[ovs-discuss] null ptr exception in ovs_vport_get_stats+0x6a/0x130 [openvswitch]

[Flavio Fernandes]

Hi folks,

wrt http://openvswitch.org/pipermail/discuss/2015-December/019760.html

I got some answers, followed by another question. :) It now makes sense to
me why folks create a tap port instead of simply using the 'br' internal
interface.

The issue is indeed related to using VMs on top of  an internal ovs
interface.

As soon as I:

1) use tap port as the bridged port from vagrant (virtualBox provider);
2) did _not_ set type of tap to internal
[3]) enabled firewall to allow packets from network 192.168.50.x -- w/out
specifying
    any particular interface

I was able to create multiple VMs on 'top' of that same tap ovs port w/out
any problems.

So, I'm a happy camper, but can't help but worry a little about the
fragility of the
system when one attempts to use a port type internal 'directly' as bridged.
The fix
I have in mind is relatively simple:  add a check in  internal_dev_get_stats
<http://lxr.oss.org.cn/ident?i=internal_dev_get_stats>
to gracefully handle cases when ovs_internal_dev_get_vport
<http://lxr.oss.org.cn/ident?i=ovs_internal_dev_get_vport> returns null. Too
simple?

Thanks,

-- flaviof

[3]:   firewall-cmd --permanent --zone=trusted --add-source=192.168.50.0/24




On Mon, Dec 28, 2015 at 11:08 AM, FlaviofOvsMl <ovs at flaviof.com> wrote:

> Hello ovs gurus,
>
> I seem to have hit a null exception in openvswitch that has not
> been mentioned in the mailing list [ml] for the last 3 months.
> The full stack trace and the kdump core are here [1] and here [2].
>
>    ovs_vport_get_stats+0x6a/0x130 [openvswitch]
>    internal_dev_get_stats+0x39/0xb0 [openvswitch]
>    dev_get_stats+0x6e/0x200
>    rtnl_fill_ifinfo+0x459/0xf60
>    blk_rq_map_sg+0x9b/0x210
>    swiotlb_map_sg_attrs+0x78/0x150
>    update_curr+0xcc/0x150
>    account_entity_dequeue+0xae/0xd0
>    dequeue_entity+0x106/0x510
>
> I'm running centos 7.2 (3.10.0-327.3.1.el7.x86_64 x86_64) with
> virtualBox 5 (VirtualBox-5.0-5.0.12_104815_el7-1.x86_64). In
> my setup, I start with an ovs bridge br1 which has the internal
> address 192.168.50.1  [3]. With that, I see the address added
> as the internal port br1 of bridge br1, as shown here [4]. I believe
> that is the right config for using a vm with ovs. I have come to
> that conclusion based on the faq [faq] by Ben.
>
> I attempt to start a vm using Vagrant, using the following
> Vagrantfile [5]. The important line in that Vagrant file that causes
> the issue is this:
>
>    node.vm.network "public_network", ip: "192.168.50.254", bridge: "br1"
>
> with that, I'm attempting to make a bridged port on br1 for
> the vm, which is given the ip address 192.168.50.254.
> I have also seen  folks using the vm on top of a tap [tap] [tap2], but I
> don't
> quite understand why that is needed. Independent of that, I hit this
> kernel panic whether I use the tap interface or the bridge's internal
> port.
>
> As the trace show, the crux of the problem is that the code is attempting
> to extract stats from an internal ovs port, where the function
> ovs_vport_get_stats <http://lxr.oss.org.cn/ident?i=ovs_vport_get_stats>
> is given null as the vport param. One
> potential fix would be to check for null in
> internal_dev_get_stats
> <http://lxr.oss.org.cn/ident?i=internal_dev_get_stats> but I would like
> to hear from the ovs gurus
> in this mailing list if that is the best way to go. I have tried both ovs
> 2.3.2
> and 2.4.0 and hit this issue on both versions. Given this is a kernel
> crash,
> the user-space does not really matter.
>
> I'd be happy to propose a patch in the code, once I hear from you.
>
> Thanks and happy new year!
>
> -- flavio
>
> [ml]: http://openvswitch.org/pipermail/discuss/
> [1]: https://gist.github.com/daf71120b99a5d57cb26
> [2]:
> https://drive.google.com/folderview?id=0BxKY67eIRXcDNmNYU05aVWxWQmc&usp=sharing
> [3]: https://gist.github.com/d6fbf119c99dbba61b43
> [4]: https://gist.github.com/1a3d0795574c42a6530f
> [faq]: https://github.com/openvswitch/ovs/blame/master/FAQ.md#L971
> [tap]: https://youtu.be/rYW7kQRyUvA?t=7m22s
> [tap2]:
> https://ariscahyadi.wordpress.com/2013/07/16/virtual-networking-for-virtualbox-using-open-vswitch/
> [5]: https://gist.github.com/446d7b5aa59164dd2bc0
> [6]: http://lxr.oss.org.cn/source/net/openvswitch/vport.c#L291
> [7]: http://lxr.oss.org.cn/source/net/openvswitch/vport-internal_dev.c#L53
>
>
>
> _______________________________________________
> discuss mailing list
> discuss at openvswitch.org
> http://openvswitch.org/mailman/listinfo/discuss
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20160104/6013e212/attachment-0002.html>