[ipxe-devel] Curious routing problem with iPXE

[Michael Brown]

On Wednesday 09 Feb 2011 23:25:18 Lee Bradshaw wrote:
> I've now set up a native Windows iSCSI server and am booting from it. All
>  is working fine, except that the routing to the server is odd. My PC is
>  .244; the server is .2. When I list the routing table I get:
> 
> IPv4 Route Table
> 
===========================================================================
> Active Routes:
> Network Destination        Netmask          Gateway       Interface  Metric
>           0.0.0.0          0.0.0.0       10.10.10.1     10.10.10.244    266
>        10.10.10.0    255.255.255.0         On-link      10.10.10.244    266
>        10.10.10.2  255.255.255.255       10.10.10.1     10.10.10.244    266
> ...
> 
> You will notice the curious route to the server via a gateway at
> 10.10.10.1. This seems to be caused by iPXE not correctly processing the
> subnet from DHCP, which was 255.255.255.0. I know this is what it read
> from the DHCP server because it says so during the boot.

Interesting.  I'm pretty sure if you type "route" at the iPXE command line, 
you'll find that iPXE's routing table doesn't include a dedicated host route to 
10.10.10.2, if only because iPXE doesn't even support a concept of routing 
beyond a simple "network address, subnet mask, gateway address" triplet. :)

iPXE passes IP configuration information to Windows via the iBFT, which can 
contain only a local IP address, subnet prefix, and gateway address; there's no 
way that iPXE could pass host-specific routes to Windows even if it wanted to.
 
> In many installations this wouldn't be a problem - although it's an
> unnecessary overhead - but in my case the gateway at 10.10.10.1 is on a
> 100Mb connection to the switch, whereas the rest of the network is 1000Mb.
> As you can imagine, this means the performance of my disks is not what it
> should be.
> 
> I've been able to work around it by adding a static route (DHCP 249) to
> override the errant entry, but it would be nice if I didn't have to do
>  this. Any ideas?

I vaguely (and quite possibly inaccurately) remember that Shao Miller has 
encountered this problem before.  Shao: any thoughts?

Michael