diff options
Diffstat (limited to 'net/bridge')
-rw-r--r-- | net/bridge/br_forward.c | 9 | ||||
-rw-r--r-- | net/bridge/br_private.h | 31 | ||||
-rw-r--r-- | net/bridge/br_switchdev.c | 68 | ||||
-rw-r--r-- | net/bridge/br_vlan.c | 10 |
4 files changed, 113 insertions, 5 deletions
diff --git a/net/bridge/br_forward.c b/net/bridge/br_forward.c index bfdbaf3015b9..bc14b1b384e9 100644 --- a/net/bridge/br_forward.c +++ b/net/bridge/br_forward.c @@ -48,6 +48,8 @@ int br_dev_queue_push_xmit(struct net *net, struct sock *sk, struct sk_buff *skb skb_set_network_header(skb, depth); } + skb->offload_fwd_mark = br_switchdev_frame_uses_tx_fwd_offload(skb); + dev_queue_xmit(skb); return 0; @@ -76,6 +78,11 @@ static void __br_forward(const struct net_bridge_port *to, struct net *net; int br_hook; + /* Mark the skb for forwarding offload early so that br_handle_vlan() + * can know whether to pop the VLAN header on egress or keep it. + */ + nbp_switchdev_frame_mark_tx_fwd_offload(to, skb); + vg = nbp_vlan_group_rcu(to); skb = br_handle_vlan(to->br, to, vg, skb); if (!skb) @@ -174,6 +181,8 @@ static struct net_bridge_port *maybe_deliver( if (!should_deliver(p, skb)) return prev; + nbp_switchdev_frame_mark_tx_fwd_to_hwdom(p, skb); + if (!prev) goto out; diff --git a/net/bridge/br_private.h b/net/bridge/br_private.h index 2f32d330b648..86ca617fec7a 100644 --- a/net/bridge/br_private.h +++ b/net/bridge/br_private.h @@ -552,12 +552,20 @@ struct br_input_skb_cb { #endif #ifdef CONFIG_NET_SWITCHDEV + /* Set if TX data plane offloading is used towards at least one + * hardware domain. + */ + u8 tx_fwd_offload:1; /* The switchdev hardware domain from which this packet was received. * If skb->offload_fwd_mark was set, then this packet was already * forwarded by hardware to the other ports in the source hardware * domain, otherwise it wasn't. */ int src_hwdom; + /* Bit mask of hardware domains towards this packet has already been + * transmitted using the TX data plane offload. + */ + unsigned long fwd_hwdoms; #endif }; @@ -1871,6 +1879,12 @@ static inline void br_sysfs_delbr(struct net_device *dev) { return; } /* br_switchdev.c */ #ifdef CONFIG_NET_SWITCHDEV +bool br_switchdev_frame_uses_tx_fwd_offload(struct sk_buff *skb); + +void nbp_switchdev_frame_mark_tx_fwd_offload(const struct net_bridge_port *p, + struct sk_buff *skb); +void nbp_switchdev_frame_mark_tx_fwd_to_hwdom(const struct net_bridge_port *p, + struct sk_buff *skb); void nbp_switchdev_frame_mark(const struct net_bridge_port *p, struct sk_buff *skb); bool nbp_switchdev_allowed_egress(const struct net_bridge_port *p, @@ -1891,6 +1905,23 @@ static inline void br_switchdev_frame_unmark(struct sk_buff *skb) skb->offload_fwd_mark = 0; } #else +static inline bool br_switchdev_frame_uses_tx_fwd_offload(struct sk_buff *skb) +{ + return false; +} + +static inline void +nbp_switchdev_frame_mark_tx_fwd_offload(const struct net_bridge_port *p, + struct sk_buff *skb) +{ +} + +static inline void +nbp_switchdev_frame_mark_tx_fwd_to_hwdom(const struct net_bridge_port *p, + struct sk_buff *skb) +{ +} + static inline void nbp_switchdev_frame_mark(const struct net_bridge_port *p, struct sk_buff *skb) { diff --git a/net/bridge/br_switchdev.c b/net/bridge/br_switchdev.c index 6bfff28ede23..96ce069d0c8c 100644 --- a/net/bridge/br_switchdev.c +++ b/net/bridge/br_switchdev.c @@ -8,6 +8,46 @@ #include "br_private.h" +static struct static_key_false br_switchdev_tx_fwd_offload; + +static bool nbp_switchdev_can_offload_tx_fwd(const struct net_bridge_port *p, + const struct sk_buff *skb) +{ + if (!static_branch_unlikely(&br_switchdev_tx_fwd_offload)) + return false; + + return (p->flags & BR_TX_FWD_OFFLOAD) && + (p->hwdom != BR_INPUT_SKB_CB(skb)->src_hwdom); +} + +bool br_switchdev_frame_uses_tx_fwd_offload(struct sk_buff *skb) +{ + if (!static_branch_unlikely(&br_switchdev_tx_fwd_offload)) + return