Sequence Number Attacks By Rik Farrow Kevin Mitnick's alleged attack on Tsutomu Shimomura's Computers used a vulnerability in TCP/IP and mistaken trust. Questions regarding this article should be directed to the author at rik@spirit.com. December 25, 1994 found Tsutomu Shimomura, a computational physicist for the San Diego Supercomputer Center, on his way to the Sierra Nevadas to go skiing. He had left his personal network of computers running at his beach cottage in Del Mar, just north of San Diego. Perhaps it is fortunate for us he did so. Just after two o'clock in the afternoon, Shimomura's home systems were probed, then successfully attacked using something new in Internet attacks, sequence number guessing. Shimomura also works as a security expert, which made his systems both desirable targets for attack, and allows us to understand in detail what happened. Because, unlike most networks, Shimomura was using tcpdump to monitor traffic incoming from his Internet connection, and routinely sent his logs to an offsite location. Sequence number guessing is not really new. Steve Bellovin, a researcher at Bell Labs, and co-author of the Firewalls and Internet Security book (Addison-Wesley, 1994, ISBN 0-201-63357-4), included details of an attack scenario in his 1989 paper entitled ``Security Problems in the TCP/IP Protocol Suite''. But the Christmas day attack is the first known use of the technique. To better understand what happened, it helps to understand a little about how TCP (Transport Control Protocol) works. TCP is used for establishing bidirectional streams, like those used for remote terminal connections (established with telnet or rlogin utilities). TCP is also used for transferring large amounts of data, for example with FTP or connecting to a Web server. TCP provides a reliable connection. That is, unlike most other parts of the Internet Protocol suite (such as ICMP, Internet Control Message Protocol, or UDP, User Datagram Protocol), TCP establishes a connection between the local and remote site. Once the connection has been successfully established, groups of bytes of data are acknowledged by sending a sequence number back to the sending site. If the sending site does not receive an acknowledgement quickly enough, it will resend the data. If the sending site has resent the same data several times unsuccessfully, it will send an error to the application saying that the connection has been broken. The sequence number is used to acknowledge receipt of data. At the beginning of a TCP connection, the client sends a TCP packet with an initial sequence number, but no acknowledgement (there can't be one yet). If there is a server application running at the other end of the connection, the server sends back a TCP packet with its own initial sequence number, and an acknowledgement: the initial sequence number from the client's packet plus one. When the client system receives this packet, it must send back its own acknowledgement: the server's initial sequence number plus one. Thus, it takes three packets to establish a TCP connection (see Part A of Figure 1 which shows the time-line diagram. There's more to TCP, of course. You won't learn all about TCP in this short article (try Doug Comer's book Internetworking with TCP/IP, Volume 1, Principles, Protocols, and Architecture. Second Edition (Prentice Hall, 1991 ISBN 0-13-468505-9) or W. Richard Steven's TCP/IP Illustrated, Volume 1 (Addison-Wesley, 1993, ISBN 0-201-63346-9). For now, it's important to understand that TCP packets include flag bits that get set to indicate conditions. When you read Shimomura's account of the attack, he makes reference to several flags. The SYN flag (shown as a capital ``S'' in tcpdump command output) indicates the initiation of a connection, and that an initial sequence number is included. When the first packet is sent from the client, only the SYN flag is set. When the server responds, both the SYN flag and the ACK flag, indicating that a valid acknowledgement is included, are set. From then on, the ACK flag will be set, showing that each packet includes an acknowledgement of a received packet. The PUSH (shown as a capital ``P'') flag means that the data in this packet should be pushed to the application, rathered than queued until more data arrives. The RESET (``R'') flag tells TCP to break (reset) the connection, and is sent when a client attempts to connect to a server application that is not running. In the attack, RESETs are used to close the half-open connections used to keep the server busy. The FIN bit (``F'') is used to close a connection. Each end of the connection sends a packet with the FIN flag, which must be acknowledged, so four packets are used to close a TCP connection. Of course in the attack you won't see two FIN packets, because the attacker never sees the responses from the target system, the X terminal. With this background, you are ready to read Shimomura's own description, if you haven't already. Essentially, the attack begins when several probes were launched from toad.com (a site registered by Nebula Consulting). Although I don't know this for sure, it is likely that toad.com had been broken into