| ==Phrack Magazine== | |
| Volume Seven, Issue Forty-Eight, File 14 of 18 | |
| [ IP-spoofing Demystified ] | |
| (Trust-Relationship Exploitation) | |
| by daemon9 / route / infinity | |
| for Phrack Magazine | |
| June 1996 Guild Productions, kid | |
| comments to route@infonexus.com | |
| The purpose of this paper is to explain IP-spoofing to the | |
| masses. It assumes little more than a working knowledge of Unix and | |
| TCP/IP. Oh, and that yur not a moron... | |
| IP-spoofing is complex technical attack that is made up of | |
| several components. (In actuality, IP-spoofing is not the attack, but | |
| a step in the attack. The attack is actually trust-relationship | |
| exploitation. However, in this paper, IP-spoofing will refer to the | |
| whole attack.) In this paper, I will explain the attack in detail, | |
| including the relevant operating system and networking information. | |
| [SECTION I. BACKGROUND INFORMATION] | |
| --[ The Players ]-- | |
| A: Target host | |
| B: Trusted host | |
| X: Unreachable host | |
| Z: Attacking host | |
| (1)2: Host 1 masquerading as host 2 | |
| --[ The Figures ]-- | |
| There are several figures in the paper and they are to be | |
| interpreted as per the following example: | |
| ick host a control host b | |
| 1 A ---SYN---> B | |
| tick: A tick of time. There is no distinction made as to *how* | |
| much time passes between ticks, just that time passes. It's generally | |
| not a great deal. | |
| host a: A machine particpating in a TCP-based conversation. | |
| control: This field shows any relevant control bits set in the TCP | |
| header and the direction the data is flowing | |
| host b: A machine particpating in a TCP-based conversation. | |
| In this case, at the first refrenced point in time host a is sending | |
| a TCP segment to host b with the SYN bit on. Unless stated, we are | |
| generally not concerned with the data portion of the TCP segment. | |
| --[ Trust Relationships ]-- | |
| In the Unix world, trust can be given all too easily. Say you | |
| have an account on machine A, and on machine B. To facilitate going | |
| betwixt the two with a minimum amount of hassle, you want to setup a | |
| full-duplex trust relationship between them. In your home directory | |
| at A you create a .rhosts file: `echo "B username" > ~/.rhosts` In | |
| your home directory at B you create a .rhosts file: `echo "A username" | |
| > ~/.rhosts` (Alternately, root can setup similar rules in | |
| /etc/hosts.equiv, the difference being that the rules are hostwide, | |
| rather than just on an individual basis.) Now, you can use any of the | |
| r* commands without that annoying hassle of password authentication. | |
| These commands will allow address-based authentication, which will | |
| grant or deny access based off of the IP address of the service | |
| requestor. | |
| --[ Rlogin ]-- | |
| Rlogin is a simple client-server based protocol that uses TCP | |
| as it's transport. Rlogin allows a user to login remotely from one | |
| host to another, and, if the target machine trusts the other, rlogin | |
| will allow the convienience of not prompting for a password. It will | |
| instead have authenticated the client via the source IP address. So, | |
| from our example above, we can use rlogin to remotely login to A from | |
| B (or vice-versa) and not be prompted for a password. | |
| --[ Internet Protocol ]-- | |
| IP is the connectionless, unreliable network protocol in the | |
| TCP/IP suite. It has two 32-bit header fields to hold address | |
| information. IP is also the busiest of all the TCP/IP protocols as | |
| almost all TCP/IP traffic is encapsulated in IP datagrams. IP's job | |
| is to route packets around the network. It provides no mechanism for | |
| reliability or accountability, for that, it relies on the upper | |
| layers. IP simply sends out datagrams and hopes they make it intact. | |
| If they don't, IP can try to send an ICMP error message back to the | |
| source, however this packet can get lost as well. (ICMP is Internet | |
| Control Message Protocol and it is used to relay network conditions | |
| and different errors to IP and the other layers.) IP has no means to | |
| guarantee delivery. Since IP is connectionless, it does not maintain | |
| any connection state information. Each IP datagram is sent out without | |
| regard to the last one or the next one. This, along with the fact that | |
| it is trivial to modify the IP stack to allow an arbitrarily choosen IP | |
| address in the source (and destination) fields make IP easily subvertable. | |
| --[ Transmission Control Protocol ]-- | |
| TCP is the connection-oriented, reliable transport protocol | |
| in the TCP/IP suite. Connection-oriented simply means that the two | |
| hosts participating in a discussion must first establish a connection | |
| before data may change hands. Reliability is provided in a number of | |
| ways but the only two we are concerned with are data sequencing and | |
| acknowledgement. TCP assigns sequence numbers to every segment and | |
