[ovs-dev] [PATCH ovn v2 2/3] actions: Add new OVN logical fields 'ip.src' and ip.dst'.

Dumitru Ceara dceara at redhat.com
Wed Oct 14 12:03:38 UTC 2020


On 10/14/20 1:54 PM, Numan Siddique wrote:
> On Wed, Oct 14, 2020 at 4:40 PM Dumitru Ceara <dceara at redhat.com> wrote:
>>
>> On 10/14/20 12:50 PM, Numan Siddique wrote:
>>> On Wed, Oct 14, 2020 at 3:50 PM Dumitru Ceara <dceara at redhat.com> wrote:
>>>>
>>>> On 10/14/20 11:15 AM, numans at ovn.org wrote:
>>>>> From: Numan Siddique <numans at ovn.org>
>>>>>
>>>>> Thee new fields are version independent and these can be used in any
>>>>
>>>> Hi Numan,
>>>>
>>>> Nit: s/Thee/These
>>>>
>>>>> OVN action. Right now the usage of these fields are restricted to
>>>>> exchanging the IP source and destination fields.
>>>>>
>>>>> Eg. reject {... ip.src <-> ip.dst ... };
>>>>>
>>>>> "ip.src <-> ip.dst" translates to controller action with "pause" flag set.
>>>>> When pinctrl thread receives the packet in, it checks the IP version of the
>>>>> packet and either does - "ip4.src <-> ip4.dst" or "ip6.src <-> ip6.dst" and
>>>>> resumes the packet to continue with the pipeline.
>>>>>
>>>>> Upcoming patch will make use of these fields.
>>>>>
>>>>> Signed-off-by: Numan Siddique <numans at ovn.org>
>>>>> ---
>>>>>  controller/pinctrl.c         |  49 +++++++++++++++
>>>>>  include/ovn/actions.h        |   4 ++
>>>>>  include/ovn/logical-fields.h |  18 ++++++
>>>>>  lib/actions.c                | 118 +++++++++++++++++++++++++++++++++++
>>>>>  lib/logical-fields.c         |  10 +++
>>>>>  ovn-sb.xml                   |  37 +++++++++++
>>>>>  tests/ovn.at                 |  27 ++++++++
>>>>>  utilities/ovn-trace.c        |  28 +++++++++
>>>>>  8 files changed, 291 insertions(+)
>>>>>
>>>>> diff --git a/controller/pinctrl.c b/controller/pinctrl.c
>>>>> index 631d058458..bc16a82404 100644
>>>>> --- a/controller/pinctrl.c
>>>>> +++ b/controller/pinctrl.c
>>>>> @@ -233,6 +233,11 @@ static void pinctrl_handle_put_icmp_frag_mtu(struct rconn *swconn,
>>>>>                                               struct ofputil_packet_in *pin,
>>>>>                                               struct ofpbuf *userdata,
>>>>>                                               struct ofpbuf *continuation);
>>>>> +static void pinctrl_handle_swap_src_dst_ip(struct rconn *swconn,
>>>>> +                                           const struct flow *in_flow,
>>>>> +                                           struct dp_packet *pkt_in,
>>>>> +                                           struct ofputil_packet_in *pin,
>>>>> +                                           struct ofpbuf *continuation);
>>>>>  static void
>>>>>  pinctrl_handle_event(struct ofpbuf *userdata)
>>>>>      OVS_REQUIRES(pinctrl_mutex);
>>>>> @@ -2835,6 +2840,11 @@ process_packet_in(struct rconn *swconn, const struct ofp_header *msg)
>>>>>          ovs_mutex_unlock(&pinctrl_mutex);
>>>>>          break;
>>>>>
>>>>> +    case ACTION_OPCODE_SWAP_SRC_DST_IP:
>>>>> +        pinctrl_handle_swap_src_dst_ip(swconn, &headers, &packet, &pin,
