On 19.03.2019 18:51, Ilya Maximets wrote:
> On 19.03.2019 12:23, Eelco Chaudron wrote:
>> When debugging an issue we noticed that by accident someone has changes the
>> bridge datapath_type to netdev, where it's clearly a kernel (system bridge)
>> with physical and tap devices. Unfortunately, this is not something you
>> will easily spot, as the bridge datapath type value is not shown by default.
>>
>> In addition, OVS is not warning you about this potential mismatch in
>> interface and bridge datapath.
>>
>> I'm sending out this patch as an RFC as I could not find a clear
>> demarcation between bridge datapath types and interface datapath types.
>> The patch below will at least warn for netdev bridges with system
>> interfaces. But no warning will be given for some unsupported virtual
>> interfaces. For system bridges, the dpdk types will no be recognized as
>> system/virtual interfaces (unless the name exists) which will result in
>> an error.
>>
>> Signed-off-by: Eelco Chaudron <[email protected]>
>> ---
>
> Hi.
>
> Hmm. What do you mean under misconfiguration?
> In practice, userspace datapath is able to open "system" type netdevs.
> It's supported by netdev-linux to open system network devices via socket.
> And these devices has "system" type.
> On 'datapath_type' changes, bridge/ofproto will be destroyed and re-created.
> All the ports from kernel datapath will be destroyed and new ones created
> in userspace. All the ports should still work as usual.
>
> The only possible issue will be that this bridge will no longer communicate
> with other bridges, because they're in different datapaths now.
For this issue, probably, following warning will give a clue to a user:
diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c
index 21dd54bab..df51aaa3a 100644
--- a/ofproto/ofproto-dpif.c
+++ b/ofproto/ofproto-dpif.c
@@ -3557,6 +3557,10 @@ ofport_update_peer(struct ofport_dpif *ofport)
break;
}
+ if (!ofport->peer) {
+ VLOG_WARN_RL(&rl, "No usable peer '%s' exist for patch port '%s'.",
+ peer_name, netdev_get_name(ofport->up.netdev));
+ }
free(peer_name);
}
---
I could send this as a proper patch.
Best regards, Ilya Maximets.
>
> So, below warning will be printed on any attempt to add legitimate
> system network interface to the userspace bridge.
> "system" devices supported by both datapaths, but "dpdk" devices supported
> only by userspace, that is why you see the error in case of changing to
> 'datapath_type=system'.
>
> Maybe I'm missing something? What is the issue you're trying to address?
>
> Best regards, Ilya Maximets.
>
>> vswitchd/bridge.c | 6 ++++++
>> 1 file changed, 6 insertions(+)
>>
>> diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c
>> index a427b0122..42c33d1d9 100644
>> --- a/vswitchd/bridge.c
>> +++ b/vswitchd/bridge.c
>> @@ -1808,6 +1808,12 @@ iface_do_create(const struct bridge *br,
>> goto error;
>> }
>>
>> + if (!iface_is_internal(iface_cfg, br->cfg)
>> + && !strcmp(br->type, "netdev")
>> + && !strcmp(netdev_get_type(netdev), "system")) {
>> + VLOG_WARN("bridge %s: interface %s is a system type where the
>> bridge "
>> + "is a netdev one", br->name, iface_cfg->name);
>> + }
>> VLOG_INFO("bridge %s: added interface %s on port %d",
>> br->name, iface_cfg->name, *ofp_portp);
>>
>>
>
>
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
