net/mlx5e: CT: Don't offload tuple rewrites for established tuples
authorPaul Blakey <paulb@mellanox.com>
Wed, 22 Apr 2020 15:00:25 +0000 (18:00 +0300)
committerSaeed Mahameed <saeedm@mellanox.com>
Fri, 10 Jul 2020 02:51:14 +0000 (19:51 -0700)
Next patches will remove the tupleid registers that is used
to restore the ct state on miss, and instead use the tuple on
the missed packet to lookup which state to restore.
Disable tuple rewrites after connection tracking.

For tuple rewrites, inject a ct_state=-trk match so it won't
change the tuple for established flows (+trk) that passed connection
tracking, and instead miss to software.

Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.h
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
drivers/net/ethernet/mellanox/mlx5/core/en_tc.h

index 55402b1..08ebce3 100644 (file)
@@ -894,6 +894,24 @@ mlx5_tc_ct_block_flow_offload(enum tc_setup_type type, void *type_data,
        return -EOPNOTSUPP;
 }
 
+int
+mlx5_tc_ct_add_no_trk_match(struct mlx5e_priv *priv,
+                           struct mlx5_flow_spec *spec)
+{
+       u32 ctstate = 0, ctstate_mask = 0;
+
+       mlx5e_tc_match_to_reg_get_match(spec, CTSTATE_TO_REG,
+                                       &ctstate, &ctstate_mask);
+       if (ctstate_mask)
+               return -EOPNOTSUPP;
+
+       ctstate_mask |= MLX5_CT_STATE_TRK_BIT;
+       mlx5e_tc_match_to_reg_match(spec, CTSTATE_TO_REG,
+                                   ctstate, ctstate_mask);
+
+       return 0;
+}
+
 int
 mlx5_tc_ct_parse_match(struct mlx5e_priv *priv,
                       struct mlx5_flow_spec *spec,
index 626f6c0..94f74cf 100644 (file)
@@ -91,6 +91,9 @@ mlx5_tc_ct_parse_match(struct mlx5e_priv *priv,
                       struct flow_cls_offload *f,
                       struct netlink_ext_ack *extack);
 int
+mlx5_tc_ct_add_no_trk_match(struct mlx5e_priv *priv,
+                           struct mlx5_flow_spec *spec);
+int
 mlx5_tc_ct_parse_action(struct mlx5e_priv *priv,
                        struct mlx5_esw_flow_attr *attr,
                        const struct flow_action_entry *act,
@@ -140,6 +143,13 @@ mlx5_tc_ct_parse_match(struct mlx5e_priv *priv,
        return -EOPNOTSUPP;
 }
 
+static inline int
+mlx5_tc_ct_add_no_trk_match(struct mlx5e_priv *priv,
+                           struct mlx5_flow_spec *spec)
+{
+       return 0;
+}
+
 static inline int
 mlx5_tc_ct_parse_action(struct mlx5e_priv *priv,
                        struct mlx5_esw_flow_attr *attr,
index a6cb5d8..fd984ef 100644 (file)
@@ -219,6 +219,28 @@ mlx5e_tc_match_to_reg_match(struct mlx5_flow_spec *spec,
        spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS_2;
 }
 
+void
+mlx5e_tc_match_to_reg_get_match(struct mlx5_flow_spec *spec,
+                               enum mlx5e_tc_attr_to_reg type,
+                               u32 *data,
+                               u32 *mask)
+{
+       int soffset = mlx5e_tc_attr_to_reg_mappings[type].soffset;
+       int match_len = mlx5e_tc_attr_to_reg_mappings[type].mlen;
+       void *headers_c = spec->match_criteria;
+       void *headers_v = spec->match_value;
+       void *fmask, *fval;
+
+       fmask = headers_c + soffset;
+       fval = headers_v + soffset;
+
+       memcpy(mask, fmask, match_len);
+       memcpy(data, fval, match_len);
+
+       *mask = be32_to_cpu((__force __be32)(*mask << (32 - (match_len * 8))));
+       *data = be32_to_cpu((__force __be32)(*data << (32 - (match_len * 8))));
+}
+
 int
 mlx5e_tc_match_to_reg_set(struct mlx5_core_dev *mdev,
                          struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts,
@@ -3086,6 +3108,7 @@ struct ipv6_hoplimit_word {
 
 static int is_action_keys_supported(const struct flow_action_entry *act,
                                    bool ct_flow, bool *modify_ip_header,
+                                   bool *modify_tuple,
                                    struct netlink_ext_ack *extack)
 {
        u32 mask, offset;
@@ -3108,7 +3131,10 @@ static int is_action_keys_supported(const struct flow_action_entry *act,
                        *modify_ip_header = true;
                }
 
-               if (ct_flow && offset >= offsetof(struct iphdr, saddr)) {
+               if (offset >= offsetof(struct iphdr, saddr))
+                       *modify_tuple = true;
+
+               if (ct_flow && *modify_tuple) {
                        NL_SET_ERR_MSG_MOD(extack,
                                           "can't offload re-write of ipv4 address with action ct");
                        return -EOPNOTSUPP;
@@ -3123,16 +3149,22 @@ static int is_action_keys_supported(const struct flow_action_entry *act,
                        *modify_ip_header = true;
                }
 
