switchdev: add new switchdev_port_bridge_dellink
[deliverable/linux.git] / net / switchdev / switchdev.c
index 52613ed49a8c9d59e4961443417dfc985f35f3e7..8ce678e397b4d90e40464f989e2b3a8576b1c4a4 100644 (file)
 #include <linux/mutex.h>
 #include <linux/notifier.h>
 #include <linux/netdevice.h>
+#include <linux/if_bridge.h>
 #include <net/ip_fib.h>
 #include <net/switchdev.h>
 
 /**
- *     switchdev_parent_id_get - Get ID of a switch
+ *     switchdev_port_attr_get - Get port attribute
+ *
  *     @dev: port device
- *     @psid: switch ID
+ *     @attr: attribute to get
+ */
+int switchdev_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
+{
+       const struct switchdev_ops *ops = dev->switchdev_ops;
+       struct net_device *lower_dev;
+       struct list_head *iter;
+       struct switchdev_attr first = {
+               .id = SWITCHDEV_ATTR_UNDEFINED
+       };
+       int err = -EOPNOTSUPP;
+
+       if (ops && ops->switchdev_port_attr_get)
+               return ops->switchdev_port_attr_get(dev, attr);
+
+       if (attr->flags & SWITCHDEV_F_NO_RECURSE)
+               return err;
+
+       /* Switch device port(s) may be stacked under
+        * bond/team/vlan dev, so recurse down to get attr on
+        * each port.  Return -ENODATA if attr values don't
+        * compare across ports.
+        */
+
+       netdev_for_each_lower_dev(dev, lower_dev, iter) {
+               err = switchdev_port_attr_get(lower_dev, attr);
+               if (err)
+                       break;
+               if (first.id == SWITCHDEV_ATTR_UNDEFINED)
+                       first = *attr;
+               else if (memcmp(&first, attr, sizeof(*attr)))
+                       return -ENODATA;
+       }
+
+       return err;
+}
+EXPORT_SYMBOL_GPL(switchdev_port_attr_get);
+
+static int __switchdev_port_attr_set(struct net_device *dev,
+                                    struct switchdev_attr *attr)
+{
+       const struct switchdev_ops *ops = dev->switchdev_ops;
+       struct net_device *lower_dev;
+       struct list_head *iter;
+       int err = -EOPNOTSUPP;
+
+       if (ops && ops->switchdev_port_attr_set)
+               return ops->switchdev_port_attr_set(dev, attr);
+
+       if (attr->flags & SWITCHDEV_F_NO_RECURSE)
+               return err;
+
+       /* Switch device port(s) may be stacked under
+        * bond/team/vlan dev, so recurse down to set attr on
+        * each port.
+        */
+
+       netdev_for_each_lower_dev(dev, lower_dev, iter) {
+               err = __switchdev_port_attr_set(lower_dev, attr);
+               if (err)
+                       break;
+       }
+
+       return err;
+}
+
+struct switchdev_attr_set_work {
+       struct work_struct work;
+       struct net_device *dev;
+       struct switchdev_attr attr;
+};
+
+static void switchdev_port_attr_set_work(struct work_struct *work)
+{
+       struct switchdev_attr_set_work *asw =
+               container_of(work, struct switchdev_attr_set_work, work);
+       int err;
+
+       rtnl_lock();
+       err = switchdev_port_attr_set(asw->dev, &asw->attr);
+       BUG_ON(err);
+       rtnl_unlock();
+
+       dev_put(asw->dev);
+       kfree(work);
+}
+
+static int switchdev_port_attr_set_defer(struct net_device *dev,
+                                        struct switchdev_attr *attr)
