diff options
Diffstat (limited to 'module/zfs')
32 files changed, 2409 insertions, 514 deletions
diff --git a/module/zfs/Makefile.in b/module/zfs/Makefile.in index 98576d1d2..f1e32a19d 100644 --- a/module/zfs/Makefile.in +++ b/module/zfs/Makefile.in @@ -8,6 +8,7 @@ $(MODULE)-objs += @top_srcdir@/module/zfs/arc.o $(MODULE)-objs += @top_srcdir@/module/zfs/bplist.o $(MODULE)-objs += @top_srcdir@/module/zfs/bpobj.o $(MODULE)-objs += @top_srcdir@/module/zfs/dbuf.o +$(MODULE)-objs += @top_srcdir@/module/zfs/bptree.o $(MODULE)-objs += @top_srcdir@/module/zfs/ddt.o $(MODULE)-objs += @top_srcdir@/module/zfs/ddt_zap.o $(MODULE)-objs += @top_srcdir@/module/zfs/dmu.o @@ -59,6 +60,8 @@ $(MODULE)-objs += @top_srcdir@/module/zfs/vdev_root.o $(MODULE)-objs += @top_srcdir@/module/zfs/zap.o $(MODULE)-objs += @top_srcdir@/module/zfs/zap_leaf.o $(MODULE)-objs += @top_srcdir@/module/zfs/zap_micro.o +$(MODULE)-objs += @top_srcdir@/module/zfs/zfeature.o +$(MODULE)-objs += @top_srcdir@/module/zfs/zfeature_common.o $(MODULE)-objs += @top_srcdir@/module/zfs/zfs_acl.o $(MODULE)-objs += @top_srcdir@/module/zfs/zfs_byteswap.o $(MODULE)-objs += @top_srcdir@/module/zfs/zfs_ctldir.o diff --git a/module/zfs/arc.c b/module/zfs/arc.c index 6ec9f04b7..5e21b1b71 100644 --- a/module/zfs/arc.c +++ b/module/zfs/arc.c @@ -189,6 +189,7 @@ unsigned long zfs_arc_meta_limit = 0; int zfs_arc_grow_retry = 0; int zfs_arc_shrink_shift = 0; int zfs_arc_p_min_shift = 0; +int zfs_disable_dup_eviction = 0; int zfs_arc_meta_prune = 0; /* @@ -307,6 +308,9 @@ typedef struct arc_stats { kstat_named_t arcstat_l2_size; kstat_named_t arcstat_l2_hdr_size; kstat_named_t arcstat_memory_throttle_count; + kstat_named_t arcstat_duplicate_buffers; + kstat_named_t arcstat_duplicate_buffers_size; + kstat_named_t arcstat_duplicate_reads; kstat_named_t arcstat_memory_direct_count; kstat_named_t arcstat_memory_indirect_count; kstat_named_t arcstat_no_grow; @@ -387,6 +391,9 @@ static arc_stats_t arc_stats = { { "l2_size", KSTAT_DATA_UINT64 }, { "l2_hdr_size", KSTAT_DATA_UINT64 }, { "memory_throttle_count", KSTAT_DATA_UINT64 }, + { "duplicate_buffers", KSTAT_DATA_UINT64 }, + { "duplicate_buffers_size", KSTAT_DATA_UINT64 }, + { "duplicate_reads", KSTAT_DATA_UINT64 }, { "memory_direct_count", KSTAT_DATA_UINT64 }, { "memory_indirect_count", KSTAT_DATA_UINT64 }, { "arc_no_grow", KSTAT_DATA_UINT64 }, @@ -1369,6 +1376,17 @@ arc_buf_clone(arc_buf_t *from) hdr->b_buf = buf; arc_get_data_buf(buf); bcopy(from->b_data, buf->b_data, size); + + /* + * This buffer already exists in the arc so create a duplicate + * copy for the caller. If the buffer is associated with user data + * then track the size and number of duplicates. These stats will be + * updated as duplicate buffers are created and destroyed. + */ + if (hdr->b_type == ARC_BUFC_DATA) { + ARCSTAT_BUMP(arcstat_duplicate_buffers); + ARCSTAT_INCR(arcstat_duplicate_buffers_size, size); + } hdr->b_datacnt += 1; return (buf); } @@ -1467,6 +1485,16 @@ arc_buf_destroy(arc_buf_t *buf, boolean_t recycle, boolean_t all) ASSERT3U(state->arcs_size, >=, size); atomic_add_64(&state->arcs_size, -size); buf->b_data = NULL; + + /* + * If we're destroying a duplicate buffer make sure + * that the appropriate statistics are updated. + */ + if (buf->b_hdr->b_datacnt > 1 && + buf->b_hdr->b_type == ARC_BUFC_DATA) { + ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers); + ARCSTAT_INCR(arcstat_duplicate_buffers_size, -size); + } ASSERT(buf->b_hdr->b_datacnt > 0); buf->b_hdr->b_datacnt -= 1; } @@ -1652,6 +1680,48 @@ arc_buf_size(arc_buf_t *buf) } /* + * Called from the DMU to determine if the current buffer should be + * evicted. In order to ensure proper locking, the eviction must be initiated + * from the DMU. Return true if the buffer is associated with user data and + * duplicate buffers still exist. + */ +boolean_t +arc_buf_eviction_needed(arc_buf_t *buf) +{ + arc_buf_hdr_t *hdr; + boolean_t evict_needed = B_FALSE; + + if (zfs_disable_dup_eviction) + return (B_FALSE); + + mutex_enter(&buf->b_evict_lock); + hdr = buf->b_hdr; + if (hdr == NULL) { + /* + * We are in arc_do_user_evicts(); let that function + * perform the eviction. + */ + ASSERT(buf->b_data == NULL); + mutex_exit(&buf->b_evict_lock); + return (B_FALSE); + } else if (buf->b_data == NULL) { + /* + * We have already been added to the arc eviction list; + * recommend eviction. + */ + ASSERT3P(hdr, ==, &arc_eviction_hdr); + mutex_exit(&buf->b_evict_lock); + return (B_TRUE); + } + + if (hdr->b_datacnt > 1 && hdr->b_type == ARC_BUFC_DATA) + evict_needed = B_TRUE; + + mutex_exit(&buf->b_evict_lock); + return (evict_needed); +} + +/* * Evict buffers from list until we've removed the specified number of * bytes. Move the removed buffers to the appropriate evict state. * If the recycle flag is set, then attempt to "recycle" a buffer: @@ -2729,9 +2799,11 @@ arc_read_done(zio_t *zio) callback_list = hdr->b_acb; ASSERT(callback_list != NULL); if (BP_SHOULD_BYTESWAP(zio->io_bp) && zio->io_error == 0) { + dmu_object_byteswap_t bswap = + DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp)); arc_byteswap_func_t *func = BP_GET_LEVEL(zio->io_bp) > 0 ? byteswap_uint64_array : - dmu_ot[BP_GET_TYPE(zio->io_bp)].ot_byteswap; + dmu_ot_byteswap[bswap].ob_func; func(buf->b_data, hdr->b_size); } @@ -2751,8 +2823,10 @@ arc_read_done(zio_t *zio) abuf = buf; for (acb = callback_list; acb; acb = acb->acb_next) { if (acb->acb_done) { - if (abuf == NULL) + if (abuf == NULL) { + ARCSTAT_BUMP(arcstat_duplicate_reads); abuf = arc_buf_clone(buf); + } acb->acb_buf = abuf; abuf = NULL; } @@ -3322,6 +3396,16 @@ arc_release(arc_buf_t *buf, void *tag) ASSERT3U(*size, >=, hdr->b_size); atomic_add_64(size, -hdr->b_size); } + + /* + * We're releasing a duplicate user data buffer, update + * our statistics accordingly. + */ + if (hdr->b_type == ARC_BUFC_DATA) { + ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers); + ARCSTAT_INCR(arcstat_duplicate_buffers_size, + -hdr->b_size); + } hdr->b_datacnt -= 1; arc_cksum_verify(buf); diff --git a/module/zfs/bpobj.c b/module/zfs/bpobj.c index 022921c66..d5f8d4072 100644 --- a/module/zfs/bpobj.c +++ b/module/zfs/bpobj.c @@ -20,13 +20,61 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011 by Delphix. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/bpobj.h> #include <sys/zfs_context.h> #include <sys/refcount.h> #include <sys/dsl_pool.h> +#include <sys/zfeature.h> +#include <sys/zap.h> + +/* + * Return an empty bpobj, preferably the empty dummy one (dp_empty_bpobj). + */ +uint64_t +bpobj_alloc_empty(objset_t *os, int blocksize, dmu_tx_t *tx) +{ + zfeature_info_t *empty_bpobj_feat = + &spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ]; + spa_t *spa = dmu_objset_spa(os); + dsl_pool_t *dp = dmu_objset_pool(os); + + if (spa_feature_is_enabled(spa, empty_bpobj_feat)) { + if (!spa_feature_is_active(spa, empty_bpobj_feat)) { + ASSERT3U(dp->dp_empty_bpobj, ==, 0); + dp->dp_empty_bpobj = + bpobj_alloc(os, SPA_MAXBLOCKSIZE, tx); + VERIFY(zap_add(os, + DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, + &dp->dp_empty_bpobj, tx) == 0); + } + spa_feature_incr(spa, empty_bpobj_feat, tx); + ASSERT(dp->dp_empty_bpobj != 0); + return (dp->dp_empty_bpobj); + } else { + return (bpobj_alloc(os, blocksize, tx)); + } +} + +void +bpobj_decr_empty(objset_t *os, dmu_tx_t *tx) +{ + zfeature_info_t *empty_bpobj_feat = + &spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ]; + dsl_pool_t *dp = dmu_objset_pool(os); + + spa_feature_decr(dmu_objset_spa(os), empty_bpobj_feat, tx); + if (!spa_feature_is_active(dmu_objset_spa(os), empty_bpobj_feat)) { + VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset, + DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_EMPTY_BPOBJ, tx)); + VERIFY3U(0, ==, dmu_object_free(os, dp->dp_empty_bpobj, tx)); + dp->dp_empty_bpobj = 0; + } +} uint64_t bpobj_alloc(objset_t *os, int blocksize, dmu_tx_t *tx) @@ -53,6 +101,7 @@ bpobj_free(objset_t *os, uint64_t obj, dmu_tx_t *tx) int epb; dmu_buf_t *dbuf = NULL; + ASSERT(obj != dmu_objset_pool(os)->dp_empty_bpobj); VERIFY3U(0, ==, bpobj_open(&bpo, os, obj)); mutex_enter(&bpo.bpo_lock); @@ -320,6 +369,12 @@ bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx) ASSERT(bpo->bpo_havesubobj); ASSERT(bpo->bpo_havecomp); + ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj); + + if (subobj == dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj) { + bpobj_decr_empty(bpo->bpo_os, tx); + return; + } VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj)); VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp)); @@ -388,6 +443,7 @@ bpobj_enqueue(bpobj_t *bpo, const blkptr_t *bp, dmu_tx_t *tx) blkptr_t *bparray; ASSERT(!BP_IS_HOLE(bp)); + ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj); /* We never need the fill count. */ stored_bp.blk_fill = 0; diff --git a/module/zfs/bptree.c b/module/zfs/bptree.c new file mode 100644 index 000000000..8c5a7d40e --- /dev/null +++ b/module/zfs/bptree.c @@ -0,0 +1,224 @@ +/* + * 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) 2012 by Delphix. All rights reserved. + */ + +#include <sys/arc.h> +#include <sys/bptree.h> +#include <sys/dmu.h> +#include <sys/dmu_objset.h> +#include <sys/dmu_tx.h> +#include <sys/dmu_traverse.h> +#include <sys/dsl_dataset.h> +#include <sys/dsl_dir.h> +#include <sys/dsl_pool.h> +#include <sys/dnode.h> +#include <sys/refcount.h> +#include <sys/spa.h> + +/* + * A bptree is a queue of root block pointers from destroyed datasets. When a + * dataset is destroyed its root block pointer is put on the end of the pool's + * bptree queue so the dataset's blocks can be freed asynchronously by + * dsl_scan_sync. This allows the delete operation to finish without traversing + * all the dataset's blocks. + * + * Note that while bt_begin and bt_end are only ever incremented in this code + * they are effectively reset to 0 every time the entire bptree is freed because + * the bptree's object is destroyed and re-created. + */ + +struct bptree_args { + bptree_phys_t *ba_phys; /* data in bonus buffer, dirtied if freeing */ + boolean_t ba_free; /* true if freeing during traversal */ + + bptree_itor_t *ba_func; /* function to call for each blockpointer */ + void *ba_arg; /* caller supplied argument to ba_func */ + dmu_tx_t *ba_tx; /* caller supplied tx, NULL if not freeing */ +} bptree_args_t; + +uint64_t +bptree_alloc(objset_t *os, dmu_tx_t *tx) +{ + uint64_t obj; + dmu_buf_t *db; + bptree_phys_t *bt; + + obj = dmu_object_alloc(os, DMU_OTN_UINT64_METADATA, + SPA_MAXBLOCKSIZE, DMU_OTN_UINT64_METADATA, + sizeof (bptree_phys_t), tx); + + /* + * Bonus buffer contents are already initialized to 0, but for + * readability we make it explicit. + */ + VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db)); + dmu_buf_will_dirty(db, tx); + bt = db->db_data; + bt->bt_begin = 0; + bt->bt_end = 0; + bt->bt_bytes = 0; + bt->bt_comp = 0; + bt->bt_uncomp = 0; + dmu_buf_rele(db, FTAG); + + return (obj); +} + +int +bptree_free(objset_t *os, uint64_t obj, dmu_tx_t *tx) +{ + dmu_buf_t *db; + bptree_phys_t *bt; + + VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db)); + bt = db->db_data; + ASSERT3U(bt->bt_begin, ==, bt->bt_end); + ASSERT3U(bt->bt_bytes, ==, 0); + ASSERT3U(bt->bt_comp, ==, 0); + ASSERT3U(bt->bt_uncomp, ==, 0); + dmu_buf_rele(db, FTAG); + + return (dmu_object_free(os, obj, tx)); +} + +void +bptree_add(objset_t *os, uint64_t obj, blkptr_t *bp, uint64_t birth_txg, + uint64_t bytes, uint64_t comp, uint64_t uncomp, dmu_tx_t *tx) +{ + dmu_buf_t *db; + bptree_phys_t *bt; + bptree_entry_phys_t bte; + + /* + * bptree objects are in the pool mos, therefore they can only be + * modified in syncing context. Furthermore, this is only modified + * by the sync thread, so no locking is necessary. + */ + ASSERT(dmu_tx_is_syncing(tx)); + + VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db)); + bt = db->db_data; + + bte.be_birth_txg = birth_txg; + bte.be_bp = *bp; + bzero(&bte.be_zb, sizeof (bte.be_zb)); + dmu_write(os, obj, bt->bt_end * sizeof (bte), sizeof (bte), &bte, tx); + + dmu_buf_will_dirty(db, tx); + bt->bt_end++; + bt->bt_bytes += bytes; + bt->bt_comp += comp; + bt->bt_uncomp += uncomp; + dmu_buf_rele(db, FTAG); +} + +/* ARGSUSED */ +static int +bptree_visit_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf, + const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) +{ + int err; + struct bptree_args *ba = arg; + + if (bp == NULL) + return (0); + + err = ba->ba_func(ba->ba_arg, bp, ba->ba_tx); + if (err == 0 && ba->ba_free) { + ba->ba_phys->bt_bytes -= bp_get_dsize_sync(spa, bp); + ba->ba_phys->bt_comp -= BP_GET_PSIZE(bp); + ba->ba_phys->bt_uncomp -= BP_GET_UCSIZE(bp); + } + return (err); +} + +int +bptree_iterate(objset_t *os, uint64_t obj, boolean_t free, bptree_itor_t func, + void *arg, dmu_tx_t *tx) +{ + int err; + uint64_t i; + dmu_buf_t *db; + struct bptree_args ba; + + ASSERT(!free || dmu_tx_is_syncing(tx)); + + err = dmu_bonus_hold(os, obj, FTAG, &db); + if (err != 0) + return (err); + + if (free) + dmu_buf_will_dirty(db, tx); + + ba.ba_phys = db->db_data; + ba.ba_free = free; + ba.ba_func = func; + ba.ba_arg = arg; + ba.ba_tx = tx; + + err = 0; + for (i = ba.ba_phys->bt_begin; i < ba.ba_phys->bt_end; i++) { + bptree_entry_phys_t bte; + + ASSERT(!free || i == ba.ba_phys->bt_begin); + + err = dmu_read(os, obj, i * sizeof (bte), sizeof (bte), + &bte, DMU_READ_NO_PREFETCH); + if (err != 0) + break; + + err = traverse_dataset_destroyed(os->os_spa, &bte.be_bp, + bte.be_birth_txg, &bte.be_zb, TRAVERSE_POST, + bptree_visit_cb, &ba); + if (free) { + ASSERT(err == 0 || err == ERESTART); + if (err != 0) { + /* save bookmark for future resume */ + ASSERT3U(bte.be_zb.zb_objset, ==, + ZB_DESTROYED_OBJSET); + ASSERT3U(bte.be_zb.zb_level, ==, 0); + dmu_write(os, obj, i * sizeof (bte), + sizeof (bte), &bte, tx); + break; + } else { + ba.ba_phys->bt_begin++; + (void) dmu_free_range(os, obj, + i * sizeof (bte), sizeof (bte), tx); + } + } + } + + ASSERT(!free || err != 0 || ba.ba_phys->bt_begin == ba.ba_phys->bt_end); + + /* if all blocks are free there should be no used space */ + if (ba.ba_phys->bt_begin == ba.ba_phys->bt_end) { + ASSERT3U(ba.ba_phys->bt_bytes, ==, 0); + ASSERT3U(ba.ba_phys->bt_comp, ==, 0); + ASSERT3U(ba.ba_phys->bt_uncomp, ==, 0); + } + + dmu_buf_rele(db, FTAG); + + return (err); +} diff --git a/module/zfs/dbuf.c b/module/zfs/dbuf.c index 1f6fa9340..205abaada 100644 --- a/module/zfs/dbuf.c +++ b/module/zfs/dbuf.c @@ -21,6 +21,7 @@ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/zfs_context.h> @@ -260,7 +261,7 @@ dbuf_is_metadata(dmu_buf_impl_t *db) boolean_t is_metadata; DB_DNODE_ENTER(db); - is_metadata = dmu_ot[DB_DNODE(db)->dn_type].ot_metadata; + is_metadata = DMU_OT_IS_METADATA(DB_DNODE(db)->dn_type); DB_DNODE_EXIT(db); return (is_metadata); @@ -2188,7 +2189,24 @@ dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag) dbuf_evict(db); } else { VERIFY(arc_buf_remove_ref(db->db_buf, db) == 0); - if (!DBUF_IS_CACHEABLE(db)) + + /* + * A dbuf will be eligible for eviction if either the + * 'primarycache' property is set or a duplicate + * copy of this buffer is already cached in the arc. + * + * In the case of the 'primarycache' a buffer + * is considered for eviction if it matches the + * criteria set in the property. + * + * To decide if our buffer is considered a + * duplicate, we must call into the arc to determine + * if multiple buffers are referencing the same + * block on-disk. If so, then we simply evict + * ourselves. + */ + if (!DBUF_IS_CACHEABLE(db) || + arc_buf_eviction_needed(db->db_buf)) dbuf_clear(db); else mutex_exit(&db->db_mtx); diff --git a/module/zfs/ddt.c b/module/zfs/ddt.c index 6221157f8..ef868619a 100644 --- a/module/zfs/ddt.c +++ b/module/zfs/ddt.c @@ -21,6 +21,7 @@ /* * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/zfs_context.h> @@ -1120,11 +1121,9 @@ ddt_sync_table(ddt_t *ddt, dmu_tx_t *tx, uint64_t txg) ASSERT(spa->spa_uberblock.ub_version >= SPA_VERSION_DEDUP); if (spa->spa_ddt_stat_object == 0) { - spa->spa_ddt_stat_object = zap_create(ddt->ddt_os, - DMU_OT_DDT_STATS, DMU_OT_NONE, 0, tx); - VERIFY(zap_add(ddt->ddt_os, DMU_POOL_DIRECTORY_OBJECT, - DMU_POOL_DDT_STATS, sizeof (uint64_t), 1, - &spa->spa_ddt_stat_object, tx) == 0); + spa->spa_ddt_stat_object = zap_create_link(ddt->ddt_os, + DMU_OT_DDT_STATS, DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_DDT_STATS, tx); } while ((dde = avl_destroy_nodes(&ddt->ddt_tree, &cookie)) != NULL) { diff --git a/module/zfs/dmu.c b/module/zfs/dmu.c index 00a7a07f4..e85635654 100644 --- a/module/zfs/dmu.c +++ b/module/zfs/dmu.