[v5,17/24] ssi-sd: fix buffer overrun on invalid state load

CVE-2013-4537

s->arglen is taken from wire and used as idx
in ssi_sd_transfer().

Validate it before access.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---
 hw/sd/ssi-sd.c | 8 ++++++++
 1 file changed, 8 insertions(+)

On 3 April 2014 17:52, Michael S. Tsirkin <mst@redhat.com> wrote:
> CVE-2013-4537
>
> s->arglen is taken from wire and used as idx
> in ssi_sd_transfer().
>
> Validate it before access.
>
> Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
> ---
>  hw/sd/ssi-sd.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
>
> diff --git a/hw/sd/ssi-sd.c b/hw/sd/ssi-sd.c
> index 3273c8a..2fa2b2b 100644
> --- a/hw/sd/ssi-sd.c
> +++ b/hw/sd/ssi-sd.c
> @@ -230,6 +230,14 @@ static int ssi_sd_load(QEMUFile *f, void *opaque, int version_id)
>      for (i = 0; i < 5; i++)
>          s->response[i] = qemu_get_be32(f);
>      s->arglen = qemu_get_be32(f);
> +    if (s->mode == SSI_SD_CMDARG &&
> +        (s->arglen < 0 || s->arglen > ARRAY_SIZE(s->cmdarg))) {
> +        return -EINVAL;
> +    }
> +    if (s->mode == SSI_SD_RESPONSE &&
> +        (s->response_pos < 0 || s->response_pos > ARRAY_SIZE(s->response))) {
> +        return -EINVAL;
> +    }
>      s->response_pos = qemu_get_be32(f);
>      s->stopping = qemu_get_be32(f);

Surely we should read s->response_pos off the wire before
validating it rather than after?

Also, your checks on arglen aren't sufficient. Consider
the case where the attacker:
 * sets mode to SSI_SD_RESPONSE
 * sets arglen to something huge
 * sets response_pos to 0
 * gets the guest to repeatedly provoke calls to ssi_sd_transfer

We'll happily read off the end of the s->response[] buffer,
because our "when do we stop returning response bytes" check
is "s->response_pos >= s->arglen".

thanks
-- PMM

On Thu, Apr 03, 2014 at 06:05:03PM +0100, Peter Maydell wrote:
> On 3 April 2014 17:52, Michael S. Tsirkin <mst@redhat.com> wrote:
> > CVE-2013-4537
> >
> > s->arglen is taken from wire and used as idx
> > in ssi_sd_transfer().
> >
> > Validate it before access.
> >
> > Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
> > ---
> >  hw/sd/ssi-sd.c | 8 ++++++++
> >  1 file changed, 8 insertions(+)
> >
> > diff --git a/hw/sd/ssi-sd.c b/hw/sd/ssi-sd.c
> > index 3273c8a..2fa2b2b 100644
> > --- a/hw/sd/ssi-sd.c
> > +++ b/hw/sd/ssi-sd.c
> > @@ -230,6 +230,14 @@ static int ssi_sd_load(QEMUFile *f, void *opaque, int version_id)
> >      for (i = 0; i < 5; i++)
> >          s->response[i] = qemu_get_be32(f);
> >      s->arglen = qemu_get_be32(f);
> > +    if (s->mode == SSI_SD_CMDARG &&
> > +        (s->arglen < 0 || s->arglen > ARRAY_SIZE(s->cmdarg))) {
> > +        return -EINVAL;
> > +    }
> > +    if (s->mode == SSI_SD_RESPONSE &&
> > +        (s->response_pos < 0 || s->response_pos > ARRAY_SIZE(s->response))) {

And actually it should be s->response_pos >= ARRAY_SIZE(s->response),
right?

> > +        return -EINVAL;
> > +    }
> >      s->response_pos = qemu_get_be32(f);
> >      s->stopping = qemu_get_be32(f);
> 
> Surely we should read s->response_pos off the wire before
> validating it rather than after?
> 
> Also, your checks on arglen aren't sufficient. Consider
> the case where the attacker:
>  * sets mode to SSI_SD_RESPONSE
>  * sets arglen to something huge
>  * sets response_pos to 0
>  * gets the guest to repeatedly provoke calls to ssi_sd_transfer
> 
> We'll happily read off the end of the s->response[] buffer,
> because our "when do we stop returning response bytes" check
> is "s->response_pos >= s->arglen".
> 
> thanks
> -- PMM