BSD: IPv4 forwarding doesn't consult inbound SPD in KAME-derived IPsec

[Greg Troxel <gdt () ir bbn com>]

  IPv4 forwarding doesn't consult inbound SPD in KAME-derived IPsec

                     Greg Troxel <gdt () ir bbn com>
              Bill Chiarchiaro <wjc () work cleartech com>
                              2002-02-24

SUMMARY

NetBSD 1.5.2 and -current, FreeBSD 4.5 and -current, and the KAME
versions of NetBSD and FreeBSD fail to perform inbound policy checks
on packets which are forwarded, in violation of RFC2401 sections 4.4.1
and 5.2.1 (steps 3 and 4).  NetBSD, FreeBSD, and KAME have already
committed changes to address this problem.

The typical case where one might care is at a Security Gateway where
packets from a peer gateway are required to be ESP in order to provide
confidentiality and integrity.  If the SG peer fails to use ESP,
confidentiality is lost.  If the local SG follows RFC2401 and
performs inbound SPD checking, the packet will be determined to
require ESP and be dropped.  However, with the present code, no SPD
check is performed, and the packets -- possibly forged -- will be
forwarded, presumably to hosts relying on the SG.

DETAILS

The defect was observed in attempting to control network access from
an 802.11 LAN by requiring tunnel-mode ESP of all traffic from the
wireless host, with tunnel endpoints of the host and an SG with a
wired interface and an 802.11 interface.  We observed that non-ESP
packets (emitted due to configuration errors) were forwarded by the
SG; this is a real failure that caused a security policy to be
violated, although in a situation with no adverse consequences.
The policy called for ESP for confidentiality over 802.11 and access
control to the wired network and resulting Internet connection.
With the defect, an attacker could have injected packets into the
Internet, but not received packets because the outbound SPD check
on the SG functioned correctly.

Examine src/sys/netinet/ip_input.c, and follow the code from the call
to ip_forward().  Note that inbound SPD check is done only for packets
for this host, and only for protocols with PR_LASTHDR; the check is
deferred to protocol implementation to enable per-socket policy.

Patches that fix (at least in the normal case) the problem for IPv4
are enclosed for fairly recent points along NetBSD's netbsd-1-5 branch
and FreeBSD's RELENG_4 branch.  The KAME versions of ip_input.c are
very similar.  These patches are probably not sufficient; the check
should probably be in ip_forward (aligned with the IPv6
implementation).  A more careful analysis is needed to consider all
the code paths; we have not considered interaction of inbound SPD
checks with tunneling (e.g. gif) and other features.  In particular,
forwarding of packets via source routing options seems to be handled
incorrectly.

OTHER SYSTEMS

FreeBSD RELENG_4 and NetBSD netbsd-1-5 appear to perform this check
for IPv6 in ip6_forward; KAME has had the relevant code since
1999-09-01.

OpenBSD began to perform this check as of revision 1.58 of
src/sys/netinet/in_input.c on 2000-09-01; we have not tested OpenBSD.

DETAILS

We give details for a simple setup with alice and bob on two networks
connected by two SGs alice-gw and bob-gw, nominally intended to use
tunnel-mode ESP for all traffic between the networks.  (We present
results from an actual experiment, but with the IP addresses obscured
and irrelevant traffic removed.)

10.1.1.48/28 is Alice's network, and 10.1.1.64/28 is Bob's (both red).
10.1.1.32/28 is the black network connecting Alice (.34) and Bob (.35).

alice-gw SPD

  # to/from bob
  spdadd 10.1.1.48/28[any] 10.1.1.64/28[any] any
          -P out ipsec esp/tunnel/10.1.1.34-10.1.1.35/require ;
  spdadd 10.1.1.64/28[any] 10.1.1.48/28[any] any
          -P in ipsec esp/tunnel/10.1.1.35-10.1.1.34/require ;

bob-gw SPD

  # to/from alice
  # intentional misconfiguration for testing - no ESP to alice
  spdadd 10.1.1.64/28[any] 10.1.1.48/28[any] any
          -P out none ;
  #spdadd 10.1.1.64/28[any] 10.1.1.48/28[any] any
  #        -P out ipsec esp/tunnel/10.1.1.35-10.1.1.34/require ;
  spdadd 10.1.1.48/28[any] 10.1.1.64/28[any] any
          -P in ipsec esp/tunnel/10.1.1.34-10.1.1.35/require ;