+{
+       struct switchdev_attr_set_work *asw;
+
+       asw = kmalloc(sizeof(*asw), GFP_ATOMIC);
+       if (!asw)
+               return -ENOMEM;
+
+       INIT_WORK(&asw->work, switchdev_port_attr_set_work);
+
+       dev_hold(dev);
+       asw->dev = dev;
+       memcpy(&asw->attr, attr, sizeof(asw->attr));
+
+       schedule_work(&asw->work);
+
+       return 0;
+}
+
+/**
+ *     switchdev_port_attr_set - Set port attribute
+ *
+ *     @dev: port device
+ *     @attr: attribute to set
  *
- *     Get ID of a switch this port is part of.
+ *     Use a 2-phase prepare-commit transaction model to ensure
+ *     system is not left in a partially updated state due to
+ *     failure from driver/device.
  */
-int switchdev_parent_id_get(struct net_device *dev,
-                           struct netdev_phys_item_id *psid)
+int switchdev_port_attr_set(struct net_device *dev, struct switchdev_attr *attr)
+{
+       int err;
+
+       if (!rtnl_is_locked()) {
+               /* Running prepare-commit transaction across stacked
+                * devices requires nothing moves, so if rtnl_lock is
+                * not held, schedule a worker thread to hold rtnl_lock
+                * while setting attr.
+                */
+
+               return switchdev_port_attr_set_defer(dev, attr);
+       }
+
+       /* Phase I: prepare for attr set. Driver/device should fail
+        * here if there are going to be issues in the commit phase,
+        * such as lack of resources or support.  The driver/device
+        * should reserve resources needed for the commit phase here,
+        * but should not commit the attr.
+        */
+
+       attr->trans = SWITCHDEV_TRANS_PREPARE;
+       err = __switchdev_port_attr_set(dev, attr);
+       if (err) {
+               /* Prepare phase failed: abort the transaction.  Any
+                * resources reserved in the prepare phase are
+                * released.
+                */
+
+               attr->trans = SWITCHDEV_TRANS_ABORT;
+               __switchdev_port_attr_set(dev, attr);
+
+               return err;
+       }
+
+       /* Phase II: commit attr set.  This cannot fail as a fault
+        * of driver/device.  If it does, it's a bug in the driver/device
+        * because the driver said everythings was OK in phase I.
+        */
+
+       attr->trans = SWITCHDEV_TRANS_COMMIT;
+       err = __switchdev_port_attr_set(dev, attr);
+       BUG_ON(err);
+
+       return err;
+}
+EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
+
+int __switchdev_port_obj_add(struct net_device *dev, struct switchdev_obj *obj)
 {
-       const struct swdev_ops *ops = dev->swdev_ops;
+       const struct switchdev_ops *ops = dev->switchdev_ops;
+       struct net_device *lower_dev;
+       struct list_head *iter;
+       int err = -EOPNOTSUPP;
+
+       if (ops && ops->switchdev_port_obj_add)
+               return ops->switchdev_port_obj_add(dev, obj);
+
+       /* Switch device port(s) may be stacked under
+        * bond/team/vlan dev, so recurse down to add object on
+        * each port.
+        */
+
+       netdev_for_each_lower_dev(dev, lower_dev, iter) {
+               err = __switchdev_port_obj_add(lower_dev, obj);
+               if (err)
+                       break;
+       }
 
