_limit_min, 4297468SMark.Maybee@Sun.COM dp->dp_throughput * zfs_txg_synctime)); 4307468SMark.Maybee@Sun.COM } 431789Sahrens } 432789Sahrens 433789Sahrens void 43410922SJeff.Bonwick@Sun.COM dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) 435789Sahrens { 436789Sahrens dsl_dataset_t *ds; 43710922SJeff.Bonwick@Sun.COM objset_t *os; 438789Sahrens 4395367Sahrens while (ds = list_head(&dp->dp_synced_datasets)) { 4405367Sahrens list_remove(&dp->dp_synced_datasets, ds); 44110922SJeff.Bonwick@Sun.COM os = ds->ds_objset; 44210922SJeff.Bonwick@Sun.COM zil_clean(os->os_zil); 44310922SJeff.Bonwick@Sun.COM ASSERT(!dmu_objset_is_dirty(os, txg)); 4443897Smaybee dmu_buf_rele(ds->ds_dbuf, ds); 445789Sahrens } 44610922SJeff.Bonwick@Sun.COM ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); 447789Sahrens } 448789Sahrens 4493547Smaybee /* 4503547Smaybee * TRUE if the current thread is the tx_sync_thread or if we 4513547Smaybee * are being called from SPA context during pool initialization. 4523547Smaybee */ 453789Sahrens int 454789Sahrens dsl_pool_sync_context(dsl_pool_t *dp) 455789Sahrens { 456789Sahrens return (curthread == dp->dp_tx.tx_sync_thread || 4573547Smaybee spa_get_dsl(dp->dp_spa) == NULL); 458789Sahrens } 459789Sahrens 460789Sahrens uint64_t 461789Sahrens dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) 462789Sahrens { 463789Sahrens uint64_t space, resv; 464789Sahrens 465789Sahrens /* 4661775Sbillm * Reserve about 1.6% (1/64), or at least 32MB, for allocation 467789Sahrens * efficiency. 468789Sahrens * XXX The intent log is not accounted for, so it must fit 469789Sahrens * within this slop. 470789Sahrens * 471789Sahrens * If we're trying to assess whether it's OK to do a free, 472789Sahrens * cut the reservation in half to allow forward progress 473789Sahrens * (e.g. make it possible to rm(1) files from a full pool). 474789Sahrens */ 475*10956SGeorge.Wilson@Sun.COM space = spa_get_dspace(dp->dp_spa); 4761775Sbillm resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1); 477789Sahrens if (netfree) 478789Sahrens resv >>= 1; 479789Sahrens 480789Sahrens return (space - resv); 481789Sahrens } 4826245Smaybee 4836245Smaybee int 4846245Smaybee dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx) 4856245Smaybee { 4866245Smaybee uint64_t reserved = 0; 4876245Smaybee uint64_t write_limit = (zfs_write_limit_override ? 4886245Smaybee zfs_write_limit_override : dp->dp_write_limit); 4896245Smaybee 4906245Smaybee if (zfs_no_write_throttle) { 4916643Seschrock atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], 4926643Seschrock space); 4936245Smaybee return (0); 4946245Smaybee } 4956245Smaybee 4966245Smaybee /* 4976245Smaybee * Check to see if we have exceeded the maximum allowed IO for 4986245Smaybee * this transaction group. We can do this without locks since 4996245Smaybee * a little slop here is ok. Note that we do the reserved check 5006245Smaybee * with only half the requested reserve: this is because the 5016245Smaybee * reserve requests are worst-case, and we really don't want to 5026245Smaybee * throttle based off of worst-case estimates. 5036245Smaybee */ 5046245Smaybee if (write_limit > 0) { 5056245Smaybee reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK] 5066245Smaybee + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2; 5076245Smaybee 5086245Smaybee if (reserved && reserved > write_limit) 5096245Smaybee return (ERESTART); 5106245Smaybee } 5116245Smaybee 5126245Smaybee atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space); 5136245Smaybee 5146245Smaybee /* 5156245Smaybee * If this transaction group is over 7/8ths capacity, delay 5166245Smaybee * the caller 1 clock tick. This will slow down the "fill" 5176245Smaybee * rate until the sync process can catch up with us. 5186245Smaybee */ 5196740Sgw25295 if (reserved && reserved > (write_limit - (write_limit >> 3))) 5206245Smaybee txg_delay(dp, tx->tx_txg, 1); 5216245Smaybee 5226245Smaybee return (0); 5236245Smaybee } 5246245Smaybee 5256245Smaybee void 5266245Smaybee dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 5276245Smaybee { 5286245Smaybee ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space); 5296245Smaybee atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space); 5306245Smaybee } 5316245Smaybee 5326245Smaybee void 5336245Smaybee dsl_pool_memory_pressure(dsl_pool_t *dp) 5346245Smaybee { 5356245Smaybee uint64_t space_inuse = 0; 5366245Smaybee int i; 5376245Smaybee 5386245Smaybee if (dp->dp_write_limit == zfs_write_limit_min) 5396245Smaybee return; 5406245Smaybee 5416245Smaybee for (i = 0; i < TXG_SIZE; i++) { 5426245Smaybee space_inuse += dp->dp_space_towrite[i]; 5436245Smaybee space_inuse += dp->dp_tempreserved[i]; 5446245Smaybee } 5456245Smaybee dp->dp_write_limit = MAX(zfs_write_limit_min, 5466245Smaybee MIN(dp->dp_write_limit, space_inuse / 4)); 5476245Smaybee } 5486245Smaybee 5496245Smaybee void 5506245Smaybee dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 5516245Smaybee { 5526245Smaybee if (space > 0) { 5536245Smaybee mutex_enter(&dp->dp_lock); 5546245Smaybee dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space; 5556245Smaybee mutex_exit(&dp->dp_lock); 5566245Smaybee } 5576245Smaybee } 5587046Sahrens 5597046Sahrens /* ARGSUSED */ 5607046Sahrens static int 5617046Sahrens upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) 5627046Sahrens { 5637046Sahrens dmu_tx_t *tx = arg; 5647046Sahrens dsl_dataset_t *ds, *prev = NULL; 5657046Sahrens int err; 5667046Sahrens dsl_pool_t *dp = spa_get_dsl(spa); 5677046Sahrens 5687046Sahrens err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); 5697046Sahrens if (err) 5707046Sahrens return (err); 5717046Sahrens 5727046Sahrens while (ds->ds_phys->ds_prev_snap_obj != 0) { 5737046Sahrens err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, 5747046Sahrens FTAG, &prev); 5757046Sahrens if (err) { 5767046Sahrens dsl_dataset_rele(ds, FTAG); 5777046Sahrens return (err); 5787046Sahrens } 5797046Sahrens 5807046Sahrens if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) 5817046Sahrens break; 5827046Sahrens dsl_dataset_rele(ds, FTAG); 5837046Sahrens ds = prev; 5847046Sahrens prev = NULL; 5857046Sahrens } 5867046Sahrens 5877046Sahrens if (prev == NULL) { 5887046Sahrens prev = dp->dp_origin_snap; 5897046Sahrens 5907046Sahrens /* 5917046Sahrens * The $ORIGIN can't have any data, or the accounting 5927046Sahrens * will be wrong. 5937046Sahrens */ 5947046Sahrens ASSERT(prev->ds_phys->ds_bp.blk_birth == 0); 5957046Sahrens 5967046Sahrens /* The origin doesn't get attached to itself */ 5977046Sahrens if (ds->ds_object == prev->ds_object) { 5987046Sahrens dsl_dataset_rele(ds, FTAG); 5997046Sahrens return (0); 6007046Sahrens } 6017046Sahrens 6027046Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 6037046Sahrens ds->ds_phys->ds_prev_snap_obj = prev->ds_object; 6047046Sahrens ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg; 6057046Sahrens 6067046Sahrens dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 6077046Sahrens ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object; 6087046Sahrens 6097046Sahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 6107046Sahrens prev->ds_phys->ds_num_children++; 6117046Sahrens 6127046Sahrens if (ds->ds_phys->ds_next_snap_obj == 0) { 6137046Sahrens ASSERT(ds->ds_prev == NULL); 6147046Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, 6157046Sahrens ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 6167046Sahrens } 6177046Sahrens } 6187046Sahrens 6197046Sahrens ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object); 6207046Sahrens ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object); 6217046Sahrens 6227046Sahrens if (prev->ds_phys->ds_next_clones_obj == 0) { 62310801SMatthew.Ahrens@Sun.COM dmu_buf_will_dirty(prev->ds_dbuf, tx); 6247046Sahrens prev->ds_phys->ds_next_clones_obj = 6257046Sahrens zap_create(dp->dp_meta_objset, 6267046Sahrens DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 6277046Sahrens } 6287046Sahrens VERIFY(0 == zap_add_int(dp->dp_meta_objset, 6297046Sahrens prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx)); 6307046Sahrens 6317046Sahrens dsl_dataset_rele(ds, FTAG); 6327046Sahrens if (prev != dp->dp_origin_snap) 6337046Sahrens dsl_dataset_rele(prev, FTAG); 6347046Sahrens return (0); 6357046Sahrens } 6367046Sahrens 6377046Sahrens void 6387046Sahrens dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) 6397046Sahrens { 6407046Sahrens ASSERT(dmu_tx_is_syncing(tx)); 6417046Sahrens ASSERT(dp->dp_origin_snap != NULL); 6427046Sahrens 64310801SMatthew.Ahrens@Sun.COM VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb, 64410801SMatthew.Ahrens@Sun.COM tx, DS_FIND_CHILDREN)); 6457046Sahrens } 6467046Sahrens 6477046Sahrens void 6487046Sahrens dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) 6497046Sahrens { 6507046Sahrens uint64_t dsobj; 6517046Sahrens dsl_dataset_t *ds; 6527046Sahrens 6537046Sahrens ASSERT(dmu_tx_is_syncing(tx)); 6547046Sahrens ASSERT(dp->dp_origin_snap == NULL); 6557046Sahrens 6567046Sahrens /* create the origin dir, ds, & snap-ds */ 6577046Sahrens rw_enter(&dp->dp_config_rwlock, RW_WRITER); 6587046Sahrens dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, 6597046Sahrens NULL, 0, kcred, tx); 6607046Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 6617046Sahrens dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, kcred, tx); 6627046Sahrens VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, 6637046Sahrens dp, &dp->dp_origin_snap)); 6647046Sahrens dsl_dataset_rele(ds, FTAG); 6657046Sahrens rw_exit(&dp->dp_config_rwlock); 6667046Sahrens } 6679321SNeil.Perrin@Sun.COM 6689321SNeil.Perrin@Sun.COM taskq_t * 6699321SNeil.Perrin@Sun.COM dsl_pool_vnrele_taskq(dsl_pool_t *dp) 6709321SNeil.Perrin@Sun.COM { 6719321SNeil.Perrin@Sun.COM return (dp->dp_vnrele_taskq); 6729321SNeil.Perrin@Sun.COM } 67310342Schris.kirby@sun.com 67410342Schris.kirby@sun.com /* 67510342Schris.kirby@sun.com * Walk through the pool-wide zap object of temporary snapshot user holds 67610342Schris.kirby@sun.com * and release them. 67710342Schris.kirby@sun.com */ 67810342Schris.kirby@sun.com void 67910342Schris.kirby@sun.com dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) 68010342Schris.kirby@sun.com { 68110342Schris.kirby@sun.com zap_attribute_t za; 68210342Schris.kirby@sun.com zap_cursor_t zc; 68310342Schris.kirby@sun.com objset_t *mos = dp->dp_meta_objset; 68410342Schris.kirby@sun.com uint64_t zapobj = dp->dp_tmp_userrefs_obj; 68510342Schris.kirby@sun.com 68610342Schris.kirby@sun.com if (zapobj == 0) 68710342Schris.kirby@sun.com return; 68810342Schris.kirby@sun.com ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 68910342Schris.kirby@sun.com 69010342Schris.kirby@sun.com for (zap_cursor_init(&zc, mos, zapobj); 69110342Schris.kirby@sun.com zap_cursor_retrieve(&zc, &za) == 0; 69210342Schris.kirby@sun.com zap_cursor_advance(&zc)) { 69310342Schris.kirby@sun.com char *htag; 69410342Schris.kirby@sun.com uint64_t dsobj; 69510342Schris.kirby@sun.com 69610342Schris.kirby@sun.com htag = strchr(za.za_name, '-'); 69710342Schris.kirby@sun.com *htag = '\0'; 69810342Schris.kirby@sun.com ++htag; 69910342Schris.kirby@sun.com dsobj = strtonum(za.za_name, NULL); 70010342Schris.kirby@sun.com (void) dsl_dataset_user_release_tmp(dp, dsobj, htag); 70110342Schris.kirby@sun.com } 70210342Schris.kirby@sun.com zap_cursor_fini(&zc); 70310342Schris.kirby@sun.com } 70410342Schris.kirby@sun.com 70510342Schris.kirby@sun.com /* 70610342Schris.kirby@sun.com * Create the pool-wide zap object for storing temporary snapshot holds. 