diff --git a/module/zfs/dmu_objset.c b/module/zfs/dmu_objset.c index ed0e8de38e5c..5ad3b2f269d6 100644 --- a/module/zfs/dmu_objset.c +++ b/module/zfs/dmu_objset.c @@ -1,3074 +1,3081 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012, 2020 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2013, Joyent, Inc. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. * Copyright (c) 2016 Actifio, Inc. All rights reserved. * Copyright 2017 Nexenta Systems, Inc. * Copyright (c) 2017 Open-E, Inc. All Rights Reserved. * Copyright (c) 2018, loli10K . All rights reserved. * Copyright (c) 2019, Klara Inc. * Copyright (c) 2019, Allan Jude * Copyright (c) 2022 Hewlett Packard Enterprise Development LP. */ /* Portions Copyright 2010 Robert Milkowski */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "zfs_namecheck.h" #include #include /* * Needed to close a window in dnode_move() that allows the objset to be freed * before it can be safely accessed. */ krwlock_t os_lock; /* * Tunable to overwrite the maximum number of threads for the parallelization * of dmu_objset_find_dp, needed to speed up the import of pools with many * datasets. * Default is 4 times the number of leaf vdevs. */ int dmu_find_threads = 0; /* * Backfill lower metadnode objects after this many have been freed. * Backfilling negatively impacts object creation rates, so only do it * if there are enough holes to fill. */ int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; static char *upgrade_tag = "upgrade_tag"; static void dmu_objset_find_dp_cb(void *arg); static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); static void dmu_objset_upgrade_stop(objset_t *os); void dmu_objset_init(void) { rw_init(&os_lock, NULL, RW_DEFAULT, NULL); } void dmu_objset_fini(void) { rw_destroy(&os_lock); } spa_t * dmu_objset_spa(objset_t *os) { return (os->os_spa); } zilog_t * dmu_objset_zil(objset_t *os) { return (os->os_zil); } dsl_pool_t * dmu_objset_pool(objset_t *os) { dsl_dataset_t *ds; if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) return (ds->ds_dir->dd_pool); else return (spa_get_dsl(os->os_spa)); } dsl_dataset_t * dmu_objset_ds(objset_t *os) { return (os->os_dsl_dataset); } dmu_objset_type_t dmu_objset_type(objset_t *os) { return (os->os_phys->os_type); } void dmu_objset_name(objset_t *os, char *buf) { dsl_dataset_name(os->os_dsl_dataset, buf); } uint64_t dmu_objset_id(objset_t *os) { dsl_dataset_t *ds = os->os_dsl_dataset; return (ds ? ds->ds_object : 0); } uint64_t dmu_objset_dnodesize(objset_t *os) { return (os->os_dnodesize); } zfs_sync_type_t dmu_objset_syncprop(objset_t *os) { return (os->os_sync); } zfs_logbias_op_t dmu_objset_logbias(objset_t *os) { return (os->os_logbias); } static void checksum_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance should have been done by now. */ ASSERT(newval != ZIO_CHECKSUM_INHERIT); os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); } static void compression_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval != ZIO_COMPRESS_INHERIT); os->os_compress = zio_compress_select(os->os_spa, ZIO_COMPRESS_ALGO(newval), ZIO_COMPRESS_ON); os->os_complevel = zio_complevel_select(os->os_spa, os->os_compress, ZIO_COMPRESS_LEVEL(newval), ZIO_COMPLEVEL_DEFAULT); } static void copies_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval > 0); ASSERT(newval <= spa_max_replication(os->os_spa)); os->os_copies = newval; } static void dedup_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; spa_t *spa = os->os_spa; enum zio_checksum checksum; /* * Inheritance should have been done by now. */ ASSERT(newval != ZIO_CHECKSUM_INHERIT); checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); } static void primary_cache_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || newval == ZFS_CACHE_METADATA); os->os_primary_cache = newval; } static void secondary_cache_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || newval == ZFS_CACHE_METADATA); os->os_secondary_cache = newval; } static void sync_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || newval == ZFS_SYNC_DISABLED); os->os_sync = newval; if (os->os_zil) zil_set_sync(os->os_zil, newval); } static void redundant_metadata_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || newval == ZFS_REDUNDANT_METADATA_MOST || newval == ZFS_REDUNDANT_METADATA_SOME || newval == ZFS_REDUNDANT_METADATA_NONE); os->os_redundant_metadata = newval; } static void dnodesize_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; switch (newval) { case ZFS_DNSIZE_LEGACY: os->os_dnodesize = DNODE_MIN_SIZE; break; case ZFS_DNSIZE_AUTO: /* * Choose a dnode size that will work well for most * workloads if the user specified "auto". Future code * improvements could dynamically select a dnode size * based on observed workload patterns. */ os->os_dnodesize = DNODE_MIN_SIZE * 2; break; case ZFS_DNSIZE_1K: case ZFS_DNSIZE_2K: case ZFS_DNSIZE_4K: case ZFS_DNSIZE_8K: case ZFS_DNSIZE_16K: os->os_dnodesize = newval; break; } } static void smallblk_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval <= SPA_MAXBLOCKSIZE); ASSERT(ISP2(newval)); os->os_zpl_special_smallblock = newval; } static void logbias_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; ASSERT(newval == ZFS_LOGBIAS_LATENCY || newval == ZFS_LOGBIAS_THROUGHPUT); os->os_logbias = newval; if (os->os_zil) zil_set_logbias(os->os_zil, newval); } static void recordsize_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; os->os_recordsize = newval; } void dmu_objset_byteswap(void *buf, size_t size) { objset_phys_t *osp = buf; ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 || size == sizeof (objset_phys_t)); dnode_byteswap(&osp->os_meta_dnode); byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); osp->os_type = BSWAP_64(osp->os_type); osp->os_flags = BSWAP_64(osp->os_flags); if (size >= OBJSET_PHYS_SIZE_V2) { dnode_byteswap(&osp->os_userused_dnode); dnode_byteswap(&osp->os_groupused_dnode); if (size >= sizeof (objset_phys_t)) dnode_byteswap(&osp->os_projectused_dnode); } } /* * The hash is a CRC-based hash of the objset_t pointer and the object number. */ static uint64_t dnode_hash(const objset_t *os, uint64_t obj) { uintptr_t osv = (uintptr_t)os; uint64_t crc = -1ULL; ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); /* * The low 6 bits of the pointer don't have much entropy, because * the objset_t is larger than 2^6 bytes long. */ crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; crc ^= (osv>>14) ^ (obj>>24); return (crc); } static unsigned int dnode_multilist_index_func(multilist_t *ml, void *obj) { dnode_t *dn = obj; /* * The low order bits of the hash value are thought to be * distributed evenly. Otherwise, in the case that the multilist * has a power of two number of sublists, each sublists' usage * would not be evenly distributed. In this context full 64bit * division would be a waste of time, so limit it to 32 bits. */ return ((unsigned int)dnode_hash(dn->dn_objset, dn->dn_object) % multilist_get_num_sublists(ml)); } static inline boolean_t dmu_os_is_l2cacheable(objset_t *os) { vdev_t *vd = NULL; zfs_cache_type_t cache = os->os_secondary_cache; blkptr_t *bp = os->os_rootbp; if (bp != NULL && !BP_IS_HOLE(bp)) { uint64_t vdev = DVA_GET_VDEV(bp->blk_dva); vdev_t *rvd = os->os_spa->spa_root_vdev; if (vdev < rvd->vdev_children) vd = rvd->vdev_child[vdev]; if (cache == ZFS_CACHE_ALL || cache == ZFS_CACHE_METADATA) { if (vd == NULL) return (B_TRUE); if ((vd->vdev_alloc_bias != VDEV_BIAS_SPECIAL && vd->vdev_alloc_bias != VDEV_BIAS_DEDUP) || l2arc_exclude_special == 0) return (B_TRUE); } } return (B_FALSE); } /* * Instantiates the objset_t in-memory structure corresponding to the * objset_phys_t that's pointed to by the specified blkptr_t. */ int dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, objset_t **osp) { objset_t *os; int i, err; ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); ASSERT(!BP_IS_REDACTED(bp)); /* * We need the pool config lock to get properties. */ ASSERT(ds == NULL || dsl_pool_config_held(ds->ds_dir->dd_pool)); /* * The $ORIGIN dataset (if it exists) doesn't have an associated * objset, so there's no reason to open it. The $ORIGIN dataset * will not exist on pools older than SPA_VERSION_ORIGIN. */ if (ds != NULL && spa_get_dsl(spa) != NULL && spa_get_dsl(spa)->dp_origin_snap != NULL) { ASSERT3P(ds->ds_dir, !=, spa_get_dsl(spa)->dp_origin_snap->ds_dir); } os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); os->os_dsl_dataset = ds; os->os_spa = spa; os->os_rootbp = bp; if (!BP_IS_HOLE(os->os_rootbp)) { arc_flags_t aflags = ARC_FLAG_WAIT; zbookmark_phys_t zb; int size; enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); if (dmu_os_is_l2cacheable(os)) aflags |= ARC_FLAG_L2CACHE; if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); ASSERT(BP_IS_AUTHENTICATED(bp)); zio_flags |= ZIO_FLAG_RAW; } dprintf_bp(os->os_rootbp, "reading %s", ""); err = arc_read(NULL, spa, os->os_rootbp, arc_getbuf_func, &os->os_phys_buf, ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); if (err != 0) { kmem_free(os, sizeof (objset_t)); /* convert checksum errors into IO errors */ if (err == ECKSUM) err = SET_ERROR(EIO); return (err); } if (spa_version(spa) < SPA_VERSION_USERSPACE) size = OBJSET_PHYS_SIZE_V1; else if (!spa_feature_is_enabled(spa, SPA_FEATURE_PROJECT_QUOTA)) size = OBJSET_PHYS_SIZE_V2; else size = sizeof (objset_phys_t); /* Increase the blocksize if we are permitted. */ if (arc_buf_size(os->os_phys_buf) < size) { arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, ARC_BUFC_METADATA, size); bzero(buf->b_data, size); bcopy(os->os_phys_buf->b_data, buf->b_data, arc_buf_size(os->os_phys_buf)); arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); os->os_phys_buf = buf; } os->os_phys = os->os_phys_buf->b_data; os->os_flags = os->os_phys->os_flags; } else { int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1; os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, ARC_BUFC_METADATA, size); os->os_phys = os->os_phys_buf->b_data; bzero(os->os_phys, size); } /* * These properties will be filled in by the logic in zfs_get_zplprop() * when they are queried for the first time. */ os->os_version = OBJSET_PROP_UNINITIALIZED; os->os_normalization = OBJSET_PROP_UNINITIALIZED; os->os_utf8only = OBJSET_PROP_UNINITIALIZED; os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; /* * Note: the changed_cb will be called once before the register * func returns, thus changing the checksum/compression from the * default (fletcher2/off). Snapshots don't need to know about * checksum/compression/copies. */ if (ds != NULL) { os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), primary_cache_changed_cb, os); if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), secondary_cache_changed_cb, os); } if (!ds->ds_is_snapshot) { if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_COMPRESSION), compression_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_COPIES), copies_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_DEDUP), dedup_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_LOGBIAS), logbias_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_SYNC), sync_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name( ZFS_PROP_REDUNDANT_METADATA), redundant_metadata_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE), recordsize_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_DNODESIZE), dnodesize_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name( ZFS_PROP_SPECIAL_SMALL_BLOCKS), smallblk_changed_cb, os); } } if (err != 0) { arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); kmem_free(os, sizeof (objset_t)); return (err); } } else { /* It's the meta-objset. */ os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; os->os_compress = ZIO_COMPRESS_ON; os->os_complevel = ZIO_COMPLEVEL_DEFAULT; os->os_encrypted = B_FALSE; os->os_copies = spa_max_replication(spa); os->os_dedup_checksum = ZIO_CHECKSUM_OFF; os->os_dedup_verify = B_FALSE; os->os_logbias = ZFS_LOGBIAS_LATENCY; os->os_sync = ZFS_SYNC_STANDARD; os->os_primary_cache = ZFS_CACHE_ALL; os->os_secondary_cache = ZFS_CACHE_ALL; os->os_dnodesize = DNODE_MIN_SIZE; } if (ds == NULL || !ds->ds_is_snapshot) os->os_zil_header = os->os_phys->os_zil_header; os->os_zil = zil_alloc(os, &os->os_zil_header); for (i = 0; i < TXG_SIZE; i++) { multilist_create(&os->os_dirty_dnodes[i], sizeof (dnode_t), offsetof(dnode_t, dn_dirty_link[i]), dnode_multilist_index_func); } list_create(&os->os_dnodes, sizeof (dnode_t), offsetof(dnode_t, dn_link)); list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), offsetof(dmu_buf_impl_t, db_link)); list_link_init(&os->os_evicting_node); mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); os->os_obj_next_percpu_len = boot_ncpus; os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); dnode_special_open(os, &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT, &os->os_meta_dnode); if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) { dnode_special_open(os, &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT, &os->os_userused_dnode); dnode_special_open(os, &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf)) dnode_special_open(os, &os->os_phys->os_projectused_dnode, DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode); } mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); *osp = os; return (0); } int dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) { int err = 0; /* * We need the pool_config lock to manipulate the dsl_dataset_t. * Even if the dataset is long-held, we need the pool_config lock * to open the objset, as it needs to get properties. */ ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); mutex_enter(&ds->ds_opening_lock); if (ds->ds_objset == NULL) { objset_t *os; rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), ds, dsl_dataset_get_blkptr(ds), &os); rrw_exit(&ds->ds_bp_rwlock, FTAG); if (err == 0) { mutex_enter(&ds->ds_lock); ASSERT(ds->ds_objset == NULL); ds->ds_objset = os; mutex_exit(&ds->ds_lock); } } *osp = ds->ds_objset; mutex_exit(&ds->ds_opening_lock); return (err); } /* * Holds the pool while the objset is held. Therefore only one objset * can be held at a time. */ int dmu_objset_hold_flags(const char *name, boolean_t decrypt, void *tag, objset_t **osp) { dsl_pool_t *dp; dsl_dataset_t *ds; int err; ds_hold_flags_t flags; flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; err = dsl_pool_hold(name, tag, &dp); if (err != 0) return (err); err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); if (err != 0) { dsl_pool_rele(dp, tag); return (err); } err = dmu_objset_from_ds(ds, osp); if (err != 0) { dsl_dataset_rele(ds, tag); dsl_pool_rele(dp, tag); } return (err); } int dmu_objset_hold(const char *name, void *tag, objset_t **osp) { return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); } static int dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) { (void) tag; int err = dmu_objset_from_ds(ds, osp); if (err != 0) { return (err); } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { return (SET_ERROR(EINVAL)); } else if (!readonly && dsl_dataset_is_snapshot(ds)) { return (SET_ERROR(EROFS)); } else if (!readonly && decrypt && dsl_dir_incompatible_encryption_version(ds->ds_dir)) { return (SET_ERROR(EROFS)); } /* if we are decrypting, we can now check MACs in os->os_phys_buf */ if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { zbookmark_phys_t zb; SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, &zb, B_FALSE); if (err != 0) return (err); ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); } return (0); } /* * dsl_pool must not be held when this is called. * Upon successful return, there will be a longhold on the dataset, * and the dsl_pool will not be held. */ int dmu_objset_own(const char *name, dmu_objset_type_t type, boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) { dsl_pool_t *dp; dsl_dataset_t *ds; int err; ds_hold_flags_t flags; flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; err = dsl_pool_hold(name, FTAG, &dp); if (err != 0) return (err); err = dsl_dataset_own(dp, name, flags, tag, &ds); if (err != 0) { dsl_pool_rele(dp, FTAG); return (err); } err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); if (err != 0) { dsl_dataset_disown(ds, flags, tag); dsl_pool_rele(dp, FTAG); return (err); } /* * User accounting requires the dataset to be decrypted and rw. * We also don't begin user accounting during claiming to help * speed up pool import times and to keep this txg reserved * completely for recovery work. */ if (!readonly && !dp->dp_spa->spa_claiming && (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) { if (dmu_objset_userobjspace_upgradable(*osp) || dmu_objset_projectquota_upgradable(*osp)) { dmu_objset_id_quota_upgrade(*osp); } else if (dmu_objset_userused_enabled(*osp)) { dmu_objset_userspace_upgrade(*osp); } } dsl_pool_rele(dp, FTAG); return (0); } int dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) { dsl_dataset_t *ds; int err; ds_hold_flags_t flags; flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); if (err != 0) return (err); err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); if (err != 0) { dsl_dataset_disown(ds, flags, tag); return (err); } return (0); } void dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, void *tag) { ds_hold_flags_t flags; dsl_pool_t *dp = dmu_objset_pool(os); flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); dsl_pool_rele(dp, tag); } void dmu_objset_rele(objset_t *os, void *tag) { dmu_objset_rele_flags(os, B_FALSE, tag); } /* * When we are called, os MUST refer to an objset associated with a dataset * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner * == tag. We will then release and reacquire ownership of the dataset while * holding the pool config_rwlock to avoid intervening namespace or ownership * changes may occur. * * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to * release the hold on its dataset and acquire a new one on the dataset of the * same name so that it can be partially torn down and reconstructed. */ void dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, boolean_t decrypt, void *tag) { dsl_pool_t *dp; char name[ZFS_MAX_DATASET_NAME_LEN]; ds_hold_flags_t flags; flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; VERIFY3P(ds, !=, NULL); VERIFY3P(ds->ds_owner, ==, tag); VERIFY(dsl_dataset_long_held(ds)); dsl_dataset_name(ds, name); dp = ds->ds_dir->dd_pool; dsl_pool_config_enter(dp, FTAG); dsl_dataset_disown(ds, flags, tag); VERIFY0(dsl_dataset_own(dp, name, flags, tag, newds)); dsl_pool_config_exit(dp, FTAG); } void dmu_objset_disown(objset_t *os, boolean_t decrypt, void *tag) { ds_hold_flags_t flags; flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; /* * Stop upgrading thread */ dmu_objset_upgrade_stop(os); dsl_dataset_disown(os->os_dsl_dataset, flags, tag); } void dmu_objset_evict_dbufs(objset_t *os) { dnode_t *dn_marker; dnode_t *dn; dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); mutex_enter(&os->os_lock); dn = list_head(&os->os_dnodes); while (dn != NULL) { /* * Skip dnodes without holds. We have to do this dance * because dnode_add_ref() only works if there is already a * hold. If the dnode has no holds, then it has no dbufs. */ if (dnode_add_ref(dn, FTAG)) { list_insert_after(&os->os_dnodes, dn, dn_marker); mutex_exit(&os->os_lock); dnode_evict_dbufs(dn); dnode_rele(dn, FTAG); mutex_enter(&os->os_lock); dn = list_next(&os->os_dnodes, dn_marker); list_remove(&os->os_dnodes, dn_marker); } else { dn = list_next(&os->os_dnodes, dn); } } mutex_exit(&os->os_lock); kmem_free(dn_marker, sizeof (dnode_t)); if (DMU_USERUSED_DNODE(os) != NULL) { if (DMU_PROJECTUSED_DNODE(os) != NULL) dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os)); dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); } dnode_evict_dbufs(DMU_META_DNODE(os)); } /* * Objset eviction processing is split into into two pieces. * The first marks the objset as evicting, evicts any dbufs that * have a refcount of zero, and then queues up the objset for the * second phase of eviction. Once os->os_dnodes has been cleared by * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. * The second phase closes the special dnodes, dequeues the objset from * the list of those undergoing eviction, and finally frees the objset. * * NOTE: Due to asynchronous eviction processing (invocation of * dnode_buf_pageout()), it is possible for the meta dnode for the * objset to have no holds even though os->os_dnodes is not empty. */ void dmu_objset_evict(objset_t *os) { dsl_dataset_t *ds = os->os_dsl_dataset; for (int t = 0; t < TXG_SIZE; t++) ASSERT(!dmu_objset_is_dirty(os, t)); if (ds) dsl_prop_unregister_all(ds, os); if (os->os_sa) sa_tear_down(os); dmu_objset_evict_dbufs(os); mutex_enter(&os->os_lock); spa_evicting_os_register(os->os_spa, os); if (list_is_empty(&os->os_dnodes)) { mutex_exit(&os->os_lock); dmu_objset_evict_done(os); } else { mutex_exit(&os->os_lock); } } void dmu_objset_evict_done(objset_t *os) { ASSERT3P(list_head(&os->os_dnodes), ==, NULL); dnode_special_close(&os->os_meta_dnode); if (DMU_USERUSED_DNODE(os)) { if (DMU_PROJECTUSED_DNODE(os)) dnode_special_close(&os->os_projectused_dnode); dnode_special_close(&os->os_userused_dnode); dnode_special_close(&os->os_groupused_dnode); } zil_free(os->os_zil); arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); /* * This is a barrier to prevent the objset from going away in * dnode_move() until we can safely ensure that the objset is still in * use. We consider the objset valid before the barrier and invalid * after the barrier. */ rw_enter(&os_lock, RW_READER); rw_exit(&os_lock); kmem_free(os->os_obj_next_percpu, os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); mutex_destroy(&os->os_lock); mutex_destroy(&os->os_userused_lock); mutex_destroy(&os->os_obj_lock); mutex_destroy(&os->os_user_ptr_lock); mutex_destroy(&os->os_upgrade_lock); for (int i = 0; i < TXG_SIZE; i++) multilist_destroy(&os->os_dirty_dnodes[i]); spa_evicting_os_deregister(os->os_spa, os); kmem_free(os, sizeof (objset_t)); } inode_timespec_t dmu_objset_snap_cmtime(objset_t *os) { return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); } objset_t * dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) { objset_t *os; dnode_t *mdn; ASSERT(dmu_tx_is_syncing(tx)); if (blksz == 0) blksz = DNODE_BLOCK_SIZE; if (ibs == 0) ibs = DN_MAX_INDBLKSHIFT; if (ds != NULL) VERIFY0(dmu_objset_from_ds(ds, &os)); else VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); mdn = DMU_META_DNODE(os); dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0, DNODE_MIN_SLOTS, tx); /* * We don't want to have to increase the meta-dnode's nlevels * later, because then we could do it in quiescing context while * we are also accessing it in open context. * * This precaution is not necessary for the MOS (ds == NULL), * because the MOS is only updated in syncing context. * This is most fortunate: the MOS is the only objset that * needs to be synced multiple times as spa_sync() iterates * to convergence, so minimizing its dn_nlevels matters. */ if (ds != NULL) { if (levels == 0) { levels = 1; /* * Determine the number of levels necessary for the * meta-dnode to contain DN_MAX_OBJECT dnodes. Note * that in order to ensure that we do not overflow * 64 bits, there has to be a nlevels that gives us a * number of blocks > DN_MAX_OBJECT but < 2^64. * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) * (10) must be less than (64 - log2(DN_MAX_OBJECT)) * (16). */ while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < DN_MAX_OBJECT) levels++; } mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = mdn->dn_nlevels = levels; } ASSERT(type != DMU_OST_NONE); ASSERT(type != DMU_OST_ANY); ASSERT(type < DMU_OST_NUMTYPES); os->os_phys->os_type = type; /* * Enable user accounting if it is enabled and this is not an * encrypted receive. */ if (dmu_objset_userused_enabled(os) && (!os->os_encrypted || !dmu_objset_is_receiving(os))) { os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; if (dmu_objset_userobjused_enabled(os)) { ds->ds_feature_activation[ SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; os->os_phys->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; } if (dmu_objset_projectquota_enabled(os)) { ds->ds_feature_activation[ SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; os->os_phys->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; } os->os_flags = os->os_phys->os_flags; } dsl_dataset_dirty(ds, tx); return (os); } /* called from dsl for meta-objset */ objset_t * dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, dmu_objset_type_t type, dmu_tx_t *tx) { return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); } typedef struct dmu_objset_create_arg { const char *doca_name; cred_t *doca_cred; proc_t *doca_proc; void (*doca_userfunc)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx); void *doca_userarg; dmu_objset_type_t doca_type; uint64_t doca_flags; dsl_crypto_params_t *doca_dcp; } dmu_objset_create_arg_t; static int dmu_objset_create_check(void *arg, dmu_tx_t *tx) { dmu_objset_create_arg_t *doca = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dir_t *pdd; dsl_dataset_t *parentds; objset_t *parentos; const char *tail; int error; if (strchr(doca->doca_name, '@') != NULL) return (SET_ERROR(EINVAL)); if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) return (SET_ERROR(ENAMETOOLONG)); if (dataset_nestcheck(doca->doca_name) != 0) return (SET_ERROR(ENAMETOOLONG)); error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); if (error != 0) return (error); if (tail == NULL) { dsl_dir_rele(pdd, FTAG); return (SET_ERROR(EEXIST)); } error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL); if (error != 0) { dsl_dir_rele(pdd, FTAG); return (error); } error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, doca->doca_cred, doca->doca_proc); if (error != 0) { dsl_dir_rele(pdd, FTAG); return (error); } /* can't create below anything but filesystems (eg. no ZVOLs) */ error = dsl_dataset_hold_obj(pdd->dd_pool, dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds); if (error != 0) { dsl_dir_rele(pdd, FTAG); return (error); } error = dmu_objset_from_ds(parentds, &parentos); if (error != 0) { dsl_dataset_rele(parentds, FTAG); dsl_dir_rele(pdd, FTAG); return (error); } if (dmu_objset_type(parentos) != DMU_OST_ZFS) { dsl_dataset_rele(parentds, FTAG); dsl_dir_rele(pdd, FTAG); return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); } dsl_dataset_rele(parentds, FTAG); dsl_dir_rele(pdd, FTAG); return (error); } static void dmu_objset_create_sync(void *arg, dmu_tx_t *tx) { dmu_objset_create_arg_t *doca = arg; dsl_pool_t *dp = dmu_tx_pool(tx); spa_t *spa = dp->dp_spa; dsl_dir_t *pdd; const char *tail; dsl_dataset_t *ds; uint64_t obj; blkptr_t *bp; objset_t *os; zio_t *rzio; VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, doca->doca_cred, doca->doca_dcp, tx); VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); bp = dsl_dataset_get_blkptr(ds); os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx); rrw_exit(&ds->ds_bp_rwlock, FTAG); if (doca->doca_userfunc != NULL) { doca->doca_userfunc(os, doca->doca_userarg, doca->doca_cred, tx); } /* * The doca_userfunc() may write out some data that needs to be * encrypted if the dataset is encrypted (specifically the root * directory). This data must be written out before the encryption * key mapping is removed by dsl_dataset_rele_flags(). Force the * I/O to occur immediately by invoking the relevant sections of * dsl_pool_sync(). */ if (os->os_encrypted) { dsl_dataset_t *tmpds = NULL; boolean_t need_sync_done = B_FALSE; mutex_enter(&ds->ds_lock); ds->ds_owner = FTAG; mutex_exit(&ds->ds_lock); rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, tx->tx_txg); if (tmpds != NULL) { dsl_dataset_sync(ds, rzio, tx); need_sync_done = B_TRUE; } VERIFY0(zio_wait(rzio)); dmu_objset_sync_done(os, tx); taskq_wait(dp->dp_sync_taskq); if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { ASSERT3P(ds->ds_key_mapping, !=, NULL); key_mapping_rele(spa, ds->ds_key_mapping, ds); } rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, tx->tx_txg); if (tmpds != NULL) { dmu_buf_rele(ds->ds_dbuf, ds); dsl_dataset_sync(ds, rzio, tx); } VERIFY0(zio_wait(rzio)); if (need_sync_done) { ASSERT3P(ds->ds_key_mapping, !=, NULL); key_mapping_rele(spa, ds->ds_key_mapping, ds); dsl_dataset_sync_done(ds, tx); } mutex_enter(&ds->ds_lock); ds->ds_owner = NULL; mutex_exit(&ds->ds_lock); } spa_history_log_internal_ds(ds, "create", tx, " "); dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); dsl_dir_rele(pdd, FTAG); } int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) { dmu_objset_create_arg_t doca; dsl_crypto_params_t tmp_dcp = { 0 }; doca.doca_name = name; doca.doca_cred = CRED(); doca.doca_proc = curproc; doca.doca_flags = flags; doca.doca_userfunc = func; doca.doca_userarg = arg; doca.doca_type = type; /* * Some callers (mostly for testing) do not provide a dcp on their * own but various code inside the sync task will require it to be * allocated. Rather than adding NULL checks throughout this code * or adding dummy dcp's to all of the callers we simply create a * dummy one here and use that. This zero dcp will have the same * effect as asking for inheritance of all encryption params. */ doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; int rv = dsl_sync_task(name, dmu_objset_create_check, dmu_objset_create_sync, &doca, 6, ZFS_SPACE_CHECK_NORMAL); if (rv == 0) zvol_create_minor(name); return (rv); } typedef struct dmu_objset_clone_arg { const char *doca_clone; const char *doca_origin; cred_t *doca_cred; proc_t *doca_proc; } dmu_objset_clone_arg_t; static int dmu_objset_clone_check(void *arg, dmu_tx_t *tx) { dmu_objset_clone_arg_t *doca = arg; dsl_dir_t *pdd; const char *tail; int error; dsl_dataset_t *origin; dsl_pool_t *dp = dmu_tx_pool(tx); if (strchr(doca->doca_clone, '@') != NULL) return (SET_ERROR(EINVAL)); if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) return (SET_ERROR(ENAMETOOLONG)); error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); if (error != 0) return (error); if (tail == NULL) { dsl_dir_rele(pdd, FTAG); return (SET_ERROR(EEXIST)); } error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, doca->doca_cred, doca->doca_proc); if (error != 0) { dsl_dir_rele(pdd, FTAG); return (SET_ERROR(EDQUOT)); } error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); if (error != 0) { dsl_dir_rele(pdd, FTAG); return (error); } /* You can only clone snapshots, not the head datasets. */ if (!origin->ds_is_snapshot) { dsl_dataset_rele(origin, FTAG); dsl_dir_rele(pdd, FTAG); return (SET_ERROR(EINVAL)); } dsl_dataset_rele(origin, FTAG); dsl_dir_rele(pdd, FTAG); return (0); } static void dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) { dmu_objset_clone_arg_t *doca = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dir_t *pdd; const char *tail; dsl_dataset_t *origin, *ds; uint64_t obj; char namebuf[ZFS_MAX_DATASET_NAME_LEN]; VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); obj = dsl_dataset_create_sync(pdd, tail, origin, 0, doca->doca_cred, NULL, tx); VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); dsl_dataset_name(origin, namebuf); spa_history_log_internal_ds(ds, "clone", tx, "origin=%s (%llu)", namebuf, (u_longlong_t)origin->ds_object); dsl_dataset_rele(ds, FTAG); dsl_dataset_rele(origin, FTAG); dsl_dir_rele(pdd, FTAG); } int dmu_objset_clone(const char *clone, const char *origin) { dmu_objset_clone_arg_t doca; doca.doca_clone = clone; doca.doca_origin = origin; doca.doca_cred = CRED(); doca.doca_proc = curproc; int rv = dsl_sync_task(clone, dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 6, ZFS_SPACE_CHECK_NORMAL); if (rv == 0) zvol_create_minor(clone); return (rv); } int dmu_objset_snapshot_one(const char *fsname, const char *snapname) { int err; char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); nvlist_t *snaps = fnvlist_alloc(); fnvlist_add_boolean(snaps, longsnap); kmem_strfree(longsnap); err = dsl_dataset_snapshot(snaps, NULL, NULL); fnvlist_free(snaps); return (err); } static void dmu_objset_upgrade_task_cb(void *data) { objset_t *os = data; mutex_enter(&os->os_upgrade_lock); os->os_upgrade_status = EINTR; if (!os->os_upgrade_exit) { int status; mutex_exit(&os->os_upgrade_lock); status = os->os_upgrade_cb(os); mutex_enter(&os->os_upgrade_lock); os->os_upgrade_status = status; } os->os_upgrade_exit = B_TRUE; os->os_upgrade_id = 0; mutex_exit(&os->os_upgrade_lock); dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); } static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) { if (os->os_upgrade_id != 0) return; ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag); mutex_enter(&os->os_upgrade_lock); if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { os->os_upgrade_exit = B_FALSE; os->os_upgrade_cb = cb; os->os_upgrade_id = taskq_dispatch( os->os_spa->spa_upgrade_taskq, dmu_objset_upgrade_task_cb, os, TQ_SLEEP); if (os->os_upgrade_id == TASKQID_INVALID) { dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); os->os_upgrade_status = ENOMEM; } } else { dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); } mutex_exit(&os->os_upgrade_lock); } static void dmu_objset_upgrade_stop(objset_t *os) { mutex_enter(&os->os_upgrade_lock); os->os_upgrade_exit = B_TRUE; if (os->os_upgrade_id != 0) { taskqid_t id = os->os_upgrade_id; os->os_upgrade_id = 0; mutex_exit(&os->os_upgrade_lock); if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) { dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); } txg_wait_synced(os->os_spa->spa_dsl_pool, 0); } else { mutex_exit(&os->os_upgrade_lock); } } static void dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) { dnode_t *dn; while ((dn = multilist_sublist_head(list)) != NULL) { ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); ASSERT(dn->dn_dbuf->db_data_pending); /* * Initialize dn_zio outside dnode_sync() because the * meta-dnode needs to set it outside dnode_sync(). */ dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; ASSERT(dn->dn_zio); ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); multilist_sublist_remove(list, dn); /* * See the comment above dnode_rele_task() for an explanation * of why this dnode hold is always needed (even when not * doing user accounting). */ multilist_t *newlist = &dn->dn_objset->os_synced_dnodes; (void) dnode_add_ref(dn, newlist); multilist_insert(newlist, dn); dnode_sync(dn, tx); } } static void dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) { (void) abuf; blkptr_t *bp = zio->io_bp; objset_t *os = arg; dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; uint64_t fill = 0; ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); ASSERT0(BP_GET_LEVEL(bp)); /* * Update rootbp fill count: it should be the number of objects * allocated in the object set (not counting the "special" * objects that are stored in the objset_phys_t -- the meta * dnode and user/group/project accounting objects). */ for (int i = 0; i < dnp->dn_nblkptr; i++) fill += BP_GET_FILL(&dnp->dn_blkptr[i]); BP_SET_FILL(bp, fill); if (os->os_dsl_dataset != NULL) rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); *os->os_rootbp = *bp; if (os->os_dsl_dataset != NULL) rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); } static void dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) { (void) abuf; blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; objset_t *os = arg; if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { ASSERT(BP_EQUAL(bp, bp_orig)); } else { dsl_dataset_t *ds = os->os_dsl_dataset; dmu_tx_t *tx = os->os_synctx; (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); dsl_dataset_block_born(ds, bp, tx); } kmem_free(bp, sizeof (*bp)); } typedef struct sync_dnodes_arg { multilist_t *sda_list; int sda_sublist_idx; multilist_t *sda_newlist; dmu_tx_t *sda_tx; } sync_dnodes_arg_t; static void sync_dnodes_task(void *arg) { sync_dnodes_arg_t *sda = arg; multilist_sublist_t *ms = multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); dmu_objset_sync_dnodes(ms, sda->sda_tx); multilist_sublist_unlock(ms); kmem_free(sda, sizeof (*sda)); } /* called from dsl */ void dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) { int txgoff; zbookmark_phys_t zb; zio_prop_t zp; zio_t *zio; list_t *list; dbuf_dirty_record_t *dr; int num_sublists; multilist_t *ml; blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); *blkptr_copy = *os->os_rootbp; dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", (u_longlong_t)tx->tx_txg); ASSERT(dmu_tx_is_syncing(tx)); /* XXX the write_done callback should really give us the tx... */ os->os_synctx = tx; if (os->os_dsl_dataset == NULL) { /* * This is the MOS. If we have upgraded, * spa_max_replication() could change, so reset * os_copies here. */ os->os_copies = spa_max_replication(os->os_spa); } /* * Create the root block IO */ SET_BOOKMARK(&zb, os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : DMU_META_OBJSET, ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); arc_release(os->os_phys_buf, &os->os_phys_buf); dmu_write_policy(os, NULL, 0, 0, &zp); /* * If we are either claiming the ZIL or doing a raw receive, write * out the os_phys_buf raw. Neither of these actions will effect the * MAC at this point. */ if (os->os_raw_receive || os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { ASSERT(os->os_encrypted); arc_convert_to_raw(os->os_phys_buf, os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, DMU_OT_OBJSET, NULL, NULL, NULL); } zio = arc_write(pio, os->os_spa, tx->tx_txg, blkptr_copy, os->os_phys_buf, dmu_os_is_l2cacheable(os), &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); /* * Sync special dnodes - the parent IO for the sync is the root block */ DMU_META_DNODE(os)->dn_zio = zio; dnode_sync(DMU_META_DNODE(os), tx); os->os_phys->os_flags = os->os_flags; if (DMU_USERUSED_DNODE(os) && DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { DMU_USERUSED_DNODE(os)->dn_zio = zio; dnode_sync(DMU_USERUSED_DNODE(os), tx); DMU_GROUPUSED_DNODE(os)->dn_zio = zio; dnode_sync(DMU_GROUPUSED_DNODE(os), tx); } if (DMU_PROJECTUSED_DNODE(os) && DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) { DMU_PROJECTUSED_DNODE(os)->dn_zio = zio; dnode_sync(DMU_PROJECTUSED_DNODE(os), tx); } txgoff = tx->tx_txg & TXG_MASK; /* * We must create the list here because it uses the * dn_dirty_link[] of this txg. But it may already * exist because we call dsl_dataset_sync() twice per txg. */ if (os->os_synced_dnodes.ml_sublists == NULL) { multilist_create(&os->os_synced_dnodes, sizeof (dnode_t), offsetof(dnode_t, dn_dirty_link[txgoff]), dnode_multilist_index_func); } else { ASSERT3U(os->os_synced_dnodes.ml_offset, ==, offsetof(dnode_t, dn_dirty_link[txgoff])); } ml = &os->os_dirty_dnodes[txgoff]; num_sublists = multilist_get_num_sublists(ml); for (int i = 0; i < num_sublists; i++) { if (multilist_sublist_is_empty_idx(ml, i)) continue; sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); sda->sda_list = ml; sda->sda_sublist_idx = i; sda->sda_tx = tx; (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, sync_dnodes_task, sda, 0); /* callback frees sda */ } taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; while ((dr = list_head(list)) != NULL) { ASSERT0(dr->dr_dbuf->db_level); list_remove(list, dr); zio_nowait(dr->dr_zio); } /* Enable dnode backfill if enough objects have been freed. */ if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { os->os_rescan_dnodes = B_TRUE; os->os_freed_dnodes = 0; } /* * Free intent log blocks up to this tx. */ zil_sync(os->os_zil, tx); os->os_phys->os_zil_header = os->os_zil_header; zio_nowait(zio); } boolean_t dmu_objset_is_dirty(objset_t *os, uint64_t txg) { return (!multilist_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK])); } static file_info_cb_t *file_cbs[DMU_OST_NUMTYPES]; void dmu_objset_register_type(dmu_objset_type_t ost, file_info_cb_t *cb) { file_cbs[ost] = cb; } int dmu_get_file_info(objset_t *os, dmu_object_type_t bonustype, const void *data, zfs_file_info_t *zfi) { file_info_cb_t *cb = file_cbs[os->os_phys->os_type]; if (cb == NULL) return (EINVAL); return (cb(bonustype, data, zfi)); } boolean_t dmu_objset_userused_enabled(objset_t *os) { return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && file_cbs[os->os_phys->os_type] != NULL && DMU_USERUSED_DNODE(os) != NULL); } boolean_t dmu_objset_userobjused_enabled(objset_t *os) { return (dmu_objset_userused_enabled(os) && spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); } boolean_t dmu_objset_projectquota_enabled(objset_t *os) { return (file_cbs[os->os_phys->os_type] != NULL && DMU_PROJECTUSED_DNODE(os) != NULL && spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA)); } typedef