[PATCH v4 net-next 0/4] Faster SO_REUSEPORT

[PATCH v4 net-next 0/4] Faster SO_REUSEPORT

[Craig Gallek]

From: Craig Gallek <kraig@google.com>

This series contains two optimizations for the SO_REUSEPORT feature:
Faster lookup when selecting a socket for an incoming packet and
the ability to select the socket from the group using a BPF program.

This series only includes the UDP path.  I plan to submit a follow-up
including the TCP path if the implementation in this series is
acceptable.

Changes in v4:
- pskb_may_pull is unnecessary with pskb_pull (per Alexei Starovoitov)

Changes in v3:
- skb_pull_inline -> pskb_pull (per Alexei Starovoitov)
- reuseport_attach* -> sk_reuseport_attach* and simple return statement
  syntax change (per Daniel Borkmann)

Changes in v2:
- Fix ARM build; remove unnecessary include.
- Handle case where protocol header is not in linear section (per
  Alexei Starovoitov).

Craig Gallek (4):
  soreuseport: define reuseport groups
  soreuseport: fast reuseport UDP socket selection
  soreuseport: setsockopt SO_ATTACH_REUSEPORT_[CE]BPF
  soreuseport: BPF selection functional test

 arch/alpha/include/uapi/asm/socket.h        |   3 +
 arch/avr32/include/uapi/asm/socket.h        |   3 +
 arch/frv/include/uapi/asm/socket.h          |   3 +
 arch/ia64/include/uapi/asm/socket.h         |   3 +
 arch/m32r/include/uapi/asm/socket.h         |   3 +
 arch/mips/include/uapi/asm/socket.h         |   3 +
 arch/mn10300/include/uapi/asm/socket.h      |   3 +
 arch/parisc/include/uapi/asm/socket.h       |   3 +
 arch/powerpc/include/uapi/asm/socket.h      |   3 +
 arch/s390/include/uapi/asm/socket.h         |   3 +
 arch/sparc/include/uapi/asm/socket.h        |   3 +
 arch/xtensa/include/uapi/asm/socket.h       |   3 +
 include/linux/filter.h                      |   2 +
 include/net/addrconf.h                      |   3 +-
 include/net/sock.h                          |   2 +
 include/net/sock_reuseport.h                |  28 ++
 include/net/udp.h                           |   7 +-
 include/uapi/asm-generic/socket.h           |   3 +
 net/core/Makefile                           |   2 +-
 net/core/filter.c                           | 121 +++++--
 net/core/sock.c                             |  29 ++
 net/core/sock_reuseport.c                   | 251 +++++++++++++++
 net/ipv4/udp.c                              | 127 ++++++--
 net/ipv4/udp_diag.c                         |   4 +-
 net/ipv6/inet6_connection_sock.c            |   4 +-
 net/ipv6/udp.c                              |  56 +++-
 tools/testing/selftests/net/.gitignore      |   1 +
 tools/testing/selftests/net/Makefile        |   2 +-
 tools/testing/selftests/net/reuseport_bpf.c | 467 ++++++++++++++++++++++++++++
 29 files changed, 1076 insertions(+), 69 deletions(-)
 create mode 100644 include/net/sock_reuseport.h
 create mode 100644 net/core/sock_reuseport.c
 create mode 100644 tools/testing/selftests/net/reuseport_bpf.c

-- 
2.6.0.rc2.230.g3dd15c0

[PATCH v4 net-next 1/4] soreuseport: define reuseport groups

[Craig Gallek]

From: Craig Gallek <kraig@google.com>

struct sock_reuseport is an optional shared structure referenced by each
socket belonging to a reuseport group.  When a socket is bound to an
address/port not yet in use and the reuseport flag has been set, the
structure will be allocated and attached to the newly bound socket.
When subsequent calls to bind are made for the same address/port, the
shared structure will be updated to include the new socket and the
newly bound socket will reference the group structure.

Usually, when an incoming packet was destined for a reuseport group,
all sockets in the same group needed to be considered before a
dispatching decision was made.  With this structure, an appropriate
socket can be found after looking up just one socket in the group.

This shared structure will also allow for more complicated decisions to
be made when selecting a socket (eg a BPF filter).

This work is based off a similar implementation written by
Ying Cai <ycai@google.com> for implementing policy-based reuseport
selection.

Signed-off-by: Craig Gallek <kraig@google.com>
---
 include/net/sock.h           |   2 +
 include/net/sock_reuseport.h |  20 +++++
 net/core/Makefile            |   2 +-
 net/core/sock_reuseport.c    | 173 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 196 insertions(+), 1 deletion(-)
 create mode 100644 include/net/sock_reuseport.h
 create mode 100644 net/core/sock_reuseport.c

diff --git a/include/net/sock.h b/include/net/sock.h
index 3794cdd..e830c10 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -318,6 +318,7 @@ struct cg_proto;
   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
   *	@sk_backlog_rcv: callback to process the backlog
   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
+  *	@sk_reuseport_cb: reuseport group container
  */
 struct sock {
 	/*
@@ -453,6 +454,7 @@ struct sock {
 	int			(*sk_backlog_rcv)(struct sock *sk,
 						  struct sk_buff *skb);
 	void                    (*sk_destruct)(struct sock *sk);
+	struct sock_reuseport __rcu	*sk_reuseport_cb;
 };
 
 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
diff --git a/include/net/sock_reuseport.h b/include/net/sock_reuseport.h
new file mode 100644
index 0000000..67d1eb8
--- /dev/null
+++ b/include/net/sock_reuseport.h
@@ -0,0 +1,20 @@
+#ifndef _SOCK_REUSEPORT_H
+#define _SOCK_REUSEPORT_H
+
+#include <linux/types.h>
+#include <net/sock.h>
+
+struct sock_reuseport {
+	struct rcu_head		rcu;
+
+	u16			max_socks;	/* length of socks */
+	u16			num_socks;	/* elements in socks */
+	struct sock		*socks[0];	/* array of sock pointers */
+};
+
+extern int reuseport_alloc(struct sock *sk);
+extern int reuseport_add_sock(struct sock *sk, const struct sock *sk2);
+extern void reuseport_detach_sock(struct sock *sk);
+extern struct sock *reuseport_select_sock(struct sock *sk, u32 hash);
+
+#endif  /* _SOCK_REUSEPORT_H */
diff --git a/net/core/Makefile b/net/core/Makefile
index 086b01f..0b835de 100644
--- a/net/core/Makefile
+++ b/net/core/Makefile
@@ -9,7 +9,7 @@ obj-$(CONFIG_SYSCTL) += sysctl_net_core.o
 
 obj-y		     += dev.o ethtool.o dev_addr_lists.o dst.o netevent.o \
 			neighbour.o rtnetlink.o utils.o link_watch.o filter.o \
-			sock_diag.o dev_ioctl.o tso.o
+			sock_diag.o dev_ioctl.o tso.o sock_reuseport.o
 