>>>>> +                                       &continuation);
>>>>> +        break;
>>>>> +
>>>>>      default:
>>>>>          VLOG_WARN_RL(&rl, "unrecognized packet-in opcode %"PRIu32,
>>>>>                       ntohl(ah->opcode));
>>>>> @@ -6508,3 +6518,42 @@ pinctrl_handle_svc_check(struct rconn *swconn, const struct flow *ip_flow,
>>>>>          svc_mon->next_send_time = time_msec() + svc_mon->interval;
>>>>>      }
>>>>>  }
>>>>> +
>>>>> +static void
>>>>> +pinctrl_handle_swap_src_dst_ip(struct rconn *swconn,
>>>>> +                               const struct flow *in_flow,
>>>>> +                               struct dp_packet *pkt_in,
>>>>> +                               struct ofputil_packet_in *pin,
>>>>> +                               struct ofpbuf *continuation)
>>>>> +{
>>>>> +    enum ofp_version version = rconn_get_version(swconn);
>>>>> +    enum ofputil_protocol proto = ofputil_protocol_from_ofp_version(version);
>>>>> +    struct dp_packet *pkt_out;
>>>>> +
>>>>> +    pkt_out = dp_packet_clone(pkt_in);
>>>>> +    pkt_out->l2_5_ofs = pkt_in->l2_5_ofs;
>>>>> +    pkt_out->l2_pad_size = pkt_in->l2_pad_size;
>>>>> +    pkt_out->l3_ofs = pkt_in->l3_ofs;
>>>>> +    pkt_out->l4_ofs = pkt_in->l4_ofs;
>>>>> +
>>>>> +    if (get_dl_type(in_flow) == htons(ETH_TYPE_IP)) {
>>>>> +        /* IPv4 packet. Swap nw_src with nw_dst. */
>>>>> +        packet_set_ipv4(pkt_out, in_flow->nw_dst, in_flow->nw_src,
>>>>> +                        in_flow->nw_tos, in_flow->nw_ttl);
>>>>> +    } else {
>>>>> +        /* IPv6 packet. Swap ip6_src with ip6_dst.
>>>>> +         * We could also use packet_set_ipv6() here, but that would require
>>>>> +         * to extract the 'tc' and 'label' from in_nh->ip6_flow which seems
>>>>> +         * unnecessary. */
>>>>> +        struct ovs_16aligned_ip6_hdr *out_nh = dp_packet_l3(pkt_out);
>>>>> +        union ovs_16aligned_in6_addr tmp = out_nh->ip6_src;
>>>>> +        out_nh->ip6_src = out_nh->ip6_dst;
>>>>> +        out_nh->ip6_dst = tmp;
>>>>> +    }
>>>>> +
>>>>> +    pin->packet = dp_packet_data(pkt_out);
>>>>> +    pin->packet_len = dp_packet_size(pkt_out);
>>>>> +
>>>>> +    queue_msg(swconn, ofputil_encode_resume(pin, continuation, proto));
>>>>> +    dp_packet_delete(pkt_out);
>>>>> +}
>>>>> diff --git a/include/ovn/actions.h b/include/ovn/actions.h
>>>>> index b4e5acabb9..bf1fe848b7 100644
>>>>> --- a/include/ovn/actions.h
>>>>> +++ b/include/ovn/actions.h
>>>>> @@ -97,6 +97,7 @@ struct ovn_extend_table;
>>>>>      OVNACT(DHCP6_REPLY,       ovnact_null)            \
>>>>>      OVNACT(ICMP6_ERROR,       ovnact_nest)            \
>>>>>      OVNACT(REJECT,            ovnact_nest)            \
>>>>> +    OVNACT(OVNFIELD_EXCHANGE, ovnact_move)            \
>>>>>
>>>>>  /* enum ovnact_type, with a member OVNACT_<ENUM> for each action. */
>>>>>  enum OVS_PACKED_ENUM ovnact_type {
>>>>> @@ -612,6 +613,9 @@ enum action_opcode {
>>>>>       * The actions, in OpenFlow 1.3 format, follow the action_header.