struct userquota_node { /* must be in the first filed, see userquota_update_cache() */ char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; int64_t uqn_delta; avl_node_t uqn_node; } userquota_node_t; typedef struct userquota_cache { avl_tree_t uqc_user_deltas; avl_tree_t uqc_group_deltas; avl_tree_t uqc_project_deltas; } userquota_cache_t; static int userquota_compare(const void *l, const void *r) { const userquota_node_t *luqn = l; const userquota_node_t *ruqn = r; int rv; /* * NB: can only access uqn_id because userquota_update_cache() doesn't * pass in an entire userquota_node_t. */ rv = strcmp(luqn->uqn_id, ruqn->uqn_id); return (TREE_ISIGN(rv)); } static void do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) { void *cookie; userquota_node_t *uqn; ASSERT(dmu_tx_is_syncing(tx)); cookie = NULL; while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, &cookie)) != NULL) { /* * os_userused_lock protects against concurrent calls to * zap_increment_int(). It's needed because zap_increment_int() * is not thread-safe (i.e. not atomic). */ mutex_enter(&os->os_userused_lock); VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, uqn->uqn_id, uqn->uqn_delta, tx)); mutex_exit(&os->os_userused_lock); kmem_free(uqn, sizeof (*uqn)); } avl_destroy(&cache->uqc_user_deltas); cookie = NULL; while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, &cookie)) != NULL) { mutex_enter(&os->os_userused_lock); VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, uqn->uqn_id, uqn->uqn_delta, tx)); mutex_exit(&os->os_userused_lock); kmem_free(uqn, sizeof (*uqn)); } avl_destroy(&cache->uqc_group_deltas); if (dmu_objset_projectquota_enabled(os)) { cookie = NULL; while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas, &cookie)) != NULL) { mutex_enter(&os->os_userused_lock); VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT, uqn->uqn_id, uqn->uqn_delta, tx)); mutex_exit(&os->os_userused_lock); kmem_free(uqn, sizeof (*uqn)); } avl_destroy(&cache->uqc_project_deltas); } } static void userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) { userquota_node_t *uqn; avl_index_t idx; ASSERT(strlen(id) < sizeof (uqn->uqn_id)); /* * Use id directly for searching because uqn_id is the first field of * userquota_node_t and fields after uqn_id won't be accessed in * avl_find(). */ uqn = avl_find(avl, (const void *)id, &idx); if (uqn == NULL) { uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); avl_insert(avl, uqn, idx); } uqn->uqn_delta += delta; } static void do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used, uint64_t flags, uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) { if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) { int64_t delta = DNODE_MIN_SIZE + used; char name[20]; if (subtract) delta = -delta; (void) snprintf(name, sizeof (name), "%llx", (longlong_t)user); userquota_update_cache(&cache->uqc_user_deltas, name, delta); (void) snprintf(name, sizeof (name), "%llx", (longlong_t)group); userquota_update_cache(&cache->uqc_group_deltas, name, delta); if (dmu_objset_projectquota_enabled(os)) { (void) snprintf(name, sizeof (name), "%llx", (longlong_t)project); userquota_update_cache(&cache->uqc_project_deltas, name, delta); } } } static void do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags, uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) { if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { char name[20 + DMU_OBJACCT_PREFIX_LEN]; int delta = subtract ? -1 : 1; (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", (longlong_t)user); userquota_update_cache(&cache->uqc_user_deltas, name, delta); (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", (longlong_t)group); userquota_update_cache(&cache->uqc_group_deltas, name, delta); if (dmu_objset_projectquota_enabled(os)) { (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", (longlong_t)project); userquota_update_cache(&cache->uqc_project_deltas, name, delta); } } } typedef struct userquota_updates_arg { objset_t *uua_os; int uua_sublist_idx; dmu_tx_t *uua_tx; } userquota_updates_arg_t; static void userquota_updates_task(void *arg) { userquota_updates_arg_t *uua = arg; objset_t *os = uua->uua_os; dmu_tx_t *tx = uua->uua_tx; dnode_t *dn; userquota_cache_t cache = { { 0 } }; multilist_sublist_t *list = multilist_sublist_lock(&os->os_synced_dnodes, uua->uua_sublist_idx); ASSERT(multilist_sublist_head(list) == NULL || dmu_objset_userused_enabled(os)); avl_create(&cache.uqc_user_deltas, userquota_compare, sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); avl_create(&cache.uqc_group_deltas, userquota_compare, sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); if (dmu_objset_projectquota_enabled(os)) avl_create(&cache.uqc_project_deltas, userquota_compare, sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); while ((dn = multilist_sublist_head(list)) != NULL) { int flags; ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED); flags = dn->dn_id_flags; ASSERT(flags); if (flags & DN_ID_OLD_EXIST) { do_userquota_update(os, &cache, dn->dn_oldused, dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, dn->dn_oldprojid, B_TRUE); do_userobjquota_update(os, &cache, dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, dn->dn_oldprojid, B_TRUE); } if (flags & DN_ID_NEW_EXIST) { do_userquota_update(os, &cache, DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, dn->dn_newprojid, B_FALSE); do_userobjquota_update(os, &cache, dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, dn->dn_newprojid, B_FALSE); } mutex_enter(&dn->dn_mtx); dn->dn_oldused = 0; dn->dn_oldflags = 0; if (dn->dn_id_flags & DN_ID_NEW_EXIST) { dn->dn_olduid = dn->dn_newuid; dn->dn_oldgid = dn->dn_newgid; dn->dn_oldprojid = dn->dn_newprojid; dn->dn_id_flags |= DN_ID_OLD_EXIST; if (dn->dn_bonuslen == 0) dn->dn_id_flags |= DN_ID_CHKED_SPILL; else dn->dn_id_flags |= DN_ID_CHKED_BONUS; } dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); mutex_exit(&dn->dn_mtx); multilist_sublist_remove(list, dn); dnode_rele(dn, &os->os_synced_dnodes); } do_userquota_cacheflush(os, &cache, tx); multilist_sublist_unlock(list); kmem_free(uua, sizeof (*uua)); } /* * Release dnode holds from dmu_objset_sync_dnodes(). When the dnode is being * synced (i.e. we have issued the zio's for blocks in the dnode), it can't be * evicted because the block containing the dnode can't be evicted until it is * written out. However, this hold is necessary to prevent the dnode_t from * being moved (via dnode_move()) while it's still referenced by * dbuf_dirty_record_t:dr_dnode. And dr_dnode is needed for * dirty_lightweight_leaf-type dirty records. * * If we are doing user-object accounting, the dnode_rele() happens from * userquota_updates_task() instead. */ static void dnode_rele_task(void *arg) { userquota_updates_arg_t *uua = arg; objset_t *os = uua->uua_os; multilist_sublist_t *list = multilist_sublist_lock(&os->os_synced_dnodes, uua->uua_sublist_idx); dnode_t *dn; while ((dn = multilist_sublist_head(list)) != NULL) { multilist_sublist_remove(list, dn); dnode_rele(dn, &os->os_synced_dnodes); } multilist_sublist_unlock(list); kmem_free(uua, sizeof (*uua)); } /* * Return TRUE if userquota updates are needed. */ static boolean_t dmu_objset_do_userquota_updates_prep(objset_t *os, dmu_tx_t *tx) { if (!dmu_objset_userused_enabled(os)) return (B_FALSE); /* * If this is a raw receive just return and handle accounting * later when we have the keys loaded. We also don't do user * accounting during claiming since the datasets are not owned * for the duration of claiming and this txg should only be * used for recovery. */ if (os->os_encrypted && dmu_objset_is_receiving(os)) return (B_FALSE); if (tx->tx_txg <= os->os_spa->spa_claim_max_txg) return (B_FALSE); /* Allocate the user/group/project used objects if necessary. */ if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { VERIFY0(zap_create_claim(os, DMU_USERUSED_OBJECT, DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); VERIFY0(zap_create_claim(os, DMU_GROUPUSED_OBJECT, DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); } if (dmu_objset_projectquota_enabled(os) && DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) { VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT, DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); } return (B_TRUE); } /* * Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and * also release the holds on the dnodes from dmu_objset_sync_dnodes(). * The caller must taskq_wait(dp_sync_taskq). */ void dmu_objset_sync_done(objset_t *os, dmu_tx_t *tx) { boolean_t need_userquota = dmu_objset_do_userquota_updates_prep(os, tx); int num_sublists = multilist_get_num_sublists(&os->os_synced_dnodes); for (int i = 0; i < num_sublists; i++) { userquota_updates_arg_t *uua = kmem_alloc(sizeof (*uua), KM_SLEEP); uua->uua_os = os; uua->uua_sublist_idx = i; uua->uua_tx = tx; /* * If we don't need to update userquotas, use * dnode_rele_task() to call dnode_rele() */ (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, need_userquota ? userquota_updates_task : dnode_rele_task, uua, 0); /* callback frees uua */ } } /* * Returns a pointer to data to find uid/gid from * * If a dirty record for transaction group that is syncing can't * be found then NULL is returned. In the NULL case it is assumed * the uid/gid aren't changing. */ static void * dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) { dbuf_dirty_record_t *dr; void *data; if (db->db_dirtycnt == 0) return (db->db.db_data); /* Nothing is changing */ dr = dbuf_find_dirty_eq(db, tx->tx_txg); if (dr == NULL) { data = NULL; } else { if (dr->dr_dnode->dn_bonuslen == 0 && dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) data = dr->dt.dl.dr_data->b_data; else data = dr->dt.dl.dr_data; } return (data); } void dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) { objset_t *os = dn->dn_objset; void *data = NULL; dmu_buf_impl_t *db = NULL; int flags = dn->dn_id_flags; int error; boolean_t have_spill = B_FALSE; if (!dmu_objset_userused_enabled(dn->dn_objset)) return; /* * Raw receives introduce a problem with user accounting. Raw * receives cannot update the user accounting info because the * user ids and the sizes are encrypted. To guarantee that we * never end up with bad user accounting, we simply disable it * during raw receives. We also disable this for normal receives * so that an incremental raw receive may be done on top of an * existing non-raw receive. */ if (os->os_encrypted && dmu_objset_is_receiving(os)) return; if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| DN_ID_CHKED_SPILL))) return; if (before && dn->dn_bonuslen != 0) data = DN_BONUS(dn->dn_phys); else if (!before && dn->dn_bonuslen != 0) { if (dn->dn_bonus) { db = dn->dn_bonus; mutex_enter(&db->db_mtx); data = dmu_objset_userquota_find_data(db, tx); } else { data = DN_BONUS(dn->dn_phys); } } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { int rf = 0; if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) rf |= DB_RF_HAVESTRUCT; error = dmu_spill_hold_by_dnode(dn, rf | DB_RF_MUST_SUCCEED, FTAG, (dmu_buf_t **)&db); ASSERT(error == 0); mutex_enter(&db->db_mtx); data = (before) ? db->db.db_data : dmu_objset_userquota_find_data(db, tx); have_spill = B_TRUE; } else { mutex_enter(&dn->dn_mtx); dn->dn_id_flags |= DN_ID_CHKED_BONUS; mutex_exit(&dn->dn_mtx); return; } /* * Must always call the callback in case the object * type has changed and that type isn't an object type to track */ zfs_file_info_t zfi; error = file_cbs[os->os_phys->os_type](dn->dn_bonustype, data, &zfi); if (before) { ASSERT(data); dn->dn_olduid = zfi.zfi_user; dn->dn_oldgid = zfi.zfi_group; dn->dn_oldprojid = zfi.zfi_project; } else if (data) { dn->dn_newuid = zfi.zfi_user; dn->dn_newgid = zfi.zfi_group; dn->dn_newprojid = zfi.zfi_project; } /* * Preserve existing uid/gid when the callback can't determine * what the new uid/gid are and the callback returned EEXIST. * The EEXIST error tells us to just use the existing uid/gid. * If we don't know what the old values are then just assign * them to 0, since that is a new file being created. */ if (!before && data == NULL && error == EEXIST) { if (flags & DN_ID_OLD_EXIST) { dn->dn_newuid = dn->dn_olduid; dn->dn_newgid = dn->dn_oldgid; dn->dn_newprojid = dn->dn_oldprojid; } else { dn->dn_newuid = 0; dn->dn_newgid = 0; dn->dn_newprojid = ZFS_DEFAULT_PROJID; } error = 0; } if (db) mutex_exit(&db->db_mtx); mutex_enter(&dn->dn_mtx); if (error == 0 && before) dn->dn_id_flags |= DN_ID_OLD_EXIST; if (error == 0 && !before) dn->dn_id_flags |= DN_ID_NEW_EXIST; if (have_spill) { dn->dn_id_flags |= DN_ID_CHKED_SPILL; } else { dn->dn_id_flags |= DN_ID_CHKED_BONUS; } mutex_exit(&dn->dn_mtx); if (have_spill) dmu_buf_rele((dmu_buf_t *)db, FTAG); } boolean_t dmu_objset_userspace_present(objset_t *os) { return (os->os_phys->os_flags & OBJSET_FLAG_USERACCOUNTING_COMPLETE); } boolean_t dmu_objset_userobjspace_present(objset_t *os) { return (os->os_phys->os_flags & OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); } boolean_t dmu_objset_projectquota_present(objset_t *os) { return (os->os_phys->os_flags & OBJSET_FLAG_PROJECTQUOTA_COMPLETE); } static int dmu_objset_space_upgrade(objset_t *os) { uint64_t obj; int err = 0; /* * We simply need to mark every object dirty, so that it will be * synced out and now accounted. If this is called * concurrently, or if we already did some work before crashing, * that's fine, since we track each object's accounted state * independently. */ for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { dmu_tx_t *tx; dmu_buf_t *db; int objerr; mutex_enter(&os->os_upgrade_lock); if (os->os_upgrade_exit) err = SET_ERROR(EINTR); mutex_exit(&os->os_upgrade_lock); if (err != 0) return (err); if (issig(JUSTLOOKING) && issig(FORREAL)) return (SET_ERROR(EINTR)); objerr = dmu_bonus_hold(os, obj, FTAG, &db); if (objerr != 0) continue; tx = dmu_tx_create(os); dmu_tx_hold_bonus(tx, obj); objerr = dmu_tx_assign(tx, TXG_WAIT); if (objerr != 0) { dmu_buf_rele(db, FTAG); dmu_tx_abort(tx); continue; } dmu_buf_will_dirty(db, tx); dmu_buf_rele(db, FTAG); dmu_tx_commit(tx); } return (0); } static int dmu_objset_userspace_upgrade_cb(objset_t *os) { int err = 0; if (dmu_objset_userspace_present(os)) return (0); if (dmu_objset_is_snapshot(os)) return (SET_ERROR(EINVAL)); if (!dmu_objset_userused_enabled(os)) return (SET_ERROR(ENOTSUP)); err = dmu_objset_space_upgrade(os); if (err) return (err); os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; txg_wait_synced(dmu_objset_pool(os), 0); return (0); } void dmu_objset_userspace_upgrade(objset_t *os) { dmu_objset_upgrade(os, dmu_objset_userspace_upgrade_cb); } static int dmu_objset_id_quota_upgrade_cb(objset_t *os) { int err = 0; if (dmu_objset_userobjspace_present(os) && dmu_objset_projectquota_present(os)) return (0); if (dmu_objset_is_snapshot(os)) return (SET_ERROR(EINVAL)); if (!dmu_objset_userused_enabled(os)) return (SET_ERROR(ENOTSUP)); if (!dmu_objset_projectquota_enabled(os) && dmu_objset_userobjspace_present(os)) return (SET_ERROR(ENOTSUP)); + if (dmu_objset_userobjused_enabled(os)) + dmu_objset_ds(os)->ds_feature_activation[ + SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; + if (dmu_objset_projectquota_enabled(os)) + dmu_objset_ds(os)->ds_feature_activation[ + SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; + err = dmu_objset_space_upgrade(os); if (err) return (err); os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; if (dmu_objset_userobjused_enabled(os)) os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; if (dmu_objset_projectquota_enabled(os)) os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; txg_wait_synced(dmu_objset_pool(os), 0); return (0); } void dmu_objset_id_quota_upgrade(objset_t *os) { dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb); } boolean_t dmu_objset_userobjspace_upgradable(objset_t *os) { return (dmu_objset_type(os) == DMU_OST_ZFS && !dmu_objset_is_snapshot(os) && dmu_objset_userobjused_enabled(os) && !dmu_objset_userobjspace_present(os) && spa_writeable(dmu_objset_spa(os))); } boolean_t dmu_objset_projectquota_upgradable(objset_t *os) { return (dmu_objset_type(os) == DMU_OST_ZFS && !dmu_objset_is_snapshot(os) && dmu_objset_projectquota_enabled(os) && !dmu_objset_projectquota_present(os) && spa_writeable(dmu_objset_spa(os))); } void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, uint64_t *usedobjsp, uint64_t *availobjsp) { dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, usedobjsp, availobjsp); } uint64_t dmu_objset_fsid_guid(objset_t *os) { return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); } void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) { stat->dds_type = os->os_phys->os_type; if (os->os_dsl_dataset) dsl_dataset_fast_stat(os->os_dsl_dataset, stat); } void dmu_objset_stats(objset_t *os, nvlist_t *nv) { ASSERT(os->os_dsl_dataset || os->os_phys->os_type == DMU_OST_META); if (os->os_dsl_dataset != NULL) dsl_dataset_stats(os->os_dsl_dataset, nv); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, os->os_phys->os_type); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, dmu_objset_userspace_present(os)); } int dmu_objset_is_snapshot(objset_t *os) { if (os->os_dsl_dataset != NULL) return (os->os_dsl_dataset->ds_is_snapshot); else return (B_FALSE); } int dmu_snapshot_realname(objset_t *os, const char *name, char *real, int maxlen, boolean_t *conflict) { dsl_dataset_t *ds = os->os_dsl_dataset; uint64_t ignored; if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) return (SET_ERROR(ENOENT)); return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_NORMALIZE, real, maxlen, conflict)); } int dmu_snapshot_list_next(objset_t *os, int namelen, char *name, uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) { dsl_dataset_t *ds = os->os_dsl_dataset; zap_cursor_t cursor; zap_attribute_t attr; ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) return (SET_ERROR(ENOENT)); zap_cursor_init_serialized(&cursor, ds->ds_dir->dd_pool->dp_meta_objset, dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); if (zap_cursor_retrieve(&cursor, &attr) != 0) { zap_cursor_fini(&cursor); return (SET_ERROR(ENOENT)); } if (strlen(attr.za_name) + 1 > namelen) { zap_cursor_fini(&cursor); return (SET_ERROR(ENAMETOOLONG)); } (void) strlcpy(name, attr.za_name, namelen); if (idp) *idp = attr.za_first_integer; if (case_conflict) *case_conflict = attr.za_normalization_conflict; zap_cursor_advance(&cursor); *offp = zap_cursor_serialize(&cursor); zap_cursor_fini(&cursor); return (0); } int dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) { return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); } int dmu_dir_list_next(objset_t *os, int namelen, char *name, uint64_t *idp, uint64_t *offp) { dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; zap_cursor_t cursor; zap_attribute_t attr; /* there is no next dir on a snapshot! */ if (os->os_dsl_dataset->ds_object != dsl_dir_phys(dd)->dd_head_dataset_obj) return (SET_ERROR(ENOENT)); zap_cursor_init_serialized(&cursor, dd->dd_pool->dp_meta_objset, dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); if (zap_cursor_retrieve(&cursor, &attr) != 0) { zap_cursor_fini(&cursor); return (SET_ERROR(ENOENT)); } if (strlen(attr.za_name) + 1 > namelen) { zap_cursor_fini(&cursor); return (SET_ERROR(ENAMETOOLONG)); } (void) strlcpy(name, attr.za_name, namelen); if (idp) *idp = attr.za_first_integer; zap_cursor_advance(&cursor); *offp = zap_cursor_serialize(&cursor); zap_cursor_fini(&cursor); return (0); } typedef struct dmu_objset_find_ctx { taskq_t *dc_tq; dsl_pool_t *dc_dp; uint64_t dc_ddobj; char *dc_ddname; /* last component of ddobj's name */ int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); void *dc_arg; int dc_flags; kmutex_t *dc_error_lock; int *dc_error; } dmu_objset_find_ctx_t; static void dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) { dsl_pool_t *dp = dcp->dc_dp; dsl_dir_t *dd; dsl_dataset_t *ds; zap_cursor_t zc; zap_attribute_t *attr; uint64_t thisobj; int err = 0; /* don't process if there already was an error */ if (*dcp->dc_error != 0) goto out; /* * Note: passing the name (dc_ddname) here is optional, but it * improves performance because we don't need to call * zap_value_search() to determine the name. */ err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); if (err != 0) goto out; /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ if (dd->dd_myname[0] == '$') { dsl_dir_rele(dd, FTAG); goto out; } thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); /* * Iterate over all children. */ if (dcp->dc_flags & DS_FIND_CHILDREN) { for (zap_cursor_init(&zc, dp->dp_meta_objset, dsl_dir_phys(dd)->dd_child_dir_zapobj); zap_cursor_retrieve(&zc, attr) == 0; (void) zap_cursor_advance(&zc)) { ASSERT3U(attr->za_integer_length, ==, sizeof (uint64_t)); ASSERT3U(attr->za_num_integers, ==, 1); dmu_objset_find_ctx_t *child_dcp = kmem_alloc(sizeof (*child_dcp), KM_SLEEP); *child_dcp = *dcp; child_dcp->dc_ddobj = attr->za_first_integer; child_dcp->dc_ddname = spa_strdup(attr->za_name); if (dcp->dc_tq != NULL) (void) taskq_dispatch(dcp->dc_tq, dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); else dmu_objset_find_dp_impl(child_dcp); } zap_cursor_fini(&zc); } /* * Iterate over all snapshots. */ if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { dsl_dataset_t *ds; err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); if (err == 0) { uint64_t snapobj; snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; dsl_dataset_rele(ds, FTAG); for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); zap_cursor_retrieve(&zc, attr) == 0; (void) zap_cursor_advance(&zc)) { ASSERT3U(attr->za_integer_length, ==, sizeof (uint64_t)); ASSERT3U(attr->za_num_integers, ==, 1); err = dsl_dataset_hold_obj(dp, attr->za_first_integer, FTAG, &ds); if (err != 0) break; err = dcp->dc_func(dp, ds, dcp->dc_arg); dsl_dataset_rele(ds, FTAG); if (err != 0) break; } zap_cursor_fini(&zc); } } kmem_free(attr, sizeof (zap_attribute_t)); if (err != 0) { dsl_dir_rele(dd, FTAG); goto out; } /* * Apply to self. */ err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); /* * Note: we hold the dir while calling dsl_dataset_hold_obj() so * that the dir will remain cached, and we won't have to re-instantiate * it (which could be expensive due to finding its name via * zap_value_search()). */ dsl_dir_rele(dd, FTAG); if (err != 0) goto out; err = dcp->dc_func(dp, ds, dcp->dc_arg); dsl_dataset_rele(ds, FTAG); out: if (err != 0) { mutex_enter(dcp->dc_error_lock); /* only keep first error */ if (*dcp->dc_error == 0) *dcp->dc_error = err; mutex_exit(dcp->dc_error_lock); } if (dcp->dc_ddname != NULL) spa_strfree(dcp->dc_ddname); kmem_free(dcp, sizeof (*dcp)); } static void dmu_objset_find_dp_cb(void *arg) { dmu_objset_find_ctx_t *dcp = arg; dsl_pool_t *dp = dcp->dc_dp; /* * We need to get a pool_config_lock here, as there are several * assert(pool_config_held) down the stack. Getting a lock via * dsl_pool_config_enter is risky, as it might be stalled by a * pending writer. This would deadlock, as the write lock can * only be granted when our parent thread gives up the lock. * The _prio interface gives us priority over a pending writer. */ dsl_pool_config_enter_prio(dp, FTAG); dmu_objset_find_dp_impl(dcp); dsl_pool_config_exit(dp, FTAG); } /* * Find objsets under and including ddobj, call func(ds) on each. * The order for the enumeration is completely undefined. * func is called with dsl_pool_config held. */ int dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) { int error = 0; taskq_t *tq = NULL; int ntasks; dmu_objset_find_ctx_t *dcp; kmutex_t err_lock; mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); dcp->dc_tq = NULL; dcp->dc_dp = dp; dcp->dc_ddobj = ddobj; dcp->dc_ddname = NULL; dcp->dc_func = func; dcp->dc_arg = arg; dcp->dc_flags = flags; dcp->dc_error_lock = &err_lock; dcp->dc_error = &error; if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { /* * In case a write lock is held we can't make use of * parallelism, as down the stack of the worker threads * the lock is asserted via dsl_pool_config_held. * In case of a read lock this is solved by getting a read * lock in each worker thread, which isn't possible in case * of a writer lock. So we fall back to the synchronous path * here. * In the future it might be possible to get some magic into * dsl_pool_config_held in a way that it returns true for * the worker threads so that a single lock held from this * thread suffices. For now, stay single threaded. */ dmu_objset_find_dp_impl(dcp); mutex_destroy(&err_lock); return (error); } ntasks = dmu_find_threads; if (ntasks == 0) ntasks = vdev_count_leaves(dp->dp_spa) * 4; tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, INT_MAX, 0); if (tq == NULL) { kmem_free(dcp, sizeof (*dcp)); mutex_destroy(&err_lock); return (SET_ERROR(ENOMEM)); } dcp->dc_tq = tq; /* dcp will be freed by task */ (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); /* * PORTING: this code relies on the property of taskq_wait to wait * until no more tasks are queued and no more tasks are active. As * we always queue new tasks from within other tasks, task_wait * reliably waits for the full recursion to finish, even though we * enqueue new tasks after taskq_wait has been called. * On platforms other than illumos, taskq_wait may not have this * property. */ taskq_wait(tq); taskq_destroy(tq); mutex_destroy(&err_lock); return (error); } /* * Find all objsets under name, and for each, call 'func(child_name, arg)'. * The dp_config_rwlock must not be held when this is called, and it * will not be held when the callback is called. * Therefore this function should only be used when the pool is not changing * (e.g. in syncing context), or the callback can deal with the possible races. */ static int dmu_objset_find_impl(spa_t *spa, const char *name, int func(const char *, void *), void *arg, int flags) { dsl_dir_t *dd; dsl_pool_t *dp = spa_get_dsl(spa); dsl_dataset_t *ds; zap_cursor_t zc; zap_attribute_t *attr; char *child; uint64_t thisobj; int err; dsl_pool_config_enter(dp, FTAG); err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); if (err != 0) { dsl_pool_config_exit(dp, FTAG); return (err); } /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ if (dd->dd_myname[0] == '$') { dsl_dir_rele(dd, FTAG); dsl_pool_config_exit(dp, FTAG); return (0); } thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); /* * Iterate over all children. */ if (flags & DS_FIND_CHILDREN) { for (zap_cursor_init(&zc, dp->dp_meta_objset, dsl_dir_phys(dd)->dd_child_dir_zapobj); zap_cursor_retrieve(&zc, attr) == 0; (void) zap_cursor_advance(&zc)) { ASSERT3U(attr->za_integer_length, ==, sizeof (uint64_t)); ASSERT3U(attr->za_num_integers, ==, 1); child = kmem_asprintf("%s/%s", name, attr->za_name); dsl_pool_config_exit(dp, FTAG); err = dmu_objset_find_impl(spa, child, func, arg, flags); dsl_pool_config_enter(dp, FTAG); kmem_strfree(child); if (err != 0) break; } zap_cursor_fini(&zc); if (err != 0) { dsl_dir_rele(dd, FTAG); dsl_pool_config_exit(dp, FTAG); kmem_free(attr, sizeof (zap_attribute_t)); return (err); } } /* * Iterate over all snapshots. */ if (flags & DS_FIND_SNAPSHOTS) { err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); if (err == 0) { uint64_t snapobj; snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; dsl_dataset_rele(ds, FTAG); for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); zap_cursor_retrieve(&zc, attr) == 0; (void) zap_cursor_advance(&zc)) { ASSERT3U(attr->za_integer_length, ==, sizeof (uint64_t)); ASSERT3U(attr->za_num_integers, ==, 1); child = kmem_asprintf("%s@%s", name, attr->za_name); dsl_pool_config_exit(dp, FTAG); err = func(child, arg); dsl_pool_config_enter(dp, FTAG); kmem_strfree(child); if (err != 0) break; } zap_cursor_fini(&zc); } } dsl_dir_rele(dd, FTAG); kmem_free(attr, sizeof (zap_attribute_t)); dsl_pool_config_exit(dp, FTAG); if (err != 0) return (err); /* Apply to self. */ return (func(name, arg)); } /* * See comment above dmu_objset_find_impl(). */ int dmu_objset_find(const char *name, int func(const char *, void *), void *arg, int flags) { spa_t *spa; int error; error = spa_open(name, &spa, FTAG); if (error != 0) return (error); error = dmu_objset_find_impl(spa, name, func, arg, flags); spa_close(spa, FTAG); return (error); } boolean_t dmu_objset_incompatible_encryption_version(objset_t *os) { return (dsl_dir_incompatible_encryption_version( os->os_dsl_dataset->ds_dir)); } void dmu_objset_set_user(objset_t *os, void *user_ptr) { ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); os->os_user_ptr = user_ptr; } void * dmu_objset_get_user(objset_t *os) { ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); return (os->os_user_ptr); } /* * Determine name of filesystem, given name of snapshot. * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */ int dmu_fsname(const char *snapname, char *buf) { char *atp = strchr(snapname, '@'); if (atp == NULL) return (SET_ERROR(EINVAL)); if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) return (SET_ERROR(ENAMETOOLONG)); (void) strlcpy(buf, snapname, atp - snapname + 1); return (0); } /* * Call when we think we're going to write/free space in open context * to track the amount of dirty data in the open txg, which is also the * amount of memory that can not be evicted until this txg syncs. * * Note that there are two conditions where this can be called from * syncing context: * * [1] When we just created the dataset, in which case we go on with * updating any accounting of dirty data as usual. * [2] When we are dirtying MOS data, in which case we only update the * pool's accounting of dirty data. */ void dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) { dsl_dataset_t *ds = os->os_dsl_dataset; int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); if (ds != NULL) { dsl_dir_willuse_space(ds->ds_dir, aspace, tx); } dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); } #if defined(_KERNEL) EXPORT_SYMBOL(dmu_objset_zil); EXPORT_SYMBOL(dmu_objset_pool); EXPORT_SYMBOL(dmu_objset_ds); EXPORT_SYMBOL(dmu_objset_type); EXPORT_SYMBOL(dmu_objset_name); EXPORT_SYMBOL(dmu_objset_hold); EXPORT_SYMBOL(dmu_objset_hold_flags); EXPORT_SYMBOL(dmu_objset_own); EXPORT_SYMBOL(dmu_objset_rele); EXPORT_SYMBOL(dmu_objset_rele_flags); EXPORT_SYMBOL(dmu_objset_disown); EXPORT_SYMBOL(dmu_objset_from_ds); EXPORT_SYMBOL(dmu_objset_create); EXPORT_SYMBOL(dmu_objset_clone); EXPORT_SYMBOL(dmu_objset_stats); EXPORT_SYMBOL(dmu_objset_fast_stat); EXPORT_SYMBOL(dmu_objset_spa); EXPORT_SYMBOL(dmu_objset_space); EXPORT_SYMBOL(dmu_objset_fsid_guid); EXPORT_SYMBOL(dmu_objset_find); EXPORT_SYMBOL(dmu_objset_byteswap); EXPORT_SYMBOL(dmu_objset_evict_dbufs); EXPORT_SYMBOL(dmu_objset_snap_cmtime); EXPORT_SYMBOL(dmu_objset_dnodesize); EXPORT_SYMBOL(dmu_objset_sync); EXPORT_SYMBOL(dmu_objset_is_dirty); EXPORT_SYMBOL(dmu_objset_create_impl_dnstats); EXPORT_SYMBOL(dmu_objset_create_impl); EXPORT_SYMBOL(dmu_objset_open_impl); EXPORT_SYMBOL(dmu_objset_evict); EXPORT_SYMBOL(dmu_objset_register_type); EXPORT_SYMBOL(dmu_objset_sync_done); EXPORT_SYMBOL(dmu_objset_userquota_get_ids); EXPORT_SYMBOL(dmu_objset_userused_enabled); EXPORT_SYMBOL(dmu_objset_userspace_upgrade); EXPORT_SYMBOL(dmu_objset_userspace_present); EXPORT_SYMBOL(dmu_objset_userobjused_enabled); EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); EXPORT_SYMBOL(dmu_objset_userobjspace_present); EXPORT_SYMBOL(dmu_objset_projectquota_enabled); EXPORT_SYMBOL(dmu_objset_projectquota_present); EXPORT_SYMBOL(dmu_objset_projectquota_upgradable); EXPORT_SYMBOL(dmu_objset_id_quota_upgrade); #endif diff --git a/module/zfs/dsl_dataset.c b/module/zfs/dsl_dataset.c index 08de2546a55c..a844059de495 100644 --- a/module/zfs/dsl_dataset.c +++ b/module/zfs/dsl_dataset.c @@ -1,5025 +1,5006 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2020 by Delphix. All rights reserved. * Copyright (c) 2014, Joyent, Inc. All rights reserved. * Copyright (c) 2014 RackTop Systems. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2016 Actifio, Inc. All rights reserved. * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved. * Copyright 2017 Nexenta Systems, Inc. * Copyright (c) 2019, Klara Inc. * Copyright (c) 2019, Allan Jude * Copyright (c) 2020 The FreeBSD Foundation [1] * * [1] Portions of this software were developed by Allan Jude * under sponsorship from the FreeBSD Foundation. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * The SPA supports block sizes up to 16MB. However, very large blocks * can have an impact on i/o latency (e.g. tying up a spinning disk for * ~300ms), and also potentially on the memory allocator. Therefore, * we do not allow the recordsize to be set larger than zfs_max_recordsize * (default 1MB). Larger blocks can be created by changing this tunable, * and pools with larger blocks can always be imported and used, regardless * of this setting. */ int zfs_max_recordsize = 1 * 1024 * 1024; int zfs_allow_redacted_dataset_mount = 0; #define SWITCH64(x, y) \ { \ uint64_t __tmp = (x); \ (x) = (y); \ (y) = __tmp; \ } #define DS_REF_MAX (1ULL << 62) static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx); static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx); static void unload_zfeature(dsl_dataset_t *ds, spa_feature_t f); extern int spa_asize_inflation; static zil_header_t zero_zil; /* * Figure out how much of this delta should be propagated to the dsl_dir * layer. If there's a refreservation, that space has already been * partially accounted for in our ancestors. */ static int64_t parent_delta(dsl_dataset_t *ds, int64_t delta) { dsl_dataset_phys_t *ds_phys; uint64_t old_bytes, new_bytes; if (ds->ds_reserved == 0) return (delta); ds_phys = dsl_dataset_phys(ds); old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); return (new_bytes - old_bytes); } void dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) { spa_t *spa = dmu_tx_pool(tx)->dp_spa; int used = bp_get_dsize_sync(spa, bp); int compressed = BP_GET_PSIZE(bp); int uncompressed = BP_GET_UCSIZE(bp); int64_t delta; spa_feature_t f; dprintf_bp(bp, "ds=%p", ds); ASSERT(dmu_tx_is_syncing(tx)); /* It could have been compressed away to nothing */ if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) return; ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); if (ds == NULL) { dsl_pool_mos_diduse_space(tx->tx_pool, used, compressed, uncompressed); return; } ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); dmu_buf_will_dirty(ds->ds_dbuf, tx); mutex_enter(&ds->ds_lock); delta = parent_delta(ds, used); dsl_dataset_phys(ds)->ds_referenced_bytes += used; dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; dsl_dataset_phys(ds)->ds_unique_bytes += used; if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) { ds->ds_feature_activation[SPA_FEATURE_LARGE_BLOCKS] = (void *)B_TRUE; } f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); if (f != SPA_FEATURE_NONE) { ASSERT3S(spa_feature_table[f].fi_type, ==, ZFEATURE_TYPE_BOOLEAN); ds->ds_feature_activation[f] = (void *)B_TRUE; } f = zio_compress_to_feature(BP_GET_COMPRESS(bp)); if (f != SPA_FEATURE_NONE) { ASSERT3S(spa_feature_table[f].fi_type, ==, ZFEATURE_TYPE_BOOLEAN); ds->ds_feature_activation[f] = (void *)B_TRUE; } /* * Track block for livelist, but ignore embedded blocks because * they do not need to be freed. */ if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && bp->blk_birth > ds->ds_dir->dd_origin_txg && !(BP_IS_EMBEDDED(bp))) { ASSERT(dsl_dir_is_clone(ds->ds_dir)); ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_LIVELIST)); bplist_append(&ds->ds_dir->dd_pending_allocs, bp); } mutex_exit(&ds->ds_lock); dsl_dir_diduse_transfer_space(ds->ds_dir, delta, compressed, uncompressed, used, DD_USED_REFRSRV, DD_USED_HEAD, tx); } /* * Called when the specified segment has been remapped, and is thus no * longer referenced in the head dataset. The vdev must be indirect. * * If the segment is referenced by a snapshot, put it on the remap deadlist. * Otherwise, add this segment to the obsolete spacemap. */ void dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset, uint64_t size, uint64_t birth, dmu_tx_t *tx) { spa_t *spa = ds->ds_dir->dd_pool->dp_spa; ASSERT(dmu_tx_is_syncing(tx)); ASSERT(birth <= tx->tx_txg); ASSERT(!ds->ds_is_snapshot); if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx); } else { blkptr_t fakebp; dva_t *dva = &fakebp.blk_dva[0]; ASSERT(ds != NULL); mutex_enter(&ds->ds_remap_deadlist_lock); if (!dsl_dataset_remap_deadlist_exists(ds)) { dsl_dataset_create_remap_deadlist(ds, tx); } mutex_exit(&ds->ds_remap_deadlist_lock); BP_ZERO(&fakebp); fakebp.blk_birth = birth; DVA_SET_VDEV(dva, vdev); DVA_SET_OFFSET(dva, offset); DVA_SET_ASIZE(dva, size); dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, B_FALSE, tx); } } int dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, boolean_t async) { spa_t *spa = dmu_tx_pool(tx)->dp_spa; int used = bp_get_dsize_sync(spa, bp); int compressed = BP_GET_PSIZE(bp); int uncompressed = BP_GET_UCSIZE(bp); if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) return (0); ASSERT(dmu_tx_is_syncing(tx)); ASSERT(bp->blk_birth <= tx->tx_txg); if (ds == NULL) { dsl_free(tx->tx_pool, tx->tx_txg, bp); dsl_pool_mos_diduse_space(tx->tx_pool, -used, -compressed, -uncompressed); return (used); } ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); ASSERT(!ds->ds_is_snapshot); dmu_buf_will_dirty(ds->ds_dbuf, tx); /* * Track block for livelist, but ignore embedded blocks because * they do not need to be freed. */ if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && bp->blk_birth > ds->ds_dir->dd_origin_txg && !