| acknowledges any and all data segments recieved from the other end. | |
| (ACK's consume a sequence number, but are not themselves ACK'd.) | |
| This reliability makes TCP harder to fool than IP. | |
| --[ Sequence Numbers, Acknowledgements and other flags ]-- | |
| Since TCP is reliable, it must be able to recover from | |
| lost, duplicated, or out-of-order data. By assigning a sequence | |
| number to every byte transfered, and requiring an acknowledgement from | |
| the other end upon receipt, TCP can guarantee reliable delivery. The | |
| receiving end uses the sequence numbers to ensure proper ordering of | |
| the data and to eliminate duplicate data bytes. | |
| TCP sequence numbers can simply be thought of as 32-bit | |
| counters. They range from 0 to 4,294,967,295. Every byte of | |
| data exchanged across a TCP connection (along with certain flags) | |
| is sequenced. The sequence number field in the TCP header will | |
| contain the sequence number of the *first* byte of data in the | |
| TCP segment. The acknowledgement number field in the TCP header | |
| holds the value of next *expected* sequence number, and also | |
| acknowledges *all* data up through this ACK number minus one. | |
| TCP uses the concept of window advertisement for flow | |
| control. It uses a sliding window to tell the other end how much | |
| data it can buffer. Since the window size is 16-bits a receiving TCP | |
| can advertise up to a maximum of 65535 bytes. Window advertisement | |
| can be thought of an advertisment from one TCP to the other of how | |
| high acceptable sequence numbers can be. | |
| Other TCP header flags of note are RST (reset), PSH (push) | |
| and FIN (finish). If a RST is received, the connection is | |
| immediately torn down. RSTs are normally sent when one end | |
| receives a segment that just doesn't jive with current connection | |
| (we will encounter an example below). The PSH flag tells the | |
| reciever to pass all the data is has queued to the aplication, as | |
| soon as possible. The FIN flag is the way an application begins a | |
| graceful close of a connection (connection termination is a 4-way | |
| process). When one end recieves a FIN, it ACKs it, and does not | |
| expect to receive any more data (sending is still possible, however). | |
| --[ TCP Connection Establishment ]-- | |
| In order to exchange data using TCP, hosts must establish a | |
| a connection. TCP establishes a connection in a 3 step process called | |
| the 3-way handshake. If machine A is running an rlogin client and | |
| wishes to conect to an rlogin daemon on machine B, the process is as | |
| follows: | |
| fig(1) | |
| 1 A ---SYN---> B | |
| 2 A <---SYN/ACK--- B | |
| 3 A ---ACK---> B | |
| At (1) the client is telling the server that it wants a connection. | |
| This is the SYN flag's only purpose. The client is telling the | |
| server that the sequence number field is valid, and should be checked. | |
| The client will set the sequence number field in the TCP header to | |
| it's ISN (initial sequence number). The server, upon receiving this | |
| segment (2) will respond with it's own ISN (therefore the SYN flag is | |
| on) and an ACKnowledgement of the clients first segment (which is the | |
| client's ISN+1). The client then ACK's the server's ISN (3). Now, | |
| data transfer may take place. | |
| --[ The ISN and Sequence Number Incrementation ]-- | |
| It is important to understand how sequence numbers are | |
| initially choosen, and how they change with respect to time. The | |
| initial sequence number when a host is bootstraped is initialized | |
| to 1. (TCP actually calls this variable 'tcp_iss' as it is the initial | |
| *send* sequence number. The other sequence number variable, | |
| 'tcp_irs' is the initial *receive* sequence number and is learned | |
| during the 3-way connection establishment. We are not going to worry | |
| about the distinction.) This practice is wrong, and is acknowledged | |
| as so in a comment the tcp_init() function where it appears. The ISN | |
| is incremented by 128,000 every second, which causes the 32-bit ISN | |
| counter to wrap every 9.32 hours if no connections occur. However, | |
| each time a connect() is issued, the counter is incremented by | |
| 64,000. | |
| One important reason behind this predictibility is to | |
| minimize the chance that data from an older stale incarnation | |
| (that is, from the same 4-tuple of the local and remote | |
| IP-addresses TCP ports) of the current connection could arrive | |
| and foul things up. The concept of the 2MSL wait time applies | |
| here, but is beyond the scope of this paper. If sequence | |
| numbers were choosen at random when a connection arrived, no | |
| guarantees could be made that the sequence numbers would be different | |
| from a previous incarnation. If some data that was stuck in a | |
| routing loop somewhere finally freed itself and wandered into the new | |
| incarnation of it's old connection, it could really foul things up. | |
