[ovs-dev] [PATCH] ovn: Fix IPv6 DAD failure for child ports

Guru Shetty guru at ovn.org
Thu Oct 4 21:02:10 UTC 2018


On Thu, 4 Oct 2018 at 13:29, Numan Siddique <nusiddiq at redhat.com> wrote:

>
> Thanks for the review Guru.
>
> Please see below for the comments.
>
>
>
> On Fri, Oct 5, 2018 at 12:09 AM Guru Shetty <guru at ovn.org> wrote:
>
>>
>>
>> On Mon, 17 Sep 2018 at 02:24, <nusiddiq at redhat.com> wrote:
>>
>>> From: Numan Siddique <nusiddiq at redhat.com>
>>>
>>> When a child vlan interface is created inside a VM, the below kernel
>>> message
>>> is seen and IPv6 doesn't work on that interface.
>>>
>>
>> On which interface doesn't IPv6 work? On the Vm's interface or on the
>> container's interface?
>>
>>
>
> It is seen on the container's interface.
>
>>
>>> [  138.000753] IPv6: vlan4: IPv6 duplicate address <IPv6 LLA> detected!
>>>
>>> When a child port sends a broadcast packet, OVN delivers the same
>>> packet back to the child port (and hence the DAD check fails).
>>>
>>
>> Is this limited to IP broadcast only? Or does this happen with mac
>> broadcast too? (I am surprised why I hadn't seen this behavior with ARP
>> packets of containers).
>>
>
> It happens with mac broadcast. You can actually reproduce the issue by
> applying the code in tests/ovn.at without this patch.
>
> Looks like I have confused you with the patch.  Let me try to give an
> example. Suppose we have switch - sw0 - with ports p1, p2 and p3 and all of
> them reside in the same chassis.
> When p1 sends a broadcast/multicast packet, the below logical flow is hit
>
> table=17(ls_in_l2_lkup      ), priority=100  , match=(eth.mcast),
> action=(outport = "_MC_flood"; output;)
>
> This translates to the OF flow
> - metadata=0x1,dl_dst=01:00:00:00:00:00/01:00:00:00:00:00
> actions=load:0xffff->NXM_NX_REG15[],resubmit(,32)
>
> In the table 33 if reg15 has 0xffff, it will resubmit the flow with reg15
> set to each logical port
>
> i.e reg15=0xffff,metadata=0x1
> actions=load:0x1->NXM_NX_REG15[],resubmit(,34),load:0x2>NXM_NX_REG15[],resubmit(,34),load:0x3->NXM_NX_REG15[],resubmit(,34),load:0xffff->NXM_NX_REG15[]
>
> In the normal case, the packet will not be delivered back to p1 and
> instead will be dropped in table 34 (OFTABLE_CHECK_LOOPBACK)
> since reg10[0] is not set.
>
> table=34, n_packets=0, n_bytes=0,
> priority=100,reg10=0/0x1,reg14=0x2,reg15=0x2,metadata=0x1 actions=drop
>
> When the register reg10[0] is set, it means the packet needs to be loop
> backed to the source port. So table 64 has flows to clear the in_port
> i,.e
> table=64,  priority=100,reg10=0x1/0x1,reg15=0x1,metadata=0x1
> actions=push:NXM_OF_IN_PORT[],load:0->NXM_OF_IN_PORT[],resubmit(,65),pop:NXM_OF_IN_PORT[]
>
> Now if we take the example of child ports, let's say "p2" is the child
> port of "p1" and in_port of p1 is 1
> In table 0, we will see 2 flows for in_port=1
>
> match=(in_port=1) -
> actions=(load:0x1->OXM_OF_METADATA[],load:0x1->NXM_NX_REG14[] and submit to
> table 8)
> match=(in_port=1, dl_vlan=4) actions(load:0x1->NXM_NX_REG10[0],
> strip_vlan, load:0x1->OXM_OF_METADATA[],load:0x2->NXM_NX_REG14[] , and
> submit to table 8)
>
> Suppose if p2 sends a braodcast packet, reg10[0] is set in table 0.
