[PATCH net-next v3 0/3] New openvswitch MPLS actions for layer 2 tunnelling

[PATCH net-next v3 0/3] New openvswitch MPLS actions for layer 2 tunnelling

[Martin Varghese]

From: Martin Varghese <martin.varghese@nokia.com>

The existing PUSH MPLS action inserts MPLS header between ethernet header
and the IP header. Though this behaviour is fine for L3 VPN where an IP
packet is encapsulated inside a MPLS tunnel, it does not suffice the L2
VPN (l2 tunnelling) requirements. In L2 VPN the MPLS header should
encapsulate the ethernet packet.

The new mpls action PTAP_PUSH_MPLS inserts MPLS header at the start of the
packet or at the start of the l3 header depending on the value of l2 tunnel
flag in the PTAP_PUSH_MPLS arguments.

POP_MPLS action is extended to support ethertype 0x6558

OVS userspace changes -
---------------------
Encap & Decap ovs actions are extended to support MPLS packet type. The encap & decap
adds and removes MPLS header at the start of packet as depicted below.

Actions - encap(mpls(ether_type=0x8847)),encap(ethernet)

Incoming packet -> | ETH | IP | Payload |

1 Actions -  encap(mpls(ether_type=0x8847)) [Kernel action - ptap_push_mpls:0x8847]

        Outgoing packet -> | MPLS | ETH | Payload|

2 Actions - encap(ethernet) [ Kernel action - push_eth ]

        Outgoing packet -> | ETH | MPLS | ETH | Payload|

Decapsulation:

Incoming packet -> | ETH | MPLS | ETH | IP | Payload |

Actions - decap(),decap(packet_type(ns=0,type=0)

1 Actions -  decap() [Kernel action - pop_eth)

        Outgoing packet -> | MPLS | ETH | IP | Payload|

2 Actions - decap(packet_type(ns=0,type=0) [Kernel action - pop_mpls:0]

        Outgoing packet -> | ETH  | IP | Payload

Martin Varghese (3):
  net: skb_mpls_push() modified to allow MPLS header push at start of
    packet.
  net: Rephrased comments section of skb_mpls_pop()
  openvswitch: New MPLS actions for layer 2 tunnelling

 include/uapi/linux/openvswitch.h | 23 ++++++++++++++++++++++-
 net/core/skbuff.c                | 10 +++++++---
 net/openvswitch/actions.c        | 30 ++++++++++++++++++++++++------
 net/openvswitch/flow_netlink.c   | 34 ++++++++++++++++++++++++++++++++++
 4 files changed, 87 insertions(+), 10 deletions(-)

-- 
1.8.3.1

[PATCH net-next v3 1/3] net: skb_mpls_push() modified to allow MPLS header push at start of packet.

[Martin Varghese]

From: Martin Varghese <martin.varghese@nokia.com>

The existing skb_mpls_push() implementation always inserts mpls header
after the mac header. L2 VPN use cases requires MPLS header to be
inserted before the ethernet header as the ethernet packet gets tunnelled
inside MPLS header in those cases.

Signed-off-by: Martin Varghese <martin.varghese@nokia.com>
---
Changes in v2:
    - Fixed comments section of skb_mpls_push().
    - Added skb_reset_mac_len() in skb_mpls_push(). The mac len changes
      when MPLS header in inserted at the start of the packet.

 net/core/skbuff.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 973a71f..d90c827 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -5472,12 +5472,15 @@ static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
 }
 
 /**
- * skb_mpls_push() - push a new MPLS header after the mac header
+ * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
+ *                   the packet
  *
  * @skb: buffer
  * @mpls_lse: MPLS label stack entry to push
  * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
  * @mac_len: length of the MAC header
+ * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
+ *            ethernet
  *
  * Expects skb->data at mac header.
  *
@@ -5501,7 +5504,7 @@ int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
 		return err;
 
 	if (!skb->inner_protocol) {
-		skb_set_inner_network_header(skb, mac_len);
+		skb_set_inner_network_header(skb, skb_network_offset(skb));
 		skb_set_inner_protocol(skb, skb->protocol);
 	}
 
@@ -5510,6 +5513,7 @@ int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
 		mac_len);
 	skb_reset_mac_header(skb);
 	skb_set_network_header(skb, mac_len);
+	skb_reset_mac_len(skb);
 
 	lse = mpls_hdr(skb);
 	lse->label_stack_entry = mpls_lse;
-- 
1.8.3.1

[PATCH net-next v3 2/3] net: Rephrased comments section of skb_mpls_pop()

[Martin Varghese]

From: Martin Varghese <martin.varghese@nokia.com>

Rephrased comments section of skb_mpls_pop() to align it with
comments section of skb_mpls_push().