false; + + return BR_INPUT_SKB_CB(skb)->tx_fwd_offload; +} + +/* Mark the frame for TX forwarding offload if this egress port supports it */ +void nbp_switchdev_frame_mark_tx_fwd_offload(const struct net_bridge_port *p, + struct sk_buff *skb) +{ + if (nbp_switchdev_can_offload_tx_fwd(p, skb)) + BR_INPUT_SKB_CB(skb)->tx_fwd_offload = true; +} + +/* Lazily adds the hwdom of the egress bridge port to the bit mask of hwdoms + * that the skb has been already forwarded to, to avoid further cloning to + * other ports in the same hwdom by making nbp_switchdev_allowed_egress() + * return false. + */ +void nbp_switchdev_frame_mark_tx_fwd_to_hwdom(const struct net_bridge_port *p, + struct sk_buff *skb) +{ + if (nbp_switchdev_can_offload_tx_fwd(p, skb)) + set_bit(p->hwdom, &BR_INPUT_SKB_CB(skb)->fwd_hwdoms); +} + void nbp_switchdev_frame_mark(const struct net_bridge_port *p, struct sk_buff *skb) { @@ -18,8 +58,10 @@ void nbp_switchdev_frame_mark(const struct net_bridge_port *p, bool nbp_switchdev_allowed_egress(const struct net_bridge_port *p, const struct sk_buff *skb) { - return !skb->offload_fwd_mark || - BR_INPUT_SKB_CB(skb)->src_hwdom != p->hwdom; + struct br_input_skb_cb *cb = BR_INPUT_SKB_CB(skb); + + return !test_bit(p->hwdom, &cb->fwd_hwdoms) && + (!skb->offload_fwd_mark || cb->src_hwdom != p->hwdom); } /* Flags that can be offloaded to hardware */ @@ -164,8 +206,11 @@ static void nbp_switchdev_hwdom_put(struct net_bridge_port *leaving) static int nbp_switchdev_add(struct net_bridge_port *p, struct netdev_phys_item_id ppid, + bool tx_fwd_offload, struct netlink_ext_ack *extack) { + int err; + if (p->offload_count) { /* Prevent unsupported configurations such as a bridge port * which is a bonding interface, and the member ports are from @@ -189,7 +234,16 @@ static int nbp_switchdev_add(struct net_bridge_port *p, p->ppid = ppid; p->offload_count = 1; - return nbp_switchdev_hwdom_set(p); + err = nbp_switchdev_hwdom_set(p); + if (err) + return err; + + if (tx_fwd_offload) { + p->flags |= BR_TX_FWD_OFFLOAD; + static_branch_inc(&br_switchdev_tx_fwd_offload); + } + + return 0; } static void nbp_switchdev_del(struct net_bridge_port *p) @@ -204,6 +258,11 @@ static void nbp_switchdev_del(struct net_bridge_port *p) if (p->hwdom) nbp_switchdev_hwdom_put(p); + + if (p->flags & BR_TX_FWD_OFFLOAD) { + p->flags &= ~BR_TX_FWD_OFFLOAD; + static_branch_dec(&br_switchdev_tx_fwd_offload); + } } static int nbp_switchdev_sync_objs(struct net_bridge_port *p, const void *ctx, @@ -262,6 +321,7 @@ int switchdev_bridge_port_offload(struct net_device *brport_dev, struct net_device *dev, const void *ctx, struct notifier_block *atomic_nb, struct notifier_block *blocking_nb, + bool tx_fwd_offload, struct netlink_ext_ack *extack) { struct netdev_phys_item_id ppid; @@ -278,7 +338,7 @@ int switchdev_bridge_port_offload(struct net_device *brport_dev, if (err) return err; - err = nbp_switchdev_add(p, ppid, extack); + err = nbp_switchdev_add(p, ppid, tx_fwd_offload, extack); if (err) return err; diff --git a/net/bridge/br_vlan.c b/net/bridge/br_vlan.c index 382ab992badf..325600361487 100644 --- a/net/bridge/br_vlan.c +++ b/net/bridge/br_vlan.c @@ -465,7 +465,15 @@ struct sk_buff *br_handle_vlan(struct net_bridge *br, u64_stats_update_end(&stats->syncp); } - if (v->flags & BRIDGE_VLAN_INFO_UNTAGGED) + /* If the skb will be sent using forwarding offload, the assumption is + * that the switchdev will inject the packet into hardware together + * with the bridge VLAN, so that it can be forwarded according to that + * VLAN. The switchdev should deal with popping the VLAN header in + * hardware on each egress port as appropriate. So only strip the VLAN + * header if forwarding offload is not being used. + */ + if (v->flags & BRIDGE_VLAN_INFO_UNTAGGED && + !br_switchdev_frame_uses_tx_fwd_offload(skb)) __vlan_hwaccel_clear_tag(skb); if (p && (p->flags & BR_VLAN_TUNNEL) && |