[libvirt] [RFC] VM which uses macvtap will not respond ping request when being migrated
A vm which uses macvtap will not respond ping request, when the vm is being migrated. I found that on the destination side the macvtap will send a IPv6 packet at the begin of migration to update the route table in switch, however VM is still on the src. In this case , what can I do to avoid VM's network disconnection Regards, Wangrui
On 03/18/2014 02:12 AM, Wangrui (K) wrote:
A vm which uses macvtap will not respond ping request, when the vm is being migrated.
I found that on the destination side the macvtap will send a IPv6 packet at the begin of migration to update the route table in switch, however VM is still on the src.
In this case , what can I do to avoid VM's network disconnection
Can you provide more details on this? What are the versions of libvirt, qemu, and kernel on your host system? Can you provide the <interface> section of the guest's config? Does this only occur when the guest has IPv6 enabled, and only for an IPv6 ping? What is the packet that is sent? At the beginning of the migration the guest is not running on the destination, and the host has no idea what the guest's IPv6 (or v4) address might be, so it can't send a IPv6 packet on behalf of the guest. The only thing I can think of could be something to do with the link-local address (the fe80:... address that is used for IPv6's discovery and autoconf protocols).
Thanks. libvirt 1.1.0 qemu 1.5.1 host kernel 3.0.58 guest hasn't ipv6 enabled. macvtap0 Link encap:Ethernet HWaddr 52:54:00:37:CD:68 inet6 addr: fe80::5054:ff:fe37:cd68/64 Scope:Link UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:0 errors:0 dropped:0 overruns:0 frame:0 TX packets:6 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:500 RX bytes:0 (0.0 b) TX bytes:468 (468.0 b) macvtap0 is belong to vm at the dest side. When migrate begins macvtap0 will send the packet like this: 07:57:59.233270 IP6 :: > ff02::16: HBH ICMP6, multicast listener report v2, 1 group record(s), lengt 28 07:58:00.096088 IP6 :: > ff02::1:ff6d:4665: ICMP6, neighbor solicitation, who has fe80::5054:ff:fe6d4665, length 24 07:58:00.123900 IP6 fe80::2a6e:d4ff:fe50:8d7 > ipv6-allrouters: ICMP6, router solicitation, length 1 07:58:00.175891 IP6 fe80::2a6e:d4ff:fe50:8d7 > ff02::16: HBH ICMP6, multicast listener report v2, 1 roup record(s), length 28 07:58:01.096089 IP6 fe80::5054:ff:fe6d:4665 > ipv6-allrouters: ICMP6, router solicitation, length 16 the virNetDevSetOnline in the libvirt makes macvtap0 up on the dest side, so macvtap0 sends the packets. I do nothing when VM is migrating. The IPv6 packets are sent automatically from the dest host when it is up. I guess this makes route table updated in switch. the configure of macvtap in xml like this <interface type='direct'> <source dev='eth4' mode='bridge'/> <model type='virtio'/> </interface> -----Original Message----- From: sendmail [mailto:justsendmailnothingelse@gmail.com] On Behalf Of Laine Stump Sent: Saturday, March 22, 2014 2:39 AM To: libvir-list@redhat.com Cc: Wangrui (K); chenliang (T); Zhaoyanbin (A); Michael S. Tsirkin Subject: Re: [libvirt] [RFC] VM which uses macvtap will not respond ping request when being migrated On 03/18/2014 02:12 AM, Wangrui (K) wrote:
A vm which uses macvtap will not respond ping request, when the vm is being migrated.
I found that on the destination side the macvtap will send a IPv6 packet at the begin of migration to update the route table in switch, however VM is still on the src.
In this case , what can I do to avoid VM's network disconnection
Can you provide more details on this? What are the versions of libvirt, qemu, and kernel on your host system? Can you provide the <interface> section of the guest's config? Does this only occur when the guest has IPv6 enabled, and only for an IPv6 ping? What is the packet that is sent? At the beginning of the migration the guest is not running on the destination, and the host has no idea what the guest's IPv6 (or v4) address might be, so it can't send a IPv6 packet on behalf of the guest. The only thing I can think of could be something to do with the link-local address (the fe80:... address that is used for IPv6's discovery and autoconf protocols).
On 03/25/2014 07:15 AM, Wangrui (K) wrote:
macvtap0 is belong to vm at the dest side. When migrate begins macvtap0 will send the packet like this:
07:57:59.233270 IP6 :: > ff02::16: HBH ICMP6, multicast listener report v2, 1 group record(s), lengt 28 07:58:00.096088 IP6 :: > ff02::1:ff6d:4665: ICMP6, neighbor solicitation, who has fe80::5054:ff:fe6d4665, length 24 07:58:00.123900 IP6 fe80::2a6e:d4ff:fe50:8d7 > ipv6-allrouters: ICMP6, router solicitation, length 1 07:58:00.175891 IP6 fe80::2a6e:d4ff:fe50:8d7 > ff02::16: HBH ICMP6, multicast listener report v2, 1 roup record(s), length 28 07:58:01.096089 IP6 fe80::5054:ff:fe6d:4665 > ipv6-allrouters: ICMP6, router solicitation, length 16
the virNetDevSetOnline in the libvirt makes macvtap0 up on the dest side, so macvtap0 sends the packets.
well, what happens here is that the newly created interface is subject to IPv6 autoconfiguration (you can see that it has obtained an IPv6 link local address). The resulting ICMP6 traffic will update your switch's MAC address tables as you have correctly observed. A workaround (but no more) *could be* to disable the IPv6 autoconfiguration for the interface in question (awkward/racy). A better approach would probably be to not online the macvtap interface until the domain is really running (here: after the migration was completed). Would need a rewrite of the network interface code in the qemu driver. -- Mit freundlichen Grüßen/Kind Regards Viktor Mihajlovski IBM Deutschland Research & Development GmbH Vorsitzender des Aufsichtsrats: Martina Köderitz Geschäftsführung: Dirk Wittkopp Sitz der Gesellschaft: Böblingen Registergericht: Amtsgericht Stuttgart, HRB 243294
participants (3)
-
Laine Stump -
Viktor Mihajlovski -
Wangrui (K)