Tarpit (networking)
From Wikipedia, the free encyclopedia
A tarpit (also known as Teergrube, the German word for tarpit) is a service on a computer system (usually a server) that delays incoming connections for as long as possible. The technique was developed as a defense against a computer worm, and the idea is that network abuses such as spamming or broad scanning are less effective if they take too long. The name is analogous with a tar pit, in which animals can get bogged down and slowly sink under the surface.
Contents |
[edit] SMTP tarpits
Various methods have been discussed and implemented for SMTP tarpits, systems that plug into the Mail Transfer Agent (MTA, i.e. the mail server software) or sit in front of it as a proxy.
One method increases transfer time for all mails by a few seconds by delaying the initial greeting message ("greet delay"). The idea is that it will not matter if a legitimate mail takes a little longer to deliver, but due to the high volume, it will make a difference for spammers. The downside of this is that mailing lists and other legitimate mass-mailings will have to be explicitly whitelisted or they will suffer too.
Another method is to delay only known spammers, e.g. by using a blacklist (see Spamming, DNSBL). OpenBSD has recently integrated this method into their core system, with a special-purpose daemon (spamd) and functionality in the firewall (pf) to redirect known spammers to this tarpit.
A more subtle idea is greylisting, which, in simple terms, rejects the first connection attempt from any previously-unseen IP address. The assumption is that most spammers make only one connection attempt (or a few attempts over a short period of time) to send each message, whereas legitimate mail delivery systems will keep retrying over a longer period. After they retry, they will eventually be allowed in without any further impediments.
Finally, a more elaborate method tries to glue tarpits and filtering software together, by filtering e-mail in realtime, while it is being transmitted, and adding delays to the communication in response to the filter's "spam likeliness" indicator. For example, the spam filter would make a "guess" after each line or after every x bytes received as to how likely this message is going to be spam. The more likely this is, the more the MTA will delay the transmission.
[edit] Background
The SMTP protocol consists of requests, which are four-letter words such as MAIL, and replies, which are three-digit numbers. In the last line of the reply, the number is followed by a space; in the preceding lines it is followed by a hyphen. Thus, on determining that a message being attempted to send is spam, a mail server can reply:
451-Ophiomyia prima is an agromyzid fly 451-Ophiomyia secunda is an agromyzid fly 451-Ophiomyia tertia is an agromyzid fly 451-Ophiomyia quarta is an agromyzid fly 451-Ophiomyia quinta is an agromyzid fly 451-Ophiomyia sexta is an agromyzid fly 451-Ophiomyia septima is an agromyzid fly 451 Your IP address is listed in the DNSBL. Please try again later.
The tarpit waits fifteen or more seconds between lines (long delays are allowed in SMTP, as humans sometimes send mail manually to test mail servers). This ties up the SMTP sending process on the spammer's box so that he cannot send as much spam.
[edit] IP-level tarpits
The Linux kernel can now be patched to allow tarpitting of incoming connections instead of the more usual dropping of packets. This is implemented in iptables by the addition of a TARPIT target. The same packet inspection and matching features can be applied to tarpit targets as are applied to other targets.
[edit] The original tarpit idea
Tom Liston developed the original tarpitting program LaBrea. It can protect an entire network with a tarpit run from a single machine. The machine listens for ARP requests that go unanswered (indicating unused addresses), then replies to those requests, receives the initial SYN packet of the scanner and sends a SYN/ACK in response. It does not open a socket or prepare a connection, in fact it can forget all about the connection after sending the SYN/ACK.
However, the remote site sends its ACK (which gets ignored) and believes the 3-way-handshake to be complete. Then it starts to send data, which never reaches a destination. The connection will time out after a while, but since the system believes it is dealing with a live, i.e. established connection, it is conservative in timing it out and will instead try to retransmit, back-off, retransmit, etc. for quite a while.
Later versions of LaBrea also added functionality to reply to the incoming data, again using raw IP packets and no sockets or other resources of the tarpit server, with bogus packets that request that the sending site "slow down". This will keep the connection established and waste even more time of the scanner.
[edit] Commercial implementations of tar-pitting
There have been two successful commercial implementations of the tar pit idea. The first was developed by TurnTide, a Philadelphia-based startup company, which was acquired by Symantec in 2004 for $28 million in cash [1]. The TurnTide Anti Spam Router contains a modified Linux kernel which allows it to play various tricks with TCP traffic, such as varying the TCP window size. By grouping various email senders into different traffic classes and limiting the bandwidth for each class, the amount of abusive traffic is reduced - particularly when the abusive traffic is coming from single sources which are easily identified by their high traffic volume.
After the Symantec acquisition, a Canadian startup company called MailChannels released their "Traffic Control" software, which uses a slightly different approach to achieve similar results. Traffic Control is a semi-realtime SMTP Proxy. Unlike the TurnTide appliance, which applies Traffic Shaping at the network layer, Traffic Control applies traffic shaping to individual senders at the application layer. This approach results in a somewhat more effective handling of spam traffic originating from Botnets because it allows the software to slow traffic from individual spam zombies, rather than requiring zombie traffic to be aggregated into a class.
[edit] See also
[edit] External links
This article was originally based on material from the Free On-line Dictionary of Computing, which is licensed under the GFDL.
