Do not iterate over data blocks scanning for bh's to forget as they're
never exist. This improves time taken by unlink / truncate syscall.
Tested by continuously truncating file that is being written by dd.
Another test is rm -rf of linux tree while tar unpacks it. With
ordered data mode condition unlikely(!tbh) was always met in
ext4_free_blocks. With journal data mode tbh was found only few times,
so optimisation is also possible.
Unlinking fallocated 60G file after doing sync && echo 3 >
/proc/sys/vm/drop_caches && time rm --help
X86 before (linux 3.6-rc4):
# time rm -f test1
real 0m2.710s
user 0m0.000s
sys 0m1.530s
X86 after:
# time rm -f test1
real 0m0.644s
user 0m0.003s
sys 0m0.060s
MIPS before (linux 2.6.37):
# time rm -f test1
real 0m 4.93s
user 0m 0.00s
sys 0m 4.61s
MIPS after:
# time rm -f test1
real 0m 0.16s
user 0m 0.00s
sys 0m 0.06s
---
fs/ext4/extents.c | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)
Index: linux-3.6-rc6/fs/ext4/extents.c
===================================================================
--- linux-3.6-rc6.orig/fs/ext4/extents.c 2012-09-17 12:08:48.842901966 -0400
+++ linux-3.6-rc6/fs/ext4/extents.c 2012-09-17 12:09:43.403630792 -0400
@@ -2274,10 +2274,13 @@ static int ext4_remove_blocks(handle_t *
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
unsigned short ee_len = ext4_ext_get_actual_len(ex);
ext4_fsblk_t pblk;
- int flags = EXT4_FREE_BLOCKS_FORGET;
+ int flags = 0;
if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
- flags |= EXT4_FREE_BLOCKS_METADATA;
+ flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
+ else if (ext4_should_journal_data(inode))
+ flags |= EXT4_FREE_BLOCKS_FORGET;
+
/*
* For bigalloc file systems, we never free a partial cluster
* at the beginning of the extent. Instead, we make a note
On Tue, Sep 18, 2012 at 01:00:30AM +0400, Andrey Sidorov wrote:
> Do not iterate over data blocks scanning for bh's to forget as they're
> never exist. This improves time taken by unlink / truncate syscall.
> Tested by continuously truncating file that is being written by dd.
> Another test is rm -rf of linux tree while tar unpacks it. With
> ordered data mode condition unlikely(!tbh) was always met in
> ext4_free_blocks. With journal data mode tbh was found only few times,
> so optimisation is also possible.
Thanks for this patch. It's good you did the testing, although from a
theoretical point of view I'm sure it's sound because the only case
where journal=data mode will there be data blocks in the buffer cache.
In the other cases, the data is cached in the page cache, so it's a
waste of time looking up the blocks so they can be bforgotten.
The one thing which is missing from your patch is a Signed-off-by:
footer. This has a specific legal meaning. See:
http://elinux.org/Developer_Certificate_Of_Origin
You don't need to resend this patch, though; just reply with an
acknowledgement that you agree to add your DCO to the patch:
Signed-off-by: Andrey Sidorov <[email protected]>
Thanks for contributing to the ext4 file system!
- Ted
On Tue, Sep 18, 2012 at 12:10 AM, Theodore Ts'o <[email protected]> wrote:
> The one thing which is missing from your patch is a Signed-off-by:
> footer. This has a specific legal meaning. See:
>
> http://elinux.org/Developer_Certificate_Of_Origin
>
> You don't need to resend this patch, though; just reply with an
> acknowledgement that you agree to add your DCO to the patch:
>
> Signed-off-by: Andrey Sidorov <[email protected]>
>
>
> Thanks for contributing to the ext4 file system!
>
> - Ted
Hi Ted,
Yes, please add
Signed-off-by: Andrey Sidorov <[email protected]>
Thanks!
On Tue, Sep 18, 2012 at 02:17:09PM -0400, Andrey Sidorov wrote:
>
> Yes, please add
> Signed-off-by: Andrey Sidorov <[email protected]>
Thanks, applied.
I am interested in your proposed patch 2/2 as well, but it's going to
require more detailed review, and it sounds like you were planning on
doing some work to better handle corner cases as well?
- Ted