Re: Why the "Linux goofiness" socket writability check?

[David Fifield <david () bamsoftware com>]

On Sat, Jul 18, 2009 at 09:55:57PM -0600, David Fifield wrote:
> On Sat, Jul 18, 2009 at 10:26:47PM -0500, Kris Katterjohn wrote:
> > On 07/18/2009 10:15 PM, David Fifield wrote:
> > > Fyodor didn't remember what this code was for but it dates back to 2000
> > > at least. (There was a problem with the permissions in the Subversion
> > > repository, but Fyodor changed it and now you can see log messages going
> > > back further.) We decided to remove it and see if it causes problems. It
> > > is removed in r14426.
> >
> > Upon first glance, the code starting around line 3692 of scan_engine.cc in
> > trunk contains the same logic (same comment anyway).  Just missed?
>
> Thanks for seeing that. I didn't know it was in scan_engine.cc too.
> There's a lot else besides--the Linux-specific code is almost 50 lines.
> The zero-length write and this message:
>       log_write(LOG_STDOUT, "Bad port %hu caught by 0-byte write: ", pport);
> is present in Nmap from the original Phrack article.

There's another reference to the test in the Phrack article.

http://nmap.org/p51-11.html

signal(SIGPIPE, SIG_IGN); /* ignore SIGPIPE so our 'write 0 bytes' test
                             doesn't crash our program!*/

I think it is probably about detecting connect errors with nonblocking
sockets, which appears to be nontrivial. Here's an article.

http://cr.yp.to/docs/connect.html

David Fifield

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