 obj-$(CONFIG_XFRM) += flow.o
 obj-y += net-sysfs.o
diff --git a/net/core/sock_reuseport.c b/net/core/sock_reuseport.c
new file mode 100644
index 0000000..963c8d5
--- /dev/null
+++ b/net/core/sock_reuseport.c
@@ -0,0 +1,173 @@
+/*
+ * To speed up listener socket lookup, create an array to store all sockets
+ * listening on the same port.  This allows a decision to be made after finding
+ * the first socket.
+ */
+
+#include <net/sock_reuseport.h>
+#include <linux/rcupdate.h>
+
+#define INIT_SOCKS 128
+
+static DEFINE_SPINLOCK(reuseport_lock);
+
+static struct sock_reuseport *__reuseport_alloc(u16 max_socks)
+{
+	size_t size = sizeof(struct sock_reuseport) +
+		      sizeof(struct sock *) * max_socks;
+	struct sock_reuseport *reuse = kzalloc(size, GFP_ATOMIC);
+
+	if (!reuse)
+		return NULL;
+
+	reuse->max_socks = max_socks;
+
+	return reuse;
+}
+
+int reuseport_alloc(struct sock *sk)
+{
+	struct sock_reuseport *reuse;
+
+	/* bh lock used since this function call may precede hlist lock in
+	 * soft irq of receive path or setsockopt from process context
+	 */
+	spin_lock_bh(&reuseport_lock);
+	WARN_ONCE(rcu_dereference_protected(sk->sk_reuseport_cb,
+					    lockdep_is_held(&reuseport_lock)),
+		  "multiple allocations for the same socket");
+	reuse = __reuseport_alloc(INIT_SOCKS);
+	if (!reuse) {
+		spin_unlock_bh(&reuseport_lock);
+		return -ENOMEM;
+	}
+
+	reuse->socks[0] = sk;
+	reuse->num_socks = 1;
+	rcu_assign_pointer(sk->sk_reuseport_cb, reuse);
+
+	spin_unlock_bh(&reuseport_lock);
+
+	return 0;
+}
+EXPORT_SYMBOL(reuseport_alloc);
+
+static struct sock_reuseport *reuseport_grow(struct sock_reuseport *reuse)
+{
+	struct sock_reuseport *more_reuse;
+	u32 more_socks_size, i;
+
+	more_socks_size = reuse->max_socks * 2U;
+	if (more_socks_size > U16_MAX)
+		return NULL;
+
+	more_reuse = __reuseport_alloc(more_socks_size);
+	if (!more_reuse)
+		return NULL;
+
+	more_reuse->max_socks = more_socks_size;
+	more_reuse->num_socks = reuse->num_socks;
+
+	memcpy(more_reuse->socks, reuse->socks,
+	       reuse->num_socks * sizeof(struct sock *));
+
+	for (i = 0; i < reuse->num_socks; ++i)
+		rcu_assign_pointer(reuse->socks[i]->sk_reuseport_cb,
+				   more_reuse);
+
+	kfree_rcu(reuse, rcu);
+	return more_reuse;
+}
+
+/**
+ *  reuseport_add_sock - Add a socket to the reuseport group of another.
+ *  @sk:  New socket to add to the group.
+ *  @sk2: Socket belonging to the existing reuseport group.
+ *  May return ENOMEM and not add socket to group under memory pressure.
+ */
+int reuseport_add_sock(struct sock *sk, const struct sock *sk2)
+{
+	struct sock_reuseport *reuse;
+
+	spin_lock_bh(&reuseport_lock);
+	reuse = rcu_dereference_protected(sk2->sk_reuseport_cb,
+					  lockdep_is_held(&reuseport_lock)),
+	WARN_ONCE(rcu_dereference_protected(sk->sk_reuseport_cb,
+					    lockdep_is_held(&reuseport_lock)),
+		  "socket already in reuseport group");
+
+	if (reuse->num_socks == reuse->max_socks) {
+		reuse = reuseport_grow(reuse);
+		if (!reuse) {
+			spin_unlock_bh(&reuseport_lock);
+			return -ENOMEM;
+		}
+	}
+
+	reuse->socks[reuse->num_socks] = sk;
+	/* paired with smp_rmb() in reuseport_select_sock() */
+	smp_wmb();
+	reuse->num_socks++;
+	rcu_assign_pointer(sk->sk_reuseport_cb, reuse);
+
+	spin_unlock_bh(&reuseport_lock);
+
+	return 0;
+}
+EXPORT_SYMBOL(reuseport_add_sock);
+
+void reuseport_detach_sock(struct sock *sk)
+{
+	struct sock_reuseport *reuse;
+	int i;
+
+	spin_lock_bh(&reuseport_lock);
+	reuse = rcu_dereference_protected(sk->sk_reuseport_cb,
+					  lockdep_is_held(&reuseport_lock));
+	rcu_assign_pointer(sk->sk_reuseport_cb, NULL);
+
+	for (i = 0; i < reuse->num_socks; i++) {
+		if (reuse->socks[i] == sk) {
+			reuse->socks[i] = reuse->socks[reuse->num_socks - 1];
+			reuse->num_socks--;
+			if (reuse->num_socks == 0)
+				kfree_rcu(reuse, rcu);
+			break;
+		}
+	}
+	spin_unlock_bh(&reuseport_lock);
+}
+EXPORT_SYMBOL(reuseport_detach_sock);
+
+/**
+ *  reuseport_select_sock - Select a socket from an SO_REUSEPORT group.
+ *  @sk: First socket in the group.
+ *  @hash: Use this hash to select.
+ *  Returns a socket that should receive the packet (or NULL on error).
+ */
+struct sock *reuseport_select_sock(struct sock *sk, u32 hash)
+{
+	struct sock_reuseport *reuse;
+	struct sock *sk2 = NULL;
+	u16 socks;
+
+	rcu_read_lock();
+	reuse = rcu_dereference(sk->sk_reuseport_cb);
+
+	/* if memory allocation failed or add call is not yet complete */
+	if (!reuse)
+		goto out;
+
+	socks = READ_ONCE(reuse->num_socks);
+	if (likely(socks)) {
+		/* paired with smp_wmb() in reuseport_add_sock() */
+		smp_rmb();
+
+		sk2 = reuse->socks[reciprocal_scale(hash, socks)];
+	}
+
+out:
+	rcu_read_unlock();
+	return sk2;
+}
+EXPORT_SYMBOL(reuseport_select_sock);
-- 
2.6.0.rc2.230.g3dd15c0

Re: [PATCH v4 net-next 1/4] soreuseport: define reuseport groups

[Eric Dumazet]

On Mon, 2016-01-04 at 17:41 -0500, Craig Gallek wrote:
> From: Craig Gallek <kraig@google.com>
> 


Acked-by: Eric Dumazet <edumazet@google.com>

[PATCH v4 net-next 2/4] soreuseport: fast reuseport UDP socket selection

[Craig Gallek]

From: Craig Gallek <kraig@google.com>

Include a struct sock_reuseport instance when a UDP socket binds to
a specific address for the first time with the reuseport flag set.
When selecting a socket for an incoming UDP packet, use the information
available in sock_reuseport if present.

This required adding an additional field to the UDP source address
equality function to differentiate between exact and wildcard matches.
The original use case allowed wildcard matches when checking for
existing port uses during bind.  The new use case of adding a socket
to a reuseport group requires exact address matching.

Performance test (using a machine with 2 CPU sockets and a total of
48 cores):  Create reuseport groups of varying size.  Use one socket
from this group per user thread (pinning each thread to a different
core) calling recvmmsg in a tight loop.  Record number of messages
received per second while saturating a 10G link.
  10 sockets: 18% increase (~2.8M -> 3.3M pkts/s)
  20 sockets: 14% increase (~2.9M -> 3.3M pkts/s)
  40 sockets: 13% increase (~3.0M -> 3.4M pkts/s)

This work is based off a similar implementation written by
Ying Cai <ycai@google.com> for implementing policy-based reuseport
selection.

Signed-off-by: Craig Gallek <kraig@google.com>
---
 include/net/addrconf.h           |   3 +-
 include/net/udp.h                |   2 +-
 net/ipv4/udp.c                   | 119 +++++++++++++++++++++++++++++++--------
 net/ipv6/inet6_connection_sock.c |   4 +-
 net/ipv6/udp.c                   |  48 +++++++++++++---
 5 files changed, 141 insertions(+), 35 deletions(-)

diff --git a/include/net/addrconf.h b/include/net/addrconf.h
index 78003df..47f52d3 100644
--- a/include/net/addrconf.h
+++ b/include/net/addrconf.h
@@ -87,7 +87,8 @@ int __ipv6_get_lladdr(struct inet6_dev *idev, struct in6_addr *addr,
 		      u32 banned_flags);
 int ipv6_get_lladdr(struct net_device *dev, struct in6_addr *addr,
 		    u32 banned_flags);
-int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2);
+int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
+			 bool match_wildcard);
 void addrconf_join_solict(struct net_device *dev, const struct in6_addr *addr);
 void addrconf_leave_solict(struct inet6_dev *idev, const struct in6_addr *addr);
 
diff --git a/include/net/udp.h b/include/net/udp.h
index 6d4ed18..3b5d7f9 100644
--- a/include/net/udp.h
+++ b/include/net/udp.h
@@ -191,7 +191,7 @@ static inline void udp_lib_close(struct sock *sk, long timeout)
 }
 
 int udp_lib_get_port(struct sock *sk, unsigned short snum,
-		     int (*)(const struct sock *, const struct sock *),
+		     int (*)(const struct sock *, const struct sock *, bool),
 		     unsigned int hash2_nulladdr);
 
 u32 udp_flow_hashrnd(void);
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index ac14ae4..762b01f 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -113,6 +113,7 @@
 #include <trace/events/skb.h>
 #include <net/busy_poll.h>
 #include "udp_impl.h"
+#include <net/sock_reuseport.h>
 
 struct udp_table udp_table __read_mostly;
 EXPORT_SYMBOL(udp_table);
@@ -137,7 +138,8 @@ static int udp_lib_lport_inuse(struct net *net, __u16 num,
 			       unsigned long *bitmap,
 			       struct sock *sk,
 			       int (*saddr_comp)(const struct sock *sk1,
-						 const struct sock *sk2),
+						 const struct sock *sk2,
+						 bool match_wildcard),
 			       unsigned int log)
 {
 	struct sock *sk2;
@@ -152,8 +154,9 @@ static int udp_lib_lport_inuse(struct net *net, __u16 num,
 		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 		    (!sk2->sk_reuseport || !sk->sk_reuseport ||
+		     rcu_access_pointer(sk->sk_reuseport_cb) ||
 		     !uid_eq(uid, sock_i_uid(sk2))) &&
-		    saddr_comp(sk, sk2)) {
+		    saddr_comp(sk, sk2, true)) {
 			if (!bitmap)
 				return 1;
 			__set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap);
@@ -170,7 +173,8 @@ static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 				struct udp_hslot *hslot2,
 				struct sock *sk,
 				int (*saddr_comp)(const struct sock *sk1,
-						  const struct sock *sk2))
+						  const struct sock *sk2,
+						  bool match_wildcard))
 {
 	struct sock *sk2;
 	struct hlist_nulls_node *node;
@@ -186,8 +190,9 @@ static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 		    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 		     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 		    (!sk2->sk_reuseport || !sk->sk_reuseport ||
+		     rcu_access_pointer(sk->sk_reuseport_cb) ||
 		     !uid_eq(uid, sock_i_uid(sk2))) &&
-		    saddr_comp(sk, sk2)) {
+		    saddr_comp(sk, sk2, true)) {
 			res = 1;
 			break;
 		}
@@ -196,6 +201,35 @@ static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 	return res;
 }
 
