[Cluster-devel] [libgfs2 PATCH] libgfs2: Check block range when inserting into rgrp tree

[Andrew Price]

On 14/09/15 18:19, Bob Peterson wrote:
> Hi,
>
> This patch adds checks to function rindex_read to make sure the
> rgrp starting address isn't grossly outside the file system.
> It may be in the case of severely corrupt file systems from fsck.
> If we added them to the rgrp tree, our calculations will get
> screwed up, eventually causing a segfault.
>
> Regards,
>
> Bob Peterson
> Red Hat File Systems
>
> Signed-off-by: Bob Peterson <rpeterso at redhat.com>
> ---
> diff --git a/gfs2/libgfs2/super.c b/gfs2/libgfs2/super.c
> index b956366..73354ff 100644
> --- a/gfs2/libgfs2/super.c
> +++ b/gfs2/libgfs2/super.c
> @@ -166,6 +166,12 @@ int rindex_read(struct gfs2_sbd *sdp, int fd, int *count1, int *sane)
>   			return -1;
>
>   		gfs2_rindex_in(&ri, (char *)&buf.bufgfs2);
> +		if (gfs2_check_range(sdp, ri.ri_addr) != 0) {

I guess since we don't have the full rindex yet to get the fs size it's 
checking against the device size here. Assuming other tools using this 
function have initialized sdp->fssize sensibly before calling it: ACK.

Andy

> +			*sane = 0;
> +			if (prev_rgd == NULL)
> +				return -1;
> +			ri.ri_addr = prev_rgd->ri.ri_addr + prev_rgd->length;
> +		}
>   		rgd = rgrp_insert(&sdp->rgtree, ri.ri_addr);
>   		memcpy(&rgd->ri, &ri, sizeof(struct gfs2_rindex));
>
>