For some as-yet-unknown reason, the IP assigned by the user-net DHCP
server occasionally switches to 10.0.2.16. I suspect this happens while
renewing the lease while I'm away, since that would explain why the
headless VM is unreachable when I come back on mondays - then when booting
with SDL display, I find out the IP is 10.0.2.16.
This would not be a problem at all, if the port redirections would know
about that, but a tcpdump in the VM shows that my "-redir
tcp:10080::10080" is indeed taken as "tcp:10080:10.0.2.15:10080", even
though the attributed IP is 10.0.2.16. Not sure how qemu could be taught
to do something sensible here, without breaking multi-NIC setups.
But with some luck we should not have to bother with it, if only the
problem is just that the lease renewal is broken.
Bringing the (gentoo/dhcpcd) VM back on its feet seems to require all of
the following steps, which is quite error-prone (and time-consuming when a
single step is missed;):
- "dhcpcd -k", which clears /var/cache/dhcpcd-eth0.cache
- removing /var/lib/dhcpc/*
- restarting qemu itself so that it forgets about the lease as well
Would it make sense to be able to tell qemu "do not ever issue a lease on
an IP different from 10.0.2.15", and possibly make that the default for
single-NIC VMs ?
--
Yann Dirson <[EMAIL PROTECTED]>
<[EMAIL PROTECTED]>
_______________________________________________
Qemu-devel mailing list
Qemu-devel@nongnu.org
http://lists.nongnu.org/mailman/listinfo/qemu-devel