EXAMPLE WITH THE INCORRECT CODE

With NetBSD along netbsd-1-5 as of 2001-11-26, examining the black
network at alice-gw, we see a cleartext SYN, followed by alice-gw
initiating IKE due to the response from alice, followed by a
retransmitted SYN and the reset in ESP:

09:44:45.258912 bob.65514 > alice.8001: S 406012389:406012389(0) win 16384 <mss 1460,nop,wscale 0,nop,nop,timestamp 
2214891 0>
09:44:45.260843 alice-gw.isakmp > bob-gw.isakmp: isakmp: phase 1 I ident:
    (sa: doi=ipsec situation=identity
        (p: #1 protoid=isakmp transform=1
            (t: #1 id=ike (type=lifetype value=sec)(type=lifeduration value=0258)(type=enc value=3des)(type=auth 
value=rsa sig)(type=hash value=sha1)(type=group desc value=modp1024))))
09:44:45.262261 bob-gw.isakmp > alice-gw.isakmp: isakmp: phase 1 R ident:
    (sa: doi=ipsec situation=identity
        (p: #1 protoid=isakmp transform=1
            (t: #1 id=ike (type=lifetype value=sec)(type=lifeduration value=0258)(type=enc value=3des)(type=auth 
value=rsa sig)(type=hash value=sha1)(type=group desc value=modp1024))))
    (vid: len=16)
09:44:45.345226 alice-gw.isakmp > bob-gw.isakmp: isakmp: phase 1 I ident:
    (ke: key len=128)
    (nonce: n len=16)
    (vid: len=16)
09:44:45.426774 bob-gw.isakmp > alice-gw.isakmp: isakmp: phase 1 R ident:
    (ke: key len=128)
    (nonce: n len=16)
    (vid: len=16)
    (cr: len=1 type=x509sign)
09:44:45.545348 alice-gw.isakmp > bob-gw.isakmp: isakmp: phase 1 I ident[E]: [|id]
09:44:45.576906 bob-gw.isakmp > alice-gw.isakmp: isakmp: phase 1 R ident[E]: [|id]
09:44:45.577171 bob-gw.isakmp > alice-gw.isakmp: isakmp: phase 2/others R inf[E]: [|hash]
09:44:45.582749 alice-gw.isakmp > bob-gw.isakmp: isakmp: phase 2/others I inf[E]: [|hash]
09:44:46.588822 alice-gw.isakmp > bob-gw.isakmp: isakmp: phase 2/others I oakley-quick[E]: [|hash]
09:44:46.590693 bob-gw.isakmp > alice-gw.isakmp: isakmp: phase 2/others R oakley-quick[E]: [|hash]
09:44:46.591449 alice-gw.isakmp > bob-gw.isakmp: isakmp: phase 2/others I oakley-quick[E]: [|hash]
09:44:51.217057 bob.65514 > alice.8001: S 406012389:406012389(0) win 16384 <mss 1460,nop,wscale 0,nop,nop,timestamp 
2214902 0>
09:44:51.217686 alice-gw > bob-gw: ESP(spi=29712172,seq=0x1)

On the Alice red network we see the SYNs and the response (a TCP
reset):

09:44:45.258949 bob.65514 > alice.8001: S 406012389:406012389(0) win 16384 <mss 1460,nop,wscale 0,nop,nop,timestamp 
2214891 0>
09:44:45.259415 alice.8001 > bob.65514: R 0:0(0) ack 406012390 win 0
09:44:51.217090 bob.65514 > alice.8001: S 406012389:406012389(0) win 16384 <mss 1460,nop,wscale 0,nop,nop,timestamp 
2214902 0>
09:44:51.217540 alice.8001 > bob.65514: R 0:0(0) ack 1 win 0

EXAMPLE WITH THE MODIFIED CODE

Here we see the black network observed at alice-gw.  One NTP exchange
has been left in, since otherwise the red tcpdump would be empty.
(The security policy does not require the observed NTP traffic to be
in ESP.)

09:58:25.886014 bob.65513 > alice.8001: S 2189637913:2189637913(0) win 16384 <mss 1460,nop,wscale 0,nop,nop,timestamp 
2216532 0>
09:58:31.664971 bob.65513 > alice.8001: S 2189637913:2189637913(0) win 16384 <mss 1460,nop,wscale 0,nop,nop,timestamp 
2216543 0>
09:58:43.664091 bob.65513 > alice.8001: S 2189637913:2189637913(0) win 16384 <mss 1460,nop,wscale 0,nop,nop,timestamp 
2216567 0>
10:00:06.991381 alice.ntp > tai.ntp:  v4 client strat 3 poll 6 prec -17 [tos 0x10]
10:00:06.991559 tai.ntp > alice.ntp:  v4 server strat 3 poll 6 prec -18 [tos 0x10]
09:59:07.662368 bob.65513 > alice.8001: S 2189637913:2189637913(0) win 16384 <mss 1460,nop,wscale 0,nop,nop,timestamp 
2216615 0>