-       if (!ops || !ops->swdev_parent_id_get)
-               return -EOPNOTSUPP;
-       return ops->swdev_parent_id_get(dev, psid);
+       return err;
 }
-EXPORT_SYMBOL_GPL(switchdev_parent_id_get);
 
 /**
- *     switchdev_port_stp_update - Notify switch device port of STP
- *                                     state change
+ *     switchdev_port_obj_add - Add port object
+ *
  *     @dev: port device
- *     @state: port STP state
+ *     @obj: object to add
+ *
+ *     Use a 2-phase prepare-commit transaction model to ensure
+ *     system is not left in a partially updated state due to
+ *     failure from driver/device.
  *
- *     Notify switch device port of bridge port STP state change.
+ *     rtnl_lock must be held.
  */
-int switchdev_port_stp_update(struct net_device *dev, u8 state)
+int switchdev_port_obj_add(struct net_device *dev, struct switchdev_obj *obj)
 {
-       const struct swdev_ops *ops = dev->swdev_ops;
+       int err;
+
+       ASSERT_RTNL();
+
+       /* Phase I: prepare for obj add. Driver/device should fail
+        * here if there are going to be issues in the commit phase,
+        * such as lack of resources or support.  The driver/device
+        * should reserve resources needed for the commit phase here,
+        * but should not commit the obj.
+        */
+
+       obj->trans = SWITCHDEV_TRANS_PREPARE;
+       err = __switchdev_port_obj_add(dev, obj);
+       if (err) {
+               /* Prepare phase failed: abort the transaction.  Any
+                * resources reserved in the prepare phase are
+                * released.
+                */
+
+               obj->trans = SWITCHDEV_TRANS_ABORT;
+               __switchdev_port_obj_add(dev, obj);
+
+               return err;
+       }
+
+       /* Phase II: commit obj add.  This cannot fail as a fault
+        * of driver/device.  If it does, it's a bug in the driver/device
+        * because the driver said everythings was OK in phase I.
+        */
+
+       obj->trans = SWITCHDEV_TRANS_COMMIT;
+       err = __switchdev_port_obj_add(dev, obj);
+       WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
+
+       return err;
+}
+EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
+
+/**
+ *     switchdev_port_obj_del - Delete port object
+ *
+ *     @dev: port device
+ *     @obj: object to delete
+ */
+int switchdev_port_obj_del(struct net_device *dev, struct switchdev_obj *obj)
+{
+       const struct switchdev_ops *ops = dev->switchdev_ops;
        struct net_device *lower_dev;
        struct list_head *iter;
        int err = -EOPNOTSUPP;
 
-       if (ops && ops->swdev_port_stp_update)
-               return ops->swdev_port_stp_update(dev, state);
+       if (ops && ops->switchdev_port_obj_del)
+               return ops->switchdev_port_obj_del(dev, obj);
+
+       /* Switch device port(s) may be stacked under
+        * bond/team/vlan dev, so recurse down to delete object on
+        * each port.
+        */
 
        netdev_for_each_lower_dev(dev, lower_dev, iter) {
-               err = switchdev_port_stp_update(lower_dev, state);
-               if (err && err != -EOPNOTSUPP)
-                       return err;
+               err = switchdev_port_obj_del(lower_dev, obj);
+               if (err)
+                       break;
        }
 
        return err;
 }
-EXPORT_SYMBOL_GPL(switchdev_port_stp_update);
+EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
 
 static DEFINE_MUTEX(switchdev_mutex);
 static RAW_NOTIFIER_HEAD(switchdev_notif_chain);
@@ -127,85 +358,183 @@ int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
 }
 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
 
-/**
- *     switchdev_port_bridge_setlink - Notify switch device port of bridge
- *     port attributes
- *
- *     @dev: port device
- *     @nlh: netlink msg with bridge port attributes
- *     @flags: bridge setlink flags
- *
- *     Notify switch device port of bridge port attributes
- */
-int switchdev_port_bridge_setlink(struct net_device *dev,
-                                 struct nlmsghdr *nlh, u16 flags)
+static int switchdev_port_br_setflag(struct net_device *dev,
+                                    struct nlattr *nlattr,
+                                    unsigned long brport_flag)
 {
-       const struct net_device_ops *ops = dev->netdev_ops;
+       struct switchdev_attr attr = {
+               .id = SWITCHDEV_ATTR_PORT_BRIDGE_FLAGS,
+       };
+       u8 flag = nla_get_u8(nlattr);
+       int err;
 
-       if (!(dev->features & NETIF_F_HW_SWITCH_OFFLOAD))
-               return 0;
+       err = switchdev_port_attr_get(dev, &attr);
+       if (err)
+               return err;
 
-       if (!ops->ndo_bridge_setlink)
-               return -EOPNOTSUPP;
+       if (flag)
+               attr.brport_flags |= brport_flag;
+       else
+               attr.brport_flags &= ~brport_flag;
 
