There are a surprising number of ways to get false positives in images, and I'm not sure how useful the current code is in stopping threats.
Posted: March 2nd, 2012, 10:55 AM
I remember discussing this issue with several people at SMF when this originally blew up way back.
The problem that isn't really being solved here is that you have legitimate images (and other files) being uploaded and checked for common things that could be malicious - but not nearly all of the things that could be, and unfortunately blocking others in the process like this one. The problem is that you can't realistically blacklist certain constructions and then provide exceptions, unless you're making very specific exceptions - and even then it's not reliable.
What might be better to do is to do separate types of validation based on the file type (assuming it's an image that you're trying to validate) and attempt to make sense of the file itself, e.g. if it's a GIF, step through and validate the image contents vs its headers and if there's any extra content, dump it. (Being sure to validate for animated images, of course)
It doesn't help that GIF, PNG and JPEG all legitimately allow for extra non-image information to be embedded in less than pleasant ways.