Merge branch 'ila-early-demux'

Tom Herbert says:

====================
ila: Optimization to preserve value of early demux

In the current implementation of ILA, LWT is used to perform
translation on both the input and output paths. This is functional,
however there is a big performance hit in the receive path. Early
demux occurs before the routing lookup (a hit actually obviates the
route lookup). Therefore the stack currently performs early
demux before translation so that a local connection with ILA
addresses is never matched. Note that this issue is not just
with ILA, but pretty much any translated or encapsulated packet
handled by LWT would miss the opportunity for early demux. Solving
the general problem seems non trivial since we would need to move
the route lookup before early demx thereby mitigating the value.

This patch set addresses the issue for ILA by adding a fast locator
lookup that occurs before early demux. This done by hooking in to
NF_INET_PRE_ROUTING

For the backend we implement an rhashtable that contains identifier
to locator to mappings. The table also allows more specific matches
that include original locator and interface.

This patch set:
 - Add an rhashtable function to atomically replace and element.
   This is useful to implement sub-trees from a table entry
   without needing to use a special anchor structure as the
   table entry.
 - Add a start callback for starting a netlink dump.
 - Creates an ila directory under net/ipv6 and moves ila.c to it.
   ila.c is split into ila_common.c and ila_lwt.c.
 - Implement a table to do identifier->locator mapping. This is
   an rhashtable (in ila_xlat.c).
 - Configuration for the table with netlink.
 - Add a hook into NF_INET_PRE_ROUTING to perform ILA translation
   before early demux.

Changes in v2:
 - Use iptables targets instead of a new xfrm function

Changes in v3:
 - Add __rcu to next pointer in struct ila_map

Changes in v4:
 - Use hook for NF_INET_PRE_ROUTING

Changed in v5:
 - Register hooks per namespace using nf_register_net_hooks
 - Only register hooks when first mapping is actually added

Changed in v6:
  - Remove gfp argument in alloc_ila_locks, it is unnecessary
  - Set registered_hooks properly when hooks are registered

Testing:
   Running 200 netperf TCP_RR streams

No ILA, baseline
   79.26% CPU utilization
   1678282 tps
   104/189/390 50/90/99% latencies

ILA before fix (LWT on both input and output)
   81.91% CPU utilization
   1464723 tps (-14.5% from baseline)
   121/215/411 50/90/99% latencies

ILA after fix
   80.62% CPU utilization
   1622985 (-3.4% from baseline)
   110/191/347 50/90/99% latencies
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/netlink.h

@@ -131,6 +131,7 @@ netlink_skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
 struct netlink_callback {
 	struct sk_buff		*skb;
 	const struct nlmsghdr	*nlh;
+	int			(*start)(struct netlink_callback *);
 	int			(*dump)(struct sk_buff * skb,
 					struct netlink_callback *cb);
 	int			(*done)(struct netlink_callback *cb);
@@ -153,6 +154,7 @@ struct nlmsghdr *
 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags);
 
 struct netlink_dump_control {
+	int (*start)(struct netlink_callback *);
 	int (*dump)(struct sk_buff *skb, struct netlink_callback *);
 	int (*done)(struct netlink_callback *);
 	void *data;