Signed-off-by: Martin Varghese <martin.varghese@nokia.com>
---
 net/core/skbuff.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index d90c827..44b0894 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -5533,7 +5533,7 @@ int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
  * @skb: buffer
  * @next_proto: ethertype of header after popped MPLS header
  * @mac_len: length of the MAC header
- * @ethernet: flag to indicate if ethernet header is present in packet
+ * @ethernet: flag to indicate if the packet is ethernet
  *
  * Expects skb->data at mac header.
  *
-- 
1.8.3.1

[PATCH net-next v3 3/3] openvswitch: New MPLS actions for layer 2 tunnelling

[Martin Varghese]

From: Martin Varghese <martin.varghese@nokia.com>

The existing PUSH MPLS action inserts MPLS header between ethernet header
and the IP header. Though this behaviour is fine for L3 VPN where an IP
packet is encapsulated inside a MPLS tunnel, it does not suffice the L2
VPN (l2 tunnelling) requirements. In L2 VPN the MPLS header should
encapsulate the ethernet packet.

The new mpls action PTAP_PUSH_MPLS inserts MPLS header at the start of the
packet or at the start of the l3 header depending on the value of l2 tunnel
flag in the PTAP_PUSH_MPLS arguments.

POP_MPLS action is extended to support ethertype 0x6558.

Signed-off-by: Martin Varghese <martin.varghese@nokia.com>
---
Changes in v2:
   - PTAP_POP_MPLS action removed.
   - Special handling for ethertype 0 added in PUSH_MPLS.
   - Refactored push_mpls function to cater existing push_mpls and
     ptap_push_mpls actions.
   - mac len to specify the MPLS header location added in PTAP_PUSH_MPLS
     arguments.
 
Changes in v3:
   - Special handling for ethertype 0 removed.
   - Added support for ether type 0x6558.
   - Removed mac len from PTAP_PUSH_MPLS argument list
   - used l2_tun flag to distinguish l2 and l3 tunnelling.
   - Extended PTAP_PUSH_MPLS handling to cater PUSH_MPLS action also.
   

 include/uapi/linux/openvswitch.h | 23 ++++++++++++++++++++++-
 net/openvswitch/actions.c        | 30 ++++++++++++++++++++++++------
 net/openvswitch/flow_netlink.c   | 34 ++++++++++++++++++++++++++++++++++
 3 files changed, 80 insertions(+), 7 deletions(-)

diff --git a/include/uapi/linux/openvswitch.h b/include/uapi/linux/openvswitch.h
index a87b44c..b7221ad 100644
--- a/include/uapi/linux/openvswitch.h
+++ b/include/uapi/linux/openvswitch.h
@@ -673,6 +673,25 @@ struct ovs_action_push_mpls {
 };
 
 /**
+ * struct ovs_action_ptap_push_mpls - %OVS_ACTION_ATTR_PTAP_PUSH_MPLS action
+ * argument.
+ * @mpls_lse: MPLS label stack entry to push.
+ * @mpls_ethertype: Ethertype to set in the encapsulating ethernet frame.
+ * @l2_tun: Flag to specify the place of insertion of MPLS header.
+ * When true, the MPLS header will be inserted at the start of the packet.
+ * When false, the MPLS header will be inserted at the start of the l3 header.
+ *
+ * The only values @mpls_ethertype should ever be given are %ETH_P_MPLS_UC and
+ * %ETH_P_MPLS_MC, indicating MPLS unicast or multicast. Other are rejected.
+ */
+struct ovs_action_ptap_push_mpls {
+	__be32 mpls_lse;
+	__be16 mpls_ethertype; /* Either %ETH_P_MPLS_UC or %ETH_P_MPLS_MC */
+	bool l2_tun;
+};
+
+
+/**
  * struct ovs_action_push_vlan - %OVS_ACTION_ATTR_PUSH_VLAN action argument.
  * @vlan_tpid: Tag protocol identifier (TPID) to push.
  * @vlan_tci: Tag control identifier (TCI) to push.  The CFI bit must be set
@@ -892,7 +911,8 @@ struct check_pkt_len_arg {
  * @OVS_ACTION_ATTR_CHECK_PKT_LEN: Check the packet length and execute a set
  * of actions if greater than the specified packet length, else execute
  * another set of actions.
- *
+ * @OVS_ACTION_ATTR_PUSH_MPLS: Push a new MPLS label stack entry at specified
+ * offset from start of packet.
  * Only a single header can be set with a single %OVS_ACTION_ATTR_SET.  Not all
  * fields within a header are modifiable, e.g. the IPv4 protocol and fragment
  * type may not be changed.
@@ -927,6 +947,7 @@ enum ovs_action_attr {
 	OVS_ACTION_ATTR_METER,        /* u32 meter ID. */
 	OVS_ACTION_ATTR_CLONE,        /* Nested OVS_CLONE_ATTR_*.  */
 	OVS_ACTION_ATTR_CHECK_PKT_LEN, /* Nested OVS_CHECK_PKT_LEN_ATTR_*. */
+	OVS_ACTION_ATTR_PTAP_PUSH_MPLS, /* struct ovs_action_ptap_push_mpls. */
 