(BP_IS_EMBEDDED(bp))) { ASSERT(dsl_dir_is_clone(ds->ds_dir)); ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_LIVELIST)); bplist_append(&ds->ds_dir->dd_pending_frees, bp); } if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { int64_t delta; dprintf_bp(bp, "freeing ds=%llu", (u_longlong_t)ds->ds_object); dsl_free(tx->tx_pool, tx->tx_txg, bp); mutex_enter(&ds->ds_lock); ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || !DS_UNIQUE_IS_ACCURATE(ds)); delta = parent_delta(ds, -used); dsl_dataset_phys(ds)->ds_unique_bytes -= used; mutex_exit(&ds->ds_lock); dsl_dir_diduse_transfer_space(ds->ds_dir, delta, -compressed, -uncompressed, -used, DD_USED_REFRSRV, DD_USED_HEAD, tx); } else { dprintf_bp(bp, "putting on dead list: %s", ""); if (async) { /* * We are here as part of zio's write done callback, * which means we're a zio interrupt thread. We can't * call dsl_deadlist_insert() now because it may block * waiting for I/O. Instead, put bp on the deferred * queue and let dsl_pool_sync() finish the job. */ bplist_append(&ds->ds_pending_deadlist, bp); } else { dsl_deadlist_insert(&ds->ds_deadlist, bp, B_FALSE, tx); } ASSERT3U(ds->ds_prev->ds_object, ==, dsl_dataset_phys(ds)->ds_prev_snap_obj); ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object && bp->blk_birth > dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); mutex_enter(&ds->ds_prev->ds_lock); dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; mutex_exit(&ds->ds_prev->ds_lock); } if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { dsl_dir_transfer_space(ds->ds_dir, used, DD_USED_HEAD, DD_USED_SNAP, tx); } } dsl_bookmark_block_killed(ds, bp, tx); mutex_enter(&ds->ds_lock); ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); dsl_dataset_phys(ds)->ds_referenced_bytes -= used; ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; mutex_exit(&ds->ds_lock); return (used); } struct feature_type_uint64_array_arg { uint64_t length; uint64_t *array; }; static void unload_zfeature(dsl_dataset_t *ds, spa_feature_t f) { switch (spa_feature_table[f].fi_type) { case ZFEATURE_TYPE_BOOLEAN: break; case ZFEATURE_TYPE_UINT64_ARRAY: { struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f]; kmem_free(ftuaa->array, ftuaa->length * sizeof (uint64_t)); kmem_free(ftuaa, sizeof (*ftuaa)); break; } default: panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); } } static int load_zfeature(objset_t *mos, dsl_dataset_t *ds, spa_feature_t f) { int err = 0; switch (spa_feature_table[f].fi_type) { case ZFEATURE_TYPE_BOOLEAN: err = zap_contains(mos, ds->ds_object, spa_feature_table[f].fi_guid); if (err == 0) { ds->ds_feature[f] = (void *)B_TRUE; } else { ASSERT3U(err, ==, ENOENT); err = 0; } break; case ZFEATURE_TYPE_UINT64_ARRAY: { uint64_t int_size, num_int; uint64_t *data; err = zap_length(mos, ds->ds_object, spa_feature_table[f].fi_guid, &int_size, &num_int); if (err != 0) { ASSERT3U(err, ==, ENOENT); err = 0; break; } ASSERT3U(int_size, ==, sizeof (uint64_t)); data = kmem_alloc(int_size * num_int, KM_SLEEP); VERIFY0(zap_lookup(mos, ds->ds_object, spa_feature_table[f].fi_guid, int_size, num_int, data)); struct feature_type_uint64_array_arg *ftuaa = kmem_alloc(sizeof (*ftuaa), KM_SLEEP); ftuaa->length = num_int; ftuaa->array = data; ds->ds_feature[f] = ftuaa; break; } default: panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); } return (err); } /* * We have to release the fsid synchronously or we risk that a subsequent * mount of the same dataset will fail to unique_insert the fsid. This * failure would manifest itself as the fsid of this dataset changing * between mounts which makes NFS clients quite unhappy. */ static void dsl_dataset_evict_sync(void *dbu) { dsl_dataset_t *ds = dbu; ASSERT(ds->ds_owner == NULL); unique_remove(ds->ds_fsid_guid); } static void dsl_dataset_evict_async(void *dbu) { dsl_dataset_t *ds = dbu; ASSERT(ds->ds_owner == NULL); ds->ds_dbuf = NULL; if (ds->ds_objset != NULL) dmu_objset_evict(ds->ds_objset); if (ds->ds_prev) { dsl_dataset_rele(ds->ds_prev, ds); ds->ds_prev = NULL; } dsl_bookmark_fini_ds(ds); bplist_destroy(&ds->ds_pending_deadlist); if (dsl_deadlist_is_open(&ds->ds_deadlist)) dsl_deadlist_close(&ds->ds_deadlist); if (dsl_deadlist_is_open(&ds->ds_remap_deadlist)) dsl_deadlist_close(&ds->ds_remap_deadlist); if (ds->ds_dir) dsl_dir_async_rele(ds->ds_dir, ds); ASSERT(!list_link_active(&ds->ds_synced_link)); for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (dsl_dataset_feature_is_active(ds, f)) unload_zfeature(ds, f); } list_destroy(&ds->ds_prop_cbs); mutex_destroy(&ds->ds_lock); mutex_destroy(&ds->ds_opening_lock); mutex_destroy(&ds->ds_sendstream_lock); mutex_destroy(&ds->ds_remap_deadlist_lock); zfs_refcount_destroy(&ds->ds_longholds); rrw_destroy(&ds->ds_bp_rwlock); kmem_free(ds, sizeof (dsl_dataset_t)); } int dsl_dataset_get_snapname(dsl_dataset_t *ds) { dsl_dataset_phys_t *headphys; int err; dmu_buf_t *headdbuf; dsl_pool_t *dp = ds->ds_dir->dd_pool; objset_t *mos = dp->dp_meta_objset; if (ds->ds_snapname[0]) return (0); if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) return (0); err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &headdbuf); if (err != 0) return (err); headphys = headdbuf->db_data; err = zap_value_search(dp->dp_meta_objset, headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); if (err != 0 && zfs_recover == B_TRUE) { err = 0; (void) snprintf(ds->ds_snapname, sizeof (ds->ds_snapname), "SNAPOBJ=%llu-ERR=%d", (unsigned long long)ds->ds_object, err); } dmu_buf_rele(headdbuf, FTAG); return (err); } int dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) { objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; matchtype_t mt = 0; int err; if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) mt = MT_NORMALIZE; err = zap_lookup_norm(mos, snapobj, name, 8, 1, value, mt, NULL, 0, NULL); if (err == ENOTSUP && (mt & MT_NORMALIZE)) err = zap_lookup(mos, snapobj, name, 8, 1, value); return (err); } int dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, boolean_t adj_cnt) { objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; matchtype_t mt = 0; int err; dsl_dir_snap_cmtime_update(ds->ds_dir); if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) mt = MT_NORMALIZE; err = zap_remove_norm(mos, snapobj, name, mt, tx); if (err == ENOTSUP && (mt & MT_NORMALIZE)) err = zap_remove(mos, snapobj, name, tx); if (err == 0 && adj_cnt) dsl_fs_ss_count_adjust(ds->ds_dir, -1, DD_FIELD_SNAPSHOT_COUNT, tx); return (err); } boolean_t dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) { dmu_buf_t *dbuf = ds->ds_dbuf; boolean_t result = B_FALSE; if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, ds->ds_object, DMU_BONUS_BLKID, tag)) { if (ds == dmu_buf_get_user(dbuf)) result = B_TRUE; else dmu_buf_rele(dbuf, tag); } return (result); } int dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, dsl_dataset_t **dsp) { objset_t *mos = dp->dp_meta_objset; dmu_buf_t *dbuf; dsl_dataset_t *ds; int err; dmu_object_info_t doi; ASSERT(dsl_pool_config_held(dp)); err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); if (err != 0) return (err); /* Make sure dsobj has the correct object type. */ dmu_object_info_from_db(dbuf, &doi); if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { dmu_buf_rele(dbuf, tag); return (SET_ERROR(EINVAL)); } ds = dmu_buf_get_user(dbuf); if (ds == NULL) { dsl_dataset_t *winner = NULL; ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); ds->ds_dbuf = dbuf; ds->ds_object = dsobj; ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; list_link_init(&ds->ds_synced_link); err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir); if (err != 0) { kmem_free(ds, sizeof (dsl_dataset_t)); dmu_buf_rele(dbuf, tag); return (err); } mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&ds->ds_remap_deadlist_lock, NULL, MUTEX_DEFAULT, NULL); rrw_init(&ds->ds_bp_rwlock, B_FALSE); zfs_refcount_create(&ds->ds_longholds); bplist_create(&ds->ds_pending_deadlist); list_create(&ds->ds_sendstreams, sizeof (dmu_sendstatus_t), offsetof(dmu_sendstatus_t, dss_link)); list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t), offsetof(dsl_prop_cb_record_t, cbr_ds_node)); if (doi.doi_type == DMU_OTN_ZAP_METADATA) { spa_feature_t f; for (f = 0; f < SPA_FEATURES; f++) { if (!(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET)) continue; err = load_zfeature(mos, ds, f); } } if (!ds->ds_is_snapshot) { ds->ds_snapname[0] = '\0'; if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev); } err = dsl_bookmark_init_ds(ds); } else { if (zfs_flags & ZFS_DEBUG_SNAPNAMES) err = dsl_dataset_get_snapname(ds); if (err == 0 && dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { err = zap_count( ds->ds_dir->dd_pool->dp_meta_objset, dsl_dataset_phys(ds)->ds_userrefs_obj, &ds->ds_userrefs); } } if (err == 0 && !ds->ds_is_snapshot) { err = dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &ds->ds_reserved); if (err == 0) { err = dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_REFQUOTA), &ds->ds_quota); } } else { ds->ds_reserved = ds->ds_quota = 0; } if (err == 0 && ds->ds_dir->dd_crypto_obj != 0 && ds->ds_is_snapshot && zap_contains(mos, dsobj, DS_FIELD_IVSET_GUID) != 0) { dp->dp_spa->spa_errata = ZPOOL_ERRATA_ZOL_8308_ENCRYPTION; } dsl_deadlist_open(&ds->ds_deadlist, mos, dsl_dataset_phys(ds)->ds_deadlist_obj); uint64_t remap_deadlist_obj = dsl_dataset_get_remap_deadlist_object(ds); if (remap_deadlist_obj != 0) { dsl_deadlist_open(&ds->ds_remap_deadlist, mos, remap_deadlist_obj); } dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync, dsl_dataset_evict_async, &ds->ds_dbuf); if (err == 0) winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); if (err != 0 || winner != NULL) { bplist_destroy(&ds->ds_pending_deadlist); dsl_deadlist_close(&ds->ds_deadlist); if (dsl_deadlist_is_open(&ds->ds_remap_deadlist)) dsl_deadlist_close(&ds->ds_remap_deadlist); dsl_bookmark_fini_ds(ds); if (ds->ds_prev) dsl_dataset_rele(ds->ds_prev, ds); dsl_dir_rele(ds->ds_dir, ds); for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (dsl_dataset_feature_is_active(ds, f)) unload_zfeature(ds, f); } list_destroy(&ds->ds_prop_cbs); list_destroy(&ds->ds_sendstreams); mutex_destroy(&ds->ds_lock); mutex_destroy(&ds->ds_opening_lock); mutex_destroy(&ds->ds_sendstream_lock); mutex_destroy(&ds->ds_remap_deadlist_lock); zfs_refcount_destroy(&ds->ds_longholds); rrw_destroy(&ds->ds_bp_rwlock); kmem_free(ds, sizeof (dsl_dataset_t)); if (err != 0) { dmu_buf_rele(dbuf, tag); return (err); } ds = winner; } else { ds->ds_fsid_guid = unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); if (ds->ds_fsid_guid != dsl_dataset_phys(ds)->ds_fsid_guid) { zfs_dbgmsg("ds_fsid_guid changed from " "%llx to %llx for pool %s dataset id %llu", (long long) dsl_dataset_phys(ds)->ds_fsid_guid, (long long)ds->ds_fsid_guid, spa_name(dp->dp_spa), (u_longlong_t)dsobj); } } } ASSERT3P(ds->ds_dbuf, ==, dbuf); ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); *dsp = ds; return (0); } int dsl_dataset_create_key_mapping(dsl_dataset_t *ds) { dsl_dir_t *dd = ds->ds_dir; if (dd->dd_crypto_obj == 0) return (0); return (spa_keystore_create_mapping(dd->dd_pool->dp_spa, ds, ds, &ds->ds_key_mapping)); } int dsl_dataset_hold_obj_flags(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp) { int err; err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); if (err != 0) return (err); ASSERT3P(*dsp, !=, NULL); if (flags & DS_HOLD_FLAG_DECRYPT) { err = dsl_dataset_create_key_mapping(*dsp); if (err != 0) dsl_dataset_rele(*dsp, tag); } return (err); } int dsl_dataset_hold_flags(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp) { dsl_dir_t *dd; const char *snapname; uint64_t obj; int err = 0; dsl_dataset_t *ds; err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); if (err != 0) return (err); ASSERT(dsl_pool_config_held(dp)); obj = dsl_dir_phys(dd)->dd_head_dataset_obj; if (obj != 0) err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, &ds); else err = SET_ERROR(ENOENT); /* we may be looking for a snapshot */ if (err == 0 && snapname != NULL) { dsl_dataset_t *snap_ds; if (*snapname++ != '@') { dsl_dataset_rele_flags(ds, flags, tag); dsl_dir_rele(dd, FTAG); return (SET_ERROR(ENOENT)); } dprintf("looking for snapshot '%s'\n", snapname); err = dsl_dataset_snap_lookup(ds, snapname, &obj); if (err == 0) { err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, &snap_ds); } dsl_dataset_rele_flags(ds, flags, tag); if (err == 0) { mutex_enter(&snap_ds->ds_lock); if (snap_ds->ds_snapname[0] == 0) (void) strlcpy(snap_ds->ds_snapname, snapname, sizeof (snap_ds->ds_snapname)); mutex_exit(&snap_ds->ds_lock); ds = snap_ds; } } if (err == 0) *dsp = ds; dsl_dir_rele(dd, FTAG); return (err); } int dsl_dataset_hold(dsl_pool_t *dp, const char *name, void *tag, dsl_dataset_t **dsp) { return (dsl_dataset_hold_flags(dp, name, 0, tag, dsp)); } static int dsl_dataset_own_obj_impl(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags, void *tag, boolean_t override, dsl_dataset_t **dsp) { int err = dsl_dataset_hold_obj_flags(dp, dsobj, flags, tag, dsp); if (err != 0) return (err); if (!dsl_dataset_tryown(*dsp, tag, override)) { dsl_dataset_rele_flags(*dsp, flags, tag); *dsp = NULL; return (SET_ERROR(EBUSY)); } return (0); } int dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp) { return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_FALSE, dsp)); } int dsl_dataset_own_obj_force(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp) { return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_TRUE, dsp)); } static int dsl_dataset_own_impl(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, void *tag, boolean_t override, dsl_dataset_t **dsp) { int err = dsl_dataset_hold_flags(dp, name, flags, tag, dsp); if (err != 0) return (err); if (!dsl_dataset_tryown(*dsp, tag, override)) { dsl_dataset_rele_flags(*dsp, flags, tag); return (SET_ERROR(EBUSY)); } return (0); } int dsl_dataset_own_force(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp) { return (dsl_dataset_own_impl(dp, name, flags, tag, B_TRUE, dsp)); } int dsl_dataset_own(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp) { return (dsl_dataset_own_impl(dp, name, flags, tag, B_FALSE, dsp)); } /* * See the comment above dsl_pool_hold() for details. In summary, a long * hold is used to prevent destruction of a dataset while the pool hold * is dropped, allowing other concurrent operations (e.g. spa_sync()). * * The dataset and pool must be held when this function is called. After it * is called, the pool hold may be released while the dataset is still held * and accessed. */ void dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) { ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); (void) zfs_refcount_add(&ds->ds_longholds, tag); } void dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) { (void) zfs_refcount_remove(&ds->ds_longholds, tag); } /* Return B_TRUE if there are any long holds on this dataset. */ boolean_t dsl_dataset_long_held(dsl_dataset_t *ds) { return (!zfs_refcount_is_zero(&ds->ds_longholds)); } void dsl_dataset_name(dsl_dataset_t *ds, char *name) { if (ds == NULL) { (void) strlcpy(name, "mos", ZFS_MAX_DATASET_NAME_LEN); } else { dsl_dir_name(ds->ds_dir, name); VERIFY0(dsl_dataset_get_snapname(ds)); if (ds->ds_snapname[0]) { VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN), <, ZFS_MAX_DATASET_NAME_LEN); /* * We use a "recursive" mutex so that we * can call dprintf_ds() with ds_lock held. */ if (!MUTEX_HELD(&ds->ds_lock)) { mutex_enter(&ds->ds_lock); VERIFY3U(strlcat(name, ds->ds_snapname, ZFS_MAX_DATASET_NAME_LEN), <, ZFS_MAX_DATASET_NAME_LEN); mutex_exit(&ds->ds_lock); } else { VERIFY3U(strlcat(name, ds->ds_snapname, ZFS_MAX_DATASET_NAME_LEN), <, ZFS_MAX_DATASET_NAME_LEN); } } } } int dsl_dataset_namelen(dsl_dataset_t *ds) { VERIFY0(dsl_dataset_get_snapname(ds)); mutex_enter(&ds->ds_lock); int len = strlen(ds->ds_snapname); mutex_exit(&ds->ds_lock); /* add '@' if ds is a snap */ if (len > 0) len++; len += dsl_dir_namelen(ds->ds_dir); return (len); } void dsl_dataset_rele(dsl_dataset_t *ds, void *tag) { dmu_buf_rele(ds->ds_dbuf, tag); } void dsl_dataset_remove_key_mapping(dsl_dataset_t *ds) { dsl_dir_t *dd = ds->ds_dir; if (dd == NULL || dd->dd_crypto_obj == 0) return; (void) spa_keystore_remove_mapping(dd->dd_pool->dp_spa, ds->ds_object, ds); } void dsl_dataset_rele_flags(dsl_dataset_t *ds, ds_hold_flags_t flags, void *tag) { if (flags & DS_HOLD_FLAG_DECRYPT) dsl_dataset_remove_key_mapping(ds); dsl_dataset_rele(ds, tag); } void dsl_dataset_disown(dsl_dataset_t *ds, ds_hold_flags_t flags, void *tag) { ASSERT3P(ds->ds_owner, ==, tag); ASSERT(ds->ds_dbuf != NULL); mutex_enter(&ds->ds_lock); ds->ds_owner = NULL; mutex_exit(&ds->ds_lock); dsl_dataset_long_rele(ds, tag); dsl_dataset_rele_flags(ds, flags, tag); } boolean_t dsl_dataset_tryown(dsl_dataset_t *ds, void *tag, boolean_t override) { boolean_t gotit = FALSE; ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); mutex_enter(&ds->ds_lock); if (ds->ds_owner == NULL && (override || !(DS_IS_INCONSISTENT(ds) || (dsl_dataset_feature_is_active(ds, SPA_FEATURE_REDACTED_DATASETS) && !zfs_allow_redacted_dataset_mount)))) { ds->ds_owner = tag; dsl_dataset_long_hold(ds, tag); gotit = TRUE; } mutex_exit(&ds->ds_lock); return (gotit); } boolean_t dsl_dataset_has_owner(dsl_dataset_t *ds) { boolean_t rv; mutex_enter(&ds->ds_lock); rv = (ds->ds_owner != NULL); mutex_exit(&ds->ds_lock); return (rv); } boolean_t zfeature_active(spa_feature_t f, void *arg) { switch (spa_feature_table[f].fi_type) { case ZFEATURE_TYPE_BOOLEAN: { boolean_t val = (boolean_t)(uintptr_t)arg; ASSERT(val == B_FALSE || val == B_TRUE); return (val); } case ZFEATURE_TYPE_UINT64_ARRAY: /* * In this case, arg is a uint64_t array. The feature is active * if the array is non-null. */ return (arg != NULL); default: panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); return (B_FALSE); } } boolean_t dsl_dataset_feature_is_active(dsl_dataset_t *ds, spa_feature_t f) { return (zfeature_active(f, ds->ds_feature[f])); } /* * The buffers passed out by this function are references to internal buffers; * they should not be freed by callers of this function, and they should not be * used after the dataset has been released. */ boolean_t dsl_dataset_get_uint64_array_feature(dsl_dataset_t *ds, spa_feature_t f, uint64_t *outlength, uint64_t **outp) { VERIFY(spa_feature_table[f].fi_type & ZFEATURE_TYPE_UINT64_ARRAY); if (!dsl_dataset_feature_is_active(ds, f)) { return (B_FALSE); } struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f]; *outp = ftuaa->array; *outlength = ftuaa->length; return (B_TRUE); } void dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, void *arg, dmu_tx_t *tx) { spa_t *spa = dmu_tx_pool(tx)->dp_spa; objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; uint64_t zero = 0; VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); spa_feature_incr(spa, f, tx); dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); switch (spa_feature_table[f].fi_type) { case ZFEATURE_TYPE_BOOLEAN: ASSERT3S((boolean_t)(uintptr_t)arg, ==, B_TRUE); VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, sizeof (zero), 1, &zero, tx)); break; case ZFEATURE_TYPE_UINT64_ARRAY: { struct feature_type_uint64_array_arg *ftuaa = arg; VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, sizeof (uint64_t), ftuaa->length, ftuaa->array, tx)); break; } default: panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); } } static void dsl_dataset_deactivate_feature_impl(dsl_dataset_t *ds, spa_feature_t f, dmu_tx_t *tx) { spa_t *spa = dmu_tx_pool(tx)->dp_spa; objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; uint64_t dsobj = ds->ds_object; VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx)); spa_feature_decr(spa, f, tx); ds->ds_feature[f] = NULL; } void dsl_dataset_deactivate_feature(dsl_dataset_t *ds, spa_feature_t f, dmu_tx_t *tx) { unload_zfeature(ds, f); dsl_dataset_deactivate_feature_impl(ds, f, tx); } uint64_t dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, dsl_crypto_params_t *dcp, uint64_t flags, dmu_tx_t *tx) { dsl_pool_t *dp = dd->dd_pool; dmu_buf_t *dbuf; dsl_dataset_phys_t *dsphys; uint64_t dsobj; objset_t *mos = dp->dp_meta_objset; if (origin == NULL) origin = dp->dp_origin_snap; ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); ASSERT(dmu_tx_is_syncing(tx)); ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); dmu_buf_will_dirty(dbuf, tx); dsphys = dbuf->db_data; bzero(dsphys, sizeof (dsl_dataset_phys_t)); dsphys->ds_dir_obj = dd->dd_object; dsphys->ds_flags = flags; dsphys->ds_fsid_guid = unique_create(); (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, sizeof (dsphys->ds_guid)); dsphys->ds_snapnames_zapobj = zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx); dsphys->ds_creation_time = gethrestime_sec(); dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; if (origin == NULL) { dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); } else { dsl_dataset_t *ohds; /* head of the origin snapshot */ dsphys->ds_prev_snap_obj = origin->ds_object; dsphys->ds_prev_snap_txg = dsl_dataset_phys(origin)->ds_creation_txg; dsphys->ds_referenced_bytes = dsl_dataset_phys(origin)->ds_referenced_bytes; dsphys->ds_compressed_bytes = dsl_dataset_phys(origin)->ds_compressed_bytes; dsphys->ds_uncompressed_bytes = dsl_dataset_phys(origin)->ds_uncompressed_bytes; rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG); dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; rrw_exit(&origin->ds_bp_rwlock, FTAG); /* * Inherit flags that describe the dataset's contents * (INCONSISTENT) or properties (Case Insensitive). */ dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (zfeature_active(f, origin->ds_feature[f])) { dsl_dataset_activate_feature(dsobj, f, origin->ds_feature[f], tx); } } dmu_buf_will_dirty(origin->ds_dbuf, tx); dsl_dataset_phys(origin)->ds_num_children++; VERIFY0(dsl_dataset_hold_obj(dp, dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, FTAG, &ohds)); dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); dsl_dataset_rele(ohds, FTAG); if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { dsl_dataset_phys(origin)->ds_next_clones_obj = zap_create(mos, DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); } VERIFY0(zap_add_int(mos, dsl_dataset_phys(origin)->ds_next_clones_obj, dsobj, tx)); } dmu_buf_will_dirty(dd->dd_dbuf, tx); dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); dsl_dir_phys(origin->ds_dir)->dd_clones = zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); } VERIFY0(zap_add_int(mos, dsl_dir_phys(origin->ds_dir)->dd_clones, dsobj, tx)); } } /* handle encryption */ dsl_dataset_create_crypt_sync(dsobj, dd, origin, dcp, tx); if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; dmu_buf_rele(dbuf, FTAG); dmu_buf_will_dirty(dd->dd_dbuf, tx); dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; return (dsobj); } static void dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) { objset_t *os; VERIFY0(dmu_objset_from_ds(ds, &os)); if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) { dsl_pool_t *dp = ds->ds_dir->dd_pool; zio_t *zio; bzero(&os->os_zil_header, sizeof (os->os_zil_header)); if (os->os_encrypted) os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE; zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); dsl_dataset_sync(ds, zio, tx); VERIFY0(zio_wait(zio)); /* dsl_dataset_sync_done will drop this reference. */ dmu_buf_add_ref(ds->ds_dbuf, ds); dsl_dataset_sync_done(ds, tx); } } uint64_t dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dsl_crypto_params_t *dcp, dmu_tx_t *tx) { dsl_pool_t *dp = pdd->dd_pool; uint64_t dsobj, ddobj; dsl_dir_t *dd; ASSERT(dmu_tx_is_syncing(tx)); ASSERT(lastname[0] != '@'); /* * Filesystems will eventually have their origin set to dp_origin_snap, * but that's taken care of in dsl_dataset_create_sync_dd. When * creating a filesystem, this function is called with origin equal to * NULL. */ if (origin != NULL) ASSERT3P(origin, !=, dp->dp_origin_snap); ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); dsobj = dsl_dataset_create_sync_dd(dd, origin, dcp, flags & ~DS_CREATE_FLAG_NODIRTY, tx); dsl_deleg_set_create_perms(dd, tx, cr); /* * If we are creating a clone and the livelist feature is enabled, * add the entry DD_FIELD_LIVELIST to ZAP. */ if (origin != NULL && spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LIVELIST)) { objset_t *mos = dd->dd_pool->dp_meta_objset; dsl_dir_zapify(dd, tx); uint64_t obj = dsl_deadlist_alloc(mos, tx); VERIFY0(zap_add(mos, dd->dd_object, DD_FIELD_LIVELIST, sizeof (uint64_t), 1, &obj, tx)); spa_feature_incr(dp->dp_spa, SPA_FEATURE_LIVELIST, tx); } /* * Since we're creating a new node we know it's a leaf, so we can * initialize the counts if the limit feature is active. */ if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { uint64_t cnt = 0; objset_t *os = dd->dd_pool->dp_meta_objset; dsl_dir_zapify(dd, tx); VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, sizeof (cnt), 1, &cnt, tx)); VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, sizeof (cnt), 1, &cnt, tx)); } dsl_dir_rele(dd, FTAG); /* * If we are creating a clone, make sure we zero out any stale * data from the origin snapshots zil header. */ if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { dsl_dataset_t *ds; VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); dsl_dataset_zero_zil(ds, tx); dsl_dataset_rele(ds, FTAG); } return (dsobj); } /* * The unique space in the head dataset can be calculated by subtracting * the space used in the most recent snapshot, that is still being used * in this file system, from the space currently in use. To figure out * the space in the most recent snapshot still in use, we need to take * the total space used in the snapshot and subtract out the space that * has been freed up since the snapshot was taken. */ void dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) { uint64_t mrs_used; uint64_t dlused, dlcomp, dluncomp; ASSERT(!ds->ds_is_snapshot); if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; else mrs_used = 0; dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); ASSERT3U(dlused, <=, mrs_used); dsl_dataset_phys(ds)->ds_unique_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; } void dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx) { objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; uint64_t count __maybe_unused; int err; ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, obj, tx); /* * The err should not be ENOENT, but a bug in a previous version * of the code could cause upgrade_clones_cb() to not set * ds_next_snap_obj when it should, leading to a missing entry. * If we knew that the pool was created after * SPA_VERSION_NEXT_CLONES, we could assert that it isn't * ENOENT. However, at least we can check that we don't have * too many entries in the next_clones_obj even after failing to * remove this one. */ if (err != ENOENT) VERIFY0(err); ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, &count)); ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); } blkptr_t * dsl_dataset_get_blkptr(dsl_dataset_t *ds) { return (&dsl_dataset_phys(ds)->ds_bp); } spa_t * dsl_dataset_get_spa(dsl_dataset_t *ds) { return (ds->ds_dir->dd_pool->dp_spa); } void dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_pool_t *dp; if (ds == NULL) /* this is the meta-objset */ return; ASSERT(ds->ds_objset != NULL); if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) panic("dirtying snapshot!"); /* Must not dirty a dataset in the same txg where it got snapshotted. */ ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); dp = ds->ds_dir->dd_pool; if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { objset_t *os = ds->ds_objset; /* up the hold count until we can be written out */ dmu_buf_add_ref(ds->ds_dbuf, ds); /* if this dataset is encrypted, grab a reference to the DCK */ if (ds->ds_dir->dd_crypto_obj != 0 && !os->os_raw_receive && !os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { ASSERT3P(ds->ds_key_mapping, !=, NULL); key_mapping_add_ref(ds->ds_key_mapping, ds); } } } static int dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) { uint64_t asize; if (!dmu_tx_is_syncing(tx)) return (0); /* * If there's an fs-only reservation, any blocks that might become * owned by the snapshot dataset must be accommodated by space * outside of the reservation. */ ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) return (SET_ERROR(ENOSPC)); /* * Propagate any reserved space for this snapshot to other * snapshot checks in this sync group. */ if (asize > 0) dsl_dir_willuse_space(ds->ds_dir, asize, tx); return (0); } int dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr, proc_t *proc) { int error; uint64_t value; ds->ds_trysnap_txg = tx->tx_txg; if (!dmu_tx_is_syncing(tx)) return (0); /* * We don't allow multiple snapshots of the same txg. If there * is already one, try again. */ if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) return (SET_ERROR(EAGAIN)); /* * Check for conflicting snapshot name. */ error = dsl_dataset_snap_lookup(ds, snapname, &value); if (error == 0) return (SET_ERROR(EEXIST)); if (error != ENOENT) return (error); /* * We don't allow taking snapshots of inconsistent datasets, such as * those into which we are currently receiving. However, if we are * creating this snapshot as part of a receive, this check will be * executed atomically with respect to the completion of the receive * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this * case we ignore this, knowing it will be fixed up for us shortly in * dmu_recv_end_sync(). */ if (!recv && DS_IS_INCONSISTENT(ds)) return (SET_ERROR(EBUSY)); /* * Skip the check for temporary snapshots or if we have already checked * the counts in dsl_dataset_snapshot_check. This means we really only * check the count here when we're receiving a stream. */ if (cnt != 0 && cr != NULL) { error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr, proc); if (error != 0) return (error); } error = dsl_dataset_snapshot_reserve_space(ds, tx); if (error != 0) return (error); return (0); } int dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) { dsl_dataset_snapshot_arg_t *ddsa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); nvpair_t *pair; int rv = 0; /* * Pre-compute how many total new snapshots will be created for each * level in the tree and below. This is needed for validating the * snapshot limit when either taking a recursive snapshot or when * taking multiple snapshots. * * The problem is that the counts are not actually adjusted when * we are checking, only when we finally sync. For a single snapshot, * this is easy, the count will increase by 1 at each node up the tree, * but its more complicated for the recursive/multiple snapshot case. * * The dsl_fs_ss_limit_check function does recursively check the count * at each level up the tree but since it is validating each snapshot * independently we need to be sure that we are validating the complete * count for the entire set of snapshots. We do this by rolling up the * counts for each component of the name into an nvlist and then * checking each of those cases with the aggregated count. * * This approach properly handles not only the recursive snapshot * case (where we get all of those on the ddsa_snaps list) but also * the sibling case (e.g. snapshot a/b and a/c so that we will also * validate the limit on 'a' using a count of 2). * * We validate the snapshot names in the third loop and only report * name errors once. */ if (dmu_tx_is_syncing(tx)) { char *nm; nvlist_t *cnt_track = NULL; cnt_track = fnvlist_alloc(); nm = kmem_alloc(MAXPATHLEN, KM_SLEEP); /* Rollup aggregated counts into the cnt_track list */ for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { char *pdelim; uint64_t val; (void) strlcpy(nm, nvpair_name(pair), MAXPATHLEN); pdelim = strchr(nm, '@'); if (pdelim == NULL) continue; *pdelim = '\0'; do { if (nvlist_lookup_uint64(cnt_track, nm, &val) == 0) { /* update existing entry */ fnvlist_add_uint64(cnt_track, nm, val + 1); } else { /* add to list */ fnvlist_add_uint64(cnt_track, nm, 1); } pdelim = strrchr(nm, '/'); if (pdelim != NULL) *pdelim = '\0'; } while (pdelim != NULL); } kmem_free(nm, MAXPATHLEN); /* Check aggregated counts at each level */ for (pair = nvlist_next_nvpair(cnt_track, NULL); pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { int error = 0; char *name; uint64_t cnt = 0; dsl_dataset_t *ds; name = nvpair_name(pair); cnt = fnvpair_value_uint64(pair); ASSERT(cnt > 0); error = dsl_dataset_hold(dp, name, FTAG, &ds); if (error == 0) { error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, ZFS_PROP_SNAPSHOT_LIMIT, NULL, ddsa->ddsa_cr, ddsa->ddsa_proc); dsl_dataset_rele(ds, FTAG); } if (error != 0) { if (ddsa->ddsa_errors != NULL) fnvlist_add_int32(ddsa->ddsa_errors, name, error); rv = error; /* only report one error for this check */ break; } } nvlist_free(cnt_track); } for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { int error = 0; dsl_dataset_t *ds; char *name, *atp = NULL; char dsname[ZFS_MAX_DATASET_NAME_LEN]; name = nvpair_name(pair); if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN) error = SET_ERROR(ENAMETOOLONG); if (error == 0) { atp = strchr(name, '@'); if (atp == NULL) error = SET_ERROR(EINVAL); if (error == 0) (void) strlcpy(dsname, name, atp - name + 1); } if (error == 0) error = dsl_dataset_hold(dp, dsname, FTAG, &ds); if (error == 0) { /* passing 0/NULL skips dsl_fs_ss_limit_check */ error = dsl_dataset_snapshot_check_impl(ds, atp + 1, tx, B_FALSE, 0, NULL, NULL); dsl_dataset_rele(ds, FTAG); } if (error != 0) { if (ddsa->ddsa_errors != NULL) { fnvlist_add_int32(ddsa->ddsa_errors, name, error); } rv = error; } } return (rv); } void dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, dmu_tx_t *tx) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dmu_buf_t *dbuf; dsl_dataset_phys_t *dsphys; uint64_t dsobj, crtxg; objset_t *mos = dp->dp_meta_objset; static zil_header_t zero_zil __maybe_unused; objset_t *os __maybe_unused; ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); /* * If we are on an old pool, the zil must not be active, in which * case it will be zeroed. Usually zil_suspend() accomplishes this. */ ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || dmu_objset_from_ds(ds, &os) != 0 || bcmp(&os->os_phys->os_zil_header, &zero_zil, sizeof (zero_zil)) == 0); /* Should not snapshot a dirty dataset. */ ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, ds, tx->tx_txg)); dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); /* * The origin's ds_creation_txg has to be < TXG_INITIAL */ if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) crtxg = 1; else crtxg = tx->tx_txg; dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); dmu_buf_will_dirty(dbuf, tx); dsphys = dbuf->db_data; bzero(dsphys, sizeof (dsl_dataset_phys_t)); dsphys->ds_dir_obj = ds->ds_dir->dd_object; dsphys->ds_fsid_guid = unique_create(); (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, sizeof (dsphys->ds_guid)); dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; dsphys->ds_next_snap_obj = ds->ds_object; dsphys->ds_num_children = 1; dsphys->ds_creation_time = gethrestime_sec(); dsphys->ds_creation_txg = crtxg; dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; dsphys->ds_uncompressed_bytes = dsl_dataset_phys(ds)->ds_uncompressed_bytes; dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; rrw_exit(&ds->ds_bp_rwlock, FTAG); dmu_buf_rele(dbuf, FTAG); for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (zfeature_active(f, ds->ds_feature[f])) { dsl_dataset_activate_feature(dsobj, f, ds->ds_feature[f], tx); } } - /* - * We are not allowed to dirty a filesystem when done receiving - * a snapshot. In this case some flags such as SPA_FEATURE_LARGE_BLOCKS - * will not be set and a subsequent encrypted raw send will fail. Hence - * activate this feature if needed here. This needs to happen only in - * syncing context. - */ - if (dmu_tx_is_syncing(tx)) { - for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { - if (zfeature_active(f, ds->ds_feature_activation[f]) && - !(zfeature_active(f, ds->ds_feature[f]))) { - dsl_dataset_activate_feature(dsobj, f, - ds->ds_feature_activation[f], tx); - ds->ds_feature[f] = - ds->ds_feature_activation[f]; - } - } - } - ASSERT3U(ds->ds_prev != 0, ==, dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); if (ds->ds_prev) { uint64_t next_clones_obj = dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object || dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object) { dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; } else if (next_clones_obj != 0) { dsl_dataset_remove_from_next_clones(ds->ds_prev, dsphys->ds_next_snap_obj, tx); VERIFY0(zap_add_int(mos, next_clones_obj, dsobj, tx)); } } /* * If we have a reference-reservation on this dataset, we will * need to increase the amount of refreservation being charged * since our unique space is going to zero. */ if (ds->ds_reserved) { int64_t delta; ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); } dmu_buf_will_dirty(ds->ds_dbuf, tx); dsl_dataset_phys(ds)->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); dsl_deadlist_close(&ds->ds_deadlist); dsl_deadlist_open(&ds->ds_deadlist, mos, dsl_dataset_phys(ds)->ds_deadlist_obj); dsl_deadlist_add_key(&ds->ds_deadlist, dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); dsl_bookmark_snapshotted(ds, tx); if (dsl_dataset_remap_deadlist_exists(ds)) { uint64_t remap_deadlist_obj = dsl_dataset_get_remap_deadlist_object(ds); /* * Move the remap_deadlist to the snapshot. The head * will create a new remap deadlist on demand, from * dsl_dataset_block_remapped(). */ dsl_dataset_unset_remap_deadlist_object(ds, tx); dsl_deadlist_close(&ds->ds_remap_deadlist); dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx)); } /* * Create a ivset guid for this snapshot if the dataset is * encrypted. This may be overridden by a raw receive. A * previous implementation of this code did not have this * field as part of the on-disk format for ZFS encryption * (see errata #4). As part of the remediation for this * issue, we ask the user to enable the bookmark_v2 feature * which is now a dependency of the encryption feature. We * use this as a heuristic to determine when the user has * elected to correct any datasets created with the old code. * As a result, we only do this step if the bookmark_v2 * feature is enabled, which limits the number of states a * given pool / dataset can be in with regards to terms of * correcting the issue. */ if (ds->ds_dir->dd_crypto_obj != 0 && spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARK_V2)) { uint64_t ivset_guid = unique_create(); dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); VERIFY0(zap_add(mos, dsobj, DS_FIELD_IVSET_GUID, sizeof (ivset_guid), 1, &ivset_guid, tx)); } ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; dsl_dataset_phys(ds)->ds_unique_bytes = 0; if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, snapname, 8, 1, &dsobj, tx)); if (ds->ds_prev) dsl_dataset_rele(ds->ds_prev, ds); VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); dsl_scan_ds_snapshotted(ds, tx); dsl_dir_snap_cmtime_update(ds->ds_dir); spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, " "); } void dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_snapshot_arg_t *ddsa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); nvpair_t *pair; for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { dsl_dataset_t *ds; char *name, *atp; char dsname[ZFS_MAX_DATASET_NAME_LEN]; name = nvpair_name(pair); atp = strchr(name, '@'); (void) strlcpy(dsname, name, atp - name + 1); VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); if (ddsa->ddsa_props != NULL) { dsl_props_set_sync_impl(ds->ds_prev, ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); } dsl_dataset_rele(ds, FTAG); } } /* * The snapshots must all be in the same pool. * All-or-nothing: if there are any failures, nothing will be modified. */ int dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) { dsl_dataset_snapshot_arg_t ddsa; nvpair_t *pair; boolean_t needsuspend; int error; spa_t *spa; char *firstname; nvlist_t *suspended = NULL; pair = nvlist_next_nvpair(snaps, NULL); if (pair == NULL) return (0); firstname = nvpair_name(pair); error = spa_open(firstname, &spa, FTAG); if (error != 0) return (error); needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); spa_close(spa, FTAG); if (needsuspend) { suspended = fnvlist_alloc(); for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; pair = nvlist_next_nvpair(snaps, pair)) { char fsname[ZFS_MAX_DATASET_NAME_LEN]; char *snapname = nvpair_name(pair); char *atp; void *cookie; atp = strchr(snapname, '@'); if (atp == NULL) { error = SET_ERROR(EINVAL); break; } (void) strlcpy(fsname, snapname, atp - snapname + 1); error = zil_suspend(fsname, &cookie); if (error != 0) break; fnvlist_add_uint64(suspended, fsname, (uintptr_t)cookie); } } ddsa.ddsa_snaps = snaps; ddsa.ddsa_props = props; ddsa.ddsa_errors = errors; ddsa.ddsa_cr = CRED(); ddsa.ddsa_proc = curproc; if (error == 0) { error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, dsl_dataset_snapshot_sync, &ddsa, fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); } if (suspended != NULL) { for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; pair = nvlist_next_nvpair(suspended, pair)) { zil_resume((void *)(uintptr_t) fnvpair_value_uint64(pair)); } fnvlist_free(suspended); } if (error == 0) { for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; pair = nvlist_next_nvpair(snaps, pair)) { zvol_create_minor(nvpair_name(pair)); } } return (error); } typedef struct dsl_dataset_snapshot_tmp_arg { const char *ddsta_fsname; const char *ddsta_snapname; minor_t ddsta_cleanup_minor; const char *ddsta_htag; } dsl_dataset_snapshot_tmp_arg_t; static int dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) { dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; int error; error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); if (error != 0) return (error); /* NULL cred means no limit check for tmp snapshot */ error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, tx, B_FALSE, 0, NULL, NULL); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(ENOTSUP)); } error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, B_TRUE, tx); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds = NULL; VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); dsl_dataset_rele(ds, FTAG); } int dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, minor_t cleanup_minor, const char *htag) { dsl_dataset_snapshot_tmp_arg_t ddsta; int error; spa_t *spa; boolean_t needsuspend; void *cookie; ddsta.ddsta_fsname = fsname; ddsta.ddsta_snapname = snapname; ddsta.ddsta_cleanup_minor = cleanup_minor; ddsta.ddsta_htag = htag; error = spa_open(fsname, &spa, FTAG); if (error != 0) return (error); needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); spa_close(spa, FTAG); if (needsuspend) { error = zil_suspend(fsname, &cookie); if (error != 0) return (error); } error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); if (needsuspend) zil_resume(cookie); return (error); } void dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) { ASSERT(dmu_tx_is_syncing(tx)); ASSERT(ds->ds_objset != NULL); ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); /* * in case we had to change ds_fsid_guid when we opened it, * sync it out now. */ dmu_buf_will_dirty(ds->ds_dbuf, tx); dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) { VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1, &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx)); VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1, &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx)); VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1, &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx)); ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0; ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0; ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0; } dmu_objset_sync(ds->ds_objset, zio, tx); } /* * Check if the percentage of blocks shared between the clone and the * snapshot (as opposed to those that are clone only) is below a certain * threshold */ static boolean_t dsl_livelist_should_disable(dsl_dataset_t *ds) { uint64_t used, referenced; int percent_shared; used = dsl_dir_get_usedds(ds->ds_dir); referenced = dsl_get_referenced(ds); ASSERT3U(referenced, >=, 0); ASSERT3U(used, >=, 0); if (referenced == 0) return (B_FALSE); percent_shared = (100 * (referenced - used)) / referenced; if (percent_shared <= zfs_livelist_min_percent_shared) return (B_TRUE); return (B_FALSE); } /* * Check if it is possible to combine two livelist entries into one. * This is the case if the combined number of 'live' blkptrs (ALLOCs that * don't have a matching FREE) is under the maximum sublist size. * We check this by subtracting twice the total number of frees from the total * number of blkptrs. FREEs are counted twice because each FREE blkptr * will cancel out an ALLOC blkptr when the livelist is processed. */ static boolean_t dsl_livelist_should_condense(dsl_deadlist_entry_t *first, dsl_deadlist_entry_t *next) { uint64_t total_free = first->dle_bpobj.bpo_phys->bpo_num_freed + next->dle_bpobj.bpo_phys->bpo_num_freed; uint64_t total_entries = first->dle_bpobj.bpo_phys->bpo_num_blkptrs + next->dle_bpobj.bpo_phys->bpo_num_blkptrs; if ((total_entries - (2 * total_free)) < zfs_livelist_max_entries) return (B_TRUE); return (B_FALSE); } typedef struct try_condense_arg { spa_t *spa; dsl_dataset_t *ds; } try_condense_arg_t; /* * Iterate over the livelist entries, searching for a pair to condense. * A nonzero return value means stop, 0 means keep looking. */ static int dsl_livelist_try_condense(void *arg, dsl_deadlist_entry_t *first) { try_condense_arg_t *tca = arg; spa_t *spa = tca->spa; dsl_dataset_t *ds = tca->ds; dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist; dsl_deadlist_entry_t *next; /* The condense thread has not yet been created at import */ if (spa->spa_livelist_condense_zthr == NULL) return (1); /* A condense is already in progress */ if (spa->spa_to_condense.ds != NULL) return (1); next = AVL_NEXT(&ll->dl_tree, &first->dle_node); /* The livelist has only one entry - don't condense it */ if (next == NULL) return (1); /* Next is the newest entry - don't condense it */ if (AVL_NEXT(&ll->dl_tree, &next->dle_node) == NULL) return (1); /* This pair is not ready to condense but keep looking */ if (!dsl_livelist_should_condense(first, next)) return (0); /* * Add a ref to prevent the dataset from being evicted while * the condense zthr or synctask are running. Ref will be * released at the end of the condense synctask */ dmu_buf_add_ref(ds->ds_dbuf, spa); spa->spa_to_condense.ds = ds; spa->spa_to_condense.first = first; spa->spa_to_condense.next = next; spa->spa_to_condense.syncing = B_FALSE; spa->spa_to_condense.cancelled = B_FALSE; zthr_wakeup(spa->spa_livelist_condense_zthr); return (1); } static void dsl_flush_pending_livelist(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_dir_t *dd = ds->ds_dir; spa_t *spa = ds->ds_dir->dd_pool->dp_spa; dsl_deadlist_entry_t *last = dsl_deadlist_last(&dd->dd_livelist); /* Check if we need to add a new sub-livelist */ if (last == NULL) { /* The livelist is empty */ dsl_deadlist_add_key(&dd->dd_livelist, tx->tx_txg - 1, tx); } else if (spa_sync_pass(spa) == 1) { /* * Check if the newest entry is full. If it is, make a new one. * We only do this once per sync because we could overfill a * sublist in one sync pass and don't want to add another entry * for a txg that is already represented. This ensures that * blkptrs born in the same txg are stored in the same sublist. */ bpobj_t bpobj = last->dle_bpobj; uint64_t all = bpobj.bpo_phys->bpo_num_blkptrs; uint64_t free = bpobj.bpo_phys->bpo_num_freed; uint64_t alloc = all - free; if (alloc > zfs_livelist_max_entries) { dsl_deadlist_add_key(&dd->dd_livelist, tx->tx_txg - 1, tx); } } /* Insert each entry into the on-disk livelist */ bplist_iterate(&dd->dd_pending_allocs, dsl_deadlist_insert_alloc_cb, &dd->dd_livelist, tx); bplist_iterate(&dd->dd_pending_frees, dsl_deadlist_insert_free_cb, &dd->dd_livelist, tx); /* Attempt to condense every pair of adjacent entries */ try_condense_arg_t arg = { .spa = spa, .ds = ds }; dsl_deadlist_iterate(&dd->dd_livelist, dsl_livelist_try_condense, &arg); } void dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx) { objset_t *os = ds->ds_objset; bplist_iterate(&ds->ds_pending_deadlist, dsl_deadlist_insert_alloc_cb, &ds->ds_deadlist, tx); if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist)) { dsl_flush_pending_livelist(ds, tx); if (dsl_livelist_should_disable(ds)) { dsl_dir_remove_livelist(ds->ds_dir, tx, B_TRUE); } } dsl_bookmark_sync_done(ds, tx); multilist_destroy(&os->os_synced_dnodes); if (os->os_encrypted) os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE; else ASSERT0(os->os_next_write_raw[tx->tx_txg & TXG_MASK]); ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx))); dmu_buf_rele(ds->ds_dbuf, ds); for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (zfeature_active(f, ds->ds_feature_activation[f])) { if (zfeature_active(f, ds->ds_feature[f])) continue; dsl_dataset_activate_feature(ds->ds_object, f, ds->ds_feature_activation[f], tx); ds->ds_feature[f] = ds->ds_feature_activation[f]; } } } int get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val) { uint64_t count = 0; objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; zap_cursor_t zc; zap_attribute_t za; ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); /* * There may be missing entries in ds_next_clones_obj * due to a bug in a previous version of the code. * Only trust it if it has the right number of entries. */ if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, &count)); } if (count != dsl_dataset_phys(ds)->ds_num_children - 1) { return (SET_ERROR(ENOENT)); } for (zap_cursor_init(&zc, mos, dsl_dataset_phys(ds)->ds_next_clones_obj); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { dsl_dataset_t *clone; char buf[ZFS_MAX_DATASET_NAME_LEN]; VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, za.za_first_integer, FTAG, &clone)); dsl_dir_name(clone->ds_dir, buf); fnvlist_add_boolean(val, buf); dsl_dataset_rele(clone, FTAG); } zap_cursor_fini(&zc); return (0); } void get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) { nvlist_t *propval = fnvlist_alloc(); nvlist_t *val = fnvlist_alloc(); if (get_clones_stat_impl(ds, val) == 0) { fnvlist_add_nvlist(propval, ZPROP_VALUE, val); fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); } nvlist_free(val); nvlist_free(propval); } /* * Returns a string that represents the receive resume stats token. It should * be freed with strfree(). */ char * get_receive_resume_stats_impl(dsl_dataset_t *ds) { dsl_pool_t *dp = ds->ds_dir->dd_pool; if (dsl_dataset_has_resume_receive_state(ds)) { char *str; void *packed; uint8_t *compressed; uint64_t val; nvlist_t *token_nv = fnvlist_alloc(); size_t packed_size, compressed_size; if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) { fnvlist_add_uint64(token_nv, "fromguid", val); } if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) { fnvlist_add_uint64(token_nv, "object", val); } if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) { fnvlist_add_uint64(token_nv, "offset", val); } if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) { fnvlist_add_uint64(token_nv, "bytes", val); } if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) { fnvlist_add_uint64(token_nv, "toguid", val); } char buf[MAXNAMELEN]; if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) { fnvlist_add_string(token_nv, "toname", buf); } if (zap_contains(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_LARGEBLOCK) == 0) { fnvlist_add_boolean(token_nv, "largeblockok"); } if (zap_contains(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_EMBEDOK) == 0) { fnvlist_add_boolean(token_nv, "embedok"); } if (zap_contains(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_COMPRESSOK) == 0) { fnvlist_add_boolean(token_nv, "compressok"); } if (zap_contains(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_RAWOK) == 0) { fnvlist_add_boolean(token_nv, "rawok"); } if (dsl_dataset_feature_is_active(ds, SPA_FEATURE_REDACTED_DATASETS)) { uint64_t num_redact_snaps; uint64_t *redact_snaps; VERIFY(dsl_dataset_get_uint64_array_feature(ds, SPA_FEATURE_REDACTED_DATASETS, &num_redact_snaps, &redact_snaps)); fnvlist_add_uint64_array(token_nv, "redact_snaps", redact_snaps, num_redact_snaps); } if (zap_contains(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS) == 0) { uint64_t num_redact_snaps, int_size; uint64_t *redact_snaps; VERIFY0(zap_length(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, &int_size, &num_redact_snaps)); ASSERT3U(int_size, ==, sizeof (uint64_t)); redact_snaps = kmem_alloc(int_size * num_redact_snaps, KM_SLEEP); VERIFY0(zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, int_size, num_redact_snaps, redact_snaps)); fnvlist_add_uint64_array(token_nv, "book_redact_snaps", redact_snaps, num_redact_snaps); kmem_free(redact_snaps, int_size * num_redact_snaps); } packed = fnvlist_pack(token_nv, &packed_size); fnvlist_free(token_nv); compressed = kmem_alloc(packed_size, KM_SLEEP); compressed_size = gzip_compress(packed, compressed, packed_size, packed_size, 6); zio_cksum_t cksum; fletcher_4_native_varsize(compressed, compressed_size, &cksum); size_t alloc_size = compressed_size * 2 + 1; str = kmem_alloc(alloc_size, KM_SLEEP); for (int i = 0; i < compressed_size; i++) { size_t offset = i * 2; (void) snprintf(str + offset, alloc_size - offset, "%02x", compressed[i]); } str[compressed_size * 2] = '\0'; char *propval = kmem_asprintf("%u-%llx-%llx-%s", ZFS_SEND_RESUME_TOKEN_VERSION, (longlong_t)cksum.zc_word[0], (longlong_t)packed_size, str); kmem_free(packed, packed_size); kmem_free(str, alloc_size); kmem_free(compressed, packed_size); return (propval); } return (kmem_strdup("")); } /* * Returns a string that represents the receive resume stats token of the * dataset's child. It should be freed with strfree(). */ char * get_child_receive_stats(dsl_dataset_t *ds) { char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; dsl_dataset_t *recv_ds; dsl_dataset_name(ds, recvname); if (strlcat(recvname, "/", sizeof (recvname)) < sizeof (recvname) && strlcat(recvname, recv_clone_name, sizeof (recvname)) < sizeof (recvname) && dsl_dataset_hold(ds->ds_dir->dd_pool, recvname, FTAG, &recv_ds) == 0) { char *propval = get_receive_resume_stats_impl(recv_ds); dsl_dataset_rele(recv_ds, FTAG); return (propval); } return (kmem_strdup("")); } static void get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv) { char *propval = get_receive_resume_stats_impl(ds); if (strcmp(propval, "") != 0) { dsl_prop_nvlist_add_string(nv, ZFS_PROP_RECEIVE_RESUME_TOKEN, propval); } else { char *childval = get_child_receive_stats(ds); if (strcmp(childval, "") != 0) { dsl_prop_nvlist_add_string(nv, ZFS_PROP_RECEIVE_RESUME_TOKEN, childval); } kmem_strfree(childval); } kmem_strfree(propval); } uint64_t dsl_get_refratio(dsl_dataset_t *ds) { uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / dsl_dataset_phys(ds)->ds_compressed_bytes); return (ratio); } uint64_t dsl_get_logicalreferenced(dsl_dataset_t *ds) { return (dsl_dataset_phys(ds)->ds_uncompressed_bytes); } uint64_t dsl_get_compressratio(dsl_dataset_t *ds) { if (ds->ds_is_snapshot) { return (dsl_get_refratio(ds)); } else { dsl_dir_t *dd = ds->ds_dir; mutex_enter(&dd->dd_lock); uint64_t val = dsl_dir_get_compressratio(dd); mutex_exit(&dd->dd_lock); return (val); } } uint64_t dsl_get_used(dsl_dataset_t *ds) { if (ds->ds_is_snapshot) { return (dsl_dataset_phys(ds)->ds_unique_bytes); } else { dsl_dir_t *dd = ds->ds_dir; mutex_enter(&dd->dd_lock); uint64_t val = dsl_dir_get_used(dd); mutex_exit(&dd->dd_lock); return (val); } } uint64_t dsl_get_creation(dsl_dataset_t *ds) { return (dsl_dataset_phys(ds)->ds_creation_time); } uint64_t dsl_get_creationtxg(dsl_dataset_t *ds) { return (dsl_dataset_phys(ds)->ds_creation_txg); } uint64_t dsl_get_refquota(dsl_dataset_t *ds) { return (ds->ds_quota); } uint64_t dsl_get_refreservation(dsl_dataset_t *ds) { return (ds->ds_reserved); } uint64_t dsl_get_guid(dsl_dataset_t *ds) { return (dsl_dataset_phys(ds)->ds_guid); } uint64_t dsl_get_unique(dsl_dataset_t *ds) { return (dsl_dataset_phys(ds)->ds_unique_bytes); } uint64_t dsl_get_objsetid(dsl_dataset_t *ds) { return (ds->ds_object); } uint64_t dsl_get_userrefs(dsl_dataset_t *ds) { return (ds->ds_userrefs); } uint64_t dsl_get_defer_destroy(dsl_dataset_t *ds) { return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0); } uint64_t dsl_get_referenced(dsl_dataset_t *ds) { return (dsl_dataset_phys(ds)->ds_referenced_bytes); } uint64_t dsl_get_numclones(dsl_dataset_t *ds) { ASSERT(ds->ds_is_snapshot); return (dsl_dataset_phys(ds)->ds_num_children - 1); } uint64_t dsl_get_inconsistent(dsl_dataset_t *ds) { return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ? 1 : 0); } uint64_t dsl_get_redacted(dsl_dataset_t *ds) { return (dsl_dataset_feature_is_active(ds, SPA_FEATURE_REDACTED_DATASETS)); } uint64_t dsl_get_available(dsl_dataset_t *ds) { uint64_t refdbytes = dsl_get_referenced(ds); uint64_t availbytes = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { availbytes += ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; } if (ds->ds_quota != 0) { /* * Adjust available bytes according to refquota */ if (refdbytes < ds->ds_quota) { availbytes = MIN(availbytes, ds->ds_quota - refdbytes); } else { availbytes = 0; } } return (availbytes); } int dsl_get_written(dsl_dataset_t *ds, uint64_t *written) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_dataset_t *prev; int err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); if (err == 0) { uint64_t comp, uncomp; err = dsl_dataset_space_written(prev, ds, written, &comp, &uncomp); dsl_dataset_rele(prev, FTAG); } return (err); } /* * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN. */ int dsl_get_prev_snap(dsl_dataset_t *ds, char *snap) { dsl_pool_t *dp = ds->ds_dir->dd_pool; if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { dsl_dataset_name(ds->ds_prev, snap); return (0); } else { return (SET_ERROR(ENOENT)); } } void dsl_get_redact_snaps(dsl_dataset_t *ds, nvlist_t *propval) { uint64_t nsnaps; uint64_t *snaps; if (dsl_dataset_get_uint64_array_feature(ds, SPA_FEATURE_REDACTED_DATASETS, &nsnaps, &snaps)) { fnvlist_add_uint64_array(propval, ZPROP_VALUE, snaps, nsnaps); } } /* * Returns the mountpoint property and source for the given dataset in the value * and source buffers. The value buffer must be at least as large as MAXPATHLEN * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN. * Returns 0 on success and an error on failure. */ int dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value, char *source) { int error; dsl_pool_t *dp = ds->ds_dir->dd_pool; /* Retrieve the mountpoint value stored in the zap object */ error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1, ZAP_MAXVALUELEN, value, source); if (error != 0) { return (error); } /* * Process the dsname and source to find the full mountpoint string. * Can be skipped for 'legacy' or 'none'. */ if (value[0] == '/') { char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP); char *root = buf; const char *relpath; /* * If we inherit the mountpoint, even from a dataset * with a received value, the source will be the path of * the dataset we inherit from. If source is * ZPROP_SOURCE_VAL_RECVD, the received value is not * inherited. */ if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) { relpath = ""; } else { ASSERT0(strncmp(dsname, source, strlen(source))); relpath = dsname + strlen(source); if (relpath[0] == '/') relpath++; } spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN); /* * Special case an alternate root of '/'. This will * avoid having multiple leading slashes in the * mountpoint path. */ if (strcmp(root, "/") == 0) root++; /* * If the mountpoint is '/' then skip over this * if we are obtaining either an alternate root or * an inherited mountpoint. */ char *mnt = value; if (value[1] == '\0' && (root[0] != '\0' || relpath[0] != '\0')) mnt = value + 1; if (relpath[0] == '\0') { (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s", root, mnt); } else { (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s", root, mnt, relpath[0] == '@' ? "" : "/", relpath); } kmem_free(buf, ZAP_MAXVALUELEN); } return (0); } void dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) { dsl_pool_t *dp = ds->ds_dir->dd_pool; ASSERT(dsl_pool_config_held(dp)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, dsl_get_refratio(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, dsl_get_logicalreferenced(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, dsl_get_compressratio(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, dsl_get_used(ds)); if (ds->ds_is_snapshot) { get_clones_stat(ds, nv); } else { char buf[ZFS_MAX_DATASET_NAME_LEN]; if (dsl_get_prev_snap(ds, buf) == 0) dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); dsl_dir_stats(ds->ds_dir, nv); } nvlist_t *propval = fnvlist_alloc(); dsl_get_redact_snaps(ds, propval); fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS), propval); nvlist_free(propval); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, dsl_get_available(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, dsl_get_referenced(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, dsl_get_creation(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, dsl_get_creationtxg(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, dsl_get_refquota(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, dsl_get_refreservation(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, dsl_get_guid(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, dsl_get_unique(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, dsl_get_objsetid(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, dsl_get_userrefs(ds)); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, dsl_get_defer_destroy(ds)); dsl_dataset_crypt_stats(ds, nv); if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { uint64_t written; if (dsl_get_written(ds, &written) == 0) { dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, written); } } if (!dsl_dataset_is_snapshot(ds)) { /* * A failed "newfs" (e.g. full) resumable receive leaves * the stats set on this dataset. Check here for the prop. */ get_receive_resume_stats(ds, nv); /* * A failed incremental resumable receive leaves the * stats set on our child named "%recv". Check the child * for the prop. */ /* 6 extra bytes for /%recv */ char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; dsl_dataset_t *recv_ds; dsl_dataset_name(ds, recvname); if (strlcat(recvname, "/", sizeof (recvname)) < sizeof (recvname) && strlcat(recvname, recv_clone_name, sizeof (recvname)) < sizeof (recvname) && dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) { get_receive_resume_stats(recv_ds, nv); dsl_dataset_rele(recv_ds, FTAG); } } } void dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) { dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool; ASSERT(dsl_pool_config_held(dp)); stat->dds_creation_txg = dsl_get_creationtxg(ds); stat->dds_inconsistent = dsl_get_inconsistent(ds); stat->dds_guid = dsl_get_guid(ds); stat->dds_redacted = dsl_get_redacted(ds); stat->dds_origin[0] = '\0'; if (ds->ds_is_snapshot) { stat->dds_is_snapshot = B_TRUE; stat->dds_num_clones = dsl_get_numclones(ds); } else { stat->dds_is_snapshot = B_FALSE; stat->dds_num_clones = 0; if (dsl_dir_is_clone(ds->ds_dir)) { dsl_dir_get_origin(ds->ds_dir, stat->dds_origin); } } } uint64_t dsl_dataset_fsid_guid(dsl_dataset_t *ds) { return (ds->ds_fsid_guid); } void dsl_dataset_space(dsl_dataset_t *ds, uint64_t *refdbytesp, uint64_t *availbytesp, uint64_t *usedobjsp, uint64_t *availobjsp) { *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) *availbytesp += ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; if (ds->ds_quota != 0) { /* * Adjust available bytes according to refquota */ if (*refdbytesp < ds->ds_quota) *availbytesp = MIN(*availbytesp, ds->ds_quota - *refdbytesp); else *availbytesp = 0; } rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); rrw_exit(&ds->ds_bp_rwlock, FTAG); *availobjsp = DN_MAX_OBJECT - *usedobjsp; } boolean_t dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) { dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool; uint64_t birth; ASSERT(dsl_pool_config_held(dp)); if (snap == NULL) return (B_FALSE); rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); birth = dsl_dataset_get_blkptr(ds)->blk_birth; rrw_exit(&ds->ds_bp_rwlock, FTAG); if (birth > dsl_dataset_phys(snap)->ds_creation_txg) { objset_t *os, *os_snap; /* * It may be that only the ZIL differs, because it was * reset in the head. Don't count that as being * modified. */ if (dmu_objset_from_ds(ds, &os) != 0) return (B_TRUE); if (dmu_objset_from_ds(snap, &os_snap) != 0) return (B_TRUE); return (bcmp(&os->os_phys->os_meta_dnode, &os_snap->os_phys->os_meta_dnode, sizeof (os->os_phys->os_meta_dnode)) != 0); } return (B_FALSE); } typedef struct dsl_dataset_rename_snapshot_arg { const char *ddrsa_fsname; const char *ddrsa_oldsnapname; const char *ddrsa_newsnapname; boolean_t ddrsa_recursive; dmu_tx_t *ddrsa_tx; } dsl_dataset_rename_snapshot_arg_t; static int dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { (void) dp; dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; int error; uint64_t val; error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); if (error != 0) { /* ignore nonexistent snapshots */ return (error == ENOENT ? 0 : error); } /* new name should not exist */ error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); if (error == 0) error = SET_ERROR(EEXIST); else if (error == ENOENT) error = 0; /* dataset name + 1 for the "@" + the new snapshot name must fit */ if (dsl_dir_namelen(hds->ds_dir) + 1 + strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN) error = SET_ERROR(ENAMETOOLONG); return (error); } static int dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) { dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *hds; int error; error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); if (error != 0) return (error); if (ddrsa->ddrsa_recursive) { error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, dsl_dataset_rename_snapshot_check_impl, ddrsa, DS_FIND_CHILDREN); } else { error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); } dsl_dataset_rele(hds, FTAG); return (error); } static int dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; dsl_dataset_t *ds; uint64_t val; dmu_tx_t *tx = ddrsa->ddrsa_tx; int error; error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); ASSERT(error == 0 || error == ENOENT); if (error == ENOENT) { /* ignore nonexistent snapshots */ return (0); } VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); /* log before we change the name */ spa_history_log_internal_ds(ds, "rename", tx, "-> @%s", ddrsa->ddrsa_newsnapname); VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, B_FALSE)); mutex_enter(&ds->ds_lock); (void) strlcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname, sizeof (ds->ds_snapname)); mutex_exit(&ds->ds_lock); VERIFY0(zap_add(dp->dp_meta_objset, dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 8, 1, &ds->ds_object, tx)); zvol_rename_minors(dp->dp_spa, ddrsa->ddrsa_oldsnapname, ddrsa->ddrsa_newsnapname, B_TRUE); dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *hds = NULL; VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); ddrsa->ddrsa_tx = tx; if (ddrsa->ddrsa_recursive) { VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, dsl_dataset_rename_snapshot_sync_impl, ddrsa, DS_FIND_CHILDREN)); } else { VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); } dsl_dataset_rele(hds, FTAG); } int dsl_dataset_rename_snapshot(const char *fsname, const char *oldsnapname, const char *newsnapname, boolean_t recursive) { dsl_dataset_rename_snapshot_arg_t ddrsa; ddrsa.ddrsa_fsname = fsname; ddrsa.ddrsa_oldsnapname = oldsnapname; ddrsa.ddrsa_newsnapname = newsnapname; ddrsa.ddrsa_recursive = recursive; return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, dsl_dataset_rename_snapshot_sync, &ddrsa, 1, ZFS_SPACE_CHECK_RESERVED)); } /* * If we're doing an ownership handoff, we need to make sure that there is * only one long hold on the dataset. We're not allowed to change anything here * so we don't permanently release the long hold or regular hold here. We want * to do this only when syncing to avoid the dataset unexpectedly going away * when we release the long hold. */ static int dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) { boolean_t held = B_FALSE; if (!dmu_tx_is_syncing(tx)) return (0); dsl_dir_t *dd = ds->ds_dir; mutex_enter(&dd->dd_activity_lock); uint64_t holds = zfs_refcount_count(&ds->ds_longholds) - (owner != NULL ? 1 : 0); /* * The value of dd_activity_waiters can chance as soon as we drop the * lock, but we're fine with that; new waiters coming in or old * waiters leaving doesn't cause problems, since we're going to cancel * waiters later anyway. The goal of this check is to verify that no * non-waiters have long-holds, and all new long-holds will be * prevented because we're holding the pool config as writer. */ if (holds != dd->dd_activity_waiters) held = B_TRUE; mutex_exit(&dd->dd_activity_lock); if (held) return (SET_ERROR(EBUSY)); return (0); } int dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) { dsl_dataset_rollback_arg_t *ddra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; int64_t unused_refres_delta; int error; error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); if (error != 0) return (error); /* must not be a snapshot */ if (ds->ds_is_snapshot) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } /* must have a most recent snapshot */ if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(ESRCH)); } /* * No rollback to a snapshot created in the current txg, because * the rollback may dirty the dataset and create blocks that are * not reachable from the rootbp while having a birth txg that * falls into the snapshot's range. */ if (dmu_tx_is_syncing(tx) && dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EAGAIN)); } /* * If the expected target snapshot is specified, then check that * the latest snapshot is it. */ if (ddra->ddra_tosnap != NULL) { dsl_dataset_t *snapds; /* Check if the target snapshot exists at all. */ error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds); if (error != 0) { /* * ESRCH is used to signal that the target snapshot does * not exist, while ENOENT is used to report that * the rolled back dataset does not exist. * ESRCH is also used to cover other cases where the * target snapshot is not related to the dataset being * rolled back such as being in a different pool. */ if (error == ENOENT || error == EXDEV) error = SET_ERROR(ESRCH); dsl_dataset_rele(ds, FTAG); return (error); } ASSERT(snapds->ds_is_snapshot); /* Check if the snapshot is the latest snapshot indeed. */ if (snapds != ds->ds_prev) { /* * Distinguish between the case where the only problem * is intervening snapshots (EEXIST) vs the snapshot * not being a valid target for rollback (ESRCH). */ if (snapds->ds_dir == ds->ds_dir || (dsl_dir_is_clone(ds->ds_dir) && dsl_dir_phys(ds->ds_dir)->dd_origin_obj == snapds->ds_object)) { error = SET_ERROR(EEXIST); } else { error = SET_ERROR(ESRCH); } dsl_dataset_rele(snapds, FTAG); dsl_dataset_rele(ds, FTAG); return (error); } dsl_dataset_rele(snapds, FTAG); } /* must not have any bookmarks after the most recent snapshot */ if (dsl_bookmark_latest_txg(ds) > dsl_dataset_phys(ds)->ds_prev_snap_txg) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EEXIST)); } error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } /* * Check if the snap we are rolling back to uses more than * the refquota. */ if (ds->ds_quota != 0 && dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EDQUOT)); } /* * When we do the clone swap, we will temporarily use more space * due to the refreservation (the head will no longer have any * unique space, so the entire amount of the refreservation will need * to be free). We will immediately destroy the clone, freeing * this space, but the freeing happens over many txg's. */ unused_refres_delta = (int64_t)MIN(ds->ds_reserved, dsl_dataset_phys(ds)->ds_unique_bytes); if (unused_refres_delta > 0 && unused_refres_delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(ENOSPC)); } dsl_dataset_rele(ds, FTAG); return (0); } void dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_rollback_arg_t *ddra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds, *clone; uint64_t cloneobj; char namebuf[ZFS_MAX_DATASET_NAME_LEN]; VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); dsl_dataset_name(ds->ds_prev, namebuf); fnvlist_add_string(ddra->ddra_result, "target", namebuf); cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, NULL, tx); VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); dsl_dataset_clone_swap_sync_impl(clone, ds, tx); dsl_dataset_zero_zil(ds, tx); dsl_destroy_head_sync_impl(clone, tx); dsl_dataset_rele(clone, FTAG); dsl_dataset_rele(ds, FTAG); } /* * Rolls back the given filesystem or volume to the most recent snapshot. * The name of the most recent snapshot will be returned under key "target" * in the result nvlist. * * If owner != NULL: * - The existing dataset MUST be owned by the specified owner at entry * - Upon return, dataset will still be held by the same owner, whether we * succeed or not. * * This mode is required any time the existing filesystem is mounted. See * notes above zfs_suspend_fs() for further details. */ int dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner, nvlist_t *result) { dsl_dataset_rollback_arg_t ddra; ddra.ddra_fsname = fsname; ddra.ddra_tosnap = tosnap; ddra.ddra_owner = owner; ddra.ddra_result = result; return (dsl_sync_task(fsname, dsl_dataset_rollback_check, dsl_dataset_rollback_sync, &ddra, 1, ZFS_SPACE_CHECK_RESERVED)); } struct promotenode { list_node_t link; dsl_dataset_t *ds; }; static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag); static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); int dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) { dsl_dataset_promote_arg_t *ddpa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *hds; struct promotenode *snap; int err; uint64_t unused; uint64_t ss_mv_cnt; size_t max_snap_len; boolean_t conflicting_snaps; err = promote_hold(ddpa, dp, FTAG); if (err != 0) return (err); hds = ddpa->ddpa_clone; max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { promote_rele(ddpa, FTAG); return (SET_ERROR(EXDEV)); } snap = list_head(&ddpa->shared_snaps); if (snap == NULL) { err = SET_ERROR(ENOENT); goto out; } dsl_dataset_t *const origin_ds = snap->ds; /* * Encrypted clones share a DSL Crypto Key with their origin's dsl dir. * When doing a promote we must make sure the encryption root for * both the target and the target's origin does not change to avoid * needing to rewrap encryption keys */ err = dsl_dataset_promote_crypt_check(hds->ds_dir, origin_ds->ds_dir); if (err != 0) goto out; /* * Compute and check the amount of space to transfer. Since this is * so expensive, don't do the preliminary check. */ if (!dmu_tx_is_syncing(tx)) { promote_rele(ddpa, FTAG); return (0); } /* compute origin's new unique space */ snap = list_tail(&ddpa->clone_snaps); ASSERT(snap != NULL); ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, origin_ds->ds_object); dsl_deadlist_space_range(&snap->ds->ds_deadlist, dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, &ddpa->unique, &unused, &unused); /* * Walk the snapshots that we are moving * * Compute space to transfer. Consider the incremental changes * to used by each snapshot: * (my used) = (prev's used) + (blocks born) - (blocks killed) * So each snapshot gave birth to: * (blocks born) = (my used) - (prev's used) + (blocks killed) * So a sequence would look like: * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) * Which simplifies to: * uN + kN + kN-1 + ... + k1 + k0 * Note however, if we stop before we reach the ORIGIN we get: * uN + kN + kN-1 + ... + kM - uM-1 */ conflicting_snaps = B_FALSE; ss_mv_cnt = 0; ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; for (snap = list_head(&ddpa->shared_snaps); snap; snap = list_next(&ddpa->shared_snaps, snap)) { uint64_t val, dlused, dlcomp, dluncomp; dsl_dataset_t *ds = snap->ds; ss_mv_cnt++; /* * If there are long holds, we won't be able to evict * the objset. */ if (dsl_dataset_long_held(ds)) { err = SET_ERROR(EBUSY); goto out; } /* Check that the snapshot name does not conflict */ VERIFY0(dsl_dataset_get_snapname(ds)); if (strlen(ds->ds_snapname) >= max_snap_len) { err = SET_ERROR(ENAMETOOLONG); goto out; } err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); if (err == 0) { fnvlist_add_boolean(ddpa->err_ds, snap->ds->ds_snapname); conflicting_snaps = B_TRUE; } else if (err != ENOENT) { goto out; } /* The very first snapshot does not have a deadlist */ if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) continue; dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); ddpa->used += dlused; ddpa->comp += dlcomp; ddpa->uncomp += dluncomp; } /* * Check that bookmarks that are being transferred don't have * name conflicts. */ for (dsl_bookmark_node_t *dbn = avl_first(&origin_ds->ds_bookmarks); dbn != NULL && dbn->dbn_phys.zbm_creation_txg <= dsl_dataset_phys(origin_ds)->ds_creation_txg; dbn = AVL_NEXT(&origin_ds->ds_bookmarks, dbn)) { if (strlen(dbn->dbn_name) >= max_snap_len) { err = SET_ERROR(ENAMETOOLONG); goto out; } zfs_bookmark_phys_t bm; err = dsl_bookmark_lookup_impl(ddpa->ddpa_clone, dbn->dbn_name, &bm); if (err == 0) { fnvlist_add_boolean(ddpa->err_ds, dbn->dbn_name); conflicting_snaps = B_TRUE; } else if (err == ESRCH) { err = 0; } else if (err != 0) { goto out; } } /* * In order to return the full list of conflicting snapshots, we check * whether there was a conflict after traversing all of them. */ if (conflicting_snaps) { err = SET_ERROR(EEXIST); goto out; } /* * If we are a clone of a clone then we never reached ORIGIN, * so we need to subtract out the clone origin's used space. */ if (ddpa->origin_origin) { ddpa->used -= dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; ddpa->comp -= dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; ddpa->uncomp -= dsl_dataset_phys(ddpa->origin_origin)-> ds_uncompressed_bytes; } /* Check that there is enough space and limit headroom here */ err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 0, ss_mv_cnt, ddpa->used, ddpa->cr, ddpa->proc); if (err != 0) goto out; /* * Compute the amounts of space that will be used by snapshots * after the promotion (for both origin and clone). For each, * it is the amount of space that will be on all of their * deadlists (that was not born before their new origin). */ if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { uint64_t space; /* * Note, typically this will not be a clone of a clone, * so dd_origin_txg will be < TXG_INITIAL, so * these snaplist_space() -> dsl_deadlist_space_range() * calls will be fast because they do not have to * iterate over all bps. */ snap = list_head(&ddpa->origin_snaps); if (snap == NULL) { err = SET_ERROR(ENOENT); goto out; } err = snaplist_space(&ddpa->shared_snaps, snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); if (err != 0) goto out; err = snaplist_space(&ddpa->clone_snaps, snap->ds->ds_dir->dd_origin_txg, &space); if (err != 0) goto out; ddpa->cloneusedsnap += space; } if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { err = snaplist_space(&ddpa->origin_snaps, dsl_dataset_phys(origin_ds)->ds_creation_txg, &ddpa->originusedsnap); if (err != 0) goto out; } out: promote_rele(ddpa, FTAG); return (err); } void dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_promote_arg_t *ddpa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *hds; struct promotenode *snap; dsl_dataset_t *origin_ds; dsl_dataset_t *origin_head; dsl_dir_t *dd; dsl_dir_t *odd = NULL; uint64_t oldnext_obj; int64_t delta; ASSERT(nvlist_empty(ddpa->err_ds)); VERIFY0(promote_hold(ddpa, dp, FTAG)); hds = ddpa->ddpa_clone; ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); snap = list_head(&ddpa->shared_snaps); origin_ds = snap->ds; dd = hds->ds_dir; snap = list_head(&ddpa->origin_snaps); origin_head = snap->ds; /* * We need to explicitly open odd, since origin_ds's dd will be * changing. */ VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, NULL, FTAG, &odd)); dsl_dataset_promote_crypt_sync(hds->ds_dir, odd, tx); /* change origin's next snap */ dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; snap = list_tail(&ddpa->clone_snaps); ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, origin_ds->ds_object); dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; /* change the origin's next clone */ if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { dsl_dataset_remove_from_next_clones(origin_ds, snap->ds->ds_object, tx); VERIFY0(zap_add_int(dp->dp_meta_objset, dsl_dataset_phys(origin_ds)->ds_next_clones_obj, oldnext_obj, tx)); } /* change origin */ dmu_buf_will_dirty(dd->dd_dbuf, tx); ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; dmu_buf_will_dirty(odd->dd_dbuf, tx); dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; origin_head->ds_dir->dd_origin_txg = dsl_dataset_phys(origin_ds)->ds_creation_txg; /* change dd_clone entries */ if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { VERIFY0(zap_remove_int(dp->dp_meta_objset, dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); VERIFY0(zap_add_int(dp->dp_meta_objset, dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, hds->ds_object, tx)); VERIFY0(zap_remove_int(dp->dp_meta_objset, dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, origin_head->ds_object, tx)); if (dsl_dir_phys(dd)->dd_clones == 0) { dsl_dir_phys(dd)->dd_clones = zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); } VERIFY0(zap_add_int(dp->dp_meta_objset, dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); } /* * Move bookmarks to this dir. */ dsl_bookmark_node_t *dbn_next; for (dsl_bookmark_node_t *dbn = avl_first(&origin_head->ds_bookmarks); dbn != NULL && dbn->dbn_phys.zbm_creation_txg <= dsl_dataset_phys(origin_ds)->ds_creation_txg; dbn = dbn_next) { dbn_next = AVL_NEXT(&origin_head->ds_bookmarks, dbn); avl_remove(&origin_head->ds_bookmarks, dbn); VERIFY0(zap_remove(dp->dp_meta_objset, origin_head->ds_bookmarks_obj, dbn->dbn_name, tx)); dsl_bookmark_node_add(hds, dbn, tx); } dsl_bookmark_next_changed(hds, origin_ds, tx); /* move snapshots to this dir */ for (snap = list_head(&ddpa->shared_snaps); snap; snap = list_next(&ddpa->shared_snaps, snap)) { dsl_dataset_t *ds = snap->ds; /* * Property callbacks are registered to a particular * dsl_dir. Since ours is changing, evict the objset * so that they will be unregistered from the old dsl_dir. */ if (ds->ds_objset) { dmu_objset_evict(ds->ds_objset); ds->ds_objset = NULL; } /* move snap name entry */ VERIFY0(dsl_dataset_get_snapname(ds)); VERIFY0(dsl_dataset_snap_remove(origin_head, ds->ds_snapname, tx, B_TRUE)); VERIFY0(zap_add(dp->dp_meta_objset, dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 8, 1, &ds->ds_object, tx)); dsl_fs_ss_count_adjust(hds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); /* change containing dsl_dir */ dmu_buf_will_dirty(ds->ds_dbuf, tx); ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; ASSERT3P(ds->ds_dir, ==, odd); dsl_dir_rele(ds->ds_dir, ds); VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, NULL, ds, &ds->ds_dir)); /* move any clone references */ if (dsl_dataset_phys(ds)->ds_next_clones_obj && spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { zap_cursor_t zc; zap_attribute_t za; for (zap_cursor_init(&zc, dp->dp_meta_objset, dsl_dataset_phys(ds)->ds_next_clones_obj); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { dsl_dataset_t *cnds; uint64_t o; if (za.za_first_integer == oldnext_obj) { /* * We've already moved the * origin's reference. */ continue; } VERIFY0(dsl_dataset_hold_obj(dp, za.za_first_integer, FTAG, &cnds)); o = dsl_dir_phys(cnds->ds_dir)-> dd_head_dataset_obj; VERIFY0(zap_remove_int(dp->dp_meta_objset, dsl_dir_phys(odd)->dd_clones, o, tx)); VERIFY0(zap_add_int(dp->dp_meta_objset, dsl_dir_phys(dd)->dd_clones, o, tx)); dsl_dataset_rele(cnds, FTAG); } zap_cursor_fini(&zc); } ASSERT(!dsl_prop_hascb(ds)); } /* * Change