>
> In this case, since reg10[0] is set, the broadcast packet is not dropped
> in table 34 for port "p2" and it gets delivered back.
> So this patch instead of using reg10[0], it uses a new register bit
> -MLF_NESTED_CONTAINER_BIT
>

Thanks for the detailed response. I understand till the above point. So not
using NXM_NX_REG10[0] (or MLF_ALLOW_LOOPBACK_BIT) actually lets the packet
drop in table 34.  Can you please add that information to the commit
message. I think it will be helpful in the future.

I then don't understand, what we are doing with the new code
in OFTABLE_SAVE_INPORT.  I would expect that the new code is needed to save
inport if the destination is a nested container. Not quite understand why
we have a match for parent port. I will wait for the comment in the code to
understand it.





>
> We also need to clear the in_port in table 64 if MLF_NESTED_CONTAINER_BIT
> is set.
>
> I will address the review comments and submit p2.
>
> Hope this clears your confusion.
>
> Thanks
> Numan
>
>
>
>>
>>
>> This is because 'MLF_ALLOW_LOOPBACK_BIT' is set in REG10 in table 0 when
>>> the packet is received from any child port and table
>>> 'OFTABLE_CHECK_LOOPBACK'
>>> doesn't drop the packet.
>>>
>>
>> I had thought that the previous tables only replicate packets to ports
>> other than the inport. But I may be wrong.
>> Some of my comments further down could be stupid because I may have not
>> understood the problem well.
>>
>>
>>>
>>> This patch fixes the issue by using a new register bit
>>> (MLF_NESTED_CONTAINER_BIT)
>>> instead of 'MLF_ALLOW_LOOPBACK_BIT' and sets it in REG10 for the packets
>>> received
>>> from child ports.
>>>
>>> Signed-off-by: Numan Siddique <nusiddiq at redhat.com>
>>> ---
>>>  ovn/controller/ofctrl.c   | 29 ++++++++++++++++++++---------
>>>  ovn/controller/ofctrl.h   |  5 +++++
>>>  ovn/controller/physical.c | 38 ++++++++++++++++++++++++++++++++------
>>>  ovn/lib/logical-fields.h  |  4 ++++
>>>  tests/ovn.at              | 11 +++++++++++
>>>  5 files changed, 72 insertions(+), 15 deletions(-)
>>>
>>> diff --git a/ovn/controller/ofctrl.c b/ovn/controller/ofctrl.c
>>> index 96c57f143..2f2b185ae 100644
>>> --- a/ovn/controller/ofctrl.c
>>> +++ b/ovn/controller/ofctrl.c
>>> @@ -630,9 +630,11 @@ ofctrl_recv(const struct ofp_header *oh, enum
>>> ofptype type)
>>>   *
>>>   * The caller should initialize its own hmap to hold the flows. */
>>>  void
>>> -ofctrl_add_flow(struct hmap *desired_flows,
>>> -                uint8_t table_id, uint16_t priority, uint64_t cookie,
>>> -                const struct match *match, const struct ofpbuf *actions)
>>> +ofctrl_check_and_add_flow(struct hmap *desired_flows,
>>> +                          uint8_t table_id, uint16_t priority, uint64_t
>>> cookie,
>>> +                          const struct match *match,
>>> +                          const struct ofpbuf *actions,
>>> +                          bool log_duplicate_flow)
>>>  {
>>>      struct ovn_flow *f = xmalloc(sizeof *f);
>>>      f->table_id = table_id;
>>> @@ -644,13 +646,14 @@ ofctrl_add_flow(struct hmap *desired_flows,
>>>      f->cookie = cookie;
>>>
>>>      if (ovn_flow_lookup(desired_flows, f)) {
>>> -        static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