 	__OVS_ACTION_ATTR_MAX,	      /* Nothing past this will be accepted
 				       * from userspace. */
diff --git a/net/openvswitch/actions.c b/net/openvswitch/actions.c
index 4c83954..5138473 100644
--- a/net/openvswitch/actions.c
+++ b/net/openvswitch/actions.c
@@ -161,16 +161,17 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 			      const struct nlattr *attr, int len);
 
 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
-		     const struct ovs_action_push_mpls *mpls)
+		     __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
 {
 	int err;
 
-	err = skb_mpls_push(skb, mpls->mpls_lse, mpls->mpls_ethertype,
-			    skb->mac_len,
-			    ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
+	err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
 	if (err)
 		return err;
 
+	if (!mac_len)
+		key->mac_proto = MAC_PROTO_NONE;
+
 	invalidate_flow_key(key);
 	return 0;
 }
@@ -185,6 +186,9 @@ static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 	if (err)
 		return err;
 
+	if (ethertype == htons(ETH_P_TEB))
+		key->mac_proto = MAC_PROTO_ETHERNET;
+
 	invalidate_flow_key(key);
 	return 0;
 }
@@ -1229,10 +1233,24 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 			execute_hash(skb, key, a);
 			break;
 
-		case OVS_ACTION_ATTR_PUSH_MPLS:
-			err = push_mpls(skb, key, nla_data(a));
+		case OVS_ACTION_ATTR_PUSH_MPLS: {
+			struct ovs_action_push_mpls *mpls = nla_data(a);
+
+			err = push_mpls(skb, key, mpls->mpls_lse,
+					mpls->mpls_ethertype, skb->mac_len);
 			break;
+		}
+		case OVS_ACTION_ATTR_PTAP_PUSH_MPLS: {
+			struct ovs_action_ptap_push_mpls *mpls = nla_data(a);
+			__u16 mac_len = 0;
+
+			if (!mpls->l2_tun)
+				mac_len = skb->mac_len;
 
+			err = push_mpls(skb, key, mpls->mpls_lse,
+					mpls->mpls_ethertype, mac_len);
+			break;
+		}
 		case OVS_ACTION_ATTR_POP_MPLS:
 			err = pop_mpls(skb, key, nla_get_be16(a));
 			break;
diff --git a/net/openvswitch/flow_netlink.c b/net/openvswitch/flow_netlink.c
index 65c2e34..8d650e3 100644
--- a/net/openvswitch/flow_netlink.c
+++ b/net/openvswitch/flow_netlink.c
@@ -79,6 +79,7 @@ static bool actions_may_change_flow(const struct nlattr *actions)
 		case OVS_ACTION_ATTR_SET_MASKED:
 		case OVS_ACTION_ATTR_METER:
 		case OVS_ACTION_ATTR_CHECK_PKT_LEN:
+		case OVS_ACTION_ATTR_PTAP_PUSH_MPLS:
 		default:
 			return true;
 		}
@@ -3005,6 +3006,7 @@ static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
 			[OVS_ACTION_ATTR_METER] = sizeof(u32),
 			[OVS_ACTION_ATTR_CLONE] = (u32)-1,
 			[OVS_ACTION_ATTR_CHECK_PKT_LEN] = (u32)-1,
+			[OVS_ACTION_ATTR_PTAP_PUSH_MPLS] = sizeof(struct ovs_action_ptap_push_mpls),
 		};
 		const struct ovs_action_push_vlan *vlan;
 		int type = nla_type(a);
@@ -3072,6 +3074,33 @@ static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
 		case OVS_ACTION_ATTR_RECIRC:
 			break;
 
