Re:Cisco routers in at least 4 countries infected by highly ...
by roel_v on Wednesday September 16, 2015 @04:24AM
(#10225339)
Sure, my point was that open source doesn't matter anything for this attack vector. I have yet to see anyone with reproducible builds being used and cared about in production. It's too expensive, and impractical. How are you going to verify that a system you just installed is OK? An attacker would've patched the md5sum (or other hashing) utility as well, so you'd need to mount you whole filesystem from another (trusted) machine, run the checksum on each file from that other machine, and compare with a known-good (through a reputation network) list.
And in the mean time, a million monkeys have social-engineered your secretary's machine through a Farmville js trick and pwned your network 3 ways from Sunday.
I have an example of a binary where the difference between a remotely exploitable service and a fixed version of the same service is a single bit. That bit (or some other bit that also changes the code's functionality in an important way) can be changed by any person or (potentially compromised) computer in the chain of compilation, packaging, and distribution unless someone has a way to tell whether what those people and computers assert is "correct". If the binaries routinely change appreciably every time someone cuts a software release, it's going to be expensive to go back and check whether there's also a malicious change, bearing in mind that the person whose computer made the change doesn't have to have been malicious.