Now we have block_lock_hole_extend clearing the dirty flag of
buffer_heads outside i_size we should not find buffer_heads
which are unmapped and dirty in writepage. If we find do a WARN_ON.
We can still continue because block_write_full page look at the mapped
flag only.
Following sequence of events would result in the above condition.
1) truncate(f, 1024)
2) mmap(f, 0, 4096)
3) a[0] = 'a'
4) truncate(f, 4096)
5) writepage(...)
After step 3 we would have unmapped buffer_heads outside i_size.
After step 4 we would have unmapped buffer_heads within i_size.
Now that truncate is calling block_lock_hole_extend which in turn
is clearing the dirty flag, we can safely assume that we won't
find unmapped dirty buffer_heads in write page. If we did find one
we should find out why.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
---
fs/ext4/inode.c | 13 +++++++++++++
1 files changed, 13 insertions(+), 0 deletions(-)
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index f6f2202..ca26bdd 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -2509,6 +2509,11 @@ static int __ext4_journalled_writepage(struct page *page,
return ret;
}
+static int ext4_bh_unmapped_and_dirty(handle_t *handle, struct buffer_head *bh)
+{
+ return (!buffer_mapped(bh)) && buffer_dirty(bh);
+}
+
/*
* Note that we don't need to start a transaction unless we're journaling data
* because we should have holes filled from ext4_page_mkwrite(). We even don't
@@ -2622,6 +2627,14 @@ static int ext4_writepage(struct page *page,
/* now mark the buffer_heads as dirty and uptodate */
block_commit_write(page, 0, len);
}
+ /*
+ * There should not be any unmapped and dirty
+ * buffer_heads at this point. Look at block_lock_hole_extend
+ * for more info. If we find one print more info
+ */
+ WARN(walk_page_buffers(NULL, page_bufs, 0, len, NULL,
+ ext4_bh_unmapped_and_dirty),
+ "Unmapped dirty buffer_heads found in %s\n", __func__);
if (PageChecked(page) && ext4_should_journal_data(inode)) {
/*
--
1.6.3.1.145.gb74d77
On Thu 28-05-09 13:57:44, Aneesh Kumar K.V wrote:
> Now we have block_lock_hole_extend clearing the dirty flag of
> buffer_heads outside i_size we should not find buffer_heads
> which are unmapped and dirty in writepage. If we find do a WARN_ON.
> We can still continue because block_write_full page look at the mapped
> flag only.
>
> Following sequence of events would result in the above condition.
> 1) truncate(f, 1024)
> 2) mmap(f, 0, 4096)
> 3) a[0] = 'a'
> 4) truncate(f, 4096)
> 5) writepage(...)
>
> After step 3 we would have unmapped buffer_heads outside i_size.
> After step 4 we would have unmapped buffer_heads within i_size.
>
> Now that truncate is calling block_lock_hole_extend which in turn
> is clearing the dirty flag, we can safely assume that we won't
> find unmapped dirty buffer_heads in write page. If we did find one
> we should find out why.
Thanks. Apart from one coding-style nitpick, you can add:
Acked-by: Jan Kara <[email protected]>
Honza
>
> Signed-off-by: Aneesh Kumar K.V <[email protected]>
> ---
> fs/ext4/inode.c | 13 +++++++++++++
> 1 files changed, 13 insertions(+), 0 deletions(-)
>
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index f6f2202..ca26bdd 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -2509,6 +2509,11 @@ static int __ext4_journalled_writepage(struct page *page,
> return ret;
> }
>
> +static int ext4_bh_unmapped_and_dirty(handle_t *handle, struct buffer_head *bh)
> +{
> + return (!buffer_mapped(bh)) && buffer_dirty(bh);
^ ^
looks a bit strange
> +}
> +
> /*
> * Note that we don't need to start a transaction unless we're journaling data
> * because we should have holes filled from ext4_page_mkwrite(). We even don't
> @@ -2622,6 +2627,14 @@ static int ext4_writepage(struct page *page,
> /* now mark the buffer_heads as dirty and uptodate */
> block_commit_write(page, 0, len);
> }
> + /*
> + * There should not be any unmapped and dirty
> + * buffer_heads at this point. Look at block_lock_hole_extend
> + * for more info. If we find one print more info
> + */
> + WARN(walk_page_buffers(NULL, page_bufs, 0, len, NULL,
> + ext4_bh_unmapped_and_dirty),
> + "Unmapped dirty buffer_heads found in %s\n", __func__);
>
> if (PageChecked(page) && ext4_should_journal_data(inode)) {
> /*
> --
> 1.6.3.1.145.gb74d77
>
--
Jan Kara <[email protected]>
SUSE Labs, CR