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/dmu.h> @@ -46,60 +47,73 @@ #endif const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { - { byteswap_uint8_array, TRUE, "unallocated" }, - { zap_byteswap, TRUE, "object directory" }, - { byteswap_uint64_array, TRUE, "object array" }, - { byteswap_uint8_array, TRUE, "packed nvlist" }, - { byteswap_uint64_array, TRUE, "packed nvlist size" }, - { byteswap_uint64_array, TRUE, "bpobj" }, - { byteswap_uint64_array, TRUE, "bpobj header" }, - { byteswap_uint64_array, TRUE, "SPA space map header" }, - { byteswap_uint64_array, TRUE, "SPA space map" }, - { byteswap_uint64_array, TRUE, "ZIL intent log" }, - { dnode_buf_byteswap, TRUE, "DMU dnode" }, - { dmu_objset_byteswap, TRUE, "DMU objset" }, - { byteswap_uint64_array, TRUE, "DSL directory" }, - { zap_byteswap, TRUE, "DSL directory child map"}, - { zap_byteswap, TRUE, "DSL dataset snap map" }, - { zap_byteswap, TRUE, "DSL props" }, - { byteswap_uint64_array, TRUE, "DSL dataset" }, - { zfs_znode_byteswap, TRUE, "ZFS znode" }, - { zfs_oldacl_byteswap, TRUE, "ZFS V0 ACL" }, - { byteswap_uint8_array, FALSE, "ZFS plain file" }, - { zap_byteswap, TRUE, "ZFS directory" }, - { zap_byteswap, TRUE, "ZFS master node" }, - { zap_byteswap, TRUE, "ZFS delete queue" }, - { byteswap_uint8_array, FALSE, "zvol object" }, - { zap_byteswap, TRUE, "zvol prop" }, - { byteswap_uint8_array, FALSE, "other uint8[]" }, - { byteswap_uint64_array, FALSE, "other uint64[]" }, - { zap_byteswap, TRUE, "other ZAP" }, - { zap_byteswap, TRUE, "persistent error log" }, - { byteswap_uint8_array, TRUE, "SPA history" }, - { byteswap_uint64_array, TRUE, "SPA history offsets" }, - { zap_byteswap, TRUE, "Pool properties" }, - { zap_byteswap, TRUE, "DSL permissions" }, - { zfs_acl_byteswap, TRUE, "ZFS ACL" }, - { byteswap_uint8_array, TRUE, "ZFS SYSACL" }, - { byteswap_uint8_array, TRUE, "FUID table" }, - { byteswap_uint64_array, TRUE, "FUID table size" }, - { zap_byteswap, TRUE, "DSL dataset next clones"}, - { zap_byteswap, TRUE, "scan work queue" }, - { zap_byteswap, TRUE, "ZFS user/group used" }, - { zap_byteswap, TRUE, "ZFS user/group quota" }, - { zap_byteswap, TRUE, "snapshot refcount tags"}, - { zap_byteswap, TRUE, "DDT ZAP algorithm" }, - { zap_byteswap, TRUE, "DDT statistics" }, - { byteswap_uint8_array, TRUE, "System attributes" }, - { zap_byteswap, TRUE, "SA master node" }, - { zap_byteswap, TRUE, "SA attr registration" }, - { zap_byteswap, TRUE, "SA attr layouts" }, - { zap_byteswap, TRUE, "scan translations" }, - { byteswap_uint8_array, FALSE, "deduplicated block" }, - { zap_byteswap, TRUE, "DSL deadlist map" }, - { byteswap_uint64_array, TRUE, "DSL deadlist map hdr" }, - { zap_byteswap, TRUE, "DSL dir clones" }, - { byteswap_uint64_array, TRUE, "bpobj subobj" }, + { DMU_BSWAP_UINT8, TRUE, "unallocated" }, + { DMU_BSWAP_ZAP, TRUE, "object directory" }, + { DMU_BSWAP_UINT64, TRUE, "object array" }, + { DMU_BSWAP_UINT8, TRUE, "packed nvlist" }, + { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" }, + { DMU_BSWAP_UINT64, TRUE, "bpobj" }, + { DMU_BSWAP_UINT64, TRUE, "bpobj header" }, + { DMU_BSWAP_UINT64, TRUE, "SPA space map header" }, + { DMU_BSWAP_UINT64, TRUE, "SPA space map" }, + { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" }, + { DMU_BSWAP_DNODE, TRUE, "DMU dnode" }, + { DMU_BSWAP_OBJSET, TRUE, "DMU objset" }, + { DMU_BSWAP_UINT64, TRUE, "DSL directory" }, + { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"}, + { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" }, + { DMU_BSWAP_ZAP, TRUE, "DSL props" }, + { DMU_BSWAP_UINT64, TRUE, "DSL dataset" }, + { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" }, + { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" }, + { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" }, + { DMU_BSWAP_ZAP, TRUE, "ZFS directory" }, + { DMU_BSWAP_ZAP, TRUE, "ZFS master node" }, + { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" }, + { DMU_BSWAP_UINT8, FALSE, "zvol object" }, + { DMU_BSWAP_ZAP, TRUE, "zvol prop" }, + { DMU_BSWAP_UINT8, FALSE, "other uint8[]" }, + { DMU_BSWAP_UINT64, FALSE, "other uint64[]" }, + { DMU_BSWAP_ZAP, TRUE, "other ZAP" }, + { DMU_BSWAP_ZAP, TRUE, "persistent error log" }, + { DMU_BSWAP_UINT8, TRUE, "SPA history" }, + { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" }, + { DMU_BSWAP_ZAP, TRUE, "Pool properties" }, + { DMU_BSWAP_ZAP, TRUE, "DSL permissions" }, + { DMU_BSWAP_ACL, TRUE, "ZFS ACL" }, + { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" }, + { DMU_BSWAP_UINT8, TRUE, "FUID table" }, + { DMU_BSWAP_UINT64, TRUE, "FUID table size" }, + { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"}, + { DMU_BSWAP_ZAP, TRUE, "scan work queue" }, + { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" }, + { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" }, + { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"}, + { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" }, + { DMU_BSWAP_ZAP, TRUE, "DDT statistics" }, + { DMU_BSWAP_UINT8, TRUE, "System attributes" }, + { DMU_BSWAP_ZAP, TRUE, "SA master node" }, + { DMU_BSWAP_ZAP, TRUE, "SA attr registration" }, + { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" }, + { DMU_BSWAP_ZAP, TRUE, "scan translations" }, + { DMU_BSWAP_UINT8, FALSE, "deduplicated block" }, + { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" }, + { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" }, + { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" }, + { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" } +}; + +const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { + { byteswap_uint8_array, "uint8" }, + { byteswap_uint16_array, "uint16" }, + { byteswap_uint32_array, "uint32" }, + { byteswap_uint64_array, "uint64" }, + { zap_byteswap, "zap" }, + { dnode_buf_byteswap, "dnode" }, + { dmu_objset_byteswap, "objset" }, + { zfs_znode_byteswap, "znode" }, + { zfs_oldacl_byteswap, "oldacl" }, + { zfs_acl_byteswap, "acl" } }; int @@ -176,7 +190,7 @@ dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) DB_DNODE_ENTER(db); dn = DB_DNODE(db); - if (type > DMU_OT_NUMTYPES) { + if (!DMU_OT_IS_VALID(type)) { error = EINVAL; } else if (dn->dn_bonus != db) { error = EINVAL; @@ -1695,7 +1709,7 @@ void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) { dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; - boolean_t ismd = (level > 0 || dmu_ot[type].ot_metadata || + boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)); enum zio_checksum checksum = os->os_checksum; enum zio_compress compress = os->os_compress; @@ -1939,15 +1953,15 @@ dmu_init(void) dbuf_init(); zfetch_init(); dmu_tx_init(); - arc_init(); l2arc_init(); + arc_init(); } void dmu_fini(void) { - l2arc_fini(); arc_fini(); + l2arc_fini(); dmu_tx_fini(); zfetch_fini(); dbuf_fini(); diff --git a/module/zfs/dmu_send.c b/module/zfs/dmu_send.c index 949f4d773..0cf3c4a9a 100644 --- a/module/zfs/dmu_send.c +++ b/module/zfs/dmu_send.c @@ -1077,8 +1077,8 @@ restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) void *data = NULL; if (drro->drr_type == DMU_OT_NONE || - drro->drr_type >= DMU_OT_NUMTYPES || - drro->drr_bonustype >= DMU_OT_NUMTYPES || + !DMU_OT_IS_VALID(drro->drr_type) || + !DMU_OT_IS_VALID(drro->drr_bonustype) || drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || @@ -1143,7 +1143,9 @@ restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro) ASSERT3U(db->db_size, >=, drro->drr_bonuslen); bcopy(data, db->db_data, drro->drr_bonuslen); if (ra->byteswap) { - dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data, + dmu_object_byteswap_t byteswap = + DMU_OT_BYTESWAP(drro->drr_bonustype); + dmu_ot_byteswap[byteswap].ob_func(db->db_data, drro->drr_bonuslen); } dmu_buf_rele(db, FTAG); @@ -1186,7 +1188,7 @@ restore_write(struct restorearg *ra, objset_t *os, int err; if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset || - drrw->drr_type >= DMU_OT_NUMTYPES) + !DMU_OT_IS_VALID(drrw->drr_type)) return (EINVAL); data = restore_read(ra, drrw->drr_length); @@ -1205,8 +1207,11 @@ restore_write(struct restorearg *ra, objset_t *os, dmu_tx_abort(tx); return (err); } - if (ra->byteswap) - dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length); + if (ra->byteswap) { + dmu_object_byteswap_t byteswap = + DMU_OT_BYTESWAP(drrw->drr_type); + dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length); + } dmu_write(os, drrw->drr_object, drrw->drr_offset, drrw->drr_length, data, tx); dmu_tx_commit(tx); @@ -1604,13 +1609,6 @@ dmu_recv_existing_end(dmu_recv_cookie_t *drc) dsl_dataset_t *ds = drc->drc_logical_ds; int err, myerr; - /* - * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean() - * expects it to have a ds_user_ptr (and zil), but clone_swap() - * can close it. - */ - txg_wait_synced(ds->ds_dir->dd_pool, 0); - if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) { err = dsl_dataset_clone_swap(drc->drc_real_ds, ds, drc->drc_force); diff --git a/module/zfs/dmu_traverse.c b/module/zfs/dmu_traverse.c index 376f60f82..980c1aa2f 100644 --- a/module/zfs/dmu_traverse.c +++ b/module/zfs/dmu_traverse.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/zfs_context.h> @@ -53,6 +54,7 @@ typedef struct traverse_data { uint64_t td_objset; blkptr_t *td_rootbp; uint64_t td_min_txg; + zbookmark_t *td_resume; int td_flags; prefetch_data_t *td_pfd; blkptr_cb_t *td_func; @@ -128,6 +130,54 @@ traverse_zil(traverse_data_t *td, zil_header_t *zh) zil_free(zilog); } +typedef enum resume_skip { + RESUME_SKIP_ALL, + RESUME_SKIP_NONE, + RESUME_SKIP_CHILDREN +} resume_skip_t; + +/* + * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and + * the block indicated by zb does not need to be visited at all. Returns + * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the + * resume point. This indicates that this block should be visited but not its + * children (since they must have been visited in a previous traversal). + * Otherwise returns RESUME_SKIP_NONE. + */ +static resume_skip_t +resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp, + const zbookmark_t *zb) +{ + if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) { + /* + * If we already visited this bp & everything below, + * don't bother doing it again. + */ + if (zbookmark_is_before(dnp, zb, td->td_resume)) + return (RESUME_SKIP_ALL); + + /* + * If we found the block we're trying to resume from, zero + * the bookmark out to indicate that we have resumed. + */ + ASSERT3U(zb->zb_object, <=, td->td_resume->zb_object); + if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) { + bzero(td->td_resume, sizeof (*zb)); + if (td->td_flags & TRAVERSE_POST) + return (RESUME_SKIP_CHILDREN); + } + } + return (RESUME_SKIP_NONE); +} + +static void +traverse_pause(traverse_data_t *td, const zbookmark_t *zb) +{ + ASSERT(td->td_resume != NULL); + ASSERT3U(zb->zb_level, ==, 0); + bcopy(zb, td->td_resume, sizeof (*td->td_resume)); +} + static int traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb) @@ -137,8 +187,20 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, arc_buf_t *buf = NULL; prefetch_data_t *pd = td->td_pfd; boolean_t hard = td->td_flags & TRAVERSE_HARD; + boolean_t pause = B_FALSE; + + switch (resume_skip_check(td, dnp, zb)) { + case RESUME_SKIP_ALL: + return (0); + case RESUME_SKIP_CHILDREN: + goto post; + case RESUME_SKIP_NONE: + break; + default: + ASSERT(0); + } - if (bp->blk_birth == 0) { + if (BP_IS_HOLE(bp)) { err = td->td_func(td->td_spa, NULL, NULL, pbuf, zb, dnp, td->td_arg); return (err); @@ -164,8 +226,10 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, td->td_arg); if (err == TRAVERSE_VISIT_NO_CHILDREN) return (0); - if (err) - return (err); + if (err == ERESTART) + pause = B_TRUE; /* handle pausing at a common point */ + if (err != 0) + goto post; } if (BP_GET_LEVEL(bp) > 0) { @@ -253,9 +317,18 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, if (buf) (void) arc_buf_remove_ref(buf, &buf); +post: if (err == 0 && lasterr == 0 && (td->td_flags & TRAVERSE_POST)) { err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp, td->td_arg); + if (err == ERESTART) + pause = B_TRUE; + } + + if (pause && td->td_resume != NULL) { + ASSERT3U(err, ==, ERESTART); + ASSERT(!hard); + traverse_pause(td, zb); } return (err != 0 ? err : lasterr); @@ -353,22 +426,27 @@ traverse_prefetch_thread(void *arg) * in syncing context). */ static int -traverse_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *rootbp, - uint64_t txg_start, int flags, blkptr_cb_t func, void *arg) +traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp, + uint64_t txg_start, zbookmark_t *resume, int flags, + blkptr_cb_t func, void *arg) { traverse_data_t *td; prefetch_data_t *pd; zbookmark_t *czb; int err; + ASSERT(ds == NULL || objset == ds->ds_object); + ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST)); + td = kmem_alloc(sizeof(traverse_data_t), KM_PUSHPAGE); pd = kmem_zalloc(sizeof(prefetch_data_t), KM_PUSHPAGE); czb = kmem_alloc(sizeof(zbookmark_t), KM_PUSHPAGE); td->td_spa = spa; - td->td_objset = ds ? ds->ds_object : 0; + td->td_objset = objset; td->td_rootbp = rootbp; td->td_min_txg = txg_start; + td->td_resume = resume; td->td_func = func; td->td_arg = arg; td->td_pfd = pd; @@ -424,8 +502,17 @@ int traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags, blkptr_cb_t func, void *arg) { - return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, - &ds->ds_phys->ds_bp, txg_start, flags, func, arg)); + return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object, + &ds->ds_phys->ds_bp, txg_start, NULL, flags, func, arg)); +} + +int +traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr, + uint64_t txg_start, zbookmark_t *resume, int flags, + blkptr_cb_t func, void *arg) +{ + return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET, + blkptr, txg_start, resume, flags, func, arg)); } /* @@ -442,8 +529,8 @@ traverse_pool(spa_t *spa, uint64_t txg_start, int flags, boolean_t hard = (flags & TRAVERSE_HARD); /* visit the MOS */ - err = traverse_impl(spa, NULL, spa_get_rootblkptr(spa), - txg_start, flags, func, arg); + err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa), + txg_start, NULL, flags, func, arg); if (err) return (err); diff --git a/module/zfs/dmu_tx.c b/module/zfs/dmu_tx.c index 81c6dfea2..47ec4c109 100644 --- a/module/zfs/dmu_tx.c +++ b/module/zfs/dmu_tx.c @@ -20,9 +20,8 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - */ -/* * Copyright 2011 Nexenta Systems, Inc. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/dmu.h> @@ -693,7 +692,7 @@ dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name) return; } - ASSERT3P(dmu_ot[dn->dn_type].ot_byteswap, ==, zap_byteswap); + ASSERT3U(DMU_OT_BYTESWAP(dn->dn_type), ==, DMU_BSWAP_ZAP); if (dn->dn_maxblkid == 0 && !add) { blkptr_t *bp; diff --git a/module/zfs/dnode.c b/module/zfs/dnode.c index 99ac62565..3a8a5e32e 100644 --- a/module/zfs/dnode.c +++ b/module/zfs/dnode.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/zfs_context.h> @@ -194,7 +195,7 @@ dnode_verify(dnode_t *dn) ASSERT(dn->dn_objset); ASSERT(dn->dn_handle->dnh_dnode == dn); - ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); + ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type)); if (!(zfs_flags & ZFS_DEBUG_DNODE_VERIFY)) return; @@ -212,7 +213,7 @@ dnode_verify(dnode_t *dn) ASSERT3U(1<<dn->dn_datablkshift, ==, dn->dn_datablksz); } ASSERT3U(dn->dn_nlevels, <=, 30); - ASSERT3U(dn->dn_type, <=, DMU_OT_NUMTYPES); + ASSERT(DMU_OT_IS_VALID(dn->dn_type)); ASSERT3U(dn->dn_nblkptr, >=, 1); ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); ASSERT3U(dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); @@ -278,8 +279,10 @@ dnode_byteswap(dnode_phys_t *dnp) */ int off = (dnp->dn_nblkptr-1) * sizeof (blkptr_t); size_t len = DN_MAX_BONUSLEN - off; - ASSERT3U(dnp->dn_bonustype, <, DMU_OT_NUMTYPES); - dmu_ot[dnp->dn_bonustype].ot_byteswap(dnp->dn_bonus + off, len); + dmu_object_byteswap_t byteswap; + ASSERT(DMU_OT_IS_VALID(dnp->dn_bonustype)); + byteswap = DMU_OT_BYTESWAP(dnp->dn_bonustype); + dmu_ot_byteswap[byteswap].ob_func(dnp->dn_bonus + off, len); } /* Swap SPILL block if we have one */ @@ -407,7 +410,7 @@ dnode_create(objset_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, dmu_zfetch_init(&dn->dn_zfetch, dn); - ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); + ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type)); mutex_enter(&os->os_lock); list_insert_head(&os->os_dnodes, dn); @@ -496,11 +499,11 @@ dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); ASSERT(ot != DMU_OT_NONE); - ASSERT3U(ot, <, DMU_OT_NUMTYPES); + ASSERT(DMU_OT_IS_VALID(ot)); ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || (bonustype == DMU_OT_SA && bonuslen == 0) || (bonustype != DMU_OT_NONE && bonuslen != 0)); - ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); + ASSERT(DMU_OT_IS_VALID(bonustype)); ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); ASSERT(dn->dn_type == DMU_OT_NONE); ASSERT3U(dn->dn_maxblkid, ==, 0); @@ -568,7 +571,7 @@ dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || (bonustype != DMU_OT_NONE && bonuslen != 0) || (bonustype == DMU_OT_SA && bonuslen == 0)); - ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); + ASSERT(DMU_OT_IS_VALID(bonustype)); ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); /* clean up any unreferenced dbufs */ diff --git a/module/zfs/dnode_sync.c b/module/zfs/dnode_sync.c index f2dda867d..58fa473cb 100644 --- a/module/zfs/dnode_sync.c +++ b/module/zfs/dnode_sync.c @@ -18,8 +18,10 @@ * * CDDL HEADER END */ + /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/zfs_context.h> @@ -598,7 +600,7 @@ dnode_sync(dnode_t *dn, dmu_tx_t *tx) } if (dn->dn_next_bonustype[txgoff]) { - ASSERT(dn->dn_next_bonustype[txgoff] < DMU_OT_NUMTYPES); + ASSERT(DMU_OT_IS_VALID(dn->dn_next_bonustype[txgoff])); dnp->dn_bonustype = dn->dn_next_bonustype[txgoff]; dn->dn_next_bonustype[txgoff] = 0; } diff --git a/module/zfs/dsl_dataset.c b/module/zfs/dsl_dataset.c index 21fdd081c..c5b84a26c 100644 --- a/module/zfs/dsl_dataset.c +++ b/module/zfs/dsl_dataset.c @@ -20,7 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011 by Delphix. