2007-02-10 14:44:10

by OGAWA Hirofumi

Subject: [PATCH] FAT: DIO-write fallback to normal buffered

[RESEND: forget to add [email protected]]

If the DIO write on FAT is expanding the size, it will be fail by -EINVAL,
because FAT can't handle it now.

This patch fallback it to the normal buffered-write and would return
success.

Signed-off-by: OGAWA Hirofumi <[email protected]>
---

fs/fat/inode.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff -puN fs/fat/inode.c~fat_direct-io-fallback fs/fat/inode.c
--- linux-2.6/fs/fat/inode.c~fat_direct-io-fallback 2007-02-10 22:08:33.000000000 +0900
+++ linux-2.6-hirofumi/fs/fat/inode.c 2007-02-10 22:08:33.000000000 +0900
@@ -173,10 +173,12 @@ static ssize_t fat_direct_IO(int rw, str
*
* But we must fill the remaining area or hole by nul for
* updating ->mmu_private.
+ *
+ * Return 0, and fallback to normal buffered write.
*/
loff_t size = offset + iov_length(iov, nr_segs);
if (MSDOS_I(inode)->mmu_private < size)
- return -EINVAL;
+ return 0;
}

/*
_

2007-02-14 10:51:59

by Jan Kara

Subject: Re: [PATCH] FAT: DIO-write fallback to normal buffered

On Sat 10-02-07 23:44:01, OGAWA Hirofumi wrote:
> [RESEND: forget to add [email protected]]
>
> If the DIO write on FAT is expanding the size, it will be fail by -EINVAL,
> because FAT can't handle it now.
>
> This patch fallback it to the normal buffered-write and would return
> success.
>
> Signed-off-by: OGAWA Hirofumi <[email protected]>
Signed-off-by: Jan Kara <[email protected]>

Just to explain a bit: I think that returning EINVAL is quite unexpected
for users in this case (I actually got a bugreport which turned out to be
this problem) and fallback to buffered IO seems to be a reasonable thing to
do. Probably it's not the cleanest solution but for FAT I think it's good
enough ;).

Honza

> ---
>
> fs/fat/inode.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff -puN fs/fat/inode.c~fat_direct-io-fallback fs/fat/inode.c
> --- linux-2.6/fs/fat/inode.c~fat_direct-io-fallback 2007-02-10 22:08:33.000000000 +0900
> +++ linux-2.6-hirofumi/fs/fat/inode.c 2007-02-10 22:08:33.000000000 +0900
> @@ -173,10 +173,12 @@ static ssize_t fat_direct_IO(int rw, str
> *
> * But we must fill the remaining area or hole by nul for
> * updating ->mmu_private.
> + *
> + * Return 0, and fallback to normal buffered write.
> */
> loff_t size = offset + iov_length(iov, nr_segs);
> if (MSDOS_I(inode)->mmu_private < size)
> - return -EINVAL;
> + return 0;
> }
>
> /*
> _
--
Jan Kara <[email protected]>
SuSE CR Labs