previously using other techniques. The probes, using finger, showmount, and rpcinfo (similar to probes from SATAN, but not automatic) apparently helped the attacker to determine a trust relationship between Shimomura's X terminal (actually another workstation used as an X terminal), and a local server. This was the real weakness exploited in the attack. Shimomura's systems trusted one another, using the trust mechanism exploited in the ``r'' commands like rsh and rcp. Although convenient, and safe behind a strong perimeter defense, trust has been used to break into systems for many years. The November 1988 Internet Worm exploited trust in its automated attacks. In the next phase of the attack, thirty TCP SYN packets are sent to the rlogin port of Shimomura's server. These packets come from an unused Internet address, and their initial sequence numbers are incremented by one instead of the more common 128,000. The purpose of these packets are to fill the queue on the server with half-open connections, so when the spoofing begins, the server won't be able to respond to the packets being sent as acknowledgements from the X terminal. Next, a system at Loyola University of Chicago (apollo.it.luc.edu) was used to probe the X terminal. Once again, a synthetic series of TCP packets (initial sequence numbers incremented by one) gets sent, but this time responses get sent back to a real site. It is the responses that the attacker is after, because each response contains an initial sequence number from the X terminal. In Shimomura's paper tcpdump labels the responses with x-terminal.shell > apollo.it.luc.edu.1000 S, indicating a packet with the SYN flag set. If we subtract the first initial sequence number from the second, 2021952000-2021824000, we get 128,000, a pattern that holds for all twenty probes. So now the attacker knows that the next initial sequence number will be 128,000 greater than the previous one. The stage is now set for abusing the trust between the X terminal and the server. The attacker generates packets that appear to come from the server to open a TCP connection with the X terminal rshell daemon. The X terminal sends an acknowledgement back to the server, but this acknowledgement gets stuck in the queue. Next, the attacker generates the acknowledgement the server might have sent (if it had really been the server initiating the connection). Now there is an open TCP connection from the X terminal to the server, which is being spoofed by the server. Figure 1B shows how the packets actually traveled. Because the X terminal trusts the server, the attacker, masquerading as root, sends the command "echo + + >> /.rhosts" to the X terminal, extending trust to ANY root user with access to this system. Then the attacker closes the connection by sending a packet with FIN set, and acknowledging the FIN packet never seen from the X terminal. Finally, thirty RESETs are sent to the server to clear its queue. At this point, the server would send a RESET to the X terminal to close the rshell connection it never made (the one spoofed by the attacker), but it is now too late. The attacked can now log in as root on the X terminal using rlogin. There was more to the attack, but I'd like to save taking over TCP connections for another day. There is also a CERT advisory about this attack, which you can read. Copyright © 1995-1997 The McGraw-Hill Companies, Inc. All Rights Reserved. Edited by Becca Thomas / editor@unixworld.com ------------------------------------------------------------------------------- Network Working Group S. Bellovin Request for Comments: 1948 AT&T Research Category: Informational May 1996 Defending Against Sequence Number Attacks Status of This Memo This memo provides information for the Internet community. This memo does not specify an Internet standard of any kind. Distribution of this memo is unlimited. Abstract IP spoofing attacks based on sequence number spoofing have become a serious threat on the Internet (CERT Advisory CA-95:01). While ubiquitous crypgraphic authentication is the right answer, we propose a simple modification to TCP implementations that should be a very substantial block to the current wave of attacks. Overview and Rational In 1985, Morris [1] described a form of attack based on guessing what sequence numbers TCP [2] will use for new connections. Briefly, the attacker gags a host trusted by the target, impersonates the IP address of the trusted host when talking to the target, and completes the 3-way handshake based on its guess at the next initial sequence number to be used. An ordinary connection to the target is used to gather sequence number state information. This entire sequence, coupled with address-based authentication, allows the attacker to execute commands on the target host. Clearly, the proper solution is cryptographic authentication [3,4]. But it will quite a long time before that is deployed. It has therefore been necessary for many sites to restrict use of protocols that rely on address-based authentication, such as rlogin and rsh. Unfortunately, the prevalence of "sniffer attacks" -- network eavesdropping (CERT Advisory CA-94:01) -- has rendered ordinary TELNET [5] very dangerous as well. The Internet is thus left without a safe, secure mechanism for remote login. We propose a simple change to TCP implementations that will block most sequence number guessing attacks. More precisely, such attacks will remain possible if and only if the Bad Guy already has the ability to launch even more devastating attacks. Bellovin Informational [Page 1] ---------------------------------------------------------------------------- RFC 1948 Sequence Number Attacks May 1996 Details of the Attack In order to understand the particular case of sequence number guessing, one must look at the 3-way handshake used in the TCP open sequence [2]. Suppose client machine A wants to talk to rsh server B. It sends the following message: A->B: SYN, ISNa That is, it sends a packet with the SYN ("synchronize sequence number") bit set and an initial sequence number ISNa. B replies with B->A: SYN, ISNb, ACK(ISNa) In addition to sending its own initial sequence number, it acknowledges A's. Note that the actual numeric value ISNa must appear in the message. A concludes the handshake by sending A->B: ACK(ISNb) The initial sequence numbers are intended to be more or less random. More precisely, RFC 793 specifies that the 32-bit counter be incremented by 1 in the low-order position about every 4 microseconds. Instead, Berkeley-derived kernels increment it by a constant every second, and by another constant for each new connection. Thus, if you open a connection to a machine, you know to a very high degree of confidence what sequence number it will use for its next connection. And therein lies the attack. The attacker X first opens a real connection to its target B -- say, to the mail port or the TCP echo port. This gives ISNb. It then impersonates A and sends Ax->B: SYN, ISNx where "Ax" denotes a packet sent by X pretending to be A. B's response to X's original SYN (so to speak) B->A: SYN, ISNb', ACK(ISNx) Bellovin Informational [Page 2] ---------------------------------------------------------------------------- RFC 1948 Sequence Number Attacks May 1996 goes to the legitimate A, about which more anon. X never sees that message but can still send Ax->B: ACK(ISNb') using the predicted value for ISNb'. If the guess is right -- and usually it will be -- B's rsh server thinks it has a legitimate connection with A, when in fact X is sending the packets. X can't see the output from this session, but it can execute commands as more or less any user -- and in that case, the game is over and X has won. There is a minor difficulty here. If A sees B's message, it will realize that B is acknowledging something it never sent, and will send a RST packet in response to tear down the connection. There are a variety of ways to prevent this; the easiest is to wait until the real A is down (possibly as a result of enemy action, of course). In actual practice, X can gag A by exploiting a very common implementation bug; this is described below. The Fix The choice of initial sequence numbers for a connection is not random. Rather, it must be chosen so as to minimize the probability of old stale packets being accepted by new incarnations of the same connection [6, Appendix A]. Furthermore, implementations of TCP derived from 4.2BSD contain special code to deal with such reincarnations when the server end of the original connection is still in TIMEWAIT state [7, pp. 945]. Accordingly, simple randomization, as suggested in [8], will not work well. But duplicate packets, and hence the restrictions on the initial sequence number for reincarnations, are peculiar to individual connections. That is, there is no connection, syntactic or semantic, between the sequence numbers used for two different connections. We can prevent sequence number guessing attacks by giving each connection -- that is, each 4-tuple of -- a separate sequence number space. Within each space, the initial sequence number is incremented according to [2]; however, there is no obvious relationship between the numbering in different spaces. The obvious way to do this is to maintain state for dead connections, and the easiest way to do that is to change the TCP state transition diagram so that both ends of all connections go to TIMEWAIT state. That would work, but it's inelegant and consumes storage space. Instead, we use the current 4 microsecond timer M and set ISN = M + F(localhost, localport, remotehost, remoteport). Bellovin Informational [Page 3] ---------------------------------------------------------------------------- RFC 1948 Sequence Number Attacks May 1996 It is vital that F not be computable from the outside, or an attacker could still guess at sequence numbers from the initial sequence number used for some other connection. We therefore suggest that F be a cryptographic hash function of the connection-id and some secret data. MD5 [9] is a good choice, since the code is widely available. The secret data can either be a true random number [10], or it can be the combination of some per-host secret and the boot time of the machine. The boot time is included to ensure that the secret is changed on occasion. Other data, such as the host's IP address and name, may