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. * Copyright (c) 2012, Joyent, Inc. All rights reserved. */ @@ -35,6 +35,7 @@ #include <sys/arc.h> #include <sys/zio.h> #include <sys/zap.h> +#include <sys/zfeature.h> #include <sys/unique.h> #include <sys/zfs_context.h> #include <sys/zfs_ioctl.h> @@ -102,16 +103,10 @@ dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) if (BP_IS_HOLE(bp)) return; ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); - ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES); + ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); if (ds == NULL) { - /* - * Account for the meta-objset space in its placeholder - * dsl_dir. - */ - ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */ - dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, - used, compressed, uncompressed, tx); - dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); + dsl_pool_mos_diduse_space(tx->tx_pool, + used, compressed, uncompressed); return; } dmu_buf_will_dirty(ds->ds_dbuf, tx); @@ -119,7 +114,7 @@ dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) mutex_enter(&ds->ds_dir->dd_lock); mutex_enter(&ds->ds_lock); delta = parent_delta(ds, used); - ds->ds_phys->ds_used_bytes += used; + ds->ds_phys->ds_referenced_bytes += used; ds->ds_phys->ds_compressed_bytes += compressed; ds->ds_phys->ds_uncompressed_bytes += uncompressed; ds->ds_phys->ds_unique_bytes += used; @@ -149,15 +144,9 @@ dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, ASSERT(used > 0); if (ds == NULL) { - /* - * Account for the meta-objset space in its placeholder - * dataset. - */ dsl_free(tx->tx_pool, tx->tx_txg, bp); - - dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, - -used, -compressed, -uncompressed, tx); - dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); + dsl_pool_mos_diduse_space(tx->tx_pool, + -used, -compressed, -uncompressed); return (used); } ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); @@ -215,8 +204,8 @@ dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, } } mutex_enter(&ds->ds_lock); - ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); - ds->ds_phys->ds_used_bytes -= used; + ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used); + ds->ds_phys->ds_referenced_bytes -= used; ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); ds->ds_phys->ds_compressed_bytes -= compressed; ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); @@ -823,8 +812,8 @@ dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, dsphys->ds_prev_snap_obj = origin->ds_object; dsphys->ds_prev_snap_txg = origin->ds_phys->ds_creation_txg; - dsphys->ds_used_bytes = - origin->ds_phys->ds_used_bytes; + dsphys->ds_referenced_bytes = + origin->ds_phys->ds_referenced_bytes; dsphys->ds_compressed_bytes = origin->ds_phys->ds_compressed_bytes; dsphys->ds_uncompressed_bytes = @@ -938,7 +927,6 @@ dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer, char *failed) for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; pair = nvlist_next_nvpair(snaps, pair)) { dsl_dataset_t *ds; - int err; err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds); if (err == 0) { @@ -1074,55 +1062,55 @@ dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer) dummy_ds->ds_dir = dd; dummy_ds->ds_object = ds->ds_object; - /* - * Check for errors and mark this ds as inconsistent, in - * case we crash while freeing the objects. - */ - err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check, - dsl_dataset_destroy_begin_sync, ds, NULL, 0); - if (err) - goto out_free; + if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds), + &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) { + /* + * Check for errors and mark this ds as inconsistent, in + * case we crash while freeing the objects. + */ + err = dsl_sync_task_do(dd->dd_pool, + dsl_dataset_destroy_begin_check, + dsl_dataset_destroy_begin_sync, ds, NULL, 0); + if (err) + goto out_free; - err = dmu_objset_from_ds(ds, &os); - if (err) - goto out_free; + err = dmu_objset_from_ds(ds, &os); + if (err) + goto out_free; - /* - * remove the objects in open context, so that we won't - * have too much to do in syncing context. - */ - for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, - ds->ds_phys->ds_prev_snap_txg)) { /* - * Ignore errors, if there is not enough disk space - * we will deal with it in dsl_dataset_destroy_sync(). + * Remove all objects while in the open context so that + * there is less work to do in the syncing context. */ - (void) dmu_free_object(os, obj); - } - if (err != ESRCH) - goto out_free; - - /* - * Only the ZIL knows how to free log blocks. - */ - zil_destroy(dmu_objset_zil(os), B_FALSE); + for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, + ds->ds_phys->ds_prev_snap_txg)) { + /* + * Ignore errors, if there is not enough disk space + * we will deal with it in dsl_dataset_destroy_sync(). + */ + (void) dmu_free_object(os, obj); + } + if (err != ESRCH) + goto out_free; - /* - * Sync out all in-flight IO. - */ - txg_wait_synced(dd->dd_pool, 0); + /* + * Sync out all in-flight IO. + */ + txg_wait_synced(dd->dd_pool, 0); - /* - * If we managed to free all the objects in open - * context, the user space accounting should be zero. - */ - if (ds->ds_phys->ds_bp.blk_fill == 0 && - dmu_objset_userused_enabled(os)) { - ASSERTV(uint64_t count); - ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 || - count == 0); - ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 || - count == 0); + /* + * If we managed to free all the objects in open + * context, the user space accounting should be zero. + */ + if (ds->ds_phys->ds_bp.blk_fill == 0 && + dmu_objset_userused_enabled(os)) { + ASSERTV(uint64_t count); + + ASSERT(zap_count(os, DMU_USERUSED_OBJECT, + &count) != 0 || count == 0); + ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, + &count) != 0 || count == 0); + } } rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); @@ -1261,7 +1249,7 @@ dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) ASSERT(!dsl_dataset_is_snapshot(ds)); if (ds->ds_phys->ds_prev_snap_obj != 0) - mrs_used = ds->ds_prev->ds_phys->ds_used_bytes; + mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes; else mrs_used = 0; @@ -1269,7 +1257,7 @@ dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) ASSERT3U(dlused, <=, mrs_used); ds->ds_phys->ds_unique_bytes = - ds->ds_phys->ds_used_bytes - (mrs_used - dlused); + ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused); if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) @@ -1627,12 +1615,36 @@ process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev, ds_next->ds_phys->ds_deadlist_obj); } +static int +old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx) +{ + int err; + struct killarg ka; + + /* + * Free everything that we point to (that's born after + * the previous snapshot, if we are a clone) + * + * NB: this should be very quick, because we already + * freed all the objects in open context. + */ + ka.ds = ds; + ka.tx = tx; + err = traverse_dataset(ds, + ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST, + kill_blkptr, &ka); + ASSERT3U(err, ==, 0); + ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0); + + return (err); +} + void dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) { struct dsl_ds_destroyarg *dsda = arg1; dsl_dataset_t *ds = dsda->ds; - int err; + int err = 0; int after_branch_point = FALSE; dsl_pool_t *dp = ds->ds_dir->dd_pool; objset_t *mos = dp->dp_meta_objset; @@ -1773,7 +1785,6 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) tx); dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD, used, comp, uncomp, tx); - dsl_dir_dirty(tx->tx_pool->dp_free_dir, tx); /* Merge our deadlist into next's and free it. */ dsl_deadlist_merge(&ds_next->ds_deadlist, @@ -1849,32 +1860,57 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) } dsl_dataset_rele(ds_next, FTAG); } else { + zfeature_info_t *async_destroy = + &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY]; + objset_t *os; + /* * There's no next snapshot, so this is a head dataset. * Destroy the deadlist. Unless it's a clone, the * deadlist should be empty. (If it's a clone, it's * safe to ignore the deadlist contents.) */ - struct killarg ka; - dsl_deadlist_close(&ds->ds_deadlist); dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); ds->ds_phys->ds_deadlist_obj = 0; - /* - * Free everything that we point to (that's born after - * the previous snapshot, if we are a clone) - * - * NB: this should be very quick, because we already - * freed all the objects in open context. - */ - ka.ds = ds; - ka.tx = tx; - err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg, - TRAVERSE_POST, kill_blkptr, &ka); - ASSERT3U(err, ==, 0); - ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || - ds->ds_phys->ds_unique_bytes == 0); + VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os)); + + if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) { + err = old_synchronous_dataset_destroy(ds, tx); + } else { + /* + * Move the bptree into the pool's list of trees to + * clean up and update space accounting information. + */ + uint64_t used, comp, uncomp; + + zil_destroy_sync(dmu_objset_zil(os), tx); + + if (!spa_feature_is_active(dp->dp_spa, async_destroy)) { + spa_feature_incr(dp->dp_spa, async_destroy, tx); + dp->dp_bptree_obj = bptree_alloc(mos, tx); + VERIFY(zap_add(mos, + DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, + &dp->dp_bptree_obj, tx) == 0); + } + + used = ds->ds_dir->dd_phys->dd_used_bytes; + comp = ds->ds_dir->dd_phys->dd_compressed_bytes; + uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes; + + ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || + ds->ds_phys->ds_unique_bytes == used); + + bptree_add(mos, dp->dp_bptree_obj, + &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg, + used, comp, uncomp, tx); + dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, + -used, -comp, -uncomp, tx); + dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD, + used, comp, uncomp, tx); + } if (ds->ds_prev != NULL) { if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { @@ -2065,7 +2101,7 @@ dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx) dsphys->ds_creation_time = gethrestime_sec(); dsphys->ds_creation_txg = crtxg; dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; - dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; + dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes; dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; dsphys->ds_flags = ds->ds_phys->ds_flags; @@ -2154,7 +2190,6 @@ dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) dmu_buf_will_dirty(ds->ds_dbuf, tx); ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; - dsl_dir_dirty(ds->ds_dir, tx); dmu_objset_sync(ds->ds_objset, zio, tx); } @@ -2189,10 +2224,22 @@ get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) zap_cursor_advance(&zc)) { dsl_dataset_t *clone; char buf[ZFS_MAXNAMELEN]; + /* + * Even though we hold the dp_config_rwlock, the dataset + * may fail to open, returning ENOENT. If there is a + * thread concurrently attempting to destroy this + * dataset, it will have the ds_rwlock held for + * RW_WRITER. Our call to dsl_dataset_hold_obj() -> + * dsl_dataset_hold_ref() will fail its + * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the + * dp_config_rwlock, and wait for the destroy progress + * and signal ds_exclusive_cv. If the destroy was + * successful, we will see that + * DSL_DATASET_IS_DESTROYED(), and return ENOENT. + */ if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool, - za.za_first_integer, FTAG, &clone) != 0) { - goto fail; - } + za.za_first_integer, FTAG, &clone) != 0) + continue; dsl_dir_name(clone->ds_dir, buf); VERIFY(nvlist_add_boolean(val, buf) == 0); dsl_dataset_rele(clone, FTAG); @@ -2316,7 +2363,7 @@ dsl_dataset_space(dsl_dataset_t *ds, uint64_t *refdbytesp, uint64_t *availbytesp, uint64_t *usedobjsp, uint64_t *availobjsp) { - *refdbytesp = ds->ds_phys->ds_used_bytes; + *refdbytesp = ds->ds_phys->ds_referenced_bytes; *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; @@ -2652,7 +2699,7 @@ dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) * Note however, if we stop before we reach the ORIGIN we get: * uN + kN + kN-1 + ... + kM - uM-1 */ - pa->used = origin_ds->ds_phys->ds_used_bytes; + pa->used = origin_ds->ds_phys->ds_referenced_bytes; pa->comp = origin_ds->ds_phys->ds_compressed_bytes; pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; for (snap = list_head(&pa->shared_snaps); snap; @@ -2686,7 +2733,7 @@ dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) * so we need to subtract out the clone origin's used space. */ if (pa->origin_origin) { - pa->used -= pa->origin_origin->ds_phys->ds_used_bytes; + pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes; pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes; pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes; } @@ -3203,8 +3250,8 @@ dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx) dsl_deadlist_space(&csa->ohds->ds_deadlist, &odl_used, &odl_comp, &odl_uncomp); - dused = csa->cds->ds_phys->ds_used_bytes + cdl_used - - (csa->ohds->ds_phys->ds_used_bytes + odl_used); + dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used - + (csa->ohds->ds_phys->ds_referenced_bytes + odl_used); dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp - (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp); duncomp = csa->cds->ds_phys->ds_uncompressed_bytes + @@ -3233,8 +3280,8 @@ dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx) } /* swap ds_*_bytes */ - SWITCH64(csa->ohds->ds_phys->ds_used_bytes, - csa->cds->ds_phys->ds_used_bytes); + SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes, + csa->cds->ds_phys->ds_referenced_bytes); SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes, csa->cds->ds_phys->ds_compressed_bytes); SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes, @@ -3363,8 +3410,9 @@ dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, * on-disk is over quota and there are no pending changes (which * may free up space for us). */ - if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) { - if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota) + if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) { + if (inflight > 0 || + ds->ds_phys->ds_referenced_bytes < ds->ds_quota) error = ERESTART; else error = EDQUOT; @@ -3393,7 +3441,7 @@ dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) if (psa->psa_effective_value == 0) return (0); - if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes || + if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes || psa->psa_effective_value < ds->ds_reserved) return (ENOSPC); @@ -4141,8 +4189,8 @@ dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, dsl_pool_t *dp = new->ds_dir->dd_pool; *usedp = 0; - *usedp += new->ds_phys->ds_used_bytes; - *usedp -= oldsnap->ds_phys->ds_used_bytes; + *usedp += new->ds_phys->ds_referenced_bytes; + *usedp -= oldsnap->ds_phys->ds_referenced_bytes; *compp = 0; *compp += new->ds_phys->ds_compressed_bytes; @@ -4158,9 +4206,13 @@ dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, dsl_dataset_t *snap; uint64_t used, comp, uncomp; - err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); - if (err != 0) - break; + if (snapobj == new->ds_object) { + snap = new; + } else { + err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); + if (err != 0) + break; + } if (snap->ds_phys->ds_prev_snap_txg == oldsnap->ds_phys->ds_creation_txg) { @@ -4189,7 +4241,8 @@ dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, * was not a snapshot of/before new. */ snapobj = snap->ds_phys->ds_prev_snap_obj; - dsl_dataset_rele(snap, FTAG); + if (snap != new) + dsl_dataset_rele(snap, FTAG); if (snapobj == 0) { err = EINVAL; break; diff --git a/module/zfs/dsl_deadlist.c b/module/zfs/dsl_deadlist.c index 1e89a68d7..909b5f8fc 100644 --- a/module/zfs/dsl_deadlist.c +++ b/module/zfs/dsl_deadlist.c @@ -20,7 +20,7 @@ */ /* * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011 by Delphix. