switchdev: add offload_fwd_mark generator helper
[deliverable/linux.git] / net / switchdev / switchdev.c
index 84f77a0540251cbe127785de670e5b0a41a3bdac..33bafa2e703e299f3b423d1f95b2a21cf177c634 100644 (file)
@@ -171,8 +171,10 @@ int switchdev_port_attr_set(struct net_device *dev, struct switchdev_attr *attr)
                 * released.
                 */
 
-               attr->trans = SWITCHDEV_TRANS_ABORT;
-               __switchdev_port_attr_set(dev, attr);
+               if (err != -EOPNOTSUPP) {
+                       attr->trans = SWITCHDEV_TRANS_ABORT;
+                       __switchdev_port_attr_set(dev, attr);
+               }
 
                return err;
        }
@@ -249,8 +251,10 @@ int switchdev_port_obj_add(struct net_device *dev, struct switchdev_obj *obj)
                 * released.
                 */
 
-               obj->trans = SWITCHDEV_TRANS_ABORT;
-               __switchdev_port_obj_add(dev, obj);
+               if (err != -EOPNOTSUPP) {
+                       obj->trans = SWITCHDEV_TRANS_ABORT;
+                       __switchdev_port_obj_add(dev, obj);
+               }
 
                return err;
        }
@@ -906,13 +910,9 @@ static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
                if (switchdev_port_attr_get(dev, &attr))
                        return NULL;
 
-               if (nhsel > 0) {
-                       if (prev_attr.u.ppid.id_len != attr.u.ppid.id_len)
-                               return NULL;
-                       if (memcmp(prev_attr.u.ppid.id, attr.u.ppid.id,
-                                  attr.u.ppid.id_len))
+               if (nhsel > 0 &&
+                   !netdev_phys_item_id_same(&prev_attr.u.ppid, &attr.u.ppid))
                                return NULL;
-               }
 
                prev_attr = attr;
        }
@@ -1039,3 +1039,106 @@ void switchdev_fib_ipv4_abort(struct fib_info *fi)
        fi->fib_net->ipv4.fib_offload_disabled = true;
 }
 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_abort);
+
+static bool switchdev_port_same_parent_id(struct net_device *a,
+                                         struct net_device *b)
+{
+       struct switchdev_attr a_attr = {
+               .id = SWITCHDEV_ATTR_PORT_PARENT_ID,
+               .flags = SWITCHDEV_F_NO_RECURSE,
+       };
+       struct switchdev_attr b_attr = {
+               .id = SWITCHDEV_ATTR_PORT_PARENT_ID,
+               .flags = SWITCHDEV_F_NO_RECURSE,
+       };
+
+       if (switchdev_port_attr_get(a, &a_attr) ||
+           switchdev_port_attr_get(b, &b_attr))
+               return false;
+
+       return netdev_phys_item_id_same(&a_attr.u.ppid, &b_attr.u.ppid);
+}
+
+static u32 switchdev_port_fwd_mark_get(struct net_device *dev,
+                                      struct net_device *group_dev)
+{
+       struct net_device *lower_dev;
+       struct list_head *iter;
+
+       netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
+               if (lower_dev == dev)
+                       continue;
+               if (switchdev_port_same_parent_id(dev, lower_dev))
+                       return lower_dev->offload_fwd_mark;
+               return switchdev_port_fwd_mark_get(dev, lower_dev);
+       }
+
+       return dev->ifindex;
+}
+
+static void switchdev_port_fwd_mark_reset(struct net_device *group_dev,
+                                         u32 old_mark, u32 *reset_mark)
+{
+       struct net_device *lower_dev;
+       struct list_head *iter;
+
+       netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
+               if (lower_dev->offload_fwd_mark == old_mark) {
+                       if (!*reset_mark)
+                               *reset_mark = lower_dev->ifindex;
+                       lower_dev->offload_fwd_mark = *reset_mark;
+               }
+               switchdev_port_fwd_mark_reset(lower_dev, old_mark, reset_mark);
+       }
+}
+
+/**
+ *     switchdev_port_fwd_mark_set - Set port offload forwarding mark
+ *
+ *     @dev: port device
+ *     @group_dev: containing device
+ *     @joining: true if dev is joining group; false if leaving group
+ *
+ *     An ungrouped port's offload mark is just its ifindex.  A grouped
+ *     port's (member of a bridge, for example) offload mark is the ifindex
+ *     of one of the ports in the group with the same parent (switch) ID.
+ *     Ports on the same device in the same group will have the same mark.
+ *
+ *     Example:
+ *
+ *             br0             ifindex=9
+ *               sw1p1         ifindex=2       mark=2
+ *               sw1p2         ifindex=3       mark=2
+ *               sw2p1         ifindex=4       mark=5
+ *               sw2p2         ifindex=5       mark=5
+ *
+ *     If sw2p2 leaves the bridge, we'll have:
+ *
+ *             br0             ifindex=9
+ *               sw1p1         ifindex=2       mark=2
+ *               sw1p2         ifindex=3       mark=2
+ *               sw2p1         ifindex=4       mark=4
+ *             sw2p2           ifindex=5       mark=5
+ */
+void switchdev_port_fwd_mark_set(struct net_device *dev,
+                                struct net_device *group_dev,
+                                bool joining)
+{
+       u32 mark = dev->ifindex;
+       u32 reset_mark = 0;
+
+       if (group_dev && joining) {
+               mark = switchdev_port_fwd_mark_get(dev, group_dev);
+       } else if (group_dev && !joining) {
+               if (dev->offload_fwd_mark == mark)
+                       /* Ohoh, this port was the mark reference port,
+                        * but it's leaving the group, so reset the
+                        * mark for the remaining ports in the group.
+                        */
+                       switchdev_port_fwd_mark_reset(group_dev, mark,
+                                                     &reset_mark);
+       }
+
+       dev->offload_fwd_mark = mark;
+}
+EXPORT_SYMBOL_GPL(switchdev_port_fwd_mark_set);
This page took 0.027154 seconds and 5 git commands to generate.