Mercurial > illumos > illumos-gate
changeset 13782:8f78aae28a63
3104 eliminate empty bpobjs
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <chris.siden@delphix.com>
Reviewed by: Garrett D'Amore <garrett@damore.org>
Approved by: Eric Schrock <eric.schrock@delphix.com>
| author | Matthew Ahrens <mahrens@delphix.com> |
|---|---|
| date | Mon, 27 Aug 2012 03:58:14 -0700 |
| parents | 64a1ab954737 |
| children | b3333788688b |
| files | usr/src/common/zfs/zfeature_common.c usr/src/common/zfs/zfeature_common.h usr/src/man/man5/zpool-features.5 usr/src/uts/common/fs/zfs/bpobj.c usr/src/uts/common/fs/zfs/dsl_deadlist.c usr/src/uts/common/fs/zfs/dsl_pool.c usr/src/uts/common/fs/zfs/sys/bpobj.h usr/src/uts/common/fs/zfs/sys/dmu.h usr/src/uts/common/fs/zfs/sys/dsl_pool.h usr/src/uts/common/fs/zfs/sys/zap.h usr/src/uts/common/fs/zfs/zap.c usr/src/uts/common/fs/zfs/zfeature.c |
| diffstat | 12 files changed, 166 insertions(+), 11 deletions(-) [+] |
line wrap: on
line diff
--- a/usr/src/common/zfs/zfeature_common.c Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/common/zfs/zfeature_common.c Mon Aug 27 03:58:14 2012 -0700 @@ -153,4 +153,7 @@ 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); }
--- a/usr/src/common/zfs/zfeature_common.h Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/common/zfs/zfeature_common.h Mon Aug 27 03:58:14 2012 -0700 @@ -51,6 +51,7 @@ enum spa_feature { SPA_FEATURE_ASYNC_DESTROY, + SPA_FEATURE_EMPTY_BPOBJ, SPA_FEATURES } spa_feature_t;
--- a/usr/src/man/man5/zpool-features.5 Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/man/man5/zpool-features.5 Mon Aug 27 03:58:14 2012 -0700 @@ -169,5 +169,33 @@ This feature is only \fBactive\fR while \fBfreeing\fR is non\-zero. .RE + +.sp +.ne 2 +.na +\fB\fBempty_bpobj\fR\fR +.ad +.RS 4n +.TS +l l . +GUID com.delphix:empty_bpobj +READ\-ONLY COMPATIBLE yes +DEPENDENCIES none +.TE + +This feature increases the performance of creating and using a large +number of snapshots of a single filesystem or volume, and also reduces +the disk space required. + +When there are many snapshots, each snapshot uses many Block Pointer +Objects (bpobj's) to track blocks associated with that snapshot. +However, in common use cases, most of these bpobj's are empty. This +feature allows us to create each bpobj on-demand, thus eliminating the +empty bpobjs. + +This feature is \fBactive\fR while there are any filesystems, volumes, +or snapshots which were created after enabling this feature. +.RE + .SH "SEE ALSO" \fBzpool\fR(1M)
--- a/usr/src/uts/common/fs/zfs/bpobj.c Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/uts/common/fs/zfs/bpobj.c Mon Aug 27 03:58:14 2012 -0700 @@ -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 @@ 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 @@ 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 @@ 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;
--- a/usr/src/uts/common/fs/zfs/dsl_deadlist.c Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/uts/common/fs/zfs/dsl_deadlist.c Mon Aug 27 03:58:14 2012 -0700 @@ -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> @@ -163,12 +163,49 @@ 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) { @@ -197,7 +234,7 @@ 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); } /* @@ -217,7 +254,7 @@ dle = kmem_alloc(sizeof (*dle), KM_SLEEP); 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); @@ -243,8 +280,7 @@ 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); @@ -302,7 +338,7 @@ 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)); } @@ -400,7 +436,7 @@ 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
--- a/usr/src/uts/common/fs/zfs/dsl_pool.c Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/uts/common/fs/zfs/dsl_pool.c Mon Aug 27 03:58:14 2012 -0700 @@ -182,6 +182,15 @@ 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);
--- a/usr/src/uts/common/fs/zfs/sys/bpobj.h Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/uts/common/fs/zfs/sys/bpobj.h Mon Aug 27 03:58:14 2012 -0700 @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ #ifndef _SYS_BPOBJ_H @@ -67,7 +68,9 @@ typedef int bpobj_itor_t(void *arg, const blkptr_t *bp, dmu_tx_t *tx); uint64_t bpobj_alloc(objset_t *mos, int blocksize, dmu_tx_t *tx); +uint64_t bpobj_alloc_empty(objset_t *os, int blocksize, dmu_tx_t *tx); void bpobj_free(objset_t *os, uint64_t obj, dmu_tx_t *tx); +void bpobj_decr_empty(objset_t *os, dmu_tx_t *tx); int bpobj_open(bpobj_t *bpo, objset_t *mos, uint64_t object); void bpobj_close(bpobj_t *bpo);
--- a/usr/src/uts/common/fs/zfs/sys/dmu.h Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/uts/common/fs/zfs/sys/dmu.h Mon Aug 27 03:58:14 2012 -0700 @@ -305,6 +305,7 @@ #define DMU_POOL_SCAN "scan" #define DMU_POOL_FREE_BPOBJ "free_bpobj" #define DMU_POOL_BPTREE_OBJ "bptree_obj" +#define DMU_POOL_EMPTY_BPOBJ "empty_bpobj" /* * Allocate an object from this objset. The range of object numbers
--- a/usr/src/uts/common/fs/zfs/sys/dsl_pool.h Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/uts/common/fs/zfs/sys/dsl_pool.h Mon Aug 27 03:58:14 2012 -0700 @@ -88,6 +88,7 @@ uint64_t dp_tmp_userrefs_obj; bpobj_t dp_free_bpobj; uint64_t dp_bptree_obj; + uint64_t dp_empty_bpobj; struct dsl_scan *dp_scan;
--- a/usr/src/uts/common/fs/zfs/sys/zap.h Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/uts/common/fs/zfs/sys/zap.h Mon Aug 27 03:58:14 2012 -0700 @@ -300,6 +300,8 @@ /* Here the key is an int and the value is a different int. */ int zap_add_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t value, dmu_tx_t *tx); +int zap_update_int_key(objset_t *os, uint64_t obj, + uint64_t key, uint64_t value, dmu_tx_t *tx); int zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep);
--- a/usr/src/uts/common/fs/zfs/zap.c Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/uts/common/fs/zfs/zap.c Mon Aug 27 03:58:14 2012 -0700 @@ -1094,6 +1094,16 @@ } 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];
--- a/usr/src/uts/common/fs/zfs/zfeature.c Sat Aug 25 02:44:53 2012 -0500 +++ b/usr/src/uts/common/fs/zfs/zfeature.c Mon Aug 27 03:58:14 2012 -0700 @@ -221,7 +221,12 @@ uint64_t refcount; uint64_t zapobj = feature->fi_can_readonly ? write_obj : read_obj; - ASSERT(0 != zapobj); + /* + * 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);