Re: Counter-intuitive handling of --min-parallelism argument without --max-parallelism

[Chris Woodbury <woodbusy () gmail com>]

David-

Sorry for the extreme delay - your email slipped through the cracks.
You are correct that the simple check would solve the problem;
however, you would then have a situation where the scan engines could
have minimum values that were above the maximums. For example, if I
ran with --min-parallelism 350 (not necessarily a good idea, but for
the sake of discussion), then NmapOps::min_parallelism would be 350
and ::max_parallelism would be 0. In scan_engine, this would result in
low_cwnd being 350 and max_cwnd being 300. In idle_scan, max_groupsz
would be 100 and min_groupsz would be 350. Etc.

Off the top of my head, I'm not sure what the effects of this would
be, but it sounds like it could cause problems. If that is indeed the
case, I think it would be better to have each scan engine do its own
min < max checking.

What are your thoughts?
-chris

On Tue, Aug 2, 2011 at 12:11 AM, David Fifield <david () bamsoftware com> wrote:
> On Sun, Jul 10, 2011 at 01:02:47PM -0500, Chris Woodbury wrote:
> > I have been using the --min-parallelism argument for some time now to
> > speed up my scans, but I just recently discovered that it does not
> > have the effect I thought it did. My intention was to use
> > --min-parallelism to set a floor on the scan's parallelism but to let
> > it go higher dynamically. However, I discovered that using
> > --min-parallelism without --max-parallelism does not have this effect,
> > and that instead it sets the maximum to be the same as the minimum,
> > potentially slowing down scans on fast networks. This behavior is not
> > noted in the Nmap documentation and seems counter-intuitive to me.
> >
> > The reason for this is lines 482-483 of NmapOps.cc. The
> > NmapOps::min_parallelism and NmapOps::max_parallelism variables are
> > initially set to 0, which NmapOps.h defines as meaning their value
> > "has not been set." If no values for these are provided on the command
> > line, the various scan engines see the zeroes and use default values
> > instead. However, if only the minimum is provided on the command line,
> > when NmapOps.cc does a check on line 482 to see whether
> > min_parallelism > max_parallelism, it sets max_parallelism to be the
> > same as min_parallelism. This changes its meaning from "has not been
> > set" to 'has been set to X' and the engines' defaults are no longer
> > used.
> >
> > I can think of a few solutions off the top of my head:
> > 1) Move the min > max checks into the scan engines - Users would be
> > able to set a minimum while still using the default maximum. This is
> > my favorite and seems to best align with my intuition of what
> > --min-parallelism and --max-parallelism should be.
>
> Thanks for noticing this. Couldn't option #1 be done much more simply
> with something along the lines of
>
>   /* Prevent performance values from getting out of whack */
> -  if (min_parallelism > max_parallelism)
> +  if (max_parallelism != 0 && min_parallelism > max_parallelism)
>     max_parallelism = min_parallelism;
>
> David Fifield
_______________________________________________
Sent through the nmap-dev mailing list
http://cgi.insecure.org/mailman/listinfo/nmap-dev
Archived at http://seclists.org/nmap-dev/