From: Saugata Das <[email protected]>
Today, storage devices like eMMC has special features like data tagging
(introduced in MMC-4.5 version) in order to improve performance of some
specific writes. On MMC stack, data tagging is used for all writes which has
REQ_META flag set. On EXT4, however, currently REQ_META is set only for read.
This patch adds the capability to add REQ_META flag during meta data write.
This patch adds new function set_buffer_meta, which is used to set a bit
BH_Meta in the b_state field when meta data and super block is made dirty.
Signed-off-by: Saugata Das <[email protected]>
Acked-by: Arnd Bergmann <[email protected]>
---
fs/buffer.c | 3 +++
fs/ext4/ext4_jbd2.c | 4 ++++
fs/ext4/inode.c | 4 +++-
include/linux/buffer_head.h | 2 ++
4 files changed, 12 insertions(+), 1 deletions(-)
diff --git a/fs/buffer.c b/fs/buffer.c
index 19d8eb7..71c8c74 100644
--- a/fs/buffer.c
+++ b/fs/buffer.c
@@ -2926,6 +2926,9 @@ int submit_bh(int rw, struct buffer_head * bh)
if (test_set_buffer_req(bh) && (rw & WRITE))
clear_buffer_write_io_error(bh);
+ if (buffer_meta(bh) && (rw & WRITE))
+ rw |= REQ_META;
+
/*
* from here on down, it's all bio -- do the initial mapping,
* submit_bio -> generic_make_request may further map this bio around
diff --git a/fs/ext4/ext4_jbd2.c b/fs/ext4/ext4_jbd2.c
index aca1790..097c062 100644
--- a/fs/ext4/ext4_jbd2.c
+++ b/fs/ext4/ext4_jbd2.c
@@ -107,6 +107,8 @@ int __ext4_handle_dirty_metadata(const char *where, unsigned int line,
{
int err = 0;
+ set_buffer_meta(bh);
+
if (ext4_handle_valid(handle)) {
err = jbd2_journal_dirty_metadata(handle, bh);
if (err) {
@@ -143,6 +145,8 @@ int __ext4_handle_dirty_super(const char *where, unsigned int line,
struct buffer_head *bh = EXT4_SB(sb)->s_sbh;
int err = 0;
+ set_buffer_meta(bh);
+
if (ext4_handle_valid(handle)) {
err = jbd2_journal_dirty_metadata(handle, bh);
if (err)
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 92655fd..105e1d0 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -4144,8 +4144,10 @@ int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
err = __ext4_get_inode_loc(inode, &iloc, 0);
if (err)
return err;
- if (wbc->sync_mode == WB_SYNC_ALL)
+ if (wbc->sync_mode == WB_SYNC_ALL) {
+ set_buffer_meta(iloc.bh);
sync_dirty_buffer(iloc.bh);
+ }
if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
"IO error syncing inode");
diff --git a/include/linux/buffer_head.h b/include/linux/buffer_head.h
index 458f497..13bba17 100644
--- a/include/linux/buffer_head.h
+++ b/include/linux/buffer_head.h
@@ -34,6 +34,7 @@ enum bh_state_bits {
BH_Write_EIO, /* I/O error on write */
BH_Unwritten, /* Buffer is allocated on disk but not written */
BH_Quiet, /* Buffer Error Prinks to be quiet */
+ BH_Meta, /* Is meta */
BH_PrivateStart,/* not a state bit, but the first bit available
* for private allocation by other entities
@@ -124,6 +125,7 @@ BUFFER_FNS(Delay, delay)
BUFFER_FNS(Boundary, boundary)
BUFFER_FNS(Write_EIO, write_io_error)
BUFFER_FNS(Unwritten, unwritten)
+BUFFER_FNS(Meta, meta)
#define bh_offset(bh) ((unsigned long)(bh)->b_data & ~PAGE_MASK)
#define touch_buffer(bh) mark_page_accessed(bh->b_page)
--
1.7.4.3
On Thu, Mar 01, 2012 at 11:41:03PM +0530, Saugata Das wrote:
> From: Saugata Das <[email protected]>
>
> Today, storage devices like eMMC has special features like data tagging
> (introduced in MMC-4.5 version) in order to improve performance of some
> specific writes. On MMC stack, data tagging is used for all writes which has
> REQ_META flag set. On EXT4, however, currently REQ_META is set only for read.
> This patch adds the capability to add REQ_META flag during meta data write.
> This patch adds new function set_buffer_meta, which is used to set a bit
> BH_Meta in the b_state field when meta data and super block is made dirty.
>
> --- a/fs/buffer.c
> +++ b/fs/buffer.c
> @@ -2926,6 +2926,9 @@ int submit_bh(int rw, struct buffer_head * bh)
> if (test_set_buffer_req(bh) && (rw & WRITE))
> clear_buffer_write_io_error(bh);
>
> + if (buffer_meta(bh) && (rw & WRITE))
> + rw |= REQ_META;
> +
I know it's not necessary for your purposes as stated in the commit
description, but wouldn't it be better if buffer_meta() always caused
REQ_META to be set, regardless of whether we are doing a read or a
write?
- Ted
On 2 March 2012 00:03, Ted Ts'o <[email protected]> wrote:
> On Thu, Mar 01, 2012 at 11:41:03PM +0530, Saugata Das wrote:
>> From: Saugata Das <[email protected]>
>>
>> Today, storage devices like eMMC has special features like data tagging
>> (introduced in MMC-4.5 version) in order to improve performance of some
>> specific writes. On MMC stack, data tagging is used for all writes which has
>> REQ_META flag set. On EXT4, however, currently REQ_META is set only for read.
