Re: [PATCH] Fix corrupt TCP packets when options space overflows with MD5SIG enabled

["Adam Langley" <agl@imperialviolet.org>]

On Sun, Jun 1, 2008 at 4:40 PM, James Morris <jmorris@namei.org> wrote:
> Reviewed-by: James Morris <jmorris@namei.org>


Looking at this code some more, I fear that I've fucked up that patch.
The logic for TCP options seems to be duplicated three times in the
code: the size calculation (patched), the building of the actual
options (tcp_build_and_update_options / tcp_syn_build_options) and the
MSS calculations (tcp_current_mss).

It looks, on second glance, that this code in
tcp_build_and_update_options will include the options even though we
calculated the size without:

if (tp->rx_opt.eff_sacks) {
	struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack :
tp->selective_acks;
	int this_sack;

	*ptr++ = htonl((TCPOPT_NOP  << 24) |
		       (TCPOPT_NOP  << 16) |
		       (TCPOPT_SACK <<  8) |
		       (TCPOLEN_SACK_BASE + (tp->rx_opt.eff_sacks *
			     TCPOLEN_SACK_PERBLOCK)));
...

Unless I'm missing something, that patch was incomplete and we're
still sending invalid packets on in the MD5SIG + SACK case.

If so, I'll try and get the other two cases with another patch.

Additionally, it would appear that it would be useful to pull this
logic into a single place: maybe a function which runs multiple times
(to calculate the MSS / header size and a second time to actually
perform the options writes). But that'll be a different patch.

Cheers,

AGL

-- 
Adam Langley agl@imperialviolet.org http://www.imperialviolet.org