flow_offload: support get multi-subsystem block
authorwenxu <wenxu@ucloud.cn>
Wed, 7 Aug 2019 01:13:53 +0000 (09:13 +0800)
committerDavid S. Miller <davem@davemloft.net>
Fri, 9 Aug 2019 01:44:30 +0000 (18:44 -0700)
It provide a callback list to find the blocks of tc
and nft subsystems

Signed-off-by: wenxu <wenxu@ucloud.cn>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/flow_offload.h
net/core/flow_offload.c
net/sched/cls_api.c

index 46b8777..e8069b6 100644 (file)
@@ -379,6 +379,15 @@ typedef void flow_indr_block_ing_cmd_t(struct net_device *dev,
                                        void *cb_priv,
                                        enum flow_block_command command);
 
+struct flow_indr_block_ing_entry {
+       flow_indr_block_ing_cmd_t *cb;
+       struct list_head        list;
+};
+
+void flow_indr_add_block_ing_cb(struct flow_indr_block_ing_entry *entry);
+
+void flow_indr_del_block_ing_cb(struct flow_indr_block_ing_entry *entry);
+
 int __flow_indr_block_cb_register(struct net_device *dev, void *cb_priv,
                                  flow_indr_block_bind_cb_t *cb,
                                  void *cb_ident);
@@ -395,7 +404,6 @@ void flow_indr_block_cb_unregister(struct net_device *dev,
                                   void *cb_ident);
 
 void flow_indr_block_call(struct net_device *dev,
-                         flow_indr_block_ing_cmd_t *cb,
                          struct flow_block_offload *bo,
                          enum flow_block_command command);
 
index 4cc18e4..64c3d4d 100644 (file)
@@ -3,6 +3,7 @@
 #include <linux/slab.h>
 #include <net/flow_offload.h>
 #include <linux/rtnetlink.h>
+#include <linux/mutex.h>
 
 struct flow_rule *flow_rule_alloc(unsigned int num_actions)
 {
@@ -282,6 +283,8 @@ int flow_block_cb_setup_simple(struct flow_block_offload *f,
 }
 EXPORT_SYMBOL(flow_block_cb_setup_simple);
 
+static LIST_HEAD(block_ing_cb_list);
+
 static struct rhashtable indr_setup_block_ht;
 
 struct flow_indr_block_cb {
@@ -295,7 +298,6 @@ struct flow_indr_block_dev {
        struct rhash_head ht_node;
        struct net_device *dev;
        unsigned int refcnt;
-       flow_indr_block_ing_cmd_t  *block_ing_cmd_cb;
        struct list_head cb_list;
 };
 
@@ -389,6 +391,20 @@ static void flow_indr_block_cb_del(struct flow_indr_block_cb *indr_block_cb)
        kfree(indr_block_cb);
 }
 
+static void flow_block_ing_cmd(struct net_device *dev,
+                              flow_indr_block_bind_cb_t *cb,
+                              void *cb_priv,
+                              enum flow_block_command command)
+{
+       struct flow_indr_block_ing_entry *entry;
+
+       rcu_read_lock();
+       list_for_each_entry_rcu(entry, &block_ing_cb_list, list) {
+               entry->cb(dev, cb, cb_priv, command);
+       }
+       rcu_read_unlock();
+}
+
 int __flow_indr_block_cb_register(struct net_device *dev, void *cb_priv,
                                  flow_indr_block_bind_cb_t *cb,
                                  void *cb_ident)
@@ -406,10 +422,8 @@ int __flow_indr_block_cb_register(struct net_device *dev, void *cb_priv,
        if (err)
                goto err_dev_put;
 
-       if (indr_dev->block_ing_cmd_cb)
-               indr_dev->block_ing_cmd_cb(dev, indr_block_cb->cb,
-                                          indr_block_cb->cb_priv,
-                                          FLOW_BLOCK_BIND);
+       flow_block_ing_cmd(dev, indr_block_cb->cb, indr_block_cb->cb_priv,
+                          FLOW_BLOCK_BIND);
 
        return 0;
 