-       return ops->ndo_bridge_setlink(dev, nlh, flags);
+       return switchdev_port_attr_set(dev, &attr);
+}
+
+static const struct nla_policy
+switchdev_port_bridge_policy[IFLA_BRPORT_MAX + 1] = {
+       [IFLA_BRPORT_STATE]             = { .type = NLA_U8 },
+       [IFLA_BRPORT_COST]              = { .type = NLA_U32 },
+       [IFLA_BRPORT_PRIORITY]          = { .type = NLA_U16 },
+       [IFLA_BRPORT_MODE]              = { .type = NLA_U8 },
+       [IFLA_BRPORT_GUARD]             = { .type = NLA_U8 },
+       [IFLA_BRPORT_PROTECT]           = { .type = NLA_U8 },
+       [IFLA_BRPORT_FAST_LEAVE]        = { .type = NLA_U8 },
+       [IFLA_BRPORT_LEARNING]          = { .type = NLA_U8 },
+       [IFLA_BRPORT_LEARNING_SYNC]     = { .type = NLA_U8 },
+       [IFLA_BRPORT_UNICAST_FLOOD]     = { .type = NLA_U8 },
+};
+
+static int switchdev_port_br_setlink_protinfo(struct net_device *dev,
+                                             struct nlattr *protinfo)
+{
+       struct nlattr *attr;
+       int rem;
+       int err;
+
+       err = nla_validate_nested(protinfo, IFLA_BRPORT_MAX,
+                                 switchdev_port_bridge_policy);
+       if (err)
+               return err;
+
+       nla_for_each_nested(attr, protinfo, rem) {
+               switch (nla_type(attr)) {
+               case IFLA_BRPORT_LEARNING:
+                       err = switchdev_port_br_setflag(dev, attr,
+                                                       BR_LEARNING);
+                       break;
+               case IFLA_BRPORT_LEARNING_SYNC:
+                       err = switchdev_port_br_setflag(dev, attr,
+                                                       BR_LEARNING_SYNC);
+                       break;
+               default:
+                       err = -EOPNOTSUPP;
+                       break;
+               }
+               if (err)
+                       return err;
+       }
+
+       return 0;
+}
+
+static int switchdev_port_br_afspec(struct net_device *dev,
+                                   struct nlattr *afspec,
+                                   int (*f)(struct net_device *dev,
+                                            struct switchdev_obj *obj))
+{
+       struct nlattr *attr;
+       struct bridge_vlan_info *vinfo;
+       struct switchdev_obj obj = {
+               .id = SWITCHDEV_OBJ_PORT_VLAN,
+       };
+       int rem;
+       int err;
+
+       nla_for_each_nested(attr, afspec, rem) {
+               if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
+                       continue;
+               if (nla_len(attr) != sizeof(struct bridge_vlan_info))
+                       return -EINVAL;
+               vinfo = nla_data(attr);
+               obj.vlan.flags = vinfo->flags;
+               if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
+                       if (obj.vlan.vid_start)
+                               return -EINVAL;
+                       obj.vlan.vid_start = vinfo->vid;
+               } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
+                       if (!obj.vlan.vid_start)
+                               return -EINVAL;
+                       obj.vlan.vid_end = vinfo->vid;
+                       if (obj.vlan.vid_end <= obj.vlan.vid_start)
+                               return -EINVAL;
+                       err = f(dev, &obj);
+                       if (err)
+                               return err;
+                       memset(&obj.vlan, 0, sizeof(obj.vlan));
+               } else {
+                       if (obj.vlan.vid_start)
+                               return -EINVAL;
+                       obj.vlan.vid_start = vinfo->vid;
+                       obj.vlan.vid_end = vinfo->vid;
+                       err = f(dev, &obj);
+                       if (err)
+                               return err;
+                       memset(&obj.vlan, 0, sizeof(obj.vlan));
+               }
+       }
+
+       return 0;
 }
-EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
 
