query: redundant tcp header length checks?

From: William Allen Simpson
Date: Sun Jan 10 2010 - 08:03:10 EST


On a related note, back on Nov 10th, Ilpo brought to my attention --
*hidden* inside the pskb_may_pull() -- the tcp header length is range
checked for being too short (skb->len < th->doff * 4).

tcp_v4_rcv()
...
1585 if (!pskb_may_pull(skb, th->doff * 4))
1586 goto discard_it;
1587

But the next comment in tcp_ipv4.c (missing from tcp_ipv6.c) says:

1588 /* An explanation is required here, I think.
1589 * Packet length and doff are validated by header prediction,
1590 * provided case of th->doff==0 is eliminated.
1591 * So, we defer the checks. */

That's not correct anymore, as the checks aren't deferred! But the
redundant deferred checks _are_ still present in tcp_input.c:

tcp_rcv_established()
...
5210 if (len <= tcp_header_len) {
5211 /* Bulk data transfer: sender */
5212 if (len == tcp_header_len) {
...
5229 } else { /* Header too small */
5230 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
5231 goto discard;
5232 }

Also:

5325 slow_path:
5326 if (len < (th->doff << 2) || tcp_checksum_complete_user(sk, skb))
5327 goto csum_error;

Apparently tcp_rcv_established() might be called through some other path
that requires this extra header length checking?

Its callers also have another reference, for example in tcp_ipv4.c:

2432 .backlog_rcv = tcp_v4_do_rcv,

And tcp_ipv6.c:

213 sk->sk_backlog_rcv = tcp_v4_do_rcv;

223 sk->sk_backlog_rcv = tcp_v6_do_rcv;

1302 newsk->sk_backlog_rcv = tcp_v4_do_rcv;

2073 .backlog_rcv = tcp_v6_do_rcv,

Can anybody explain these?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/