Skip to contentSkip to conversation

Open vSwitch zerocopy upcall

PATCH Superseded Thomas Graf 1 participant 4 messages Nov 11, 2013

Subsystems (4)

v5 of 9 Superseded Latest is v9 →

lore
Share & tools
lore
#0 lore raw

v5 cover Open vSwitch zerocopy upcall

Respin of the zerocopy patches for the openvswitch upcall.

V5: - Removed padding requirement in user space
- Added OVS_DP_F_UNALIGNED flag let user space signal ability to receive unaligned Netlink messages, fall back to linear copy otherwise.
V4: - Daniel Borkmann pointed out that the style in skbuff.h has changed
in net-next, adapted to no longer using extern in headers.
V3: - Removed unneeded alignment of nlmsg_len after padding
V2: - Added skb_zerocopy_headlen() to calculate headroom of destination
buffer. This also takes care of the from->head_frag issue.
- Attribute alignment for frag_list case
- API documentation
- performance data for CHECKSUM_PARTIAL tx case

Thomas Graf (3):
net: Export skb_zerocopy() to zerocopy from one skb to another
openvswitch: Allow user space to announce ability to accept unaligned
Netlink messages
openvswitch: Use skb_zerocopy() for upcall

 include/linux/skbuff.h               |  3 ++
 include/uapi/linux/openvswitch.h     |  4 ++
 net/core/skbuff.c                    | 85 ++++++++++++++++++++++++++++++++++++
 net/netfilter/nfnetlink_queue_core.c | 59 ++-----------------------
 net/openvswitch/datapath.c           | 58 ++++++++++++++++--------
 net/openvswitch/datapath.h           |  2 +
 6 files changed, 139 insertions(+), 72 deletions(-)
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH v5 1/3 net: Export skb_zerocopy() to zerocopy from one skb to another

Make the skb zerocopy logic written for nfnetlink queue available for use by other modules.

---
 include/linux/skbuff.h               |  3 ++
 net/core/skbuff.c                    | 85 ++++++++++++++++++++++++++++++++++++
 net/netfilter/nfnetlink_queue_core.c | 59 ++-----------------------
 3 files changed, 92 insertions(+), 55 deletions(-)
include/linux/skbuff.h +3 -0
index 036ec7d..9e6a293 100644
@@ -2372,6 +2372,9 @@ int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
 		    struct pipe_inode_info *pipe, unsigned int len,
 		    unsigned int flags);
 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
+unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
+void skb_zerocopy(struct sk_buff *to, const struct sk_buff *from,
+		  int len, int hlen);
 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
net/core/skbuff.c +85 -0

91 lines not on this page · raw message

net/netfilter/nfnetlink_queue_core.c +4 -55

83 lines not on this page · raw message

Signature · 1 line
1.8.3.1
#2 lore raw

PATCH v5 2/3 openvswitch: Allow user space to announce ability to accept unaligned Netlink messages

In reply to Thomas Graf
---
 include/uapi/linux/openvswitch.h | 4 ++++
 net/openvswitch/datapath.c       | 4 ++++
 net/openvswitch/datapath.h       | 2 ++
 3 files changed, 10 insertions(+)
include/uapi/linux/openvswitch.h +4 -0
index d120f9f..07ef2c3 100644
@@ -75,6 +75,7 @@ enum ovs_datapath_attr {
 	OVS_DP_ATTR_UPCALL_PID,		/* Netlink PID to receive upcalls */
 	OVS_DP_ATTR_STATS,		/* struct ovs_dp_stats */
 	OVS_DP_ATTR_MEGAFLOW_STATS,	/* struct ovs_dp_megaflow_stats */
+	OVS_DP_ATTR_USER_FEATURES,	/* OVS_DP_F_*  */
 	__OVS_DP_ATTR_MAX
 };
 
@@ -106,6 +107,9 @@ struct ovs_vport_stats {
 	__u64   tx_dropped;		/* no space available in linux  */
 };
 
+/* Allow last Netlink attribute to be unaligned */
+#define OVS_DP_F_UNALIGNED	(1 << 0)
+
 /* Fixed logical ports. */
 #define OVSP_LOCAL      ((__u32)0)
 
net/openvswitch/datapath.c +4 -0
index 1408adc..0a50574 100644
@@ -1061,6 +1061,7 @@ static struct genl_ops dp_flow_genl_ops[] = {
 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
+	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
 };
 
 static struct genl_family dp_datapath_genl_family = {
@@ -1217,6 +1218,9 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
 	parms.port_no = OVSP_LOCAL;
 	parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
 
+	if (a[OVS_DP_ATTR_USER_FEATURES])
+		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
+
 	vport = new_vport(&parms);
 	if (IS_ERR(vport)) {
 		err = PTR_ERR(vport);
net/openvswitch/datapath.h +2 -0
index d3d14a58..ecdf44c 100644
@@ -88,6 +88,8 @@ struct datapath {
 	/* Network namespace ref. */
 	struct net *net;
 #endif
+
+	u32 user_features;
 };
 
 /**
Signature · 1 line
1.8.3.1
#3 lore raw

PATCH v5 3/3 openvswitch: Use skb_zerocopy() for upcall

In reply to Thomas Graf

Use of skb_zerocopy() avoids the expensive call to memcpy() when copying the packet data into the Netlink skb. Completes checksum through skb_checksum_help() if needed.

