Some errors were not properly propagated. Ensure proxy_resolve() returns
a negative error code and make the caller fatal() when unable to complete
proxy_chain initialization for whatever reason.
Report finished (likely failed) events as early as possible. Because the
corresponding IODs can have no I/O activity, they wouldn't be flagged as active
by the I/O engines and the events could end up being delivered at shutdown only
(or never).
Reported by d33tah on http://seclists.org/nmap-dev/2014/q2/409
Only FD_SET and FD_CLR were available. Added a FD_ISSET equivalent.
Implemented them as static inline instead of macros for consistency.
These functions abort() if the FD number is higher than FD_SETSIZE
(except on windows where no check is performed).
This wasn't using Nmap's included libpcap if no options were specified,
instead compiling nsock without pcap support if no libpcap could be
found (which breaks Nmap)
Using offsetof instead of actually dereferencing a 0 pointer is still
standards compliant (works under gcc -ansi, clang -ansi), and is more in
line with how Linux kernel defines this macro (kernel.h). Can't exactly
lift the kernel implementation because it uses gnu-extension typeof.
Not really a necessary change, but it was causing clang's address
sanitizer to complain.
I had intended the unlink to appy only to unix domain sockets, but it
was being called for every kind of local address.
http://seclists.org/nmap-dev/2013/q3/647
To avoid new GCC warnings about tempnam:
ncat_connect.c:789: warning: the use of `tempnam' is dangerous, better use `mkstemp'
Doing things this way has the same race condition as tempnam did,
because we are unlinking the file before binding it. (The race window is
smaller now.) The file must not exist before binding the Unix socket, or
else you get an "address already in use" error. Unlinking before binding
is the same thing that netcat-openbsd does. See this earlier thread:
http://seclists.org/nmap-dev/2012/q4/336.
This fixes a really bad bug that seems to have been there for a while. Canceling
a nsock connect SSL operation fails with fatal(). I have never seen it in real
life though.
Added a corresponding unit test.
This prevents nsock from iterating over the whole list of events at
each runloop, thus improving performance.
It made it necessary to have pointers from the msevents to the event
lists they belong to. The patch therefore also changes gh_list from
autonomous containers to embedded structures.
Added unit tests accordingly and cosmetic changes to make things look
more consistent.
These includes seem useless and nsock and nmap happily compile w/o them but the
nsock test suite doesn't. Temporarily restore them to their pre-r36100 state
until I can clean and streamline the inclusion tree.
