> Peter Memishian writes:
> >
> > > The problem: packets are leaving interface e1000g2 with the source IP
> > > of interface e1000g0. Is it ever possible?
> >
> > Yes, completely -- outbound interface selection and source address
> > selection are completely decoupled, and your netmask for e1000g2 causes it
> > to overlap with e1000g0. Perhaps you meant to configure e1000g2 with a
> > /24?
>
> Actually, no, they don't overlap:
>
> > > inet 10.24.1.101 netmask ffffff00 broadcast 10.24.1.255
> > > inet 10.21.3.101 netmask ffff0000 broadcast 10.21.255.255
>
> 24 != 21.
>
> But the rest is true, and this is just normal IP operation. We choose
> a source address to use based on the output interface _only if_ the
> sender doesn't already have a source address set because of a prior
> bind() or connect(), or because of IP_PKTINFO use.
>
> IP routes by destination address, not by source.
>
Hi, James

You are right, it was not a Solaris problem. As I found later, windows box had 
chosen wrong source IP for iscsi communication.
It happend because iscsi targets were not restricted to a portal IP.

So, Solaris correctly answered: it accepted packets on a correct interface, but 
send out to a wrong one since it saw the wrong source IP.

I had to find out how iscsi tptg works for Windows initiator.

--
Roman Naumenko
-- 
This message posted from opensolaris.org
_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to