space accounting. * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either * both be valid, or both be 0 (resulting in delta == 0). This * is true for each of {clone,origin} independently. */ delta = ddpa->cloneusedsnap - dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; ASSERT3S(delta, >=, 0); ASSERT3U(ddpa->used, >=, delta); dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); dsl_dir_diduse_space(dd, DD_USED_HEAD, ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); delta = ddpa->originusedsnap - dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; ASSERT3S(delta, <=, 0); ASSERT3U(ddpa->used, >=, -delta); dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); dsl_dir_diduse_space(odd, DD_USED_HEAD, -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; /* * Since livelists are specific to a clone's origin txg, they * are no longer accurate. Destroy the livelist from the clone being * promoted. If the origin dataset is a clone, destroy its livelist * as well. */ dsl_dir_remove_livelist(dd, tx, B_TRUE); dsl_dir_remove_livelist(odd, tx, B_TRUE); /* log history record */ spa_history_log_internal_ds(hds, "promote", tx, " "); dsl_dir_rele(odd, FTAG); promote_rele(ddpa, FTAG); } /* * Make a list of dsl_dataset_t's for the snapshots between first_obj * (exclusive) and last_obj (inclusive). The list will be in reverse * order (last_obj will be the list_head()). If first_obj == 0, do all * snapshots back to this dataset's origin. */ static int snaplist_make(dsl_pool_t *dp, uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) { uint64_t obj = last_obj; list_create(l, sizeof (struct promotenode), offsetof(struct promotenode, link)); while (obj != first_obj) { dsl_dataset_t *ds; struct promotenode *snap; int err; err = dsl_dataset_hold_obj(dp, obj, tag, &ds); ASSERT(err != ENOENT); if (err != 0) return (err); if (first_obj == 0) first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; snap = kmem_alloc(sizeof (*snap), KM_SLEEP); snap->ds = ds; list_insert_tail(l, snap); obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; } return (0); } static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) { struct promotenode *snap; *spacep = 0; for (snap = list_head(l); snap; snap = list_next(l, snap)) { uint64_t used, comp, uncomp; dsl_deadlist_space_range(&snap->ds->ds_deadlist, mintxg, UINT64_MAX, &used, &comp, &uncomp); *spacep += used; } return (0); } static void snaplist_destroy(list_t *l, void *tag) { struct promotenode *snap; if (l == NULL || !list_link_active(&l->list_head)) return; while ((snap = list_tail(l)) != NULL) { list_remove(l, snap); dsl_dataset_rele(snap->ds, tag); kmem_free(snap, sizeof (*snap)); } list_destroy(l); } static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) { int error; dsl_dir_t *dd; struct promotenode *snap; error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, &ddpa->ddpa_clone); if (error != 0) return (error); dd = ddpa->ddpa_clone->ds_dir; if (ddpa->ddpa_clone->ds_is_snapshot || !dsl_dir_is_clone(dd)) { dsl_dataset_rele(ddpa->ddpa_clone, tag); return (SET_ERROR(EINVAL)); } error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, &ddpa->shared_snaps, tag); if (error != 0) goto out; error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, &ddpa->clone_snaps, tag); if (error != 0) goto out; snap = list_head(&ddpa->shared_snaps); ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, &ddpa->origin_snaps, tag); if (error != 0) goto out; if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { error = dsl_dataset_hold_obj(dp, dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, tag, &ddpa->origin_origin); if (error != 0) goto out; } out: if (error != 0) promote_rele(ddpa, tag); return (error); } static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) { snaplist_destroy(&ddpa->shared_snaps, tag); snaplist_destroy(&ddpa->clone_snaps, tag); snaplist_destroy(&ddpa->origin_snaps, tag); if (ddpa->origin_origin != NULL) dsl_dataset_rele(ddpa->origin_origin, tag); dsl_dataset_rele(ddpa->ddpa_clone, tag); } /* * Promote a clone. * * If it fails due to a conflicting snapshot name, "conflsnap" will be filled * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.) */ int dsl_dataset_promote(const char *name, char *conflsnap) { dsl_dataset_promote_arg_t ddpa = { 0 }; uint64_t numsnaps; int error; nvpair_t *snap_pair; objset_t *os; /* * We will modify space proportional to the number of * snapshots. Compute numsnaps. */ error = dmu_objset_hold(name, FTAG, &os); if (error != 0) return (error); error = zap_count(dmu_objset_pool(os)->dp_meta_objset, dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, &numsnaps); dmu_objset_rele(os, FTAG); if (error != 0) return (error); ddpa.ddpa_clonename = name; ddpa.err_ds = fnvlist_alloc(); ddpa.cr = CRED(); ddpa.proc = curproc; error = dsl_sync_task(name, dsl_dataset_promote_check, dsl_dataset_promote_sync, &ddpa, 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED); /* * Return the first conflicting snapshot found. */ snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL); if (snap_pair != NULL && conflsnap != NULL) (void) strlcpy(conflsnap, nvpair_name(snap_pair), ZFS_MAX_DATASET_NAME_LEN); fnvlist_free(ddpa.err_ds); return (error); } int dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) { /* * "slack" factor for received datasets with refquota set on them. * See the bottom of this function for details on its use. */ uint64_t refquota_slack = (uint64_t)DMU_MAX_ACCESS * spa_asize_inflation; int64_t unused_refres_delta; /* they should both be heads */ if (clone->ds_is_snapshot || origin_head->ds_is_snapshot) return (SET_ERROR(EINVAL)); /* if we are not forcing, the branch point should be just before them */ if (!force && clone->ds_prev != origin_head->ds_prev) return (SET_ERROR(EINVAL)); /* clone should be the clone (unless they are unrelated) */ if (clone->ds_prev != NULL && clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && origin_head->ds_dir != clone->ds_prev->ds_dir) return (SET_ERROR(EINVAL)); /* the clone should be a child of the origin */ if (clone->ds_dir->dd_parent != origin_head->ds_dir) return (SET_ERROR(EINVAL)); /* origin_head shouldn't be modified unless 'force' */ if (!force && dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) return (SET_ERROR(ETXTBSY)); /* origin_head should have no long holds (e.g. is not mounted) */ if (dsl_dataset_handoff_check(origin_head, owner, tx)) return (SET_ERROR(EBUSY)); /* check amount of any unconsumed refreservation */ unused_refres_delta = (int64_t)MIN(origin_head->ds_reserved, dsl_dataset_phys(origin_head)->ds_unique_bytes) - (int64_t)MIN(origin_head->ds_reserved, dsl_dataset_phys(clone)->ds_unique_bytes); if (unused_refres_delta > 0 && unused_refres_delta > dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) return (SET_ERROR(ENOSPC)); /* * The clone can't be too much over the head's refquota. * * To ensure that the entire refquota can be used, we allow one * transaction to exceed the refquota. Therefore, this check * needs to also allow for the space referenced to be more than the * refquota. The maximum amount of space that one transaction can use * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this * overage ensures that we are able to receive a filesystem that * exceeds the refquota on the source system. * * So that overage is the refquota_slack we use below. */ if (origin_head->ds_quota != 0 && dsl_dataset_phys(clone)->ds_referenced_bytes > origin_head->ds_quota + refquota_slack) return (SET_ERROR(EDQUOT)); return (0); } static void dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone, dsl_dataset_t *origin, dmu_tx_t *tx) { uint64_t clone_remap_dl_obj, origin_remap_dl_obj; dsl_pool_t *dp = dmu_tx_pool(tx); ASSERT(dsl_pool_sync_context(dp)); clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone); origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin); if (clone_remap_dl_obj != 0) { dsl_deadlist_close(&clone->ds_remap_deadlist); dsl_dataset_unset_remap_deadlist_object(clone, tx); } if (origin_remap_dl_obj != 0) { dsl_deadlist_close(&origin->ds_remap_deadlist); dsl_dataset_unset_remap_deadlist_object(origin, tx); } if (clone_remap_dl_obj != 0) { dsl_dataset_set_remap_deadlist_object(origin, clone_remap_dl_obj, tx); dsl_deadlist_open(&origin->ds_remap_deadlist, dp->dp_meta_objset, clone_remap_dl_obj); } if (origin_remap_dl_obj != 0) { dsl_dataset_set_remap_deadlist_object(clone, origin_remap_dl_obj, tx); dsl_deadlist_open(&clone->ds_remap_deadlist, dp->dp_meta_objset, origin_remap_dl_obj); } } void dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, dsl_dataset_t *origin_head, dmu_tx_t *tx) { dsl_pool_t *dp = dmu_tx_pool(tx); int64_t unused_refres_delta; ASSERT(clone->ds_reserved == 0); /* * NOTE: On DEBUG kernels there could be a race between this and * the check function if spa_asize_inflation is adjusted... */ ASSERT(origin_head->ds_quota == 0 || dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota + DMU_MAX_ACCESS * spa_asize_inflation); ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); dsl_dir_cancel_waiters(origin_head->ds_dir); /* * Swap per-dataset feature flags. */ for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (!(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET)) { ASSERT(!dsl_dataset_feature_is_active(clone, f)); ASSERT(!dsl_dataset_feature_is_active(origin_head, f)); continue; } boolean_t clone_inuse = dsl_dataset_feature_is_active(clone, f); void *clone_feature = clone->ds_feature[f]; boolean_t origin_head_inuse = dsl_dataset_feature_is_active(origin_head, f); void *origin_head_feature = origin_head->ds_feature[f]; if (clone_inuse) dsl_dataset_deactivate_feature_impl(clone, f, tx); if (origin_head_inuse) dsl_dataset_deactivate_feature_impl(origin_head, f, tx); if (clone_inuse) { dsl_dataset_activate_feature(origin_head->ds_object, f, clone_feature, tx); origin_head->ds_feature[f] = clone_feature; } if (origin_head_inuse) { dsl_dataset_activate_feature(clone->ds_object, f, origin_head_feature, tx); clone->ds_feature[f] = origin_head_feature; } } dmu_buf_will_dirty(clone->ds_dbuf, tx); dmu_buf_will_dirty(origin_head->ds_dbuf, tx); if (clone->ds_objset != NULL) { dmu_objset_evict(clone->ds_objset); clone->ds_objset = NULL; } if (origin_head->ds_objset != NULL) { dmu_objset_evict(origin_head->ds_objset); origin_head->ds_objset = NULL; } unused_refres_delta = (int64_t)MIN(origin_head->ds_reserved, dsl_dataset_phys(origin_head)->ds_unique_bytes) - (int64_t)MIN(origin_head->ds_reserved, dsl_dataset_phys(clone)->ds_unique_bytes); /* * Reset origin's unique bytes. */ { dsl_dataset_t *origin = clone->ds_prev; uint64_t comp, uncomp; dmu_buf_will_dirty(origin->ds_dbuf, tx); dsl_deadlist_space_range(&clone->ds_deadlist, dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); } /* swap blkptrs */ { rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG); rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG); blkptr_t tmp; tmp = dsl_dataset_phys(origin_head)->ds_bp; dsl_dataset_phys(origin_head)->ds_bp = dsl_dataset_phys(clone)->ds_bp; dsl_dataset_phys(clone)->ds_bp = tmp; rrw_exit(&origin_head->ds_bp_rwlock, FTAG); rrw_exit(&clone->ds_bp_rwlock, FTAG); } /* set dd_*_bytes */ { int64_t dused, dcomp, duncomp; uint64_t cdl_used, cdl_comp, cdl_uncomp; uint64_t odl_used, odl_comp, odl_uncomp; ASSERT3U(dsl_dir_phys(clone->ds_dir)-> dd_used_breakdown[DD_USED_SNAP], ==, 0); dsl_deadlist_space(&clone->ds_deadlist, &cdl_used, &cdl_comp, &cdl_uncomp); dsl_deadlist_space(&origin_head->ds_deadlist, &odl_used, &odl_comp, &odl_uncomp); dused = dsl_dataset_phys(clone)->ds_referenced_bytes + cdl_used - (dsl_dataset_phys(origin_head)->ds_referenced_bytes + odl_used); dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + cdl_comp - (dsl_dataset_phys(origin_head)->ds_compressed_bytes + odl_comp); duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + cdl_uncomp - (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + odl_uncomp); dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, dused, dcomp, duncomp, tx); dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, -dused, -dcomp, -duncomp, tx); /* * The difference in the space used by snapshots is the * difference in snapshot space due to the head's * deadlist (since that's the only thing that's * changing that affects the snapused). */ dsl_deadlist_space_range(&clone->ds_deadlist, origin_head->ds_dir->dd_origin_txg, UINT64_MAX, &cdl_used, &cdl_comp, &cdl_uncomp); dsl_deadlist_space_range(&origin_head->ds_deadlist, origin_head->ds_dir->dd_origin_txg, UINT64_MAX, &odl_used, &odl_comp, &odl_uncomp); dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, DD_USED_HEAD, DD_USED_SNAP, tx); } /* swap ds_*_bytes */ SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, dsl_dataset_phys(clone)->ds_referenced_bytes); SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, dsl_dataset_phys(clone)->ds_compressed_bytes); SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, dsl_dataset_phys(clone)->ds_uncompressed_bytes); SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, dsl_dataset_phys(clone)->ds_unique_bytes); /* apply any parent delta for change in unconsumed refreservation */ dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, unused_refres_delta, 0, 0, tx); /* * Swap deadlists. */ dsl_deadlist_close(&clone->ds_deadlist); dsl_deadlist_close(&origin_head->ds_deadlist); SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, dsl_dataset_phys(clone)->ds_deadlist_obj); dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, dsl_dataset_phys(clone)->ds_deadlist_obj); dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, dsl_dataset_phys(origin_head)->ds_deadlist_obj); dsl_dataset_swap_remap_deadlists(clone, origin_head, tx); /* * If there is a bookmark at the origin, its "next dataset" is * changing, so we need to reset its FBN. */ dsl_bookmark_next_changed(origin_head, origin_head->ds_prev, tx); dsl_scan_ds_clone_swapped(origin_head, clone, tx); /* * Destroy any livelists associated with the clone or the origin, * since after the swap the corresponding livelists are no longer * valid. */ dsl_dir_remove_livelist(clone->ds_dir, tx, B_TRUE); dsl_dir_remove_livelist(origin_head->ds_dir, tx, B_TRUE); spa_history_log_internal_ds(clone, "clone swap", tx, "parent=%s", origin_head->ds_dir->dd_myname); } /* * Given a pool name and a dataset object number in that pool, * return the name of that dataset. */ int dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) { dsl_pool_t *dp; dsl_dataset_t *ds; int error; error = dsl_pool_hold(pname, FTAG, &dp); if (error != 0) return (error); error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); if (error == 0) { dsl_dataset_name(ds, buf); dsl_dataset_rele(ds, FTAG); } dsl_pool_rele(dp, FTAG); return (error); } int dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) { int error = 0; ASSERT3S(asize, >, 0); /* * *ref_rsrv is the portion of asize that will come from any * unconsumed refreservation space. */ *ref_rsrv = 0; mutex_enter(&ds->ds_lock); /* * Make a space adjustment for reserved bytes. */ if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { ASSERT3U(*used, >=, ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); *used -= (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); *ref_rsrv = asize - MIN(asize, parent_delta(ds, asize + inflight)); } if (!check_quota || ds->ds_quota == 0) { mutex_exit(&ds->ds_lock); return (0); } /* * If they are requesting more space, and our current estimate * is over quota, they get to try again unless the actual * on-disk is over quota and there are no pending changes (which * may free up space for us). */ if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= ds->ds_quota) { if (inflight > 0 || dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) error = SET_ERROR(ERESTART); else error = SET_ERROR(EDQUOT); } mutex_exit(&ds->ds_lock); return (error); } typedef struct dsl_dataset_set_qr_arg { const char *ddsqra_name; zprop_source_t ddsqra_source; uint64_t ddsqra_value; } dsl_dataset_set_qr_arg_t; static int dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) { dsl_dataset_set_qr_arg_t *ddsqra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; int error; uint64_t newval; if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) return (SET_ERROR(ENOTSUP)); error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); if (error != 0) return (error); if (ds->ds_is_snapshot) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } error = dsl_prop_predict(ds->ds_dir, zfs_prop_to_name(ZFS_PROP_REFQUOTA), ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } if (newval == 0) { dsl_dataset_rele(ds, FTAG); return (0); } if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || newval < ds->ds_reserved) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(ENOSPC)); } dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_set_qr_arg_t *ddsqra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds = NULL; uint64_t newval; VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFQUOTA), ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, &ddsqra->ddsqra_value, tx); VERIFY0(dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); if (ds->ds_quota != newval) { dmu_buf_will_dirty(ds->ds_dbuf, tx); ds->ds_quota = newval; } dsl_dataset_rele(ds, FTAG); } int dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, uint64_t refquota) { dsl_dataset_set_qr_arg_t ddsqra; ddsqra.ddsqra_name = dsname; ddsqra.ddsqra_source = source; ddsqra.ddsqra_value = refquota; return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_EXTRA_RESERVED)); } static int dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) { dsl_dataset_set_qr_arg_t *ddsqra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; int error; uint64_t newval, unique; if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) return (SET_ERROR(ENOTSUP)); error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); if (error != 0) return (error); if (ds->ds_is_snapshot) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } error = dsl_prop_predict(ds->ds_dir, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } /* * If we are doing the preliminary check in open context, the * space estimates may be inaccurate. */ if (!dmu_tx_is_syncing(tx)) { dsl_dataset_rele(ds, FTAG); return (0); } mutex_enter(&ds->ds_lock); if (!DS_UNIQUE_IS_ACCURATE(ds)) dsl_dataset_recalc_head_uniq(ds); unique = dsl_dataset_phys(ds)->ds_unique_bytes; mutex_exit(&ds->ds_lock); if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { uint64_t delta = MAX(unique, newval) - MAX(unique, ds->ds_reserved); if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || (ds->ds_quota > 0 && newval > ds->ds_quota)) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(ENOSPC)); } } dsl_dataset_rele(ds, FTAG); return (0); } void dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, zprop_source_t source, uint64_t value, dmu_tx_t *tx) { uint64_t newval; uint64_t unique; int64_t delta; dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), source, sizeof (value), 1, &value, tx); VERIFY0(dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); dmu_buf_will_dirty(ds->ds_dbuf, tx); mutex_enter(&ds->ds_dir->dd_lock); mutex_enter(&ds->ds_lock); ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); unique = dsl_dataset_phys(ds)->ds_unique_bytes; delta = MAX(0, (int64_t)(newval - unique)) - MAX(0, (int64_t)(ds->ds_reserved - unique)); ds->ds_reserved = newval; mutex_exit(&ds->ds_lock); dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); mutex_exit(&ds->ds_dir->dd_lock); } static void dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_set_qr_arg_t *ddsqra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds = NULL; VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); dsl_dataset_set_refreservation_sync_impl(ds, ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); dsl_dataset_rele(ds, FTAG); } int dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, uint64_t refreservation) { dsl_dataset_set_qr_arg_t ddsqra; ddsqra.ddsqra_name = dsname; ddsqra.ddsqra_source = source; ddsqra.ddsqra_value = refreservation; return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, dsl_dataset_set_refreservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_EXTRA_RESERVED)); } typedef struct dsl_dataset_set_compression_arg { const char *ddsca_name; zprop_source_t ddsca_source; uint64_t ddsca_value; } dsl_dataset_set_compression_arg_t; static int dsl_dataset_set_compression_check(void *arg, dmu_tx_t *tx) { dsl_dataset_set_compression_arg_t *ddsca = arg; dsl_pool_t *dp = dmu_tx_pool(tx); uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value); spa_feature_t f = zio_compress_to_feature(compval); if (f == SPA_FEATURE_NONE) return (SET_ERROR(EINVAL)); if (!spa_feature_is_enabled(dp->dp_spa, f)) return (SET_ERROR(ENOTSUP)); return (0); } static void dsl_dataset_set_compression_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_set_compression_arg_t *ddsca = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds = NULL; uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value); spa_feature_t f = zio_compress_to_feature(compval); ASSERT3S(spa_feature_table[f].fi_type, ==, ZFEATURE_TYPE_BOOLEAN); VERIFY0(dsl_dataset_hold(dp, ddsca->ddsca_name, FTAG, &ds)); if (zfeature_active(f, ds->ds_feature[f]) != B_TRUE) { ds->ds_feature_activation[f] = (void *)B_TRUE; dsl_dataset_activate_feature(ds->ds_object, f, ds->ds_feature_activation[f], tx); ds->ds_feature[f] = ds->ds_feature_activation[f]; } dsl_dataset_rele(ds, FTAG); } int dsl_dataset_set_compression(const char *dsname, zprop_source_t source, uint64_t compression) { dsl_dataset_set_compression_arg_t ddsca; /* * The sync task is only required for zstd in order to activate * the feature flag when the property is first set. */ if (ZIO_COMPRESS_ALGO(compression) != ZIO_COMPRESS_ZSTD) return (0); ddsca.ddsca_name = dsname; ddsca.ddsca_source = source; ddsca.ddsca_value = compression; return (dsl_sync_task(dsname, dsl_dataset_set_compression_check, dsl_dataset_set_compression_sync, &ddsca, 0, ZFS_SPACE_CHECK_EXTRA_RESERVED)); } /* * Return (in *usedp) the amount of space referenced by "new" that was not * referenced at the time the bookmark corresponds to. "New" may be a * snapshot or a head. The bookmark must be before new, in * new's filesystem (or its origin) -- caller verifies this. * * The written space is calculated by considering two components: First, we * ignore any freed space, and calculate the written as new's used space * minus old's used space. Next, we add in the amount of space that was freed * between the two time points, thus reducing new's used space relative to * old's. Specifically, this is the space that was born before * zbm_creation_txg, and freed before new (ie. on new's deadlist or a * previous deadlist). * * space freed [---------------------] * snapshots ---O-------O--------O-------O------ * bookmark new * * Note, the bookmark's zbm_*_bytes_refd must be valid, but if the HAS_FBN * flag is not set, we will calculate the freed_before_next based on the * next snapshot's deadlist, rather than using zbm_*_freed_before_next_snap. */ static int dsl_dataset_space_written_impl(zfs_bookmark_phys_t *bmp, dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) { int err = 0; dsl_pool_t *dp = new->ds_dir->dd_pool; ASSERT(dsl_pool_config_held(dp)); if (dsl_dataset_is_snapshot(new)) { ASSERT3U(bmp->zbm_creation_txg, <, dsl_dataset_phys(new)->ds_creation_txg); } *usedp = 0; *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; *usedp -= bmp->zbm_referenced_bytes_refd; *compp = 0; *compp += dsl_dataset_phys(new)->ds_compressed_bytes; *compp -= bmp->zbm_compressed_bytes_refd; *uncompp = 0; *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; *uncompp -= bmp->zbm_uncompressed_bytes_refd; dsl_dataset_t *snap = new; while (dsl_dataset_phys(snap)->ds_prev_snap_txg > bmp->zbm_creation_txg) { uint64_t used, comp, uncomp; dsl_deadlist_space_range(&snap->ds_deadlist, 0, bmp->zbm_creation_txg, &used, &comp, &uncomp); *usedp += used; *compp += comp; *uncompp += uncomp; uint64_t snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; if (snap != new) dsl_dataset_rele(snap, FTAG); err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); if (err != 0) break; } /* * We might not have the FBN if we are calculating written from * a snapshot (because we didn't know the correct "next" snapshot * until now). */ if (bmp->zbm_flags & ZBM_FLAG_HAS_FBN) { *usedp += bmp->zbm_referenced_freed_before_next_snap; *compp += bmp->zbm_compressed_freed_before_next_snap; *uncompp += bmp->zbm_uncompressed_freed_before_next_snap; } else { ASSERT3U(dsl_dataset_phys(snap)->ds_prev_snap_txg, ==, bmp->zbm_creation_txg); uint64_t used, comp, uncomp; dsl_deadlist_space(&snap->ds_deadlist, &used, &comp, &uncomp); *usedp += used; *compp += comp; *uncompp += uncomp; } if (snap != new) dsl_dataset_rele(snap, FTAG); return (err); } /* * Return (in *usedp) the amount of space written in new that was not * present at the time the bookmark corresponds to. New may be a * snapshot or the head. Old must be a bookmark before new, in * new's filesystem (or its origin) -- caller verifies this. */ int dsl_dataset_space_written_bookmark(zfs_bookmark_phys_t *bmp, dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) { if (!(bmp->zbm_flags & ZBM_FLAG_HAS_FBN)) return (SET_ERROR(ENOTSUP)); return (dsl_dataset_space_written_impl(bmp, new, usedp, compp, uncompp)); } /* * Return (in *usedp) the amount of space written in new that is not * present in oldsnap. New may be a snapshot or the head. Old must be * a snapshot before new, in new's filesystem (or its origin). If not then * fail and return EINVAL. */ int dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) { if (!dsl_dataset_is_before(new, oldsnap, 0)) return (SET_ERROR(EINVAL)); zfs_bookmark_phys_t zbm = { 0 }; dsl_dataset_phys_t *dsp = dsl_dataset_phys(oldsnap); zbm.zbm_guid = dsp->ds_guid; zbm.zbm_creation_txg = dsp->ds_creation_txg; zbm.zbm_creation_time = dsp->ds_creation_time; zbm.zbm_referenced_bytes_refd = dsp->ds_referenced_bytes; zbm.zbm_compressed_bytes_refd = dsp->ds_compressed_bytes; zbm.zbm_uncompressed_bytes_refd = dsp->ds_uncompressed_bytes; /* * If oldsnap is the origin (or origin's origin, ...) of new, * we can't easily calculate the effective FBN. Therefore, * we do not set ZBM_FLAG_HAS_FBN, so that the _impl will calculate * it relative to the correct "next": the next snapshot towards "new", * rather than the next snapshot in oldsnap's dsl_dir. */ return (dsl_dataset_space_written_impl(&zbm, new, usedp, compp, uncompp)); } /* * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, * lastsnap, and all snapshots in between are deleted. * * blocks that would be freed [---------------------------] * snapshots ---O-------O--------O-------O--------O * firstsnap lastsnap * * This is the set of blocks that were born after the snap before firstsnap, * (birth > firstsnap->prev_snap_txg) and died before the snap after the * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). * We calculate this by iterating over the relevant deadlists (from the snap * after lastsnap, backward to the snap after firstsnap), summing up the * space on the deadlist that was born after the snap before firstsnap. */ int dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, dsl_dataset_t *lastsnap, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) { int err = 0; uint64_t snapobj; dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; ASSERT(firstsnap->ds_is_snapshot); ASSERT(lastsnap->ds_is_snapshot); /* * Check that the snapshots are in the same dsl_dir, and firstsnap * is before lastsnap. */ if (firstsnap->ds_dir != lastsnap->ds_dir || dsl_dataset_phys(firstsnap)->ds_creation_txg > dsl_dataset_phys(lastsnap)->ds_creation_txg) return (SET_ERROR(EINVAL)); *usedp = *compp = *uncompp = 0; snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; while (snapobj != firstsnap->ds_object) { dsl_dataset_t *ds; uint64_t used, comp, uncomp; err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); if (err != 0) break; dsl_deadlist_space_range(&ds->ds_deadlist, dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, &used, &comp, &uncomp); *usedp += used; *compp += comp; *uncompp += uncomp; snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; ASSERT3U(snapobj, !=, 0); dsl_dataset_rele(ds, FTAG); } return (err); } /* * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. * For example, they could both be snapshots of the same filesystem, and * 'earlier' is before 'later'. Or 'earlier' could be the origin of * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's * filesystem. Or 'earlier' could be the origin's origin. * * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. */ boolean_t dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, uint64_t earlier_txg) { dsl_pool_t *dp = later->ds_dir->dd_pool; int error; boolean_t ret; ASSERT(dsl_pool_config_held(dp)); ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); if (earlier_txg == 0) earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; if (later->ds_is_snapshot && earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) return (B_FALSE); if (later->ds_dir == earlier->ds_dir) return (B_TRUE); /* * We check dd_origin_obj explicitly here rather than using * dsl_dir_is_clone() so that we will return TRUE if "earlier" * is $ORIGIN@$ORIGIN. dsl_dataset_space_written() depends on * this behavior. */ if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == 0) return (B_FALSE); dsl_dataset_t *origin; error = dsl_dataset_hold_obj(dp, dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); if (error != 0) return (B_FALSE); if (dsl_dataset_phys(origin)->ds_creation_txg == earlier_txg && origin->ds_dir == earlier->ds_dir) { dsl_dataset_rele(origin, FTAG); return (B_TRUE); } ret = dsl_dataset_is_before(origin, earlier, earlier_txg); dsl_dataset_rele(origin, FTAG); return (ret); } void dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) { objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); } boolean_t dsl_dataset_is_zapified(dsl_dataset_t *ds) { dmu_object_info_t doi; dmu_object_info_from_db(ds->ds_dbuf, &doi); return (doi.doi_type == DMU_OTN_ZAP_METADATA); } boolean_t dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds) { return (dsl_dataset_is_zapified(ds) && zap_contains(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0); } uint64_t dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds) { uint64_t remap_deadlist_obj; int err; if (!dsl_dataset_is_zapified(ds)) return (0); err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj); if (err != 0) { VERIFY3S(err, ==, ENOENT); return (0); } ASSERT(remap_deadlist_obj != 0); return (remap_deadlist_obj); } boolean_t dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds) { EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist), dsl_dataset_get_remap_deadlist_object(ds) != 0); return (dsl_deadlist_is_open(&ds->ds_remap_deadlist)); } static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx) { ASSERT(obj != 0); dsl_dataset_zapify(ds, tx); VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx)); } static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx) { VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx)); } void dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx) { uint64_t remap_deadlist_object; spa_t *spa = ds->ds_dir->dd_pool->dp_spa; ASSERT(dmu_tx_is_syncing(tx)); ASSERT(dsl_dataset_remap_deadlist_exists(ds)); remap_deadlist_object = ds->ds_remap_deadlist.dl_object; dsl_deadlist_close(&ds->ds_remap_deadlist); dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx); dsl_dataset_unset_remap_deadlist_object(ds, tx); spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); } void dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx) { uint64_t remap_deadlist_obj; spa_t *spa = ds->ds_dir->dd_pool->dp_spa; ASSERT(dmu_tx_is_syncing(tx)); ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock)); /* * Currently we only create remap deadlists when there are indirect * vdevs with referenced mappings. */ ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL)); remap_deadlist_obj = dsl_deadlist_clone( &ds->ds_deadlist, UINT64_MAX, dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); dsl_dataset_set_remap_deadlist_object(ds, remap_deadlist_obj, tx); dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa), remap_deadlist_obj); spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); } void dsl_dataset_activate_redaction(dsl_dataset_t *ds, uint64_t *redact_snaps, uint64_t num_redact_snaps, dmu_tx_t *tx) { uint64_t dsobj = ds->ds_object; struct feature_type_uint64_array_arg *ftuaa = kmem_zalloc(sizeof (*ftuaa), KM_SLEEP); ftuaa->length = (int64_t)num_redact_snaps; if (num_redact_snaps > 0) { ftuaa->array = kmem_alloc(num_redact_snaps * sizeof (uint64_t), KM_SLEEP); bcopy(redact_snaps, ftuaa->array, num_redact_snaps * sizeof (uint64_t)); } dsl_dataset_activate_feature(dsobj, SPA_FEATURE_REDACTED_DATASETS, ftuaa, tx); ds->ds_feature[SPA_FEATURE_REDACTED_DATASETS] = ftuaa; } /* BEGIN CSTYLED */ #if defined(_LP64) #define RECORDSIZE_PERM ZMOD_RW #else /* Limited to 1M on 32-bit platforms due to lack of virtual address space */ #define RECORDSIZE_PERM ZMOD_RD #endif ZFS_MODULE_PARAM(zfs, zfs_, max_recordsize, INT, RECORDSIZE_PERM, "Max allowed record size"); ZFS_MODULE_PARAM(zfs, zfs_, allow_redacted_dataset_mount, INT, ZMOD_RW, "Allow mounting of redacted datasets"); /* END CSTYLED */ EXPORT_SYMBOL(dsl_dataset_hold); EXPORT_SYMBOL(dsl_dataset_hold_flags); EXPORT_SYMBOL(dsl_dataset_hold_obj); EXPORT_SYMBOL(dsl_dataset_hold_obj_flags); EXPORT_SYMBOL(dsl_dataset_own); EXPORT_SYMBOL(dsl_dataset_own_obj); EXPORT_SYMBOL(dsl_dataset_name); EXPORT_SYMBOL(dsl_dataset_rele); EXPORT_SYMBOL(dsl_dataset_rele_flags); EXPORT_SYMBOL(dsl_dataset_disown); EXPORT_SYMBOL(dsl_dataset_tryown); EXPORT_SYMBOL(dsl_dataset_create_sync); EXPORT_SYMBOL(dsl_dataset_create_sync_dd); EXPORT_SYMBOL(dsl_dataset_snapshot_check); EXPORT_SYMBOL(dsl_dataset_snapshot_sync); EXPORT_SYMBOL(dsl_dataset_promote); EXPORT_SYMBOL(dsl_dataset_user_hold); EXPORT_SYMBOL(dsl_dataset_user_release); EXPORT_SYMBOL(dsl_dataset_get_holds); EXPORT_SYMBOL(dsl_dataset_get_blkptr); EXPORT_SYMBOL(dsl_dataset_get_spa); EXPORT_SYMBOL(dsl_dataset_modified_since_snap); EXPORT_SYMBOL(dsl_dataset_space_written); EXPORT_SYMBOL(dsl_dataset_space_wouldfree); EXPORT_SYMBOL(dsl_dataset_sync); EXPORT_SYMBOL(dsl_dataset_block_born); EXPORT_SYMBOL(dsl_dataset_block_kill); EXPORT_SYMBOL(dsl_dataset_dirty); EXPORT_SYMBOL(dsl_dataset_stats); EXPORT_SYMBOL(dsl_dataset_fast_stat); EXPORT_SYMBOL(dsl_dataset_space); EXPORT_SYMBOL(dsl_dataset_fsid_guid); EXPORT_SYMBOL(dsl_dsobj_to_dsname); EXPORT_SYMBOL(dsl_dataset_check_quota); EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl); EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl); diff --git a/module/zfs/dsl_scan.c b/module/zfs/dsl_scan.c index 8d7c16341cd1..f3c639b0d04e 100644 --- a/module/zfs/dsl_scan.c +++ b/module/zfs/dsl_scan.c @@ -1,4495 +1,4490 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2021 by Delphix. All rights reserved. * Copyright 2016 Gary Mills * Copyright (c) 2017, 2019, Datto Inc. All rights reserved. * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. * Copyright 2019 Joyent, Inc. