net/mlx5e: Configure IPsec packet offload flow steering
authorLeon Romanovsky <leonro@nvidia.com>
Fri, 2 Dec 2022 20:14:48 +0000 (22:14 +0200)
committerSteffen Klassert <steffen.klassert@secunet.com>
Thu, 8 Dec 2022 09:36:05 +0000 (10:36 +0100)
In packet offload mode, the HW is responsible to handle ESP headers,
SPI numbers and trailers (ICV) together with different logic for
RX and TX paths.

In order to support packet offload mode, special logic is added
to flow steering rules.

Reviewed-by: Raed Salem <raeds@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
drivers/net/ethernet/mellanox/mlx5/core/en_accel/ipsec.c
drivers/net/ethernet/mellanox/mlx5/core/en_accel/ipsec.h
drivers/net/ethernet/mellanox/mlx5/core/en_accel/ipsec_fs.c

index b22a31e..65f73a5 100644 (file)
@@ -162,6 +162,8 @@ mlx5e_ipsec_build_accel_xfrm_attrs(struct mlx5e_ipsec_sa_entry *sa_entry,
        memcpy(&aes_gcm->salt, x->aead->alg_key + key_len,
               sizeof(aes_gcm->salt));
 
+       attrs->authsize = crypto_aead_authsize(aead) / 4; /* in dwords */
+
        /* iv len */
        aes_gcm->icv_len = x->aead->alg_icv_len;
 
index 8d1a0d0..25b8655 100644 (file)
@@ -76,6 +76,7 @@ struct mlx5_accel_esp_xfrm_attrs {
        u8 type : 2;
        u8 family;
        u32 replay_window;
+       u32 authsize;
 };
 
 enum mlx5_ipsec_cap {
@@ -127,6 +128,7 @@ struct mlx5e_ipsec_esn_state {
 struct mlx5e_ipsec_rule {
        struct mlx5_flow_handle *rule;
        struct mlx5_modify_hdr *modify_hdr;
+       struct mlx5_pkt_reformat *pkt_reformat;
 };
 
 struct mlx5e_ipsec_modify_state_work {
index 893c186..dbe35ac 100644 (file)
@@ -480,6 +480,48 @@ static int setup_modify_header(struct mlx5_core_dev *mdev, u32 val, u8 dir,
        return 0;
 }
 
+static int setup_pkt_reformat(struct mlx5_core_dev *mdev,
+                             struct mlx5_accel_esp_xfrm_attrs *attrs,
+                             struct mlx5_flow_act *flow_act)
+{
+       enum mlx5_flow_namespace_type ns_type = MLX5_FLOW_NAMESPACE_EGRESS;
+       struct mlx5_pkt_reformat_params reformat_params = {};
+       struct mlx5_pkt_reformat *pkt_reformat;
+       u8 reformatbf[16] = {};
+       __be32 spi;
+
+       if (attrs->dir == XFRM_DEV_OFFLOAD_IN) {
+               reformat_params.type = MLX5_REFORMAT_TYPE_DEL_ESP_TRANSPORT;
+               ns_type = MLX5_FLOW_NAMESPACE_KERNEL;
+               goto cmd;
+       }
+
+       if (attrs->family == AF_INET)
+               reformat_params.type =
+                       MLX5_REFORMAT_TYPE_ADD_ESP_TRANSPORT_OVER_IPV4;
+       else
+               reformat_params.type =
+                       MLX5_REFORMAT_TYPE_ADD_ESP_TRANSPORT_OVER_IPV6;
+
+       /* convert to network format */
+       spi = htonl(attrs->spi);
+       memcpy(reformatbf, &spi, 4);
+
+       reformat_params.param_0 = attrs->authsize;
+       reformat_params.size = sizeof(reformatbf);
+       reformat_params.data = &reformatbf;
+
+cmd:
+       pkt_reformat =
+               mlx5_packet_reformat_alloc(mdev, &reformat_params, ns_type);
+       if (IS_ERR(pkt_reformat))
+               return PTR_ERR(pkt_reformat);
+
+       flow_act->pkt_reformat = pkt_reformat;
+       flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT;
+       return 0;
+}
+
 static int rx_add_rule(struct mlx5e_ipsec_sa_entry *sa_entry)
 {
        struct mlx5_accel_esp_xfrm_attrs *attrs = &sa_entry->attrs;
@@ -516,6 +558,16 @@ static int rx_add_rule(struct mlx5e_ipsec_sa_entry *sa_entry)
        if (err)
                goto err_mod_header;
 
