Re: BUG: ncat does not work when file is used as input

[Henri Doreau <henri.doreau () gmail com>]

Hi,

I had to make a tough choice: either introduce a cheesy hack into the
code of ncat or bloat nsock.

As explained in the link I posted earlier[1], epoll(7) can't deal with
regular files the same way it does with sockets and pipes. The effect
is what Michal could observe:
"ncat localhost 1234 < file.c" fails, although "cat file.c | ncat
localhost 1234" would work.

The right way(tm) to do polling on regular file would involve two kind
of recent linux-specific mechanisms which are eventfd and aio. I first
tried this approach, and despite being fun this brings an absolutely
undesired level of complexity to nsock. Especially since it's only for
this specific case. Nsock, as name suggests, mostly handles sockets.

To sum up what I did (maybe someone comes up with a smarter approach):
  - hook epoll_ctl() when registering/unregistering events for STDIN_FILENO
  - create an eventfd(), stored in the engine_info structure (hurk!)
  - Schedule an aio_read() as a replacement for the actual read().
This is where it started to become way too intrusive.

The second approach I tried (patch attached) is hackish but simple,
efficient and far less intrusive. If we're on linux (then epoll() is
expected to be used, but not even necessarily) and stdin is a regular
file, then we enforce the use of the select(2)-based nsock engine.

Regards.

[1] https://bugzilla.kernel.org/show_bug.cgi?id=15272

-- 
Henri

Attachment: ncat_reg_stdin.diff
Description:

_______________________________________________
Sent through the nmap-dev mailing list
http://cgi.insecure.org/mailman/listinfo/nmap-dev
Archived at http://seclists.org/nmap-dev/