-               if (ct_flow && offset >= offsetof(struct ipv6hdr, saddr)) {
+               if (ct_flow && offset >= offsetof(struct ipv6hdr, saddr))
+                       *modify_tuple = true;
+
+               if (ct_flow && *modify_tuple) {
                        NL_SET_ERR_MSG_MOD(extack,
                                           "can't offload re-write of ipv6 address with action ct");
                        return -EOPNOTSUPP;
                }
-       } else if (ct_flow && (htype == FLOW_ACT_MANGLE_HDR_TYPE_TCP ||
-                              htype == FLOW_ACT_MANGLE_HDR_TYPE_UDP)) {
-               NL_SET_ERR_MSG_MOD(extack,
-                                  "can't offload re-write of transport header ports with action ct");
-               return -EOPNOTSUPP;
+       } else if (htype == FLOW_ACT_MANGLE_HDR_TYPE_TCP ||
+                  htype == FLOW_ACT_MANGLE_HDR_TYPE_UDP) {
+               *modify_tuple = true;
+               if (ct_flow) {
+                       NL_SET_ERR_MSG_MOD(extack,
+                                          "can't offload re-write of transport header ports with action ct");
+                       return -EOPNOTSUPP;
+               }
        }
 
        return 0;
@@ -3142,10 +3174,11 @@ static bool modify_header_match_supported(struct mlx5e_priv *priv,
                                          struct mlx5_flow_spec *spec,
                                          struct flow_action *flow_action,
                                          u32 actions, bool ct_flow,
+                                         bool ct_clear,
                                          struct netlink_ext_ack *extack)
 {
        const struct flow_action_entry *act;
-       bool modify_ip_header;
+       bool modify_ip_header, modify_tuple;
        void *headers_c;
        void *headers_v;
        u16 ethertype;
@@ -3162,17 +3195,32 @@ static bool modify_header_match_supported(struct mlx5e_priv *priv,
                goto out_ok;
 
        modify_ip_header = false;
+       modify_tuple = false;
        flow_action_for_each(i, act, flow_action) {
                if (act->id != FLOW_ACTION_MANGLE &&
                    act->id != FLOW_ACTION_ADD)
                        continue;
 
                err = is_action_keys_supported(act, ct_flow,
-                                              &modify_ip_header, extack);
+                                              &modify_ip_header,
+                                              &modify_tuple, extack);
                if (err)
                        return err;
        }
 
+       /* Add ct_state=-trk match so it will be offloaded for non ct flows
+        * (or after clear action), as otherwise, since the tuple is changed,
+        *  we can't restore ct state
+        */
+       if (!ct_clear && modify_tuple &&
+           mlx5_tc_ct_add_no_trk_match(priv, spec)) {
+               NL_SET_ERR_MSG_MOD(extack,
+                                  "can't offload tuple modify header with ct matches");
+               netdev_info(priv->netdev,
+                           "can't offload tuple modify header with ct matches");
+               return false;
+       }
+
        ip_proto = MLX5_GET(fte_match_set_lyr_2_4, headers_v, ip_protocol);
        if (modify_ip_header && ip_proto != IPPROTO_TCP &&
            ip_proto != IPPROTO_UDP && ip_proto != IPPROTO_ICMP) {
@@ -3216,7 +3264,8 @@ static bool actions_match_supported(struct mlx5e_priv *priv,
        if (actions & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
                return modify_header_match_supported(priv, &parse_attr->spec,
                                                     flow_action, actions,
-                                                    ct_flow, extack);
+                                                    ct_flow, ct_clear,
+                                                    extack);
 
        return true;
 }
@@ -4483,11 +4532,12 @@ __mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
        if (err)
                goto err_free;
 
-       err = parse_tc_fdb_actions(priv, &rule->action, flow, extack, filter_dev);
+       /* actions validation depends on parsing the ct matches first */
+       err = mlx5_tc_ct_parse_match(priv, &parse_attr->spec, f, extack);
        if (err)
                goto err_free;
 
-       err = mlx5_tc_ct_parse_match(priv, &parse_attr->spec, f, extack);
+       err = parse_tc_fdb_actions(priv, &rule->action, flow, extack, filter_dev);
        if (err)
                goto err_free;
 
index 1561eaa..68d49b9 100644 (file)
@@ -164,6 +164,11 @@ void mlx5e_tc_match_to_reg_match(struct mlx5_flow_spec *spec,
                                 u32 data,
                                 u32 mask);
 
+void mlx5e_tc_match_to_reg_get_match(struct mlx5_flow_spec *spec,
+                                    enum mlx5e_tc_attr_to_reg type,
+                                    u32 *data,
+                                    u32 *mask);
+
 int alloc_mod_hdr_actions(struct mlx5_core_dev *mdev,
                          int namespace,
                          struct mlx5e_tc_mod_hdr_acts *mod_hdr_acts);