- From: Jeffrey Mogul <mogul@pa.dec.com>
- Date: Fri, 27 Oct 95 11:38:54 MDT
- To: Beth Frank <efrank@ncsa.uiuc.edu>
- Cc: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
A few days ago, Beth wrote: We've been warned repeatedly by our security guru that persistent TCP connections can by hijacked by unauthorized users without much difficulty, so there are different constraints on an authentication session than on an open (insecure?) session. This was something I had not considered before, and I'm not a real expert on the TCP hijacking problem, so I asked several people who are recognized experts. It is important to pose the question the right way. The choice is not between equal numbers of long or short TCP connections; for a given number of HTTP requests, the choice is between some number of long TCP connections or a relatively larger number of short TCP connections. Laurent Joncheray, author of "A Simple Active Attack Against TCP", Proc. 5th USENIX UNIX Security Symposium, 1995, pages: 7-19, available via http://www.merit.edu/routing.arbiter/RA/security/.index.html told me: One year ago I did some testing on hijacking a TCP connection. I used one method of my own, but I am sure there are others. The result are that the dificulty is independent of the lenght of the connection. The hijacking happens at the begining, so the more TCP connections you open the more likely you are to be hijacked. Since persistent-connection HTTP would reduce the number of times a TCP connection is opened, it reduces the number of chances to hijack a connection using Laurent's scheme. Laurent also writes: Another result was that the slower the connection (long RTT), the more dificult it is to notice the hijacking. I'm not sure if this is relevant or not. It may be that if the network has a lot of RTT variance, then the more TCP connections that go by, the more likely you are to find one with a large RTT. Steve Bellovin of AT&T (co-author of the book "Firewalls and Internet Security" and author of the article "Security Problems in the TCP/IP Protocol Suite", which appeared in CCR for April, 1989) writes: One caveat -- a supershort connection -- which does indeed characterize much Web traffic -- might be hard to seize, because the parties will be shutting it down very fast. However, it's not clear that this is really a benefit of short connections. I believe that if Laurent's scheme succeeds at all, it will succeed close enough to the beginning of the connection to make this moot. And most Web connections are not "super-short"; they last for several RTTs, at least. Anyway, I would summarize this as: (1) we should be worried about TCP connection hijacking (2) persistent connections might make the problem better (3) or (less likely) they might make it worse (4) but cryptography is the only real solution. I am not confident enough in this analysis to say that it is conclusive, since there may be other TCP-hijacking attacks that we don't yet know about. Beth, if your security guru would like to provide a more detailed analysis supporting her/his warning, I think we should all see it. -Jeff
Received on Friday, 27 October 1995 12:16:42 UTC