+static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot,
+				  int (*saddr_same)(const struct sock *sk1,
+						    const struct sock *sk2,
+						    bool match_wildcard))
+{
+	struct net *net = sock_net(sk);
+	struct hlist_nulls_node *node;
+	kuid_t uid = sock_i_uid(sk);
+	struct sock *sk2;
+
+	sk_nulls_for_each(sk2, node, &hslot->head) {
+		if (net_eq(sock_net(sk2), net) &&
+		    sk2 != sk &&
+		    sk2->sk_family == sk->sk_family &&
+		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
+		    (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
+		    (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
+		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
+		    (*saddr_same)(sk, sk2, false)) {
+			return reuseport_add_sock(sk, sk2);
+		}
+	}
+
+	/* Initial allocation may have already happened via setsockopt */
+	if (!rcu_access_pointer(sk->sk_reuseport_cb))
+		return reuseport_alloc(sk);
+	return 0;
+}
+
 /**
  *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
  *
@@ -207,7 +241,8 @@ static int udp_lib_lport_inuse2(struct net *net, __u16 num,
  */
 int udp_lib_get_port(struct sock *sk, unsigned short snum,
 		     int (*saddr_comp)(const struct sock *sk1,
-				       const struct sock *sk2),
+				       const struct sock *sk2,
+				       bool match_wildcard),
 		     unsigned int hash2_nulladdr)
 {
 	struct udp_hslot *hslot, *hslot2;
@@ -290,6 +325,14 @@ found:
 	udp_sk(sk)->udp_port_hash = snum;
 	udp_sk(sk)->udp_portaddr_hash ^= snum;
 	if (sk_unhashed(sk)) {
+		if (sk->sk_reuseport &&
+		    udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
+			inet_sk(sk)->inet_num = 0;
+			udp_sk(sk)->udp_port_hash = 0;
+			udp_sk(sk)->udp_portaddr_hash ^= snum;
+			goto fail_unlock;
+		}
+
 		sk_nulls_add_node_rcu(sk, &hslot->head);
 		hslot->count++;
 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
@@ -309,13 +352,22 @@ fail:
 }
 EXPORT_SYMBOL(udp_lib_get_port);
 
-static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
+/* match_wildcard == true:  0.0.0.0 equals to any IPv4 addresses
+ * match_wildcard == false: addresses must be exactly the same, i.e.
+ *                          0.0.0.0 only equals to 0.0.0.0
+ */
+static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
+				bool match_wildcard)
 {
 	struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
 
-	return 	(!ipv6_only_sock(sk2)  &&
-		 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
-		   inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
+	if (!ipv6_only_sock(sk2)) {
+		if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
+			return 1;
+		if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
+			return match_wildcard;
+	}
+	return 0;
 }
 
 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
@@ -459,8 +511,14 @@ begin:
 			badness = score;
 			reuseport = sk->sk_reuseport;
 			if (reuseport) {
+				struct sock *sk2;
 				hash = udp_ehashfn(net, daddr, hnum,
 						   saddr, sport);
+				sk2 = reuseport_select_sock(sk, hash);
+				if (sk2) {
+					result = sk2;
+					goto found;
+				}
 				matches = 1;
 			}
 		} else if (score == badness && reuseport) {
@@ -478,6 +536,7 @@ begin:
 	if (get_nulls_value(node) != slot2)
 		goto begin;
 	if (result) {
+found:
 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
 			result = NULL;
 		else if (unlikely(compute_score2(result, net, saddr, sport,
@@ -540,8 +599,14 @@ begin:
 			badness = score;
 			reuseport = sk->sk_reuseport;
 			if (reuseport) {
+				struct sock *sk2;
 				hash = udp_ehashfn(net, daddr, hnum,
 						   saddr, sport);
+				sk2 = reuseport_select_sock(sk, hash);
+				if (sk2) {
+					result = sk2;
+					goto found;
+				}
 				matches = 1;
 			}
 		} else if (score == badness && reuseport) {
@@ -560,6 +625,7 @@ begin:
 		goto begin;
 
 	if (result) {
+found:
 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
 			result = NULL;
 		else if (unlikely(compute_score(result, net, saddr, hnum, sport,
@@ -587,7 +653,8 @@ static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 			     __be32 daddr, __be16 dport, int dif)
 {
-	return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
+	return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif,
+				 &udp_table);
 }
 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 
@@ -1398,6 +1465,8 @@ void udp_lib_unhash(struct sock *sk)
 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 
 		spin_lock_bh(&hslot->lock);
+		if (rcu_access_pointer(sk->sk_reuseport_cb))
+			reuseport_detach_sock(sk);
 		if (sk_nulls_del_node_init_rcu(sk)) {
 			hslot->count--;
 			inet_sk(sk)->inet_num = 0;
@@ -1425,22 +1494,28 @@ void udp_lib_rehash(struct sock *sk, u16 newhash)
 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 		nhslot2 = udp_hashslot2(udptable, newhash);
 		udp_sk(sk)->udp_portaddr_hash = newhash;
-		if (hslot2 != nhslot2) {
+
+		if (hslot2 != nhslot2 ||
+		    rcu_access_pointer(sk->sk_reuseport_cb)) {
 			hslot = udp_hashslot(udptable, sock_net(sk),
 					     udp_sk(sk)->udp_port_hash);
 			/* we must lock primary chain too */
 			spin_lock_bh(&hslot->lock);
-
-			spin_lock(&hslot2->lock);
-			hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
-			hslot2->count--;
-			spin_unlock(&hslot2->lock);
-
-			spin_lock(&nhslot2->lock);
-			hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
-						 &nhslot2->head);
-			nhslot2->count++;
-			spin_unlock(&nhslot2->lock);
+			if (rcu_access_pointer(sk->sk_reuseport_cb))
+				reuseport_detach_sock(sk);
+
+			if (hslot2 != nhslot2) {
+				spin_lock(&hslot2->lock);
+				hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
+				hslot2->count--;
+				spin_unlock(&hslot2->lock);
+
+				spin_lock(&nhslot2->lock);
+				hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
+							 &nhslot2->head);
+				nhslot2->count++;
+				spin_unlock(&nhslot2->lock);
+			}
 
 			spin_unlock_bh(&hslot->lock);
 		}
diff --git a/net/ipv6/inet6_connection_sock.c b/net/ipv6/inet6_connection_sock.c
index a7ca2cd..36c3f01 100644
--- a/net/ipv6/inet6_connection_sock.c
+++ b/net/ipv6/inet6_connection_sock.c
@@ -51,12 +51,12 @@ int inet6_csk_bind_conflict(const struct sock *sk,
 			     (sk2->sk_state != TCP_TIME_WAIT &&
 			      !uid_eq(uid,
 				      sock_i_uid((struct sock *)sk2))))) {
-				if (ipv6_rcv_saddr_equal(sk, sk2))
+				if (ipv6_rcv_saddr_equal(sk, sk2, true))
 					break;
 			}
 			if (!relax && reuse && sk2->sk_reuse &&
 			    sk2->sk_state != TCP_LISTEN &&
-			    ipv6_rcv_saddr_equal(sk, sk2))
+			    ipv6_rcv_saddr_equal(sk, sk2, true))
 				break;
 		}
 	}
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 00775ee..6204b89 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -47,6 +47,7 @@
 #include <net/xfrm.h>
 #include <net/inet6_hashtables.h>
 #include <net/busy_poll.h>
+#include <net/sock_reuseport.h>
 
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
@@ -76,7 +77,14 @@ static u32 udp6_ehashfn(const struct net *net,
 			       udp_ipv6_hash_secret + net_hash_mix(net));
 }
 
-int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
+/* match_wildcard == true:  IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
+ *                          only, and any IPv4 addresses if not IPv6 only
+ * match_wildcard == false: addresses must be exactly the same, i.e.
+ *                          IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
+ *                          and 0.0.0.0 equals to 0.0.0.0 only
+ */
+int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
+			 bool match_wildcard)
 {
 	const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
 	int sk2_ipv6only = inet_v6_ipv6only(sk2);
@@ -84,16 +92,24 @@ int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
 	int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
 
 	/* if both are mapped, treat as IPv4 */
-	if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
-		return (!sk2_ipv6only &&
-			(!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr ||
-			  sk->sk_rcv_saddr == sk2->sk_rcv_saddr));
+	if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
+		if (!sk2_ipv6only) {
+			if (sk->sk_rcv_saddr == sk2->sk_rcv_saddr)
+				return 1;
+			if (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr)
+				return match_wildcard;
+		}
+		return 0;
+	}
+
+	if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
+		return 1;
 
-	if (addr_type2 == IPV6_ADDR_ANY &&
+	if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
 	    !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
 		return 1;
 
-	if (addr_type == IPV6_ADDR_ANY &&
+	if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
 	    !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED))
 		return 1;
 
@@ -253,8 +269,14 @@ begin:
 			badness = score;
 			reuseport = sk->sk_reuseport;
 			if (reuseport) {
+				struct sock *sk2;
 				hash = udp6_ehashfn(net, daddr, hnum,
 						    saddr, sport);
+				sk2 = reuseport_select_sock(sk, hash);
+				if (sk2) {
+					result = sk2;
+					goto found;
+				}
 				matches = 1;
 			}
 		} else if (score == badness && reuseport) {
@@ -273,6 +295,7 @@ begin:
 		goto begin;
 
 	if (result) {
+found:
 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
 			result = NULL;
 		else if (unlikely(compute_score2(result, net, saddr, sport,
@@ -332,8 +355,14 @@ begin:
 			badness = score;
 			reuseport = sk->sk_reuseport;
 			if (reuseport) {
+				struct sock *sk2;
 				hash = udp6_ehashfn(net, daddr, hnum,
 						    saddr, sport);
+				sk2 = reuseport_select_sock(sk, hash);
+				if (sk2) {
+					result = sk2;
+					goto found;
+				}
 				matches = 1;
 			}
 		} else if (score == badness && reuseport) {
@@ -352,6 +381,7 @@ begin:
 		goto begin;
 
 	if (result) {
+found:
 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
 			result = NULL;
 		else if (unlikely(compute_score(result, net, hnum, saddr, sport,
@@ -549,8 +579,8 @@ void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 	int err;
 	struct net *net = dev_net(skb->dev);
 
-	sk = __udp6_lib_lookup(net, daddr, uh->dest,
-			       saddr, uh->source, inet6_iif(skb), udptable);
+	sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
+			       inet6_iif(skb), udptable);
 	if (!sk) {
 		ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 				   ICMP6_MIB_INERRORS);
-- 
2.6.0.rc2.230.g3dd15c0

[PATCH v4 net-next 3/4] soreuseport: setsockopt SO_ATTACH_REUSEPORT_[CE]BPF

[Craig Gallek]

From: Craig Gallek <kraig@google.com>

Expose socket options for setting a classic or extended BPF program
for use when selecting sockets in an SO_REUSEPORT group.  These options
can be used on the first socket to belong to a group before bind or
on any socket in the group after bind.

