On 20 Mar 2023, at 11:48, Ilya Maximets wrote:
> On 3/20/23 11:44, Simon Horman wrote:
>> On Mon, Mar 20, 2023 at 09:51:48AM +0200, Roi Dayan wrote:
>>>
>>>
>>> On 14/03/2023 13:04, Simon Horman wrote:
>>>> On Mon, Mar 13, 2023 at 07:47:14PM +0200, Roi Dayan wrote:
>>>>>
>>>>>
>>>>> On 13/03/2023 14:16, Simon Horman wrote:
>>>>>> On Mon, Mar 13, 2023 at 12:31:49PM +0200, Roi Dayan wrote:
>>>>>>>
>>>>>>>
>>>>>>> On 13/03/2023 11:01, Eelco Chaudron wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>> On 13 Mar 2023, at 9:38, Roi Dayan wrote:
>>>>>>>>
>>>>>>>>> On 22/02/2023 12:30, Roi Dayan wrote:
>>>>>>>>>> Sometimes there is a need to clean empty chains as done in
>>>>>>>>>> delete_chains_from_netdev(). The cited commit doesn't remove
>>>>>>>>>> the chain completely which cause adding ingress_block later to fail.
>>>>>>>>>> This can be reproduced with adding bond as ovs port which makes ovs
>>>>>>>>>> use ingress_block for it.
>>>>>>>>>> While at it add the netdev name that fails to the log.
>>>>>>>>>>
>>>>>>>>>> Fixes: e1e5eac5b016 ("tc: Add TCA_KIND flower to delete and get
>>>>>>>>>> operation to avoid rtnl_lock().")
>>>>>>>>>> Signed-off-by: Roi Dayan <[email protected]>
>>>>>>>>>> ---
>>>>>>>>>> lib/netdev-offload-tc.c | 7 ++++---
>>>>>>>>>> lib/tc.c | 4 +++-
>>>>>>>>>> 2 files changed, 7 insertions(+), 4 deletions(-)
>>>>>>>>>>
>>>>>>>>>> diff --git a/lib/netdev-offload-tc.c b/lib/netdev-offload-tc.c
>>>>>>>>>> index 4fb9d9f2127a..9dd0aa2e2a85 100644
>>>>>>>>>> --- a/lib/netdev-offload-tc.c
>>>>>>>>>> +++ b/lib/netdev-offload-tc.c
>>>>>>>>>> @@ -524,7 +524,7 @@ delete_chains_from_netdev(struct netdev *netdev,
>>>>>>>>>> struct tcf_id *id)
>>>>>>>>>> */
>>>>>>>>>> HMAP_FOR_EACH_POP (chain_node, node, &map) {
>>>>>>>>>> id->chain = chain_node->chain;
>>>>>>>>>> - tc_del_flower_filter(id);
>>>>>>>>>> + tc_del_filter(id, NULL);
>>>>>>>>>> free(chain_node);
>>>>>>>>>> }
>>>>>>>>>> }
>>>>>>>>>> @@ -2860,8 +2860,9 @@ netdev_tc_init_flow_api(struct netdev *netdev)
>>>>>>>>>> error = tc_add_del_qdisc(ifindex, true, block_id, hook);
>>>>>>>>>>
>>>>>>>>>> if (error && error != EEXIST) {
>>>>>>>>>> - VLOG_INFO("failed adding ingress qdisc required for
>>>>>>>>>> offloading: %s",
>>>>>>>>>> - ovs_strerror(error));
>>>>>>>>>> + VLOG_INFO("failed adding ingress qdisc required for
>>>>>>>>>> offloading "
>>>>>>>>>> + "on %s: %s",
>>>>>>>>>> + netdev_get_name(netdev), ovs_strerror(error));
>>>>>>>>>> return error;
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>>> diff --git a/lib/tc.c b/lib/tc.c
>>>>>>>>>> index 4c07e22162e7..5c32c6f971da 100644
>>>>>>>>>> --- a/lib/tc.c
>>>>>>>>>> +++ b/lib/tc.c
>>>>>>>>>> @@ -2354,7 +2354,9 @@ tc_del_filter(struct tcf_id *id, const char
>>>>>>>>>> *kind)
>>>>>>>>>> struct ofpbuf request;
>>>>>>>>>>
>>>>>>>>>> request_from_tcf_id(id, 0, RTM_DELTFILTER, NLM_F_ACK, &request);
>>>>>>>>>> - nl_msg_put_string(&request, TCA_KIND, kind);
>>>>>>>>>> + if (kind) {
>>>>>>>>>> + nl_msg_put_string(&request, TCA_KIND, kind);
>>>>>>>>>> + }
>>>>>>>>>> return tc_transact(&request, NULL);
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> hi
>>>>>>>>>
>>>>>>>>> just pinging about this fix.
>>>>>>>>
>>>>>>>> Guess it’s waiting on your feedback on Simon’s reply:
>>>>>>>>
>>>>>>>> EC> The changes look good to me. Will it be worth adding a test case?
>>>>>>>>
>>>>>>>> SH> From my POV, yes, I think that would be nice.
>>>>>>>> SH> Roi, do you have any thoughts on this?
>>>>>>>>
>>>>>>>
>>>>>>> Oh. thanks for the updates.
>>>>>>> I missed the replies. if I'm not on the to/cc the mailing list emails
>>>>>>> are going to
>>>>>>> a different folder so I could catch emails when I am on to/cc better.
>>>>>>
>>>>>> Sorry, I should have taken more care to CC you.
>>>>>> I will try to do so in future.
>>>>>
>>>>> I'm having some trouble with adding a test for this.
>>>>>
>>>>> Internally I reproduce the issue with hw port with the following steps
>>>>>
>>>>> # ip l add dev bond0 type bond
>>>>> # ip l set dev enp8s0f0 master bond0
>>>>> # ovs-vsctl add-port ov1 bond0
>>>>> # tc qdisc show dev bond0
>>>>> qdisc ingress ffff: parent ffff:fff1 ingress_block 563 ----------------
>>>>> # tc filter add block 563 ingress prio 1 flower action drop
>>>>> # ovs-vsctl del-port ov1 bond0
>>>>> # ovs-vsctl add-port ov1 bond0
>>>>> # tc qdisc show dev bond0
>>>>> (no ingress_block)
>>>>>
>>>>>
>>>>> Without adding a slave the issue doesn't happen and for the autoconf
>>>>> test I wanted to use veth interface as a slave but the issue doesn't
>>>>> reproduce with it as well.
>>>>>
>>>>> So we do need the fix as it solves us the issue but there is
>>>>> something weird happening here. I'll try to look at this more
>>>>> later this week or next.
>>>>
>>>> Thanks Roi,
>>>>
>>>> I understand.
>>>>
>>>> FWIIW, I am happy to move forwards with the fix if you follow-up with a
>>>> test.
>>>>
>>>
>>> Hi Simon,
>>>
>>> I'm still trying this between other stuff i need to do.
>>> I couldn't reproduce this with veth. I'm not sure why or
>>> what it means. I'm still trying every now and then.
>>> I would be happy if we could still go with this fix to do
>>> chains cleaning without related to kind flower as it does
>>> help us and doesn't break anyone else.
>>
>> Ilya, Eelco, all,
>>
>> are there any objections to taking this patch now.
>> And allowing Roi to follow-up with a test later?
>>
>
> Fine by me. Though it's a bit concerning that the issue is not
> reproducible. Maybe we should update the comment in the code
> stating why we need to remove not only flower chains? To avoid
> messing up this part in the future again.
+1
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
