Windows doesn't support selectable pcap fds, so instead of fiddling with
adapter read timeouts via PacketSetReadTimeout, we get an event handle
with pcap_get_event and WaitForSIngleObject. This means we don't need to
extract the adapter from the pcap_t, which is not part of the libpcap
API and was causing crashes switching between the libpcap versions used
by WinPcap and Npcap.
Two issues here: First, IP protocol scan can send packets with protocol
58 (ICMPv6) even over IPv4. This led to a bad interaction where the
packet was created (in build_protoscan_packet) without a data payload,
but setIP tried to set the packet's Identifier field (present in both
ICMPv6 and ICMP Echo Request packets), leading to a heap buffer
overflow. Instead, we now only try to set this identifier when the IP
version matches the ICMP version, indicating that we set the data
payload.
The other issue was a out-of-bounds read while packet tracing when an
ICMPv6 packet without a payload was sent or received, due to trying to
read the type and code. Now we check that the data length is sufficient
to contain an ICMPv6 header before attempting to read one.
Credit LLVM/Clang's AddressSanitizer with catching these bugs.
The formats has been tested and verified on Mac OS X 10.8.5,
Mac OS X 10.11.5 and Ubuntu 14.04 LTS, all on x86_64 machines
and OSs. It mainly silences warnings. There were no warnings on
Ubuntu but a few on Mac OS, so the fix is intended to silence
warnings on Mac OS whithout triggering new warnings on other OSs.
Example of warnings previously encountered:
netutil.cc:2828:74: warning: format specifies type 'unsigned short'
but the argument has type 'int' [-Wformat]