This change includes refactoring of the existing sk_filter code to
allow reuse of the existing BPF filter validation checks.

Signed-off-by: Craig Gallek <kraig@google.com>
---
 arch/alpha/include/uapi/asm/socket.h   |   3 +
 arch/avr32/include/uapi/asm/socket.h   |   3 +
 arch/frv/include/uapi/asm/socket.h     |   3 +
 arch/ia64/include/uapi/asm/socket.h    |   3 +
 arch/m32r/include/uapi/asm/socket.h    |   3 +
 arch/mips/include/uapi/asm/socket.h    |   3 +
 arch/mn10300/include/uapi/asm/socket.h |   3 +
 arch/parisc/include/uapi/asm/socket.h  |   3 +
 arch/powerpc/include/uapi/asm/socket.h |   3 +
 arch/s390/include/uapi/asm/socket.h    |   3 +
 arch/sparc/include/uapi/asm/socket.h   |   3 +
 arch/xtensa/include/uapi/asm/socket.h  |   3 +
 include/linux/filter.h                 |   2 +
 include/net/sock_reuseport.h           |  10 ++-
 include/net/udp.h                      |   5 +-
 include/uapi/asm-generic/socket.h      |   3 +
 net/core/filter.c                      | 121 +++++++++++++++++++++++++++------
 net/core/sock.c                        |  29 ++++++++
 net/core/sock_reuseport.c              |  88 ++++++++++++++++++++++--
 net/ipv4/udp.c                         |  14 ++--
 net/ipv4/udp_diag.c                    |   4 +-
 net/ipv6/udp.c                         |  14 ++--
 22 files changed, 282 insertions(+), 44 deletions(-)

diff --git a/arch/alpha/include/uapi/asm/socket.h b/arch/alpha/include/uapi/asm/socket.h
index 9a20821..c5fb9e6 100644
--- a/arch/alpha/include/uapi/asm/socket.h
+++ b/arch/alpha/include/uapi/asm/socket.h
@@ -92,4 +92,7 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif /* _UAPI_ASM_SOCKET_H */
diff --git a/arch/avr32/include/uapi/asm/socket.h b/arch/avr32/include/uapi/asm/socket.h
index 2b65ed6..9de0796 100644
--- a/arch/avr32/include/uapi/asm/socket.h
+++ b/arch/avr32/include/uapi/asm/socket.h
@@ -85,4 +85,7 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif /* _UAPI__ASM_AVR32_SOCKET_H */
diff --git a/arch/frv/include/uapi/asm/socket.h b/arch/frv/include/uapi/asm/socket.h
index 4823ad1..f02e484 100644
--- a/arch/frv/include/uapi/asm/socket.h
+++ b/arch/frv/include/uapi/asm/socket.h
@@ -85,5 +85,8 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif /* _ASM_SOCKET_H */
 
diff --git a/arch/ia64/include/uapi/asm/socket.h b/arch/ia64/include/uapi/asm/socket.h
index 59be3d8..bce2916 100644
--- a/arch/ia64/include/uapi/asm/socket.h
+++ b/arch/ia64/include/uapi/asm/socket.h
@@ -94,4 +94,7 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif /* _ASM_IA64_SOCKET_H */
diff --git a/arch/m32r/include/uapi/asm/socket.h b/arch/m32r/include/uapi/asm/socket.h
index 7bc4cb2..14aa4a6 100644
--- a/arch/m32r/include/uapi/asm/socket.h
+++ b/arch/m32r/include/uapi/asm/socket.h
@@ -85,4 +85,7 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif /* _ASM_M32R_SOCKET_H */
diff --git a/arch/mips/include/uapi/asm/socket.h b/arch/mips/include/uapi/asm/socket.h
index dec3c85..5910fe2 100644
--- a/arch/mips/include/uapi/asm/socket.h
+++ b/arch/mips/include/uapi/asm/socket.h
@@ -103,4 +103,7 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif /* _UAPI_ASM_SOCKET_H */
diff --git a/arch/mn10300/include/uapi/asm/socket.h b/arch/mn10300/include/uapi/asm/socket.h
index cab7d6d..58b1aa0 100644
--- a/arch/mn10300/include/uapi/asm/socket.h
+++ b/arch/mn10300/include/uapi/asm/socket.h
@@ -85,4 +85,7 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif /* _ASM_SOCKET_H */
diff --git a/arch/parisc/include/uapi/asm/socket.h b/arch/parisc/include/uapi/asm/socket.h
index a5cd40c..f9cf122 100644
--- a/arch/parisc/include/uapi/asm/socket.h
+++ b/arch/parisc/include/uapi/asm/socket.h
@@ -84,4 +84,7 @@
 #define SO_ATTACH_BPF		0x402B
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	0x402C
+#define SO_ATTACH_REUSEPORT_EBPF	0x402D
+
 #endif /* _UAPI_ASM_SOCKET_H */
diff --git a/arch/powerpc/include/uapi/asm/socket.h b/arch/powerpc/include/uapi/asm/socket.h
index c046666..dd54f28 100644
--- a/arch/powerpc/include/uapi/asm/socket.h
+++ b/arch/powerpc/include/uapi/asm/socket.h
@@ -92,4 +92,7 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif	/* _ASM_POWERPC_SOCKET_H */
diff --git a/arch/s390/include/uapi/asm/socket.h b/arch/s390/include/uapi/asm/socket.h
index 296942d..d02e89d 100644
--- a/arch/s390/include/uapi/asm/socket.h
+++ b/arch/s390/include/uapi/asm/socket.h
@@ -91,4 +91,7 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif /* _ASM_SOCKET_H */
diff --git a/arch/sparc/include/uapi/asm/socket.h b/arch/sparc/include/uapi/asm/socket.h
index e6a16c4..d270ee9 100644
--- a/arch/sparc/include/uapi/asm/socket.h
+++ b/arch/sparc/include/uapi/asm/socket.h
@@ -81,6 +81,9 @@
 #define SO_ATTACH_BPF		0x0034
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	0x0035
+#define SO_ATTACH_REUSEPORT_EBPF	0x0036
+
 /* Security levels - as per NRL IPv6 - don't actually do anything */
 #define SO_SECURITY_AUTHENTICATION		0x5001
 #define SO_SECURITY_ENCRYPTION_TRANSPORT	0x5002
diff --git a/arch/xtensa/include/uapi/asm/socket.h b/arch/xtensa/include/uapi/asm/socket.h
index 4120af0..fd3b96d 100644
--- a/arch/xtensa/include/uapi/asm/socket.h
+++ b/arch/xtensa/include/uapi/asm/socket.h
@@ -96,4 +96,7 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif	/* _XTENSA_SOCKET_H */
diff --git a/include/linux/filter.h b/include/linux/filter.h
index 4165e9a..294c3cd 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -447,6 +447,8 @@ void bpf_prog_destroy(struct bpf_prog *fp);
 
 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
 int sk_attach_bpf(u32 ufd, struct sock *sk);
+int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
+int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
 int sk_detach_filter(struct sock *sk);
 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
 		  unsigned int len);
diff --git a/include/net/sock_reuseport.h b/include/net/sock_reuseport.h
index 67d1eb8..7dda3d7 100644
--- a/include/net/sock_reuseport.h
+++ b/include/net/sock_reuseport.h
@@ -1,6 +1,8 @@
 #ifndef _SOCK_REUSEPORT_H
 #define _SOCK_REUSEPORT_H
 
+#include <linux/filter.h>
+#include <linux/skbuff.h>
 #include <linux/types.h>
 #include <net/sock.h>
 
@@ -9,12 +11,18 @@ struct sock_reuseport {
 
 	u16			max_socks;	/* length of socks */
 	u16			num_socks;	/* elements in socks */
+	struct bpf_prog __rcu	*prog;		/* optional BPF sock selector */
 	struct sock		*socks[0];	/* array of sock pointers */
 };
 
 extern int reuseport_alloc(struct sock *sk);
 extern int reuseport_add_sock(struct sock *sk, const struct sock *sk2);
 extern void reuseport_detach_sock(struct sock *sk);
-extern struct sock *reuseport_select_sock(struct sock *sk, u32 hash);
+extern struct sock *reuseport_select_sock(struct sock *sk,
+					  u32 hash,
+					  struct sk_buff *skb,
+					  int hdr_len);
+extern struct bpf_prog *reuseport_attach_prog(struct sock *sk,
+					      struct bpf_prog *prog);
 
 #endif  /* _SOCK_REUSEPORT_H */
diff --git a/include/net/udp.h b/include/net/udp.h
index 3b5d7f9..2842541 100644
--- a/include/net/udp.h
+++ b/include/net/udp.h
@@ -258,7 +258,7 @@ struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 			     __be32 daddr, __be16 dport, int dif);
 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 			       __be32 daddr, __be16 dport, int dif,
-			       struct udp_table *tbl);
+			       struct udp_table *tbl, struct sk_buff *skb);
 struct sock *udp6_lib_lookup(struct net *net,
 			     const struct in6_addr *saddr, __be16 sport,
 			     const struct in6_addr *daddr, __be16 dport,
@@ -266,7 +266,8 @@ struct sock *udp6_lib_lookup(struct net *net,
 struct sock *__udp6_lib_lookup(struct net *net,
 			       const struct in6_addr *saddr, __be16 sport,
 			       const struct in6_addr *daddr, __be16 dport,
-			       int dif, struct udp_table *tbl);
+			       int dif, struct udp_table *tbl,
+			       struct sk_buff *skb);
 