| --[ Ports ]-- | |
| To grant simultaneous access to the TCP module, TCP provides | |
| a user interface called a port. Ports are used by the kernel to | |
| identify network processes. These are strictly transport layer | |
| entities (that is to say that IP could care less about them). | |
| Together with an IP address, a TCP port provides provides an endpoint | |
| for network communications. In fact, at any given moment *all* | |
| Internet connections can be described by 4 numbers: the source IP | |
| address and source port and the destination IP address and destination | |
| port. Servers are bound to 'well-known' ports so that they may be | |
| located on a standard port on different systems. For example, the | |
| rlogin daemon sits on TCP port 513. | |
| [SECTION II. THE ATTACK] | |
| ...The devil finds work for idle hands.... | |
| --[ Briefly... ]-- | |
| IP-spoofing consists of several steps, which I will | |
| briefly outline here, then explain in detail. First, the target host | |
| is choosen. Next, a pattern of trust is discovered, along with a | |
| trusted host. The trusted host is then disabled, and the target's TCP | |
| sequence numbers are sampled. The trusted host is impersonated, the | |
| sequence numbers guessed, and a connection attempt is made to a | |
| service that only requires address-based authentication. If | |
| successful, the attacker executes a simple command to leave a | |
| backdoor. | |
| --[ Needful Things ]-- | |
| There are a couple of things one needs to wage this attack: | |
| (1) brain, mind, or other thinking device | |
| (1) target host | |
| (1) trusted host | |
| (1) attacking host (with root access) | |
| (1) IP-spoofing software | |
| Generally the attack is made from the root account on the attacking | |
| host against the root account on the target. If the attacker is | |
| going to all this trouble, it would be stupid not to go for root. | |
| (Since root access is needed to wage the attack, this should not | |
| be an issue.) | |
| --[ IP-Spoofing is a 'Blind Attack' ]-- | |
| One often overlooked, but critical factor in IP-spoofing | |
| is the fact that the attack is blind. The attacker is going to be | |
| taking over the identity of a trusted host in order to subvert the | |
| security of the target host. The trusted host is disabled using the | |
| method described below. As far as the target knows, it is carrying on | |
| a conversation with a trusted pal. In reality, the attacker is | |
| sitting off in some dark corner of the Internet, forging packets | |
| puportedly from this trusted host while it is locked up in a denial | |
| of service battle. The IP datagrams sent with the forged IP-address | |
| reach the target fine (recall that IP is a connectionless-oriented | |
| protocol-- each datagram is sent without regard for the other end) | |
| but the datagrams the target sends back (destined for the trusted | |
| host) end up in the bit-bucket. The attacker never sees them. The | |
| intervening routers know where the datagrams are supposed to go. They | |
| are supposed to go the trusted host. As far as the network layer is | |
| concerned, this is where they originally came from, and this is where | |
| responses should go. Of course once the datagrams are routed there, | |
| and the information is demultiplexed up the protocol stack, and | |
| reaches TCP, it is discarded (the trusted host's TCP cannot respond-- | |
| see below). So the attacker has to be smart and *know* what was sent, | |
| and *know* what reponse the server is looking for. The attacker | |
| cannot see what the target host sends, but she can *predict* what it | |
| will send; that coupled with the knowledge of what it *will* send, | |
| allows the attacker to work around this blindness. | |
| --[ Patterns of Trust ]-- | |
| After a target is choosen the attacker must determine the | |
| patterns of trust (for the sake of argument, we are going to assume | |
| the target host *does* in fact trust somebody. If it didn't, the | |
| attack would end here). Figuring out who a host trusts may or may | |
| not be easy. A 'showmount -e' may show where filesystems are | |
| exported, and rpcinfo can give out valuable information as well. | |
| If enough background information is known about the host, it should | |
| not be too difficult. If all else fails, trying neighboring IP | |
| addresses in a brute force effort may be a viable option. | |
| --[ Trusted Host Disabling Using the Flood of Sins ]-- | |
| Once the trusted host is found, it must be disabled. Since | |
| the attacker is going to impersonate it, she must make sure this host | |
| cannot receive any network traffic and foul things up. There are | |
| many ways of doing this, the one I am going to discuss is TCP SYN | |
| flooding. | |
| A TCP connection is initiated with a client issuing a | |
| request to a server with the SYN flag on in the TCP header. Normally | |
| the server will issue a SYN/ACK back to the client identified by the | |