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/dsl_dataset.h> @@ -165,12 +165,49 @@ dsl_deadlist_free(objset_t *os, uint64_t dlobj, dmu_tx_t *tx) for (zap_cursor_init(&zc, os, dlobj); zap_cursor_retrieve(&zc, &za) == 0; - zap_cursor_advance(&zc)) - bpobj_free(os, za.za_first_integer, tx); + zap_cursor_advance(&zc)) { + uint64_t obj = za.za_first_integer; + if (obj == dmu_objset_pool(os)->dp_empty_bpobj) + bpobj_decr_empty(os, tx); + else + bpobj_free(os, obj, tx); + } zap_cursor_fini(&zc); VERIFY3U(0, ==, dmu_object_free(os, dlobj, tx)); } +static void +dle_enqueue(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle, + const blkptr_t *bp, dmu_tx_t *tx) +{ + if (dle->dle_bpobj.bpo_object == + dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) { + uint64_t obj = bpobj_alloc(dl->dl_os, SPA_MAXBLOCKSIZE, tx); + bpobj_close(&dle->dle_bpobj); + bpobj_decr_empty(dl->dl_os, tx); + VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj)); + VERIFY3U(0, ==, zap_update_int_key(dl->dl_os, dl->dl_object, + dle->dle_mintxg, obj, tx)); + } + bpobj_enqueue(&dle->dle_bpobj, bp, tx); +} + +static void +dle_enqueue_subobj(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle, + uint64_t obj, dmu_tx_t *tx) +{ + if (dle->dle_bpobj.bpo_object != + dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) { + bpobj_enqueue_subobj(&dle->dle_bpobj, obj, tx); + } else { + bpobj_close(&dle->dle_bpobj); + bpobj_decr_empty(dl->dl_os, tx); + VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj)); + VERIFY3U(0, ==, zap_update_int_key(dl->dl_os, dl->dl_object, + dle->dle_mintxg, obj, tx)); + } +} + void dsl_deadlist_insert(dsl_deadlist_t *dl, const blkptr_t *bp, dmu_tx_t *tx) { @@ -199,7 +236,7 @@ dsl_deadlist_insert(dsl_deadlist_t *dl, const blkptr_t *bp, dmu_tx_t *tx) dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE); else dle = AVL_PREV(&dl->dl_tree, dle); - bpobj_enqueue(&dle->dle_bpobj, bp, tx); + dle_enqueue(dl, dle, bp, tx); } /* @@ -219,7 +256,7 @@ dsl_deadlist_add_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx) dle = kmem_alloc(sizeof (*dle), KM_PUSHPAGE); dle->dle_mintxg = mintxg; - obj = bpobj_alloc(dl->dl_os, SPA_MAXBLOCKSIZE, tx); + obj = bpobj_alloc_empty(dl->dl_os, SPA_MAXBLOCKSIZE, tx); VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj)); avl_add(&dl->dl_tree, dle); @@ -245,8 +282,7 @@ dsl_deadlist_remove_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx) dle = avl_find(&dl->dl_tree, &dle_tofind, NULL); dle_prev = AVL_PREV(&dl->dl_tree, dle); - bpobj_enqueue_subobj(&dle_prev->dle_bpobj, - dle->dle_bpobj.bpo_object, tx); + dle_enqueue_subobj(dl, dle_prev, dle->dle_bpobj.bpo_object, tx); avl_remove(&dl->dl_tree, dle); bpobj_close(&dle->dle_bpobj); @@ -304,7 +340,7 @@ dsl_deadlist_clone(dsl_deadlist_t *dl, uint64_t maxtxg, if (dle->dle_mintxg >= maxtxg) break; - obj = bpobj_alloc(dl->dl_os, SPA_MAXBLOCKSIZE, tx); + obj = bpobj_alloc_empty(dl->dl_os, SPA_MAXBLOCKSIZE, tx); VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, newobj, dle->dle_mintxg, obj, tx)); } @@ -402,7 +438,7 @@ dsl_deadlist_insert_bpobj(dsl_deadlist_t *dl, uint64_t obj, uint64_t birth, dle = avl_find(&dl->dl_tree, &dle_tofind, &where); if (dle == NULL) dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE); - bpobj_enqueue_subobj(&dle->dle_bpobj, obj, tx); + dle_enqueue_subobj(dl, dle, obj, tx); } static int diff --git a/module/zfs/dsl_deleg.c b/module/zfs/dsl_deleg.c index a4d4e42da..294932c45 100644 --- a/module/zfs/dsl_deleg.c +++ b/module/zfs/dsl_deleg.c @@ -20,7 +20,7 @@ */ /* * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011 by Delphix. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ /* @@ -171,10 +171,8 @@ dsl_deleg_set_sync(void *arg1, void *arg2, dmu_tx_t *tx) VERIFY(nvpair_value_nvlist(whopair, &perms) == 0); if (zap_lookup(mos, zapobj, whokey, 8, 1, &jumpobj) != 0) { - jumpobj = zap_create(mos, DMU_OT_DSL_PERMS, - DMU_OT_NONE, 0, tx); - VERIFY(zap_update(mos, zapobj, - whokey, 8, 1, &jumpobj, tx) == 0); + jumpobj = zap_create_link(mos, DMU_OT_DSL_PERMS, + zapobj, whokey, tx); } while ((permpair = nvlist_next_nvpair(perms, permpair))) { diff --git a/module/zfs/dsl_dir.c b/module/zfs/dsl_dir.c index 377df40a8..741223984 100644 --- a/module/zfs/dsl_dir.c +++ b/module/zfs/dsl_dir.c @@ -189,7 +189,6 @@ errout: kmem_free(dd, sizeof (dsl_dir_t)); dmu_buf_rele(dbuf, tag); return (err); - } void @@ -223,7 +222,7 @@ dsl_dir_name(dsl_dir_t *dd, char *buf) } } -/* Calculate name legnth, avoiding all the strcat calls of dsl_dir_name */ +/* Calculate name length, avoiding all the strcat calls of dsl_dir_name */ int dsl_dir_namelen(dsl_dir_t *dd) { @@ -592,8 +591,6 @@ dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx) { ASSERT(dmu_tx_is_syncing(tx)); - dmu_buf_will_dirty(dd->dd_dbuf, tx); - mutex_enter(&dd->dd_lock); ASSERT3U(dd->dd_tempreserved[tx->tx_txg&TXG_MASK], ==, 0); dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, @@ -950,8 +947,6 @@ dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, ASSERT(dmu_tx_is_syncing(tx)); ASSERT(type < DD_USED_NUM); - dsl_dir_dirty(dd, tx); - if (needlock) mutex_enter(&dd->dd_lock); accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used); @@ -960,6 +955,7 @@ dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, dd->dd_phys->dd_compressed_bytes >= -compressed); ASSERT(uncompressed >= 0 || dd->dd_phys->dd_uncompressed_bytes >= -uncompressed); + dmu_buf_will_dirty(dd->dd_dbuf, tx); dd->dd_phys->dd_used_bytes += used; dd->dd_phys->dd_uncompressed_bytes += uncompressed; dd->dd_phys->dd_compressed_bytes += compressed; @@ -1003,13 +999,13 @@ dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN)) return; - dsl_dir_dirty(dd, tx); if (needlock) mutex_enter(&dd->dd_lock); ASSERT(delta > 0 ? dd->dd_phys->dd_used_breakdown[oldtype] >= delta : dd->dd_phys->dd_used_breakdown[newtype] >= -delta); ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta)); + dmu_buf_will_dirty(dd->dd_dbuf, tx); dd->dd_phys->dd_used_breakdown[oldtype] -= delta; dd->dd_phys->dd_used_breakdown[newtype] += delta; if (needlock) diff --git a/module/zfs/dsl_pool.c b/module/zfs/dsl_pool.c index 7e0fba589..704f034e9 100644 --- a/module/zfs/dsl_pool.c +++ b/module/zfs/dsl_pool.c @@ -20,7 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011 by Delphix. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/dsl_pool.h> @@ -40,6 +40,9 @@ #include <sys/zfs_znode.h> #include <sys/spa_impl.h> #include <sys/dsl_deadlist.h> +#include <sys/bptree.h> +#include <sys/zfeature.h> +#include <sys/zil_impl.h> int zfs_no_write_throttle = 0; int zfs_write_limit_shift = 3; /* 1/8th of physical memory */ @@ -222,12 +225,12 @@ dsl_pool_open_impl(spa_t *spa, uint64_t txg) txg_list_create(&dp->dp_dirty_datasets, offsetof(dsl_dataset_t, ds_dirty_link)); + txg_list_create(&dp->dp_dirty_zilogs, + offsetof(zilog_t, zl_dirty_link)); txg_list_create(&dp->dp_dirty_dirs, offsetof(dsl_dir_t, dd_dirty_link)); txg_list_create(&dp->dp_sync_tasks, offsetof(dsl_sync_task_group_t, dstg_node)); - list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t), - offsetof(dsl_dataset_t, ds_synced_link)); mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); @@ -240,20 +243,30 @@ dsl_pool_open_impl(spa_t *spa, uint64_t txg) } int -dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) +dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) { int err; dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); + + err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, + &dp->dp_meta_objset); + if (err != 0) + dsl_pool_close(dp); + else + *dpp = dp; + + return (err); +} + +int +dsl_pool_open(dsl_pool_t *dp) +{ + int err; dsl_dir_t *dd; dsl_dataset_t *ds; uint64_t obj; rw_enter(&dp->dp_config_rwlock, RW_WRITER); - err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, - &dp->dp_meta_objset); - if (err) - goto out; - err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &dp->dp_root_dir_obj); @@ -269,7 +282,7 @@ dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) if (err) goto out; - if (spa_version(spa) >= SPA_VERSION_ORIGIN) { + if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); if (err) goto out; @@ -286,7 +299,7 @@ dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) goto out; } - if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { + if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, &dp->dp_free_dir); if (err) @@ -300,6 +313,24 @@ dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) dp->dp_meta_objset, obj)); } + if (spa_feature_is_active(dp->dp_spa, + &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) { + err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, + &dp->dp_bptree_obj); + if (err != 0) + goto out; + } + + if (spa_feature_is_active(dp->dp_spa, + &spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ])) { + err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, + &dp->dp_empty_bpobj); + if (err != 0) + goto out; + } + err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, &dp->dp_tmp_userrefs_obj); @@ -308,15 +339,10 @@ dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) if (err) goto out; - err = dsl_scan_init(dp, txg); + err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); out: rw_exit(&dp->dp_config_rwlock); - if (err) - dsl_pool_close(dp); - else - *dpp = dp; - return (err); } @@ -346,9 +372,9 @@ dsl_pool_close(dsl_pool_t *dp) dmu_objset_evict(dp->dp_meta_objset); txg_list_destroy(&dp->dp_dirty_datasets); + txg_list_destroy(&dp->dp_dirty_zilogs); txg_list_destroy(&dp->dp_sync_tasks); txg_list_destroy(&dp->dp_dirty_dirs); - list_destroy(&dp->dp_synced_datasets); arc_flush(dp->dp_spa); txg_fini(dp); @@ -429,6 +455,21 @@ dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) return (dp); } +/* + * Account for the meta-objset space in its placeholder dsl_dir. + */ +void +dsl_pool_mos_diduse_space(dsl_pool_t *dp, + int64_t used, int64_t comp, int64_t uncomp) +{ + ASSERT3U(comp, ==, uncomp); /* it's all metadata */ + mutex_enter(&dp->dp_lock); + dp->dp_mos_used_delta += used; + dp->dp_mos_compressed_delta += comp; + dp->dp_mos_uncompressed_delta += uncomp; + mutex_exit(&dp->dp_lock); +} + static int deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { @@ -447,11 +488,14 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) dmu_tx_t *tx; dsl_dir_t *dd; dsl_dataset_t *ds; - dsl_sync_task_group_t *dstg; objset_t *mos = dp->dp_meta_objset; hrtime_t start, write_time; uint64_t data_written; int err; + list_t synced_datasets; + + list_create(&synced_datasets, sizeof (dsl_dataset_t), + offsetof(dsl_dataset_t, ds_synced_link)); /* * We need to copy dp_space_towrite() before doing @@ -474,7 +518,7 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) * may sync newly-created datasets on pass 2. */ ASSERT(!list_link_active(&ds->ds_synced_link)); - list_insert_tail(&dp->dp_synced_datasets, ds); + list_insert_tail(&synced_datasets, ds); dsl_dataset_sync(ds, zio, tx); } DTRACE_PROBE(pool_sync__1setup); @@ -484,15 +528,20 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) ASSERT(err == 0); DTRACE_PROBE(pool_sync__2rootzio); - for (ds = list_head(&dp->dp_synced_datasets); ds; - ds = list_next(&dp->dp_synced_datasets, ds)) + /* + * After the data blocks have been written (ensured by the zio_wait() + * above), update the user/group space accounting. + */ + for (ds = list_head(&synced_datasets); ds; + ds = list_next(&synced_datasets, ds)) dmu_objset_do_userquota_updates(ds->ds_objset, tx); /* * Sync the datasets again to push out the changes due to * userspace updates. This must be done before we process the - * sync tasks, because that could cause a snapshot of a dataset - * whose ds_bp will be rewritten when we do this 2nd sync. + * sync tasks, so that any snapshots will have the correct + * user accounting information (and we won't get confused + * about which blocks are part of the snapshot). */ zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg))) { @@ -503,30 +552,42 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) err = zio_wait(zio); /* - * Move dead blocks from the pending deadlist to the on-disk - * deadlist. + * Now that the datasets have been completely synced, we can + * clean up our in-memory structures accumulated while syncing: + * + * - move dead blocks from the pending deadlist to the on-disk deadlist + * - clean up zil records + * - release hold from dsl_dataset_dirty() */ - for (ds = list_head(&dp->dp_synced_datasets); ds; - ds = list_next(&dp->dp_synced_datasets, ds)) { + while ((ds = list_remove_head(&synced_datasets))) { + ASSERTV(objset_t *os = ds->ds_objset); bplist_iterate(&ds->ds_pending_deadlist, deadlist_enqueue_cb, &ds->ds_deadlist, tx); + ASSERT(!dmu_objset_is_dirty(os, txg)); + dmu_buf_rele(ds->ds_dbuf, ds); } - while ((dstg = txg_list_remove(&dp->dp_sync_tasks, txg))) { - /* - * No more sync tasks should have been added while we - * were syncing. - */ - ASSERT(spa_sync_pass(dp->dp_spa) == 1); - dsl_sync_task_group_sync(dstg, tx); - } - DTRACE_PROBE(pool_sync__3task); - start = gethrtime(); while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg))) dsl_dir_sync(dd, tx); write_time += gethrtime() - start; + /* + * The MOS's space is accounted for in the pool/$MOS + * (dp_mos_dir). We can't modify the mos while we're syncing + * it, so we remember the deltas and apply them here. + */ + if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || + dp->dp_mos_uncompressed_delta != 0) { + dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, + dp->dp_mos_used_delta, + dp->dp_mos_compressed_delta, + dp->dp_mos_uncompressed_delta, tx); + dp->dp_mos_used_delta = 0; + dp->dp_mos_compressed_delta = 0; + dp->dp_mos_uncompressed_delta = 0; + } + start = gethrtime(); if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { @@ -542,6 +603,27 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) hrtime_t, dp->dp_read_overhead); write_time -= dp->dp_read_overhead; + /* + * If we modify a dataset in the same txg that we want to destroy it, + * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. + * dsl_dir_destroy_check() will fail if there are unexpected holds. + * Therefore, we want to sync the MOS (thus syncing the dd_dbuf + * and clearing the hold on it) before we process the sync_tasks. + * The MOS data dirtied by the sync_tasks will be synced on the next + * pass. + */ + DTRACE_PROBE(pool_sync__3task); + if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { + dsl_sync_task_group_t *dstg; + /* + * No more sync tasks should have been added while we + * were syncing. + */ + ASSERT(spa_sync_pass(dp->dp_spa) == 1); + while ((dstg = txg_list_remove(&dp->dp_sync_tasks, txg))) + dsl_sync_task_group_sync(dstg, tx); + } + dmu_tx_commit(tx); dp->dp_space_towrite[txg & TXG_MASK] = 0; @@ -590,15 +672,14 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) void dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) { + zilog_t *zilog; dsl_dataset_t *ds; - objset_t *os; - while ((ds = list_head(&dp->dp_synced_datasets))) { - list_remove(&dp->dp_synced_datasets, ds); - os = ds->ds_objset; - zil_clean(os->os_zil, txg); - ASSERT(!dmu_objset_is_dirty(os, txg)); - dmu_buf_rele(ds->ds_dbuf, ds); + while ((zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg))) { + ds = dmu_objset_ds(zilog->zl_os); + zil_clean(zilog, txg); + ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); + dmu_buf_rele(ds->ds_dbuf, zilog); } ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); } @@ -611,7 +692,7 @@ int dsl_pool_sync_context(dsl_pool_t *dp) { return (curthread == dp->dp_tx.tx_sync_thread || - spa_get_dsl(dp->dp_spa) == NULL); + spa_is_initializing(dp->dp_spa)); } uint64_t @@ -932,11 +1013,8 @@ dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) ASSERT(dp->dp_tmp_userrefs_obj == 0); ASSERT(dmu_tx_is_syncing(tx)); - dp->dp_tmp_userrefs_obj = zap_create(mos, DMU_OT_USERREFS, - DMU_OT_NONE, 0, tx); - - VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, - sizeof (uint64_t), 1, &dp->dp_tmp_userrefs_obj, tx) == 0); + dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, + DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); } static int diff --git a/module/zfs/dsl_scan.c b/module/zfs/dsl_scan.c index c2386dd67..297caa067 100644 --- a/module/zfs/dsl_scan.c +++ b/module/zfs/dsl_scan.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/dsl_scan.h> @@ -44,6 +45,7 @@ #include <sys/ddt.h> #include <sys/sa.h> #include <sys/sa_impl.h> +#include <sys/zfeature.h> #ifdef _KERNEL #include <sys/zfs_vfsops.h> #endif @@ -379,55 +381,6 @@ dsl_read_nolock(zio_t *pio, spa_t *spa, const blkptr_t *bpp, priority, zio_flags, arc_flags, zb)); } -static boolean_t -bookmark_is_zero(const zbookmark_t *zb) -{ - return (zb->zb_objset == 0 && zb->zb_object == 0 && - zb->zb_level == 0 && zb->zb_blkid == 0); -} - -/* dnp is the dnode for zb1->zb_object */ -static boolean_t -bookmark_is_before(const dnode_phys_t *dnp, const zbookmark_t *zb1, - const zbookmark_t *zb2) -{ - uint64_t zb1nextL0, zb2thisobj; - - ASSERT(zb1->zb_objset == zb2->zb_objset); - ASSERT(zb2->zb_level == 0); - - /* - * A bookmark in the deadlist is considered to be after - * everything else. - */ - if (zb2->zb_object == DMU_DEADLIST_OBJECT) - return (B_TRUE); - - /* The objset_phys_t isn't before anything. */ - if (dnp == NULL) - return (B_FALSE); - - zb1nextL0 = (zb1->zb_blkid + 1) << - ((zb1->zb_level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)); - - zb2thisobj = zb2->zb_object ? zb2->zb_object : - zb2->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT); - - if (zb1->zb_object == DMU_META_DNODE_OBJECT) { - uint64_t nextobj = zb1nextL0 * - (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT) >> DNODE_SHIFT; - return (nextobj <= zb2thisobj); - } - - if (zb1->zb_object < zb2thisobj) - return (B_TRUE); - if (zb1->zb_object > zb2thisobj) - return (B_FALSE); - if (zb2->zb_object == DMU_META_DNODE_OBJECT) - return (B_FALSE); - return (zb1nextL0 <= zb2->zb_blkid); -} - static uint64_t dsl_scan_ds_maxtxg(dsl_dataset_t *ds) { @@ -459,7 +412,7 @@ dsl_scan_check_pause(dsl_scan_t *scn, const zbookmark_t *zb) if (scn->scn_pausing) return (B_TRUE); /* we're already pausing */ - if (!bookmark_is_zero(&scn->scn_phys.scn_bookmark)) + if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark)) return (B_FALSE); /* we're resuming */ /* We only know how to resume from level-0 blocks. */ @@ -614,13 +567,13 @@ dsl_scan_check_resume(dsl_scan_t *scn, const dnode_phys_t *dnp, /* * We never skip over user/group accounting objects (obj<0) */ - if (!bookmark_is_zero(&scn->scn_phys.scn_bookmark) && + 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 (bookmark_is_before(dnp, zb, &scn->scn_phys.scn_bookmark)) + if (zbookmark_is_before(dnp, zb, &scn->scn_phys.scn_bookmark)) return (B_TRUE); /* @@ -823,22 +776,6 @@ dsl_scan_visitbp(blkptr_t *bp, const zbookmark_t *zb, if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) goto out; - if (BP_GET_TYPE(bp) != DMU_OT_USERGROUP_USED) { - /* - * For non-user-accounting blocks, we need to read the - * new bp (from a deleted snapshot, found in - * check_existing_xlation). If we used the old bp, - * pointers inside this