2009-01-13 12:42:02

by Jan Kara

Subject: [PATCH] ext2: Update also inode on disk when dir is IS_DIRSYNC

We used to just write changed page for IS_DIRSYNC inodes. But we also have to
update directory inode itself just for the case that we've allocated a new
block and changed i_size.

Signed-off-by: Jan Kara <[email protected]>
CC: Pavel Machek <[email protected]>
---
fs/ext2/dir.c | 5 ++---
1 files changed, 2 insertions(+), 3 deletions(-)

Andrew, would you please merge the patch? Ext3 and ext4 don't have this
problem BTW.

diff --git a/fs/ext2/dir.c b/fs/ext2/dir.c
index 9a0fc40..7fba549 100644
--- a/fs/ext2/dir.c
+++ b/fs/ext2/dir.c
@@ -94,11 +94,10 @@ static int ext2_commit_chunk(struct page *page, loff_t pos, unsigned len)
i_size_write(dir, pos+len);
mark_inode_dirty(dir);
}
+ unlock_page(page);

if (IS_DIRSYNC(dir))
- err = write_one_page(page, 1);
- else
- unlock_page(page);
+ err = ext2_sync_inode(dir);

return err;
}
--
1.6.0.4

2009-01-13 13:08:08

by Pavel Machek

Subject: Re: [PATCH] ext2: Update also inode on disk when dir is IS_DIRSYNC

> We used to just write changed page for IS_DIRSYNC inodes. But we also have to
> update directory inode itself just for the case that we've allocated a new
> block and changed i_size.
>
> Signed-off-by: Jan Kara <[email protected]>
> CC: Pavel Machek <[email protected]>

Tested, and I can't break ext2 in simple tests now.
Pavel

--
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

2009-01-13 23:17:53

by Andrew Morton

Subject: Re: [PATCH] ext2: Update also inode on disk when dir is IS_DIRSYNC

On Tue, 13 Jan 2009 13:42:02 +0100
Jan Kara <[email protected]> wrote:

> We used to just write changed page for IS_DIRSYNC inodes. But we also have to
> update directory inode itself just for the case that we've allocated a new
> block and changed i_size.
>
> Signed-off-by: Jan Kara <[email protected]>
> CC: Pavel Machek <[email protected]>
> ---
> fs/ext2/dir.c | 5 ++---
> 1 files changed, 2 insertions(+), 3 deletions(-)
>
> Andrew, would you please merge the patch? Ext3 and ext4 don't have this
> problem BTW.
>
> diff --git a/fs/ext2/dir.c b/fs/ext2/dir.c
> index 9a0fc40..7fba549 100644
> --- a/fs/ext2/dir.c
> +++ b/fs/ext2/dir.c
> @@ -94,11 +94,10 @@ static int ext2_commit_chunk(struct page *page, loff_t pos, unsigned len)
> i_size_write(dir, pos+len);
> mark_inode_dirty(dir);
> }
> + unlock_page(page);
>
> if (IS_DIRSYNC(dir))
> - err = write_one_page(page, 1);
> - else
> - unlock_page(page);
> + err = ext2_sync_inode(dir);
>

But with this change we no longer sync the data page? Don't we need
something like this:

static int ext2_commit_chunk(struct page *page, loff_t pos, unsigned len)
{
struct address_space *mapping = page->mapping;
struct inode *dir = mapping->host;
int err = 0;

dir->i_version++;
block_write_end(NULL, mapping, pos, len, len, page, NULL);

if (pos+len > dir->i_size) {
i_size_write(dir, pos+len);
mark_inode_dirty(dir);
}

if (IS_DIRSYNC(dir)) {
err = write_one_page(page, 1);
if (err == 0)
err = ext2_sync_inode(dir);
} else {
unlock_page(page);
}

return err;
}

?

2009-01-14 15:12:36

by Jan Kara

Subject: Re: [PATCH] ext2: Update also inode on disk when dir is IS_DIRSYNC

On Tue 13-01-09 15:17:16, Andrew Morton wrote:
> On Tue, 13 Jan 2009 13:42:02 +0100
> Jan Kara <[email protected]> wrote:
>
> > We used to just write changed page for IS_DIRSYNC inodes. But we also have to
> > update directory inode itself just for the case that we've allocated a new
> > block and changed i_size.
> >
> > Signed-off-by: Jan Kara <[email protected]>
> > CC: Pavel Machek <[email protected]>
> > ---
> > fs/ext2/dir.c | 5 ++---
> > 1 files changed, 2 insertions(+), 3 deletions(-)
> >
> > Andrew, would you please merge the patch? Ext3 and ext4 don't have this
> > problem BTW.
> >
> > diff --git a/fs/ext2/dir.c b/fs/ext2/dir.c
> > index 9a0fc40..7fba549 100644
> > --- a/fs/ext2/dir.c
> > +++ b/fs/ext2/dir.c
> > @@ -94,11 +94,10 @@ static int ext2_commit_chunk(struct page *page, loff_t pos, unsigned len)
> > i_size_write(dir, pos+len);
> > mark_inode_dirty(dir);
> > }
> > + unlock_page(page);
> >
> > if (IS_DIRSYNC(dir))
> > - err = write_one_page(page, 1);
> > - else
> > - unlock_page(page);
> > + err = ext2_sync_inode(dir);
> >
>
> But with this change we no longer sync the data page? Don't we need
> something like this:
Oh, correct, we do. I was confused by sync_inode() calling fdatawrite()
but did not realize that the way ext2_sync_inode() calls it has start ==
end == 0. Thanks for spotting this.

> static int ext2_commit_chunk(struct page *page, loff_t pos, unsigned len)
> {
> struct address_space *mapping = page->mapping;
> struct inode *dir = mapping->host;
> int err = 0;
>
> dir->i_version++;
> block_write_end(NULL, mapping, pos, len, len, page, NULL);
>
> if (pos+len > dir->i_size) {
> i_size_write(dir, pos+len);
> mark_inode_dirty(dir);
> }
>
> if (IS_DIRSYNC(dir)) {
> err = write_one_page(page, 1);
> if (err == 0)
> err = ext2_sync_inode(dir);
> } else {
> unlock_page(page);
> }
>
> return err;
> }
>
> ?
Yes, I'll send you a new version of the patch and also send refreshed
second patch with blkdev_issue_flush() calls.

Honza
--
Jan Kara <[email protected]>
SUSE Labs, CR