| 32-bit source address in the IP header. The client will then send an | |
| ACK to the server (as we saw in figure 1 above) and data transfer | |
| can commence. There is an upper limit of how many concurrent SYN | |
| requests TCP can process for a given socket, however. This limit | |
| is called the backlog, and it is the length of the queue where | |
| incoming (as yet incomplete) connections are kept. This queue limit | |
| applies to both the number of imcomplete connections (the 3-way | |
| handshake is not complete) and the number of completed connections | |
| that have not been pulled from the queue by the application by way of | |
| the accept() system call. If this backlog limit is reached, TCP will | |
| silently discard all incoming SYN requests until the pending | |
| connections can be dealt with. Therein lies the attack. | |
| The attacking host sends several SYN requests to the TCP port | |
| she desires disabled. The attacking host also must make sure that | |
| the source IP-address is spoofed to be that of another, currently | |
| unreachable host (the target TCP will be sending it's response to | |
| this address. (IP may inform TCP that the host is unreachable, | |
| but TCP considers these errors to be transient and leaves the | |
| resolution of them up to IP (reroute the packets, etc) effectively | |
| ignoring them.) The IP-address must be unreachable because the | |
| attacker does not want any host to recieve the SYN/ACKs that will be | |
| coming from the target TCP (this would result in a RST being sent to | |
| the target TCP, which would foil our attack). The process is as | |
| follows: | |
| fig(2) | |
| 1 Z(x) ---SYN---> B | |
| Z(x) ---SYN---> B | |
| Z(x) ---SYN---> B | |
| Z(x) ---SYN---> B | |
| Z(x) ---SYN---> B | |
| ... | |
| 2 X <---SYN/ACK--- B | |
| X <---SYN/ACK--- B | |
| ... | |
| 3 X <---RST--- B | |
| At (1) the attacking host sends a multitude of SYN requests to the | |
| target (remember the target in this phase of the attack is the | |
| trusted host) to fill it's backlog queue with pending connections. | |
| (2) The target responds with SYN/ACKs to what it believes is the | |
| source of the incoming SYNs. During this time all further requests | |
| to this TCP port will be ignored. | |
| Different TCP implementations have different backlog sizes. | |
| BSD generally has a backlog of 5 (Linux has a backlog of 6). There | |
| is also a 'grace' margin of 3/2. That is, TCP will allow up to | |
| backlog*3/2+1 connections. This will allow a socket one connection | |
| even if it calls listen with a backlog of 0. | |
| AuthNote: [For a much more in-depth treatment of TCP SYN | |
| flooding, see my definitive paper on the subject. It covers the | |
| whole process in detail, in both theory, and practice. There is | |
| robust working code, a statistical analysis, and a legnthy paper. | |
| Look for it in issue 49 of Phrack. -daemon9 6/96] | |
| --[ Sequence Number Sampling and Prediction ]-- | |
| Now the attacker needs to get an idea of where in the 32-bit | |
| sequence number space the target's TCP is. The attacker connects to | |
| a TCP port on the target (SMTP is a good choice) just prior to launching | |
| the attack and completes the three-way handshake. The process is | |
| exactly the same as fig(1), except that the attacker will save the | |
| value of the ISN sent by the target host. Often times, this process is | |
| repeated several times and the final ISN sent is stored. The attacker | |
| needs to get an idea of what the RTT (round-trip time) from the target | |
| to her host is like. (The process can be repeated several times, and an | |
| average of the RTT's is calculated.) The RTT is necessary in being | |
| able to accuratly predict the next ISN. The attacker has the baseline | |
| (the last ISN sent) and knows how the sequence numbers are incremented | |
| (128,000/second and 64,000 per connect) and now has a good idea of | |
| how long it will take an IP datagram to travel across the Internet to | |
| reach the target (approximately half the RTT, as most times the | |
| routes are symmetrical). After the attacker has this information, she | |
| immediately proceeds to the next phase of the attack (if another TCP | |
| connection were to arrive on any port of the target before the | |
| attacker was able to continue the attack, the ISN predicted by the | |
| attacker would be off by 64,000 of what was predicted). | |
| When the spoofed segment makes it's way to the target, | |
| several different things may happen depending on the accuracy of | |
| the attacker's prediction: | |
| - If the sequence number is EXACTly where the receiving TCP expects | |
| it to be, the incoming data will be placed on the next available | |
| position in the receive buffer. | |
| - If the sequence number is LESS than the expected value the data | |
| byte is considered a retransmission, and is discarded. | |
| - If the sequence number is GREATER than the expected value but | |
| still within the bounds of the receive window, the data byte is | |
