I don't think this is an overflow. According to gdb, the spin lock
from CFSocketEnableCallBacks is crashing. Backtracing to the call of
CFSocketEnableCallBacks, I see that the two parameters (a CFSocket
pointer and a set of binary flags, respectively) are:

r3 0x0 0
r4 0x9 9

Thus CFSocketEnableCallBacks is being passed a NULL pointer (and, for
what it's worth, the flags are kCFSocketReadCallBack and
kCFSocketWriteCallBack). Why it crashes in the spinlock, I'm not sure,
but I imagine it might have to do with gdb not behaving nicely with
threads.

In any case, since I don't see any registers getting overwritten with
either 0x41 or 0x61 (since Safari converts it to lowercase), I don't
really see an overflow. It's a crash which you might want to report to
Apple, but it seems like it's probably just a case of dereferencing a
NULL pointer.

I think what's happening is that some handler somewhere for the URL is
seeing an overly large buffer and rejects it with a NULL pointer
somewhere for the socket which isn't checked for appropriately.

Thanks,
David

[ reply ]