>>>>>       */
>>>>>      ACTION_OPCODE_REJECT,
>>>>> +
>>>>> +    /* ip.src <-> ip.dst */
>>>>> +    ACTION_OPCODE_SWAP_SRC_DST_IP,
>>>>>  };
>>>>>
>>>>>  /* Header. */
>>>>> diff --git a/include/ovn/logical-fields.h b/include/ovn/logical-fields.h
>>>>> index ac6f2f909b..bb6daa50ca 100644
>>>>> --- a/include/ovn/logical-fields.h
>>>>> +++ b/include/ovn/logical-fields.h
>>>>> @@ -116,6 +116,24 @@ enum ovn_field_id {
>>>>>       */
>>>>>      OVN_ICMP6_FRAG_MTU,
>>>>>
>>>>> +    /*
>>>>> +     * Name: "ip.src"
>>>>> +     * Type: be128
>>>>> +     * Description: Sets the field MFF_IPV4_SRC if eth.type == 0x800 (IPv4)
>>>>> +     *              or sets the field MFF_IPV6_SRC if
>>>>> +     *              eth.type == 0x86dd (IPV6).
>>>>> +     */
>>>>> +    OVN_IP_SRC,
>>>>> +
>>>>> +    /*
>>>>> +     * Name: "ip.dst"
>>>>> +     * Type: be128
>>>>> +     * Description: Sets the field MFF_IPV4_DST if eth.type == 0x800 (IPv4)
>>>>> +     *              or sets the field MFF_IPV6_DST if
>>>>> +     *              eth.type == 0x86dd (IPV6).
>>>>> +     */
>>>>> +    OVN_IP_DST,
>>>>> +
>>>>>      OVN_FIELD_N_IDS
>>>>>  };
>>>>>
>>>>> diff --git a/lib/actions.c b/lib/actions.c
>>>>> index 1e1bdeff24..e9c77d2a0a 100644
>>>>> --- a/lib/actions.c
>>>>> +++ b/lib/actions.c
>>>>> @@ -371,12 +371,35 @@ ovnact_next_free(struct ovnact_next *a OVS_UNUSED)
>>>>>  {
>>>>>  }
>>>>>
>>>>> +static bool
>>>>> +check_ovnfield_load(struct action_context *ctx, const struct expr_field *field)
>>>>> +{
>>>>> +    switch (field->symbol->ovn_field->id) {
>>>>> +    case OVN_ICMP4_FRAG_MTU:
>>>>> +    case OVN_ICMP6_FRAG_MTU:
>>>>> +        return true;
>>>>> +
>>>>> +    case OVN_IP_SRC:
>>>>> +    case OVN_IP_DST:
>>>>> +        lexer_error(ctx->lexer, "Can't load a value to ovn field (%s).",
>>>>> +                    field->symbol->name);
>>>>> +        return false;
>>>>> +
>>>>> +    case OVN_FIELD_N_IDS:
>>>>> +    default:
>>>>> +        OVS_NOT_REACHED();
>>>>> +    }
>>>>> +}
>>>>> +
>>>>>  static void
>>>>>  parse_LOAD(struct action_context *ctx, const struct expr_field *lhs)
>>>>>  {
>>>>>      size_t ofs = ctx->ovnacts->size;
>>>>>      struct ovnact_load *load;
>>>>>      if (lhs->symbol->ovn_field) {
>>>>> +        if (!check_ovnfield_load(ctx, lhs)) {
>>>>> +            return;
>>>>> +        }
>>>>>          load = ovnact_put_OVNFIELD_LOAD(ctx->ovnacts);
>>>>>      } else {
>>>>>          load = ovnact_put_LOAD(ctx->ovnacts);
>>>>> @@ -474,6 +497,46 @@ format_EXCHANGE(const struct ovnact_move *move, struct ds *s)
>>>>>      format_assignment(move, "<->", s);
>>>>>  }
>>>>>
>>>>> +static void
>>>>> +parse_ovnfield_exchange(struct action_context *ctx,
>>>>> +                        const struct expr_field *lhs,
>>>>> +                        const struct expr_field *rhs)
>>>>> +{
>>>>> +    if (!lhs->symbol->ovn_field || !rhs->symbol->ovn_field) {