 /**
- *     switchdev_port_bridge_dellink - Notify switch device port of bridge
- *     port attribute delete
+ *     switchdev_port_bridge_setlink - Set bridge port attributes
  *
  *     @dev: port device
- *     @nlh: netlink msg with bridge port attributes
- *     @flags: bridge setlink flags
+ *     @nlh: netlink header
+ *     @flags: netlink flags
  *
- *     Notify switch device port of bridge port attribute delete
+ *     Called for SELF on rtnl_bridge_setlink to set bridge port
+ *     attributes.
  */
-int switchdev_port_bridge_dellink(struct net_device *dev,
+int switchdev_port_bridge_setlink(struct net_device *dev,
                                  struct nlmsghdr *nlh, u16 flags)
 {
-       const struct net_device_ops *ops = dev->netdev_ops;
+       struct nlattr *protinfo;
+       struct nlattr *afspec;
+       int err = 0;
 
-       if (!(dev->features & NETIF_F_HW_SWITCH_OFFLOAD))
-               return 0;
+       protinfo = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
+                                  IFLA_PROTINFO);
+       if (protinfo) {
+               err = switchdev_port_br_setlink_protinfo(dev, protinfo);
+               if (err)
+                       return err;
+       }
 
-       if (!ops->ndo_bridge_dellink)
-               return -EOPNOTSUPP;
+       afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
+                                IFLA_AF_SPEC);
+       if (afspec)
+               err = switchdev_port_br_afspec(dev, afspec,
+                                              switchdev_port_obj_add);
 
-       return ops->ndo_bridge_dellink(dev, nlh, flags);
+       return err;
 }
-EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
+EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
 
 /**
- *     ndo_dflt_switchdev_port_bridge_setlink - default ndo bridge setlink
- *                                              op for master devices
+ *     switchdev_port_bridge_dellink - Set bridge port attributes
  *
  *     @dev: port device
- *     @nlh: netlink msg with bridge port attributes
- *     @flags: bridge setlink flags
+ *     @nlh: netlink header
+ *     @flags: netlink flags
  *
- *     Notify master device slaves of bridge port attributes
+ *     Called for SELF on rtnl_bridge_dellink to set bridge port
+ *     attributes.
  */
-int ndo_dflt_switchdev_port_bridge_setlink(struct net_device *dev,
-                                          struct nlmsghdr *nlh, u16 flags)
+int switchdev_port_bridge_dellink(struct net_device *dev,
+                                 struct nlmsghdr *nlh, u16 flags)
 {
-       struct net_device *lower_dev;
-       struct list_head *iter;
-       int ret = 0, err = 0;
+       struct nlattr *afspec;
 
-       if (!(dev->features & NETIF_F_HW_SWITCH_OFFLOAD))
-               return ret;
-
-       netdev_for_each_lower_dev(dev, lower_dev, iter) {
-               err = switchdev_port_bridge_setlink(lower_dev, nlh, flags);
-               if (err && err != -EOPNOTSUPP)
-                       ret = err;
-       }
+       afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
+                                IFLA_AF_SPEC);
+       if (afspec)
+               return switchdev_port_br_afspec(dev, afspec,
+                                               switchdev_port_obj_del);
 
-       return ret;
+       return 0;
 }
-EXPORT_SYMBOL_GPL(ndo_dflt_switchdev_port_bridge_setlink);
+EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
 
 /**
  *     ndo_dflt_switchdev_port_bridge_dellink - default ndo bridge dellink
@@ -239,17 +568,16 @@ EXPORT_SYMBOL_GPL(ndo_dflt_switchdev_port_bridge_dellink);
 
 static struct net_device *switchdev_get_lowest_dev(struct net_device *dev)
 {
-       const struct swdev_ops *ops = dev->swdev_ops;
+       const struct switchdev_ops *ops = dev->switchdev_ops;
        struct net_device *lower_dev;
        struct net_device *port_dev;
        struct list_head *iter;
 
        /* Recusively search down until we find a sw port dev.
-        * (A sw port dev supports swdev_parent_id_get).
+        * (A sw port dev supports switchdev_port_attr_get).
         */
 
