On Tue, Oct 3, 2023 at 4:57 AM Reuben Hawkins <reubenhwk@xxxxxxxxx> wrote:
>
> Readahead was factored to call generic_fadvise. That refactor added an
> S_ISREG restriction which broke readahead on block devices.
>
> In addition to S_ISREG, this change checks S_ISBLK to fix block device
> readahead. There is no change in behavior with any file type besides block
> devices in this change.
>
> Fixes: 3d8f7615319b ("vfs: implement readahead(2) using POSIX_FADV_WILLNEED")
> Signed-off-by: Reuben Hawkins <reubenhwk@xxxxxxxxx>
Reviewed-by: Amir Goldstein <amir73il@xxxxxxxxx>
Christian,
We've had a lot of back and forth on this patch.
I've asked Reuben to post this minimal and backportable patch
which should replace the one that is currently on the head of
vfs.misc in order to only fix the regression without other changes
of behavior.
If you happen to send any vfs fixes to 6.6, this one can also
be included but no need to rush.
It would be wrong to mix the regression fix with other changes of
behavior because the latter we may be forced to revert in the future.
Either Reuben or I will follow up later with patches to change the
behavior of posix_fadvise and readahead(2) for pipes and sockets
as suggested by Matthew.
Thanks,
Amir.
> ---
> mm/readahead.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/mm/readahead.c b/mm/readahead.c
> index e815c114de21..6925e6959fd3 100644
> --- a/mm/readahead.c
> +++ b/mm/readahead.c
> @@ -735,7 +735,8 @@ ssize_t ksys_readahead(int fd, loff_t offset, size_t count)
> */
> ret = -EINVAL;
> if (!f.file->f_mapping || !f.file->f_mapping->a_ops ||
> - !S_ISREG(file_inode(f.file)->i_mode))
> + (!S_ISREG(file_inode(f.file)->i_mode) &&
> + !S_ISBLK(file_inode(f.file)->i_mode)))
> goto out;
>
> ret = vfs_fadvise(f.file, offset, count, POSIX_FADV_WILLNEED);
> --
> 2.34.1
>