>>>>> +        lexer_error(ctx->lexer,
>>>>> +                    "Can't exchange ovn field with non ovn field (%s <-> %s).",
>>>>> +                    lhs->symbol->name, rhs->symbol->name);
>>>>> +        return;
>>>>> +    }
>>>>> +
>>>>> +    if (lhs->symbol->ovn_field->id != OVN_IP_SRC &&
>>>>> +            lhs->symbol->ovn_field->id != OVN_IP_DST) {
>>>>> +        lexer_error(ctx->lexer,
>>>>> +                    "Can't exchange ovn field (%s) with ovn field (%s).",
>>>>> +                    lhs->symbol->name, rhs->symbol->name);
>>>>> +        return;
>>>>> +    }
>>>>> +
>>>>> +    if (rhs->symbol->ovn_field->id != OVN_IP_SRC &&
>>>>> +            rhs->symbol->ovn_field->id != OVN_IP_DST) {
>>>>> +        lexer_error(ctx->lexer,
>>>>> +                    "Can't exchange ovn field (%s) with ovn field (%s).",
>>>>> +                    lhs->symbol->name, rhs->symbol->name);
>>>>> +        return;
>>>>> +    }
>>>>> +
>>>>> +    if (lhs->symbol->ovn_field->id == rhs->symbol->ovn_field->id) {
>>>>> +        lexer_error(ctx->lexer,
>>>>> +                    "Can't exchange ovn field (%s) with ovn field (%s).",
>>>>> +                    lhs->symbol->name, rhs->symbol->name);
>>>>> +        return;
>>>>> +    }
>>>>> +
>>>>> +    struct ovnact_move *move = ovnact_put_OVNFIELD_EXCHANGE(ctx->ovnacts);
>>>>> +    move->lhs = *lhs;
>>>>> +    move->rhs = *rhs;
>>>>> +}
>>>>> +
>>>>>  static void
>>>>>  parse_assignment_action(struct action_context *ctx, bool exchange,
>>>>>                          const struct expr_field *lhs)
>>>>> @@ -483,6 +546,11 @@ parse_assignment_action(struct action_context *ctx, bool exchange,
>>>>>          return;
>>>>>      }
>>>>>
>>>>> +    if (exchange && (lhs->symbol->ovn_field || rhs.symbol->ovn_field)) {
>>>>> +        parse_ovnfield_exchange(ctx, lhs, &rhs);
>>>>> +        return;
>>>>> +    }
>>>>> +
>>>>>      const struct expr_symbol *ls = lhs->symbol;
>>>>>      const struct expr_symbol *rs = rhs.symbol;
>>>>>      if ((ls->width != 0) != (rs->width != 0)) {
>>>>> @@ -3128,6 +3196,8 @@ format_OVNFIELD_LOAD(const struct ovnact_load *load , struct ds *s)
>>>>>                        ntohs(load->imm.value.be16_int));
>>>>>          break;
>>>>>
>>>>> +    case OVN_IP_SRC:
>>>>> +    case OVN_IP_DST:
>>>>>      case OVN_FIELD_N_IDS:
>>>>>      default:
>>>>>          OVS_NOT_REACHED();
>>>>> @@ -3157,6 +3227,9 @@ encode_OVNFIELD_LOAD(const struct ovnact_load *load,
>>>>>          encode_finish_controller_op(oc_offset, ofpacts);
>>>>>          break;
>>>>>      }
>>>>> +
>>>>> +    case OVN_IP_SRC:
>>>>> +    case OVN_IP_DST:
>>>>>      case OVN_FIELD_N_IDS:
>>>>>      default:
>>>>>          OVS_NOT_REACHED();
>>>>> @@ -3451,6 +3524,51 @@ ovnact_fwd_group_free(struct ovnact_fwd_group *fwd_group)
>>>>>      free(fwd_group->child_ports);
>>>>>  }
>>>>>
>>>>> +static void