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef _KERNEL #include #endif /* * Grand theory statement on scan queue sorting * * Scanning is implemented by recursively traversing all indirection levels * in an object and reading all blocks referenced from said objects. This * results in us approximately traversing the object from lowest logical * offset to the highest. For best performance, we would want the logical * blocks to be physically contiguous. However, this is frequently not the * case with pools given the allocation patterns of copy-on-write filesystems. * So instead, we put the I/Os into a reordering queue and issue them in a * way that will most benefit physical disks (LBA-order). * * Queue management: * * Ideally, we would want to scan all metadata and queue up all block I/O * prior to starting to issue it, because that allows us to do an optimal * sorting job. This can however consume large amounts of memory. Therefore * we continuously monitor the size of the queues and constrain them to 5% * (zfs_scan_mem_lim_fact) of physmem. If the queues grow larger than this * limit, we clear out a few of the largest extents at the head of the queues * to make room for more scanning. Hopefully, these extents will be fairly * large and contiguous, allowing us to approach sequential I/O throughput * even without a fully sorted tree. * * Metadata scanning takes place in dsl_scan_visit(), which is called from * dsl_scan_sync() every spa_sync(). If we have either fully scanned all * metadata on the pool, or we need to make room in memory because our * queues are too large, dsl_scan_visit() is postponed and * scan_io_queues_run() is called from dsl_scan_sync() instead. This implies * that metadata scanning and queued I/O issuing are mutually exclusive. This * allows us to provide maximum sequential I/O throughput for the majority of * I/O's issued since sequential I/O performance is significantly negatively * impacted if it is interleaved with random I/O. * * Implementation Notes * * One side effect of the queued scanning algorithm is that the scanning code * needs to be notified whenever a block is freed. This is needed to allow * the scanning code to remove these I/Os from the issuing queue. Additionally, * we do not attempt to queue gang blocks to be issued sequentially since this * is very hard to do and would have an extremely limited performance benefit. * Instead, we simply issue gang I/Os as soon as we find them using the legacy * algorithm. * * Backwards compatibility * * This new algorithm is backwards compatible with the legacy on-disk data * structures (and therefore does not require a new feature flag). * Periodically during scanning (see zfs_scan_checkpoint_intval), the scan * will stop scanning metadata (in logical order) and wait for all outstanding * sorted I/O to complete. Once this is done, we write out a checkpoint * bookmark, indicating that we have scanned everything logically before it. * If the pool is imported on a machine without the new sorting algorithm, * the scan simply resumes from the last checkpoint using the legacy algorithm. */ typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *, const zbookmark_phys_t *); static scan_cb_t dsl_scan_scrub_cb; static int scan_ds_queue_compare(const void *a, const void *b); static int scan_prefetch_queue_compare(const void *a, const void *b); static void scan_ds_queue_clear(dsl_scan_t *scn); static void scan_ds_prefetch_queue_clear(dsl_scan_t *scn); static boolean_t scan_ds_queue_contains(dsl_scan_t *scn, uint64_t dsobj, uint64_t *txg); static void scan_ds_queue_insert(dsl_scan_t *scn, uint64_t dsobj, uint64_t txg); static void scan_ds_queue_remove(dsl_scan_t *scn, uint64_t dsobj); static void scan_ds_queue_sync(dsl_scan_t *scn, dmu_tx_t *tx); static uint64_t dsl_scan_count_data_disks(vdev_t *vd); extern int zfs_vdev_async_write_active_min_dirty_percent; static int zfs_scan_blkstats = 0; /* * By default zfs will check to ensure it is not over the hard memory * limit before each txg. If finer-grained control of this is needed * this value can be set to 1 to enable checking before scanning each * block. */ int zfs_scan_strict_mem_lim = B_FALSE; /* * Maximum number of parallelly executed bytes per leaf vdev. We attempt * to strike a balance here between keeping the vdev queues full of I/Os * at all times and not overflowing the queues to cause long latency, * which would cause long txg sync times. No matter what, we will not * overload the drives with I/O, since that is protected by * zfs_vdev_scrub_max_active. */ unsigned long zfs_scan_vdev_limit = 4 << 20; int zfs_scan_issue_strategy = 0; int zfs_scan_legacy = B_FALSE; /* don't queue & sort zios, go direct */ unsigned long zfs_scan_max_ext_gap = 2 << 20; /* in bytes */ /* * fill_weight is non-tunable at runtime, so we copy it at module init from * zfs_scan_fill_weight. Runtime adjustments to zfs_scan_fill_weight would * break queue sorting. */ int zfs_scan_fill_weight = 3; static uint64_t fill_weight; /* See dsl_scan_should_clear() for details on the memory limit tunables */ uint64_t zfs_scan_mem_lim_min = 16 << 20; /* bytes */ uint64_t zfs_scan_mem_lim_soft_max = 128 << 20; /* bytes */ int zfs_scan_mem_lim_fact = 20; /* fraction of physmem */ int zfs_scan_mem_lim_soft_fact = 20; /* fraction of mem lim above */ int zfs_scrub_min_time_ms = 1000; /* min millisecs to scrub per txg */ int zfs_obsolete_min_time_ms = 500; /* min millisecs to obsolete per txg */ int zfs_free_min_time_ms = 1000; /* min millisecs to free per txg */ int zfs_resilver_min_time_ms = 3000; /* min millisecs to resilver per txg */ int zfs_scan_checkpoint_intval = 7200; /* in seconds */ int zfs_scan_suspend_progress = 0; /* set to prevent scans from progressing */ int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */ int zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */ enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE; /* max number of blocks to free in a single TXG */ unsigned long zfs_async_block_max_blocks = ULONG_MAX; /* max number of dedup blocks to free in a single TXG */ unsigned long zfs_max_async_dedup_frees = 100000; int zfs_resilver_disable_defer = 0; /* set to disable resilver deferring */ /* * We wait a few txgs after importing a pool to begin scanning so that * the import / mounting code isn't held up by scrub / resilver IO. * Unfortunately, it is a bit difficult to determine exactly how long * this will take since userspace will trigger fs mounts asynchronously * and the kernel will create zvol minors asynchronously. As a result, * the value provided here is a bit arbitrary, but represents a * reasonable estimate of how many txgs it will take to finish fully * importing a pool */ #define SCAN_IMPORT_WAIT_TXGS 5 #define DSL_SCAN_IS_SCRUB_RESILVER(scn) \ ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \ (scn)->scn_phys.scn_func == POOL_SCAN_RESILVER) /* * Enable/disable the processing of the free_bpobj object. */ int zfs_free_bpobj_enabled = 1; /* the order has to match pool_scan_type */ static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = { NULL, dsl_scan_scrub_cb, /* POOL_SCAN_SCRUB */ dsl_scan_scrub_cb, /* POOL_SCAN_RESILVER */ }; /* In core node for the scn->scn_queue. Represents a dataset to be scanned */ typedef struct { uint64_t sds_dsobj; uint64_t sds_txg; avl_node_t sds_node; } scan_ds_t; /* * This controls what conditions are placed on dsl_scan_sync_state(): * SYNC_OPTIONAL) write out scn_phys iff scn_queues_pending == 0 * SYNC_MANDATORY) write out scn_phys always. scn_queues_pending must be 0. * SYNC_CACHED) if scn_queues_pending == 0, write out scn_phys. Otherwise * write out the scn_phys_cached version. * See dsl_scan_sync_state for details. */ typedef enum { SYNC_OPTIONAL, SYNC_MANDATORY, SYNC_CACHED } state_sync_type_t; /* * This struct represents the minimum information needed to reconstruct a * zio for sequential scanning. This is useful because many of these will * accumulate in the sequential IO queues before being issued, so saving * memory matters here. */ typedef struct scan_io { /* fields from blkptr_t */ uint64_t sio_blk_prop; uint64_t sio_phys_birth; uint64_t sio_birth; zio_cksum_t sio_cksum; uint32_t sio_nr_dvas; /* fields from zio_t */ uint32_t sio_flags; zbookmark_phys_t sio_zb; /* members for queue sorting */ union { avl_node_t sio_addr_node; /* link into issuing queue */ list_node_t sio_list_node; /* link for issuing to disk */ } sio_nodes; /* * There may be up to SPA_DVAS_PER_BP DVAs here from the bp, * depending on how many were in the original bp. Only the * first DVA is really used for sorting and issuing purposes. * The other DVAs (if provided) simply exist so that the zio * layer can find additional copies to repair from in the * event of an error. This array must go at the end of the * struct to allow this for the variable number of elements. */ dva_t sio_dva[0]; } scan_io_t; #define SIO_SET_OFFSET(sio, x) DVA_SET_OFFSET(&(sio)->sio_dva[0], x) #define SIO_SET_ASIZE(sio, x) DVA_SET_ASIZE(&(sio)->sio_dva[0], x) #define SIO_GET_OFFSET(sio) DVA_GET_OFFSET(&(sio)->sio_dva[0]) #define SIO_GET_ASIZE(sio) DVA_GET_ASIZE(&(sio)->sio_dva[0]) #define SIO_GET_END_OFFSET(sio) \ (SIO_GET_OFFSET(sio) + SIO_GET_ASIZE(sio)) #define SIO_GET_MUSED(sio) \ (sizeof (scan_io_t) + ((sio)->sio_nr_dvas * sizeof (dva_t))) struct dsl_scan_io_queue { dsl_scan_t *q_scn; /* associated dsl_scan_t */ vdev_t *q_vd; /* top-level vdev that this queue represents */ zio_t *q_zio; /* scn_zio_root child for waiting on IO */ /* trees used for sorting I/Os and extents of I/Os */ range_tree_t *q_exts_by_addr; zfs_btree_t q_exts_by_size; avl_tree_t q_sios_by_addr; uint64_t q_sio_memused; uint64_t q_last_ext_addr; /* members for zio rate limiting */ uint64_t q_maxinflight_bytes; uint64_t q_inflight_bytes; kcondvar_t q_zio_cv; /* used under vd->vdev_scan_io_queue_lock */ /* per txg statistics */ uint64_t q_total_seg_size_this_txg; uint64_t q_segs_this_txg; uint64_t q_total_zio_size_this_txg; uint64_t q_zios_this_txg; }; /* private data for dsl_scan_prefetch_cb() */ typedef struct scan_prefetch_ctx { zfs_refcount_t spc_refcnt; /* refcount for memory management */ dsl_scan_t *spc_scn; /* dsl_scan_t for the pool */ boolean_t spc_root; /* is this prefetch for an objset? */ uint8_t spc_indblkshift; /* dn_indblkshift of current dnode */ uint16_t spc_datablkszsec; /* dn_idatablkszsec of current dnode */ } scan_prefetch_ctx_t; /* private data for dsl_scan_prefetch() */ typedef struct scan_prefetch_issue_ctx { avl_node_t spic_avl_node; /* link into scn->scn_prefetch_queue */ scan_prefetch_ctx_t *spic_spc; /* spc for the callback */ blkptr_t spic_bp; /* bp to prefetch */ zbookmark_phys_t spic_zb; /* bookmark to prefetch */ } scan_prefetch_issue_ctx_t; static void scan_exec_io(dsl_pool_t *dp, const blkptr_t *bp, int zio_flags, const zbookmark_phys_t *zb, dsl_scan_io_queue_t *queue); static void scan_io_queue_insert_impl(dsl_scan_io_queue_t *queue, scan_io_t *sio); static dsl_scan_io_queue_t *scan_io_queue_create(vdev_t *vd); static void scan_io_queues_destroy(dsl_scan_t *scn); static kmem_cache_t *sio_cache[SPA_DVAS_PER_BP]; /* sio->sio_nr_dvas must be set so we know which cache to free from */ static void sio_free(scan_io_t *sio) { ASSERT3U(sio->sio_nr_dvas, >, 0); ASSERT3U(sio->sio_nr_dvas, <=, SPA_DVAS_PER_BP); kmem_cache_free(sio_cache[sio->sio_nr_dvas - 1], sio); } /* It is up to the caller to set sio->sio_nr_dvas for freeing */ static scan_io_t * sio_alloc(unsigned short nr_dvas) { ASSERT3U(nr_dvas, >, 0); ASSERT3U(nr_dvas, <=, SPA_DVAS_PER_BP); return (kmem_cache_alloc(sio_cache[nr_dvas - 1], KM_SLEEP)); } void scan_init(void) { /* * This is used in ext_size_compare() to weight segments * based on how sparse they are. This cannot be changed * mid-scan and the tree comparison functions don't currently * have a mechanism for passing additional context to the * compare functions. Thus we store this value globally and * we only allow it to be set at module initialization time */ fill_weight = zfs_scan_fill_weight; for (int i = 0; i < SPA_DVAS_PER_BP; i++) { char name[36]; (void) snprintf(name, sizeof (name), "sio_cache_%d", i); sio_cache[i] = kmem_cache_create(name, (sizeof (scan_io_t) + ((i + 1) * sizeof (dva_t))), 0, NULL, NULL, NULL, NULL, NULL, 0); } } void scan_fini(void) { for (int i = 0; i < SPA_DVAS_PER_BP; i++) { kmem_cache_destroy(sio_cache[i]); } } static inline boolean_t dsl_scan_is_running(const dsl_scan_t *scn) { return (scn->scn_phys.scn_state == DSS_SCANNING); } boolean_t dsl_scan_resilvering(dsl_pool_t *dp) { return (dsl_scan_is_running(dp->dp_scan) && dp->dp_scan->scn_phys.scn_func == POOL_SCAN_RESILVER); } static inline void sio2bp(const scan_io_t *sio, blkptr_t *bp) { bzero(bp, sizeof (*bp)); bp->blk_prop = sio->sio_blk_prop; bp->blk_phys_birth = sio->sio_phys_birth; bp->blk_birth = sio->sio_birth; bp->blk_fill = 1; /* we always only work with data pointers */ bp->blk_cksum = sio->sio_cksum; ASSERT3U(sio->sio_nr_dvas, >, 0); ASSERT3U(sio->sio_nr_dvas, <=, SPA_DVAS_PER_BP); bcopy(sio->sio_dva, bp->blk_dva, sio->sio_nr_dvas * sizeof (dva_t)); } static inline void bp2sio(const blkptr_t *bp, scan_io_t *sio, int dva_i) { sio->sio_blk_prop = bp->blk_prop; sio->sio_phys_birth = bp->blk_phys_birth; sio->sio_birth = bp->blk_birth; sio->sio_cksum = bp->blk_cksum; sio->sio_nr_dvas = BP_GET_NDVAS(bp); /* * Copy the DVAs to the sio. We need all copies of the block so * that the self healing code can use the alternate copies if the * first is corrupted. We want the DVA at index dva_i to be first * in the sio since this is the primary one that we want to issue. */ for (int i = 0, j = dva_i; i < sio->sio_nr_dvas; i++, j++) { sio->sio_dva[i] = bp->blk_dva[j % sio->sio_nr_dvas]; } } int dsl_scan_init(dsl_pool_t *dp, uint64_t txg) { int err; dsl_scan_t *scn; spa_t *spa = dp->dp_spa; uint64_t f; scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP); scn->scn_dp = dp; /* * It's possible that we're resuming a scan after a reboot so * make sure that the scan_async_destroying flag is initialized * appropriately. */ ASSERT(!scn->scn_async_destroying); scn->scn_async_destroying = spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY); /* * Calculate the max number of in-flight bytes for pool-wide * scanning operations (minimum 1MB). Limits for the issuing * phase are done per top-level vdev and are handled separately. */ scn->scn_maxinflight_bytes = MAX(zfs_scan_vdev_limit * dsl_scan_count_data_disks(spa->spa_root_vdev), 1ULL << 20); avl_create(&scn->scn_queue, scan_ds_queue_compare, sizeof (scan_ds_t), offsetof(scan_ds_t, sds_node)); avl_create(&scn->scn_prefetch_queue, scan_prefetch_queue_compare, sizeof (scan_prefetch_issue_ctx_t), offsetof(scan_prefetch_issue_ctx_t, spic_avl_node)); err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, "scrub_func", sizeof (uint64_t), 1, &f); if (err == 0) { /* * There was an old-style scrub in progress. Restart a * new-style scrub from the beginning. */ scn->scn_restart_txg = txg; zfs_dbgmsg("old-style scrub was in progress; " "restarting new-style scrub in txg %llu", (longlong_t)scn->scn_restart_txg); /* * Load the queue obj from the old location so that it * can be freed by dsl_scan_done(). */ (void) zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, "scrub_queue", sizeof (uint64_t), 1, &scn->scn_phys.scn_queue_obj); } else { err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys); /* * Detect if the pool contains the signature of #2094. If it * does properly update the scn->scn_phys structure and notify * the administrator by setting an errata for the pool. */ if (err == EOVERFLOW) { uint64_t zaptmp[SCAN_PHYS_NUMINTS + 1]; VERIFY3S(SCAN_PHYS_NUMINTS, ==, 24); VERIFY3S(offsetof(dsl_scan_phys_t, scn_flags), ==, (23 * sizeof (uint64_t))); err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS + 1, &zaptmp); if (err == 0) { uint64_t overflow = zaptmp[SCAN_PHYS_NUMINTS]; if (overflow & ~DSL_SCAN_FLAGS_MASK || scn->scn_async_destroying) { spa->spa_errata = ZPOOL_ERRATA_ZOL_2094_ASYNC_DESTROY; return (EOVERFLOW); } bcopy(zaptmp, &scn->scn_phys, SCAN_PHYS_NUMINTS * sizeof (uint64_t)); scn->scn_phys.scn_flags = overflow; /* Required scrub already in progress. */ if (scn->scn_phys.scn_state == DSS_FINISHED || scn->scn_phys.scn_state == DSS_CANCELED) spa->spa_errata = ZPOOL_ERRATA_ZOL_2094_SCRUB; } } if (err == ENOENT) return (0); else if (err) return (err); /* * We might be restarting after a reboot, so jump the issued * counter to how far we've scanned. We know we're consistent * up to here. */ scn->scn_issued_before_pass = scn->scn_phys.scn_examined; if (dsl_scan_is_running(scn) && spa_prev_software_version(dp->dp_spa) < SPA_VERSION_SCAN) { /* * A new-type scrub was in progress on an old * pool, and the pool was accessed by old * software. Restart from the beginning, since * the old software may have changed the pool in * the meantime. */ scn->scn_restart_txg = txg; zfs_dbgmsg("new-style scrub was modified " "by old software; restarting in txg %llu", (longlong_t)scn->scn_restart_txg); } else if (dsl_scan_resilvering(dp)) { /* * If a resilver is in progress and there are already * errors, restart it instead of finishing this scan and * then restarting it. If there haven't been any errors * then remember that the incore DTL is valid. */ if (scn->scn_phys.scn_errors > 0) { scn->scn_restart_txg = txg; zfs_dbgmsg("resilver can't excise DTL_MISSING " "when finished; restarting in txg %llu", (u_longlong_t)scn->scn_restart_txg); } else { /* it's safe to excise DTL when finished */ spa->spa_scrub_started = B_TRUE; } } } bcopy(&scn->scn_phys, &scn->scn_phys_cached, sizeof (scn->scn_phys)); /* reload the queue into the in-core state */ if (scn->scn_phys.scn_queue_obj != 0) { zap_cursor_t zc; zap_attribute_t za; for (zap_cursor_init(&zc, dp->dp_meta_objset, scn->scn_phys.scn_queue_obj); zap_cursor_retrieve(&zc, &za) == 0; (void) zap_cursor_advance(&zc)) { scan_ds_queue_insert(scn, zfs_strtonum(za.za_name, NULL), za.za_first_integer); } zap_cursor_fini(&zc); } spa_scan_stat_init(spa); return (0); } void dsl_scan_fini(dsl_pool_t *dp) { if (dp->dp_scan != NULL) { dsl_scan_t *scn = dp->dp_scan; if (scn->scn_taskq != NULL) taskq_destroy(scn->scn_taskq); scan_ds_queue_clear(scn); avl_destroy(&scn->scn_queue); scan_ds_prefetch_queue_clear(scn); avl_destroy(&scn->scn_prefetch_queue); kmem_free(dp->dp_scan, sizeof (dsl_scan_t)); dp->dp_scan = NULL; } } static boolean_t dsl_scan_restarting(dsl_scan_t *scn, dmu_tx_t *tx) { return (scn->scn_restart_txg != 0 && scn->scn_restart_txg <= tx->tx_txg); } boolean_t dsl_scan_resilver_scheduled(dsl_pool_t *dp) { return ((dp->dp_scan && dp->dp_scan->scn_restart_txg != 0) || (spa_async_tasks(dp->dp_spa) & SPA_ASYNC_RESILVER)); } boolean_t dsl_scan_scrubbing(const dsl_pool_t *dp) { dsl_scan_phys_t *scn_phys = &dp->dp_scan->scn_phys; return (scn_phys->scn_state == DSS_SCANNING && scn_phys->scn_func == POOL_SCAN_SCRUB); } boolean_t dsl_scan_is_paused_scrub(const dsl_scan_t *scn) { return (dsl_scan_scrubbing(scn->scn_dp) && scn->scn_phys.scn_flags & DSF_SCRUB_PAUSED); } /* * Writes out a persistent dsl_scan_phys_t record to the pool directory. * Because we can be running in the block sorting algorithm, we do not always * want to write out the record, only when it is "safe" to do so. This safety * condition is achieved by making sure that the sorting queues are empty * (scn_queues_pending == 0). When this condition is not true, the sync'd state * is inconsistent with how much actual scanning progress has been made. The * kind of sync to be performed is specified by the sync_type argument. If the * sync is optional, we only sync if the queues are empty. If the sync is * mandatory, we do a hard ASSERT to make sure that the queues are empty. The * third possible state is a "cached" sync. This is done in response to: * 1) The dataset that was in the last sync'd dsl_scan_phys_t having been * destroyed, so we wouldn't be able to restart scanning from it. * 2) The snapshot that was in the last sync'd dsl_scan_phys_t having been * superseded by a newer snapshot. * 3) The dataset that was in the last sync'd dsl_scan_phys_t having been * swapped with its clone. * In all cases, a cached sync simply rewrites the last record we've written, * just slightly modified. For the modifications that are performed to the * last written dsl_scan_phys_t, see dsl_scan_ds_destroyed, * dsl_scan_ds_snapshotted and dsl_scan_ds_clone_swapped. */ static void dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx, state_sync_type_t sync_type) { int i; spa_t *spa = scn->scn_dp->dp_spa; ASSERT(sync_type != SYNC_MANDATORY || scn->scn_queues_pending == 0); if (scn->scn_queues_pending == 0) { for (i = 0; i < spa->spa_root_vdev->vdev_children; i++) { vdev_t *vd = spa->spa_root_vdev->vdev_child[i]; dsl_scan_io_queue_t *q = vd->vdev_scan_io_queue; if (q == NULL) continue; mutex_enter(&vd->vdev_scan_io_queue_lock); ASSERT3P(avl_first(&q->q_sios_by_addr), ==, NULL); ASSERT3P(zfs_btree_first(&q->q_exts_by_size, NULL), ==, NULL); ASSERT3P(range_tree_first(q->q_exts_by_addr), ==, NULL); mutex_exit(&vd->vdev_scan_io_queue_lock); } if (scn->scn_phys.scn_queue_obj != 0) scan_ds_queue_sync(scn, tx); VERIFY0(zap_update(scn->scn_dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys, tx)); bcopy(&scn->scn_phys, &scn->scn_phys_cached, sizeof (scn->scn_phys)); if (scn->scn_checkpointing) zfs_dbgmsg("finish scan checkpoint"); scn->scn_checkpointing = B_FALSE; scn->scn_last_checkpoint = ddi_get_lbolt(); } else if (sync_type == SYNC_CACHED) { VERIFY0(zap_update(scn->scn_dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys_cached, tx)); } } int dsl_scan_setup_check(void *arg, dmu_tx_t *tx) { (void) arg; dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; vdev_t *rvd = scn->scn_dp->dp_spa->spa_root_vdev; if (dsl_scan_is_running(scn) || vdev_rebuild_active(rvd)) return (SET_ERROR(EBUSY)); return (0); } void dsl_scan_setup_sync(void *arg, dmu_tx_t *tx) { dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; pool_scan_func_t *funcp = arg; dmu_object_type_t ot = 0; dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; ASSERT(!dsl_scan_is_running(scn)); ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS); bzero(&scn->scn_phys, sizeof (scn->scn_phys)); scn->scn_phys.scn_func = *funcp; scn->scn_phys.scn_state = DSS_SCANNING; scn->scn_phys.scn_min_txg = 0; scn->scn_phys.scn_max_txg = tx->tx_txg; scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */ scn->scn_phys.scn_start_time = gethrestime_sec(); scn->scn_phys.scn_errors = 0; scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc; scn->scn_issued_before_pass = 0; scn->scn_restart_txg = 0; scn->scn_done_txg = 0; scn->scn_last_checkpoint = 0; scn->scn_checkpointing = B_FALSE; spa_scan_stat_init(spa); if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max; /* rewrite all disk labels */ vdev_config_dirty(spa->spa_root_vdev); if (vdev_resilver_needed(spa->spa_root_vdev, &scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) { nvlist_t *aux = fnvlist_alloc(); fnvlist_add_string(aux, ZFS_EV_RESILVER_TYPE, "healing"); spa_event_notify(spa, NULL, aux, ESC_ZFS_RESILVER_START); nvlist_free(aux); } else { spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_START); } spa->spa_scrub_started = B_TRUE; /* * If this is an incremental scrub, limit the DDT scrub phase * to just the auto-ditto class (for correctness); the rest * of the scrub should go faster using top-down pruning. */ if (scn->scn_phys.scn_min_txg > TXG_INITIAL) scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO; /* * When starting a resilver clear any existing rebuild state. * This is required to prevent stale rebuild status from * being reported when a rebuild is run, then a resilver and * finally a scrub. In which case only the scrub status * should be reported by 'zpool status'. */ if (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) { vdev_t *rvd = spa->spa_root_vdev; for (uint64_t i = 0; i < rvd->vdev_children; i++) { vdev_t *vd = rvd->vdev_child[i]; vdev_rebuild_clear_sync( (void *)(uintptr_t)vd->vdev_id, tx); } } } /* back to the generic stuff */ if (zfs_scan_blkstats) { if (dp->dp_blkstats == NULL) { dp->dp_blkstats = vmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP); } memset(&dp->dp_blkstats->zab_type, 0, sizeof (dp->dp_blkstats->zab_type)); } else { if (dp->dp_blkstats) { vmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); dp->dp_blkstats = NULL; } } if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) ot = DMU_OT_ZAP_OTHER; scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset, ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx); bcopy(&scn->scn_phys, &scn->scn_phys_cached, sizeof (scn->scn_phys)); dsl_scan_sync_state(scn, tx, SYNC_MANDATORY); spa_history_log_internal(spa, "scan setup", tx, "func=%u mintxg=%llu maxtxg=%llu", *funcp, (u_longlong_t)scn->scn_phys.scn_min_txg, (u_longlong_t)scn->scn_phys.scn_max_txg); } /* * Called by the ZFS_IOC_POOL_SCAN ioctl to start a scrub or resilver. * Can also be called to resume a paused scrub. */ int dsl_scan(dsl_pool_t *dp, pool_scan_func_t func) { spa_t *spa = dp->dp_spa; dsl_scan_t *scn = dp->dp_scan; /* * Purge all vdev caches and probe all devices. We do this here * rather than in sync context because this requires a writer lock * on the spa_config lock, which we can't do from sync context. The * spa_scrub_reopen flag indicates that vdev_open() should not * attempt to start another scrub. */ spa_vdev_state_enter(spa, SCL_NONE); spa->spa_scrub_reopen = B_TRUE; vdev_reopen(spa->spa_root_vdev); spa->spa_scrub_reopen = B_FALSE; (void) spa_vdev_state_exit(spa, NULL, 0); if (func == POOL_SCAN_RESILVER) { dsl_scan_restart_resilver(spa->spa_dsl_pool, 0); return (0); } if (func == POOL_SCAN_SCRUB && dsl_scan_is_paused_scrub(scn)) { /* got scrub start cmd, resume paused scrub */ int err = dsl_scrub_set_pause_resume(scn->scn_dp, POOL_SCRUB_NORMAL); if (err == 0) { spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_RESUME); return (SET_ERROR(ECANCELED)); } return (SET_ERROR(err)); } return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check, dsl_scan_setup_sync, &func, 0, ZFS_SPACE_CHECK_EXTRA_RESERVED)); } static void dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx) { static const char *old_names[] = { "scrub_bookmark", "scrub_ddt_bookmark", "scrub_ddt_class_max", "scrub_queue", "scrub_min_txg", "scrub_max_txg", "scrub_func", "scrub_errors", NULL }; dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; int i; /* Remove any remnants of an old-style scrub. */ for (i = 0; old_names[i]; i++) { (void) zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx); } if (scn->scn_phys.scn_queue_obj != 0) { VERIFY0(dmu_object_free(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, tx)); scn->scn_phys.scn_queue_obj = 0; } scan_ds_queue_clear(scn); scan_ds_prefetch_queue_clear(scn); scn->scn_phys.scn_flags &= ~DSF_SCRUB_PAUSED; /* * If we were "restarted" from a stopped state, don't bother * with anything else. */ if (!dsl_scan_is_running(scn)) { ASSERT(!scn->scn_is_sorted); return; } if (scn->scn_is_sorted) { scan_io_queues_destroy(scn); scn->scn_is_sorted = B_FALSE; if (scn->scn_taskq != NULL) { taskq_destroy(scn->scn_taskq); scn->scn_taskq = NULL; } } scn->scn_phys.scn_state = complete ? DSS_FINISHED : DSS_CANCELED; spa_notify_waiters(spa); if (dsl_scan_restarting(scn, tx)) spa_history_log_internal(spa, "scan aborted, restarting", tx, "errors=%llu", (u_longlong_t)spa_get_errlog_size(spa)); else if (!complete) spa_history_log_internal(spa, "scan cancelled", tx, "errors=%llu", (u_longlong_t)spa_get_errlog_size(spa)); else spa_history_log_internal(spa, "scan done", tx, "errors=%llu", (u_longlong_t)spa_get_errlog_size(spa)); if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { spa->spa_scrub_active = B_FALSE; /* * If the scrub/resilver completed, update all DTLs to * reflect this. Whether it succeeded or not, vacate * all temporary scrub DTLs. * * As the scrub does not currently support traversing * data that have been freed but are part of a checkpoint, * we don't mark the scrub as done in the DTLs as faults * may still exist in those vdevs. */ if (complete && !spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg, scn->scn_phys.scn_max_txg, B_TRUE, B_FALSE); if (scn->scn_phys.scn_min_txg) { nvlist_t *aux = fnvlist_alloc(); fnvlist_add_string(aux, ZFS_EV_RESILVER_TYPE, "healing"); spa_event_notify(spa, NULL, aux, ESC_ZFS_RESILVER_FINISH); nvlist_free(aux); } else { spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_FINISH); } } else { vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg, 0, B_TRUE, B_FALSE); } spa_errlog_rotate(spa); /* * Don't clear flag until after vdev_dtl_reassess to ensure that * DTL_MISSING will get updated when possible. */ spa->spa_scrub_started = B_FALSE; /* * We may have finished replacing a device. * Let the async thread assess this and handle the detach. */ spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); /* * Clear any resilver_deferred flags in the config. * If there are drives that need resilvering, kick * off an asynchronous request to start resilver. * vdev_clear_resilver_deferred() may update the config * before the resilver can restart. In the event of * a crash during this period, the spa loading code * will find the drives that need to be resilvered * and start the resilver then. */ if (spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER) && vdev_clear_resilver_deferred(spa->spa_root_vdev, tx)) { spa_history_log_internal(spa, "starting deferred resilver", tx, "errors=%llu", (u_longlong_t)spa_get_errlog_size(spa)); spa_async_request(spa, SPA_ASYNC_RESILVER); } /* Clear recent error events (i.e. duplicate events tracking) */ if (complete) zfs_ereport_clear(spa, NULL); } scn->scn_phys.scn_end_time = gethrestime_sec(); if (spa->spa_errata == ZPOOL_ERRATA_ZOL_2094_SCRUB) spa->spa_errata = 0; ASSERT(!dsl_scan_is_running(scn)); } static int dsl_scan_cancel_check(void *arg, dmu_tx_t *tx) { (void) arg; dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; if (!dsl_scan_is_running(scn)) return (SET_ERROR(ENOENT)); return (0); } static void dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx) { (void) arg; dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; dsl_scan_done(scn, B_FALSE, tx); dsl_scan_sync_state(scn, tx, SYNC_MANDATORY); spa_event_notify(scn->scn_dp->dp_spa, NULL, NULL, ESC_ZFS_SCRUB_ABORT); } int dsl_scan_cancel(dsl_pool_t *dp) { return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check, dsl_scan_cancel_sync, NULL, 3, ZFS_SPACE_CHECK_RESERVED)); } static int dsl_scrub_pause_resume_check(void *arg, dmu_tx_t *tx) { pool_scrub_cmd_t *cmd = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_scan_t *scn = dp->dp_scan; if (*cmd == POOL_SCRUB_PAUSE) { /* can't pause a scrub when there is no in-progress scrub */ if (!dsl_scan_scrubbing(dp)) return (SET_ERROR(ENOENT)); /* can't pause a paused scrub */ if (dsl_scan_is_paused_scrub(scn)) return (SET_ERROR(EBUSY)); } else if (*cmd != POOL_SCRUB_NORMAL) { return (SET_ERROR(ENOTSUP)); } return (0); } static void dsl_scrub_pause_resume_sync(void *arg, dmu_tx_t *tx) { pool_scrub_cmd_t *cmd = arg; dsl_pool_t *dp = dmu_tx_pool(tx); spa_t *spa = dp->dp_spa; dsl_scan_t *scn = dp->dp_scan; if (*cmd == POOL_SCRUB_PAUSE) { /* can't pause a scrub when there is no in-progress scrub */ spa->spa_scan_pass_scrub_pause = gethrestime_sec(); scn->scn_phys.scn_flags |= DSF_SCRUB_PAUSED; scn->scn_phys_cached.scn_flags |= DSF_SCRUB_PAUSED; dsl_scan_sync_state(scn, tx, SYNC_CACHED); spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_PAUSED); spa_notify_waiters(spa); } else { ASSERT3U(*cmd, ==, POOL_SCRUB_NORMAL); if (dsl_scan_is_paused_scrub(scn)) { /* * We need to keep track of how much time we spend * paused per pass so that we can adjust the scrub rate * shown in the output of 'zpool status' */ spa->spa_scan_pass_scrub_spent_paused += gethrestime_sec() - spa->spa_scan_pass_scrub_pause; spa->spa_scan_pass_scrub_pause = 0; scn->scn_phys.scn_flags &= ~DSF_SCRUB_PAUSED; scn->scn_phys_cached.scn_flags &= ~DSF_SCRUB_PAUSED; dsl_scan_sync_state(scn, tx, SYNC_CACHED); } } } /* * Set scrub pause/resume state if it makes sense to do so */ int dsl_scrub_set_pause_resume(const dsl_pool_t *dp, pool_scrub_cmd_t cmd) { return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scrub_pause_resume_check, dsl_scrub_pause_resume_sync, &cmd, 3, ZFS_SPACE_CHECK_RESERVED)); } /* start a new scan, or restart an existing one. */ void dsl_scan_restart_resilver(dsl_pool_t *dp, uint64_t txg) { if (txg == 0) { dmu_tx_t *tx; tx = dmu_tx_create_dd(dp->dp_mos_dir); VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT)); txg = dmu_tx_get_txg(tx); dp->dp_scan->scn_restart_txg = txg; dmu_tx_commit(tx); } else { dp->dp_scan->scn_restart_txg = txg; } zfs_dbgmsg("restarting resilver txg=%llu", (longlong_t)txg); } void dsl_free(dsl_pool_t *dp, uint64_t txg, const blkptr_t *bp) { zio_free(dp->dp_spa, txg, bp); } void dsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp) { ASSERT(dsl_pool_sync_context(dp)); zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, pio->io_flags)); } static int scan_ds_queue_compare(const void *a, const void *b) { const scan_ds_t *sds_a = a, *sds_b = b; if (sds_a->sds_dsobj < sds_b->sds_dsobj) return (-1); if (sds_a->sds_dsobj == sds_b->sds_dsobj) return (0); return (1); } static void scan_ds_queue_clear(dsl_scan_t *scn) { void *cookie = NULL; scan_ds_t *sds; while ((sds = avl_destroy_nodes(&scn->scn_queue, &cookie)) != NULL) { kmem_free(sds, sizeof (*sds)); } } static boolean_t scan_ds_queue_contains(dsl_scan_t *scn, uint64_t dsobj, uint64_t *txg) { scan_ds_t srch, *sds; srch.sds_dsobj = dsobj; sds = avl_find(&scn->scn_queue, &srch, NULL); if (sds != NULL && txg != NULL) *txg = sds->sds_txg; return (sds != NULL); } static void scan_ds_queue_insert(dsl_scan_t *scn, uint64_t dsobj, uint64_t txg) { scan_ds_t *sds; avl_index_t where; sds = kmem_zalloc(sizeof (*sds), KM_SLEEP); sds->sds_dsobj = dsobj; sds->sds_txg = txg; VERIFY3P(avl_find(&scn->scn_queue, sds, &where), ==, NULL); avl_insert(&scn->scn_queue, sds, where); } static void scan_ds_queue_remove(dsl_scan_t *scn, uint64_t dsobj) { scan_ds_t srch, *sds; srch.sds_dsobj = dsobj; sds = avl_find(&scn->scn_queue, &srch, NULL); VERIFY(sds != NULL); avl_remove(&scn->scn_queue, sds); kmem_free(sds, sizeof (*sds)); } static void scan_ds_queue_sync(dsl_scan_t *scn, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; dmu_object_type_t ot = (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) ? DMU_OT_SCAN_QUEUE : DMU_OT_ZAP_OTHER; ASSERT0(scn->scn_queues_pending); ASSERT(scn->scn_phys.scn_queue_obj != 0); VERIFY0(dmu_object_free(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, tx)); scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset, ot, DMU_OT_NONE, 0, tx); for (scan_ds_t *sds = avl_first(&scn->scn_queue); sds != NULL; sds = AVL_NEXT(&scn->scn_queue, sds)) { VERIFY0(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, sds->sds_dsobj, sds->sds_txg, tx)); } } /* * Computes the memory limit state that we're currently in. A sorted scan * needs quite a bit of memory to hold the sorting queue, so we need to * reasonably constrain the size so it doesn't impact overall system * performance. We compute two limits: * 1) Hard memory limit: if the amount of memory used by the sorting * queues on a pool gets above this value, we stop the metadata * scanning portion and start issuing the queued up and sorted * I/Os to reduce memory usage. * This limit is calculated as a fraction of physmem (by default 5%). * We constrain the lower bound of the hard limit to an absolute * minimum of zfs_scan_mem_lim_min (default: 16 MiB). We also constrain * the upper bound to 5% of the total pool size - no chance we'll * ever need that much memory, but just to keep the value in check. * 2) Soft memory limit: once we hit the hard memory limit, we start * issuing I/O to reduce queue memory usage, but we don't want to * completely empty out the queues, since we might be able to find I/Os * that will fill in the gaps of our non-sequential IOs at some point * in the future. So we stop the issuing of I/Os once the amount of * memory used drops below the soft limit (at which point we stop issuing * I/O and start scanning metadata again). * * This limit is calculated by subtracting a fraction of the hard * limit from the hard limit. By default this fraction is 5%, so * the soft limit is 95% of the hard limit. We cap the size of the * difference between the hard and soft limits at an absolute * maximum of zfs_scan_mem_lim_soft_max (default: 128 MiB) - this is * sufficient to not cause too frequent switching between the * metadata scan and I/O issue (even at 2k recordsize, 128 MiB's * worth of queues is about 1.2 GiB of on-pool data, so scanning * that should take at least a decent fraction of a second). */ static boolean_t dsl_scan_should_clear(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; vdev_t *rvd = scn->scn_dp->dp_spa->spa_root_vdev; uint64_t alloc, mlim_hard, mlim_soft, mused; alloc = metaslab_class_get_alloc(spa_normal_class(spa)); alloc += metaslab_class_get_alloc(spa_special_class(spa)); alloc += metaslab_class_get_alloc(spa_dedup_class(spa)); mlim_hard = MAX((physmem / zfs_scan_mem_lim_fact) * PAGESIZE, zfs_scan_mem_lim_min); mlim_hard = MIN(mlim_hard, alloc / 20); mlim_soft = mlim_hard - MIN(mlim_hard / zfs_scan_mem_lim_soft_fact, zfs_scan_mem_lim_soft_max); mused = 0; for (uint64_t i = 0; i < rvd->vdev_children; i++) { vdev_t *tvd = rvd->vdev_child[i]; dsl_scan_io_queue_t *queue; mutex_enter(&tvd->vdev_scan_io_queue_lock); queue = tvd->vdev_scan_io_queue; if (queue != NULL) { /* * # of extents in exts_by_addr = # in exts_by_size. * B-tree efficiency is ~75%, but can be as low as 50%. */ mused += zfs_btree_numnodes(&queue->q_exts_by_size) * ((sizeof (range_seg_gap_t) + sizeof (uint64_t)) * 3 / 2) + queue->q_sio_memused; } mutex_exit(&tvd->vdev_scan_io_queue_lock); } dprintf("current scan memory usage: %llu bytes\n", (longlong_t)mused); if (mused == 0) ASSERT0(scn->scn_queues_pending); /* * If we are above our hard limit, we need to clear out memory. * If we are below our soft limit, we need to accumulate sequential IOs. * Otherwise, we should keep doing whatever we are currently doing. */ if (mused >= mlim_hard) return (B_TRUE); else if (mused < mlim_soft) return (B_FALSE); else return (scn->scn_clearing); } static boolean_t dsl_scan_check_suspend(dsl_scan_t *scn, const zbookmark_phys_t *zb) { /* we never skip user/group accounting objects */ if (zb && (int64_t)zb->zb_object < 0) return (B_FALSE); if (scn->scn_suspending) return (B_TRUE); /* we're already suspending */ if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark)) return (B_FALSE); /* we're resuming */ /* We only know how to resume from level-0 and objset blocks. */ if (zb && (zb->zb_level != 0 && zb->zb_level != ZB_ROOT_LEVEL)) return (B_FALSE); /* * We suspend if: * - we have scanned for at least the minimum time (default 1 sec * for scrub, 3 sec for resilver), and either we have sufficient * dirty data that we are starting to write more quickly * (default 30%), someone is explicitly waiting for this txg * to complete, or we have used up all of the time in the txg * timeout (default 5 sec). * or * - the spa is shutting down because this pool is being exported * or the machine is rebooting. * or * - the scan queue has reached its memory use limit */ uint64_t curr_time_ns = gethrtime(); uint64_t scan_time_ns = curr_time_ns - scn->scn_sync_start_time; uint64_t sync_time_ns = curr_time_ns - scn->scn_dp->dp_spa->spa_sync_starttime; uint64_t dirty_min_bytes = zfs_dirty_data_max * zfs_vdev_async_write_active_min_dirty_percent / 100; int mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ? zfs_resilver_min_time_ms : zfs_scrub_min_time_ms; if ((NSEC2MSEC(scan_time_ns) > mintime && (scn->scn_dp->dp_dirty_total >= dirty_min_bytes || txg_sync_waiting(scn->scn_dp) || NSEC2SEC(sync_time_ns) >= zfs_txg_timeout)) || spa_shutting_down(scn->scn_dp->dp_spa) || (zfs_scan_strict_mem_lim && dsl_scan_should_clear(scn))) { if (zb && zb->zb_level == ZB_ROOT_LEVEL) { dprintf("suspending at first available bookmark " "%llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); SET_BOOKMARK(&scn->scn_phys.scn_bookmark, zb->zb_objset, 0, 0, 0); } else if (zb != NULL) { dprintf("suspending at bookmark %llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); scn->scn_phys.scn_bookmark = *zb; } else { #ifdef ZFS_DEBUG dsl_scan_phys_t *scnp = &scn->scn_phys; dprintf("suspending at at DDT bookmark " "%llx/%llx/%llx/%llx\n", (longlong_t)scnp->scn_ddt_bookmark.ddb_class, (longlong_t)scnp->scn_ddt_bookmark.ddb_type, (longlong_t)scnp->scn_ddt_bookmark.ddb_checksum, (longlong_t)scnp->scn_ddt_bookmark.ddb_cursor); #endif } scn->scn_suspending = B_TRUE; return (B_TRUE); } return (B_FALSE); } typedef struct zil_scan_arg { dsl_pool_t *zsa_dp; zil_header_t *zsa_zh; } zil_scan_arg_t; static int dsl_scan_zil_block(zilog_t *zilog, const blkptr_t *bp, void *arg, uint64_t claim_txg) { (void) zilog; zil_scan_arg_t *zsa = arg; dsl_pool_t *dp = zsa->zsa_dp; dsl_scan_t *scn = dp->dp_scan; zil_header_t *zh = zsa->zsa_zh; zbookmark_phys_t zb; ASSERT(!BP_IS_REDACTED(bp)); if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) return (0); /* * One block ("stubby") can be allocated a long time ago; we * want to visit that one because it has been allocated * (on-disk) even if it hasn't been claimed (even though for * scrub there's nothing to do to it). */ if (claim_txg == 0 && bp->blk_birth >= spa_min_claim_txg(dp->dp_spa)) return (0); SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET], ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb)); return (0); } static int dsl_scan_zil_record(zilog_t *zilog, const lr_t *lrc, void *arg, uint64_t claim_txg) { (void) zilog; if (lrc->lrc_txtype == TX_WRITE) { zil_scan_arg_t *zsa = arg; dsl_pool_t *dp = zsa->zsa_dp; dsl_scan_t *scn = dp->dp_scan; zil_header_t *zh = zsa->zsa_zh; const lr_write_t *lr = (const lr_write_t *)lrc; const blkptr_t *bp = &lr->lr_blkptr; zbookmark_phys_t zb; ASSERT(!BP_IS_REDACTED(bp)); if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) return (0); /* * birth can be < claim_txg if this record's txg is * already txg sync'ed (but this log block contains * other records that are not synced) */ if (claim_txg == 0 || bp->blk_birth < claim_txg) return (0); SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET], lr->lr_foid, ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb)); } return (0); } static void dsl_scan_zil(dsl_pool_t *dp, zil_header_t *zh) { uint64_t claim_txg = zh->zh_claim_txg; zil_scan_arg_t zsa = { dp, zh }; zilog_t *zilog; ASSERT(spa_writeable(dp->dp_spa)); /* * We only want to visit blocks that have been claimed but not yet * replayed (or, in read-only mode, blocks that *would* be claimed). */ if (claim_txg == 0) return; zilog = zil_alloc(dp->dp_meta_objset, zh); (void) zil_parse(zilog, dsl_scan_zil_block, dsl_scan_zil_record, &zsa, claim_txg, B_FALSE); zil_free(zilog); } /* * We compare scan_prefetch_issue_ctx_t's based on their bookmarks. The idea * here is to sort the AVL tree by the order each block will be needed. */ static int scan_prefetch_queue_compare(const void *a, const void *b) { const scan_prefetch_issue_ctx_t *spic_a = a, *spic_b = b; const scan_prefetch_ctx_t *spc_a = spic_a->spic_spc; const scan_prefetch_ctx_t *spc_b = spic_b->spic_spc; return (zbookmark_compare(spc_a->spc_datablkszsec, spc_a->spc_indblkshift, spc_b->spc_datablkszsec, spc_b->spc_indblkshift, &spic_a->spic_zb, &spic_b->spic_zb)); } static void scan_prefetch_ctx_rele(scan_prefetch_ctx_t *spc, void *tag) { if (zfs_refcount_remove(&spc->spc_refcnt, tag) == 0) { zfs_refcount_destroy(&spc->spc_refcnt); kmem_free(spc, sizeof (scan_prefetch_ctx_t)); } } static scan_prefetch_ctx_t * scan_prefetch_ctx_create(dsl_scan_t *scn, dnode_phys_t *dnp, void *tag) { scan_prefetch_ctx_t *spc; spc = kmem_alloc(sizeof (scan_prefetch_ctx_t), KM_SLEEP); zfs_refcount_create(&spc->spc_refcnt); zfs_refcount_add(&spc->spc_refcnt, tag); spc->spc_scn = scn; if (dnp != NULL) { spc->spc_datablkszsec = dnp->dn_datablkszsec; spc->spc_indblkshift = dnp->dn_indblkshift; spc->spc_root = B_FALSE; } else { spc->spc_datablkszsec = 0; spc->spc_indblkshift = 0; spc->spc_root = B_TRUE; } return (spc); } static void scan_prefetch_ctx_add_ref(scan_prefetch_ctx_t *spc, void *tag) { zfs_refcount_add(&spc->spc_refcnt, tag); } static void scan_ds_prefetch_queue_clear(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; void *cookie = NULL; scan_prefetch_issue_ctx_t *spic = NULL; mutex_enter(&spa->spa_scrub_lock); while ((spic = avl_destroy_nodes(&scn->scn_prefetch_queue, &cookie)) != NULL) { scan_prefetch_ctx_rele(spic->spic_spc, scn); kmem_free(spic, sizeof (scan_prefetch_issue_ctx_t)); } mutex_exit(&spa->spa_scrub_lock); } static boolean_t dsl_scan_check_prefetch_resume(scan_prefetch_ctx_t *spc, const zbookmark_phys_t *zb) { zbookmark_phys_t *last_zb = &spc->spc_scn->scn_prefetch_bookmark; dnode_phys_t tmp_dnp; dnode_phys_t *dnp = (spc->spc_root) ? NULL : &tmp_dnp; if (zb->zb_objset != last_zb->zb_objset) return (B_TRUE); if ((int64_t)zb->zb_object < 0) return (B_FALSE); tmp_dnp.dn_datablkszsec = spc->spc_datablkszsec; tmp_dnp.dn_indblkshift = spc->spc_indblkshift; if (zbookmark_subtree_completed(dnp, zb, last_zb)) return (B_TRUE); return (B_FALSE); } static void dsl_scan_prefetch(scan_prefetch_ctx_t *spc, blkptr_t *bp, zbookmark_phys_t *zb) { avl_index_t idx; dsl_scan_t *scn = spc->spc_scn; spa_t *spa = scn->scn_dp->dp_spa; scan_prefetch_issue_ctx_t *spic; if (zfs_no_scrub_prefetch || BP_IS_REDACTED(bp)) return; if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_cur_min_txg || (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) return; if (dsl_scan_check_prefetch_resume(spc, zb)) return; scan_prefetch_ctx_add_ref(spc, scn); spic = kmem_alloc(sizeof (scan_prefetch_issue_ctx_t), KM_SLEEP); spic->spic_spc = spc; spic->spic_bp = *bp; spic->spic_zb = *zb; /* * Add the IO to the queue of blocks to prefetch. This allows us to * prioritize blocks that we will need first for the main traversal * thread. */ mutex_enter(&spa->spa_scrub_lock); if (avl_find(&scn->scn_prefetch_queue, spic, &idx) != NULL) { /* this block is already queued for prefetch */ kmem_free(spic, sizeof (scan_prefetch_issue_ctx_t)); scan_prefetch_ctx_rele(spc, scn); mutex_exit(&spa->spa_scrub_lock); return; } avl_insert(&scn->scn_prefetch_queue, spic, idx); cv_broadcast(&spa->spa_scrub_io_cv); mutex_exit(&spa->spa_scrub_lock); } static void dsl_scan_prefetch_dnode(dsl_scan_t *scn, dnode_phys_t *dnp, uint64_t objset, uint64_t object) { int i; zbookmark_phys_t zb; scan_prefetch_ctx_t *spc; if (dnp->dn_nblkptr == 0 && !(dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) return; SET_BOOKMARK(&zb, objset, object, 0, 0); spc = scan_prefetch_ctx_create(scn, dnp, FTAG); for (i = 0; i < dnp->dn_nblkptr; i++) { zb.zb_level = BP_GET_LEVEL(&dnp->dn_blkptr[i]); zb.zb_blkid = i; dsl_scan_prefetch(spc, &dnp->dn_blkptr[i], &zb); } if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { zb.zb_level = 0; zb.zb_blkid = DMU_SPILL_BLKID; dsl_scan_prefetch(spc, DN_SPILL_BLKPTR(dnp), &zb); } scan_prefetch_ctx_rele(spc, FTAG); } static void dsl_scan_prefetch_cb(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp, arc_buf_t *buf, void *private) { (void) zio; scan_prefetch_ctx_t *spc = private; dsl_scan_t *scn = spc->spc_scn; spa_t *spa = scn->scn_dp->dp_spa; /* broadcast that the IO has completed for rate limiting purposes */ mutex_enter(&spa->spa_scrub_lock); ASSERT3U(spa->spa_scrub_inflight, >=, BP_GET_PSIZE(bp)); spa->spa_scrub_inflight -= BP_GET_PSIZE(bp); cv_broadcast(&spa->spa_scrub_io_cv); mutex_exit(&spa->spa_scrub_lock); /* if there was an error or we are done prefetching, just cleanup */ if (buf == NULL || scn->scn_prefetch_stop) goto out; if (BP_GET_LEVEL(bp) > 0) { int i; blkptr_t *cbp; int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; zbookmark_phys_t czb; for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) { SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, zb->zb_level - 1, zb->zb_blkid * epb + i); dsl_scan_prefetch(spc, cbp, &czb); } } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { dnode_phys_t *cdnp; int i; int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; for (i = 0, cdnp = buf->b_data; i < epb; i += cdnp->dn_extra_slots + 1, cdnp += cdnp->dn_extra_slots + 1) { dsl_scan_prefetch_dnode(scn, cdnp, zb->zb_objset, zb->zb_blkid * epb + i); } } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { objset_phys_t *osp = buf->b_data; dsl_scan_prefetch_dnode(scn, &osp->os_meta_dnode, zb->zb_objset, DMU_META_DNODE_OBJECT); if (OBJSET_BUF_HAS_USERUSED(buf)) { dsl_scan_prefetch_dnode(scn, &osp->os_groupused_dnode, zb->zb_objset, DMU_GROUPUSED_OBJECT); dsl_scan_prefetch_dnode(scn, &osp->os_userused_dnode, zb->zb_objset, DMU_USERUSED_OBJECT); } } out: if (buf != NULL) arc_buf_destroy(buf, private); scan_prefetch_ctx_rele(spc, scn); } static void dsl_scan_prefetch_thread(void *arg) { dsl_scan_t *scn = arg; spa_t *spa = scn->scn_dp->dp_spa; scan_prefetch_issue_ctx_t *spic; /* loop until we are told to stop */ while (!scn->scn_prefetch_stop) { arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PRESCIENT_PREFETCH | ARC_FLAG_PREFETCH; int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD; mutex_enter(&spa->spa_scrub_lock); /* * Wait until we have an IO to issue and are not above our * maximum in flight limit. */ while (!scn->scn_prefetch_stop && (avl_numnodes(&scn->scn_prefetch_queue) == 0 || spa->spa_scrub_inflight >= scn->scn_maxinflight_bytes)) { cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); } /* recheck if we should stop since we waited for the cv */ if (scn->scn_prefetch_stop) { mutex_exit(&spa->spa_scrub_lock); break; } /* remove the prefetch IO from the tree */ spic = avl_first(&scn->scn_prefetch_queue); spa->spa_scrub_inflight += BP_GET_PSIZE(&spic->spic_bp); avl_remove(&scn->scn_prefetch_queue, spic); mutex_exit(&spa->spa_scrub_lock); if (BP_IS_PROTECTED(&spic->spic_bp)) { ASSERT(BP_GET_TYPE(&spic->spic_bp) == DMU_OT_DNODE || BP_GET_TYPE(&spic->spic_bp) == DMU_OT_OBJSET); ASSERT3U(BP_GET_LEVEL(&spic->spic_bp), ==, 0); zio_flags |= ZIO_FLAG_RAW; } /* issue the prefetch asynchronously */ (void) arc_read(scn->scn_zio_root, scn->scn_dp->dp_spa, &spic->spic_bp, dsl_scan_prefetch_cb, spic->spic_spc, ZIO_PRIORITY_SCRUB, zio_flags, &flags, &spic->spic_zb); kmem_free(spic, sizeof (scan_prefetch_issue_ctx_t)); } ASSERT(scn->scn_prefetch_stop); /* free any prefetches we didn't get to complete */ mutex_enter(&spa->spa_scrub_lock); while ((spic = avl_first(&scn->scn_prefetch_queue)) != NULL) { avl_remove(&scn->scn_prefetch_queue, spic); scan_prefetch_ctx_rele(spic->spic_spc, scn); kmem_free(spic, sizeof (scan_prefetch_issue_ctx_t)); } ASSERT0(avl_numnodes(&scn->scn_prefetch_queue)); mutex_exit(&spa->spa_scrub_lock); } static boolean_t dsl_scan_check_resume(dsl_scan_t *scn, const dnode_phys_t *dnp, const zbookmark_phys_t *zb) { /* * We never skip over user/group accounting objects (obj<0) */ if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark) && (int64_t)zb->zb_object >= 0) { /* * If we already visited this bp & everything below (in * a prior txg sync), don't bother doing it again. */ if (zbookmark_subtree_completed(dnp, zb, &scn->scn_phys.scn_bookmark)) return (B_TRUE); /* * If we found the block we're trying to resume from, or * we went past it, zero it out to indicate that it's OK * to start checking for suspending again. */ if (zbookmark_subtree_tbd(dnp, zb, &scn->scn_phys.scn_bookmark)) { dprintf("resuming at %llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); bzero(&scn->scn_phys.scn_bookmark, sizeof (*zb)); } } return (B_FALSE); } static void dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb, dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype, dmu_tx_t *tx); inline __attribute__((always_inline)) static void dsl_scan_visitdnode( dsl_scan_t *, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, uint64_t object, dmu_tx_t *tx); /* * Return nonzero on i/o error. * Return new buf to write out in *bufp. */ inline __attribute__((always_inline)) static int dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, const blkptr_t *bp, const zbookmark_phys_t *zb, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD; int err; ASSERT(!BP_IS_REDACTED(bp)); /* * There is an unlikely case of encountering dnodes with contradicting * dn_bonuslen and DNODE_FLAG_SPILL_BLKPTR flag before in files created * or modified before commit 4254acb was merged. As it is not possible * to know which of the two is correct, report an error. */ if (dnp != NULL && dnp->dn_bonuslen > DN_MAX_BONUS_LEN(dnp)) { scn->scn_phys.scn_errors++; spa_log_error(dp->dp_spa, zb); return (SET_ERROR(EINVAL)); } if (BP_GET_LEVEL(bp) > 0) { arc_flags_t flags = ARC_FLAG_WAIT; int i; blkptr_t *cbp; int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; arc_buf_t *buf; err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_SCRUB, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) { zbookmark_phys_t czb; SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, zb->zb_level - 1, zb->zb_blkid * epb + i); dsl_scan_visitbp(cbp, &czb, dnp, ds, scn, ostype, tx); } arc_buf_destroy(buf, &buf); } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { arc_flags_t flags = ARC_FLAG_WAIT; dnode_phys_t *cdnp; int i; int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; arc_buf_t *buf; if (BP_IS_PROTECTED(bp)) { ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); zio_flags |= ZIO_FLAG_RAW; } err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_SCRUB, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } for (i = 0, cdnp = buf->b_data; i < epb; i += cdnp->dn_extra_slots + 1, cdnp += cdnp->dn_extra_slots + 1) { dsl_scan_visitdnode(scn, ds, ostype, cdnp, zb->zb_blkid * epb + i, tx); } arc_buf_destroy(buf, &buf); } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { arc_flags_t flags = ARC_FLAG_WAIT; objset_phys_t *osp; arc_buf_t *buf; err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_SCRUB, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } osp = buf->b_data; dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_meta_dnode, DMU_META_DNODE_OBJECT, tx); if (OBJSET_BUF_HAS_USERUSED(buf)) { /* * We also always visit user/group/project accounting * objects, and never skip them, even if we are * suspending. This is necessary so that the * space deltas from this txg get integrated. */ if (OBJSET_BUF_HAS_PROJECTUSED(buf)) dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_projectused_dnode, DMU_PROJECTUSED_OBJECT, tx); dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_groupused_dnode, DMU_GROUPUSED_OBJECT, tx); dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_userused_dnode, DMU_USERUSED_OBJECT, tx); } arc_buf_destroy(buf, &buf); } return (0); } inline __attribute__((always_inline)) static void dsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, uint64_t object, dmu_tx_t *tx) { int j; for (j = 0; j < dnp->dn_nblkptr; j++) { zbookmark_phys_t czb; SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object, dnp->dn_nlevels - 1, j); dsl_scan_visitbp(&dnp->dn_blkptr[j], &czb, dnp, ds, scn, ostype, tx); } if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { zbookmark_phys_t czb; SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object, 0, DMU_SPILL_BLKID); dsl_scan_visitbp(DN_SPILL_BLKPTR(dnp), &czb, dnp, ds, scn, ostype, tx); } } /* * The arguments are in this order because mdb can only print the * first 5; we want them to be useful. */ static void dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb, dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; blkptr_t *bp_toread = NULL; if (dsl_scan_check_suspend(scn, zb)) return; if (dsl_scan_check_resume(scn, dnp, zb)) return; scn->scn_visited_this_txg++; /* * This debugging is commented out to conserve stack space. This * function is called recursively and the debugging adds several * bytes to the stack for each call. It can be commented back in * if required to debug an issue in dsl_scan_visitbp(). * * dprintf_bp(bp, * "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx bp=%p", * ds, ds ? ds->ds_object : 0, * zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid, * bp); */ if (BP_IS_HOLE(bp)) { scn->scn_holes_this_txg++; return; } if (BP_IS_REDACTED(bp)) { ASSERT(dsl_dataset_feature_is_active(ds, SPA_FEATURE_REDACTED_DATASETS)); return; } /* * Check if this block contradicts any filesystem flags. */ spa_feature_t f = SPA_FEATURE_LARGE_BLOCKS; if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) ASSERT(dsl_dataset_feature_is_active(ds, f)); f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); if (f != SPA_FEATURE_NONE) ASSERT(dsl_dataset_feature_is_active(ds, f)); f = zio_compress_to_feature(BP_GET_COMPRESS(bp)); if (f != SPA_FEATURE_NONE) ASSERT(dsl_dataset_feature_is_active(ds, f)); if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) { scn->scn_lt_min_this_txg++; return; } - if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) { - scn->scn_lt_min_this_txg++; - return; - } - bp_toread = kmem_alloc(sizeof (blkptr_t), KM_SLEEP); *bp_toread = *bp; if (dsl_scan_recurse(scn, ds, ostype, dnp, bp_toread, zb, tx) != 0) goto out; /* * If dsl_scan_ddt() has already visited this block, it will have * already done any translations or scrubbing, so don't call the * callback again. */ if (ddt_class_contains(dp->dp_spa, scn->scn_phys.scn_ddt_class_max, bp)) { scn->scn_ddt_contained_this_txg++; goto out; } /* * If this block is from the future (after cur_max_txg), then we * are doing this on behalf of a deleted snapshot, and we will * revisit the future block on the next pass of this dataset. * Don't scan it now unless we need to because something * under it was modified. */ if (BP_PHYSICAL_BIRTH(bp) > scn->scn_phys.scn_cur_max_txg) { scn->scn_gt_max_this_txg++; goto out; } scan_funcs[scn->scn_phys.scn_func](dp, bp, zb); out: kmem_free(bp_toread, sizeof (blkptr_t)); } static void dsl_scan_visit_rootbp(dsl_scan_t *scn, dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) { zbookmark_phys_t zb; scan_prefetch_ctx_t *spc; SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); if (ZB_IS_ZERO(&scn->scn_phys.scn_bookmark)) { SET_BOOKMARK(&scn->scn_prefetch_bookmark, zb.zb_objset, 0, 0, 0); } else { scn->scn_prefetch_bookmark = scn->scn_phys.scn_bookmark; } scn->scn_objsets_visited_this_txg++; spc = scan_prefetch_ctx_create(scn, NULL, FTAG); dsl_scan_prefetch(spc, bp, &zb); scan_prefetch_ctx_rele(spc, FTAG); dsl_scan_visitbp(bp, &zb, NULL, ds, scn, DMU_OST_NONE, tx); dprintf_ds(ds, "finished scan%s", ""); } static void ds_destroyed_scn_phys(dsl_dataset_t *ds, dsl_scan_phys_t *scn_phys) { if (scn_phys->scn_bookmark.zb_objset == ds->ds_object) { if (ds->ds_is_snapshot) { /* * Note: * - scn_cur_{min,max}_txg stays the same. * - Setting the flag is not really necessary if * scn_cur_max_txg == scn_max_txg, because there * is nothing after this snapshot that we care * about. However, we set it anyway and then * ignore it when we retraverse it in * dsl_scan_visitds(). */ scn_phys->scn_bookmark.zb_objset = dsl_dataset_phys(ds)->ds_next_snap_obj; zfs_dbgmsg("destroying ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)dsl_dataset_phys(ds)-> ds_next_snap_obj); scn_phys->scn_flags |= DSF_VISIT_DS_AGAIN; } else { SET_BOOKMARK(&scn_phys->scn_bookmark, ZB_DESTROYED_OBJSET, 0, 0, 0); zfs_dbgmsg("destroying ds %llu; currently traversing; " "reset bookmark to -1,0,0,0", (u_longlong_t)ds->ds_object); } } } /* * Invoked when a dataset is destroyed. We need to make sure that: * * 1) If it is the dataset that was currently being scanned, we write * a new dsl_scan_phys_t and marking the objset reference in it * as destroyed. * 2) Remove it from the work queue, if it was present. * * If the dataset was actually a snapshot, instead of marking the dataset * as destroyed, we instead substitute the next snapshot in line. */ void dsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg; if (!dsl_scan_is_running(scn)) return; ds_destroyed_scn_phys(ds, &scn->scn_phys); ds_destroyed_scn_phys(ds, &scn->scn_phys_cached); if (scan_ds_queue_contains(scn, ds->ds_object, &mintxg)) { scan_ds_queue_remove(scn, ds->ds_object); if (ds->ds_is_snapshot) scan_ds_queue_insert(scn, dsl_dataset_phys(ds)->ds_next_snap_obj, mintxg); } if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) { ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1); VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, tx)); if (ds->ds_is_snapshot) { /* * We keep the same mintxg; it could be > * ds_creation_txg if the previous snapshot was * deleted too. */ VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, dsl_dataset_phys(ds)->ds_next_snap_obj, mintxg, tx) == 0); zfs_dbgmsg("destroying ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)dsl_dataset_phys(ds)-> ds_next_snap_obj); } else { zfs_dbgmsg("destroying ds %llu; in queue; removing", (u_longlong_t)ds->ds_object); } } /* * dsl_scan_sync() should be called after this, and should sync * out our changed state, but just to be safe, do it here. */ dsl_scan_sync_state(scn, tx, SYNC_CACHED); } static void ds_snapshotted_bookmark(dsl_dataset_t *ds, zbookmark_phys_t *scn_bookmark) { if (scn_bookmark->zb_objset == ds->ds_object) { scn_bookmark->zb_objset = dsl_dataset_phys(ds)->ds_prev_snap_obj; zfs_dbgmsg("snapshotting ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj); } } /* * Called when a dataset is snapshotted. If we were currently traversing * this snapshot, we reset our bookmark to point at the newly created * snapshot. We also modify our work queue to remove the old snapshot and * replace with the new one. */ void dsl_scan_ds_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg; if (!dsl_scan_is_running(scn)) return; ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); ds_snapshotted_bookmark(ds, &scn->scn_phys.scn_bookmark); ds_snapshotted_bookmark(ds, &scn->scn_phys_cached.scn_bookmark); if (scan_ds_queue_contains(scn, ds->ds_object, &mintxg)) { scan_ds_queue_remove(scn, ds->ds_object); scan_ds_queue_insert(scn, dsl_dataset_phys(ds)->ds_prev_snap_obj, mintxg); } if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) { VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, tx)); VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, dsl_dataset_phys(ds)->ds_prev_snap_obj, mintxg, tx) == 0); zfs_dbgmsg("snapshotting ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj); } dsl_scan_sync_state(scn, tx, SYNC_CACHED); } static void ds_clone_swapped_bookmark(dsl_dataset_t *ds1, dsl_dataset_t *ds2, zbookmark_phys_t *scn_bookmark) { if (scn_bookmark->zb_objset == ds1->ds_object) { scn_bookmark->zb_objset = ds2->ds_object; zfs_dbgmsg("clone_swap ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds1->ds_object, (u_longlong_t)ds2->ds_object); } else if (scn_bookmark->zb_objset == ds2->ds_object) { scn_bookmark->zb_objset = ds1->ds_object; zfs_dbgmsg("clone_swap ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds2->ds_object, (u_longlong_t)ds1->ds_object); } } /* * Called when an origin dataset and its clone are swapped. If we were * currently traversing the dataset, we need to switch to traversing the * newly promoted clone. */ void dsl_scan_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx) { dsl_pool_t *dp = ds1->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg1, mintxg2; boolean_t ds1_queued, ds2_queued; if (!dsl_scan_is_running(scn)) return; ds_clone_swapped_bookmark(ds1, ds2, &scn->scn_phys.scn_bookmark); ds_clone_swapped_bookmark(ds1, ds2, &scn->scn_phys_cached.scn_bookmark); /* * Handle the in-memory scan queue. */ ds1_queued = scan_ds_queue_contains(scn, ds1->ds_object, &mintxg1); ds2_queued = scan_ds_queue_contains(scn, ds2->ds_object, &mintxg2); /* Sanity checking. */ if (ds1_queued) { ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); } if (ds2_queued) { ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); } if (ds1_queued && ds2_queued) { /* * If both are queued, we don't need to do anything. * The swapping code below would not handle this case correctly, * since we can't insert ds2 if it is already there. That's * because scan_ds_queue_insert() prohibits a duplicate insert * and panics. */ } else if (ds1_queued) { scan_ds_queue_remove(scn, ds1->ds_object); scan_ds_queue_insert(scn, ds2->ds_object, mintxg1); } else if (ds2_queued) { scan_ds_queue_remove(scn, ds2->ds_object); scan_ds_queue_insert(scn, ds1->ds_object, mintxg2); } /* * Handle the on-disk scan queue. * The on-disk state is an out-of-date version of the in-memory state, * so the in-memory and on-disk values for ds1_queued and ds2_queued may * be different. Therefore we need to apply the swap logic to the * on-disk state independently of the in-memory state. */ ds1_queued = zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, &mintxg1) == 0; ds2_queued = zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, &mintxg2) == 0; /* Sanity checking. */ if (ds1_queued) { ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); } if (ds2_queued) { ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); } if (ds1_queued && ds2_queued) { /* * If both are queued, we don't need to do anything. * Alternatively, we could check for EEXIST from * zap_add_int_key() and back out to the original state, but * that would be more work than checking for this case upfront. */ } else if (ds1_queued) { VERIFY3S(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, tx)); VERIFY3S(0, ==, zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg1, tx)); zfs_dbgmsg("clone_swap ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds1->ds_object, (u_longlong_t)ds2->ds_object); } else if (ds2_queued) { VERIFY3S(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, tx)); VERIFY3S(0, ==, zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg2, tx)); zfs_dbgmsg("clone_swap ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds2->ds_object, (u_longlong_t)ds1->ds_object); } dsl_scan_sync_state(scn, tx, SYNC_CACHED); } static int enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { uint64_t originobj = *(uint64_t *)arg; dsl_dataset_t *ds; int err; dsl_scan_t *scn = dp->dp_scan; if (dsl_dir_phys(hds->ds_dir)->dd_origin_obj != originobj) return (0); err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); if (err) return (err); while (dsl_dataset_phys(ds)->ds_prev_snap_obj != originobj) { dsl_dataset_t *prev; err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); dsl_dataset_rele(ds, FTAG); if (err) return (err); ds = prev; } scan_ds_queue_insert(scn, ds->ds_object, dsl_dataset_phys(ds)->ds_prev_snap_txg); dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; dsl_dataset_t *ds; VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); if (scn->scn_phys.scn_cur_min_txg >= scn->scn_phys.scn_max_txg) { /* * This can happen if this snapshot was created after the * scan started, and we already completed a previous snapshot * that was created after the scan started. This snapshot * only references blocks with: * * birth < our ds_creation_txg * cur_min_txg is no less than ds_creation_txg. * We have already visited these blocks. * or * birth > scn_max_txg * The scan requested not to visit these blocks. * * Subsequent snapshots (and clones) can reference our * blocks, or blocks with even higher birth times. * Therefore we do not need to visit them either, * so we do not add them to the work queue. * * Note that checking for cur_min_txg >= cur_max_txg * is not sufficient, because in that case we may need to * visit subsequent snapshots. This happens when min_txg > 0, * which raises cur_min_txg. In this case we will visit * this dataset but skip all of its blocks, because the * rootbp's birth time is < cur_min_txg. Then we will * add the next snapshots/clones to the work queue. */ char *dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); dsl_dataset_name(ds, dsname); zfs_dbgmsg("scanning dataset %llu (%s) is unnecessary because " "cur_min_txg (%llu) >= max_txg (%llu)", (longlong_t)dsobj, dsname, (longlong_t)scn->scn_phys.scn_cur_min_txg, (longlong_t)scn->scn_phys.scn_max_txg); kmem_free(dsname, MAXNAMELEN); goto out; } /* * Only the ZIL in the head (non-snapshot) is valid. Even though * snapshots can have ZIL block pointers (which may be the same * BP as in the head), they must be ignored. In addition, $ORIGIN * doesn't have a objset (i.e. its ds_bp is a hole) so we don't * need to look for a ZIL in it either. So we traverse the ZIL here, * rather than in scan_recurse(), because the regular snapshot * block-sharing rules don't apply to it. */ if (!dsl_dataset_is_snapshot(ds) && (dp->dp_origin_snap == NULL || ds->ds_dir != dp->dp_origin_snap->ds_dir)) { objset_t *os; if (dmu_objset_from_ds(ds, &os) != 0) { goto out; } dsl_scan_zil(dp, &os->os_zil_header); } /* * Iterate over the bps in this ds. */ dmu_buf_will_dirty(ds->ds_dbuf, tx); rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); dsl_scan_visit_rootbp(scn, ds, &dsl_dataset_phys(ds)->ds_bp, tx); rrw_exit(&ds->ds_bp_rwlock, FTAG); char *dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); dsl_dataset_name(ds, dsname); zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; " "suspending=%u", (longlong_t)dsobj, dsname, (longlong_t)scn->scn_phys.scn_cur_min_txg, (longlong_t)scn->scn_phys.scn_cur_max_txg, (int)scn->scn_suspending); kmem_free(dsname, ZFS_MAX_DATASET_NAME_LEN); if (scn->scn_suspending) goto out; /* * We've finished this pass over this dataset. */ /* * If we did not completely visit this dataset, do another pass. */ if (scn->scn_phys.scn_flags & DSF_VISIT_DS_AGAIN) { zfs_dbgmsg("incomplete pass; visiting again"); scn->scn_phys.scn_flags &= ~DSF_VISIT_DS_AGAIN; scan_ds_queue_insert(scn, ds->ds_object, scn->scn_phys.scn_cur_max_txg); goto out; } /* * Add descendant datasets to work queue. */ if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) { scan_ds_queue_insert(scn, dsl_dataset_phys(ds)->ds_next_snap_obj, dsl_dataset_phys(ds)->ds_creation_txg); } if (dsl_dataset_phys(ds)->ds_num_children > 1) { boolean_t usenext = B_FALSE; if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { uint64_t count; /* * A bug in a previous version of the code could * cause upgrade_clones_cb() to not set * ds_next_snap_obj when it should, leading to a * missing entry. Therefore we can only use the * next_clones_obj when its count is correct. */ int err = zap_count(dp->dp_meta_objset, dsl_dataset_phys(ds)->ds_next_clones_obj, &count); if (err == 0 && count == dsl_dataset_phys(ds)->ds_num_children - 1) usenext = B_TRUE; } if (usenext) { zap_cursor_t zc; zap_attribute_t za; for (zap_cursor_init(&zc, dp->dp_meta_objset, dsl_dataset_phys(ds)->ds_next_clones_obj); zap_cursor_retrieve(&zc, &za) == 0; (void) zap_cursor_advance(&zc)) { scan_ds_queue_insert(scn, zfs_strtonum(za.za_name, NULL), dsl_dataset_phys(ds)->ds_creation_txg); } zap_cursor_fini(&zc); } else { VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, enqueue_clones_cb, &ds->ds_object, DS_FIND_CHILDREN)); } } out: dsl_dataset_rele(ds, FTAG); } static int enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { (void) arg; dsl_dataset_t *ds; int err; dsl_scan_t *scn = dp->dp_scan; err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); if (err) return (err); while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { dsl_dataset_t *prev; err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); if (err) { dsl_dataset_rele(ds, FTAG); return (err); } /* * If this is a clone, we don't need to worry about it for now. */ if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) { dsl_dataset_rele(ds, FTAG); dsl_dataset_rele(prev, FTAG); return (0); } dsl_dataset_rele(ds, FTAG); ds = prev; } scan_ds_queue_insert(scn, ds->ds_object, dsl_dataset_phys(ds)->ds_prev_snap_txg); dsl_dataset_rele(ds, FTAG); return (0); } void dsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum, ddt_entry_t *dde, dmu_tx_t *tx) { (void) tx; const ddt_key_t *ddk = &dde->dde_key; ddt_phys_t *ddp = dde->dde_phys; blkptr_t bp; zbookmark_phys_t zb = { 0 }; if (!dsl_scan_is_running(scn)) return; /* * This function is special because it is the only thing * that can add scan_io_t's to the vdev scan queues from * outside dsl_scan_sync(). For the most part this is ok * as long as it is called from within syncing context. * However, dsl_scan_sync() expects that no new sio's will * be added between when all the work for a scan is done * and the next txg when the scan is actually marked as * completed. This check ensures we do not issue new sio's * during this period. */ if (scn->scn_done_txg != 0) return; for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { if (ddp->ddp_phys_birth == 0 || ddp->ddp_phys_birth > scn->scn_phys.scn_max_txg) continue; ddt_bp_create(checksum, ddk, ddp, &bp); scn->scn_visited_this_txg++; scan_funcs[scn->scn_phys.scn_func](scn->scn_dp, &bp, &zb); } } /* * Scrub/dedup interaction. * * If there are N references to a deduped block, we don't want to scrub it * N times -- ideally, we should scrub it exactly once. * * We leverage the fact that the dde's replication class (enum ddt_class) * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order. * * To prevent excess scrubbing, the scrub begins by walking the DDT * to find all blocks with refcnt > 1, and scrubs each of these once. * Since there are two replication classes which contain blocks with * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first. * Finally the top-down scrub begins, only visiting blocks with refcnt == 1. * * There would be nothing more to say if a block's refcnt couldn't change * during a scrub, but of course it can so we must account for changes * in a block's replication class. * * Here's an example of what can occur: * * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1 * when visited during the top-down scrub phase, it will be scrubbed twice. * This negates our scrub optimization, but is otherwise harmless. * * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1 * on each visit during the top-down scrub phase, it will never be scrubbed. * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1 * while a scrub is in progress, it scrubs the block right then. */ static void dsl_scan_ddt(dsl_scan_t *scn, dmu_tx_t *tx) { ddt_bookmark_t *ddb = &scn->scn_phys.scn_ddt_bookmark; ddt_entry_t dde; int error; uint64_t n = 0; bzero(&dde, sizeof (ddt_entry_t)); while ((error = ddt_walk(scn->scn_dp->dp_spa, ddb, &dde)) == 0) { ddt_t *ddt; if (ddb->ddb_class > scn->scn_phys.scn_ddt_class_max) break; dprintf("visiting ddb=%llu/%llu/%llu/%llx\n", (longlong_t)ddb->ddb_class, (longlong_t)ddb->ddb_type, (longlong_t)ddb->ddb_checksum, (longlong_t)ddb->ddb_cursor); /* There should be no pending changes to the dedup table */ ddt = scn->scn_dp->dp_spa->spa_ddt[ddb->ddb_checksum]; ASSERT(avl_first(&ddt->ddt_tree) == NULL); dsl_scan_ddt_entry(scn, ddb->ddb_checksum, &dde, tx); n++; if (dsl_scan_check_suspend(scn, NULL)) break; } zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; " "suspending=%u", (longlong_t)n, (int)scn->scn_phys.scn_ddt_class_max, (int)scn->scn_suspending); ASSERT(error == 0 || error == ENOENT); ASSERT(error != ENOENT || ddb->ddb_class > scn->scn_phys.scn_ddt_class_max); } static uint64_t dsl_scan_ds_maxtxg(dsl_dataset_t *ds) { uint64_t smt = ds->ds_dir->dd_pool->dp_scan->scn_phys.scn_max_txg; if (ds->ds_is_snapshot) return (MIN(smt, dsl_dataset_phys(ds)->ds_creation_txg)); return (smt); } static void dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx) { scan_ds_t *sds; dsl_pool_t *dp = scn->scn_dp; if (scn->scn_phys.scn_ddt_bookmark.ddb_class <= scn->scn_phys.scn_ddt_class_max) { scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg; scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg; dsl_scan_ddt(scn, tx); if (scn->scn_suspending) return; } if (scn->scn_phys.scn_bookmark.zb_objset == DMU_META_OBJSET) { /* First do the MOS & ORIGIN */ scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg; scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg; dsl_scan_visit_rootbp(scn, NULL, &dp->dp_meta_rootbp, tx); spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); if (scn->scn_suspending) return; if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) { VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, enqueue_cb, NULL, DS_FIND_CHILDREN)); } else { dsl_scan_visitds(scn, dp->dp_origin_snap->ds_object, tx); } ASSERT(!scn->scn_suspending); } else if (scn->scn_phys.scn_bookmark.zb_objset != ZB_DESTROYED_OBJSET) { uint64_t dsobj = scn->scn_phys.scn_bookmark.zb_objset; /* * If we were suspended, continue from here. Note if the * ds we were suspended on was deleted, the zb_objset may * be -1, so we will skip this and find a new objset * below. */ dsl_scan_visitds(scn, dsobj, tx); if (scn->scn_suspending) return; } /* * In case we suspended right at the end of the ds, zero the * bookmark so we don't think that we're still trying to resume. */ bzero(&scn->scn_phys.scn_bookmark, sizeof (zbookmark_phys_t)); /* * Keep pulling things out of the dataset avl queue. Updates to the * persistent zap-object-as-queue happen only at checkpoints. */ while ((sds = avl_first(&scn->scn_queue)) != NULL) { dsl_dataset_t *ds; uint64_t dsobj = sds->sds_dsobj; uint64_t txg = sds->sds_txg; /* dequeue and free the ds from the queue */ scan_ds_queue_remove(scn, dsobj); sds = NULL; /* set up min / max txg */ VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); if (txg != 0) { scn->scn_phys.scn_cur_min_txg = MAX(scn->scn_phys.scn_min_txg, txg); } else { scn->scn_phys.scn_cur_min_txg = MAX(scn->scn_phys.scn_min_txg, dsl_dataset_phys(ds)->ds_prev_snap_txg); } scn->scn_phys.scn_cur_max_txg = dsl_scan_ds_maxtxg(ds); dsl_dataset_rele(ds, FTAG); dsl_scan_visitds(scn, dsobj, tx); if (scn->scn_suspending) return; } /* No more objsets to fetch, we're done */ scn->scn_phys.scn_bookmark.zb_objset = ZB_DESTROYED_OBJSET; ASSERT0(scn->scn_suspending); } static uint64_t dsl_scan_count_data_disks(vdev_t *rvd) { uint64_t i, leaves = 0; for (i = 0; i < rvd->vdev_children; i++) { vdev_t *vd = rvd->vdev_child[i]; if (vd->vdev_islog || vd->vdev_isspare || vd->vdev_isl2cache) continue; leaves += vdev_get_ndisks(vd) - vdev_get_nparity(vd); } return (leaves); } static void scan_io_queues_update_zio_stats(dsl_scan_io_queue_t *q, const blkptr_t *bp) { int i; uint64_t cur_size = 0; for (i = 0; i < BP_GET_NDVAS(bp); i++) { cur_size += DVA_GET_ASIZE(&bp->blk_dva[i]); } q->q_total_zio_size_this_txg += cur_size; q->q_zios_this_txg++; } static void scan_io_queues_update_seg_stats(dsl_scan_io_queue_t *q, uint64_t start, uint64_t end) { q->q_total_seg_size_this_txg += end - start; q->q_segs_this_txg++; } static boolean_t scan_io_queue_check_suspend(dsl_scan_t *scn) { /* See comment in dsl_scan_check_suspend() */ uint64_t curr_time_ns = gethrtime(); uint64_t scan_time_ns = curr_time_ns - scn->scn_sync_start_time; uint64_t sync_time_ns = curr_time_ns - scn->scn_dp->dp_spa->spa_sync_starttime; uint64_t dirty_min_bytes = zfs_dirty_data_max * zfs_vdev_async_write_active_min_dirty_percent / 100; int mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ? zfs_resilver_min_time_ms : zfs_scrub_min_time_ms; return ((NSEC2MSEC(scan_time_ns) > mintime && (scn->scn_dp->dp_dirty_total >= dirty_min_bytes || txg_sync_waiting(scn->scn_dp) || NSEC2SEC(sync_time_ns) >= zfs_txg_timeout)) || spa_shutting_down(scn->scn_dp->dp_spa)); } /* * Given a list of scan_io_t's in io_list, this issues the I/Os out to * disk. This consumes the io_list and frees the scan_io_t's. This is * called when emptying queues, either when we're up against the memory * limit or when we have finished scanning. Returns B_TRUE if we stopped * processing the list before we finished. Any sios that were not issued * will remain in the io_list. */ static boolean_t scan_io_queue_issue(dsl_scan_io_queue_t *queue, list_t *io_list) { dsl_scan_t *scn = queue->q_scn; scan_io_t *sio; boolean_t suspended = B_FALSE; while ((sio = list_head(io_list)) != NULL) { blkptr_t bp; if (scan_io_queue_check_suspend(scn)) { suspended = B_TRUE; break; } sio2bp(sio, &bp); scan_exec_io(scn->scn_dp, &bp, sio->sio_flags, &sio->sio_zb, queue); (void) list_remove_head(io_list); scan_io_queues_update_zio_stats(queue, &bp); sio_free(sio); } return (suspended); } /* * This function removes sios from an IO queue which reside within a given * range_seg_t and inserts them (in offset order) into a list. Note that * we only ever return a maximum of 32 sios at once. If there are more sios * to process within this segment that did not make it onto the list we * return B_TRUE and otherwise B_FALSE. */ static boolean_t scan_io_queue_gather(dsl_scan_io_queue_t *queue, range_seg_t *rs, list_t *list) { scan_io_t *srch_sio, *sio, *next_sio; avl_index_t idx; uint_t num_sios = 0; int64_t bytes_issued = 0; ASSERT(rs != NULL); ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock)); srch_sio = sio_alloc(1); srch_sio->sio_nr_dvas = 1; SIO_SET_OFFSET(srch_sio, rs_get_start(rs, queue->q_exts_by_addr)); /* * The exact start of the extent might not contain any matching zios, * so if that's the case, examine the next one in the tree. */ sio = avl_find(&queue->q_sios_by_addr, srch_sio, &idx); sio_free(srch_sio); if (sio == NULL) sio = avl_nearest(&queue->q_sios_by_addr, idx, AVL_AFTER); while (sio != NULL && SIO_GET_OFFSET(sio) < rs_get_end(rs, queue->q_exts_by_addr) && num_sios <= 32) { ASSERT3U(SIO_GET_OFFSET(sio), >=, rs_get_start(rs, queue->q_exts_by_addr)); ASSERT3U(SIO_GET_END_OFFSET(sio), <=, rs_get_end(rs, queue->q_exts_by_addr)); next_sio = AVL_NEXT(&queue->q_sios_by_addr, sio); avl_remove(&queue->q_sios_by_addr, sio); if (avl_is_empty(&queue->q_sios_by_addr)) atomic_add_64(&queue->q_scn->scn_queues_pending, -1); queue->q_sio_memused -= SIO_GET_MUSED(sio); bytes_issued += SIO_GET_ASIZE(sio); num_sios++; list_insert_tail(list, sio); sio = next_sio; } /* * We limit the number of sios we process at once to 32 to avoid * biting off more than we can chew. If we didn't take everything * in the segment we update it to reflect the work we were able to * complete. Otherwise, we remove it from the range tree entirely. */ if (sio != NULL && SIO_GET_OFFSET(sio) < rs_get_end(rs, queue->q_exts_by_addr)) { range_tree_adjust_fill(queue->q_exts_by_addr, rs, -bytes_issued); range_tree_resize_segment(queue->q_exts_by_addr, rs, SIO_GET_OFFSET(sio), rs_get_end(rs, queue->q_exts_by_addr) - SIO_GET_OFFSET(sio)); queue->q_last_ext_addr = SIO_GET_OFFSET(sio); return (B_TRUE); } else { uint64_t rstart = rs_get_start(rs, queue->q_exts_by_addr); uint64_t rend = rs_get_end(rs, queue->q_exts_by_addr); range_tree_remove(queue->q_exts_by_addr, rstart, rend - rstart); queue->q_last_ext_addr = -1; return (B_FALSE); } } /* * This is called from the queue emptying thread and selects the next * extent from which we are to issue I/Os. The behavior of this function * depends on the state of the scan, the current memory consumption and * whether or not we are performing a scan shutdown. * 1) We select extents in an elevator algorithm (LBA-order) if the scan * needs to perform a checkpoint * 2) We select the largest available extent if we are up against the * memory limit. * 3) Otherwise we don't select any extents. */ static range_seg_t * scan_io_queue_fetch_ext(dsl_scan_io_queue_t *queue) { dsl_scan_t *scn = queue->q_scn; range_tree_t *rt = queue->q_exts_by_addr; ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock)); ASSERT(scn->scn_is_sorted); if (!scn->scn_checkpointing && !scn->scn_clearing) return (NULL); /* * During normal clearing, we want to issue our largest segments * first, keeping IO as sequential as possible, and leaving the * smaller extents for later with the hope that they might eventually * grow to larger sequential segments. However, when the scan is * checkpointing, no new extents will be added to the sorting queue, * so the way we are sorted now is as good as it will ever get. * In this case, we instead switch to issuing extents in LBA order. */ if ((zfs_scan_issue_strategy < 1 && scn->scn_checkpointing) || zfs_scan_issue_strategy == 1) return (range_tree_first(rt)); /* * Try to continue previous extent if it is not completed yet. After * shrink in scan_io_queue_gather() it may no longer be the best, but * otherwise we leave shorter remnant every txg. */ uint64_t start; uint64_t size = 1 << rt->rt_shift; range_seg_t *addr_rs; if (queue->q_last_ext_addr != -1) { start = queue->q_last_ext_addr; addr_rs = range_tree_find(rt, start, size); if (addr_rs != NULL) return (addr_rs); } /* * Nothing to continue, so find new best extent. */ uint64_t *v = zfs_btree_first(&queue->q_exts_by_size, NULL); if (v == NULL) return (NULL); queue->q_last_ext_addr = start = *v << rt->rt_shift; /* * We need to get the original entry in the by_addr tree so we can * modify it. */ addr_rs = range_tree_find(rt, start, size); ASSERT3P(addr_rs, !