net/mlx5: Add support in forward to namespace
authorMaor Gottlieb <maorg@mellanox.com>
Mon, 4 May 2020 05:30:10 +0000 (08:30 +0300)
committerLeon Romanovsky <leonro@mellanox.com>
Wed, 13 May 2020 15:56:31 +0000 (18:56 +0300)
Currently, fs_core supports rule of forward the traffic
to continue matching in the next priority, now we add support
to forward the traffic matching in the next namespace.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Mark Zhang <markz@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
drivers/net/ethernet/mellanox/mlx5/core/fs_core.c
drivers/net/ethernet/mellanox/mlx5/core/fs_core.h
include/linux/mlx5/fs.h

index 705f433..41aa1fa 100644 (file)
@@ -384,6 +384,12 @@ static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
        return NULL;
 }
 
+static bool is_fwd_next_action(u32 action)
+{
+       return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
+                        MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
+}
+
 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 {
        int i;
@@ -502,7 +508,7 @@ static void del_sw_hw_rule(struct fs_node *node)
        fs_get_obj(rule, node);
        fs_get_obj(fte, rule->node.parent);
        trace_mlx5_fs_del_rule(rule);
-       if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
+       if (is_fwd_next_action(rule->sw_action)) {
                mutex_lock(&rule->dest_attr.ft->lock);
                list_del(&rule->next_ft);
                mutex_unlock(&rule->dest_attr.ft->lock);
@@ -826,6 +832,36 @@ static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
        return find_closest_ft(prio, true);
 }
 
+static struct fs_prio *find_fwd_ns_prio(struct mlx5_flow_root_namespace *root,
+                                       struct mlx5_flow_namespace *ns)
+{
+       struct mlx5_flow_namespace *root_ns = &root->ns;
+       struct fs_prio *iter_prio;
+       struct fs_prio *prio;
+
+       fs_get_obj(prio, ns->node.parent);
+       list_for_each_entry(iter_prio, &root_ns->node.children, node.list) {
+               if (iter_prio == prio &&
+                   !list_is_last(&prio->node.children, &iter_prio->node.list))
+                       return list_next_entry(iter_prio, node.list);
+       }
+       return NULL;
+}
+
+static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft,
+                                               struct mlx5_flow_act *flow_act)
+{
+       struct mlx5_flow_root_namespace *root = find_root(&ft->node);
+       struct fs_prio *prio;
+
+       if (flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS)
+               prio = find_fwd_ns_prio(root, ft->ns);
+       else
+               fs_get_obj(prio, ft->node.parent);
+
+       return (prio) ? find_next_chained_ft(prio) : NULL;
+}
+
 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
                               struct fs_prio *prio,
                               struct mlx5_flow_table *ft)
@@ -976,6 +1012,10 @@ static int connect_fwd_rules(struct mlx5_core_dev *dev,
        list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
        mutex_unlock(&old_next_ft->lock);
        list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
+               if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) &&
+                   iter->ft->ns == new_next_ft->ns)
+                       continue;
+
                err = _mlx5_modify_rule_destination(iter, &dest);
                if (err)
                        pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
@@ -1077,6 +1117,7 @@ static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespa
        next_ft = unmanaged ? ft_attr->next_ft :
                              find_next_chained_ft(fs_prio);
        ft->def_miss_action = ns->def_miss_action;
+       ft->ns = ns;
        err = root->cmds->create_flow_table(root, ft, log_table_sz, next_ft);
        if (err)
                goto free_ft;
@@ -1903,21 +1944,19 @@ mlx5_add_flow_rules(struct mlx5_flow_table *ft,
        struct mlx5_flow_table *next_ft = NULL;
        struct mlx5_flow_handle *handle = NULL;
        u32 sw_action = flow_act->action;
-       struct fs_prio *prio;
        int i;
 
        if (!spec)
                spec = &zero_spec;
 
-       if (!(sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO))
+       if (!is_fwd_next_action(sw_action))
                return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
 
        if (!fwd_next_prio_supported(ft))
                return ERR_PTR(-EOPNOTSUPP);
 
        mutex_lock(&root->chain_lock);
-       fs_get_obj(prio, ft->node.parent);
-       next_ft = find_next_chained_ft(prio);
+       next_ft = find_next_fwd_ft(ft, flow_act);
        if (!next_ft) {
                handle = ERR_PTR(-EOPNOTSUPP);
                goto unlock;
@@ -1936,8 +1975,8 @@ mlx5_add_flow_rules(struct mlx5_flow_table *ft,
        gen_dest[i].ft = next_ft;
        dest = gen_dest;
        num_dest++;
-       flow_act->action &=
-               ~MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
+       flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
+                             MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
        flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
        handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
        if (IS_ERR(handle))
@@ -1948,8 +1987,8 @@ mlx5_add_flow_rules(struct mlx5_flow_table *ft,
                list_add(&handle->rule[num_dest - 1]->next_ft,
                         &next_ft->fwd_rules);
                mutex_unlock(&next_ft->lock);
-               handle->rule[num_dest - 1]->sw_action =
-                       MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
+               handle->rule[num_dest - 1]->sw_action = sw_action;
+               handle->rule[num_dest - 1]->ft = ft;
        }
 unlock:
        mutex_unlock(&root->chain_lock);
index 508108c..825b662 100644 (file)
@@ -138,6 +138,7 @@ struct fs_node {
 
 struct mlx5_flow_rule {
        struct fs_node                          node;
+       struct mlx5_flow_table                  *ft;
        struct mlx5_flow_destination            dest_attr;
        /* next_ft should be accessed under chain_lock and only of
         * destination type is FWD_NEXT_fT.
@@ -175,6 +176,7 @@ struct mlx5_flow_table {
        u32                             flags;
        struct rhltable                 fgs_hash;
        enum mlx5_flow_table_miss_action def_miss_action;
+       struct mlx5_flow_namespace      *ns;
 };
 
 struct mlx5_ft_underlay_qp {
index e2d13e0..6c5aa0a 100644 (file)
@@ -42,6 +42,7 @@ enum {
        MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO  = 1 << 16,
        MLX5_FLOW_CONTEXT_ACTION_ENCRYPT        = 1 << 17,
        MLX5_FLOW_CONTEXT_ACTION_DECRYPT        = 1 << 18,
+       MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS    = 1 << 19,
 };
 
 enum {