[ovs-dev] [PATCH 6/6] vxlan: Create and delete tnl_backers in type_run()

Kyle Mestery (kmestery) kmestery at cisco.com
Thu Feb 14 20:19:29 UTC 2013


On Feb 14, 2013, at 12:49 PM, Ben Pfaff <blp at nicira.com> wrote:
> On Thu, Feb 14, 2013 at 06:47:56PM +0000, Kyle Mestery (kmestery) wrote:
>> On Feb 14, 2013, at 12:33 PM, Ben Pfaff <blp at nicira.com> wrote:
>>> On Thu, Feb 14, 2013 at 09:37:30AM -0500, Kyle Mestery wrote:
>>>> Garbage collect tnl_backers during type_run(). Add new
>>>> tnl_backers if a VXLAN ports UDP port changes.
>>>> 
>>>> Signed-off-by: Kyle Mestery <kmestery at cisco.com>
>>> 
>>> The error handling here is bad.  If it fails to add or remove a port,
>>> then ovs_assert() aborts the whole ovs-vswitchd process.
>>> 
>> I can change the asserts into something more sane.
>> 
>>> But I don't fully understand the code.  It looks to me like the first
>>> time we encounter a particular dp_port_name in the inner HMAP_FOR_EACH,
>>> we transfer that dp_port_name from tmp_simap to backer->tnl_backers.
>>> Straightforward enough.  However, I believe that the second time we
>>> encounter that same dp_port_name, we will not find it in tmp_simap
>>> (because we deleted it) and will therefore try to add a new port for
>>> it.   I believe that will fail, because there is already a port with
>>> that dst_port, and then we'll assert-fail the process.
>>> 
>>> Am I reading the code correctly?
>>> 
>> Actually, the second time the same dp_port_name is encountered,
>> we first look and see if we need to reconfigure it, and then we check if
>> it's already in backer->tnl_backers, which it will be since we added it
>> there the first time we found it. I don't see it adding the port twice, unless
>> I'm misreading it.
> 
> Looking again, you're right, thanks.
> 
>>> I've now pushed patch 1-4 to master, waiting for an ack from Jesse on
>>> #5, waiting for your reply on this one.
>>> 
>> Thanks Ben! Let me know if you want me to modify the error handling
>> behavior.
> 
> Please do, I'm not sure of the correct behavior.

I sent a revised version with error handling in lieu of ovs_asserts() for
port creation and deletion. For the netdev_get_tunnel_config() assert,
Ethan indicated that should never happen, so I'd rather leave that as an
ovs_assert().

Thanks,
Kyle


More information about the dev mailing list