Re: [nmap-svn] r23129 - nmap-exp/david/nmap-ipv6/libdnet-stripped/src

[David Fifield <david () bamsoftware com>]

On Sun, May 08, 2011 at 08:24:03PM -0700, David Fifield wrote:
> On Sun, May 08, 2011 at 08:21:24PM -0700, commit-mailer () insecure org wrote:
> > Author: david
> > Date: Sun May  8 20:21:24 2011
> > New Revision: 23129
> >
> > Log:
> > Use 4-byte sockaddr alignment, not 8-byte, in rt_msghdr.
> >
> > 4 bytes seems to be the right number on OS X, despite Unix Network
> > Programming that states 8 bytes.
> >
> > Modified:
> >    nmap-exp/david/nmap-ipv6/libdnet-stripped/src/route-bsd.c
> >
> > Modified: nmap-exp/david/nmap-ipv6/libdnet-stripped/src/route-bsd.c
> > ==============================================================================
> > --- nmap-exp/david/nmap-ipv6/libdnet-stripped/src/route-bsd.c       (original)
> > +++ nmap-exp/david/nmap-ipv6/libdnet-stripped/src/route-bsd.c       Sun May  8 20:21:24 2011
> > @@ -43,8 +43,14 @@
> >  
> >  #include "dnet.h"
> >  
> > +/* Unix Network Programming, 3rd edition says that sockaddr structures in
> > +   rt_msghdr should be padded so their addresses start on a multiple of
> > +   sizeof(u_long). But on 64-bit Mac OS X 10.6 at least, this is false. Apple's
> > +   netstat code uses 4-byte padding, not 8-byte. This is relevant for IPv6
> > +   addresses, for which sa_len == 28.
> > +   http://www.opensource.apple.com/source/network_cmds/network_cmds-329.2.2/netstat.tproj/route.c */
> >  #define ROUNDUP(a) \
> > -   ((a) > 0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
> > +   ((a) > 0 ? (1 + (((a) - 1) | (sizeof(uint32_t) - 1))) : sizeof(uint32_t))
> >  
> >  #ifdef HAVE_SOCKADDR_SA_LEN
> >  #define NEXTSA(s) \
>
> I spent a few hours tracking this down to make routes work in the IPv6
> branch. I am not sure this patch is universally right, because every
> source that I've found, except for Apple's source code, says that this
> rounding should be to sizeof(unsigned long), which on this platform is 8
> bytes. But 8 bytes definitely doesn't work for me, and 4 bytes does.

Daniel Roethlisberger kindly did some tests and found that 4-byte
rounding is definitely wrong on 64-bit FreeBSD. So I think we will have
to make this detectable by the configuration program.

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