>>> -        if (!VLOG_DROP_DBG(&rl)) {
>>> -            char *s = ovn_flow_to_string(f);
>>> -            VLOG_DBG("dropping duplicate flow: %s", s);
>>> -            free(s);
>>> +        if (log_duplicate_flow) {
>>> +            static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5,
>>> 5);
>>> +            if (!VLOG_DROP_DBG(&rl)) {
>>> +                char *s = ovn_flow_to_string(f);
>>> +                VLOG_DBG("dropping duplicate flow: %s", s);
>>> +                free(s);
>>> +            }
>>>          }
>>> -
>>>          ovn_flow_destroy(f);
>>>          return;
>>>      }
>>> @@ -658,6 +661,14 @@ ofctrl_add_flow(struct hmap *desired_flows,
>>>      hmap_insert(desired_flows, &f->hmap_node, f->hmap_node.hash);
>>>  }
>>>
>>> +void
>>> +ofctrl_add_flow(struct hmap *desired_flows,
>>> +                uint8_t table_id, uint16_t priority, uint64_t cookie,
>>> +                const struct match *match, const struct ofpbuf *actions)
>>> +{
>>> +    ofctrl_check_and_add_flow(desired_flows, table_id, priority, cookie,
>>> +                              match, actions, true);
>>> +}
>>>
>>>  /* ovn_flow. */
>>>
>>> diff --git a/ovn/controller/ofctrl.h b/ovn/controller/ofctrl.h
>>> index e3fc95b05..ae0cfa513 100644
>>> --- a/ovn/controller/ofctrl.h
>>> +++ b/ovn/controller/ofctrl.h
>>> @@ -55,4 +55,9 @@ void ofctrl_add_flow(struct hmap *desired_flows,
>>> uint8_t table_id,
>>>                       uint16_t priority, uint64_t cookie,
>>>                       const struct match *, const struct ofpbuf
>>> *ofpacts);
>>>
>>> +void ofctrl_check_and_add_flow(struct hmap *desired_flows, uint8_t
>>> table_id,
>>> +                               uint16_t priority, uint64_t cookie,
>>> +                               const struct match *,
>>> +                               const struct ofpbuf *ofpacts,
>>> +                               bool log_duplicate_flow);
>>>  #endif /* ovn/ofctrl.h */
>>> diff --git a/ovn/controller/physical.c b/ovn/controller/physical.c
>>> index c38d7b09f..d781ae459 100644
>>> --- a/ovn/controller/physical.c
>>> +++ b/ovn/controller/physical.c
>>> @@ -189,7 +189,8 @@ get_zone_ids(const struct sbrec_port_binding
>>> *binding,
>>>
>>>  static void
>>>  put_local_common_flows(uint32_t dp_key, uint32_t port_key,
>>> -                       bool nested_container, const struct zone_ids
>>> *zone_ids,
>>> +                       uint32_t parent_port_key,
>>> +                       const struct zone_ids *zone_ids,
>>>                         struct ofpbuf *ofpacts_p, struct hmap
>>> *flow_table)
>>>  {
>>>      struct match match;
>>> @@ -253,7 +254,7 @@ put_local_common_flows(uint32_t dp_key, uint32_t
>>> port_key,
>>>      ofpbuf_clear(ofpacts_p);
>>>      match_set_metadata(&match, htonll(dp_key));
>>>      match_set_reg(&match, MFF_LOG_OUTPORT - MFF_REG0, port_key);
>>> -    if (!nested_container) {
>>> +    if (!parent_port_key) {
>>>          match_set_reg_masked(&match, MFF_LOG_FLAGS - MFF_REG0,
>>>                               MLF_ALLOW_LOOPBACK, MLF_ALLOW_LOOPBACK);
>>>      }
>>>
>> I think before your patch, the above code would always save inport for
>> nested containers. They continue to do so now. I hope that was your
>> intention?