| considered to be a future byte, and is held by TCP, pending the | |
| arrival of the other missing bytes. If a segment arrives with a | |
| sequence number GREATER than the expected value and NOT within the | |
| bounds of the receive window the segment is dropped, and TCP will | |
| send a segment back with the *expected* sequence number. | |
| --[ Subversion... ]-- | |
| Here is where the main thrust of the attack begins: | |
| fig(3) | |
| 1 Z(b) ---SYN---> A | |
| 2 B <---SYN/ACK--- A | |
| 3 Z(b) ---ACK---> A | |
| 4 Z(b) ---PSH---> A | |
| [...] | |
| The attacking host spoofs her IP address to be that of the trusted | |
| host (which should still be in the death-throes of the D.O.S. attack) | |
| and sends it's connection request to port 513 on the target (1). At | |
| (2), the target responds to the spoofed connection request with a | |
| SYN/ACK, which will make it's way to the trusted host (which, if it | |
| *could* process the incoming TCP segment, it would consider it an | |
| error, and immediately send a RST to the target). If everything goes | |
| according to plan, the SYN/ACK will be dropped by the gagged trusted | |
| host. After (1), the attacker must back off for a bit to give the | |
| target ample time to send the SYN/ACK (the attacker cannot see this | |
| segment). Then, at (3) the attacker sends an ACK to the target with | |
| the predicted sequence number (plus one, because we're ACKing it). | |
| If the attacker is correct in her prediction, the target will accept | |
| the ACK. The target is compromised and data transfer can | |
| commence (4). | |
| Generally, after compromise, the attacker will insert a | |
| backdoor into the system that will allow a simpler way of intrusion. | |
| (Often a `cat + + >> ~/.rhosts` is done. This is a good idea for | |
| several reasons: it is quick, allows for simple re-entry, and is not | |
| interactive. Remember the attacker cannot see any traffic coming from | |
| the target, so any reponses are sent off into oblivion.) | |
| --[ Why it Works ]-- | |
| IP-Spoofing works because trusted services only rely on | |
| network address based authentication. Since IP is easily duped, | |
| address forgery is not difficult. The hardest part of the attck is | |
| in the sequence number prediction, because that is where the guesswork | |
| comes into play. Reduce unknowns and guesswork to a minimum, and | |
| the attack has a better chance of suceeding. Even a machine that | |
| wraps all it's incoming TCP bound connections with Wietse Venema's TCP | |
| wrappers, is still vulnerable to the attack. TCP wrappers rely on a | |
| hostname or an IP address for authentication... | |
| [SECTION III. PREVENTITIVE MEASURES] | |
| ...A stich in time, saves nine... | |
| --[ Be Un-trusting and Un-trustworthy ]-- | |
| One easy solution to prevent this attack is not to rely | |
| on address-based authentication. Disable all the r* commands, | |
| remove all .rhosts files and empty out the /etc/hosts.equiv file. | |
| This will force all users to use other means of remote access | |
| (telnet, ssh, skey, etc). | |
| --[ Packet Filtering ]-- | |
| If your site has a direct connect to the Internet, you | |
| can use your router to help you out. First make sure only hosts | |
| on your internal LAN can particpate in trust-relationships (no | |
| internal host should trust a host outside the LAN). Then simply | |
| filter out *all* traffic from the outside (the Internet) that | |
| puports to come from the inside (the LAN). | |
| --[ Cryptographic Methods ]-- | |
| An obvious method to deter IP-spoofing is to require | |
| all network traffic to be encrypted and/or authenticated. While | |
| several solutions exist, it will be a while before such measures are | |
| deployed as defacto standards. | |
| --[ Initial Sequence Number Randomizing ]-- | |
| Since the sequence numbers are not choosen randomly (or | |
| incremented randomly) this attack works. Bellovin describes a | |
| fix for TCP that involves partitioning the sequence number space. | |
| Each connection would have it's own seperate sequence number space. | |
| The sequence numbers would still be incremented as before, however, | |
| there would be no obvious or implied relationship between the | |
| numbering in these spaces. Suggested is the following formula: | |
| ISN=M+F(localhost,localport,remotehost,remoteport) | |
| Where M is the 4 microsecond timer and F is a cryptographic hash. | |
| F must not be computable from the outside or the attacker could | |
| still guess sequence numbers. Bellovin suggests F be a hash of | |
| the connection-id and a secret vector (a random number, or a host | |
| related secret combined with the machine's boot time). | |
| [SECTION IV. SOURCES] | |
| -Books: TCP/IP Illustrated vols. I, II & III | |
| -RFCs: 793, 1825, 1948 | |
| -People: Richard W. Stevens, and the users of the | |
| Information Nexus for proofreading | |
| -Sourcecode: rbone, mendax, SYNflood | |
| This paper made possible by a grant from the Guild Corporation. | |