>> This patch adds the capability to add REQ_META flag during meta data write.
>> This patch adds new function set_buffer_meta, which is used to set a bit
>> BH_Meta in the b_state field when meta data and super block is made dirty.
>>
>> --- a/fs/buffer.c
>> +++ b/fs/buffer.c
>> @@ -2926,6 +2926,9 @@ int submit_bh(int rw, struct buffer_head * bh)
>> ? ? ? if (test_set_buffer_req(bh) && (rw & WRITE))
>> ? ? ? ? ? ? ? clear_buffer_write_io_error(bh);
>>
>> + ? ? if (buffer_meta(bh) && (rw & WRITE))
>> + ? ? ? ? ? ? rw |= REQ_META;
>> +
>
> I know it's not necessary for your purposes as stated in the commit
> description, but wouldn't it be better if buffer_meta() always caused
> REQ_META to be set, regardless of whether we are doing a read or a
> write?
Sure, I will modify this in the next version.
>
> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?- Ted
On 3/1/12 12:11 PM, Saugata Das wrote:
> From: Saugata Das <[email protected]>
>
> Today, storage devices like eMMC has special features like data tagging
> (introduced in MMC-4.5 version) in order to improve performance of some
> specific writes. On MMC stack, data tagging is used for all writes which has
> REQ_META flag set. On EXT4, however, currently REQ_META is set only for read.
> This patch adds the capability to add REQ_META flag during meta data write.
> This patch adds new function set_buffer_meta, which is used to set a bit
> BH_Meta in the b_state field when meta data and super block is made dirty.
>
> Signed-off-by: Saugata Das <[email protected]>
> Acked-by: Arnd Bergmann <[email protected]>
I like this idea too, partly because I've had a project in the back of my
mind to teach seekwatcher to optionally distinguish between metadata &
data IO on its charts, but I had noticed that ext4 didn't always tag
things.
-Eric
> ---
> fs/buffer.c | 3 +++
> fs/ext4/ext4_jbd2.c | 4 ++++
> fs/ext4/inode.c | 4 +++-
> include/linux/buffer_head.h | 2 ++
> 4 files changed, 12 insertions(+), 1 deletions(-)
>
> diff --git a/fs/buffer.c b/fs/buffer.c
> index 19d8eb7..71c8c74 100644
> --- a/fs/buffer.c
> +++ b/fs/buffer.c
> @@ -2926,6 +2926,9 @@ int submit_bh(int rw, struct buffer_head * bh)
> if (test_set_buffer_req(bh) && (rw & WRITE))
> clear_buffer_write_io_error(bh);
>
> + if (buffer_meta(bh) && (rw & WRITE))
> + rw |= REQ_META;
> +
> /*
> * from here on down, it's all bio -- do the initial mapping,
> * submit_bio -> generic_make_request may further map this bio around
> diff --git a/fs/ext4/ext4_jbd2.c b/fs/ext4/ext4_jbd2.c
> index aca1790..097c062 100644
> --- a/fs/ext4/ext4_jbd2.c
> +++ b/fs/ext4/ext4_jbd2.c
> @@ -107,6 +107,8 @@ int __ext4_handle_dirty_metadata(const char *where, unsigned int line,
> {
> int err = 0;
>
> + set_buffer_meta(bh);
> +
> if (ext4_handle_valid(handle)) {
> err = jbd2_journal_dirty_metadata(handle, bh);
> if (err) {
> @@ -143,6 +145,8 @@ int __ext4_handle_dirty_super(const char *where, unsigned int line,
> struct buffer_head *bh = EXT4_SB(sb)->s_sbh;
> int err = 0;
>
> + set_buffer_meta(bh);
> +
> if (ext4_handle_valid(handle)) {
> err = jbd2_journal_dirty_metadata(handle, bh);
> if (err)
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index 92655fd..105e1d0 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -4144,8 +4144,10 @@ int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
> err = __ext4_get_inode_loc(inode, &iloc, 0);
> if (err)
> return err;
> - if (wbc->sync_mode == WB_SYNC_ALL)
> + if (wbc->sync_mode == WB_SYNC_ALL) {
> + set_buffer_meta(iloc.bh);
> sync_dirty_buffer(iloc.bh);
> + }
> if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
> EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
> "IO error syncing inode");
> diff --git a/include/linux/buffer_head.h b/include/linux/buffer_head.h
> index 458f497..13bba17 100644
> --- a/include/linux/buffer_head.h
> +++ b/include/linux/buffer_head.h
> @@ -34,6 +34,7 @@ enum bh_state_bits {
> BH_Write_EIO, /* I/O error on write */
> BH_Unwritten, /* Buffer is allocated on disk but not written */
> BH_Quiet, /* Buffer Error Prinks to be quiet */
> + BH_Meta, /* Is meta */
>
> BH_PrivateStart,/* not a state bit, but the first bit available
> * for private allocation by other entities
> @@ -124,6 +125,7 @@ BUFFER_FNS(Delay, delay)
> BUFFER_FNS(Boundary, boundary)
> BUFFER_FNS(Write_EIO, write_io_error)
> BUFFER_FNS(Unwritten, unwritten)
> +BUFFER_FNS(Meta, meta)
>
> #define bh_offset(bh) ((unsigned long)(bh)->b_data & ~PAGE_MASK)
> #define touch_buffer(bh) mark_page_accessed(bh->b_page)