Subject: kern/32928: bpf filter can fail to extract a 32-bit quantity
To: None <,,>
From: Pavel Cahyna <>
List: netbsd-bugs
Date: 02/25/2006 12:20:00
>Number:         32928
>Category:       kern
>Synopsis:       bpf filter can fail to extract a 32-bit quantity
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Feb 25 12:20:00 +0000 2006
>Originator:     Pavel Cahyna
>Release:        4.3BSD NET/2 - NetBSD 3.99.15
System: NetBSD beta 3.0_RC5 NetBSD 3.0_RC5 (EV56) #3: Mon Dec 12 20:28:20 CET 2005 pavel@beta:/usr/src/sys/arch/alpha/compile/EV56 alpha
Architecture: alpha
Machine: alpha
In net/bpf_filter.c there is a function m_xword() which extracts a
32-bit integer from a mbuf chain at a given offset k.

First, it determines where the integer starts:
	MINDEX(len, m, k);
m is now the first mbuf, len is its length and k is updated to the
offset of the data in the mbuf. Then we test if the whole integer is
in the mbuf:
	if (len >= k + 4) {
		*err = 0;
		return EXTRACT_LONG(cp);
if it is not, we go to the next mbuf:
	m0 = m->m_next;
	if (m0 == 0 || m0->m_len + len - k < 4)
		goto bad;

but the "goto bad" if "m0->m_len + len - k < 4" is wrong - the integer
could legally continue in a third mbuf. Of course removing this check
is not enough - some code must be added to handle this situation.
Code inspection.

This will be difficult to trigger (a 1- or 2- byte mbuf in the middle
of a mbuf chain is unlikely), but with functions like bpf_mtap2()
which prepend a short mbuf it could be possible.

I'm working on it. I propose to call m_xhalf() twice and concatenate
the results.