Hopefully, that shouldn't be exploitable. I'm pretty sure the kernel
policy is that userspace shouldn't be able to trigger BUG_ONs
deliberately, so removing one in favour of just dereferencing the null
pointer ought to be safe assuming there isn't a bug elsewhere. (Of course,
if there is, it could turn it into a security vulnerability.)
Posted Jul 20, 2009 12:46 UTC (Mon) by spender (subscriber, #23067)
[Link]
Andrew Morton is wrong, a BUG_ON is in no way equivalent to just dereferencing a null. What if I have my page mapped at NULL?
If that author adds one more call to smack_netlabel and forgets to check for sock->sk == NULL before calling it, instead of just causing an OOPS he's now potentially created a vulnerability which can result in arbitrary code execution.