=, NULL); ASSERT3U(rs_get_start(addr_rs, rt), ==, start); ASSERT3U(rs_get_end(addr_rs, rt), >, start); return (addr_rs); } static void scan_io_queues_run_one(void *arg) { dsl_scan_io_queue_t *queue = arg; kmutex_t *q_lock = &queue->q_vd->vdev_scan_io_queue_lock; boolean_t suspended = B_FALSE; range_seg_t *rs; scan_io_t *sio; zio_t *zio; list_t sio_list; ASSERT(queue->q_scn->scn_is_sorted); list_create(&sio_list, sizeof (scan_io_t), offsetof(scan_io_t, sio_nodes.sio_list_node)); zio = zio_null(queue->q_scn->scn_zio_root, queue->q_scn->scn_dp->dp_spa, NULL, NULL, NULL, ZIO_FLAG_CANFAIL); mutex_enter(q_lock); queue->q_zio = zio; /* Calculate maximum in-flight bytes for this vdev. */ queue->q_maxinflight_bytes = MAX(1, zfs_scan_vdev_limit * (vdev_get_ndisks(queue->q_vd) - vdev_get_nparity(queue->q_vd))); /* reset per-queue scan statistics for this txg */ queue->q_total_seg_size_this_txg = 0; queue->q_segs_this_txg = 0; queue->q_total_zio_size_this_txg = 0; queue->q_zios_this_txg = 0; /* loop until we run out of time or sios */ while ((rs = scan_io_queue_fetch_ext(queue)) != NULL) { uint64_t seg_start = 0, seg_end = 0; boolean_t more_left; ASSERT(list_is_empty(&sio_list)); /* loop while we still have sios left to process in this rs */ do { scan_io_t *first_sio, *last_sio; /* * We have selected which extent needs to be * processed next. Gather up the corresponding sios. */ more_left = scan_io_queue_gather(queue, rs, &sio_list); ASSERT(!list_is_empty(&sio_list)); first_sio = list_head(&sio_list); last_sio = list_tail(&sio_list); seg_end = SIO_GET_END_OFFSET(last_sio); if (seg_start == 0) seg_start = SIO_GET_OFFSET(first_sio); /* * Issuing sios can take a long time so drop the * queue lock. The sio queue won't be updated by * other threads since we're in syncing context so * we can be sure that our trees will remain exactly * as we left them. */ mutex_exit(q_lock); suspended = scan_io_queue_issue(queue, &sio_list); mutex_enter(q_lock); if (suspended) break; } while (more_left); /* update statistics for debugging purposes */ scan_io_queues_update_seg_stats(queue, seg_start, seg_end); if (suspended) break; } /* * If we were suspended in the middle of processing, * requeue any unfinished sios and exit. */ while ((sio = list_head(&sio_list)) != NULL) { list_remove(&sio_list, sio); scan_io_queue_insert_impl(queue, sio); } queue->q_zio = NULL; mutex_exit(q_lock); zio_nowait(zio); list_destroy(&sio_list); } /* * Performs an emptying run on all scan queues in the pool. This just * punches out one thread per top-level vdev, each of which processes * only that vdev's scan queue. We can parallelize the I/O here because * we know that each queue's I/Os only affect its own top-level vdev. * * This function waits for the queue runs to complete, and must be * called from dsl_scan_sync (or in general, syncing context). */ static void scan_io_queues_run(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; ASSERT(scn->scn_is_sorted); ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); if (scn->scn_queues_pending == 0) return; if (scn->scn_taskq == NULL) { int nthreads = spa->spa_root_vdev->vdev_children; /* * We need to make this taskq *always* execute as many * threads in parallel as we have top-level vdevs and no * less, otherwise strange serialization of the calls to * scan_io_queues_run_one can occur during spa_sync runs * and that significantly impacts performance. */ scn->scn_taskq = taskq_create("dsl_scan_iss", nthreads, minclsyspri, nthreads, nthreads, TASKQ_PREPOPULATE); } for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) { vdev_t *vd = spa->spa_root_vdev->vdev_child[i]; mutex_enter(&vd->vdev_scan_io_queue_lock); if (vd->vdev_scan_io_queue != NULL) { VERIFY(taskq_dispatch(scn->scn_taskq, scan_io_queues_run_one, vd->vdev_scan_io_queue, TQ_SLEEP) != TASKQID_INVALID); } mutex_exit(&vd->vdev_scan_io_queue_lock); } /* * Wait for the queues to finish issuing their IOs for this run * before we return. There may still be IOs in flight at this * point. */ taskq_wait(scn->scn_taskq); } static boolean_t dsl_scan_async_block_should_pause(dsl_scan_t *scn) { uint64_t elapsed_nanosecs; if (zfs_recover) return (B_FALSE); if (zfs_async_block_max_blocks != 0 && scn->scn_visited_this_txg >= zfs_async_block_max_blocks) { return (B_TRUE); } if (zfs_max_async_dedup_frees != 0 && scn->scn_dedup_frees_this_txg >= zfs_max_async_dedup_frees) { return (B_TRUE); } elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout || (NSEC2MSEC(elapsed_nanosecs) > scn->scn_async_block_min_time_ms && txg_sync_waiting(scn->scn_dp)) || spa_shutting_down(scn->scn_dp->dp_spa)); } static int dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { dsl_scan_t *scn = arg; if (!scn->scn_is_bptree || (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) { if (dsl_scan_async_block_should_pause(scn)) return (SET_ERROR(ERESTART)); } zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa, dmu_tx_get_txg(tx), bp, 0)); dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD, -bp_get_dsize_sync(scn->scn_dp->dp_spa, bp), -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx); scn->scn_visited_this_txg++; if (BP_GET_DEDUP(bp)) scn->scn_dedup_frees_this_txg++; return (0); } static void dsl_scan_update_stats(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; uint64_t i; uint64_t seg_size_total = 0, zio_size_total = 0; uint64_t seg_count_total = 0, zio_count_total = 0; for (i = 0; i < spa->spa_root_vdev->vdev_children; i++) { vdev_t *vd = spa->spa_root_vdev->vdev_child[i]; dsl_scan_io_queue_t *queue = vd->vdev_scan_io_queue; if (queue == NULL) continue; seg_size_total += queue->q_total_seg_size_this_txg; zio_size_total += queue->q_total_zio_size_this_txg; seg_count_total += queue->q_segs_this_txg; zio_count_total += queue->q_zios_this_txg; } if (seg_count_total == 0 || zio_count_total == 0) { scn->scn_avg_seg_size_this_txg = 0; scn->scn_avg_zio_size_this_txg = 0; scn->scn_segs_this_txg = 0; scn->scn_zios_this_txg = 0; return; } scn->scn_avg_seg_size_this_txg = seg_size_total / seg_count_total; scn->scn_avg_zio_size_this_txg = zio_size_total / zio_count_total; scn->scn_segs_this_txg = seg_count_total; scn->scn_zios_this_txg = zio_count_total; } static int bpobj_dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx) { ASSERT(!bp_freed); return (dsl_scan_free_block_cb(arg, bp, tx)); } static int dsl_scan_obsolete_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx) { ASSERT(!bp_freed); dsl_scan_t *scn = arg; const dva_t *dva = &bp->blk_dva[0]; if (dsl_scan_async_block_should_pause(scn)) return (SET_ERROR(ERESTART)); spa_vdev_indirect_mark_obsolete(scn->scn_dp->dp_spa, DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva), tx); scn->scn_visited_this_txg++; return (0); } boolean_t dsl_scan_active(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; uint64_t used = 0, comp, uncomp; boolean_t clones_left; if (spa->spa_load_state != SPA_LOAD_NONE) return (B_FALSE); if (spa_shutting_down(spa)) return (B_FALSE); if ((dsl_scan_is_running(scn) && !dsl_scan_is_paused_scrub(scn)) || (scn->scn_async_destroying && !scn->scn_async_stalled)) return (B_TRUE); if (spa_version(scn->scn_dp->dp_spa) >= SPA_VERSION_DEADLISTS) { (void) bpobj_space(&scn->scn_dp->dp_free_bpobj, &used, &comp, &uncomp); } clones_left = spa_livelist_delete_check(spa); return ((used != 0) || (clones_left)); } static boolean_t dsl_scan_check_deferred(vdev_t *vd) { boolean_t need_resilver = B_FALSE; for (int c = 0; c < vd->vdev_children; c++) { need_resilver |= dsl_scan_check_deferred(vd->vdev_child[c]); } if (!vdev_is_concrete(vd) || vd->vdev_aux || !vd->vdev_ops->vdev_op_leaf) return (need_resilver); if (!vd->vdev_resilver_deferred) need_resilver = B_TRUE; return (need_resilver); } static boolean_t dsl_scan_need_resilver(spa_t *spa, const dva_t *dva, size_t psize, uint64_t phys_birth) { vdev_t *vd; vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); if (vd->vdev_ops == &vdev_indirect_ops) { /* * The indirect vdev can point to multiple * vdevs. For simplicity, always create * the resilver zio_t. zio_vdev_io_start() * will bypass the child resilver i/o's if * they are on vdevs that don't have DTL's. */ return (B_TRUE); } if (DVA_GET_GANG(dva)) { /* * Gang members may be spread across multiple * vdevs, so the best estimate we have is the * scrub range, which has already been checked. * XXX -- it would be better to change our * allocation policy to ensure that all * gang members reside on the same vdev. */ return (B_TRUE); } /* * Check if the top-level vdev must resilver this offset. * When the offset does not intersect with a dirty leaf DTL * then it may be possible to skip the resilver IO. The psize * is provided instead of asize to simplify the check for RAIDZ. */ if (!vdev_dtl_need_resilver(vd, dva, psize, phys_birth)) return (B_FALSE); /* * Check that this top-level vdev has a device under it which * is resilvering and is not deferred. */ if (!dsl_scan_check_deferred(vd)) return (B_FALSE); return (B_TRUE); } static int dsl_process_async_destroys(dsl_pool_t *dp, dmu_tx_t *tx) { dsl_scan_t *scn = dp->dp_scan; spa_t *spa = dp->dp_spa; int err = 0; if (spa_suspend_async_destroy(spa)) return (0); if (zfs_free_bpobj_enabled && spa_version(spa) >= SPA_VERSION_DEADLISTS) { scn->scn_is_bptree = B_FALSE; scn->scn_async_block_min_time_ms = zfs_free_min_time_ms; scn->scn_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); err = bpobj_iterate(&dp->dp_free_bpobj, bpobj_dsl_scan_free_block_cb, scn, tx); VERIFY0(zio_wait(scn->scn_zio_root)); scn->scn_zio_root = NULL; if (err != 0 && err != ERESTART) zfs_panic_recover("error %u from bpobj_iterate()", err); } if (err == 0 && spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) { ASSERT(scn->scn_async_destroying); scn->scn_is_bptree = B_TRUE; scn->scn_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); err = bptree_iterate(dp->dp_meta_objset, dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb, scn, tx); VERIFY0(zio_wait(scn->scn_zio_root)); scn->scn_zio_root = NULL; if (err == EIO || err == ECKSUM) { err = 0; } else if (err != 0 && err != ERESTART) { zfs_panic_recover("error %u from " "traverse_dataset_destroyed()", err); } if (bptree_is_empty(dp->dp_meta_objset, dp->dp_bptree_obj)) { /* finished; deactivate async destroy feature */ spa_feature_decr(spa, SPA_FEATURE_ASYNC_DESTROY, tx); ASSERT(!spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)); VERIFY0(zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_BPTREE_OBJ, tx)); VERIFY0(bptree_free(dp->dp_meta_objset, dp->dp_bptree_obj, tx)); dp->dp_bptree_obj = 0; scn->scn_async_destroying = B_FALSE; scn->scn_async_stalled = B_FALSE; } else { /* * If we didn't make progress, mark the async * destroy as stalled, so that we will not initiate * a spa_sync() on its behalf. Note that we only * check this if we are not finished, because if the * bptree had no blocks for us to visit, we can * finish without "making progress". */ scn->scn_async_stalled = (scn->scn_visited_this_txg == 0); } } if (scn->scn_visited_this_txg) { zfs_dbgmsg("freed %llu blocks in %llums from " "free_bpobj/bptree txg %llu; err=%u", (longlong_t)scn->scn_visited_this_txg, (longlong_t) NSEC2MSEC(gethrtime() - scn->scn_sync_start_time), (longlong_t)tx->tx_txg, err); scn->scn_visited_this_txg = 0; scn->scn_dedup_frees_this_txg = 0; /* * Write out changes to the DDT that may be required as a * result of the blocks freed. This ensures that the DDT * is clean when a scrub/resilver runs. */ ddt_sync(spa, tx->tx_txg); } if (err != 0) return (err); if (dp->dp_free_dir != NULL && !scn->scn_async_destroying && zfs_free_leak_on_eio && (dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes != 0 || dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes != 0 || dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes != 0)) { /* * We have finished background destroying, but there is still * some space left in the dp_free_dir. Transfer this leaked * space to the dp_leak_dir. */ if (dp->dp_leak_dir == NULL) { rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); (void) dsl_dir_create_sync(dp, dp->dp_root_dir, LEAK_DIR_NAME, tx); VERIFY0(dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, &dp->dp_leak_dir)); rrw_exit(&dp->dp_config_rwlock, FTAG); } dsl_dir_diduse_space(dp->dp_leak_dir, DD_USED_HEAD, dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes, dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes, dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx); dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD, -dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes, -dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes, -dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx); } if (dp->dp_free_dir != NULL && !scn->scn_async_destroying && !spa_livelist_delete_check(spa)) { /* finished; verify that space accounting went to zero */ ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes); ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes); ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes); } spa_notify_waiters(spa); EQUIV(bpobj_is_open(&dp->dp_obsolete_bpobj), 0 == zap_contains(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_OBSOLETE_BPOBJ)); if (err == 0 && bpobj_is_open(&dp->dp_obsolete_bpobj)) { ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_OBSOLETE_COUNTS)); scn->scn_is_bptree = B_FALSE; scn->scn_async_block_min_time_ms = zfs_obsolete_min_time_ms; err = bpobj_iterate(&dp->dp_obsolete_bpobj, dsl_scan_obsolete_block_cb, scn, tx); if (err != 0 && err != ERESTART) zfs_panic_recover("error %u from bpobj_iterate()", err); if (bpobj_is_empty(&dp->dp_obsolete_bpobj)) dsl_pool_destroy_obsolete_bpobj(dp, tx); } return (0); } /* * This is the primary entry point for scans that is called from syncing * context. Scans must happen entirely during syncing context so that we * can guarantee that blocks we are currently scanning will not change out * from under us. While a scan is active, this function controls how quickly * transaction groups proceed, instead of the normal handling provided by * txg_sync_thread(). */ void dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) { int err = 0; dsl_scan_t *scn = dp->dp_scan; spa_t *spa = dp->dp_spa; state_sync_type_t sync_type = SYNC_OPTIONAL; if (spa->spa_resilver_deferred && !spa_feature_is_active(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER)) spa_feature_incr(spa, SPA_FEATURE_RESILVER_DEFER, tx); /* * Check for scn_restart_txg before checking spa_load_state, so * that we can restart an old-style scan while the pool is being * imported (see dsl_scan_init). We also restart scans if there * is a deferred resilver and the user has manually disabled * deferred resilvers via the tunable. */ if (dsl_scan_restarting(scn, tx) || (spa->spa_resilver_deferred && zfs_resilver_disable_defer)) { pool_scan_func_t func = POOL_SCAN_SCRUB; dsl_scan_done(scn, B_FALSE, tx); if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) func = POOL_SCAN_RESILVER; zfs_dbgmsg("restarting scan func=%u txg=%llu", func, (longlong_t)tx->tx_txg); dsl_scan_setup_sync(&func, tx); } /* * Only process scans in sync pass 1. */ if (spa_sync_pass(spa) > 1) return; /* * If the spa is shutting down, then stop scanning. This will * ensure that the scan does not dirty any new data during the * shutdown phase. */ if (spa_shutting_down(spa)) return; /* * If the scan is inactive due to a stalled async destroy, try again. */ if (!scn->scn_async_stalled && !dsl_scan_active(scn)) return; /* reset scan statistics */ scn->scn_visited_this_txg = 0; scn->scn_dedup_frees_this_txg = 0; scn->scn_holes_this_txg = 0; scn->scn_lt_min_this_txg = 0; scn->scn_gt_max_this_txg = 0; scn->scn_ddt_contained_this_txg = 0; scn->scn_objsets_visited_this_txg = 0; scn->scn_avg_seg_size_this_txg = 0; scn->scn_segs_this_txg = 0; scn->scn_avg_zio_size_this_txg = 0; scn->scn_zios_this_txg = 0; scn->scn_suspending = B_FALSE; scn->scn_sync_start_time = gethrtime(); spa->spa_scrub_active = B_TRUE; /* * First process the async destroys. If we suspend, don't do * any scrubbing or resilvering. This ensures that there are no * async destroys while we are scanning, so the scan code doesn't * have to worry about traversing it. It is also faster to free the * blocks than to scrub them. */ err = dsl_process_async_destroys(dp, tx); if (err != 0) return; if (!dsl_scan_is_running(scn) || dsl_scan_is_paused_scrub(scn)) return; /* * Wait a few txgs after importing to begin scanning so that * we can get the pool imported quickly. */ if (spa->spa_syncing_txg < spa->spa_first_txg + SCAN_IMPORT_WAIT_TXGS) return; /* * zfs_scan_suspend_progress can be set to disable scan progress. * We don't want to spin the txg_sync thread, so we add a delay * here to simulate the time spent doing a scan. This is mostly * useful for testing and debugging. */ if (zfs_scan_suspend_progress) { uint64_t scan_time_ns = gethrtime() - scn->scn_sync_start_time; int mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ? zfs_resilver_min_time_ms : zfs_scrub_min_time_ms; while (zfs_scan_suspend_progress && !txg_sync_waiting(scn->scn_dp) && !spa_shutting_down(scn->scn_dp->dp_spa) && NSEC2MSEC(scan_time_ns) < mintime) { delay(hz); scan_time_ns = gethrtime() - scn->scn_sync_start_time; } return; } /* * It is possible to switch from unsorted to sorted at any time, * but afterwards the scan will remain sorted unless reloaded from * a checkpoint after a reboot. */ if (!zfs_scan_legacy) { scn->scn_is_sorted = B_TRUE; if (scn->scn_last_checkpoint == 0) scn->scn_last_checkpoint = ddi_get_lbolt(); } /* * For sorted scans, determine what kind of work we will be doing * this txg based on our memory limitations and whether or not we * need to perform a checkpoint. */ if (scn->scn_is_sorted) { /* * If we are over our checkpoint interval, set scn_clearing * so that we can begin checkpointing immediately. The * checkpoint allows us to save a consistent bookmark * representing how much data we have scrubbed so far. * Otherwise, use the memory limit to determine if we should * scan for metadata or start issue scrub IOs. We accumulate * metadata until we hit our hard memory limit at which point * we issue scrub IOs until we are at our soft memory limit. */ if (scn->scn_checkpointing || ddi_get_lbolt() - scn->scn_last_checkpoint > SEC_TO_TICK(zfs_scan_checkpoint_intval)) { if (!scn->scn_checkpointing) zfs_dbgmsg("begin scan checkpoint"); scn->scn_checkpointing = B_TRUE; scn->scn_clearing = B_TRUE; } else { boolean_t should_clear = dsl_scan_should_clear(scn); if (should_clear && !scn->scn_clearing) { zfs_dbgmsg("begin scan clearing"); scn->scn_clearing = B_TRUE; } else if (!should_clear && scn->scn_clearing) { zfs_dbgmsg("finish scan clearing"); scn->scn_clearing = B_FALSE; } } } else { ASSERT0(scn->scn_checkpointing); ASSERT0(scn->scn_clearing); } if (!scn->scn_clearing && scn->scn_done_txg == 0) { /* Need to scan metadata for more blocks to scrub */ dsl_scan_phys_t *scnp = &scn->scn_phys; taskqid_t prefetch_tqid; /* * Recalculate the max number of in-flight bytes for pool-wide * scanning operations (minimum 1MB). Limits for the issuing * phase are done per top-level vdev and are handled separately. */ scn->scn_maxinflight_bytes = MAX(zfs_scan_vdev_limit * dsl_scan_count_data_disks(spa->spa_root_vdev), 1ULL << 20); if (scnp->scn_ddt_bookmark.ddb_class <= scnp->scn_ddt_class_max) { ASSERT(ZB_IS_ZERO(&scnp->scn_bookmark)); zfs_dbgmsg("doing scan sync txg %llu; " "ddt bm=%llu/%llu/%llu/%llx", (longlong_t)tx->tx_txg, (longlong_t)scnp->scn_ddt_bookmark.ddb_class, (longlong_t)scnp->scn_ddt_bookmark.ddb_type, (longlong_t)scnp->scn_ddt_bookmark.ddb_checksum, (longlong_t)scnp->scn_ddt_bookmark.ddb_cursor); } else { zfs_dbgmsg("doing scan sync txg %llu; " "bm=%llu/%llu/%llu/%llu", (longlong_t)tx->tx_txg, (longlong_t)scnp->scn_bookmark.zb_objset, (longlong_t)scnp->scn_bookmark.zb_object, (longlong_t)scnp->scn_bookmark.zb_level, (longlong_t)scnp->scn_bookmark.zb_blkid); } scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_CANFAIL); scn->scn_prefetch_stop = B_FALSE; prefetch_tqid = taskq_dispatch(dp->dp_sync_taskq, dsl_scan_prefetch_thread, scn, TQ_SLEEP); ASSERT(prefetch_tqid != TASKQID_INVALID); dsl_pool_config_enter(dp, FTAG); dsl_scan_visit(scn, tx); dsl_pool_config_exit(dp, FTAG); mutex_enter(&dp->dp_spa->spa_scrub_lock); scn->scn_prefetch_stop = B_TRUE; cv_broadcast(&spa->spa_scrub_io_cv); mutex_exit(&dp->dp_spa->spa_scrub_lock); taskq_wait_id(dp->dp_sync_taskq, prefetch_tqid); (void) zio_wait(scn->scn_zio_root); scn->scn_zio_root = NULL; zfs_dbgmsg("scan visited %llu blocks in %llums " "(%llu os's, %llu holes, %llu < mintxg, " "%llu in ddt, %llu > maxtxg)", (longlong_t)scn->scn_visited_this_txg, (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time), (longlong_t)scn->scn_objsets_visited_this_txg, (longlong_t)scn->scn_holes_this_txg, (longlong_t)scn->scn_lt_min_this_txg, (longlong_t)scn->scn_ddt_contained_this_txg, (longlong_t)scn->scn_gt_max_this_txg); if (!scn->scn_suspending) { ASSERT0(avl_numnodes(&scn->scn_queue)); scn->scn_done_txg = tx->tx_txg + 1; if (scn->scn_is_sorted) { scn->scn_checkpointing = B_TRUE; scn->scn_clearing = B_TRUE; } zfs_dbgmsg("scan complete txg %llu", (longlong_t)tx->tx_txg); } } else if (scn->scn_is_sorted && scn->scn_queues_pending != 0) { ASSERT(scn->scn_clearing); /* need to issue scrubbing IOs from per-vdev queues */ scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_CANFAIL); scan_io_queues_run(scn); (void) zio_wait(scn->scn_zio_root); scn->scn_zio_root = NULL; /* calculate and dprintf the current memory usage */ (void) dsl_scan_should_clear(scn); dsl_scan_update_stats(scn); zfs_dbgmsg("scan issued %llu blocks (%llu segs) in %llums " "(avg_block_size = %llu, avg_seg_size = %llu)", (longlong_t)scn->scn_zios_this_txg, (longlong_t)scn->scn_segs_this_txg, (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time), (longlong_t)scn->scn_avg_zio_size_this_txg, (longlong_t)scn->scn_avg_seg_size_this_txg); } else if (scn->scn_done_txg != 0 && scn->scn_done_txg <= tx->tx_txg) { /* Finished with everything. Mark the scrub as complete */ zfs_dbgmsg("scan issuing complete txg %llu", (longlong_t)tx->tx_txg); ASSERT3U(scn->scn_done_txg, !=, 0); ASSERT0(spa->spa_scrub_inflight); ASSERT0(scn->scn_queues_pending); dsl_scan_done(scn, B_TRUE, tx); sync_type = SYNC_MANDATORY; } dsl_scan_sync_state(scn, tx, sync_type); } static void count_block_issued(spa_t *spa, const blkptr_t *bp, boolean_t all) { /* * Don't count embedded bp's, since we already did the work of * scanning these when we scanned the containing block. */ if (BP_IS_EMBEDDED(bp)) return; /* * Update the spa's stats on how many bytes we have issued. * Sequential scrubs create a zio for each DVA of the bp. Each * of these will include all DVAs for repair purposes, but the * zio code will only try the first one unless there is an issue. * Therefore, we should only count the first DVA for these IOs. */ atomic_add_64(&spa->spa_scan_pass_issued, all ? BP_GET_ASIZE(bp) : DVA_GET_ASIZE(&bp->blk_dva[0])); } static void count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp) { /* * If we resume after a reboot, zab will be NULL; don't record * incomplete stats in that case. */ if (zab == NULL) return; for (int i = 0; i < 4; i++) { int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS; int t = (i & 1) ? BP_GET_TYPE(bp) : DMU_OT_TOTAL; if (t & DMU_OT_NEWTYPE) t = DMU_OT_OTHER; zfs_blkstat_t *zb = &zab->zab_type[l][t]; int equal; zb->zb_count++; zb->zb_asize += BP_GET_ASIZE(bp); zb->zb_lsize += BP_GET_LSIZE(bp); zb->zb_psize += BP_GET_PSIZE(bp); zb->zb_gangs += BP_COUNT_GANG(bp); switch (BP_GET_NDVAS(bp)) { case 2: if (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[1])) zb->zb_ditto_2_of_2_samevdev++; break; case 3: equal = (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[1])) + (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[2])) + (DVA_GET_VDEV(&bp->blk_dva[1]) == DVA_GET_VDEV(&bp->blk_dva[2])); if (equal == 1) zb->zb_ditto_2_of_3_samevdev++; else if (equal == 3) zb->zb_ditto_3_of_3_samevdev++; break; } } } static void scan_io_queue_insert_impl(dsl_scan_io_queue_t *queue, scan_io_t *sio) { avl_index_t idx; dsl_scan_t *scn = queue->q_scn; ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock)); if (unlikely(avl_is_empty(&queue->q_sios_by_addr))) atomic_add_64(&scn->scn_queues_pending, 1); if (avl_find(&queue->q_sios_by_addr, sio, &idx) != NULL) { /* block is already scheduled for reading */ sio_free(sio); return; } avl_insert(&queue->q_sios_by_addr, sio, idx); queue->q_sio_memused += SIO_GET_MUSED(sio); range_tree_add(queue->q_exts_by_addr, SIO_GET_OFFSET(sio), SIO_GET_ASIZE(sio)); } /* * Given all the info we got from our metadata scanning process, we * construct a scan_io_t and insert it into the scan sorting queue. The * I/O must already be suitable for us to process. This is controlled * by dsl_scan_enqueue(). */ static void scan_io_queue_insert(dsl_scan_io_queue_t *queue, const blkptr_t *bp, int dva_i, int zio_flags, const zbookmark_phys_t *zb) { scan_io_t *sio = sio_alloc(BP_GET_NDVAS(bp)); ASSERT0(BP_IS_GANG(bp)); ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock)); bp2sio(bp, sio, dva_i); sio->sio_flags = zio_flags; sio->sio_zb = *zb; queue->q_last_ext_addr = -1; scan_io_queue_insert_impl(queue, sio); } /* * Given a set of I/O parameters as discovered by the metadata traversal * process, attempts to place the I/O into the sorted queues (if allowed), * or immediately executes the I/O. */ static void dsl_scan_enqueue(dsl_pool_t *dp, const blkptr_t *bp, int zio_flags, const zbookmark_phys_t *zb) { spa_t *spa = dp->dp_spa; ASSERT(!BP_IS_EMBEDDED(bp)); /* * Gang blocks are hard to issue sequentially, so we just issue them * here immediately instead of queuing them. */ if (!dp->dp_scan->scn_is_sorted || BP_IS_GANG(bp)) { scan_exec_io(dp, bp, zio_flags, zb, NULL); return; } for (int i = 0; i < BP_GET_NDVAS(bp); i++) { dva_t dva; vdev_t *vdev; dva = bp->blk_dva[i]; vdev = vdev_lookup_top(spa, DVA_GET_VDEV(&dva)); ASSERT(vdev != NULL); mutex_enter(&vdev->vdev_scan_io_queue_lock); if (vdev->vdev_scan_io_queue == NULL) vdev->vdev_scan_io_queue = scan_io_queue_create(vdev); ASSERT(dp->dp_scan != NULL); scan_io_queue_insert(vdev->vdev_scan_io_queue, bp, i, zio_flags, zb); mutex_exit(&vdev->vdev_scan_io_queue_lock); } } static int dsl_scan_scrub_cb(dsl_pool_t *dp, const blkptr_t *bp, const zbookmark_phys_t *zb) { dsl_scan_t *scn = dp->dp_scan; spa_t *spa = dp->dp_spa; uint64_t phys_birth = BP_PHYSICAL_BIRTH(bp); size_t psize = BP_GET_PSIZE(bp); boolean_t needs_io = B_FALSE; int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL; count_block(dp->dp_blkstats, bp); if (phys_birth <= scn->scn_phys.scn_min_txg || phys_birth >= scn->scn_phys.scn_max_txg) { count_block_issued(spa, bp, B_TRUE); return (0); } /* Embedded BP's have phys_birth==0, so we reject them above. */ ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn)); if (scn->scn_phys.scn_func == POOL_SCAN_SCRUB) { zio_flags |= ZIO_FLAG_SCRUB; needs_io = B_TRUE; } else { ASSERT3U(scn->scn_phys.scn_func, ==, POOL_SCAN_RESILVER); zio_flags |= ZIO_FLAG_RESILVER; needs_io = B_FALSE; } /* If it's an intent log block, failure is expected. */ if (zb->zb_level == ZB_ZIL_LEVEL) zio_flags |= ZIO_FLAG_SPECULATIVE; for (int d = 0; d < BP_GET_NDVAS(bp); d++) { const dva_t *dva = &bp->blk_dva[d]; /* * Keep track of how much data we've examined so that * zpool(8) status can make useful progress reports. */ uint64_t asize = DVA_GET_ASIZE(dva); scn->scn_phys.scn_examined += asize; spa->spa_scan_pass_exam += asize; /* if it's a resilver, this may not be in the target range */ if (!needs_io) needs_io = dsl_scan_need_resilver(spa, dva, psize, phys_birth); } if (needs_io && !zfs_no_scrub_io) { dsl_scan_enqueue(dp, bp, zio_flags, zb); } else { count_block_issued(spa, bp, B_TRUE); } /* do not relocate this block */ return (0); } static void dsl_scan_scrub_done(zio_t *zio) { spa_t *spa = zio->io_spa; blkptr_t *bp = zio->io_bp; dsl_scan_io_queue_t *queue = zio->io_private; abd_free(zio->io_abd); if (queue == NULL) { mutex_enter(&spa->spa_scrub_lock); ASSERT3U(spa->spa_scrub_inflight, >=, BP_GET_PSIZE(bp)); spa->spa_scrub_inflight -= BP_GET_PSIZE(bp); cv_broadcast(&spa->spa_scrub_io_cv); mutex_exit(&spa->spa_scrub_lock); } else { mutex_enter(&queue->q_vd->vdev_scan_io_queue_lock); ASSERT3U(queue->q_inflight_bytes, >=, BP_GET_PSIZE(bp)); queue->q_inflight_bytes -= BP_GET_PSIZE(bp); cv_broadcast(&queue->q_zio_cv); mutex_exit(&queue->q_vd->vdev_scan_io_queue_lock); } if (zio->io_error && (zio->io_error != ECKSUM || !(zio->io_flags & ZIO_FLAG_SPECULATIVE))) { atomic_inc_64(&spa->spa_dsl_pool->dp_scan->scn_phys.scn_errors); } } /* * Given a scanning zio's information, executes the zio. The zio need * not necessarily be only sortable, this function simply executes the * zio, no matter what it is. The optional queue argument allows the * caller to specify that they want per top level vdev IO rate limiting * instead of the legacy global limiting. */ static void scan_exec_io(dsl_pool_t *dp, const blkptr_t *bp, int zio_flags, const zbookmark_phys_t *zb, dsl_scan_io_queue_t *queue) { spa_t *spa = dp->dp_spa; dsl_scan_t *scn = dp->dp_scan; size_t size = BP_GET_PSIZE(bp); abd_t *data = abd_alloc_for_io(size, B_FALSE); zio_t *pio; if (queue == NULL) { ASSERT3U(scn->scn_maxinflight_bytes, >, 0); mutex_enter(&spa->spa_scrub_lock); while (spa->spa_scrub_inflight >= scn->scn_maxinflight_bytes) cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); spa->spa_scrub_inflight += BP_GET_PSIZE(bp); mutex_exit(&spa->spa_scrub_lock); pio = scn->scn_zio_root; } else { kmutex_t *q_lock = &queue->q_vd->vdev_scan_io_queue_lock; ASSERT3U(queue->q_maxinflight_bytes, >, 0); mutex_enter(q_lock); while (queue->q_inflight_bytes >= queue->q_maxinflight_bytes) cv_wait(&queue->q_zio_cv, q_lock); queue->q_inflight_bytes += BP_GET_PSIZE(bp); pio = queue->q_zio; mutex_exit(q_lock); } ASSERT(pio != NULL); count_block_issued(spa, bp, queue == NULL); zio_nowait(zio_read(pio, spa, bp, data, size, dsl_scan_scrub_done, queue, ZIO_PRIORITY_SCRUB, zio_flags, zb)); } /* * This is the primary extent sorting algorithm. We balance two parameters: * 1) how many bytes of I/O are in an extent * 2) how well the extent is filled with I/O (as a fraction of its total size) * Since we allow extents to have gaps between their constituent I/Os, it's * possible to have a fairly large extent that contains the same amount of * I/O bytes than a much smaller extent, which just packs the I/O more tightly. * The algorithm sorts based on a score calculated from the extent's size, * the relative fill volume (in %) and a "fill weight" parameter that controls * the split between whether we prefer larger extents or more well populated * extents: * * SCORE = FILL_IN_BYTES + (FILL_IN_PERCENT * FILL_IN_BYTES * FILL_WEIGHT) * * Example: * 1) assume extsz = 64 MiB * 2) assume fill = 32 MiB (extent is half full) * 3) assume fill_weight = 3 * 4) SCORE = 32M + (((32M * 100) / 64M) * 3 * 32M) / 100 * SCORE = 32M + (50 * 3 * 32M) / 100 * SCORE = 32M + (4800M / 100) * SCORE = 32M + 48M * ^ ^ * | +--- final total relative fill-based score * +--------- final total fill-based score * SCORE = 80M * * As can be seen, at fill_ratio=3, the algorithm is slightly biased towards * extents that are more completely filled (in a 3:2 ratio) vs just larger. * Note that as an optimization, we replace multiplication and division by * 100 with bitshifting by 7 (which effectively multiplies and divides by 128). * * Since we do not care if one extent is only few percent better than another, * compress the score into 6 bits via binary logarithm AKA highbit64() and * put into otherwise unused due to ashift high bits of offset. This allows * to reduce q_exts_by_size B-tree elements to only 64 bits and compare them * with single operation. Plus it makes scrubs more sequential and reduces * chances that minor extent change move it within the B-tree. */ static int ext_size_compare(const void *x, const void *y) { const uint64_t *a = x, *b = y; return (TREE_CMP(*a, *b)); } static void ext_size_create(range_tree_t *rt, void *arg) { (void) rt; zfs_btree_t *size_tree = arg; zfs_btree_create(size_tree, ext_size_compare, sizeof (uint64_t)); } static void ext_size_destroy(range_tree_t *rt, void *arg) { (void) rt; zfs_btree_t *size_tree = arg; ASSERT0(zfs_btree_numnodes(size_tree)); zfs_btree_destroy(size_tree); } static uint64_t ext_size_value(range_tree_t *rt, range_seg_gap_t *rsg) { (void) rt; uint64_t size = rsg->rs_end - rsg->rs_start; uint64_t score = rsg->rs_fill + ((((rsg->rs_fill << 7) / size) * fill_weight * rsg->rs_fill) >> 7); ASSERT3U(rt->rt_shift, >=, 8); return (((uint64_t)(64 - highbit64(score)) << 56) | rsg->rs_start); } static void ext_size_add(range_tree_t *rt, range_seg_t *rs, void *arg) { zfs_btree_t *size_tree = arg; ASSERT3U(rt->rt_type, ==, RANGE_SEG_GAP); uint64_t v = ext_size_value(rt, (range_seg_gap_t *)rs); zfs_btree_add(size_tree, &v); } static void ext_size_remove(range_tree_t *rt, range_seg_t *rs, void *arg) { zfs_btree_t *size_tree = arg; ASSERT3U(rt->rt_type, ==, RANGE_SEG_GAP); uint64_t v = ext_size_value(rt, (range_seg_gap_t *)rs); zfs_btree_remove(size_tree, &v); } static void ext_size_vacate(range_tree_t *rt, void *arg) { zfs_btree_t *size_tree = arg; zfs_btree_clear(size_tree); zfs_btree_destroy(size_tree); ext_size_create(rt, arg); } static const range_tree_ops_t ext_size_ops = { .rtop_create = ext_size_create, .rtop_destroy = ext_size_destroy, .rtop_add = ext_size_add, .rtop_remove = ext_size_remove, .rtop_vacate = ext_size_vacate }; /* * Comparator for the q_sios_by_addr tree. Sorting is simply performed * based on LBA-order (from lowest to highest). */ static int sio_addr_compare(const void *x, const void *y) { const scan_io_t *a = x, *b = y; return (TREE_CMP(SIO_GET_OFFSET(a), SIO_GET_OFFSET(b))); } /* IO queues are created on demand when they are needed. */ static dsl_scan_io_queue_t * scan_io_queue_create(vdev_t *vd) { dsl_scan_t *scn = vd->vdev_spa->spa_dsl_pool->dp_scan; dsl_scan_io_queue_t *q = kmem_zalloc(sizeof (*q), KM_SLEEP); q->q_scn = scn; q->q_vd = vd; q->q_sio_memused = 0; q->q_last_ext_addr = -1; cv_init(&q->q_zio_cv, NULL, CV_DEFAULT, NULL); q->q_exts_by_addr = range_tree_create_gap(&ext_size_ops, RANGE_SEG_GAP, &q->q_exts_by_size, 0, vd->vdev_ashift, zfs_scan_max_ext_gap); avl_create(&q->q_sios_by_addr, sio_addr_compare, sizeof (scan_io_t), offsetof(scan_io_t, sio_nodes.sio_addr_node)); return (q); } /* * Destroys a scan queue and all segments and scan_io_t's contained in it. * No further execution of I/O occurs, anything pending in the queue is * simply freed without being executed. */ void dsl_scan_io_queue_destroy(dsl_scan_io_queue_t *queue) { dsl_scan_t *scn = queue->q_scn; scan_io_t *sio; void *cookie = NULL; ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock)); if (!avl_is_empty(&queue->q_sios_by_addr)) atomic_add_64(&scn->scn_queues_pending, -1); while ((sio = avl_destroy_nodes(&queue->q_sios_by_addr, &cookie)) != NULL) { ASSERT(range_tree_contains(queue->q_exts_by_addr, SIO_GET_OFFSET(sio), SIO_GET_ASIZE(sio))); queue->q_sio_memused -= SIO_GET_MUSED(sio); sio_free(sio); } ASSERT0(queue->q_sio_memused); range_tree_vacate(queue->q_exts_by_addr, NULL, queue); range_tree_destroy(queue->q_exts_by_addr); avl_destroy(&queue->q_sios_by_addr); cv_destroy(&queue->q_zio_cv); kmem_free(queue, sizeof (*queue)); } /* * Properly transfers a dsl_scan_queue_t from `svd' to `tvd'. This is * called on behalf of vdev_top_transfer when creating or destroying * a mirror vdev due to zpool attach/detach. */ void dsl_scan_io_queue_vdev_xfer(vdev_t *svd, vdev_t *tvd) { mutex_enter(&svd->vdev_scan_io_queue_lock); mutex_enter(&tvd->vdev_scan_io_queue_lock); VERIFY3P(tvd->vdev_scan_io_queue, ==, NULL); tvd->vdev_scan_io_queue = svd->vdev_scan_io_queue; svd->vdev_scan_io_queue = NULL; if (tvd->vdev_scan_io_queue != NULL) tvd->vdev_scan_io_queue->q_vd = tvd; mutex_exit(&tvd->vdev_scan_io_queue_lock); mutex_exit(&svd->vdev_scan_io_queue_lock); } static void scan_io_queues_destroy(dsl_scan_t *scn) { vdev_t *rvd = scn->scn_dp->dp_spa->spa_root_vdev; for (uint64_t i = 0; i < rvd->vdev_children; i++) { vdev_t *tvd = rvd->vdev_child[i]; mutex_enter(&tvd->vdev_scan_io_queue_lock); if (tvd->vdev_scan_io_queue != NULL) dsl_scan_io_queue_destroy(tvd->vdev_scan_io_queue); tvd->vdev_scan_io_queue = NULL; mutex_exit(&tvd->vdev_scan_io_queue_lock); } } static void dsl_scan_freed_dva(spa_t *spa, const blkptr_t *bp, int dva_i) { dsl_pool_t *dp = spa->spa_dsl_pool; dsl_scan_t *scn = dp->dp_scan; vdev_t *vdev; kmutex_t *q_lock; dsl_scan_io_queue_t *queue; scan_io_t *srch_sio, *sio; avl_index_t idx; uint64_t start, size; vdev = vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[dva_i])); ASSERT(vdev != NULL); q_lock = &vdev->vdev_scan_io_queue_lock; queue = vdev->vdev_scan_io_queue; mutex_enter(q_lock); if (queue == NULL) { mutex_exit(q_lock); return; } srch_sio = sio_alloc(BP_GET_NDVAS(bp)); bp2sio(bp, srch_sio, dva_i); start = SIO_GET_OFFSET(srch_sio); size = SIO_GET_ASIZE(srch_sio); /* * We can find the zio in two states: * 1) Cold, just sitting in the queue of zio's to be issued at * some point in the future. In this case, all we do is * remove the zio from the q_sios_by_addr tree, decrement * its data volume from the containing range_seg_t and * resort the q_exts_by_size tree to reflect that the * range_seg_t has lost some of its 'fill'. We don't shorten * the range_seg_t - this is usually rare enough not to be * worth the extra hassle of trying keep track of precise * extent boundaries. * 2) Hot, where the zio is currently in-flight in * dsl_scan_issue_ios. In this case, we can't simply * reach in and stop the in-flight zio's, so we instead * block the caller. Eventually, dsl_scan_issue_ios will * be done with issuing the zio's it gathered and will * signal us. */ sio = avl_find(&queue->q_sios_by_addr, srch_sio, &idx); sio_free(srch_sio); if (sio != NULL) { blkptr_t tmpbp; /* Got it while it was cold in the queue */ ASSERT3U(start, ==, SIO_GET_OFFSET(sio)); ASSERT3U(size, ==, SIO_GET_ASIZE(sio)); avl_remove(&queue->q_sios_by_addr, sio); if (avl_is_empty(&queue->q_sios_by_addr)) atomic_add_64(&scn->scn_queues_pending, -1); queue->q_sio_memused -= SIO_GET_MUSED(sio); ASSERT(range_tree_contains(queue->q_exts_by_addr, start, size)); range_tree_remove_fill(queue->q_exts_by_addr, start, size); /* count the block as though we issued it */ sio2bp(sio, &tmpbp); count_block_issued(spa, &tmpbp, B_FALSE); sio_free(sio); } mutex_exit(q_lock); } /* * Callback invoked when a zio_free() zio is executing. This needs to be * intercepted to prevent the zio from deallocating a particular portion * of disk space and it then getting reallocated and written to, while we * still have it queued up for processing. */ void dsl_scan_freed(spa_t *spa, const blkptr_t *bp) { dsl_pool_t *dp = spa->spa_dsl_pool; dsl_scan_t *scn = dp->dp_scan; ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(scn != NULL); if (!dsl_scan_is_running(scn)) return; for (int i = 0; i < BP_GET_NDVAS(bp); i++) dsl_scan_freed_dva(spa, bp, i); } /* * Check if a vdev needs resilvering (non-empty DTL), if so, and resilver has * not started, start it. Otherwise, only restart if max txg in DTL range is * greater than the max txg in the current scan. If the DTL max is less than * the scan max, then the vdev has not missed any new data since the resilver * started, so a restart is not needed. */ void dsl_scan_assess_vdev(dsl_pool_t *dp, vdev_t *vd) { uint64_t min, max; if (!vdev_resilver_needed(vd, &min, &max)) return; if (!dsl_scan_resilvering(dp)) { spa_async_request(dp->dp_spa, SPA_ASYNC_RESILVER); return; } if (max <= dp->dp_scan->scn_phys.scn_max_txg) return; /* restart is needed, check if it can be deferred */ if (spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER)) vdev_defer_resilver(vd); else spa_async_request(dp->dp_spa, SPA_ASYNC_RESILVER); } /* BEGIN CSTYLED */ ZFS_MODULE_PARAM(zfs, zfs_, scan_vdev_limit, ULONG, ZMOD_RW, "Max bytes in flight per leaf vdev for scrubs and resilvers"); ZFS_MODULE_PARAM(zfs, zfs_, scrub_min_time_ms, INT, ZMOD_RW, "Min millisecs to scrub per txg"); ZFS_MODULE_PARAM(zfs, zfs_, obsolete_min_time_ms, INT, ZMOD_RW, "Min millisecs to obsolete per txg"); ZFS_MODULE_PARAM(zfs, zfs_, free_min_time_ms, INT, ZMOD_RW, "Min millisecs to free per txg"); ZFS_MODULE_PARAM(zfs, zfs_, resilver_min_time_ms, INT, ZMOD_RW, "Min millisecs to resilver per txg"); ZFS_MODULE_PARAM(zfs, zfs_, scan_suspend_progress, INT, ZMOD_RW, "Set to prevent scans from progressing"); ZFS_MODULE_PARAM(zfs, zfs_, no_scrub_io, INT, ZMOD_RW, "Set to disable scrub I/O"); ZFS_MODULE_PARAM(zfs, zfs_, no_scrub_prefetch, INT, ZMOD_RW, "Set to disable scrub prefetching"); ZFS_MODULE_PARAM(zfs, zfs_, async_block_max_blocks, ULONG, ZMOD_RW, "Max number of blocks freed in one txg"); ZFS_MODULE_PARAM(zfs, zfs_, max_async_dedup_frees, ULONG, ZMOD_RW, "Max number of dedup blocks freed in one txg"); ZFS_MODULE_PARAM(zfs, zfs_, free_bpobj_enabled, INT, ZMOD_RW, "Enable processing of the free_bpobj"); ZFS_MODULE_PARAM(zfs, zfs_, scan_blkstats, INT, ZMOD_RW, "Enable block statistics calculation during scrub"); ZFS_MODULE_PARAM(zfs, zfs_, scan_mem_lim_fact, INT, ZMOD_RW, "Fraction of RAM for scan hard limit"); ZFS_MODULE_PARAM(zfs, zfs_, scan_issue_strategy, INT, ZMOD_RW, "IO issuing strategy during scrubbing. " "0 = default, 1 = LBA, 2 = size"); ZFS_MODULE_PARAM(zfs, zfs_, scan_legacy, INT, ZMOD_RW, "Scrub using legacy non-sequential method"); ZFS_MODULE_PARAM(zfs, zfs_, scan_checkpoint_intval, INT, ZMOD_RW, "Scan progress on-disk checkpointing interval"); ZFS_MODULE_PARAM(zfs, zfs_, scan_max_ext_gap, ULONG, ZMOD_RW, "Max gap in bytes between sequential scrub / resilver I/Os"); ZFS_MODULE_PARAM(zfs, zfs_, scan_mem_lim_soft_fact, INT, ZMOD_RW, "Fraction of hard limit used as soft limit"); ZFS_MODULE_PARAM(zfs, zfs_, scan_strict_mem_lim, INT, ZMOD_RW, "Tunable to attempt to reduce lock contention"); ZFS_MODULE_PARAM(zfs, zfs_, scan_fill_weight, INT, ZMOD_RW, "Tunable to adjust bias towards more filled segments during scans"); ZFS_MODULE_PARAM(zfs, zfs_, resilver_disable_defer, INT, ZMOD_RW, "Process all resilvers immediately"); /* END CSTYLED */