Cost of memcpy is significantly reduced from:

+   7.48%       vhost-8471  [k] memcpy
+   5.57%     ovs-vswitchd  [k] memcpy
+   2.81%       vhost-8471  [k] csum_partial_copy_generic

to:

+   5.72%     ovs-vswitchd  [k] memcpy
+   3.32%       vhost-5153  [k] memcpy
+   0.68%       vhost-5153  [k] skb_zerocopy

(megaflows disabled)

---
 net/openvswitch/datapath.c | 54 +++++++++++++++++++++++++++++++---------------
 1 file changed, 37 insertions(+), 17 deletions(-)
net/openvswitch/datapath.c +37 -17
index 0a50574..8dc2496 100644
@@ -108,10 +108,10 @@ int lockdep_ovsl_is_held(void)
 #endif
 
 static struct vport *new_vport(const struct vport_parms *);
-static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
-			     const struct dp_upcall_info *);
-static int queue_userspace_packet(struct net *, int dp_ifindex,
-				  struct sk_buff *,
+static int queue_gso_packets(struct datapath *, struct net *, int dp_ifindex,
+			     struct sk_buff *, const struct dp_upcall_info *);
+static int queue_userspace_packet(struct datapath *, struct net *,
+				  int dp_ifindex, struct sk_buff *,
 				  const struct dp_upcall_info *);
 
 /* Must be called with rcu_read_lock or ovs_mutex. */
@@ -292,9 +292,9 @@ int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 	}
 
 	if (!skb_is_gso(skb))
-		err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
+		err = queue_userspace_packet(dp, ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 	else
-		err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
+		err = queue_gso_packets(dp, ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 	if (err)
 		goto err;
 
@@ -310,7 +310,7 @@ err:
 	return err;
 }
 
-static int queue_gso_packets(struct net *net, int dp_ifindex,
+static int queue_gso_packets(struct datapath *dp, struct net *net, int dp_ifindex,
 			     struct sk_buff *skb,
 			     const struct dp_upcall_info *upcall_info)
 {
@@ -327,7 +327,7 @@ static int queue_gso_packets(struct net *net, int dp_ifindex,
 	/* Queue all of the segments. */
 	skb = segs;
 	do {
-		err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
+		err = queue_userspace_packet(dp, net, dp_ifindex, skb, upcall_info);
 		if (err)
 			break;
 
@@ -381,10 +381,11 @@ static size_t key_attr_size(void)
 }
 
 static size_t upcall_msg_size(const struct sk_buff *skb,
-			      const struct nlattr *userdata)
+			      const struct nlattr *userdata,
+			      unsigned int hdrlen)
 {
 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
-		+ nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
+		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 		+ nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 
 	/* OVS_PACKET_ATTR_USERDATA */
@@ -394,14 +395,15 @@ static size_t upcall_msg_size(const struct sk_buff *skb,
 	return size;
 }
 
-static int queue_userspace_packet(struct net *net, int dp_ifindex,
-				  struct sk_buff *skb,
+static int queue_userspace_packet(struct datapath *dp, struct net *net,
+				  int dp_ifindex, struct sk_buff *skb,
 				  const struct dp_upcall_info *upcall_info)
 {
 	struct ovs_header *upcall;
 	struct sk_buff *nskb = NULL;
 	struct sk_buff *user_skb; /* to be queued to userspace */
 	struct nlattr *nla;
+	unsigned int hlen;
 	int err;
 
 	if (vlan_tx_tag_present(skb)) {
@@ -422,7 +424,21 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,
 		goto out;
 	}
 
-	user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
+	/* Complete checksum if needed */
+	if (skb->ip_summed == CHECKSUM_PARTIAL &&
+	    (err = skb_checksum_help(skb)))
+		goto out;
+
+	/* Older versions of OVS user space enforce alignment of the last
+	 * Netlink attribute to NLA_ALIGNTO which would require extensive
+	 * padding logic. Only perform zerocopy if padding is not required.
+	 */
+	if (dp->user_features & OVS_DP_F_UNALIGNED)
+		hlen = skb_zerocopy_headlen(skb);
+	else
+		hlen = skb->len;
+
+	user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata, hlen), GFP_ATOMIC);
 	if (!user_skb) {
 		err = -ENOMEM;
 		goto out;
@@ -441,13 +457,17 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,
 			  nla_len(upcall_info->userdata),
 			  nla_data(upcall_info->userdata));
 
-	nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
+	/* Only reserve room for attribute header, packet data is added
+	 * in skb_zerocopy() */
+	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0)))
+		goto out;
+	nla->nla_len = nla_attr_size(skb->len);
+
+	skb_zerocopy(user_skb, skb, skb->len, hlen);
 
-	skb_copy_and_csum_dev(skb, nla_data(nla));
+	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 
-	genlmsg_end(user_skb, upcall);
 	err = genlmsg_unicast(net, user_skb, upcall_info->portid);
-
 out:
 	kfree_skb(nskb);
 	return err;
Signature · 1 line
1.8.3.1