cluster-devel.redhat.com archive mirror
 help / color / mirror / Atom feed
* [Cluster-devel] [PATCH] Fixing double brelse'ing bh allocated in gfs2_meta_read when EIO occurs
@ 2012-06-18  7:31 Masatake YAMATO
  2012-06-28 15:02 ` Steven Whitehouse
  0 siblings, 1 reply; 2+ messages in thread
From: Masatake YAMATO @ 2012-06-18  7:31 UTC (permalink / raw)
  To: cluster-devel.redhat.com

This patch fixes buffer_head double free in following code path:

gfs2_block_map
=> gfs2_meta_inode_buffer
 => gfs2_meta_indirect_buffer
  => gfs2_meta_read
=> release_metapath


gfs2_block_map calls gfs2_meta_inode_buffer with &mp.mp_bh[0]
as an argument. mp.mp_bh are filled with zero at the beginning
of gfs2_block_map.

If gfs2_meta_inode_buffer returns non-zero value, gfs2_block_map
calls release_metapath to free buffers chained to mp.mp_bh.
release_metapath checks each slot of mp.mp_bh[i] and 
free(with brelse) unless the slot is filled with NULL.


&mp.mp_bh[0] passed to gfs2_meta_inode_buffer is filled at
gfs2_meta_read. gfs2_meta_read is filled a buffer allocated with
gfs2_getbuf even if EIO occurs. When EIO occurs, the allocated buffer
is brelse'ed though the pointer(wrong poiner) points the brelse'ed is
passed back to caller via an argument bhp.

gfs2_meta_indirect_buffer, the caller also pass the wrong pointer
to its caller with EIO. Finally gfs2_block_map gets both EIO and
&mp.mp_bh[0] filled with the wrong pointer. release_metapath
calls brelse again on the wrong pointer.

Signed-off-by: Masatake YAMATO<yamato@redhat.com>

diff --git a/fs/gfs2/meta_io.c b/fs/gfs2/meta_io.c
index 6c1e5d1..3a56c8d 100644
--- a/fs/gfs2/meta_io.c
+++ b/fs/gfs2/meta_io.c
@@ -213,8 +213,10 @@ int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
 	struct gfs2_sbd *sdp = gl->gl_sbd;
 	struct buffer_head *bh;
 
-	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
+	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
+		*bhp = NULL;
 		return -EIO;
+	}
 
 	*bhp = bh = gfs2_getbuf(gl, blkno, CREATE);
 
@@ -235,6 +237,7 @@ int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
 		if (tr && tr->tr_touched)
 			gfs2_io_error_bh(sdp, bh);
 		brelse(bh);
+		*bhp = NULL;
 		return -EIO;
 	}
 



^ permalink raw reply related	[flat|nested] 2+ messages in thread

* [Cluster-devel] [PATCH] Fixing double brelse'ing bh allocated in gfs2_meta_read when EIO occurs
  2012-06-18  7:31 [Cluster-devel] [PATCH] Fixing double brelse'ing bh allocated in gfs2_meta_read when EIO occurs Masatake YAMATO
@ 2012-06-28 15:02 ` Steven Whitehouse
  0 siblings, 0 replies; 2+ messages in thread
From: Steven Whitehouse @ 2012-06-28 15:02 UTC (permalink / raw)
  To: cluster-devel.redhat.com

Hi,

Many thanks for the patch. I've added it to the -nmw git tree. Sorry for
the delay, just getting back up to speed after my holiday,

Steve.

On Mon, 2012-06-18 at 16:31 +0900, Masatake YAMATO wrote:
> This patch fixes buffer_head double free in following code path:
> 
> gfs2_block_map
> => gfs2_meta_inode_buffer
>  => gfs2_meta_indirect_buffer
>   => gfs2_meta_read
> => release_metapath
> 
> 
> gfs2_block_map calls gfs2_meta_inode_buffer with &mp.mp_bh[0]
> as an argument. mp.mp_bh are filled with zero at the beginning
> of gfs2_block_map.
> 
> If gfs2_meta_inode_buffer returns non-zero value, gfs2_block_map
> calls release_metapath to free buffers chained to mp.mp_bh.
> release_metapath checks each slot of mp.mp_bh[i] and 
> free(with brelse) unless the slot is filled with NULL.
> 
> 
> &mp.mp_bh[0] passed to gfs2_meta_inode_buffer is filled at
> gfs2_meta_read. gfs2_meta_read is filled a buffer allocated with
> gfs2_getbuf even if EIO occurs. When EIO occurs, the allocated buffer
> is brelse'ed though the pointer(wrong poiner) points the brelse'ed is
> passed back to caller via an argument bhp.
> 
> gfs2_meta_indirect_buffer, the caller also pass the wrong pointer
> to its caller with EIO. Finally gfs2_block_map gets both EIO and
> &mp.mp_bh[0] filled with the wrong pointer. release_metapath
> calls brelse again on the wrong pointer.
> 
> Signed-off-by: Masatake YAMATO<yamato@redhat.com>
> 
> diff --git a/fs/gfs2/meta_io.c b/fs/gfs2/meta_io.c
> index 6c1e5d1..3a56c8d 100644
> --- a/fs/gfs2/meta_io.c
> +++ b/fs/gfs2/meta_io.c
> @@ -213,8 +213,10 @@ int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
>  	struct gfs2_sbd *sdp = gl->gl_sbd;
>  	struct buffer_head *bh;
>  
> -	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
> +	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
> +		*bhp = NULL;
>  		return -EIO;
> +	}
>  
>  	*bhp = bh = gfs2_getbuf(gl, blkno, CREATE);
>  
> @@ -235,6 +237,7 @@ int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
>  		if (tr && tr->tr_touched)
>  			gfs2_io_error_bh(sdp, bh);
>  		brelse(bh);
> +		*bhp = NULL;
>  		return -EIO;
>  	}
>  
> 




^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2012-06-28 15:02 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-06-18  7:31 [Cluster-devel] [PATCH] Fixing double brelse'ing bh allocated in gfs2_meta_read when EIO occurs Masatake YAMATO
2012-06-28 15:02 ` Steven Whitehouse

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).