 /*
  * 	SNMP statistics for UDP and UDP-Lite
diff --git a/include/uapi/asm-generic/socket.h b/include/uapi/asm-generic/socket.h
index 5c15c2a..fb8a416 100644
--- a/include/uapi/asm-generic/socket.h
+++ b/include/uapi/asm-generic/socket.h
@@ -87,4 +87,7 @@
 #define SO_ATTACH_BPF		50
 #define SO_DETACH_BPF		SO_DETACH_FILTER
 
+#define SO_ATTACH_REUSEPORT_CBPF	51
+#define SO_ATTACH_REUSEPORT_EBPF	52
+
 #endif /* __ASM_GENERIC_SOCKET_H */
diff --git a/net/core/filter.c b/net/core/filter.c
index c770196..35e6fed 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -50,6 +50,7 @@
 #include <net/cls_cgroup.h>
 #include <net/dst_metadata.h>
 #include <net/dst.h>
+#include <net/sock_reuseport.h>
 
 /**
  *	sk_filter - run a packet through a socket filter
@@ -1167,17 +1168,32 @@ static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
 	return 0;
 }
 
-/**
- *	sk_attach_filter - attach a socket filter
- *	@fprog: the filter program
- *	@sk: the socket to use
- *
- * Attach the user's filter code. We first run some sanity checks on
- * it to make sure it does not explode on us later. If an error
- * occurs or there is insufficient memory for the filter a negative
- * errno code is returned. On success the return is zero.
- */
-int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
+static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
+{
+	struct bpf_prog *old_prog;
+	int err;
+
+	if (bpf_prog_size(prog->len) > sysctl_optmem_max)
+		return -ENOMEM;
+
+	if (sk_unhashed(sk)) {
+		err = reuseport_alloc(sk);
+		if (err)
+			return err;
+	} else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
+		/* The socket wasn't bound with SO_REUSEPORT */
+		return -EINVAL;
+	}
+
+	old_prog = reuseport_attach_prog(sk, prog);
+	if (old_prog)
+		bpf_prog_destroy(old_prog);
+
+	return 0;
+}
+
+static
+struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
 {
 	unsigned int fsize = bpf_classic_proglen(fprog);
 	unsigned int bpf_fsize = bpf_prog_size(fprog->len);
@@ -1185,19 +1201,19 @@ int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
 	int err;
 
 	if (sock_flag(sk, SOCK_FILTER_LOCKED))
-		return -EPERM;
+		return ERR_PTR(-EPERM);
 
 	/* Make sure new filter is there and in the right amounts. */
 	if (fprog->filter == NULL)
-		return -EINVAL;
+		return ERR_PTR(-EINVAL);
 
 	prog = bpf_prog_alloc(bpf_fsize, 0);
 	if (!prog)
-		return -ENOMEM;
+		return ERR_PTR(-ENOMEM);
 
 	if (copy_from_user(prog->insns, fprog->filter, fsize)) {
 		__bpf_prog_free(prog);
-		return -EFAULT;
+		return ERR_PTR(-EFAULT);
 	}
 
 	prog->len = fprog->len;
@@ -1205,13 +1221,30 @@ int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
 	err = bpf_prog_store_orig_filter(prog, fprog);
 	if (err) {
 		__bpf_prog_free(prog);
-		return -ENOMEM;
+		return ERR_PTR(-ENOMEM);
 	}
 
 	/* bpf_prepare_filter() already takes care of freeing
 	 * memory in case something goes wrong.
 	 */
-	prog = bpf_prepare_filter(prog, NULL);
+	return bpf_prepare_filter(prog, NULL);
+}
+
+/**
+ *	sk_attach_filter - attach a socket filter
+ *	@fprog: the filter program
+ *	@sk: the socket to use
+ *
+ * Attach the user's filter code. We first run some sanity checks on
+ * it to make sure it does not explode on us later. If an error
+ * occurs or there is insufficient memory for the filter a negative
+ * errno code is returned. On success the return is zero.
+ */
+int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
+{
+	struct bpf_prog *prog = __get_filter(fprog, sk);
+	int err;
+
 	if (IS_ERR(prog))
 		return PTR_ERR(prog);
 
@@ -1225,23 +1258,50 @@ int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
 }
 EXPORT_SYMBOL_GPL(sk_attach_filter);
 
-int sk_attach_bpf(u32 ufd, struct sock *sk)
+int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
 {
-	struct bpf_prog *prog;
+	struct bpf_prog *prog = __get_filter(fprog, sk);
 	int err;
 
+	if (IS_ERR(prog))
+		return PTR_ERR(prog);
+
+	err = __reuseport_attach_prog(prog, sk);
+	if (err < 0) {
+		__bpf_prog_release(prog);
+		return err;
+	}
+
+	return 0;
+}
+
+static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
+{
+	struct bpf_prog *prog;
+
 	if (sock_flag(sk, SOCK_FILTER_LOCKED))
-		return -EPERM;
+		return ERR_PTR(-EPERM);
 
 	prog = bpf_prog_get(ufd);
 	if (IS_ERR(prog))
-		return PTR_ERR(prog);
+		return prog;
 
 	if (prog->type != BPF_PROG_TYPE_SOCKET_FILTER) {
 		bpf_prog_put(prog);
-		return -EINVAL;
+		return ERR_PTR(-EINVAL);
 	}
 
+	return prog;
+}
+
+int sk_attach_bpf(u32 ufd, struct sock *sk)
+{
+	struct bpf_prog *prog = __get_bpf(ufd, sk);
+	int err;
+
+	if (IS_ERR(prog))
+		return PTR_ERR(prog);
+
 	err = __sk_attach_prog(prog, sk);
 	if (err < 0) {
 		bpf_prog_put(prog);
@@ -1251,6 +1311,23 @@ int sk_attach_bpf(u32 ufd, struct sock *sk)
 	return 0;
 }
 
+int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
+{
+	struct bpf_prog *prog = __get_bpf(ufd, sk);
+	int err;
+
+	if (IS_ERR(prog))
+		return PTR_ERR(prog);
+
+	err = __reuseport_attach_prog(prog, sk);
+	if (err < 0) {
+		bpf_prog_put(prog);
+		return err;
+	}
+
+	return 0;
+}
+
 #define BPF_RECOMPUTE_CSUM(flags)	((flags) & 1)
 #define BPF_LDST_LEN			16U
 
diff --git a/net/core/sock.c b/net/core/sock.c
index 565bab7..5127023 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -134,6 +134,7 @@
 #include <linux/sock_diag.h>
 
 #include <linux/filter.h>
+#include <net/sock_reuseport.h>
 
 #include <trace/events/sock.h>
 
@@ -932,6 +933,32 @@ set_rcvbuf:
 		}
 		break;
 
+	case SO_ATTACH_REUSEPORT_CBPF:
+		ret = -EINVAL;
+		if (optlen == sizeof(struct sock_fprog)) {
+			struct sock_fprog fprog;
+
+			ret = -EFAULT;
+			if (copy_from_user(&fprog, optval, sizeof(fprog)))
+				break;
+
+			ret = sk_reuseport_attach_filter(&fprog, sk);
+		}
+		break;
+
+	case SO_ATTACH_REUSEPORT_EBPF:
+		ret = -EINVAL;
+		if (optlen == sizeof(u32)) {
+			u32 ufd;
+
+			ret = -EFAULT;
+			if (copy_from_user(&ufd, optval, sizeof(ufd)))
+				break;
+
+			ret = sk_reuseport_attach_bpf(ufd, sk);
+		}
+		break;
+
 	case SO_DETACH_FILTER:
 		ret = sk_detach_filter(sk);
 		break;
@@ -1443,6 +1470,8 @@ void sk_destruct(struct sock *sk)
 		sk_filter_uncharge(sk, filter);
 		RCU_INIT_POINTER(sk->sk_filter, NULL);
 	}
+	if (rcu_access_pointer(sk->sk_reuseport_cb))
+		reuseport_detach_sock(sk);
 
 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
 
diff --git a/net/core/sock_reuseport.c b/net/core/sock_reuseport.c
index 963c8d5..ae0969c 100644
--- a/net/core/sock_reuseport.c
+++ b/net/core/sock_reuseport.c
@@ -1,10 +1,12 @@
 /*
  * To speed up listener socket lookup, create an array to store all sockets
  * listening on the same port.  This allows a decision to be made after finding
- * the first socket.
+ * the first socket.  An optional BPF program can also be configured for
+ * selecting the socket index from the array of available sockets.
  */
 
 #include <net/sock_reuseport.h>
+#include <linux/bpf.h>
 #include <linux/rcupdate.h>
 
 #define INIT_SOCKS 128
@@ -22,6 +24,7 @@ static struct sock_reuseport *__reuseport_alloc(u16 max_socks)
 
 	reuse->max_socks = max_socks;
 
+	RCU_INIT_POINTER(reuse->prog, NULL);
 	return reuse;
 }
 
@@ -67,6 +70,7 @@ static struct sock_reuseport *reuseport_grow(struct sock_reuseport *reuse)
 
 	more_reuse->max_socks = more_socks_size;
 	more_reuse->num_socks = reuse->num_socks;
+	more_reuse->prog = reuse->prog;
 
 	memcpy(more_reuse->socks, reuse->socks,
 	       reuse->num_socks * sizeof(struct sock *));
@@ -75,6 +79,10 @@ static struct sock_reuseport *reuseport_grow(struct sock_reuseport *reuse)
 		rcu_assign_pointer(reuse->socks[i]->sk_reuseport_cb,
 				   more_reuse);
 
+	/* Note: we use kfree_rcu here instead of reuseport_free_rcu so
+	 * that reuse and more_reuse can temporarily share a reference
+	 * to prog.
+	 */
 	kfree_rcu(reuse, rcu);
 	return more_reuse;
 }
@@ -116,6 +124,16 @@ int reuseport_add_sock(struct sock *sk, const struct sock *sk2)
 }
 EXPORT_SYMBOL(reuseport_add_sock);
 
