[lugm.org] iscsi traffic over ipv6

[Nishal Goburdhan]

On 04 Jul 2013, at 8:37 PM, "Keshwarsingh Nadan" <keshwarsingh.nadan at servihoo.net> wrote:

> Hello,
> 
> Currently facing a strange issue with iscsi over ipv6. For security reasons I've recently configured my core router to stop sending routing advertisements, & disabled servers from accepting ra packets / stateless autoconfig & link-local addresses.

servers shouldn't auto-configure themselves from RAs.  but there is other information in the RA packet that's useful.
typically, you'd want to configure your router to issue RAs with an 'a' flag - ie. do not announce prefix - but still announce that you're a router, and other details (mtu, etc)

you should not disable link-local addresses.  this is bound to create weirdness.  
link local addresses are part of the scope of IPv6.
i'm guessing you fiddled with this:  net.inet6.ip6.auto_linklocal: 1    ?


> The servers in question runs centos 6.4 with iscsi daemon set to automatically start at boot time. I've noticed it takes approx 25 mins to start the daemon during boot time.
> 
> Re enabling ra advertisements on the router results the servers to boot into OS in less than 30 secs after POST.
> Anyone here ever came across such issues ?

we've used - and ttbomk, are using - iscsi over IPv6;  we suppress the prefix advertisement (right -  who wants a server to auto-config an address?) and we've not seen issues with iscsi.  i've never really looked at it to see if the RA made a difference, to be honest, but i can't think of why it would.

can you prove more detail? 
for example - when it works, is enabling RAs on the router, the only thing that you're doing to get it to work? 

what is the topology of the NAS / server connection - same directly connected network, or do they speak via the router? 
are you sure you're not also doing something to NA? 
have you tried to packet capture during the 'outage' to see what's happening?

--n.