+		case OVS_ACTION_ATTR_PTAP_PUSH_MPLS: {
+			const struct ovs_action_ptap_push_mpls *mpls = nla_data(a);
+
+			if (!eth_p_mpls(mpls->mpls_ethertype))
+				return -EINVAL;
+
+			if (!mpls->l2_tun) {
+				if (vlan_tci & htons(VLAN_CFI_MASK) ||
+				    (eth_type != htons(ETH_P_IP) &&
+				     eth_type != htons(ETH_P_IPV6) &&
+				     eth_type != htons(ETH_P_ARP) &&
+				     eth_type != htons(ETH_P_RARP) &&
+				     !eth_p_mpls(eth_type)))
+					return -EINVAL;
+				mpls_label_count++;
+			} else {
+				if (mac_proto != MAC_PROTO_NONE) {
+					mpls_label_count = 1;
+					mac_proto = MAC_PROTO_NONE;
+				} else {
+					mpls_label_count++;
+				}
+			}
+			eth_type = mpls->mpls_ethertype;
+			break;
+		}
+
 		case OVS_ACTION_ATTR_PUSH_MPLS: {
 			const struct ovs_action_push_mpls *mpls = nla_data(a);
 
@@ -3109,6 +3138,11 @@ static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
 			 * recirculation.
 			 */
 			proto = nla_get_be16(a);
+
+			if (proto == htons(ETH_P_TEB) &&
+			    mac_proto != MAC_PROTO_NONE)
+				return -EINVAL;
+
 			mpls_label_count--;
 
 			if (!eth_p_mpls(proto) || !mpls_label_count)
-- 
1.8.3.1

Re: [PATCH net-next v3 3/3] openvswitch: New MPLS actions for layer 2 tunnelling

[Jakub Kicinski]

On Mon, 16 Dec 2019 19:33:43 +0530, Martin Varghese wrote:
> diff --git a/include/uapi/linux/openvswitch.h b/include/uapi/linux/openvswitch.h
> index a87b44c..b7221ad 100644
> --- a/include/uapi/linux/openvswitch.h
> +++ b/include/uapi/linux/openvswitch.h
> @@ -673,6 +673,25 @@ struct ovs_action_push_mpls {
>  };
>  
>  /**
> + * struct ovs_action_ptap_push_mpls - %OVS_ACTION_ATTR_PTAP_PUSH_MPLS action
> + * argument.
> + * @mpls_lse: MPLS label stack entry to push.
> + * @mpls_ethertype: Ethertype to set in the encapsulating ethernet frame.
> + * @l2_tun: Flag to specify the place of insertion of MPLS header.
> + * When true, the MPLS header will be inserted at the start of the packet.
> + * When false, the MPLS header will be inserted at the start of the l3 header.
> + *
> + * The only values @mpls_ethertype should ever be given are %ETH_P_MPLS_UC and
> + * %ETH_P_MPLS_MC, indicating MPLS unicast or multicast. Other are rejected.
> + */
> +struct ovs_action_ptap_push_mpls {
> +	__be32 mpls_lse;
> +	__be16 mpls_ethertype; /* Either %ETH_P_MPLS_UC or %ETH_P_MPLS_MC */
> +	bool l2_tun;

In file included from <command-line>:32:                                        
./usr/include/linux/openvswitch.h:674:2: error: unknown type name ‘bool’        
  674 |  bool l2_tun;                                                           
      |  ^~~~                                                                   
make[3]: *** [usr/include/linux/openvswitch.hdrtest] Error 1  

> +};
> +
> +

Why the double new line? Please use checkpatch --strict.