include/linux/rhashtable.h

@@ -819,4 +819,86 @@ static inline int rhashtable_remove_fast(
 	return err;
 }
 
+/* Internal function, please use rhashtable_replace_fast() instead */
+static inline int __rhashtable_replace_fast(
+	struct rhashtable *ht, struct bucket_table *tbl,
+	struct rhash_head *obj_old, struct rhash_head *obj_new,
+	const struct rhashtable_params params)
+{
+	struct rhash_head __rcu **pprev;
+	struct rhash_head *he;
+	spinlock_t *lock;
+	unsigned int hash;
+	int err = -ENOENT;
+
+	/* Minimally, the old and new objects must have same hash
+	 * (which should mean identifiers are the same).
+	 */
+	hash = rht_head_hashfn(ht, tbl, obj_old, params);
+	if (hash != rht_head_hashfn(ht, tbl, obj_new, params))
+		return -EINVAL;
+
+	lock = rht_bucket_lock(tbl, hash);
+
+	spin_lock_bh(lock);
+
+	pprev = &tbl->buckets[hash];
+	rht_for_each(he, tbl, hash) {
+		if (he != obj_old) {
+			pprev = &he->next;
+			continue;
+		}
+
+		rcu_assign_pointer(obj_new->next, obj_old->next);
+		rcu_assign_pointer(*pprev, obj_new);
+		err = 0;
+		break;
+	}
+
+	spin_unlock_bh(lock);
+
+	return err;
+}
+
+/**
+ * rhashtable_replace_fast - replace an object in hash table
+ * @ht:		hash table
+ * @obj_old:	pointer to hash head inside object being replaced
+ * @obj_new:	pointer to hash head inside object which is new
+ * @params:	hash table parameters
+ *
+ * Replacing an object doesn't affect the number of elements in the hash table
+ * or bucket, so we don't need to worry about shrinking or expanding the
+ * table here.
+ *
+ * Returns zero on success, -ENOENT if the entry could not be found,
+ * -EINVAL if hash is not the same for the old and new objects.
+ */
+static inline int rhashtable_replace_fast(
+	struct rhashtable *ht, struct rhash_head *obj_old,
+	struct rhash_head *obj_new,
+	const struct rhashtable_params params)
+{
+	struct bucket_table *tbl;
+	int err;
+
+	rcu_read_lock();
+
+	tbl = rht_dereference_rcu(ht->tbl, ht);
+
+	/* Because we have already taken (and released) the bucket
+	 * lock in old_tbl, if we find that future_tbl is not yet
+	 * visible then that guarantees the entry to still be in
+	 * the old tbl if it exists.
+	 */
+	while ((err = __rhashtable_replace_fast(ht, tbl, obj_old,
+						obj_new, params)) &&
+	       (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
+		;
+
+	rcu_read_unlock();
+
+	return err;
+}
+
 #endif /* _LINUX_RHASHTABLE_H */

include/net/genetlink.h

@@ -114,6 +114,7 @@ static inline void genl_info_net_set(struct genl_info *info, struct net *net)
  * @flags: flags
  * @policy: attribute validation policy
  * @doit: standard command callback
+ * @start: start callback for dumps
  * @dumpit: callback for dumpers
  * @done: completion callback for dumps
  * @ops_list: operations list
@@ -122,6 +123,7 @@ struct genl_ops {
 	const struct nla_policy	*policy;
 	int		       (*doit)(struct sk_buff *skb,
 				       struct genl_info *info);
+	int		       (*start)(struct netlink_callback *cb);
 	int		       (*dumpit)(struct sk_buff *skb,
 					 struct netlink_callback *cb);
 	int		       (*done)(struct netlink_callback *cb);

include/net/ila.h

+/*
+ * ILA kernel interface
+ *
+ * Copyright (c) 2015 Tom Herbert <tom@herbertland.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ */
+
+#ifndef _NET_ILA_H
+#define _NET_ILA_H
+
+int ila_xlat_outgoing(struct sk_buff *skb);
+int ila_xlat_incoming(struct sk_buff *skb);
+
+#endif /* _NET_ILA_H */

include/uapi/linux/ila.h

@@ -3,13 +3,35 @@
 #ifndef _UAPI_LINUX_ILA_H
 #define _UAPI_LINUX_ILA_H
 
+/* NETLINK_GENERIC related info */
+#define ILA_GENL_NAME		"ila"
+#define ILA_GENL_VERSION	0x1
+
 enum {
 	ILA_ATTR_UNSPEC,
 	ILA_ATTR_LOCATOR,			/* u64 */
+	ILA_ATTR_IDENTIFIER,			/* u64 */
+	ILA_ATTR_LOCATOR_MATCH,			/* u64 */
+	ILA_ATTR_IFINDEX,			/* s32 */
+	ILA_ATTR_DIR,				/* u32 */
 
