[ovs-dev] [PATCH V3 03/12] netdev-offload-dpdk: Support offload of set IPv6 actions

[Eli Britstein]

On 6/29/2020 2:02 AM, Ilya Maximets wrote:
> On 6/21/20 1:19 PM, Eli Britstein wrote:
>
> Some trivial commit message here?
OK.
>
>> Signed-off-by: Eli Britstein <elibr at mellanox.com>
>> Reviewed-by: Roni Bar Yanai <roniba at mellanox.com>
>> ---
>>   Documentation/howto/dpdk.rst |  1 +
>>   NEWS                         |  2 ++
>>   lib/netdev-offload-dpdk.c    | 35 +++++++++++++++++++++++++++++++++++
>>   3 files changed, 38 insertions(+)
>>
>> diff --git a/Documentation/howto/dpdk.rst b/Documentation/howto/dpdk.rst
>> index 8a0eee80c..1756a7149 100644
>> --- a/Documentation/howto/dpdk.rst
>> +++ b/Documentation/howto/dpdk.rst
>> @@ -395,6 +395,7 @@ Supported actions for hardware offload are:
>>   - Modification of Ethernet (mod_dl_src/mod_dl_dst).
>>   - Modification of IPv4 (mod_nw_src/mod_nw_dst/mod_nw_ttl).
>>   - Modification of TCP/UDP (mod_tp_src/mod_tp_dst).
>> +- Modification of IPv6 (set_field:<ADDR>->ipv6_src/ipv6_dst/mod_nw_ttl).
>>   
>>   Further Reading
>>   ---------------
>> diff --git a/NEWS b/NEWS
>> index 07e23316c..7df8b134d 100644
>> --- a/NEWS
>> +++ b/NEWS
>> @@ -10,6 +10,8 @@ Post-v2.13.0
>>        * Deprecated DPDK pdump packet capture support removed.
>>        * Deprecated DPDK ring ports (dpdkr) are no longer supported.
>>        * Add hardware offload support for matching IPv6 protocol.
>> +     * Add hardware offload support for set of IPv6 TCP/UDP ports
>> +       actions (experimental).
>>      - Linux datapath:
>>        * Support for kernel versions up to 5.5.x.
>>      - AF_XDP:
>> diff --git a/lib/netdev-offload-dpdk.c b/lib/netdev-offload-dpdk.c
>> index 6b12e9ae3..dd5e71e36 100644
>> --- a/lib/netdev-offload-dpdk.c
>> +++ b/lib/netdev-offload-dpdk.c
>> @@ -458,6 +458,23 @@ dump_flow_action(struct ds *s, const struct rte_flow_action *actions)
>>           } else {
>>               ds_put_format(s, "  Set-%s-tcp/udp-port = null\n", dirstr);
>>           }
>> +    } else if (actions->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ||
>> +               actions->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_DST) {
>> +        const struct rte_flow_action_set_ipv6 *set_ipv6 = actions->conf;
>> +
>> +        char *dirstr = actions->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_DST
>> +                       ? "dst" : "src";
>> +
>> +        ds_put_format(s, "rte flow set-ipv6-%s action:\n", dirstr);
>> +        if (set_ipv6) {
>> +            char addr_str[INET6_ADDRSTRLEN];
>> +
>> +            ipv6_string_mapped(addr_str,
>> +                               (struct in6_addr *)&set_ipv6->ipv6_addr);
>> +            ds_put_format(s, "  Set-ipv6-%s: %s\n", dirstr, addr_str);
> It seems like using ipv6_format_mapped() will result in lower number of
> lines and will also avoid introducing an array:
>
>              ds_put_format(s, "  Set-ipv6-%s: ", dirstr);
>              ipv6_format_mapped((struct in6_addr *) &set_ipv6->ipv6_addr, s);
>              ds_put_cstr(s, "\n");
OK. I'll use ipv6_format_addr.
>
> Same, actually might be applicable to the matching code from the previous patch.
> It actually takes less number of lines even with extra ones like:
>              ds_put_cstr(",  src=");
Following the suggestion about testpmd format, I'll change the order of 
commits to do the testpmd format first. For testpmd format, 
ipv6_format_mapped (or ipv6_format_addr) is not applicable, so I'll keep 
it there.
>
>
>> +        } else {
>> +            ds_put_format(s, "  Set-ipv6-%s = null\n", dirstr);
>> +        }
>>       } else {
>>           ds_put_format(s, "unknown rte flow action (%d)\n", actions->type);
>>       }
>> @@ -1004,6 +1021,12 @@ BUILD_ASSERT_DECL(sizeof(struct rte_flow_action_set_ipv4) ==
>>                     MEMBER_SIZEOF(struct ovs_key_ipv4, ipv4_dst));
>>   BUILD_ASSERT_DECL(sizeof(struct rte_flow_action_set_ttl) ==
>>                     MEMBER_SIZEOF(struct ovs_key_ipv4, ipv4_ttl));
>> +BUILD_ASSERT_DECL(sizeof(struct rte_flow_action_set_ipv6) ==
>> +                  MEMBER_SIZEOF(struct ovs_key_ipv6, ipv6_src));
>> +BUILD_ASSERT_DECL(sizeof(struct rte_flow_action_set_ipv6) ==
>> +                  MEMBER_SIZEOF(struct ovs_key_ipv6, ipv6_dst));
>> +BUILD_ASSERT_DECL(sizeof(struct rte_flow_action_set_ttl) ==
>> +                  MEMBER_SIZEOF(struct ovs_key_ipv6, ipv6_hlimit));
>>   BUILD_ASSERT_DECL(sizeof(struct rte_flow_action_set_tp) ==
>>                     MEMBER_SIZEOF(struct ovs_key_tcp, tcp_src));
>>   BUILD_ASSERT_DECL(sizeof(struct rte_flow_action_set_tp) ==
>> @@ -1053,6 +1076,18 @@ parse_set_actions(struct flow_actions *actions,
>>                   VLOG_DBG_RL(&rl, "Unsupported IPv4 set action");
>>                   return -1;
>>               }
>> +        } else if (nl_attr_type(sa) == OVS_KEY_ATTR_IPV6) {
>> +            const struct ovs_key_ipv6 *key = nl_attr_get(sa);
>> +            const struct ovs_key_ipv6 *mask = masked ? key + 1 : NULL;
>> +
>> +            add_set_flow_action(ipv6_src, RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC);
>> +            add_set_flow_action(ipv6_dst, RTE_FLOW_ACTION_TYPE_SET_IPV6_DST);
>> +            add_set_flow_action(ipv6_hlimit, RTE_FLOW_ACTION_TYPE_SET_TTL);
>> +
>> +            if (mask && !is_all_zeros(mask, sizeof *mask)) {
>> +                VLOG_DBG_RL(&rl, "Unsupported IPv6 set action");
>> +                return -1;
>> +            }
>>           } else if (nl_attr_type(sa) == OVS_KEY_ATTR_TCP) {
>>               const struct ovs_key_tcp *key = nl_attr_get(sa);
>>               const struct ovs_key_tcp *mask = masked ? key + 1 : NULL;
>>