>>
>>
>>
>>> @@ -264,6 +265,25 @@ put_local_common_flows(uint32_t dp_key, uint32_t
>>> port_key,
>>>      put_stack(MFF_IN_PORT, ofpact_put_STACK_POP(ofpacts_p));
>>>      ofctrl_add_flow(flow_table, OFTABLE_SAVE_INPORT, 100, 0,
>>>
>>
>>                    &match, ofpacts_p);
>>>
>> +
>>>
>> Can you add a comment here on what we are doing below?
>>
>> When a packet is headed to a nested container, the outport in OVN should
>> be that of the destination nested container.
>> But here it looks like we are matching on the OVN outport being the
>> parent port and saying that if the outport is parent port, we should save
>> the physical inport.
>> Is that what we are doing? I don't quite understand how this helps to
>> prevent packet from going back to the originator child port. I am sure I am
>> missing something simple here.
>>
>> +    if (parent_port_key) {
>>> +        match_init_catchall(&match);
>>> +        ofpbuf_clear(ofpacts_p);
>>> +        match_set_metadata(&match, htonll(dp_key));
>>> +        match_set_reg(&match, MFF_LOG_OUTPORT - MFF_REG0,
>>> parent_port_key);
>>> +        match_set_reg_masked(&match, MFF_LOG_FLAGS - MFF_REG0,
>>> +                             MLF_NESTED_CONTAINER,
>>> MLF_NESTED_CONTAINER);
>>> +
>>> +        put_stack(MFF_IN_PORT, ofpact_put_STACK_PUSH(ofpacts_p));
>>> +        put_load(0, MFF_IN_PORT, 0, 16, ofpacts_p);
>>> +        put_resubmit(OFTABLE_LOG_TO_PHY, ofpacts_p);
>>> +        put_stack(MFF_IN_PORT, ofpact_put_STACK_POP(ofpacts_p));
>>> +        /* When a parent port has multiple child ports, we don't want to
>>> +         * log the duplicate flow message.
>>> +         */
>>> +        ofctrl_check_and_add_flow(flow_table, OFTABLE_SAVE_INPORT, 100,
>>> 0,
>>> +                                  &match, ofpacts_p, false);
>>> +    }
>>>  }
>>>
>>>  static void
>>> @@ -328,7 +348,7 @@ consider_port_binding(struct ovsdb_idl_index
>>> *sbrec_chassis_by_name,
>>>          }
>>>
>>>          struct zone_ids binding_zones = get_zone_ids(binding, ct_zones);
>>> -        put_local_common_flows(dp_key, port_key, false, &binding_zones,
>>> +        put_local_common_flows(dp_key, port_key, 0, &binding_zones,
>>>                                 ofpacts_p, flow_table);
>>>
>>>          match_init_catchall(&match);
>>> @@ -452,6 +472,7 @@ consider_port_binding(struct ovsdb_idl_index
>>> *sbrec_chassis_by_name,
>>>
>>>      int tag = 0;
>>>      bool nested_container = false;
>>> +    const struct sbrec_port_binding *parent_port = NULL;
>>>      ofp_port_t ofport;
>>>      bool is_remote = false;
>>>      if (binding->parent_port && *binding->parent_port) {
>>> @@ -463,6 +484,8 @@ consider_port_binding(struct ovsdb_idl_index
>>> *sbrec_chassis_by_name,
>>>          if (ofport) {
>>>              tag = *binding->tag;
>>>              nested_container = true;
>>> +            parent_port = lport_lookup_by_name(
>>> +                sbrec_port_binding_by_name, binding->parent_port);
>>>          }
>>>      } else {
>>>          ofport = u16_to_ofp(simap_get(&localvif_to_ofport,
>>> @@ -523,7 +546,8 @@ consider_port_binding(struct ovsdb_idl_index
>>> *sbrec_chassis_by_name,
>>>           */
>>>
>>>          struct zone_ids zone_ids = get_zone_ids(binding, ct_zones);
>>> -        put_local_common_flows(dp_key, port_key, nested_container,
>>> &zone_ids,
>>> +        uint32_t parent_port_key = parent_port ?