Here we see the red network observed at alice-gw:

10:00:06.991356 alice.ntp > tai.ntp:  v4 client strat 3 poll 6 prec -17 [tos 0x10]
10:00:06.991579 tai.ntp > alice.ntp:  v4 server strat 3 poll 6 prec -18 [tos 0x10]

With the modified code, we captured output of "netstat -p ipsec -s" on
alice-gw immediately before and immediately after sending TCP SYNs.

  ipsec:
          0 inbound packets processed successfully
          0 inbound packets violated process security policy
          0 inbound packets with no SA available
          0 invalid inbound packets
          0 inbound packets failed due to insufficient memory
          0 inbound packets failed getting SPI
          0 inbound packets failed on AH replay check
          0 inbound packets failed on ESP replay check
          0 inbound packets considered authentic
          0 inbound packets failed on authentication
          0 outbound packets processed successfully
          0 outbound packets violated process security policy
          0 outbound packets with no SA available
          0 invalid outbound packets
          0 outbound packets failed due to insufficient memory
          0 outbound packets with no route

  ipsec:
          0 inbound packets processed successfully
          4 inbound packets violated process security policy
          0 inbound packets with no SA available
          0 invalid inbound packets
          0 inbound packets failed due to insufficient memory
          0 inbound packets failed getting SPI
          0 inbound packets failed on AH replay check
          0 inbound packets failed on ESP replay check
          0 inbound packets considered authentic
          0 inbound packets failed on authentication
          0 outbound packets processed successfully
          0 outbound packets violated process security policy
          0 outbound packets with no SA available
          0 invalid outbound packets
          0 outbound packets failed due to insufficient memory
          0 outbound packets with no route



MODIFICATIONS

For NetBSD netbsd-1-5:

Index: src/sys/netinet/ip_input.c
===================================================================
RCS file: /NETBSD-CVS/netbsd/src/sys/netinet/ip_input.c,v
retrieving revision 1.1.1.1
diff -u -r1.1.1.1 ip_input.c
--- src/sys/netinet/ip_input.c  2001/07/05 14:42:54     1.1.1.1
+++ src/sys/netinet/ip_input.c  2002/02/25 01:07:02
@@ -611,6 +611,15 @@
                ipstat.ips_cantforward++;
                m_freem(m);
        } else {
+#ifdef IPSEC
+               /*
+                * Enforce inbound IPsec SPD.
+                */
+               if (ipsec4_in_reject(m, NULL)) {
+                       ipsecstat.in_polvio++;
+                       goto bad;
+               }
+#endif
                /*
                 * If ip_dst matched any of my address on !IFF_UP interface,
                 * and there's no IFF_UP interface that matches ip_dst,

For FreeBSD RELENG_4:

Index: src/sys/netinet/ip_input.c
===================================================================
RCS file: /FREEBSD-CVS/src/sys/netinet/ip_input.c,v
retrieving revision 1.130.2.31
diff -u -r1.130.2.31 ip_input.c
--- src/sys/netinet/ip_input.c  2001/12/15 01:06:27     1.130.2.31
+++ src/sys/netinet/ip_input.c  2002/02/24 16:10:26
@@ -625,8 +625,18 @@
        if (ipforwarding == 0) {
                ipstat.ips_cantforward++;
                m_freem(m);
-       } else
+       } else {
+#ifdef IPSEC
+                /*
+                 * Enforce inbound IPsec SPD.
+                 */
+                if (ipsec4_in_reject(m, NULL)) {
+                        ipsecstat.in_polvio++;
+                       goto bad;
+                }
+#endif /* IPSEC */
                ip_forward(m, 0);
+       }
 #ifdef IPFIREWALL_FORWARD
        ip_fw_fwd_addr = NULL;
 #endif

VENDOR STATUS

This advisory was provided to KAME, NetBSD and FreeBSD before public
posting.  Our modifications above, or the equivalent, have already
been incorporated into sys/netinet/ip_input.c in KAME's CVS repository
(for NetBSD, 1.37 and for FreeBSD-4, 1.33), NetBSD -current (1.145)
and 1.5-stable (1.114.4.8), and FreeBSD -current (1.192) and -stable
(1.130.2.35).