From c1801cbe59dfc99ba7aa34f09c0b6b8f35bb2d8f Mon Sep 17 00:00:00 2001
From: Alexander Motin <mav@FreeBSD.org>
Date: Fri, 11 Aug 2023 12:04:08 -0400
Subject: [PATCH] ZIL: Avoid dbuf_read() before dmu_sync().
In most cases dmu_sync() works with dirty records directly and does
not need actual data. The only exception is dmu_sync_late_arrival().
To save some CPU time use dmu_buf_hold_noread*() in z*_get_data()
and explicitly call dbuf_read() in dmu_sync_late_arrival(). There
is also a chance that by that time TXG will already be synced and
we won't have to do it at all.
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes #15153
---
include/sys/dmu.h | 4 ++++
include/sys/dmu_impl.h | 2 --
module/zfs/dmu.c | 9 ++++++++-
module/zfs/zfs_vnops.c | 4 ++--
module/zfs/zvol.c | 4 ++--
5 files changed, 16 insertions(+), 7 deletions(-)
diff --git a/include/sys/dmu.h b/include/sys/dmu.h
index 7e57d133c2..615ba8fe74 100644
--- a/include/sys/dmu.h
+++ b/include/sys/dmu.h
@@ -572,11 +572,15 @@ int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
int dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
uint64_t length, int read, const void *tag, int *numbufsp,
dmu_buf_t ***dbpp);
+int dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
+ const void *tag, dmu_buf_t **dbp);
int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
const void *tag, dmu_buf_t **dbp, int flags);
int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset,
uint64_t length, boolean_t read, const void *tag, int *numbufsp,
dmu_buf_t ***dbpp, uint32_t flags);
+int dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, const void *tag,
+ dmu_buf_t **dbp);
/*
* Add a reference to a dmu buffer that has already been held via
* dmu_buf_hold() in the current context.
diff --git a/include/sys/dmu_impl.h b/include/sys/dmu_impl.h
index ce6ae3c665..83ae2b76ba 100644
--- a/include/sys/dmu_impl.h
+++ b/include/sys/dmu_impl.h
@@ -247,8 +247,6 @@ typedef struct dmu_sendstatus {
void dmu_object_zapify(objset_t *, uint64_t, dmu_object_type_t, dmu_tx_t *);
void dmu_object_free_zapified(objset_t *, uint64_t, dmu_tx_t *);
-int dmu_buf_hold_noread(objset_t *, uint64_t, uint64_t,
- const void *, dmu_buf_t **);
#ifdef __cplusplus
}
diff --git a/module/zfs/dmu.c b/module/zfs/dmu.c
index 078811dbf4..ddb29020b0 100644
--- a/module/zfs/dmu.c
+++ b/module/zfs/dmu.c
@@ -165,7 +165,7 @@ dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
{ zfs_acl_byteswap, "acl" }
};
-static int
+int
dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset,
const void *tag, dmu_buf_t **dbp)
{
@@ -185,6 +185,7 @@ dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset,
*dbp = &db->db;
return (0);
}
+
int
dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
const void *tag, dmu_buf_t **dbp)
@@ -1653,6 +1654,12 @@ dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
{
dmu_sync_arg_t *dsa;
dmu_tx_t *tx;
+ int error;
+
+ error = dbuf_read((dmu_buf_impl_t *)zgd->zgd_db, NULL,
+ DB_RF_CANFAIL | DB_RF_NOPREFETCH);
+ if (error != 0)
+ return (error);
tx = dmu_tx_create(os);
dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c
index 54ea43363b..07c177f3bc 100644
--- a/module/zfs/zfs_vnops.c
+++ b/module/zfs/zfs_vnops.c
@@ -917,8 +917,8 @@ zfs_get_data(void *arg, uint64_t gen, lr_write_t *lr, char *buf,
}
#endif
if (error == 0)
- error = dmu_buf_hold(os, object, offset, zgd, &db,
- DMU_READ_NO_PREFETCH);
+ error = dmu_buf_hold_noread(os, object, offset, zgd,
+ &db);
if (error == 0) {
blkptr_t *bp = &lr->lr_blkptr;
diff --git a/module/zfs/zvol.c b/module/zfs/zvol.c
index cd4e6f0c75..f44d1b7b55 100644
--- a/module/zfs/zvol.c
+++ b/module/zfs/zvol.c
@@ -727,8 +727,8 @@ zvol_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
offset = P2ALIGN_TYPED(offset, size, uint64_t);
zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset,
size, RL_READER);
- error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
- DMU_READ_NO_PREFETCH);
+ error = dmu_buf_hold_noread_by_dnode(zv->zv_dn, offset, zgd,
+ &db);
if (error == 0) {
blkptr_t *bp = &lr->lr_blkptr;