>>>>> +format_OVNFIELD_EXCHANGE(const struct ovnact_move *move , struct ds *s)
>>>>> +{
>>>>> +    const struct ovn_field *lhs = ovn_field_from_name(move->lhs.symbol->name);
>>>>> +    const struct ovn_field *rhs = ovn_field_from_name(move->rhs.symbol->name);
>>>>> +    switch (lhs->id) {
>>>>> +    case OVN_IP_SRC:
>>>>> +    case OVN_IP_DST:
>>>>> +        break;
>>>>> +
>>>>> +    case OVN_ICMP4_FRAG_MTU:
>>>>> +    case OVN_ICMP6_FRAG_MTU:
>>>>> +    case OVN_FIELD_N_IDS:
>>>>> +    default:
>>>>> +        OVS_NOT_REACHED();
>>>>> +    }
>>>>
>>>> Nit: I would use the shorter:
>>>> ovs_asssert(lhs->id == OVN_IP_SRC || lhs->id == OVN_IP_DST);
>>>
>>> You mean to drop the switch right ? If so, Ack.
>>>
>>
>> Right, I meant ovs_assert instead of switch.
>>
>>>
>>>>
>>>>> +
>>>>> +    switch (rhs->id) {
>>>>> +    case OVN_IP_SRC:
>>>>> +    case OVN_IP_DST:
>>>>> +        break;
>>>>> +
>>>>> +    case OVN_ICMP4_FRAG_MTU:
>>>>> +    case OVN_ICMP6_FRAG_MTU:
>>>>> +    case OVN_FIELD_N_IDS:
>>>>> +    default:
>>>>> +        OVS_NOT_REACHED();
>>>>> +    }
>>>>
>>>> Nit: I would use the shorter:
>>>> ovs_asssert(rhs->id == OVN_IP_SRC || rhs->id == OVN_IP_DST);
>>>>
>>>>> +
>>>>> +    ds_put_format(s, "%s <-> %s;", lhs->name, rhs->name);
>>>>> +}
>>>>> +
>>>>> +static void
>>>>> +encode_OVNFIELD_EXCHANGE(const struct ovnact_move *move OVS_UNUSED,
>>>>> +                         const struct ovnact_encode_params *ep OVS_UNUSED,
>>>>> +                         struct ofpbuf *ofpacts OVS_UNUSED)
>>>>> +{
>>>>> +    /* Right now we only support exchanging the IPs in
>>>>> +     * OVN fields (ip.src <-> ip.dst). */
>>>>> +    size_t oc_offset =
>>>>> +        encode_start_controller_op(ACTION_OPCODE_SWAP_SRC_DST_IP,
>>>>> +                                   true, NX_CTLR_NO_METER, ofpacts);
>>>>> +    encode_finish_controller_op(oc_offset, ofpacts);
>>>>
>>>> I think this means that all packets matching a flow with action "reject { ...
>>>> ip.src <-> ip.dst ... }" will generate two packet-ins, right? One for the
>>>> reject action, one for the ovnfield-exchange action.
>>>>
>>>> Is that a concern wrt. performance?
>>>
>>> Yes. It would have 2 packet-ins. Just like how we handle - icmp4 {...
>>> icmp4.frag_mtu = 1500; ...};
>>>
>>> If performance is a concern, we could just drop ... ip.src <-> ip.dst
>>> in reject and let reject action
>>> swap the ip src with ip destination. I thought about that and my take
>>> is that since reject { } results in
>>> TCP RST or ICMP unreachable packet, it should be fine. I see these
>>> packets as more of control packets.
>>>
>>> I think it should be ok. Let me know what you think.
>>>
>>
>> I think I like the approach of "reject" action implicitly swapping IPs best.
>> In the end it's not like we could implement reject withough swapping IP dst
>> with IP src so why not do it as part of the reject action.