>>> parent_port->tunnel_key : 0;
>>> +        put_local_common_flows(dp_key, port_key, parent_port_key,
>>> &zone_ids,
>>>                                 ofpacts_p, flow_table);
>>>
>>>          /* Table 0, Priority 150 and 100.
>>> @@ -553,8 +577,10 @@ consider_port_binding(struct ovsdb_idl_index
>>> *sbrec_chassis_by_name,
>>>              if (nested_container) {
>>>                  /* When a packet comes from a container sitting behind a
>>>                   * parent_port, we should let it loopback to other
>>> containers
>>> -                 * or the parent_port itself. */
>>> -                put_load(MLF_ALLOW_LOOPBACK, MFF_LOG_FLAGS, 0, 1,
>>> ofpacts_p);
>>> +                 * or the parent_port itself. Indicate this by setting
>>> the
>>> +                 * MLF_NESTED_CONTAINER_BIT in MFF_LOG_FLAGS.*/
>>> +                put_load(1, MFF_LOG_FLAGS, MLF_NESTED_CONTAINER_BIT, 1,
>>> +                         ofpacts_p);
>>>              }
>>>              ofpact_put_STRIP_VLAN(ofpacts_p);
>>>          }
>>> diff --git a/ovn/lib/logical-fields.h b/ovn/lib/logical-fields.h
>>> index b1dbb035a..ab31327e9 100644
>>> --- a/ovn/lib/logical-fields.h
>>> +++ b/ovn/lib/logical-fields.h
>>> @@ -50,6 +50,7 @@ enum mff_log_flags_bits {
>>>      MLF_FORCE_SNAT_FOR_DNAT_BIT = 2,
>>>      MLF_FORCE_SNAT_FOR_LB_BIT = 3,
>>>      MLF_LOCAL_ONLY_BIT = 4,
>>> +    MLF_NESTED_CONTAINER_BIT = 5,
>>>  };
>>>
>>>  /* MFF_LOG_FLAGS_REG flag assignments */
>>> @@ -75,6 +76,9 @@ enum mff_log_flags {
>>>       * hypervisors should instead only be output to local targets
>>>       */
>>>      MLF_LOCAL_ONLY = (1 << MLF_LOCAL_ONLY_BIT),
>>> +
>>> +    /* Indicate that a packet has received from a nested container. */
>>> +    MLF_NESTED_CONTAINER = (1 << MLF_NESTED_CONTAINER_BIT),
>>>  };
>>>
>>>  #endif /* ovn/lib/logical-fields.h */
>>> diff --git a/tests/ovn.at b/tests/ovn.at
>>> index 769e09f81..a7aba5ccd 100644
>>> --- a/tests/ovn.at
>>> +++ b/tests/ovn.at
>>> @@ -7038,6 +7038,17 @@
>>> packet=${dst_mac}${src_mac}8100000208004500001c000000003f110100${src_ip}${dst_ip
>>>  echo  $packet >> expected1
>>>  OVN_CHECK_PACKETS([hv1/vm1-tx.pcap], [expected1])
>>>
>>> +# Send broadcast packet from foo1. foo1 should not receive the same
>>> packet.
>>> +src_mac="f00000010205"
>>> +dst_mac="ffffffffffff"
>>> +src_ip=`ip_to_hex 192 168 1 2`
>>> +dst_ip=`ip_to_hex 255 255 255 255`
>>>
>>> +packet=${dst_mac}${src_mac}8100000108004500001c0000000040110000${src_ip}${dst_ip}0035111100080000
>>> +as hv1 ovs-appctl netdev-dummy/receive vm1 $packet
>>> +
>>> +# expected packet at VM1
>>> +OVN_CHECK_PACKETS([hv1/vm1-tx.pcap], [expected1])
>>> +
>>>  OVN_CLEANUP([hv1],[hv2])
>>>
>>>  AT_CLEANUP
>>> --
>>> 2.17.1
>>>
>>> _______________________________________________
>>> dev mailing list
>>> dev at openvswitch.org
>>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>>>
>>


More information about the dev mailing list