c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2026-05-24 19:28:53 +03:00
Code Issues Packages Projects Releases Wiki Activity

DDT: Switch to using ZAP _by_dnode() interfaces

Browse Source 
As was previously done for BRT, avoid holding/releasing DDT ZAP
dnodes for every access.  Instead hold the dnodes during all their
life time, never releasing.

While at this, add _by_dnode() interfaces for zap_length_uint64()
and zap_count(), actively used by DDT code.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <alexander.motin@TrueNAS.com>
Closes #18047
This commit is contained in:
Alexander Motin
2025-12-15 12:49:14 -05:00
committed by Tony Hutter
parent fa857113a3
commit 4905686e67
6 changed files with 132 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files
module/zfs/ddt_zap.c
+23 -22
View File
@@ -33,6 +33,7 @@
#include <sys/ddt_impl.h>
#include <sys/zap.h>
#include <sys/dmu_tx.h>
#include <sys/dnode.h>
#include <sys/zio_compress.h>
static unsigned int ddt_zap_default_bs = 15;
@@ -120,14 +121,13 @@ ddt_zap_destroy(objset_t *os, uint64_t object, dmu_tx_t *tx)
}
static int
ddt_zap_lookup(objset_t *os, uint64_t object,
const ddt_key_t *ddk, void *phys, size_t psize)
ddt_zap_lookup(dnode_t *dn, const ddt_key_t *ddk, void *phys, size_t psize)
{
uchar_t *cbuf;
uint64_t one, csize;
int error;
error = zap_length_uint64(os, object, (uint64_t *)ddk,
error = zap_length_uint64_by_dnode(dn, (uint64_t *)ddk,
DDT_KEY_WORDS, &one, &csize);
if (error)
return (error);
@@ -137,7 +137,7 @@ ddt_zap_lookup(objset_t *os, uint64_t object,
cbuf = kmem_alloc(csize, KM_SLEEP);
error = zap_lookup_uint64(os, object, (uint64_t *)ddk,
error = zap_lookup_uint64_by_dnode(dn, (uint64_t *)ddk,
DDT_KEY_WORDS, 1, csize, cbuf);
if (error == 0)
ddt_zap_decompress(cbuf, phys, csize, psize);
@@ -148,26 +148,27 @@ ddt_zap_lookup(objset_t *os, uint64_t object,
}
static int
ddt_zap_contains(objset_t *os, uint64_t object, const ddt_key_t *ddk)
ddt_zap_contains(dnode_t *dn, const ddt_key_t *ddk)
{
return (zap_length_uint64(os, object, (uint64_t *)ddk, DDT_KEY_WORDS,
NULL, NULL));
return (zap_length_uint64_by_dnode(dn, (uint64_t *)ddk,
DDT_KEY_WORDS, NULL, NULL));
}
static void
ddt_zap_prefetch(objset_t *os, uint64_t object, const ddt_key_t *ddk)
ddt_zap_prefetch(dnode_t *dn, const ddt_key_t *ddk)
{
(void) zap_prefetch_uint64(os, object, (uint64_t *)ddk, DDT_KEY_WORDS);
(void) zap_prefetch_uint64_by_dnode(dn, (uint64_t *)ddk,
DDT_KEY_WORDS);
}
static void
ddt_zap_prefetch_all(objset_t *os, uint64_t object)
ddt_zap_prefetch_all(dnode_t *dn)
{
(void) zap_prefetch_object(os, object);
(void) zap_prefetch_object(dn->dn_objset, dn->dn_object);
}
static int
ddt_zap_update(objset_t *os, uint64_t object, const ddt_key_t *ddk,
ddt_zap_update(dnode_t *dn, const ddt_key_t *ddk,
const void *phys, size_t psize, dmu_tx_t *tx)
{
const size_t cbuf_size = psize + 1;
@@ -176,7 +177,7 @@ ddt_zap_update(objset_t *os, uint64_t object, const ddt_key_t *ddk,
uint64_t csize = ddt_zap_compress(phys, cbuf, psize, cbuf_size);
int error = zap_update_uint64(os, object, (uint64_t *)ddk,
int error = zap_update_uint64_by_dnode(dn, (uint64_t *)ddk,
DDT_KEY_WORDS, 1, csize, cbuf, tx);
kmem_free(cbuf, cbuf_size);
@@ -185,15 +186,14 @@ ddt_zap_update(objset_t *os, uint64_t object, const ddt_key_t *ddk,
}
static int
ddt_zap_remove(objset_t *os, uint64_t object, const ddt_key_t *ddk,
dmu_tx_t *tx)
ddt_zap_remove(dnode_t *dn, const ddt_key_t *ddk, dmu_tx_t *tx)
{
return (zap_remove_uint64(os, object, (uint64_t *)ddk,
return (zap_remove_uint64_by_dnode(dn, (uint64_t *)ddk,
DDT_KEY_WORDS, tx));
}
static int
ddt_zap_walk(objset_t *os, uint64_t object, uint64_t *walk, ddt_key_t *ddk,
ddt_zap_walk(dnode_t *dn, uint64_t *walk, ddt_key_t *ddk,
void *phys, size_t psize)
{
zap_cursor_t zc;
@@ -209,9 +209,10 @@ ddt_zap_walk(objset_t *os, uint64_t object, uint64_t *walk, ddt_key_t *ddk,
* scrub I/Os for each ZAP block that we read in, so
* reading the ZAP is unlikely to be the bottleneck.
*/
zap_cursor_init_noprefetch(&zc, os, object);
zap_cursor_init_noprefetch(&zc, dn->dn_objset, dn->dn_object);
} else {
zap_cursor_init_serialized(&zc, os, object, *walk);
zap_cursor_init_serialized(&zc, dn->dn_objset, dn->dn_object,
*walk);
}
if ((error = zap_cursor_retrieve(&zc, za)) == 0) {
uint64_t csize = za->za_num_integers;
@@ -221,7 +222,7 @@ ddt_zap_walk(objset_t *os, uint64_t object, uint64_t *walk, ddt_key_t *ddk,
uchar_t *cbuf = kmem_alloc(csize, KM_SLEEP);
error = zap_lookup_uint64(os, object, (uint64_t *)za->za_name,
error = zap_lookup_uint64_by_dnode(dn, (uint64_t *)za->za_name,
DDT_KEY_WORDS, 1, csize, cbuf);
ASSERT0(error);
if (error == 0) {
@@ -240,9 +241,9 @@ ddt_zap_walk(objset_t *os, uint64_t object, uint64_t *walk, ddt_key_t *ddk,
}
static int
ddt_zap_count(objset_t *os, uint64_t object, uint64_t *count)
ddt_zap_count(dnode_t *dn, uint64_t *count)
{
return (zap_count(os, object, count));
return (zap_count_by_dnode(dn, count));
}
const ddt_ops_t ddt_zap_ops = {