-       if (dev->features & NETIF_F_HW_SWITCH_OFFLOAD &&
-           ops && ops->swdev_parent_id_get)
+       if (ops && ops->switchdev_port_attr_get)
                return dev;
 
        netdev_for_each_lower_dev(dev, lower_dev, iter) {
@@ -263,8 +591,10 @@ static struct net_device *switchdev_get_lowest_dev(struct net_device *dev)
 
 static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
 {
-       struct netdev_phys_item_id psid;
-       struct netdev_phys_item_id prev_psid;
+       struct switchdev_attr attr = {
+               .id = SWITCHDEV_ATTR_PORT_PARENT_ID,
+       };
+       struct switchdev_attr prev_attr;
        struct net_device *dev = NULL;
        int nhsel;
 
@@ -280,17 +610,18 @@ static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
                if (!dev)
                        return NULL;
 
-               if (switchdev_parent_id_get(dev, &psid))
+               if (switchdev_port_attr_get(dev, &attr))
                        return NULL;
 
                if (nhsel > 0) {
-                       if (prev_psid.id_len != psid.id_len)
+                       if (prev_attr.ppid.id_len != attr.ppid.id_len)
                                return NULL;
-                       if (memcmp(prev_psid.id, psid.id, psid.id_len))
+                       if (memcmp(prev_attr.ppid.id, attr.ppid.id,
+                                  attr.ppid.id_len))
                                return NULL;
                }
 
-               prev_psid = psid;
+               prev_attr = attr;
        }
 
        return dev;
@@ -313,7 +644,7 @@ int switchdev_fib_ipv4_add(u32 dst, int dst_len, struct fib_info *fi,
                           u8 tos, u8 type, u32 nlflags, u32 tb_id)
 {
        struct net_device *dev;
-       const struct swdev_ops *ops;
+       const struct switchdev_ops *ops;
        int err = 0;
 
        /* Don't offload route if using custom ip rules or if
@@ -331,12 +662,12 @@ int switchdev_fib_ipv4_add(u32 dst, int dst_len, struct fib_info *fi,
        dev = switchdev_get_dev_by_nhs(fi);
        if (!dev)
                return 0;
-       ops = dev->swdev_ops;
+       ops = dev->switchdev_ops;
 
-       if (ops->swdev_fib_ipv4_add) {
-               err = ops->swdev_fib_ipv4_add(dev, htonl(dst), dst_len,
-                                             fi, tos, type, nlflags,
-                                             tb_id);
+       if (ops->switchdev_fib_ipv4_add) {
+               err = ops->switchdev_fib_ipv4_add(dev, htonl(dst), dst_len,
+                                                 fi, tos, type, nlflags,
+                                                 tb_id);
                if (!err)
                        fi->fib_flags |= RTNH_F_EXTERNAL;
        }
@@ -361,7 +692,7 @@ int switchdev_fib_ipv4_del(u32 dst, int dst_len, struct fib_info *fi,
                           u8 tos, u8 type, u32 tb_id)
 {
        struct net_device *dev;
-       const struct swdev_ops *ops;
+       const struct switchdev_ops *ops;
        int err = 0;
 
        if (!(fi->fib_flags & RTNH_F_EXTERNAL))
@@ -370,11 +701,11 @@ int switchdev_fib_ipv4_del(u32 dst, int dst_len, struct fib_info *fi,
        dev = switchdev_get_dev_by_nhs(fi);
        if (!dev)
                return 0;
-       ops = dev->swdev_ops;
+       ops = dev->switchdev_ops;
 
-       if (ops->swdev_fib_ipv4_del) {
-               err = ops->swdev_fib_ipv4_del(dev, htonl(dst), dst_len,
-                                             fi, tos, type, tb_id);
+       if (ops->switchdev_fib_ipv4_del) {
+               err = ops->switchdev_fib_ipv4_del(dev, htonl(dst), dst_len,
+                                                 fi, tos, type, tb_id);
                if (!err)
                        fi->fib_flags &= ~RTNH_F_EXTERNAL;
        }
This page took 0.032389 seconds and 5 git commands to generate.