Re: [PATCH net-next v3 3/3] openvswitch: New MPLS actions for layer 2 tunnelling

[Martin Varghese]

On Mon, Dec 16, 2019 at 04:13:55PM -0800, Jakub Kicinski wrote:
> On Mon, 16 Dec 2019 19:33:43 +0530, Martin Varghese wrote:
> > diff --git a/include/uapi/linux/openvswitch.h b/include/uapi/linux/openvswitch.h
> > index a87b44c..b7221ad 100644
> > --- a/include/uapi/linux/openvswitch.h
> > +++ b/include/uapi/linux/openvswitch.h
> > @@ -673,6 +673,25 @@ struct ovs_action_push_mpls {
> >  };
> >  
> >  /**
> > + * struct ovs_action_ptap_push_mpls - %OVS_ACTION_ATTR_PTAP_PUSH_MPLS action
> > + * argument.
> > + * @mpls_lse: MPLS label stack entry to push.
> > + * @mpls_ethertype: Ethertype to set in the encapsulating ethernet frame.
> > + * @l2_tun: Flag to specify the place of insertion of MPLS header.
> > + * When true, the MPLS header will be inserted at the start of the packet.
> > + * When false, the MPLS header will be inserted at the start of the l3 header.
> > + *
> > + * The only values @mpls_ethertype should ever be given are %ETH_P_MPLS_UC and
> > + * %ETH_P_MPLS_MC, indicating MPLS unicast or multicast. Other are rejected.
> > + */
> > +struct ovs_action_ptap_push_mpls {
> > +	__be32 mpls_lse;
> > +	__be16 mpls_ethertype; /* Either %ETH_P_MPLS_UC or %ETH_P_MPLS_MC */
> > +	bool l2_tun;
> 
> In file included from <command-line>:32:                                        
> ./usr/include/linux/openvswitch.h:674:2: error: unknown type name ‘bool’        
>   674 |  bool l2_tun;                                                           
>       |  ^~~~                                                                   
> make[3]: *** [usr/include/linux/openvswitch.hdrtest] Error 1  
> 

Does that mean bool cannot be used in interface header files ? but what is
the alternative u8?
> > +};
> > +
> > +
> 
> Why the double new line? Please use checkpatch --strict.

Noted

Thanks for your time.

Re: [PATCH net-next v3 3/3] openvswitch: New MPLS actions for layer 2 tunnelling

[Jakub Kicinski]

On Tue, 17 Dec 2019 22:45:39 +0530, Martin Varghese wrote:
> On Mon, Dec 16, 2019 at 04:13:55PM -0800, Jakub Kicinski wrote:
> > On Mon, 16 Dec 2019 19:33:43 +0530, Martin Varghese wrote:  
> > > diff --git a/include/uapi/linux/openvswitch.h b/include/uapi/linux/openvswitch.h
> > > index a87b44c..b7221ad 100644
> > > --- a/include/uapi/linux/openvswitch.h
> > > +++ b/include/uapi/linux/openvswitch.h
> > > @@ -673,6 +673,25 @@ struct ovs_action_push_mpls {
> > >  };
> > >  
> > >  /**
> > > + * struct ovs_action_ptap_push_mpls - %OVS_ACTION_ATTR_PTAP_PUSH_MPLS action
> > > + * argument.
> > > + * @mpls_lse: MPLS label stack entry to push.
> > > + * @mpls_ethertype: Ethertype to set in the encapsulating ethernet frame.
> > > + * @l2_tun: Flag to specify the place of insertion of MPLS header.
> > > + * When true, the MPLS header will be inserted at the start of the packet.
> > > + * When false, the MPLS header will be inserted at the start of the l3 header.
> > > + *
> > > + * The only values @mpls_ethertype should ever be given are %ETH_P_MPLS_UC and
> > > + * %ETH_P_MPLS_MC, indicating MPLS unicast or multicast. Other are rejected.
> > > + */
> > > +struct ovs_action_ptap_push_mpls {
> > > +	__be32 mpls_lse;
> > > +	__be16 mpls_ethertype; /* Either %ETH_P_MPLS_UC or %ETH_P_MPLS_MC */
> > > +	bool l2_tun;  
> > 
> > In file included from <command-line>:32:                                        
> > ./usr/include/linux/openvswitch.h:674:2: error: unknown type name ‘bool’        
> >   674 |  bool l2_tun;                                                           
> >       |  ^~~~                                                                   
> > make[3]: *** [usr/include/linux/openvswitch.hdrtest] Error 1  
> >   
> 
> Does that mean bool cannot be used in interface header files ? but what is
> the alternative u8?

I think you have a 16 bit hole there anyway due to the alignment, so
I'd declare the space as __u16 (underscores needed since this is the
user space type), and then validate in the kernel that the higher bits
are unused i.e. return -EINVAL if the field is not 0 or 1, to allow for
a future use of those bits.