Re: [regression, 4.14] xfrm: Fix stack-out-of-bounds read in xfrm_state_find breaks selinux-testsuite

[Steffen Klassert <steffen.klassert@secunet.com>]

On Tue, Nov 14, 2017 at 03:46:30PM -0500, Stephen Smalley wrote:
> Hi,
> 
> 4.14 is failing the selinux-testsuite labeled IPSEC tests despite
> having just been fixed in commit cf37966751747727 ("xfrm: do
> unconditional template resolution before pcpu cache check").  The
> breaking commit is the very next one, commit c9f3f813d462c72d ("xfrm:
> Fix stack-out-of-bounds read in xfrm_state_find.").  Unlike the earlier
> breakage, which caused use of the wrong SA, this one leads to a failure
> on connect(). Running ip xfrm monitor during one of the failing tests
> shows the following:
> acquire proto ah 
>   sel src 127.0.0.1/32 dst 127.0.0.1/32 proto tcp sport 0 dport 65535
> dev lo 
>   policy src 127.0.0.1/32 dst 127.0.0.1/32 proto tcp 
>         security context
> unconfined_u:unconfined_r:test_inet_client_t:s0-s0:c0.c1023 
>         dir out priority 0 ptype main 
>         tmpl src 0.0.0.0 dst 0.0.0.0
>                 proto ah reqid 0 mode transport

Yes, I see. This is because there are wildcard src and dst
addresses on the template. I'll revert this one for now.

I slowly start to think that the concept of having a socket
policy on a IPv6 socket that maps to IPv4 is fundamentally
broken. The bug I tried to fix here is not the first one
that were reported from syzkaller for this szenario and I
fear it is not the last one.

Thanks for reporting this!