 	__ILA_ATTR_MAX,
 };
 
 #define ILA_ATTR_MAX		(__ILA_ATTR_MAX - 1)
 
+enum {
+	ILA_CMD_UNSPEC,
+	ILA_CMD_ADD,
+	ILA_CMD_DEL,
+	ILA_CMD_GET,
+
+	__ILA_CMD_MAX,
+};
+
+#define ILA_CMD_MAX	(__ILA_CMD_MAX - 1)
+
+#define ILA_DIR_IN	(1 << 0)
+#define ILA_DIR_OUT	(1 << 1)
+
 #endif /* _UAPI_LINUX_ILA_H */

net/ipv6/Makefile

@@ -34,7 +34,7 @@ obj-$(CONFIG_INET6_XFRM_MODE_TUNNEL) += xfrm6_mode_tunnel.o
 obj-$(CONFIG_INET6_XFRM_MODE_ROUTEOPTIMIZATION) += xfrm6_mode_ro.o
 obj-$(CONFIG_INET6_XFRM_MODE_BEET) += xfrm6_mode_beet.o
 obj-$(CONFIG_IPV6_MIP6) += mip6.o
-obj-$(CONFIG_IPV6_ILA) += ila.o
+obj-$(CONFIG_IPV6_ILA) += ila/
 obj-$(CONFIG_NETFILTER)	+= netfilter/
 
 obj-$(CONFIG_IPV6_VTI) += ip6_vti.o

net/ipv6/ila/Makefile

+#
+# Makefile for ILA module
+#
+
+obj-$(CONFIG_IPV6_ILA) += ila.o
+
+ila-objs := ila_common.o ila_lwt.o ila_xlat.o

net/ipv6/ila/ila.h

+/*
+ * Copyright (c) 2015 Tom Herbert <tom@herbertland.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ *
+ */
+
+#ifndef __ILA_H
+#define __ILA_H
+
+#include <linux/errno.h>
+#include <linux/ip.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/socket.h>
+#include <linux/skbuff.h>
+#include <linux/types.h>
+#include <net/checksum.h>
+#include <net/ip.h>
+#include <net/protocol.h>
+#include <uapi/linux/ila.h>
+
+struct ila_params {
+	__be64 locator;
+	__be64 locator_match;
+	__wsum csum_diff;
+};
+
+static inline __wsum compute_csum_diff8(const __be32 *from, const __be32 *to)
+{
+	__be32 diff[] = {
+		~from[0], ~from[1], to[0], to[1],
+	};
+
+	return csum_partial(diff, sizeof(diff), 0);
+}
+
+void update_ipv6_locator(struct sk_buff *skb, struct ila_params *p);
+
+int ila_lwt_init(void);
+void ila_lwt_fini(void);
+int ila_xlat_init(void);
+void ila_xlat_fini(void);
+
+#endif /* __ILA_H */

