pv@uscvax.UUCP (Peter Vanderbilt) (05/15/86)
We have a VAX 750 running 4.2 gatewaying between two ethernets. For some
reason, connections running over the second ethernet have the return
(source) IP address of the first ethernet. This causes minor problems like
machines on the second net generating arp requests that are not responded
to. Also the address of the first ethernet happens to end in .255 (class C
host number is all 1's) which some systems don't like (we're planning to
change this).
The code in ip_output.c looks OK which implies that higher level protocols
are (incorrectly) filling in the source IP address.
Has this been seen before? Is there a known fix?
--
Peter Vanderbilt
arpa: pv@usc-cse.usc.edu
csnet: pv@usc-cse
uucp: pv@uscvax.uucp or
...!{{decvax,ucbvax}!sdcsvax,hplabs,allegra,trwrb}!sdcrdcf!uscvax!pvjerry@wg3b20.UUCP (05/22/86)
> > We have a VAX 750 running 4.2 gatewaying between two ethernets. For some > reason, connections running over the second ethernet have the return > (source) IP address of the first ethernet. This causes minor problems like > machines on the second net generating arp requests that are not responded > to. Also the address of the first ethernet happens to end in .255 (class C > host number is all 1's) which some systems don't like (we're planning to > change this). To: pesnta!hplabs!sdcrdcf!usc-oberon!bacall!uscvax!pv Subject: Re: Gateway generates wrong IP returns address (4.2, Vax) In-reply-to: your article <29@uscvax.UUCP> This is a bug in the 4.2 code for TCP sockets. If you look at in_pcb.c in routine in_pcbconnect you will see where the source IP address is chosen for the IP header. Instead of choosing the source address based on the interface the packet will be sent to, it chooses the address of the FIRST interface in the ifnet list that has an address family of AF_INET. Of course this is the first DEUNA and thus the IP source address is that of this first DEUNA. There has been a fix for this problem (when I saw it it was combined with a TCP maxsegsize change down by Mike Muus) and I will post it to the net or to your mailbox directly. Regards, Jerry Scott