Re: [PATCH net-next-2.6] bonding: allow arp_ip_targets to be on a separate vlan from bond device

[Jay Vosburgh <fubar@us.ibm.com>]

Andy Gospodarek <andy@greyhouse.net> wrote:

>On Mon, Nov 30, 2009 at 04:00:38PM -0800, Jay Vosburgh wrote:
>> Andy Gospodarek <andy@greyhouse.net> wrote:
>> 
>> >This allows a bond device to specify an arp_ip_target as a host that is
>> >not on the same vlan as the base bond device.  A configuration like
>> >this, now works:
>> >
>> >1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue
>> >    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
>> >    inet 127.0.0.1/8 scope host lo
>> >    inet6 ::1/128 scope host
>> >       valid_lft forever preferred_lft forever
>> >2: eth1: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master bond0 qlen 1000
>> >    link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff
>> >3: eth0: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master bond0 qlen 1000
>> >    link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff
>> >8: bond0: <BROADCAST,MULTICAST,MASTER,UP,LOWER_UP> mtu 1500 qdisc noqueue
>> >    link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff
>> >    inet6 fe80::213:21ff:febe:33e9/64 scope link
>> >       valid_lft forever preferred_lft forever
>> >9: bond0.100@bond0: <BROADCAST,MULTICAST,MASTER,UP,LOWER_UP> mtu 1500 qdisc noqueue
>> >    link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff
>> >    inet 10.0.100.2/24 brd 10.0.100.255 scope global bond0.100
>> >    inet6 fe80::213:21ff:febe:33e9/64 scope link
>> >       valid_lft forever preferred_lft forever
>> 
>> 	I'm not quite clear here on exactly what it is that doesn't
>> work.
>> 
>> 	Putting the arp_ip_target on a VLAN destination already works
>> (and has for a long time); I just checked against a 2.6.32-rc to make
>> sure I wasn't misremembering.
>> 
>> 	Perhaps there's some nuance of "not on the same vlan as the base
>> bond device" that I'm missing.  What I see working before me is, e.g., a
>> bond0.777 VLAN interface atop a regular bond0 active-backup with a
>> couple of slaves; bond0 may or may not have an IP address of its own.
>> The arp_ip_target destination is on VLAN 777 somewhere.
>
>Do you have net.ipv4.conf.all.arp_ignore set to 0 and/or an IP address
>assigned on bond0?  I can easily reproduce this with no IP on bond0 and
>net.ipv4.conf.all.arp_ignore = 1.
>
>I can't say for sure that the sysctl setting makes a difference, but I
>have that on all my test rigs, so it's worth mentioning.
>
>> 	Is this what your patch is meant to enable, or is it something
>> different?  I'm pulling down today's net-next to see if this is
>> something that broke recently.
>> 
>
>I first tested and found the problem while running 2.6.30-rc series
>after it was reported to be a problem on RHEL5.  It's not clear how long
>it has been broken, but this situation is odd enough that it probably
>never worked as it was never tested.

	I tried it with both arp_ignore set to 1 and 0, and with the
bond0 interface with and without an IP address.  It works fine in all
four cases.  I'm using net-next-2.6 pulled earlier today; it claims to
be 2.6.32-rc7.

	I've tested "ARP monitor over VLAN" in the past, so it's worked
for me before.  Heck, it's working right now.

	I thought maybe you have "arp_validate" enabled (which doesn't
work over a VLAN), but your patch doesn't help there, so presumably not.
Fixing that is a totally separate adventure into hook-ville; I'd briefly
hoped you'd found a better way.

	When it's failing, are you getting any messages in dmesg?  I'm
wondering specifically about any of the various routing-related things
that bond_arp_send_all might kick out.

	Do you have any of the other arp_whatever sysctls enabled?  My
configuration has them all set to zero; maybe one of those is messing
something up.

	-J

---
	-Jay Vosburgh, IBM Linux Technology Center, fubar@us.ibm.com