net/ipv6/ila/ila_common.c

+#include <linux/errno.h>
+#include <linux/ip.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/skbuff.h>
+#include <linux/socket.h>
+#include <linux/types.h>
+#include <net/checksum.h>
+#include <net/ip.h>
+#include <net/ip6_fib.h>
+#include <net/lwtunnel.h>
+#include <net/protocol.h>
+#include <uapi/linux/ila.h>
+#include "ila.h"
+
+static __wsum get_csum_diff(struct ipv6hdr *ip6h, struct ila_params *p)
+{
+	if (*(__be64 *)&ip6h->daddr == p->locator_match)
+		return p->csum_diff;
+	else
+		return compute_csum_diff8((__be32 *)&ip6h->daddr,
+					  (__be32 *)&p->locator);
+}
+
+void update_ipv6_locator(struct sk_buff *skb, struct ila_params *p)
+{
+	__wsum diff;
+	struct ipv6hdr *ip6h = ipv6_hdr(skb);
+	size_t nhoff = sizeof(struct ipv6hdr);
+
+	/* First update checksum */
+	switch (ip6h->nexthdr) {
+	case NEXTHDR_TCP:
+		if (likely(pskb_may_pull(skb, nhoff + sizeof(struct tcphdr)))) {
+			struct tcphdr *th = (struct tcphdr *)
+					(skb_network_header(skb) + nhoff);
+
+			diff = get_csum_diff(ip6h, p);
+			inet_proto_csum_replace_by_diff(&th->check, skb,
+							diff, true);
+		}
+		break;
+	case NEXTHDR_UDP:
+		if (likely(pskb_may_pull(skb, nhoff + sizeof(struct udphdr)))) {
+			struct udphdr *uh = (struct udphdr *)
+					(skb_network_header(skb) + nhoff);
+
+			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
+				diff = get_csum_diff(ip6h, p);
+				inet_proto_csum_replace_by_diff(&uh->check, skb,
+								diff, true);
+				if (!uh->check)
+					uh->check = CSUM_MANGLED_0;
+			}
+		}
+		break;
+	case NEXTHDR_ICMP:
+		if (likely(pskb_may_pull(skb,
+					 nhoff + sizeof(struct icmp6hdr)))) {
+			struct icmp6hdr *ih = (struct icmp6hdr *)
+					(skb_network_header(skb) + nhoff);
+
+			diff = get_csum_diff(ip6h, p);
+			inet_proto_csum_replace_by_diff(&ih->icmp6_cksum, skb,
+							diff, true);
+		}
+		break;
+	}
+
+	/* Now change destination address */
+	*(__be64 *)&ip6h->daddr = p->locator;
+}
+
+static int __init ila_init(void)
+{
+	int ret;
+
+	ret = ila_lwt_init();
+
+	if (ret)
+		goto fail_lwt;
+
+	ret = ila_xlat_init();
+	if (ret)
+		goto fail_xlat;
+
+	return 0;
+fail_xlat:
+	ila_lwt_fini();
+fail_lwt:
+	return ret;
+}
+
+static void __exit ila_fini(void)
+{
+	ila_xlat_fini();
+	ila_lwt_fini();
+}
+
+module_init(ila_init);
+module_exit(ila_fini);
+MODULE_AUTHOR("Tom Herbert <tom@herbertland.com>");
+MODULE_LICENSE("GPL");

net/ipv6/ila.c → net/ipv6/ila/ila_lwt.c

@@ -11,12 +11,7 @@
 #include <net/lwtunnel.h>
 #include <net/protocol.h>
 #include <uapi/linux/ila.h>
-
-struct ila_params {
-	__be64 locator;
-	__be64 locator_match;
-	__wsum csum_diff;
-};
+#include "ila.h"
 
 static inline struct ila_params *ila_params_lwtunnel(
 	struct lwtunnel_state *lwstate)
@@ -24,73 +19,6 @@ static inline struct ila_params *ila_params_lwtunnel(
 	return (struct ila_params *)lwstate->data;
 }
 
