Re: [PATCH] Possible fix for 2.6.24 xfs_file_readdir crash

[Tobias Ulmer <tu@emlix.com>]

On Tue, Feb 05, 2008 at 04:24:18PM +1100, David Chinner wrote:
> Sven, Tomas, Andrea:
> 
> Can you try the patch attached below to see if it fixes the
> xfs_file_readdir() oops you are seeing and let me know if it fixes
> the problem?

Works for me(TM) :)

My testbox survived 24h with this patch, no problems.

Tobias

> 
> It looks like we're deferencing a pointer beyond the end of a buffer
> if the buffer is filled exactly. This bug does not crash ia64 (even
> with memory poisoning enabled), which is why the targeted corner
> case testing I did a while back did not pick this up when fixing a
> similar bug a month ago.
> 
> Cheers,
> 
> Dave.
> -- 
> Dave Chinner
> Principal Engineer
> SGI Australian Software Group
> 
> ---
> Fix yet another corner case oops in xfs_file_readdir().
> 
> Signed-off-by: Dave Chinner <dgc@sgi.com>
> ---
>  fs/xfs/linux-2.6/xfs_file.c |    3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
> 
> Index: 2.6.x-xfs-new/fs/xfs/linux-2.6/xfs_file.c
> ===================================================================
> --- 2.6.x-xfs-new.orig/fs/xfs/linux-2.6/xfs_file.c	2008-01-16 16:24:01.000000000 +1100
> +++ 2.6.x-xfs-new/fs/xfs/linux-2.6/xfs_file.c	2008-02-05 15:13:17.153110696 +1100
> @@ -351,8 +351,8 @@ xfs_file_readdir(
>  
>  		size = buf.used;
>  		de = (struct hack_dirent *)buf.dirent;
> -		curr_offset = de->offset /* & 0x7fffffff */;
>  		while (size > 0) {
> +			curr_offset = de->offset /* & 0x7fffffff */;
>  			if (filldir(dirent, de->name, de->namlen,
>  					curr_offset & 0x7fffffff,
>  					de->ino, de->d_type)) {
> @@ -363,7 +363,6 @@ xfs_file_readdir(
>  				       sizeof(u64));
>  			size -= reclen;
>  			de = (struct hack_dirent *)((char *)de + reclen);
> -			curr_offset = de->offset /* & 0x7fffffff */;
>  		}
>  	}
>