Re: [dpdk-stable] [PATCH v2] vhost: add header check in dequeue offload

[David Marchand <david.marchand@redhat.com>]

On Mon, Mar 15, 2021 at 4:52 PM Xiao Wang <xiao.w.wang@intel.com> wrote:
>
> When parsing the virtio net header and packet header for dequeue offload,
> we need to perform sanity check on the packet header to ensure:
>   - No out-of-boundary memory access.
>   - The packet header and virtio_net header are valid and aligned.
>
> Fixes: d0cf91303d73 ("vhost: add Tx offload capabilities")
> Cc: stable@dpdk.org
>
> Signed-off-by: Xiao Wang <xiao.w.wang@intel.com>
> ---
> v2:
> Allow empty L4 payload for cksum offload.
> ---
>  lib/librte_vhost/virtio_net.c | 49 +++++++++++++++++++++++++++++++++++++------
>  1 file changed, 43 insertions(+), 6 deletions(-)
>
> diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
> index 583bf379c6..53a8ff2898 100644
> --- a/lib/librte_vhost/virtio_net.c
> +++ b/lib/librte_vhost/virtio_net.c
> @@ -1821,44 +1821,64 @@ virtio_net_with_host_offload(struct virtio_net *dev)
>         return false;
>  }
>
> -static void
> -parse_ethernet(struct rte_mbuf *m, uint16_t *l4_proto, void **l4_hdr)
> +static int
> +parse_ethernet(struct rte_mbuf *m, uint16_t *l4_proto, void **l4_hdr,
> +               uint16_t *len)
>  {
>         struct rte_ipv4_hdr *ipv4_hdr;
>         struct rte_ipv6_hdr *ipv6_hdr;
>         void *l3_hdr = NULL;
>         struct rte_ether_hdr *eth_hdr;
>         uint16_t ethertype;
> +       uint16_t data_len = m->data_len;

>
>         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
>
> +       if (data_len <= sizeof(struct rte_ether_hdr))
> +               return -EINVAL;

On principle, the check should happen before calling rte_pktmbuf_mtod,
like what rte_pktmbuf_read does.

Looking at the rest of the patch, does this helper function only
handle mono segment mbufs?
My reading of copy_desc_to_mbuf() was that it could generate multi
segments mbufs...


[snip]

>         case RTE_ETHER_TYPE_IPV4:
> +               if (data_len <= sizeof(struct rte_ipv4_hdr))
> +                       return -EINVAL;
>                 ipv4_hdr = l3_hdr;
>                 *l4_proto = ipv4_hdr->next_proto_id;
>                 m->l3_len = rte_ipv4_hdr_len(ipv4_hdr);
> +               if (data_len <= m->l3_len) {
> +                       m->l3_len = 0;
> +                       return -EINVAL;
> +               }

... so here, comparing l3 length to only the first segment length
(data_len) would be invalid.

If this helper must deal with multi segments, why not use rte_pktmbuf_read?
This function returns access to mbuf data after checking offset and
length are contiguous, else copy the needed data in a passed buffer.


>                 *l4_hdr = (char *)l3_hdr + m->l3_len;
>                 m->ol_flags |= PKT_TX_IPV4;
> +               data_len -= m->l3_len;
>                 break;


-- 
David Marchand