>>
> 
> Agree. reject action is expected to be used to send the generated tcp
> rst/icmp unreachable
> packet back to the sender.
> 
> I'm ok to change the reject action to just swap the IPs internally if
> everyone is fine.
> 
> @Mark Michelson  @Han Zhou  Do you have any comments here ?
> 
> Option 1 - reject { ...}    -> The reject action handling in pinctrl.c
> will swap the ip source and destination
> 
> Option 2: reject {... ip.src <-> ip.dst ...} which is the proposed
> approach in this patch series. This results in
> 2 packet-ins.
> 
> I personally prefer (2) since in OVN we normally tell what inner
> actions to apply for many OVN actions.
> But I have no strong preference and I'm fine changing to (1).
> 
> If we chose option (1), we could add a comment inside the action like
> - reject { /* ip.src <-> ip.dst is done implicitly*/,  eth.src <->
> eth.dst; output ; }
> 

Then, why not swap ETH addresses implicitly too?

> 
>> But on the other hand, I think there might be other simpler ways to generate a
>> lot of packet-ins towards ovn-controller that I don't have a strong preference
>> about either solution.  I just wanted to make sure we discuss the performance
>> implications.
> 
> In my opinion there should not be any performance implications.
> 
> 
>>
>>>
>>>>
>>>>> +}
>>>>> +
>>>>>  /* Parses an assignment or exchange or put_dhcp_opts action. */
>>>>>  static void
>>>>>  parse_set_action(struct action_context *ctx)
>>>>> diff --git a/lib/logical-fields.c b/lib/logical-fields.c
>>>>> index bf61df7719..d6f1981f52 100644
>>>>> --- a/lib/logical-fields.c
>>>>> +++ b/lib/logical-fields.c
>>>>> @@ -33,6 +33,14 @@ const struct ovn_field ovn_fields[OVN_FIELD_N_IDS] = {
>>>>>          OVN_ICMP6_FRAG_MTU,
>>>>>          "icmp6.frag_mtu",
>>>>>          4, 32,
>>>>> +    }, {
>>>>> +        OVN_IP_SRC,
>>>>> +        "ip.src",
>>>>> +        16, 128,
>>>>> +    }, {
>>>>> +        OVN_IP_DST,
>>>>> +        "ip.dst",
>>>>> +        16, 128,
>>>>
>>>> Just partially related to your change, I'm a bit confused by the meaning of
>>>> n_bytes and n_bits in "struct ovn_field".  It seems to me that we never use
>>>> those values.  Is that the case or am I missing something?
>>>
>>> These OVN fields are defined more like the way - mf_fields are defined.
>>> Like icmp4.frag_mtu is 2 bytes and 16 bits field.
>>>
>>> So ip.src/ip.dst can be used for both ipv4 and ipv6, I have set the
>>> size of ipv6.
>>>
>>> Since we don't support ip.src = 10.0.0.4 or ip.src = aef00::4 for
>>> example, these fields
>>> are not used. Later on if there is a need to support assigning a value
>>> to these fields, these fields
>>> could be used. Although I don't see a need to support it since we
>>> already have ip4.src/ip
> 
>>>
>>
>> Ok, my question was more about the fact that I don't see the ovn_field.n_bytes
>> or ovn_fields.n_bits being used anywhere in general (not referring to
>> ip.src/ip.dst only).
>>
> 
> Actually they are being used. See
> https://github.com/ovn-org/ovn/blob/master/lib/actions.c#L394
> and https://github.com/ovn-org/ovn/blob/master/lib/expr.c#L569
> 
> If you add the below in the ovn -- action parsing test case in ovn.at
> ---
> icmp4.frag_mtu = 65536;
> ---
> 
> You'll see the below error:
> 17-bit constant is not compatible with 16-bit field icmp4.frag_mtu.
> 

Ah, you're right, sorry about the noise, I see it now.

Thanks,
Dumitru



More information about the dev mailing list