Linux ipchains Firewall Vulnerability

data protect GmbH - Advisory #2 July 27, 1999

Authors: Thomas Lopatic <tl@dataprotect.com>
         John McDonald  <jm@dataprotect.com>

Overview
--------

data protect has discovered a potential vulnerability in the Linux ipchains firewall implementation. In certain situations, it is possible for an attacker to bypass the packet filter when communicating with machines that allow incoming packets to specific ports. This attack is a variation of previously discussed fragmentation attacks, where the attacker uses fragments to rewrite parts of the TCP or UDP protocol header. In this case port information is rewritten in order to gain access to ports that should be blocked by the firewall.

Included in this advisory is a patch to the 2.2.10 Linux kernel that corrects this vulnerability, and a pointer to example code that demonstrates the problem.

Problem Description
-------------------

The Linux ipchains firewall code has special provisions for IP fragments that do not contain enough information for transport protocol header analysis. Fragments that start at offset 0, and are not long enough to provide complete transport header information are treated like fragments with an offset > 0 (> 1 in the TCP case). This is the relevant code from ip_fw.c:

        if (offset == 0) {
                unsigned int size_req;
                switch (ip->protocol) {
                case IPPROTO_TCP:
                        /* Don't care about things past flags word */
                        size_req = 16;
                        break;

                case IPPROTO_UDP:
                case IPPROTO_ICMP:
                        size_req = 8;
                        break;

                default:
                        size_req = 0;
                }
                offset = (ntohs(ip->tot_len) < (ip->ihl<<2)+size_req);
        }

As mentioned above, fragments with an offset of 0, that are too short to provide a full transport protocol header, are treated like non-first fragments. This allows an attacker to perform the following port rewriting attack:

1. Attacker sends a fragment, with offset 0, a set IP_MF bit, and a full transport protocol header which meets the packet filter and is passed to the victim machine.
2. Attacker sends a fragment, with offset 0, a set IP_MF bit, and a length of 4 bytes. This contains the (blocked) ports that the attacker wishes to access on the victim machine. This fragment will be accepted by the firewall and overlap - in the victim machine's reassembly chain - the port information contained in the fragment sent in step 1.
3. Attacker sends a fragment with a cleared IP_MF bit, starting where the first fragment left off, that completes the set of fragments.

Depending on the defragmentation strategy of the victim machine's operating system, it might be necessary to swap steps 1 and 2.

It is important to note that there are two conditions that must be met for a particular ipchains packet filter to be vulnerable:

1. The packet filter must not be configured with the Linux kernel option CONFIG_IP_ALWAYS_DEFRAG. If the packet filter reassembles the fragments before doing the firewall checks, then this attack will fail.
2. The packet filter must have a rule to allow non-first fragments to pass. The Linux ipchains how-to suggests that either an administrator selects CONFIG_IP_ALWAYS_DEFRAG, or implements such a rule. This rule was considered to be safe because fragments with an offset of 1 are blocked by the packet filter, which prevents attacks based on rewriting the TCP flags.

Fix Information
---------------

The following Linux kernel patch (against version 2.2.10) will close this vulnerability by blocking packets that could be used to rewrite header information in this fashion.

It is also possible to reconfigure the ipchains machine to always defragment packets, or to remove any rule which passes non-first IP fragments through the firewall ("-f" option of the "ipchains" command). The latter, however, might introduce incompatibilities, e.g. with applications that transmit large UDP datagrams across the firewall and hence cause IP fragmentation.

*** linux.old/net/ipv4/ip_fw.c  Wed Jun  9 05:33:07 1999
--- linux/net/ipv4/ip_fw.c      Fri Jul 23 19:20:45 1999
***************
*** 37,42 ****
--- 37,45 ----
   * 19-May-1999: Star Wars: The Phantom Menace opened.  Rule num
   *            printed in log (modified from Michael Hasenstein's patch).
   *            Added SYN in log message. --RR
+  * 23-Jul-1999: Fixed small fragment security exposure opened on 15-May-1998.
+  *              John McDonald <jm@dataprotect.com>
+  *              Thomas Lopatic <tl@dataprotect.com>
   */

  /*
***************
*** 644,650 ****
                default:
                        size_req = 0;
                }
!               offset = (ntohs(ip->tot_len) < (ip->ihl<<2)+size_req);
        }

        src = ip->saddr;
--- 647,666 ----
                default:
                        size_req = 0;
                }
!
!               /* If it is a truncated first fragment then it can be
!                * used to rewrite port information, and thus should
!                * be blocked.
!                */
!
!               if (ntohs(ip->tot_len) < (ip->ihl<<2)+size_req)
!               {
!                       if (!testing && net_ratelimit()) {
!                               printk("Suspect short first fragment.\n");
!                               dump_packet(ip,rif,NULL,NULL,0,0,0,0);
!                       }
!                       return FW_BLOCK;
!               }
        }

        src = ip->saddr;

Demonstration Code
------------------

fragrouter, a component of Nidsbench, has been updated to perform this attack transparently. This is an excellent open source tool for testing intrusion detection systems and packet filters provided by Anzen Computing. The version of fragrouter that performs this attack should be available shortly, at http://www.anzen.com/research/nidsbench/.

Additional Information
----------------------

data protect would like to thank Dug Song <dugsong@anzen.com> for his help in implementing this attack.

For information regarding this advisory, please contact
Thomas Lopatic <tl@dataprotect.com> or John McDonald <jm@dataprotect.com>.

The contents of this advisory are Copyright (C) 1999 data protect GmbH, and may be distributed freely provided that no fee is charged for distribution, and that proper credit is given.