70710342Schris.kirby@sun.com */ 70810342Schris.kirby@sun.com void 70910342Schris.kirby@sun.com dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) 71010342Schris.kirby@sun.com { 71110342Schris.kirby@sun.com objset_t *mos = dp->dp_meta_objset; 71210342Schris.kirby@sun.com 71310342Schris.kirby@sun.com ASSERT(dp->dp_tmp_userrefs_obj == 0); 71410342Schris.kirby@sun.com ASSERT(dmu_tx_is_syncing(tx)); 71510342Schris.kirby@sun.com 71610342Schris.kirby@sun.com dp->dp_tmp_userrefs_obj = zap_create(mos, DMU_OT_USERREFS, 71710342Schris.kirby@sun.com DMU_OT_NONE, 0, tx); 71810342Schris.kirby@sun.com 71910342Schris.kirby@sun.com VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, 72010342Schris.kirby@sun.com sizeof (uint64_t), 1, &dp->dp_tmp_userrefs_obj, tx) == 0); 72110342Schris.kirby@sun.com } 72210342Schris.kirby@sun.com 72310342Schris.kirby@sun.com static int 72410342Schris.kirby@sun.com dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, 72510951SChris.Kirby@sun.com const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding) 72610342Schris.kirby@sun.com { 72710342Schris.kirby@sun.com objset_t *mos = dp->dp_meta_objset; 72810342Schris.kirby@sun.com uint64_t zapobj = dp->dp_tmp_userrefs_obj; 72910342Schris.kirby@sun.com char *name; 73010342Schris.kirby@sun.com int error; 73110342Schris.kirby@sun.com 73210342Schris.kirby@sun.com ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 73310342Schris.kirby@sun.com ASSERT(dmu_tx_is_syncing(tx)); 73410342Schris.kirby@sun.com 73510342Schris.kirby@sun.com /* 73610342Schris.kirby@sun.com * If the pool was created prior to SPA_VERSION_USERREFS, the 73710342Schris.kirby@sun.com * zap object for temporary holds might not exist yet. 73810342Schris.kirby@sun.com */ 73910342Schris.kirby@sun.com if (zapobj == 0) { 74010342Schris.kirby@sun.com if (holding) { 74110342Schris.kirby@sun.com dsl_pool_user_hold_create_obj(dp, tx); 74210342Schris.kirby@sun.com zapobj = dp->dp_tmp_userrefs_obj; 74310342Schris.kirby@sun.com } else { 74410342Schris.kirby@sun.com return (ENOENT); 74510342Schris.kirby@sun.com } 74610342Schris.kirby@sun.com } 74710342Schris.kirby@sun.com 74810342Schris.kirby@sun.com name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); 74910342Schris.kirby@sun.com if (holding) 75010951SChris.Kirby@sun.com error = zap_add(mos, zapobj, name, 8, 1, now, tx); 75110342Schris.kirby@sun.com else 75210342Schris.kirby@sun.com error = zap_remove(mos, zapobj, name, tx); 75310342Schris.kirby@sun.com strfree(name); 75410342Schris.kirby@sun.com 75510342Schris.kirby@sun.com return (error); 75610342Schris.kirby@sun.com } 75710342Schris.kirby@sun.com 75810342Schris.kirby@sun.com /* 75910342Schris.kirby@sun.com * Add a temporary hold for the given dataset object and tag. 76010342Schris.kirby@sun.com */ 76110342Schris.kirby@sun.com int 76210342Schris.kirby@sun.com dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 76310951SChris.Kirby@sun.com uint64_t *now, dmu_tx_t *tx) 76410342Schris.kirby@sun.com { 76510951SChris.Kirby@sun.com return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); 76610342Schris.kirby@sun.com } 76710342Schris.kirby@sun.com 76810342Schris.kirby@sun.com /* 76910342Schris.kirby@sun.com * Release a temporary hold for the given dataset object and tag. 77010342Schris.kirby@sun.com */ 77110342Schris.kirby@sun.com int 77210342Schris.kirby@sun.com dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 77310342Schris.kirby@sun.com dmu_tx_t *tx) 77410342Schris.kirby@sun.com { 77510342Schris.kirby@sun.com return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL, 77610342Schris.kirby@sun.com tx, B_FALSE)); 77710342Schris.kirby@sun.com } 778