*/
/*
* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
*/
#ifndef _SYS_BPOBJ_H
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);
#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
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;
/* 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);
typedef enum spa_feature {
SPA_FEATURE_ASYNC_DESTROY,
+ SPA_FEATURE_EMPTY_BPOBJ,
SPA_FEATURES
} spa_feature_t;
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)
*/
/*
* 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)
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);
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));
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;
*/
/*
* 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>
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)
{
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);
}
/*
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);
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);
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));
}
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
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);
return (zap_add(os, obj, name, 8, 1, &value, tx));
}
+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)
{
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);
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);
}
projects / mirror_zfs-debian.git / commitdiff