| Abstract |
The paper describes a weakness in the pseudo random number generator (PRNG) in use by OpenBSD, Mac OS X, Mac OS X Server, Darwin, NetBSD, FreeBSD and DragonFlyBSD to produce random DNS transaction IDs (OpenBSD) and random IP fragmentation IDs (OpenBSD, Mac OS X, Mac OS X Server, Darwin, NetBSD, FreeBSD and DragonFlyBSD – the latter three only if the kernel flag net.inet.ip.random_id is 1). A technique is disclosed that allows an attacker to detect the algorithm used and predict its next values. This technique can be used to conduct DNS cache poisoning attack on OpenBSD DNS server (which is a modified BIND 9 server) in caching mode. A predictability algorithm is described that typically provides 8-10 possible guesses for the next transaction ID value, thereby overcoming whatever protection offered by the transaction ID mechanism. This enables a much more effective DNS cache poisoning than the currently known attacks against the OpenBSD DNS server. The net effect is that pharming attacks are feasible against OpenBSD caching DNS servers, without the need to directly attack neither DNS servers nor clients (PCs). A similar technique is disclosed to detect the algorithm used for the IP fragmentation ID generation (thereby enabling fingerprinting, traffic analysis and host alias detection for OpenBSD, Mac OS X, Mac OS X Server and Darwin (and NetBSD, FreeBSD, DragonFlyBSD, if the kernel flag net.inet.ip.random_id is 1), as well as detecting “missing” IDs, which can be used in nmap’s IdleScan method (as the “zombie” machine whose IP is used to scan the actual target host). IP fragmentation ID can also be used (in some cases) for TCP blind data injection.
|
top