-static inline __wsum compute_csum_diff8(const __be32 *from, const __be32 *to)
-{
-	__be32 diff[] = {
-		~from[0], ~from[1], to[0], to[1],
-	};
-
-	return csum_partial(diff, sizeof(diff), 0);
-}
-
-static inline __wsum get_csum_diff(struct ipv6hdr *ip6h, struct ila_params *p)
-{
-	if (*(__be64 *)&ip6h->daddr == p->locator_match)
-		return p->csum_diff;
-	else
-		return compute_csum_diff8((__be32 *)&ip6h->daddr,
-					  (__be32 *)&p->locator);
-}
-
-static void update_ipv6_locator(struct sk_buff *skb, struct ila_params *p)
-{
-	__wsum diff;
-	struct ipv6hdr *ip6h = ipv6_hdr(skb);
-	size_t nhoff = sizeof(struct ipv6hdr);
-
-	/* First update checksum */
-	switch (ip6h->nexthdr) {
-	case NEXTHDR_TCP:
-		if (likely(pskb_may_pull(skb, nhoff + sizeof(struct tcphdr)))) {
-			struct tcphdr *th = (struct tcphdr *)
-					(skb_network_header(skb) + nhoff);
-
-			diff = get_csum_diff(ip6h, p);
-			inet_proto_csum_replace_by_diff(&th->check, skb,
-							diff, true);
-		}
-		break;
-	case NEXTHDR_UDP:
-		if (likely(pskb_may_pull(skb, nhoff + sizeof(struct udphdr)))) {
-			struct udphdr *uh = (struct udphdr *)
-					(skb_network_header(skb) + nhoff);
-
-			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
-				diff = get_csum_diff(ip6h, p);
-				inet_proto_csum_replace_by_diff(&uh->check, skb,
-								diff, true);
-				if (!uh->check)
-					uh->check = CSUM_MANGLED_0;
-			}
-		}
-		break;
-	case NEXTHDR_ICMP:
-		if (likely(pskb_may_pull(skb,
-					 nhoff + sizeof(struct icmp6hdr)))) {
-			struct icmp6hdr *ih = (struct icmp6hdr *)
-					(skb_network_header(skb) + nhoff);
-
-			diff = get_csum_diff(ip6h, p);
-			inet_proto_csum_replace_by_diff(&ih->icmp6_cksum, skb,
-							diff, true);
-		}
-		break;
-	}
-
-	/* Now change destination address */
-	*(__be64 *)&ip6h->daddr = p->locator;
-}
-
 static int ila_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 {
 	struct dst_entry *dst = skb_dst(skb);
@@ -213,17 +141,12 @@ static const struct lwtunnel_encap_ops ila_encap_ops = {
 	.cmp_encap = ila_encap_cmp,
 };
 
-static int __init ila_init(void)
+int ila_lwt_init(void)
 {
 	return lwtunnel_encap_add_ops(&ila_encap_ops, LWTUNNEL_ENCAP_ILA);
 }
 
-static void __exit ila_fini(void)
+void ila_lwt_fini(void)
 {
 	lwtunnel_encap_del_ops(&ila_encap_ops, LWTUNNEL_ENCAP_ILA);
 }
-
-module_init(ila_init);
-module_exit(ila_fini);
-MODULE_AUTHOR("Tom Herbert <tom@herbertland.com>");
-MODULE_LICENSE("GPL");

net/netlink/af_netlink.c

@@ -2915,6 +2915,7 @@ int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
 
 	cb = &nlk->cb;
 	memset(cb, 0, sizeof(*cb));
+	cb->start = control->start;
 	cb->dump = control->dump;
 	cb->done = control->done;
 	cb->nlh = nlh;
@@ -2927,6 +2928,9 @@ int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
 
 	mutex_unlock(nlk->cb_mutex);
 
+	if (cb->start)
+		cb->start(cb);
+
 	ret = netlink_dump(sk);
 	sock_put(sk);
 

net/netlink/genetlink.c

@@ -513,6 +513,20 @@ void *genlmsg_put(struct sk_buff *skb, u32 portid, u32 seq,
 }
 EXPORT_SYMBOL(genlmsg_put);
 
+static int genl_lock_start(struct netlink_callback *cb)
+{
+	/* our ops are always const - netlink API doesn't propagate that */
+	const struct genl_ops *ops = cb->data;
+	int rc = 0;
+
+	if (ops->start) {
+		genl_lock();
+		rc = ops->start(cb);
+		genl_unlock();
+	}
+	return rc;
+}
+
 static int genl_lock_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
 {
 	/* our ops are always const - netlink API doesn't propagate that */
@@ -577,6 +591,7 @@ static int genl_family_rcv_msg(struct genl_family *family,
 				.module = family->module,
 				/* we have const, but the netlink API doesn't */
 				.data = (void *)ops,
+				.start = genl_lock_start,
 				.dump = genl_lock_dumpit,
 				.done = genl_lock_done,
 			};
@@ -588,6 +603,7 @@ static int genl_family_rcv_msg(struct genl_family *family,
 		} else {
 			struct netlink_dump_control c = {
 				.module = family->module,
+				.start = ops->start,
 				.dump = ops->dumpit,
 				.done = ops->done,
 			};