On Tuesday 16 October 2007 08:42, Eric W. Biederman wrote:
quoted text
> I have not observed this case but it is possible to get a dirty page
> cache with clean buffer heads if we get a clean ramdisk page with
> buffer heads generated by a filesystem calling __getblk and then write
> to that page from user space through the block device.  Then we just
> need to hit the proper window and try_to_free_buffers() will mark that
> page clean and eventually drop it.  Ouch!
>
> To fix this use the generic __set_page_dirty_buffers in the ramdisk
> code so that when we mark a page dirty we also mark it's buffer heads
> dirty.

Hmm, so we can also have some filesystems writing their own buffers
out by hand (clear_buffer_dirty, submit buffer for IO). Other places
will do similarly dodgy things with filesystem metadata
(fsync_buffers_list, for example).

So your buffers get cleaned again, then your pages get cleaned.

While I said it was a good fix when I saw the patch earlier, I think
it's not closing the entire hole, and as such, Christian's patch is
probably the way to go for stable.

For mainline, *if* we want to keep the old rd.c around at all, I
don't see any harm in this patch so long as Christian's is merged
as well. Sharing common code is always good.

quoted text
>
> Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
> ---
>  drivers/block/rd.c |   13 +------------
>  1 files changed, 1 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/block/rd.c b/drivers/block/rd.c
> index 701ea77..84163da 100644
> --- a/drivers/block/rd.c
> +++ b/drivers/block/rd.c
> @@ -178,23 +178,12 @@ static int ramdisk_writepages(struct address_space
> *mapping, return 0;
>  }
>
> -/*
> - * ramdisk blockdev pages have their own ->set_page_dirty() because we
> don't - * want them to contribute to dirty memory accounting.
> - */
> -static int ramdisk_set_page_dirty(struct page *page)
> -{
> -	if (!TestSetPageDirty(page))
> -		return 1;
> -	return 0;
> -}
> -
>  static const struct address_space_operations ramdisk_aops = {
>  	.readpage	= ramdisk_readpage,
>  	.prepare_write	= ramdisk_prepare_write,
>  	.commit_write	= ramdisk_commit_write,
>  	.writepage	= ramdisk_writepage,
> -	.set_page_dirty	= ramdisk_set_page_dirty,
> +	.set_page_dirty	= __set_page_dirty_buffers,
>  	.writepages	= ramdisk_writepages,
>  };
-
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/