@@ -448,10 +462,8 @@ void __flow_indr_block_cb_unregister(struct net_device *dev,
        if (!indr_block_cb)
                return;
 
-       if (indr_dev->block_ing_cmd_cb)
-               indr_dev->block_ing_cmd_cb(dev, indr_block_cb->cb,
-                                          indr_block_cb->cb_priv,
-                                          FLOW_BLOCK_UNBIND);
+       flow_block_ing_cmd(dev, indr_block_cb->cb, indr_block_cb->cb_priv,
+                          FLOW_BLOCK_UNBIND);
 
        flow_indr_block_cb_del(indr_block_cb);
        flow_indr_block_dev_put(indr_dev);
@@ -469,7 +481,6 @@ void flow_indr_block_cb_unregister(struct net_device *dev,
 EXPORT_SYMBOL_GPL(flow_indr_block_cb_unregister);
 
 void flow_indr_block_call(struct net_device *dev,
-                         flow_indr_block_ing_cmd_t cb,
                          struct flow_block_offload *bo,
                          enum flow_block_command command)
 {
@@ -480,15 +491,29 @@ void flow_indr_block_call(struct net_device *dev,
        if (!indr_dev)
                return;
 
-       indr_dev->block_ing_cmd_cb = command == FLOW_BLOCK_BIND
-                                    ? cb : NULL;
-
        list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
                indr_block_cb->cb(dev, indr_block_cb->cb_priv, TC_SETUP_BLOCK,
                                  bo);
 }
 EXPORT_SYMBOL_GPL(flow_indr_block_call);
 
+static DEFINE_MUTEX(flow_indr_block_ing_cb_lock);
+void flow_indr_add_block_ing_cb(struct flow_indr_block_ing_entry *entry)
+{
+       mutex_lock(&flow_indr_block_ing_cb_lock);
+       list_add_tail_rcu(&entry->list, &block_ing_cb_list);
+       mutex_unlock(&flow_indr_block_ing_cb_lock);
+}
+EXPORT_SYMBOL_GPL(flow_indr_add_block_ing_cb);
+
+void flow_indr_del_block_ing_cb(struct flow_indr_block_ing_entry *entry)
+{
+       mutex_lock(&flow_indr_block_ing_cb_lock);
+       list_del_rcu(&entry->list);
+       mutex_unlock(&flow_indr_block_ing_cb_lock);
+}
+EXPORT_SYMBOL_GPL(flow_indr_del_block_ing_cb);
+
 static int __init init_flow_indr_rhashtable(void)
 {
        return rhashtable_init(&indr_setup_block_ht,
index 0b0dde2..e0d8b45 100644 (file)
@@ -621,7 +621,7 @@ static void tc_indr_block_call(struct tcf_block *block,
        };
        INIT_LIST_HEAD(&bo.cb_list);
 
-       flow_indr_block_call(dev, tc_indr_block_get_and_ing_cmd, &bo, command);
+       flow_indr_block_call(dev, &bo, command);
        tcf_block_setup(block, &bo);
 }
 
@@ -3183,6 +3183,11 @@ static struct pernet_operations tcf_net_ops = {
        .size = sizeof(struct tcf_net),
 };
 
+static struct flow_indr_block_ing_entry block_ing_entry = {
+       .cb = tc_indr_block_get_and_ing_cmd,
+       .list = LIST_HEAD_INIT(block_ing_entry.list),
+};
+
 static int __init tc_filter_init(void)
 {
        int err;
@@ -3195,6 +3200,8 @@ static int __init tc_filter_init(void)
        if (err)
                goto err_register_pernet_subsys;
 
+       flow_indr_add_block_ing_cb(&block_ing_entry);
+
        rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
                      RTNL_FLAG_DOIT_UNLOCKED);
        rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,