be included in the hash as well; this eases administration by permitting a network of workstations to share the same secret data while still giving them separate sequence number spaces. Our recommendation, in fact, is to use all three of these items: as random a number as the hardware can generate, an administratively- installed pass phrase, and the machine's IP address. This allows for local choice on how secure the secret is. Note that the secret cannot easily be changed on a live machine. Doing so would change the initial sequence numbers used for reincarnated connections; to maintain safety, either dead connection state must be kept or a quiet time observed for two maximum segment lifetimes after such a change. A Common TCP Bug As mentioned earlier, attackers using sequence number guessing have to "gag" the trusted machine first. While a number of strategies are possible, most of the attacks detected thus far rely on an implementation bug. When SYN packets are received for a connection, the receiving system creates a new TCB in SYN-RCVD state. To avoid overconsumption of resources, 4.2BSD-derived systems permit only a limited number of TCBs in this state per connection. Once this limit is reached, future SYN packets for new connections are discarded; it is assumed that the client will retransmit them as needed. When a packet is received, the first thing that must be done is a search for the TCB for that connection. If no TCB is found, the kernel searches for a "wild card" TCB used by servers to accept connections from all clients. Unfortunately, in many kernels this code is invoked for any incoming packets, not just for initial SYN packets. If the SYN-RCVD queue is full for the wildcard TCB, any new packets specifying just that host and port number will be discarded, even if they aren't SYN packets. Bellovin Informational [Page 4] ---------------------------------------------------------------------------- RFC 1948 Sequence Number Attacks May 1996 To gag a host, then, the attacker sends a few dozen SYN packets to the rlogin port from different port numbers on some non-existent machine. This fills up the SYN-RCVD queue, while the SYN+ACK packets go off to the bit bucket. The attack on the target machine then appears to come from the rlogin port on the trusted machine. The replies -- the SYN+ACKs from the target -- will be perceived as packets belonging to a full queue, and will be dropped silently. This could be avoided if the full queue code checked for the ACK bit, which cannot legally be on for legitimate open requests. If it is on, RST should be sent in reply. Security Considerations Good sequence numbers are not a replacement for cryptographic authentication. At best, they're a palliative measure. An eavesdropper who can observe the initial messages for a connection can determine its sequence number state, and may still be able to launch sequence number guessing attacks by impersonating that connection. However, such an eavesdropper can also hijack existing connections [11], so the incremental threat isn't that high. Still, since the offset between a fake connection and a given real connection will be more or less constant for the lifetime of the secret, it is important to ensure that attackers can never capture such packets. Typical attacks that could disclose them include both eavesdropping and the variety of routing attacks discussed in [8]. If random numbers are used as the sole source of the secret, they MUST be chosen in accordance with the recommendations given in [10]. Acknowledgments Matt Blaze and Jim Ellis contributed some crucial ideas to this RFC. Frank Kastenholz contributed constructive comments to this memo. References [1] R.T. Morris, "A Weakness in the 4.2BSD UNIX TCP/IP Software", CSTR 117, 1985, AT&T Bell Laboratories, Murray Hill, NJ. [2] Postel, J., "Transmission Control Protocol", STD 7, RFC 793, September 1981. [3] Kohl, J., and C. Neuman, "The Kerberos Network Authentication Service (V5)", RFC 1510, September 1993. [4] Atkinson, R., "Security Architecture for the Internet Protocol", RFC 1825, August 1995. Bellovin Informational [Page 5] ---------------------------------------------------------------------------- RFC 1948 Sequence Number Attacks May 1996 [5] Postel, J., and J. Reynolds, "Telnet Protocol Specification", STD 8, RFC 854, May 1983. [6] Jacobson, V., Braden, R., and L. Zhang, "TCP Extension for High-Speed Paths", RFC 1885, October 1990. [7] G.R. Wright, W. R. Stevens, "TCP/IP Illustrated, Volume 2", 1995. Addison-Wesley. [8] S. Bellovin, "Security Problems in the TCP/IP Protocol Suite", April 1989, Computer Communications Review, vol. 19, no. 2, pp. 32-48. [9] Rivest, R., "The MD5 Message-Digest Algorithm", RFC 1321, April 1992. [10] Eastlake, D., Crocker, S., and J. Schiller, "Randomness Recommendations for Security", RFC 1750, December 1994. [11] L. Joncheray, "A Simple Active Attack Against TCP, 1995, Proc. Fifth Usenix UNIX Security Symposium. Author's Address Steven M. Bellovin AT&T Research 600 Mountain Avenue Murray Hill, NJ 07974 Phone: (908) 582-5886 EMail: smb@research.att.com ----------------------------------------------------------------------------