+static void reuseport_free_rcu(struct rcu_head *head)
+{
+	struct sock_reuseport *reuse;
+
+	reuse = container_of(head, struct sock_reuseport, rcu);
+	if (reuse->prog)
+		bpf_prog_destroy(reuse->prog);
+	kfree(reuse);
+}
+
 void reuseport_detach_sock(struct sock *sk)
 {
 	struct sock_reuseport *reuse;
@@ -131,7 +149,7 @@ void reuseport_detach_sock(struct sock *sk)
 			reuse->socks[i] = reuse->socks[reuse->num_socks - 1];
 			reuse->num_socks--;
 			if (reuse->num_socks == 0)
-				kfree_rcu(reuse, rcu);
+				call_rcu(&reuse->rcu, reuseport_free_rcu);
 			break;
 		}
 	}
@@ -139,15 +157,53 @@ void reuseport_detach_sock(struct sock *sk)
 }
 EXPORT_SYMBOL(reuseport_detach_sock);
 
+static struct sock *run_bpf(struct sock_reuseport *reuse, u16 socks,
+			    struct bpf_prog *prog, struct sk_buff *skb,
+			    int hdr_len)
+{
+	struct sk_buff *nskb = NULL;
+	u32 index;
+
+	if (skb_shared(skb)) {
+		nskb = skb_clone(skb, GFP_ATOMIC);
+		if (!nskb)
+			return NULL;
+		skb = nskb;
+	}
+
+	/* temporarily advance data past protocol header */
+	if (!pskb_pull(skb, hdr_len)) {
+		consume_skb(nskb);
+		return NULL;
+	}
+	index = bpf_prog_run_save_cb(prog, skb);
+	__skb_push(skb, hdr_len);
+
+	consume_skb(nskb);
+
+	if (index >= socks)
+		return NULL;
+
+	return reuse->socks[index];
+}
+
 /**
  *  reuseport_select_sock - Select a socket from an SO_REUSEPORT group.
  *  @sk: First socket in the group.
- *  @hash: Use this hash to select.
+ *  @hash: When no BPF filter is available, use this hash to select.
+ *  @skb: skb to run through BPF filter.
+ *  @hdr_len: BPF filter expects skb data pointer at payload data.  If
+ *    the skb does not yet point at the payload, this parameter represents
+ *    how far the pointer needs to advance to reach the payload.
  *  Returns a socket that should receive the packet (or NULL on error).
  */
-struct sock *reuseport_select_sock(struct sock *sk, u32 hash)
+struct sock *reuseport_select_sock(struct sock *sk,
+				   u32 hash,
+				   struct sk_buff *skb,
+				   int hdr_len)
 {
 	struct sock_reuseport *reuse;
+	struct bpf_prog *prog;
 	struct sock *sk2 = NULL;
 	u16 socks;
 
@@ -158,12 +214,16 @@ struct sock *reuseport_select_sock(struct sock *sk, u32 hash)
 	if (!reuse)
 		goto out;
 
+	prog = rcu_dereference(reuse->prog);
 	socks = READ_ONCE(reuse->num_socks);
 	if (likely(socks)) {
 		/* paired with smp_wmb() in reuseport_add_sock() */
 		smp_rmb();
 
-		sk2 = reuse->socks[reciprocal_scale(hash, socks)];
+		if (prog && skb)
+			sk2 = run_bpf(reuse, socks, prog, skb, hdr_len);
+		else
+			sk2 = reuse->socks[reciprocal_scale(hash, socks)];
 	}
 
 out:
@@ -171,3 +231,21 @@ out:
 	return sk2;
 }
 EXPORT_SYMBOL(reuseport_select_sock);
+
+struct bpf_prog *
+reuseport_attach_prog(struct sock *sk, struct bpf_prog *prog)
+{
+	struct sock_reuseport *reuse;
+	struct bpf_prog *old_prog;
+
+	spin_lock_bh(&reuseport_lock);
+	reuse = rcu_dereference_protected(sk->sk_reuseport_cb,
+					  lockdep_is_held(&reuseport_lock));
+	old_prog = rcu_dereference_protected(reuse->prog,
+					     lockdep_is_held(&reuseport_lock));
+	rcu_assign_pointer(reuse->prog, prog);
+	spin_unlock_bh(&reuseport_lock);
+
+	return old_prog;
+}
+EXPORT_SYMBOL(reuseport_attach_prog);
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index 762b01f..8353783 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -514,7 +514,7 @@ begin:
 				struct sock *sk2;
 				hash = udp_ehashfn(net, daddr, hnum,
 						   saddr, sport);
-				sk2 = reuseport_select_sock(sk, hash);
+				sk2 = reuseport_select_sock(sk, hash, NULL, 0);
 				if (sk2) {
 					result = sk2;
 					goto found;
@@ -553,7 +553,7 @@ found:
  */
 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 		__be16 sport, __be32 daddr, __be16 dport,
-		int dif, struct udp_table *udptable)
+		int dif, struct udp_table *udptable, struct sk_buff *skb)
 {
 	struct sock *sk, *result;
 	struct hlist_nulls_node *node;
@@ -602,7 +602,8 @@ begin:
 				struct sock *sk2;
 				hash = udp_ehashfn(net, daddr, hnum,
 						   saddr, sport);
-				sk2 = reuseport_select_sock(sk, hash);
+				sk2 = reuseport_select_sock(sk, hash, skb,
+							sizeof(struct udphdr));
 				if (sk2) {
 					result = sk2;
 					goto found;
@@ -647,14 +648,14 @@ static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 
 	return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
 				 iph->daddr, dport, inet_iif(skb),
-				 udptable);
+				 udptable, skb);
 }
 
 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 			     __be32 daddr, __be16 dport, int dif)
 {
 	return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif,
-				 &udp_table);
+				 &udp_table, NULL);
 }
 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 
@@ -702,7 +703,8 @@ void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 	struct net *net = dev_net(skb->dev);
 
 	sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
-			iph->saddr, uh->source, skb->dev->ifindex, udptable);
+			iph->saddr, uh->source, skb->dev->ifindex, udptable,
+			NULL);
 	if (!sk) {
 		ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 		return;	/* No socket for error */
diff --git a/net/ipv4/udp_diag.c b/net/ipv4/udp_diag.c
index 6116604..df1966f 100644
--- a/net/ipv4/udp_diag.c
+++ b/net/ipv4/udp_diag.c
@@ -44,7 +44,7 @@ static int udp_dump_one(struct udp_table *tbl, struct sk_buff *in_skb,
 		sk = __udp4_lib_lookup(net,
 				req->id.idiag_src[0], req->id.idiag_sport,
 				req->id.idiag_dst[0], req->id.idiag_dport,
-				req->id.idiag_if, tbl);
+				req->id.idiag_if, tbl, NULL);
 #if IS_ENABLED(CONFIG_IPV6)
 	else if (req->sdiag_family == AF_INET6)
 		sk = __udp6_lib_lookup(net,
@@ -52,7 +52,7 @@ static int udp_dump_one(struct udp_table *tbl, struct sk_buff *in_skb,
 				req->id.idiag_sport,
 				(struct in6_addr *)req->id.idiag_dst,
 				req->id.idiag_dport,
-				req->id.idiag_if, tbl);
+				req->id.idiag_if, tbl, NULL);
 #endif
 	else
 		goto out_nosk;
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 6204b89..56fcb55 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -272,7 +272,7 @@ begin:
 				struct sock *sk2;
 				hash = udp6_ehashfn(net, daddr, hnum,
 						    saddr, sport);
-				sk2 = reuseport_select_sock(sk, hash);
+				sk2 = reuseport_select_sock(sk, hash, NULL, 0);
 				if (sk2) {
 					result = sk2;
 					goto found;
@@ -310,7 +310,8 @@ found:
 struct sock *__udp6_lib_lookup(struct net *net,
 				      const struct in6_addr *saddr, __be16 sport,
 				      const struct in6_addr *daddr, __be16 dport,
-				      int dif, struct udp_table *udptable)
+				      int dif, struct udp_table *udptable,
+				      struct sk_buff *skb)
 {
 	struct sock *sk, *result;
 	struct hlist_nulls_node *node;
@@ -358,7 +359,8 @@ begin:
 				struct sock *sk2;
 				hash = udp6_ehashfn(net, daddr, hnum,
 						    saddr, sport);
-				sk2 = reuseport_select_sock(sk, hash);
+				sk2 = reuseport_select_sock(sk, hash, skb,
+							sizeof(struct udphdr));
 				if (sk2) {
 					result = sk2;
 					goto found;
@@ -407,13 +409,13 @@ static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
 		return sk;
 	return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
 				 &iph->daddr, dport, inet6_iif(skb),
-				 udptable);
+				 udptable, skb);
 }
 
 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
 			     const struct in6_addr *daddr, __be16 dport, int dif)
 {
-	return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
+	return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table, NULL);
 }
 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
 
@@ -580,7 +582,7 @@ void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 	struct net *net = dev_net(skb->dev);
 
 	sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
-			       inet6_iif(skb), udptable);
+			       inet6_iif(skb), udptable, skb);
 	if (!sk) {
 		ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 				   ICMP6_MIB_INERRORS);
-- 
2.6.0.rc2.230.g3dd15c0

Re: [PATCH v4 net-next 3/4] soreuseport: setsockopt SO_ATTACH_REUSEPORT_[CE]BPF

[Alexei Starovoitov]

