CVE request -- memcached

[Jamie Strandboge <jamie () canonical com>]

FYI, this issue was recently pointed out to me:
http://code.google.com/p/memcached/issues/detail?id=102

A remote attacker who is allowed to connect to memcached can crash the
server by sending bad input. I've not investigated this to see if it is
more than a DoS.

People wanting to fix this may want to more thoroughly look at the
patch[1]. After a cursory glance at it, I'm not sure it is enough:
1. it uses:
  if (strcmp(ptr, "get ") && strcmp(ptr, "gets ")) {

Why not use something like (*totally* untested):
  if (strncmp(ptr, "get ", 5) && strncmp(ptr, "gets ", 5)) {

just in case ptr is not NULL terminated? I haven't checked if this is an
actual issue, but it certainly wouldn't hurt. '5' should probably be
changed to something more reasonable.

2. As I read the patch, couldn't an attacker send crafted input after
the 4 reallocs and then achieve the same thing (a DoS)?. Perhaps this
isn't a problem since it limits the object size to 1MB (according to the
FAQ [2]).


[1]http://github.com/memcached/memcached/commit/75cc83685e103bc8ba380a57468c8f04413033f9
[2]http://code.google.com/p/memcached/wiki/FAQ

-- 
Jamie Strandboge             | http://www.canonical.com

Attachment: signature.asc
Description: This is a digitally signed message part