block from before we resumed - * would be untranslated. - * - * For user-accounting blocks, we need to read the old - * bp, because we will apply the entire space delta to - * it (original untranslated -> translations from - * deleted snap -> now). - */ - *bp_toread = *bp; - } - if (dsl_scan_recurse(scn, ds, ostype, dnp, bp_toread, zb, tx, &buf) != 0) goto out; @@ -1414,19 +1351,28 @@ out: kmem_free(zc, sizeof(zap_cursor_t)); } -static int -dsl_scan_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) +static boolean_t +dsl_scan_free_should_pause(dsl_scan_t *scn) { - dsl_scan_t *scn = arg; uint64_t elapsed_nanosecs; elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; - - if (elapsed_nanosecs / NANOSEC > zfs_txg_timeout || + return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout || (elapsed_nanosecs / MICROSEC > zfs_free_min_time_ms && txg_sync_waiting(scn->scn_dp)) || - spa_shutting_down(scn->scn_dp->dp_spa)) - return (ERESTART); + 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_free_should_pause(scn)) + return (ERESTART); + } zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa, dmu_tx_get_txg(tx), bp, 0)); @@ -1451,6 +1397,10 @@ dsl_scan_active(dsl_scan_t *scn) if (scn->scn_phys.scn_state == DSS_SCANNING) return (B_TRUE); + if (spa_feature_is_active(spa, + &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) { + 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); @@ -1497,14 +1447,40 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) * traversing it. */ if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { + scn->scn_is_bptree = B_FALSE; scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); err = bpobj_iterate(&dp->dp_free_bpobj, - dsl_scan_free_cb, scn, tx); + dsl_scan_free_block_cb, scn, tx); VERIFY3U(0, ==, zio_wait(scn->scn_zio_root)); + + if (err == 0 && spa_feature_is_active(spa, + &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) { + scn->scn_is_bptree = B_TRUE; + scn->scn_zio_root = zio_root(dp->dp_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); + VERIFY3U(0, ==, zio_wait(scn->scn_zio_root)); + if (err != 0) + return; + + /* disable async destroy feature */ + spa_feature_decr(spa, + &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY], tx); + ASSERT(!spa_feature_is_active(spa, + &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])); + VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset, + DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_BPTREE_OBJ, tx)); + VERIFY3U(0, ==, bptree_free(dp->dp_meta_objset, + dp->dp_bptree_obj, tx)); + dp->dp_bptree_obj = 0; + } if (scn->scn_visited_this_txg) { zfs_dbgmsg("freed %llu blocks in %llums from " - "free_bpobj txg %llu", + "free_bpobj/bptree txg %llu", (longlong_t)scn->scn_visited_this_txg, (longlong_t) (gethrtime() - scn->scn_sync_start_time) / MICROSEC, @@ -1619,9 +1595,13 @@ count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp) for (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; - zfs_blkstat_t *zb = &zab->zab_type[l][t]; int equal; + zfs_blkstat_t *zb; + + if (t & DMU_OT_NEWTYPE) + t = DMU_OT_OTHER; + zb = &zab->zab_type[l][t]; zb->zb_count++; zb->zb_asize += BP_GET_ASIZE(bp); zb->zb_lsize += BP_GET_LSIZE(bp); diff --git a/module/zfs/sa.c b/module/zfs/sa.c index d4b28cc90..240a683d6 100644 --- a/module/zfs/sa.c +++ b/module/zfs/sa.c @@ -18,8 +18,10 @@ * * CDDL HEADER END */ + /* * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/zfs_context.h> @@ -446,10 +448,9 @@ sa_add_layout_entry(objset_t *os, sa_attr_type_t *attrs, int attr_count, char attr_name[8]; if (sa->sa_layout_attr_obj == 0) { - sa->sa_layout_attr_obj = zap_create(os, - DMU_OT_SA_ATTR_LAYOUTS, DMU_OT_NONE, 0, tx); - VERIFY(zap_add(os, sa->sa_master_obj, SA_LAYOUTS, 8, 1, - &sa->sa_layout_attr_obj, tx) == 0); + sa->sa_layout_attr_obj = zap_create_link(os, + DMU_OT_SA_ATTR_LAYOUTS, + sa->sa_master_obj, SA_LAYOUTS, tx); } (void) snprintf(attr_name, sizeof (attr_name), @@ -1583,10 +1584,9 @@ sa_attr_register_sync(sa_handle_t *hdl, dmu_tx_t *tx) } if (sa->sa_reg_attr_obj == 0) { - sa->sa_reg_attr_obj = zap_create(hdl->sa_os, - DMU_OT_SA_ATTR_REGISTRATION, DMU_OT_NONE, 0, tx); - VERIFY(zap_add(hdl->sa_os, sa->sa_master_obj, - SA_REGISTRY, 8, 1, &sa->sa_reg_attr_obj, tx) == 0); + sa->sa_reg_attr_obj = zap_create_link(hdl->sa_os, + DMU_OT_SA_ATTR_REGISTRATION, + sa->sa_master_obj, SA_REGISTRY, tx); } for (i = 0; i != sa->sa_num_attrs; i++) { if (sa->sa_attr_table[i].sa_registered) diff --git a/module/zfs/spa.c b/module/zfs/spa.c index b610a0dae..5b6465f2e 100644 --- a/module/zfs/spa.c +++ b/module/zfs/spa.c @@ -63,6 +63,7 @@ #include <sys/spa_boot.h> #include <sys/zfs_ioctl.h> #include <sys/dsl_scan.h> +#include <sys/zfeature.h> #ifdef _KERNEL #include <sys/bootprops.h> @@ -114,7 +115,10 @@ const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, }; +static dsl_syncfunc_t spa_sync_version; static dsl_syncfunc_t spa_sync_props; +static dsl_checkfunc_t spa_change_guid_check; +static dsl_syncfunc_t spa_change_guid_sync; static boolean_t spa_has_active_shared_spare(spa_t *spa); static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, @@ -169,6 +173,7 @@ static void spa_prop_get_config(spa_t *spa, nvlist_t **nvp) { vdev_t *rvd = spa->spa_root_vdev; + dsl_pool_t *pool = spa->spa_dsl_pool; uint64_t size; uint64_t alloc; uint64_t space; @@ -216,6 +221,22 @@ spa_prop_get_config(spa_t *spa, nvlist_t **nvp) spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); } + if (pool != NULL) { + dsl_dir_t *freedir = pool->dp_free_dir; + + /* + * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, + * when opening pools before this version freedir will be NULL. + */ + if (freedir != NULL) { + spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, + freedir->dd_phys->dd_used_bytes, src); + } else { + spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, + NULL, 0, src); + } + } + spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); if (spa->spa_comment != NULL) { @@ -357,25 +378,55 @@ spa_prop_validate(spa_t *spa, nvlist_t *props) nvpair_t *elem; int error = 0, reset_bootfs = 0; uint64_t objnum = 0; + boolean_t has_feature = B_FALSE; elem = NULL; while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { - zpool_prop_t prop; - char *propname, *strval; uint64_t intval; - objset_t *os; - char *slash, *check; + char *strval, *slash, *check, *fname; + const char *propname = nvpair_name(elem); + zpool_prop_t prop = zpool_name_to_prop(propname); + + switch ((int)prop) { + case ZPROP_INVAL: + if (!zpool_prop_feature(propname)) { + error = EINVAL; + break; + } + + /* + * Sanitize the input. + */ + if (nvpair_type(elem) != DATA_TYPE_UINT64) { + error = EINVAL; + break; + } + + if (nvpair_value_uint64(elem, &intval) != 0) { + error = EINVAL; + break; + } - propname = nvpair_name(elem); + if (intval != 0) { + error = EINVAL; + break; + } + + fname = strchr(propname, '@') + 1; + if (zfeature_lookup_name(fname, NULL) != 0) { + error = EINVAL; + break; + } - if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL) - return (EINVAL); + has_feature = B_TRUE; + break; - switch (prop) { case ZPOOL_PROP_VERSION: error = nvpair_value_uint64(elem, &intval); if (!error && - (intval < spa_version(spa) || intval > SPA_VERSION)) + (intval < spa_version(spa) || + intval > SPA_VERSION_BEFORE_FEATURES || + has_feature)) error = EINVAL; break; @@ -412,6 +463,7 @@ spa_prop_validate(spa_t *spa, nvlist_t *props) error = nvpair_value_string(elem, &strval); if (!error) { + objset_t *os; uint64_t compress; if (strval == NULL || strval[0] == '\0') { @@ -558,33 +610,58 @@ int spa_prop_set(spa_t *spa, nvlist_t *nvp) { int error; - nvpair_t *elem; + nvpair_t *elem = NULL; boolean_t need_sync = B_FALSE; - zpool_prop_t prop; if ((error = spa_prop_validate(spa, nvp)) != 0) return (error); - elem = NULL; while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { - if ((prop = zpool_name_to_prop( - nvpair_name(elem))) == ZPROP_INVAL) - return (EINVAL); + zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT || prop == ZPOOL_PROP_READONLY) continue; + if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { + uint64_t ver; + + if (prop == ZPOOL_PROP_VERSION) { + VERIFY(nvpair_value_uint64(elem, &ver) == 0); + } else { + ASSERT(zpool_prop_feature(nvpair_name(elem))); + ver = SPA_VERSION_FEATURES; + need_sync = B_TRUE; + } + + /* Save time if the version is already set. */ + if (ver == spa_version(spa)) + continue; + + /* + * In addition to the pool directory object, we might + * create the pool properties object, the features for + * read object, the features for write object, or the + * feature descriptions object. + */ + error = dsl_sync_task_do(spa_get_dsl(spa), NULL, + spa_sync_version, spa, &ver, 6); + if (error) + return (error); + continue; + } + need_sync = B_TRUE; break; } - if (need_sync) + if (need_sync) { return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props, - spa, nvp, 3)); - else - return (0); + spa, nvp, 6)); + } + + return (0); } /* @@ -601,6 +678,47 @@ spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) } } +/*ARGSUSED*/ +static int +spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx) +{ + spa_t *spa = arg1; + vdev_t *rvd = spa->spa_root_vdev; + uint64_t vdev_state; + ASSERTV(uint64_t *newguid = arg2); + + spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); + vdev_state = rvd->vdev_state; + spa_config_exit(spa, SCL_STATE, FTAG); + + if (vdev_state != VDEV_STATE_HEALTHY) + return (ENXIO); + + ASSERT3U(spa_guid(spa), !=, *newguid); + + return (0); +} + +static void +spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx) +{ + spa_t *spa = arg1; + uint64_t *newguid = arg2; + uint64_t oldguid; + vdev_t *rvd = spa->spa_root_vdev; + + oldguid = spa_guid(spa); + + spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); + rvd->vdev_guid = *newguid; + rvd->vdev_guid_sum += (*newguid - oldguid); + vdev_config_dirty(rvd); + spa_config_exit(spa, SCL_STATE, FTAG); + + spa_history_log_internal(LOG_POOL_GUID_CHANGE, spa, tx, + "old=%lld new=%lld", oldguid, *newguid); +} + /* * Change the GUID for the pool. This is done so that we can later * re-import a pool built from a clone of our own vdevs. We will modify @@ -613,29 +731,23 @@ spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) int spa_change_guid(spa_t *spa) { - uint64_t oldguid, newguid; - uint64_t txg; - - if (!(spa_mode_global & FWRITE)) - return (EROFS); - - txg = spa_vdev_enter(spa); - - if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY) - return (spa_vdev_exit(spa, NULL, txg, ENXIO)); + int error; + uint64_t guid; - oldguid = spa_guid(spa); - newguid = spa_generate_guid(NULL); - ASSERT3U(oldguid, !=, newguid); + mutex_enter(&spa_namespace_lock); + guid = spa_generate_guid(NULL); - spa->spa_root_vdev->vdev_guid = newguid; - spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid); + error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check, + spa_change_guid_sync, spa, &guid, 5); - vdev_config_dirty(spa->spa_root_vdev); + if (error == 0) { + spa_config_sync(spa, B_FALSE, B_TRUE); + spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_REGUID); + } - spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_REGUID); + mutex_exit(&spa_namespace_lock); - return (spa_vdev_exit(spa, NULL, txg, 0)); + return (error); } /* @@ -1628,7 +1740,7 @@ spa_load_verify_done(zio_t *zio) int error = zio->io_error; if (error) { - if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) && + if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && type != DMU_OT_INTENT_LOG) atomic_add_64(&sle->sle_meta_count, 1); else @@ -1858,6 +1970,9 @@ spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, KM_PUSHPAGE) == 0); } + nvlist_free(spa->spa_load_info); + spa->spa_load_info = fnvlist_alloc(); + gethrestime(&spa->spa_loaded_ts); error = spa_load_impl(spa, pool_guid, config, state, type, mosconfig, &ereport); @@ -1891,12 +2006,14 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, { int error = 0; nvlist_t *nvroot = NULL; + nvlist_t *label; vdev_t *rvd; uberblock_t *ub = &spa->spa_uberblock; uint64_t children, config_cache_txg = spa->spa_config_txg; int orig_mode = spa->spa_mode; int parse; uint64_t obj; + boolean_t missing_feat_write = B_FALSE; /* * If this is an untrusted config, access the pool in read-only mode. @@ -1976,19 +2093,79 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, /* * Find the best uberblock. */ - vdev_uberblock_load(NULL, rvd, ub); + vdev_uberblock_load(rvd, ub, &label); /* * If we weren't able to find a single valid uberblock, return failure. */ - if (ub->ub_txg == 0) + if (ub->ub_txg == 0) { + nvlist_free(label); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); + } /* - * If the pool is newer than the code, we can't open it. + * If the pool has an unsupported version we can't open it. */ - if (ub->ub_version > SPA_VERSION) + if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { + nvlist_free(label); return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); + } + + if (ub->ub_version >= SPA_VERSION_FEATURES) { + nvlist_t *features; + + /* + * If we weren't able to find what's necessary for reading the + * MOS in the label, return failure. + */ + if (label == NULL || nvlist_lookup_nvlist(label, + ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { + nvlist_free(label); + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, + ENXIO)); + } + + /* + * Update our in-core representation with the definitive values + * from the label. + */ + nvlist_free(spa->spa_label_features); + VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); + } + + nvlist_free(label); + + /* + * Look through entries in the label nvlist's features_for_read. If + * there is a feature listed there which we don't understand then we + * cannot open a pool. + */ + if (ub->ub_version >= SPA_VERSION_FEATURES) { + nvlist_t *unsup_feat; + nvpair_t *nvp; + + VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == + 0); + + for (nvp = nvlist_next_nvpair(spa->spa_label_features, NULL); + nvp != NULL; + nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { + if (!zfeature_is_supported(nvpair_name(nvp))) { + VERIFY(nvlist_add_string(unsup_feat, + nvpair_name(nvp), "") == 0); + } + } + + if (!nvlist_empty(unsup_feat)) { + VERIFY(nvlist_add_nvlist(spa->spa_load_info, + ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); + nvlist_free(unsup_feat); + return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, + ENOTSUP)); + } + + nvlist_free(unsup_feat); + } /* * If the vdev guid sum doesn't match the uberblock, we have an @@ -2022,7 +2199,7 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, spa->spa_claim_max_txg = spa->spa_first_txg; spa->spa_prev_software_version = ub->ub_software_version; - error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool); + error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); if (error) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; @@ -2030,6 +2207,89 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); + if (spa_version(spa) >= SPA_VERSION_FEATURES) { + boolean_t missing_feat_read = B_FALSE; + nvlist_t *unsup_feat, *enabled_feat; + + if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, + &spa->spa_feat_for_read_obj) != 0) { + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); + } + + if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, + &spa->spa_feat_for_write_obj) != 0) { + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); + } + + if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, + &spa->spa_feat_desc_obj) != 0) { + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); + } + + enabled_feat = fnvlist_alloc(); + unsup_feat = fnvlist_alloc(); + + if (!feature_is_supported(spa->spa_meta_objset, + spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj, + unsup_feat, enabled_feat)) + missing_feat_read = B_TRUE; + + if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { + if (!feature_is_supported(spa->spa_meta_objset, + spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj, + unsup_feat, enabled_feat)) { + missing_feat_write = B_TRUE; + } + } + + fnvlist_add_nvlist(spa->spa_load_info, + ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); + + if (!nvlist_empty(unsup_feat)) { + fnvlist_add_nvlist(spa->spa_load_info, + ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); + } + + fnvlist_free(enabled_feat); + fnvlist_free(unsup_feat); + + if (!missing_feat_read) { + fnvlist_add_boolean(spa->spa_load_info, + ZPOOL_CONFIG_CAN_RDONLY); + } + + /* + * If the state is SPA_LOAD_TRYIMPORT, our objective is + * twofold: to determine whether the pool is available for + * import in read-write mode and (if it is not) whether the + * pool is available for import in read-only mode. If the pool + * is available for import in read-write mode, it is displayed + * as available in userland; if it is not available for import + * in read-only mode, it is displayed as unavailable in + * userland. If the pool is available for import in read-only + * mode but not read-write mode, it is displayed as unavailable + * in userland with a special note that the pool is actually + * available for open in read-only mode. + * + * As a result, if the state is SPA_LOAD_TRYIMPORT and we are + * missing a feature for write, we must first determine whether + * the pool can be opened read-only before returning to + * userland in order to know whether to display the + * abovementioned note. + */ + if (missing_feat_read || (missing_feat_write && + spa_writeable(spa))) { + return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, + ENOTSUP)); + } + } + + spa->spa_is_initializing = B_TRUE; + error = dsl_pool_open(spa->spa_dsl_pool); + spa->spa_is_initializing = B_FALSE; + if (error != 0) + return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); + if (!mosconfig) { uint64_t hostid; nvlist_t *policy = NULL, *nvconfig; @@ -2247,7 +2507,7 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, nvlist_free(nvconfig); /* - * Now that we've validate the config, check the state of the + * Now that we've validated the config, check the state of the * root vdev. If it can't be opened, it indicates one or * more toplevel vdevs are faulted. */ @@ -2260,6 +2520,17 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, } } + if (missing_feat_write) { + ASSERT(state == SPA_LOAD_TRYIMPORT); + + /* + * At this point, we know that we can open the pool in + * read-only mode but not read-write mode. We now have enough + * information and can return to userland. + */ + return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); + } + /* * We've successfully opened the pool, verify that we're ready * to start pushing transactions. @@ -2370,10 +2641,18 @@ spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); } +/* + * If spa_load() fails this function will try loading prior txg's. If + * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool + * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this + * function will not rewind the pool and will return the same error as + * spa_load(). + */ static int spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, uint64_t max_request, int rewind_flags) { + nvlist_t *loadinfo = NULL; nvlist_t *config = NULL; int load_error, rewind_error; uint64_t safe_rewind_txg; @@ -2402,9 +2681,18 @@ spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, return (load_error); } - /* Price of rolling back is discarding txgs, including log */ - if (state == SPA_LOAD_RECOVER) + if (state == SPA_LOAD_RECOVER) { + /* Price of rolling back is discarding txgs, including log */ spa_set_log_state(spa, SPA_LOG_CLEAR); + } else { + /* + * If we aren't rolling back save the load info from our first + * import attempt so that we can restore it after attempting + * to rewind. + */ + loadinfo = spa->spa_load_info; + spa->spa_load_info = fnvlist_alloc(); + } spa->spa_load_max_txg = spa->spa_last_ubsync_txg; safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; @@ -2428,7 +2716,20 @@ spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, if (config && (rewind_error || state != SPA_LOAD_RECOVER)) spa_config_set(spa, config); - return (state == SPA_LOAD_RECOVER ? rewind_error : load_error); + if (state == SPA_LOAD_RECOVER) { + ASSERT3P(loadinfo, ==, NULL); + return (rewind_error); + } else { + /* Store the rewind info as part of the initial load info */ + fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, + spa->spa_load_info); + + /* Restore the initial load info */ + fnvlist_free(spa->spa_load_info); + spa->spa_load_info = loadinfo; + + return (load_error); + } } /* @@ -2698,8 +2999,50 @@ spa_add_l2cache(spa_t *spa, nvlist_t *config) } } +static void +spa_add_feature_stats(spa_t *spa, nvlist_t *config) +{ + nvlist_t *features; + zap_cursor_t zc; + zap_attribute_t za; + + ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); + VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); + + if (spa->spa_feat_for_read_obj != 0) { + for (zap_cursor_init(&zc, spa->spa_meta_objset, + spa->spa_feat_for_read_obj); + zap_cursor_retrieve(&zc, &za) == 0; + zap_cursor_advance(&zc)) { + ASSERT(za.za_integer_length == sizeof (uint64_t) && + za.za_num_integers == 1); + VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, + za.za_first_integer)); + } + zap_cursor_fini(&zc); + } + + if (spa->spa_feat_for_write_obj != 0) { + for (zap_cursor_init(&zc, spa->spa_meta_objset, + spa->spa_feat_for_write_obj); + zap_cursor_retrieve(&zc, &za) == 0; + zap_cursor_advance(&zc)) { + ASSERT(za.za_integer_length == sizeof (uint64_t) && + za.za_num_integers == 1); + VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, + za.za_first_integer)); + } + zap_cursor_fini(&zc); + } + + VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, + features) == 0); + nvlist_free(features); +} + int -spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) +spa_get_stats(const char *name, nvlist_t **config, + char *altroot, size_t buflen) { int error; spa_t *spa; @@ -2734,6 +3077,7 @@ spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) spa_add_spares(spa, *config); spa_add_l2cache(spa, *config); + spa_add_feature_stats(spa, *config); } } @@ -2954,6 +3298,8 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, nvlist_t **spares, **l2cache; uint_t nspares, nl2cache; uint64_t version, obj; + boolean_t has_features; + nvpair_t *elem; int c; /* @@ -2980,10 +3326,18 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, return (error); } - if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION), - &version) != 0) + has_features = B_FALSE; + for (elem = nvlist_next_nvpair(props, NULL); + elem != NULL; elem = nvlist_next_nvpair(props, elem)) { + if (zpool_prop_feature(nvpair_name(elem))) + has_features = B_TRUE; + } + + if (has_features || nvlist_lookup_uint64(props, + zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { version = SPA_VERSION; - ASSERT(version <= SPA_VERSION); + } + ASSERT(SPA_VERSION_IS_SUPPORTED(version)); spa->spa_first_txg = txg; spa->spa_uberblock.ub_txg = txg - 1; @@ -3059,8 +3413,10 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, spa->spa_l2cache.sav_sync = B_TRUE; } + spa->spa_is_initializing = B_TRUE; spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); spa->spa_meta_objset = dp->dp_meta_objset; + spa->spa_is_initializing = B_FALSE; /* * Create DDTs (dedup tables). @@ -3084,6 +3440,9 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, cmn_err(CE_PANIC, "failed to add pool config"); } + if (spa_version(spa) >= SPA_VERSION_FEATURES) + spa_feature_create_zap_objects(spa, tx); + if (zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, sizeof (uint64_t), 1, &version, tx) != 0) { @@ -3276,7 +3635,7 @@ spa_import_rootpool(char *devpath, char *devid) } #endif if (config == NULL) { - cmn_err(CE_NOTE, "Can not read the pool label from '%s'", + cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", devpath); return (EIO); } @@ -3590,6 +3949,8 @@ spa_tryimport(nvlist_t *tryconfig) state) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, spa->spa_uberblock.ub_timestamp) == 0); + VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, + spa->spa_load_info) == 0); /* * If the bootfs property exists on this pool then we @@ -5305,7 +5666,7 @@ spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) * information. This avoids the dbuf_will_dirty() path and * saves us a pre-read to get data we don't actually care about. */ - bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE); + bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); packed = vmem_alloc(bufsize, KM_PUSHPAGE); VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, @@ -5381,6 +5742,14 @@ spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) config = spa_config_generate(spa, spa->spa_root_vdev, dmu_tx_get_txg(tx), B_FALSE); + /* + * If we're upgrading the spa version then make sure that + * the config object gets updated with the correct version. + */ + if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) + fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, + spa->spa_uberblock.ub_version); + spa_config_exit(spa, SCL_STATE, FTAG); if (spa->spa_config_syncing) @@ -5390,6 +5759,24 @@ spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) spa_sync_nvlist(spa, spa->spa_config_object, config, tx); } +static void +spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx) +{ + spa_t *spa = arg1; + uint64_t version = *(uint64_t *)arg2; + + /* + * Setting the version is special cased when first creating the pool. + */ + ASSERT(tx->tx_txg != TXG_INITIAL); + + ASSERT(version <= SPA_VERSION); + ASSERT(version >= spa_version(spa)); + + spa->spa_uberblock.ub_version = version; + vdev_config_dirty(spa->spa_root_vdev); +} + /* * Set zpool properties. */ @@ -5399,32 +5786,39 @@ spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx) spa_t *spa = arg1; objset_t *mos = spa->spa_meta_objset; nvlist_t *nvp = arg2; - nvpair_t *elem; - uint64_t intval; - char *strval; - zpool_prop_t prop; - const char *propname; - zprop_type_t proptype; + nvpair_t *elem = NULL; mutex_enter(&spa->spa_props_lock); - elem = NULL; while ((elem = nvlist_next_nvpair(nvp, elem))) { - switch (prop = zpool_name_to_prop(nvpair_name(elem))) { + uint64_t intval; + char *strval, *fname; + zpool_prop_t prop; + const char *propname; + zprop_type_t proptype; + zfeature_info_t *feature; + + prop = zpool_name_to_prop(nvpair_name(elem)); + switch ((int)prop) { + case ZPROP_INVAL: + /* + * We checked this earlier in spa_prop_validate(). + */ + ASSERT(zpool_prop_feature(nvpair_name(elem))); + + fname = strchr(nvpair_name(elem), '@') + 1; + VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature)); + + spa_feature_enable(spa, feature, tx); + break; + case ZPOOL_PROP_VERSION: + VERIFY(nvpair_value_uint64(elem, &intval) == 0); /* - * Only set version for non-zpool-creation cases - * (set/import). spa_create() needs special care - * for version setting. + * The version is synced seperatly before other + * properties and should be correct by now. */ - if (tx->tx_txg != TXG_INITIAL) { - VERIFY(nvpair_value_uint64(elem, - &intval) == 0); - ASSERT(intval <= SPA_VERSION); - ASSERT(intval >= spa_version(spa)); - spa->spa_uberblock.ub_version = intval; - vdev_config_dirty(spa->spa_root_vdev); - } + ASSERT3U(spa_version(spa), >=, intval); break; case ZPOOL_PROP_ALTROOT: @@ -5461,14 +5855,10 @@ spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx) * Set pool property values in the poolprops mos object. */ if (spa->spa_pool_props_object == 0) { - VERIFY((spa->spa_pool_props_object = - zap_create(mos, DMU_OT_POOL_PROPS, - DMU_OT_NONE, 0, tx)) > 0); - - VERIFY(zap_update(mos, + spa->spa_pool_props_object = + zap_create_link(mos, DMU_OT_POOL_PROPS, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, - 8, 1, &spa->spa_pool_props_object, tx) - == 0); + tx); } /* normalize the property name */ @@ -5567,6 +5957,11 @@ spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) /* Keeping the freedir open increases spa_minref */ spa->spa_minref += 3; } + + if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && + spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { + spa_feature_create_zap_objects(spa, tx); + } } /* @@ -5738,6 +6133,9 @@ spa_sync(spa_t *spa, uint64_t txg) rvd->vdev_children, txg, B_TRUE); } + if (error == 0) + spa->spa_last_synced_guid = rvd->vdev_guid; + spa_config_exit(spa, SCL_STATE, FTAG); if (error == 0) diff --git a/module/zfs/spa_config.c b/module/zfs/spa_config.c index c86884148..09149e622 100644 --- a/module/zfs/spa_config.c +++ b/module/zfs/spa_config.c @@ -22,7 +22,7 @@ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. - * Copyright (c) 2011 by Delphix. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/spa.h> @@ -35,6 +35,7 @@ #include <sys/utsname.h> #include <sys/systeminfo.h> #include <sys/sunddi.h> +#include <sys/zfeature.h> #ifdef _KERNEL #include <sys/kobj.h> #include <sys/zone.h> @@ -408,6 +409,12 @@ spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats) VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); nvlist_free(nvroot); + /* + * Store what's necessary for reading the MOS in the label. + */ + VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, + spa->spa_label_features) == 0); + if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) { ddt_histogram_t *ddh; ddt_stat_t *dds; diff --git a/module/zfs/spa_misc.c b/module/zfs/spa_misc.c index 4a8e6adfd..5ec8e688e 100644 --- a/module/zfs/spa_misc.c +++ b/module/zfs/spa_misc.c @@ -20,7 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011 by Delphix. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. */ @@ -49,6 +49,7 @@ #include <sys/arc.h> #include <sys/ddt.h> #include "zfs_prop.h" +#include "zfeature_common.h" /* * SPA locking @@ -217,7 +218,7 @@ * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual * locking is, always, based on spa_namespace_lock and spa_config_lock[]. * - * spa_rename() is also implemented within this file since is requires + * spa_rename() is also implemented within this file since it requires * manipulation of the namespace. */ @@ -479,8 +480,22 @@ spa_add(const char *name, nvlist_t *config, const char *altroot) VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0); - if (config != NULL) + if (config != NULL) { + nvlist_t *features; + + if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, + &features) == 0) { + VERIFY(nvlist_dup(features, &spa->spa_label_features, + 0) == 0); + } + VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); + } + + if (spa->spa_label_features == NULL) { + VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, + KM_SLEEP) == 0); + } return (spa); } @@ -518,6 +533,7 @@ spa_remove(spa_t *spa) list_destroy(&spa->spa_config_list); + nvlist_free(spa->spa_label_features); nvlist_free(spa->spa_load_info); spa_config_set(spa, NULL); @@ -1025,6 +1041,20 @@ spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) * ========================================================================== */ +void +spa_activate_mos_feature(spa_t *spa, const char *feature) +{ + (void) nvlist_add_boolean(spa->spa_label_features, feature); + vdev_config_dirty(spa->spa_root_vdev); +} + +void +spa_deactivate_mos_feature(spa_t *spa, const char *feature) +{ + (void) nvlist_remove_all(spa->spa_label_features, feature); + vdev_config_dirty(spa->spa_root_vdev); +} + /* * Rename a spa_t. */ @@ -1175,12 +1205,22 @@ spa_generate_guid(spa_t *spa) void sprintf_blkptr(char *buf, const blkptr_t *bp) { - char *type = NULL; + char type[256]; char *checksum = NULL; char *compress = NULL; if (bp != NULL) { - type = dmu_ot[BP_GET_TYPE(bp)].ot_name; + if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { + dmu_object_byteswap_t bswap = + DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); + (void) snprintf(type, sizeof (type), "bswap %s %s", + DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? + "metadata" : "data", + dmu_ot_byteswap[bswap].ob_name); + } else { + (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, + sizeof (type)); + } checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; } @@ -1252,6 +1292,12 @@ spa_get_dsl(spa_t *spa) return (spa->spa_dsl_pool); } +boolean_t +spa_is_initializing(spa_t *spa) +{ + return (spa->spa_is_initializing); +} + blkptr_t * spa_get_rootblkptr(spa_t *spa) { @@ -1288,16 +1334,29 @@ spa_name(spa_t *spa) uint64_t spa_guid(spa_t *spa) { + dsl_pool_t *dp = spa_get_dsl(spa); + uint64_t guid; + /* * If we fail to parse the config during spa_load(), we can go through * the error path (which posts an ereport) and end up here with no root * vdev. We stash the original pool guid in 'spa_config_guid' to handle * this case. */ - if (spa->spa_root_vdev != NULL) + if (spa->spa_root_vdev == NULL) + return (spa->spa_config_guid); + + guid = spa->spa_last_synced_guid != 0 ? + spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; + + /* + * Return the most recently synced out guid unless we're + * in syncing context. + */ + if (dp && dsl_pool_sync_context(dp)) return (spa->spa_root_vdev->vdev_guid); else - return (spa->spa_config_guid); + return (guid); } uint64_t @@ -1532,6 +1591,7 @@ spa_init(int mode) vdev_cache_stat_init(); zfs_prop_init(); zpool_prop_init(); + zpool_feature_init(); spa_config_load(); l2arc_start(); } diff --git a/module/zfs/txg.c b/module/zfs/txg.c index 17494bcda..838a6f642 100644 --- a/module/zfs/txg.c +++ b/module/zfs/txg.c @@ -20,6 +20,8 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Portions Copyright 2011 Martin Matuska + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/zfs_context.h> @@ -669,7 +671,7 @@ txg_list_destroy(txg_list_t *tl) mutex_destroy(&tl->tl_lock); } -int +boolean_t txg_list_empty(txg_list_t *tl, uint64_t txg) { return (tl->tl_head[txg & TXG_MASK] == NULL); diff --git a/module/zfs/vdev.c b/module/zfs/vdev.c index e374f6d78..b96975103 100644 --- a/module/zfs/vdev.c +++ b/module/zfs/vdev.c @@ -1348,8 +1348,9 @@ vdev_validate(vdev_t *vd, boolean_t strict) if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { uint64_t aux_guid = 0; nvlist_t *nvl; + uint64_t txg = strict ? spa->spa_config_txg : -1ULL; - if ((label = vdev_label_read_config(vd)) == NULL) { + if ((label = vdev_label_read_config(vd, txg)) == NULL) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_BAD_LABEL); return (0); @@ -1532,7 +1533,7 @@ vdev_reopen(vdev_t *vd) !l2arc_vdev_present(vd)) l2arc_add_vdev(spa, vd); } else { - (void) vdev_validate(vd, B_TRUE); + (void) vdev_validate(vd, spa_last_synced_txg(spa)); } /* @@ -1993,14 +1994,14 @@ vdev_validate_aux(vdev_t *vd) if (!vdev_readable(vd)) return (0); - if ((label = vdev_label_read_config(vd)) == NULL) { + if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); return (-1); } if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || - version > SPA_VERSION || + !SPA_VERSION_IS_SUPPORTED(version) || nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || guid != vd->vdev_guid || nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { diff --git a/module/zfs/vdev_label.c b/module/zfs/vdev_label.c index 7ac23500f..1fe36feb2 100644 --- a/module/zfs/vdev_label.c +++ b/module/zfs/vdev_label.c @@ -18,8 +18,10 @@ * * CDDL HEADER END */ + /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ /* @@ -121,6 +123,8 @@ * txg Transaction group in which this label was written * pool_guid Unique identifier for this pool * vdev_tree An nvlist describing vdev tree. + * features_for_read + * An nvlist of the features necessary for reading the MOS. * * Each leaf device label also contains the following: * @@ -428,13 +432,23 @@ vdev_top_config_generate(spa_t *spa, nvlist_t *config) kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); } +/* + * Returns the configuration from the label of the given vdev. For vdevs + * which don't have a txg value stored on their label (i.e. spares/cache) + * or have not been completely initialized (txg = 0) just return + * the configuration from the first valid label we find. Otherwise, + * find the most up-to-date label that does not exceed the specified + * 'txg' value. + */ nvlist_t * -vdev_label_read_config(vdev_t *vd) +vdev_label_read_config(vdev_t *vd, uint64_t txg) { spa_t *spa = vd->vdev_spa; nvlist_t *config = NULL; vdev_phys_t *vp; zio_t *zio; + uint64_t best_txg = 0; + int error = 0; int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE; int l; @@ -448,6 +462,7 @@ vdev_label_read_config(vdev_t *vd) retry: for (l = 0; l < VDEV_LABELS; l++) { + nvlist_t *label = NULL; zio = zio_root(spa, NULL, NULL, flags); @@ -457,12 +472,31 @@ retry: if (zio_wait(zio) == 0 && nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), - &config, 0) == 0) - break; + &label, 0) == 0) { + uint64_t label_txg = 0; + + /* + * Auxiliary vdevs won't have txg values in their + * labels and newly added vdevs may not have been + * completely initialized so just return the + * configuration from the first valid label we + * encounter. + */ + error = nvlist_lookup_uint64(label, + ZPOOL_CONFIG_POOL_TXG, &label_txg); + if ((error || label_txg == 0) && !config) { + config = label; + break; + } else if (label_txg <= txg && label_txg > best_txg) { + best_txg = label_txg; + nvlist_free(config); + config = fnvlist_dup(label); + } + } - if (config != NULL) { - nvlist_free(config); - config = NULL; + if (label != NULL) { + nvlist_free(label); + label = NULL; } } @@ -497,7 +531,7 @@ vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, /* * Read the label, if any, and perform some basic sanity checks. */ - if ((label = vdev_label_read_config(vd)) == NULL) + if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) return (B_FALSE); (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, @@ -838,7 +872,7 @@ retry: * come back up, we fail to see the uberblock for txg + 1 because, say, * it was on a mirrored device and the replica to which we wrote txg + 1 * is now offline. If we then make some changes and sync txg + 1, and then - * the missing replica comes back, then for a new seconds we'll have two + * the missing replica comes back, then for a few seconds we'll have two * conflicting uberblocks on disk with the same txg. The solution is simple: * among uberblocks with equal txg, choose the one with the latest timestamp. */ @@ -858,47 +892,49 @@ vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) return (0); } +struct ubl_cbdata { + uberblock_t *ubl_ubbest; /* Best uberblock */ + vdev_t *ubl_vd; /* vdev associated with the above */ +}; + static void vdev_uberblock_load_done(zio_t *zio) { + vdev_t *vd = zio->io_vd; spa_t *spa = zio->io_spa; zio_t *rio = zio->io_private; uberblock_t *ub = zio->io_data; - uberblock_t *ubbest = rio->io_private; + struct ubl_cbdata *cbp = rio->io_private; - ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); + ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); if (zio->io_error == 0 && uberblock_verify(ub) == 0) { mutex_enter(&rio->io_lock); if (ub->ub_txg <= spa->spa_load_max_txg && - vdev_uberblock_compare(ub, ubbest) > 0) - *ubbest = *ub; + vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { + /* + * Keep track of the vdev in which this uberblock + * was found. We will use this information later + * to obtain the config nvlist associated with + * this uberblock. + */ + *cbp->ubl_ubbest = *ub; + cbp->ubl_vd = vd; + } mutex_exit(&rio->io_lock); } zio_buf_free(zio->io_data, zio->io_size); } -void -vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) +static void +vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, + struct ubl_cbdata *cbp) { - spa_t *spa = vd->vdev_spa; - vdev_t *rvd = spa->spa_root_vdev; - int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | - ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; int c, l, n; - if (vd == rvd) { - ASSERT(zio == NULL); - spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); - zio = zio_root(spa, NULL, ubbest, flags); - bzero(ubbest, sizeof (uberblock_t)); - } - - ASSERT(zio != NULL); - for (c = 0; c < vd->vdev_children; c++) - vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); + vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { for (l = 0; l < VDEV_LABELS; l++) { @@ -911,11 +947,46 @@ vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) } } } +} - if (vd == rvd) { - (void) zio_wait(zio); - spa_config_exit(spa, SCL_ALL, FTAG); - } +/* + * Reads the 'best' uberblock from disk along with its associated + * configuration. First, we read the uberblock array of each label of each + * vdev, keeping track of the uberblock with the highest txg in each array. + * Then, we read the configuration from the same vdev as the best uberblock. + */ +void +vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) +{ + zio_t *zio; + spa_t *spa = rvd->vdev_spa; + struct ubl_cbdata cb; + int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | + ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; + + ASSERT(ub); + ASSERT(config); + + bzero(ub, sizeof (uberblock_t)); + *config = NULL; + + cb.ubl_ubbest = ub; + cb.ubl_vd = NULL; + + spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); + zio = zio_root(spa, NULL, &cb, flags); + vdev_uberblock_load_impl(zio, rvd, flags, &cb); + (void) zio_wait(zio); + + /* + * It's possible that the best uberblock was discovered on a label + * that has a configuration which was written in a future txg. + * Search all labels on this vdev to find the configuration that + * matches the txg for our uberblock. + */ + if (cb.ubl_vd != NULL) + *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); + spa_config_exit(spa, SCL_ALL, FTAG); } /* diff --git a/module/zfs/zap.c b/module/zfs/zap.c index fac54eab0..fd3021be6 100644 --- a/module/zfs/zap.c +++ b/module/zfs/zap.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ /* @@ -946,6 +947,19 @@ fzap_prefetch(zap_name_t *zn) * Helper functions for consumers. */ +uint64_t +zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj, + const char *name, dmu_tx_t *tx) +{ + uint64_t new_obj; + + VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0); + VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj, + tx) == 0); + + return (new_obj); +} + int zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask, char *name) @@ -1080,6 +1094,16 @@ zap_add_int_key(objset_t *os, uint64_t obj, } int +zap_update_int_key(objset_t *os, uint64_t obj, + uint64_t key, uint64_t value, dmu_tx_t *tx) +{ + char name[20]; + + (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); + return (zap_update(os, obj, name, 8, 1, &value, tx)); +} + +int zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep) { char name[20]; diff --git a/module/zfs/zap_micro.c b/module/zfs/zap_micro.c index d5b97dac9..3d8cae0d3 100644 --- a/module/zfs/zap_micro.c +++ b/module/zfs/zap_micro.c @@ -20,7 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011 by Delphix. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #include <sys/zio.h> @@ -461,7 +461,7 @@ zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx, { dmu_object_info_t doi; dmu_object_info_from_db(db, &doi); - ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap); + ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP); } #endif @@ -585,7 +585,7 @@ mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags, { dmu_object_info_t doi; dmu_object_info_from_db(db, &doi); - ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap); + ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP); } #endif diff --git a/module/zfs/zfeature.c b/module/zfs/zfeature.c new file mode 100644 index 000000000..c09b32d17 --- /dev/null +++ b/module/zfs/zfeature.c @@ -0,0 +1,432 @@ +/* + * 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) 2012 by Delphix. All rights reserved. + */ + +#include <sys/zfs_context.h> +#include <sys/zfeature.h> +#include <sys/dmu.h> +#include <sys/nvpair.h> +#include <sys/zap.h> +#include <sys/dmu_tx.h> +#include "zfeature_common.h" +#include <sys/spa_impl.h> + +/* + * ZFS Feature Flags + * ----------------- + * + * ZFS feature flags are used to provide fine-grained versioning to the ZFS + * on-disk format. Once enabled on a pool feature flags replace the old + * spa_version() number. + * + * Each new on-disk format change will be given a uniquely identifying string + * guid rather than a version number. This avoids the problem of different + * organizations creating new on-disk formats with the same version number. To + * keep feature guids unique they should consist of the reverse dns name of the + * organization which implemented the feature and a short name for the feature, + * separated by a colon (e.g. com.delphix:async_destroy). + * + * Reference Counts + * ---------------- + * + * Within each pool features can be in one of three states: disabled, enabled, + * or active. These states are differentiated by a reference count stored on + * disk for each feature: + * + * 1) If there is no reference count stored on disk the feature is disabled. + * 2) If the reference count is 0 a system administrator has enabled the + * feature, but the feature has not been used yet, so no on-disk + * format changes have been made. + * 3) If the reference count is greater than 0 the feature is active. + * The format changes required by the feature are currently on disk. + * Note that if the feature's format changes are reversed the feature + * may choose to set its reference count back to 0. + * + * Feature flags makes no differentiation between non-zero reference counts + * for an active feature (e.g. a reference count of 1 means the same thing as a + * reference count of 27834721), but feature implementations may choose to use + * the reference count to store meaningful information. For example, a new RAID + * implementation might set the reference count to the number of vdevs using + * it. If all those disks are removed from the pool the feature goes back to + * having a reference count of 0. + * + * It is the responsibility of the individual features to maintain a non-zero + * reference count as long as the feature's format changes are present on disk. + * + * Dependencies + * ------------ + * + * Each feature may depend on other features. The only effect of this + * relationship is that when a feature is enabled all of its dependencies are + * automatically enabled as well. Any future work to support disabling of + * features would need to ensure that features cannot be disabled if other + * enabled features depend on them. + * + * On-disk Format + * -------------- + * + * When feature flags are enabled spa_version() is set to SPA_VERSION_FEATURES + * (5000). In order for this to work the pool is automatically upgraded to + * SPA_VERSION_BEFORE_FEATURES (28) first, so all pre-feature flags on disk + * format changes will be in use. + * + * Information about features is stored in 3 ZAP objects in the pool's MOS. + * These objects are linked to by the following names in the pool directory + * object: + * + * 1) features_for_read: feature guid -> reference count + * Features needed to open the pool for reading. + * 2) features_for_write: feature guid -> reference count + * Features needed to open the pool for writing. + * 3) feature_descriptions: feature guid -> descriptive string + * A human readable string. + * + * All enabled features appear in either features_for_read or + * features_for_write, but not both. + * + * To open a pool in read-only mode only the features listed in + * features_for_read need to be supported. + * + * To open the pool in read-write mode features in both features_for_read and + * features_for_write need to be supported. + * + * Some features may be required to read the ZAP objects containing feature + * information. To allow software to check for compatibility with these features + * before the pool is opened their names must be stored in the label in a + * new "features_for_read" entry (note that features that are only required + * to write to a pool never need to be stored in the label since the + * features_for_write ZAP object can be read before the pool is written to). + * To save space in the label features must be explicitly marked as needing to + * be written to the label. Also, reference counts are not stored in the label, + * instead any feature whose reference count drops to 0 is removed from the + * label. + * + * Adding New Features + * ------------------- + * + * Features must be registered in zpool_feature_init() function in + * zfeature_common.c using the zfeature_register() function. This function + * has arguments to specify if the feature should be stored in the + * features_for_read or features_for_write ZAP object and if it needs to be + * written to the label when active. + * + * Once a feature is registered it will appear as a "feature@<feature name>" + * property which can be set by an administrator. Feature implementors should + * use the spa_feature_is_enabled() and spa_feature_is_active() functions to + * query the state of a feature and the spa_feature_incr() and + * spa_feature_decr() functions to change an enabled feature's reference count. + * Reference counts may only be updated in the syncing context. + * + * Features may not perform enable-time initialization. Instead, any such + * initialization should occur when the feature is first used. This design + * enforces that on-disk changes be made only when features are used. Code + * should only check if a feature is enabled using spa_feature_is_enabled(), + * not by relying on any feature specific metadata existing. If a feature is + * enabled, but the feature's metadata is not on disk yet then it should be + * created as needed. + * + * As an example, consider the com.delphix:async_destroy feature. This feature + * relies on the existence of a bptree in the MOS that store blocks for + * asynchronous freeing. This bptree is not created when async_destroy is + * enabled. Instead, when a dataset is destroyed spa_feature_is_enabled() is + * called to check if async_destroy is enabled. If it is and the bptree object + * does not exist yet, the bptree object is created as part of the dataset + * destroy and async_destroy's reference count is incremented to indicate it + * has made an on-disk format change. Later, after the destroyed dataset's + * blocks have all been asynchronously freed there is no longer any use for the + * bptree object, so it is destroyed and async_destroy's reference count is + * decremented back to 0 to indicate that it has undone its on-disk format + * changes. + */ + +typedef enum { + FEATURE_ACTION_ENABLE, + FEATURE_ACTION_INCR, + FEATURE_ACTION_DECR, +} feature_action_t; + +/* + * Checks that the features active in the specified object are supported by + * this software. Adds each unsupported feature (name -> description) to + * the supplied nvlist. + */ +boolean_t +feature_is_supported(objset_t *os, uint64_t obj, uint64_t desc_obj, + nvlist_t *unsup_feat, nvlist_t *enabled_feat) +{ + boolean_t supported; + zap_cursor_t *zc; + zap_attribute_t *za; + char *buf; + + zc = kmem_alloc(sizeof(zap_cursor_t), KM_SLEEP); + za = kmem_alloc(sizeof(zap_attribute_t), KM_SLEEP); + buf = kmem_alloc(MAXPATHLEN, KM_SLEEP); + + supported = B_TRUE; + for (zap_cursor_init(zc, os, obj); + zap_cursor_retrieve(zc, za) == 0; + zap_cursor_advance(zc)) { + ASSERT(za->za_integer_length == sizeof (uint64_t) && + za->za_num_integers == 1); + + if (NULL != enabled_feat) { + fnvlist_add_uint64(enabled_feat, za->za_name, + za->za_first_integer); + } + + if (za->za_first_integer != 0 && + !zfeature_is_supported(za->za_name)) { + supported = B_FALSE; + + if (NULL != unsup_feat) { + char *desc = ""; + + if (zap_lookup(os, desc_obj, za->za_name, + 1, sizeof (buf), buf) == 0) + desc = buf; + + VERIFY(nvlist_add_string(unsup_feat, + za->za_name, desc) == 0); + } + } + } + zap_cursor_fini(zc); + + kmem_free(buf, MAXPATHLEN); + kmem_free(za, sizeof(zap_attribute_t)); + kmem_free(zc, sizeof(zap_cursor_t)); + + return (supported); +} + +static int +feature_get_refcount(objset_t *os, uint64_t read_obj, uint64_t write_obj, + zfeature_info_t *feature, uint64_t *res) +{ + int err; + uint64_t refcount; + uint64_t zapobj = feature->fi_can_readonly ? write_obj : read_obj; + + /* + * If the pool is currently being created, the feature objects may not + * have been allocated yet. Act as though all features are disabled. + */ + if (zapobj == 0) + return (ENOTSUP); + + err = zap_lookup(os, zapobj, feature->fi_guid, sizeof (uint64_t), 1, + &refcount); + if (err != 0) { + if (err == ENOENT) + return (ENOTSUP); + else + return (err); + } + *res = refcount; + return (0); +} + +static int +feature_do_action(objset_t *os, uint64_t read_obj, uint64_t write_obj, + uint64_t desc_obj, zfeature_info_t *feature, feature_action_t action, + dmu_tx_t *tx) +{ + int error; + uint64_t refcount; + uint64_t zapobj = feature->fi_can_readonly ? write_obj : read_obj; + + ASSERT(0 != zapobj); + ASSERT(zfeature_is_valid_guid(feature->fi_guid)); + + error = zap_lookup(os, zapobj, feature->fi_guid, + sizeof (uint64_t), 1, &refcount); + + /* + * If we can't ascertain the status of the specified feature, an I/O + * error occurred. + */ + if (error != 0 && error != ENOENT) + return (error); + + switch (action) { + case FEATURE_ACTION_ENABLE: + /* + * If the feature is already enabled, ignore the request. + */ + if (error == 0) + return (0); + refcount = 0; + break; + case FEATURE_ACTION_INCR: + if (error == ENOENT) + return (ENOTSUP); + if (refcount == UINT64_MAX) + return (EOVERFLOW); + refcount++; + break; + case FEATURE_ACTION_DECR: + if (error == ENOENT) + return (ENOTSUP); + if (refcount == 0) + return (EOVERFLOW); + refcount--; + break; + default: + ASSERT(0); + break; + } + + if (action == FEATURE_ACTION_ENABLE) { + int i; + + for (i = 0; feature->fi_depends[i] != NULL; i++) { + zfeature_info_t *dep = feature->fi_depends[i]; + + error = feature_do_action(os, read_obj, write_obj, + desc_obj, dep, FEATURE_ACTION_ENABLE, tx); + if (error != 0) + return (error); + } + } + + error = zap_update(os, zapobj, feature->fi_guid, + sizeof (uint64_t), 1, &refcount, tx); + if (error != 0) + return (error); + + if (action == FEATURE_ACTION_ENABLE) { + error = zap_update(os, desc_obj, + feature->fi_guid, 1, strlen(feature->fi_desc) + 1, + feature->fi_desc, tx); + if (error != 0) + return (error); + } + + if (action == FEATURE_ACTION_INCR && refcount == 1 && feature->fi_mos) { + spa_activate_mos_feature(dmu_objset_spa(os), feature->fi_guid); + } + + if (action == FEATURE_ACTION_DECR && refcount == 0) { + spa_deactivate_mos_feature(dmu_objset_spa(os), + feature->fi_guid); + } + + return (0); +} + +void +spa_feature_create_zap_objects(spa_t *spa, dmu_tx_t *tx) +{ + /* + * We create feature flags ZAP objects in two instances: during pool + * creation and during pool upgrade. + */ + ASSERT(dsl_pool_sync_context(spa_get_dsl(spa)) || (!spa->spa_sync_on && + tx->tx_txg == TXG_INITIAL)); + + spa->spa_feat_for_read_obj = zap_create_link(spa->spa_meta_objset, + DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_FEATURES_FOR_READ, tx); + spa->spa_feat_for_write_obj = zap_create_link(spa->spa_meta_objset, + DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_FEATURES_FOR_WRITE, tx); + spa->spa_feat_desc_obj = zap_create_link(spa->spa_meta_objset, + DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, + DMU_POOL_FEATURE_DESCRIPTIONS, tx); +} + +/* + * Enable any required dependencies, then enable the requested feature. + */ +void +spa_feature_enable(spa_t *spa, zfeature_info_t *feature, dmu_tx_t *tx) +{ + ASSERT3U(spa_version(spa), >=, SPA_VERSION_FEATURES); + VERIFY3U(0, ==, feature_do_action(spa->spa_meta_objset, + spa->spa_feat_for_read_obj, spa->spa_feat_for_write_obj, + spa->spa_feat_desc_obj, feature, FEATURE_ACTION_ENABLE, tx)); +} + +/* + * If the specified feature has not yet been enabled, this function returns + * ENOTSUP; otherwise, this function increments the feature's refcount (or + * returns EOVERFLOW if the refcount cannot be incremented). This function must + * be called from syncing context. + */ +void +spa_feature_incr(spa_t *spa, zfeature_info_t *feature, dmu_tx_t *tx) +{ + ASSERT3U(spa_version(spa), >=, SPA_VERSION_FEATURES); + VERIFY3U(0, ==, feature_do_action(spa->spa_meta_objset, + spa->spa_feat_for_read_obj, spa->spa_feat_for_write_obj, + spa->spa_feat_desc_obj, feature, FEATURE_ACTION_INCR, tx)); +} + +/* + * If the specified feature has not yet been enabled, this function returns + * ENOTSUP; otherwise, this function decrements the feature's refcount (or + * returns EOVERFLOW if the refcount is already 0). This function must + * be called from syncing context. + */ +void +spa_feature_decr(spa_t *spa, zfeature_info_t *feature, dmu_tx_t *tx) +{ + ASSERT3U(spa_version(spa), >=, SPA_VERSION_FEATURES); + VERIFY3U(0, ==, feature_do_action(spa->spa_meta_objset, + spa->spa_feat_for_read_obj, spa->spa_feat_for_write_obj, + spa->spa_feat_desc_obj, feature, FEATURE_ACTION_DECR, tx)); +} + +boolean_t +spa_feature_is_enabled(spa_t *spa, zfeature_info_t *feature) +{ + int err; + uint64_t refcount = 0; + + if (spa_version(spa) < SPA_VERSION_FEATURES) + return (B_FALSE); + + err = feature_get_refcount(spa->spa_meta_objset, + spa->spa_feat_for_read_obj, spa->spa_feat_for_write_obj, + feature, &refcount); + ASSERT(err == 0 || err == ENOTSUP); + return (err == 0); +} + +boolean_t +spa_feature_is_active(spa_t *spa, zfeature_info_t *feature) +{ + int err; + uint64_t refcount = 0; + + if (spa_version(spa) < SPA_VERSION_FEATURES) + return (B_FALSE); + + err = feature_get_refcount(spa->spa_meta_objset, + spa->spa_feat_for_read_obj, spa->spa_feat_for_write_obj, + feature, &refcount); + ASSERT(err == 0 || err == ENOTSUP); + return (err == 0 && refcount > 0); +} diff --git a/module/zfs/zfeature_common.c b/module/zfs/zfeature_common.c new file mode 100644 index 000000000..40066991b --- /dev/null +++ b/module/zfs/zfeature_common.c @@ -0,0 +1,163 @@ +/* + * 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) 2012 by Delphix. All rights reserved. + */ + +#ifdef _KERNEL +#include <sys/systm.h> +#else +#include <errno.h> +#include <string.h> +#endif +#include <sys/debug.h> +#include <sys/fs/zfs.h> +#include <sys/inttypes.h> +#include <sys/types.h> +#include "zfeature_common.h" + +/* + * Set to disable all feature checks while opening pools, allowing pools with + * unsupported features to be opened. Set for testing only. + */ +boolean_t zfeature_checks_disable = B_FALSE; + +zfeature_info_t spa_feature_table[SPA_FEATURES]; + +/* + * Valid characters for feature guids. This list is mainly for aesthetic + * purposes and could be expanded in the future. There are different allowed + * characters in the guids reverse dns portion (before the colon) and its + * short name (after the colon). + */ +static int +valid_char(char c, boolean_t after_colon) +{ + return ((c >= 'a' && c <= 'z') || + (c >= '0' && c <= '9') || + c == (after_colon ? '_' : '.')); +} + +/* + * Every feature guid must contain exactly one colon which separates a reverse + * dns organization name from the feature's "short" name (e.g. + * "com.company:feature_name"). + */ +boolean_t +zfeature_is_valid_guid(const char *name) +{ + int i; + boolean_t has_colon = B_FALSE; + + i = 0; + while (name[i] != '\0') { + char c = name[i++]; + if (c == ':') { + if (has_colon) + return (B_FALSE); + has_colon = B_TRUE; + continue; + } + if (!valid_char(c, has_colon)) + return (B_FALSE); + } + + return (has_colon); +} + +boolean_t +zfeature_is_supported(const char *guid) +{ + if (zfeature_checks_disable) + return (B_TRUE); + + return (0 == zfeature_lookup_guid(guid, NULL)); +} + +int +zfeature_lookup_guid(const char *guid, zfeature_info_t **res) +{ + int i; + + for (i = 0; i < SPA_FEATURES; i++) { + zfeature_info_t *feature = &spa_feature_table[i]; + if (strcmp(guid, feature->fi_guid) == 0) { + if (res != NULL) + *res = feature; + return (0); + } + } + + return (ENOENT); +} + +int +zfeature_lookup_name(const char *name, zfeature_info_t **res) +{ + int i; + + for (i = 0; i < SPA_FEATURES; i++) { + zfeature_info_t *feature = &spa_feature_table[i]; + if (strcmp(name, feature->fi_uname) == 0) { + if (res != NULL) + *res = feature; + return (0); + } + } + + return (ENOENT); +} + +static void +zfeature_register(int fid, const char *guid, const char *name, const char *desc, + boolean_t readonly, boolean_t mos, zfeature_info_t **deps) +{ + zfeature_info_t *feature = &spa_feature_table[fid]; + static zfeature_info_t *nodeps[] = { NULL }; + + ASSERT(name != NULL); + ASSERT(desc != NULL); + ASSERT(!readonly || !mos); + ASSERT3U(fid, <, SPA_FEATURES); + ASSERT(zfeature_is_valid_guid(guid)); + + if (deps == NULL) + deps = nodeps; + + feature->fi_guid = guid; + feature->fi_uname = name; + feature->fi_desc = desc; + feature->fi_can_readonly = readonly; + feature->fi_mos = mos; + feature->fi_depends = deps; +} + +void +zpool_feature_init(void) +{ + zfeature_register(SPA_FEATURE_ASYNC_DESTROY, + "com.delphix:async_destroy", "async_destroy", + "Destroy filesystems asynchronously.", B_TRUE, B_FALSE, NULL); + zfeature_register(SPA_FEATURE_EMPTY_BPOBJ, + "com.delphix:empty_bpobj", "empty_bpobj", + "Snapshots use less space.", B_TRUE, B_FALSE, NULL); +} diff --git a/module/zfs/zfs_ioctl.c b/module/zfs/zfs_ioctl.c index c609203ea..98b1dd794 100644 --- a/module/zfs/zfs_ioctl.c +++ b/module/zfs/zfs_ioctl.c @@ -18,6 +18,7 @@ * * CDDL HEADER END */ + /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Portions Copyright 2011 Martin Matuska @@ -1107,6 +1108,8 @@ get_zfs_sb(const char *dsname, zfs_sb_t **zsbp) /* * Find a zfs_sb_t for a mounted filesystem, or create our own, in which * case its z_sb will be NULL, and it will be opened as the owner. + * If 'writer' is set, the z_teardown_lock will be held for RW_WRITER, + * which prevents all inode ops from running. */ static int zfs_sb_hold(const char *name, void *tag, zfs_sb_t **zsbp, boolean_t writer) @@ -1170,7 +1173,7 @@ zfs_ioc_pool_create(zfs_cmd_t *zc) (void) nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION), &version); - if (version < SPA_VERSION_INITIAL || version > SPA_VERSION) { + if (!SPA_VERSION_IS_SUPPORTED(version)) { error = EINVAL; goto pool_props_bad; } @@ -1297,6 +1300,15 @@ zfs_ioc_pool_configs(zfs_cmd_t *zc) return (error); } +/* + * inputs: + * zc_name name of the pool + * + * outputs: + * zc_cookie real errno + * zc_nvlist_dst config nvlist + * zc_nvlist_dst_size size of config nvlist + */ static int zfs_ioc_pool_stats(zfs_cmd_t *zc) { @@ -1398,7 +1410,8 @@ zfs_ioc_pool_upgrade(zfs_cmd_t *zc) if ((error = spa_open(zc->zc_name, &spa, FTAG)) != 0) return (error); - if (zc->zc_cookie < spa_version(spa) || zc->zc_cookie > SPA_VERSION) { + if (zc->zc_cookie < spa_version(spa) || + !SPA_VERSION_IS_SUPPORTED(zc->zc_cookie)) { spa_close(spa, FTAG); return (EINVAL); } diff --git a/module/zfs/zil.c b/module/zfs/zil.c index 220f2d79e..c9618c13e 100644 --- a/module/zfs/zil.c +++ b/module/zfs/zil.c @@ -20,7 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011 by Delphix. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ /* Portions Copyright 2010 Robert Milkowski */ @@ -480,6 +480,38 @@ zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, uint64_t txg, boolean_t fastwrite) } /* + * Called when we create in-memory log transactions so that we know + * to cleanup the itxs at the end of spa_sync(). + */ +void +zilog_dirty(zilog_t *zilog, uint64_t txg) +{ + dsl_pool_t *dp = zilog->zl_dmu_pool; + dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); + + if (dsl_dataset_is_snapshot(ds)) + panic("dirtying snapshot!"); + + if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg) == 0) { + /* up the hold count until we can be written out */ + dmu_buf_add_ref(ds->ds_dbuf, zilog); + } +} + +boolean_t +zilog_is_dirty(zilog_t *zilog) +{ + dsl_pool_t *dp = zilog->zl_dmu_pool; + int t; + + for (t = 0; t < TXG_SIZE; t++) { + if (txg_list_member(&dp->dp_dirty_zilogs, zilog, t)) + return (B_TRUE); + } + return (B_FALSE); +} + +/* * Create an on-disk intent log. */ static lwb_t * @@ -601,14 +633,21 @@ zil_destroy(zilog_t *zilog, boolean_t keep_first) kmem_cache_free(zil_lwb_cache, lwb); } } else if (!keep_first) { - (void) zil_parse(zilog, zil_free_log_block, - zil_free_log_record, tx, zh->zh_claim_txg); + zil_destroy_sync(zilog, tx); } mutex_exit(&zilog->zl_lock); dmu_tx_commit(tx); } +void +zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx) +{ + ASSERT(list_is_empty(&zilog->zl_lwb_list)); + (void) zil_parse(zilog, zil_free_log_block, + zil_free_log_record, tx, zilog->zl_header->zh_claim_txg); +} + int zil_claim(const char *osname, void *txarg) { @@ -1042,6 +1081,8 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) return (NULL); ASSERT(lwb->lwb_buf != NULL); + ASSERT(zilog_is_dirty(zilog) || + spa_freeze_txg(zilog->zl_spa) != UINT64_MAX); if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY) dlen = P2ROUNDUP_TYPED( @@ -1272,7 +1313,7 @@ zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_RENAME) zil_async_to_sync(zilog, itx->itx_oid); - if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) + if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) txg = ZILTEST_TXG; else txg = dmu_tx_get_txg(tx); @@ -1323,6 +1364,7 @@ zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) } itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx); + zilog_dirty(zilog, txg); mutex_exit(&itxg->itxg_lock); /* Release the old itxs now we've dropped the lock */ @@ -1332,7 +1374,10 @@ zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) /* * If there are any in-memory intent log transactions which have now been - * synced then start up a taskq to free them. + * synced then start up a taskq to free them. We should only do this after we + * have written out the uberblocks (i.e. txg has been comitted) so that + * don't inadvertently clean out in-memory log records that would be required + * by zil_commit(). */ void zil_clean(zilog_t *zilog, uint64_t synced_txg) @@ -1837,6 +1882,7 @@ zil_close(zilog_t *zilog) mutex_exit(&zilog->zl_lock); if (txg) txg_wait_synced(zilog->zl_dmu_pool, txg); + ASSERT(!zilog_is_dirty(zilog)); taskq_destroy(zilog->zl_clean_taskq); zilog->zl_clean_taskq = NULL; diff --git a/module/zfs/zio.c b/module/zfs/zio.c index bfb817b78..638105a09 100644 --- a/module/zfs/zio.c +++ b/module/zfs/zio.c @@ -704,7 +704,7 @@ zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && zp->zp_compress >= ZIO_COMPRESS_OFF && zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && - zp->zp_type < DMU_OT_NUMTYPES && + DMU_OT_IS_VALID(zp->zp_type) && zp->zp_level < 32 && zp->zp_copies > 0 && zp->zp_copies <= spa_max_replication(spa) && @@ -988,7 +988,7 @@ zio_read_bp_init(zio_t *zio) zio_push_transform(zio, cbuf, psize, psize, zio_decompress); } - if (!dmu_ot[BP_GET_TYPE(bp)].ot_metadata && BP_GET_LEVEL(bp) == 0) + if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0) zio->io_flags |= ZIO_FLAG_DONT_CACHE; if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) @@ -1248,7 +1248,7 @@ __zio_execute(zio_t *zio) while (zio->io_stage < ZIO_STAGE_DONE) { enum zio_stage pipeline = zio->io_pipeline; enum zio_stage stage = zio->io_stage; - dsl_pool_t *dsl; + dsl_pool_t *dp; boolean_t cut; int rv; @@ -1262,7 +1262,7 @@ __zio_execute(zio_t *zio) ASSERT(stage <= ZIO_STAGE_DONE); - dsl = spa_get_dsl(zio->io_spa); + dp = spa_get_dsl(zio->io_spa); cut = (stage == ZIO_STAGE_VDEV_IO_START) ? zio_requeue_io_start_cut_in_line : B_FALSE; @@ -1272,16 +1272,24 @@ __zio_execute(zio_t *zio) * or may wait for an I/O that needs an interrupt thread * to complete, issue async to avoid deadlock. * - * If we are in the txg_sync_thread or being called - * during pool init issue async to minimize stack depth. - * Both of these call paths may be recursively called. - * * For VDEV_IO_START, we cut in line so that the io will * be sent to disk promptly. */ - if (((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && - zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) || - (dsl != NULL && dsl_pool_sync_context(dsl))) { + if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && + zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { + zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); + return; + } + + /* + * If we executing in the context of the tx_sync_thread, + * or we are performing pool initialization outside of a + * zio_taskq[ZIO_TASKQ_ISSUE] context. Then issue the zio + * async to minimize stack usage for these deep call paths. + */ + if ((dp && curthread == dp->dp_tx.tx_sync_thread) || + (dp && spa_is_initializing(dp->dp_spa) && + !zio_taskq_member(zio, ZIO_TASKQ_ISSUE))) { zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); return; } @@ -3131,6 +3139,48 @@ static zio_pipe_stage_t *zio_pipeline[] = { zio_done }; +/* dnp is the dnode for zb1->zb_object */ +boolean_t +zbookmark_is_before(const dnode_phys_t *dnp, const zbookmark_t *zb1, + const zbookmark_t *zb2) +{ + uint64_t zb1nextL0, zb2thisobj; + + ASSERT(zb1->zb_objset == zb2->zb_objset); + ASSERT(zb2->zb_level == 0); + + /* + * A bookmark in the deadlist is considered to be after + * everything else. + */ + if (zb2->zb_object == DMU_DEADLIST_OBJECT) + return (B_TRUE); + + /* The objset_phys_t isn't before anything. */ + if (dnp == NULL) + return (B_FALSE); + + zb1nextL0 = (zb1->zb_blkid + 1) << + ((zb1->zb_level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)); + + zb2thisobj = zb2->zb_object ? zb2->zb_object : + zb2->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT); + + if (zb1->zb_object == DMU_META_DNODE_OBJECT) { + uint64_t nextobj = zb1nextL0 * + (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT) >> DNODE_SHIFT; + return (nextobj <= zb2thisobj); + } + + if (zb1->zb_object < zb2thisobj) + return (B_TRUE); + if (zb1->zb_object > zb2thisobj) + return (B_FALSE); + if (zb2->zb_object == DMU_META_DNODE_OBJECT) + return (B_FALSE); + return (zb1nextL0 <= zb2->zb_blkid); +} + #if defined(_KERNEL) && defined(HAVE_SPL) /* Fault injection */ EXPORT_SYMBOL(zio_injection_enabled); |