On Mon, Jan 04, 2016 at 05:41:47PM -0500, Craig Gallek wrote:
> From: Craig Gallek <kraig@google.com>
> 
> Expose socket options for setting a classic or extended BPF program
> for use when selecting sockets in an SO_REUSEPORT group.  These options
> can be used on the first socket to belong to a group before bind or
> on any socket in the group after bind.
> 
> This change includes refactoring of the existing sk_filter code to
> allow reuse of the existing BPF filter validation checks.
> 
> Signed-off-by: Craig Gallek <kraig@google.com>

looks good to me. Thanks!
Acked-by: Alexei Starovoitov <ast@kernel.org>

Re: [PATCH v4 net-next 3/4] soreuseport: setsockopt SO_ATTACH_REUSEPORT_[CE]BPF

[Daniel Borkmann]

On 01/04/2016 11:41 PM, Craig Gallek wrote:
> From: Craig Gallek <kraig@google.com>
>
> Expose socket options for setting a classic or extended BPF program
> for use when selecting sockets in an SO_REUSEPORT group.  These options
> can be used on the first socket to belong to a group before bind or
> on any socket in the group after bind.
>
> This change includes refactoring of the existing sk_filter code to
> allow reuse of the existing BPF filter validation checks.
>
> Signed-off-by: Craig Gallek <kraig@google.com>
[...]
> +static struct sock *run_bpf(struct sock_reuseport *reuse, u16 socks,
> +			    struct bpf_prog *prog, struct sk_buff *skb,
> +			    int hdr_len)
> +{
> +	struct sk_buff *nskb = NULL;
> +	u32 index;
> +
> +	if (skb_shared(skb)) {
> +		nskb = skb_clone(skb, GFP_ATOMIC);
> +		if (!nskb)
> +			return NULL;
> +		skb = nskb;
> +	}
> +
> +	/* temporarily advance data past protocol header */
> +	if (!pskb_pull(skb, hdr_len)) {
> +		consume_skb(nskb);

Btw, this one could still be made kfree_skb() to indicate error condition here.

> +		return NULL;
> +	}
> +	index = bpf_prog_run_save_cb(prog, skb);
> +	__skb_push(skb, hdr_len);
> +
> +	consume_skb(nskb);
> +
> +	if (index >= socks)
> +		return NULL;
> +
> +	return reuse->socks[index];
> +}

Re: [PATCH v4 net-next 3/4] soreuseport: setsockopt SO_ATTACH_REUSEPORT_[CE]BPF

[Craig Gallek]

On Tue, Jan 5, 2016 at 4:38 AM, Daniel Borkmann <daniel@iogearbox.net> wrote:
> On 01/04/2016 11:41 PM, Craig Gallek wrote:
>>
>> From: Craig Gallek <kraig@google.com>
>>
>> Expose socket options for setting a classic or extended BPF program
>> for use when selecting sockets in an SO_REUSEPORT group.  These options
>> can be used on the first socket to belong to a group before bind or
>> on any socket in the group after bind.
>>
>> This change includes refactoring of the existing sk_filter code to
>> allow reuse of the existing BPF filter validation checks.
>>
>> Signed-off-by: Craig Gallek <kraig@google.com>
>
> [...]
>>
>> +static struct sock *run_bpf(struct sock_reuseport *reuse, u16 socks,
>> +                           struct bpf_prog *prog, struct sk_buff *skb,
>> +                           int hdr_len)
>> +{
>> +       struct sk_buff *nskb = NULL;
>> +       u32 index;
>> +
>> +       if (skb_shared(skb)) {
>> +               nskb = skb_clone(skb, GFP_ATOMIC);
>> +               if (!nskb)
>> +                       return NULL;
>> +               skb = nskb;
>> +       }
>> +
>> +       /* temporarily advance data past protocol header */
>> +       if (!pskb_pull(skb, hdr_len)) {
>> +               consume_skb(nskb);
>
>
> Btw, this one could still be made kfree_skb() to indicate error condition
> here.
Good point.  I'll send a follow-up.

[PATCH v4 net-next 4/4] soreuseport: BPF selection functional test

[Craig Gallek]

From: Craig Gallek <kraig@google.com>

This program will build classic and extended BPF programs and
validate the socket selection logic when used with
SO_ATTACH_REUSEPORT_CBPF and SO_ATTACH_REUSEPORT_EBPF.

It also validates the re-programing flow and several edge cases.

Signed-off-by: Craig Gallek <kraig@google.com>
---
 tools/testing/selftests/net/.gitignore      |   1 +
 tools/testing/selftests/net/Makefile        |   2 +-
 tools/testing/selftests/net/reuseport_bpf.c | 467 ++++++++++++++++++++++++++++
 3 files changed, 469 insertions(+), 1 deletion(-)
 create mode 100644 tools/testing/selftests/net/reuseport_bpf.c

diff --git a/tools/testing/selftests/net/.gitignore b/tools/testing/selftests/net/.gitignore
index 0032662..6fb2336 100644
--- a/tools/testing/selftests/net/.gitignore
+++ b/tools/testing/selftests/net/.gitignore
@@ -1,3 +1,4 @@
 socket
 psock_fanout
 psock_tpacket
+reuseport_bpf
diff --git a/tools/testing/selftests/net/Makefile b/tools/testing/selftests/net/Makefile
index fac4782..41449b5 100644
--- a/tools/testing/selftests/net/Makefile
+++ b/tools/testing/selftests/net/Makefile
@@ -4,7 +4,7 @@ CFLAGS = -Wall -O2 -g
 
 CFLAGS += -I../../../../usr/include/
 
-NET_PROGS = socket psock_fanout psock_tpacket
+NET_PROGS = socket psock_fanout psock_tpacket reuseport_bpf
 
