Hello all,
I have discovered a signed/unsigned issue in a routine responsible for
demarshalling XDR data for NFSv3 procedure calls. As far as I can tell,
this bug has existed since NFSv3 support was integrated. It has been
silently fixed in 2.4.21.
The bug is in the decode_fh routine of fs/nfsd/nfs3xdr.c under the kernel
source tree.
Vulnerable code:
static inline u32 *
decode_fh(u32 *p, struct svc_fh *fhp)
{
int size;
fh_init(fhp, NFS3_FHSIZE);
size = ntohl(*p++);
if (size > NFS3_FHSIZE)
return NULL;
memcpy(&fhp->fh_handle.fh_base, p, size);
fhp->fh_handle.fh_size = size;
return p + XDR_QUADLEN(size);
}
Where p is a packet of attacker controlled XDR data. If size is made to be
negative, the sanity check is passed and the malicious value is passed to
memcpy. Due to the behavior of the kernel's memcpy, this will cause a very
large copy in kernel space, resulting in an instant kernel panic.
The attached code is a POC of this vulnerability. It requires that the
vulnerable host has an exported directory available to the attacker. This
is probably not the only way to manifest this bug, however.
If you have any questions, please feel free to contact me.
Cheers,
Jared Stanbrough <jaredsat_private>
This archive was generated by hypermail 2b30 : Tue Jul 29 2003 - 13:11:38 PDT