+       switch (attrs->type) {
+       case XFRM_DEV_OFFLOAD_PACKET:
+               err = setup_pkt_reformat(mdev, attrs, &flow_act);
+               if (err)
+                       goto err_pkt_reformat;
+               break;
+       default:
+               break;
+       }
+
        flow_act.crypto.type = MLX5_FLOW_CONTEXT_ENCRYPT_DECRYPT_TYPE_IPSEC;
        flow_act.crypto.obj_id = sa_entry->ipsec_obj_id;
        flow_act.flags |= FLOW_ACT_NO_APPEND;
@@ -533,9 +585,13 @@ static int rx_add_rule(struct mlx5e_ipsec_sa_entry *sa_entry)
 
        sa_entry->ipsec_rule.rule = rule;
        sa_entry->ipsec_rule.modify_hdr = flow_act.modify_hdr;
+       sa_entry->ipsec_rule.pkt_reformat = flow_act.pkt_reformat;
        return 0;
 
 err_add_flow:
+       if (flow_act.pkt_reformat)
+               mlx5_packet_reformat_dealloc(mdev, flow_act.pkt_reformat);
+err_pkt_reformat:
        mlx5_modify_header_dealloc(mdev, flow_act.modify_hdr);
 err_mod_header:
        kvfree(spec);
@@ -562,7 +618,7 @@ static int tx_add_rule(struct mlx5e_ipsec_sa_entry *sa_entry)
        spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
        if (!spec) {
                err = -ENOMEM;
-               goto out;
+               goto err_alloc;
        }
 
        if (attrs->family == AF_INET)
@@ -570,29 +626,47 @@ static int tx_add_rule(struct mlx5e_ipsec_sa_entry *sa_entry)
        else
                setup_fte_addr6(spec, attrs->saddr.a6, attrs->daddr.a6);
 
-       setup_fte_spi(spec, attrs->spi);
-       setup_fte_esp(spec);
        setup_fte_no_frags(spec);
-       setup_fte_reg_a(spec);
+
+       switch (attrs->type) {
+       case XFRM_DEV_OFFLOAD_CRYPTO:
+               setup_fte_spi(spec, attrs->spi);
+               setup_fte_esp(spec);
+               setup_fte_reg_a(spec);
+               break;
+       case XFRM_DEV_OFFLOAD_PACKET:
+               err = setup_pkt_reformat(mdev, attrs, &flow_act);
+               if (err)
+                       goto err_pkt_reformat;
+               break;
+       default:
+               break;
+       }
 
        flow_act.crypto.type = MLX5_FLOW_CONTEXT_ENCRYPT_DECRYPT_TYPE_IPSEC;
        flow_act.crypto.obj_id = sa_entry->ipsec_obj_id;
        flow_act.flags |= FLOW_ACT_NO_APPEND;
-       flow_act.action = MLX5_FLOW_CONTEXT_ACTION_ALLOW |
-                         MLX5_FLOW_CONTEXT_ACTION_CRYPTO_ENCRYPT;
+       flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW |
+                          MLX5_FLOW_CONTEXT_ACTION_CRYPTO_ENCRYPT;
        rule = mlx5_add_flow_rules(tx->ft.sa, spec, &flow_act, NULL, 0);
        if (IS_ERR(rule)) {
                err = PTR_ERR(rule);
                mlx5_core_err(mdev, "fail to add TX ipsec rule err=%d\n", err);
-               goto out;
+               goto err_add_flow;
        }
 
+       kvfree(spec);
        sa_entry->ipsec_rule.rule = rule;
+       sa_entry->ipsec_rule.pkt_reformat = flow_act.pkt_reformat;
+       return 0;
 
-out:
+err_add_flow:
+       if (flow_act.pkt_reformat)
+               mlx5_packet_reformat_dealloc(mdev, flow_act.pkt_reformat);
+err_pkt_reformat:
        kvfree(spec);
-       if (err)
-               tx_ft_put(ipsec);
+err_alloc:
+       tx_ft_put(ipsec);
        return err;
 }
 
@@ -735,6 +809,9 @@ void mlx5e_accel_ipsec_fs_del_rule(struct mlx5e_ipsec_sa_entry *sa_entry)
 
        mlx5_del_flow_rules(ipsec_rule->rule);
 
+       if (ipsec_rule->pkt_reformat)
+               mlx5_packet_reformat_dealloc(mdev, ipsec_rule->pkt_reformat);
+
        if (sa_entry->attrs.dir == XFRM_DEV_OFFLOAD_OUT) {
                tx_ft_put(sa_entry->ipsec);
                return;