 all: $(NET_PROGS)
 %: %.c
diff --git a/tools/testing/selftests/net/reuseport_bpf.c b/tools/testing/selftests/net/reuseport_bpf.c
new file mode 100644
index 0000000..74ff099
--- /dev/null
+++ b/tools/testing/selftests/net/reuseport_bpf.c
@@ -0,0 +1,467 @@
+/*
+ * Test functionality of BPF filters for SO_REUSEPORT.  The tests below will use
+ * a BPF program (both classic and extended) to read the first word from an
+ * incoming packet (expected to be in network byte-order), calculate a modulus
+ * of that number, and then dispatch the packet to the Nth socket using the
+ * result.  These tests are run for each supported address family and protocol.
+ * Additionally, a few edge cases in the implementation are tested.
+ */
+
+#include <errno.h>
+#include <error.h>
+#include <linux/bpf.h>
+#include <linux/filter.h>
+#include <linux/unistd.h>
+#include <netinet/in.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/epoll.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <unistd.h>
+
+#ifndef ARRAY_SIZE
+#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
+#endif
+
+struct test_params {
+	int recv_family;
+	int send_family;
+	int protocol;
+	size_t recv_socks;
+	uint16_t recv_port;
+	uint16_t send_port_min;
+};
+
+static size_t sockaddr_size(void)
+{
+	return sizeof(struct sockaddr_storage);
+}
+
+static struct sockaddr *new_any_sockaddr(int family, uint16_t port)
+{
+	struct sockaddr_storage *addr;
+	struct sockaddr_in *addr4;
+	struct sockaddr_in6 *addr6;
+
+	addr = malloc(sizeof(struct sockaddr_storage));
+	memset(addr, 0, sizeof(struct sockaddr_storage));
+
+	switch (family) {
+	case AF_INET:
+		addr4 = (struct sockaddr_in *)addr;
+		addr4->sin_family = AF_INET;
+		addr4->sin_addr.s_addr = htonl(INADDR_ANY);
+		addr4->sin_port = htons(port);
+		break;
+	case AF_INET6:
+		addr6 = (struct sockaddr_in6 *)addr;
+		addr6->sin6_family = AF_INET6;
+		addr6->sin6_addr = in6addr_any;
+		addr6->sin6_port = htons(port);
+		break;
+	default:
+		error(1, 0, "Unsupported family %d", family);
+	}
+	return (struct sockaddr *)addr;
+}
+
+static struct sockaddr *new_loopback_sockaddr(int family, uint16_t port)
+{
+	struct sockaddr *addr = new_any_sockaddr(family, port);
+	struct sockaddr_in *addr4;
+	struct sockaddr_in6 *addr6;
+
+	switch (family) {
+	case AF_INET:
+		addr4 = (struct sockaddr_in *)addr;
+		addr4->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
+		break;
+	case AF_INET6:
+		addr6 = (struct sockaddr_in6 *)addr;
+		addr6->sin6_addr = in6addr_loopback;
+		break;
+	default:
+		error(1, 0, "Unsupported family %d", family);
+	}
+	return addr;
+}
+
+static void attach_ebpf(int fd, uint16_t mod)
+{
+	static char bpf_log_buf[65536];
+	static const char bpf_license[] = "GPL";
+
+	int bpf_fd;
+	const struct bpf_insn prog[] = {
+		/* BPF_MOV64_REG(BPF_REG_6, BPF_REG_1) */
+		{ BPF_ALU64 | BPF_MOV | BPF_X, BPF_REG_6, BPF_REG_1, 0, 0 },
+		/* BPF_LD_ABS(BPF_W, 0) R0 = (uint32_t)skb[0] */
+		{ BPF_LD | BPF_ABS | BPF_W, 0, 0, 0, 0 },
+		/* BPF_ALU64_IMM(BPF_MOD, BPF_REG_0, mod) */
+		{ BPF_ALU64 | BPF_MOD | BPF_K, BPF_REG_0, 0, 0, mod },
+		/* BPF_EXIT_INSN() */
+		{ BPF_JMP | BPF_EXIT, 0, 0, 0, 0 }
+	};
+	union bpf_attr attr;
+
+	memset(&attr, 0, sizeof(attr));
+	attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
+	attr.insn_cnt = ARRAY_SIZE(prog);
+	attr.insns = (uint64_t)prog;
+	attr.license = (uint64_t)bpf_license;
+	attr.log_buf = (uint64_t)bpf_log_buf;
+	attr.log_size = sizeof(bpf_log_buf);
+	attr.log_level = 1;
+	attr.kern_version = 0;
+
+	bpf_fd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr));
+	if (bpf_fd < 0)
+		error(1, errno, "ebpf error. log:\n%s\n", bpf_log_buf);
+
+	if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_REUSEPORT_EBPF, &bpf_fd,
+			sizeof(bpf_fd)))
+		error(1, errno, "failed to set SO_ATTACH_REUSEPORT_EBPF");
+}
+
+static void attach_cbpf(int fd, uint16_t mod)
+{
+	struct sock_filter code[] = {
+		/* A = (uint32_t)skb[0] */
+		{ BPF_LD  | BPF_W | BPF_ABS, 0, 0, 0 },
+		/* A = A % mod */
+		{ BPF_ALU | BPF_MOD, 0, 0, mod },
+		/* return A */
+		{ BPF_RET | BPF_A, 0, 0, 0 },
+	};
+	struct sock_fprog p = {
+		.len = ARRAY_SIZE(code),
+		.filter = code,
+	};
+
+	if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_REUSEPORT_CBPF, &p, sizeof(p)))
+		error(1, errno, "failed to set SO_ATTACH_REUSEPORT_CBPF");
+}
+
+static void build_recv_group(const struct test_params p, int fd[], uint16_t mod,
+			     void (*attach_bpf)(int, uint16_t))
+{
+	struct sockaddr * const addr =
+		new_any_sockaddr(p.recv_family, p.recv_port);
+	int i, opt;
+
+	for (i = 0; i < p.recv_socks; ++i) {
+		fd[i] = socket(p.recv_family, p.protocol, 0);
+		if (fd[i] < 0)
+			error(1, errno, "failed to create recv %d", i);
+
+		opt = 1;
+		if (setsockopt(fd[i], SOL_SOCKET, SO_REUSEPORT, &opt,
+			       sizeof(opt)))
+			error(1, errno, "failed to set SO_REUSEPORT on %d", i);
+
+		if (i == 0)
+			attach_bpf(fd[i], mod);
+
+		if (bind(fd[i], addr, sockaddr_size()))
+			error(1, errno, "failed to bind recv socket %d", i);
+
+		if (p.protocol == SOCK_STREAM)
+			if (listen(fd[i], p.recv_socks * 10))
+				error(1, errno, "failed to listen on socket");
+	}
+	free(addr);
+}
+
+static void send_from(struct test_params p, uint16_t sport, char *buf,
+		      size_t len)
+{
+	struct sockaddr * const saddr = new_any_sockaddr(p.send_family, sport);
+	struct sockaddr * const daddr =
+		new_loopback_sockaddr(p.send_family, p.recv_port);
+	const int fd = socket(p.send_family, p.protocol, 0);
+
+	if (fd < 0)
+		error(1, errno, "failed to create send socket");
+
+	if (bind(fd, saddr, sockaddr_size()))
+		error(1, errno, "failed to bind send socket");
+	if (connect(fd, daddr, sockaddr_size()))
+		error(1, errno, "failed to connect");
+
+	if (send(fd, buf, len, 0) < 0)
+		error(1, errno, "failed to send message");
+
+	close(fd);
+	free(saddr);
+	free(daddr);
+}
+
+static void test_recv_order(const struct test_params p, int fd[], int mod)
+{
+	char recv_buf[8], send_buf[8];
+	struct msghdr msg;
+	struct iovec recv_io = { recv_buf, 8 };
+	struct epoll_event ev;
+	int epfd, conn, i, sport, expected;
+	uint32_t data, ndata;
+
+	epfd = epoll_create(1);
+	if (epfd < 0)
+		error(1, errno, "failed to create epoll");
+	for (i = 0; i < p.recv_socks; ++i) {
+		ev.events = EPOLLIN;
+		ev.data.fd = fd[i];
+		if (epoll_ctl(epfd, EPOLL_CTL_ADD, fd[i], &ev))
+			error(1, errno, "failed to register sock %d epoll", i);
+	}
+
+	memset(&msg, 0, sizeof(msg));
+	msg.msg_iov = &recv_io;
+	msg.msg_iovlen = 1;
+
+	for (data = 0; data < p.recv_socks * 2; ++data) {
+		sport = p.send_port_min + data;
+		ndata = htonl(data);
+		memcpy(send_buf, &ndata, sizeof(ndata));
+		send_from(p, sport, send_buf, sizeof(ndata));
+
+		i = epoll_wait(epfd, &ev, 1, -1);
+		if (i < 0)
+			error(1, errno, "epoll wait failed");
+
+		if (p.protocol == SOCK_STREAM) {
+			conn = accept(ev.data.fd, NULL, NULL);
+			if (conn < 0)
+				error(1, errno, "error accepting");
+			i = recvmsg(conn, &msg, 0);
+			close(conn);
+		} else {
+			i = recvmsg(ev.data.fd, &msg, 0);
+		}
+		if (i < 0)
+			error(1, errno, "recvmsg error");
+		if (i != sizeof(ndata))
+			error(1, 0, "expected size %zd got %d",
+			      sizeof(ndata), i);
+
+		for (i = 0; i < p.recv_socks; ++i)
+			if (ev.data.fd == fd[i])
+				break;
+		memcpy(&ndata, recv_buf, sizeof(ndata));
+		fprintf(stderr, "Socket %d: %d\n", i, ntohl(ndata));
+
+		expected = (sport % mod);
+		if (i != expected)
+			error(1, 0, "expected socket %d", expected);
+	}
+}
+
+static void test_reuseport_ebpf(const struct test_params p)
+{
+	int i, fd[p.recv_socks];
+
+	fprintf(stderr, "Testing EBPF mod %zd...\n", p.recv_socks);
+	build_recv_group(p, fd, p.recv_socks, attach_ebpf);
+	test_recv_order(p, fd, p.recv_socks);
+
+	fprintf(stderr, "Reprograming, testing mod %zd...\n", p.recv_socks / 2);
+	attach_ebpf(fd[0], p.recv_socks / 2);
+	test_recv_order(p, fd, p.recv_socks / 2);
+
+	for (i = 0; i < p.recv_socks; ++i)
+		close(fd[i]);
+}
+
+static void test_reuseport_cbpf(const struct test_params p)
+{
+	int i, fd[p.recv_socks];
+
+	fprintf(stderr, "Testing CBPF mod %zd...\n", p.recv_socks);
+	build_recv_group(p, fd, p.recv_socks, attach_cbpf);
+	test_recv_order(p, fd, p.recv_socks);
+
+	fprintf(stderr, "Reprograming, testing mod %zd...\n", p.recv_socks / 2);
+	attach_cbpf(fd[0], p.recv_socks / 2);
+	test_recv_order(p, fd, p.recv_socks / 2);
+
+	for (i = 0; i < p.recv_socks; ++i)
+		close(fd[i]);
+}
+
+static void test_extra_filter(const struct test_params p)
+{
+	struct sockaddr * const addr =
+		new_any_sockaddr(p.recv_family, p.recv_port);
+	int fd1, fd2, opt;
+
+	fprintf(stderr, "Testing too many filters...\n");
+	fd1 = socket(p.recv_family, p.protocol, 0);
+	if (fd1 < 0)
+		error(1, errno, "failed to create socket 1");
+	fd2 = socket(p.recv_family, p.protocol, 0);
+	if (fd2 < 0)
+		error(1, errno, "failed to create socket 2");
+
+	opt = 1;
+	if (setsockopt(fd1, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)))
+		error(1, errno, "failed to set SO_REUSEPORT on socket 1");
+	if (setsockopt(fd2, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)))
+		error(1, errno, "failed to set SO_REUSEPORT on socket 2");
+
+	attach_ebpf(fd1, 10);
+	attach_ebpf(fd2, 10);
+
+	if (bind(fd1, addr, sockaddr_size()))
+		error(1, errno, "failed to bind recv socket 1");
+
+	if (!bind(fd2, addr, sockaddr_size()) && errno != EADDRINUSE)
+		error(1, errno, "bind socket 2 should fail with EADDRINUSE");
+
+	free(addr);
+}
+
+static void test_filter_no_reuseport(const struct test_params p)
+{
+	struct sockaddr * const addr =
+		new_any_sockaddr(p.recv_family, p.recv_port);
+	const char bpf_license[] = "GPL";
+	struct bpf_insn ecode[] = {
+		{ BPF_ALU64 | BPF_MOV | BPF_K, BPF_REG_0, 0, 0, 10 },
+		{ BPF_JMP | BPF_EXIT, 0, 0, 0, 0 }
+	};
+	struct sock_filter ccode[] = {{ BPF_RET | BPF_A, 0, 0, 0 }};
+	union bpf_attr eprog;
+	struct sock_fprog cprog;
+	int fd, bpf_fd;
+
+	fprintf(stderr, "Testing filters on non-SO_REUSEPORT socket...\n");
+
+	memset(&eprog, 0, sizeof(eprog));
+	eprog.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
+	eprog.insn_cnt = ARRAY_SIZE(ecode);
+	eprog.insns = (uint64_t)ecode;
+	eprog.license = (uint64_t)bpf_license;
+	eprog.kern_version = 0;
+
+	memset(&cprog, 0, sizeof(cprog));
+	cprog.len = ARRAY_SIZE(ccode);
+	cprog.filter = ccode;
+
+
+	bpf_fd = syscall(__NR_bpf, BPF_PROG_LOAD, &eprog, sizeof(eprog));
+	if (bpf_fd < 0)
+		error(1, errno, "ebpf error");
+	fd = socket(p.recv_family, p.protocol, 0);
+	if (fd < 0)
+		error(1, errno, "failed to create socket 1");
+
+	if (bind(fd, addr, sockaddr_size()))
+		error(1, errno, "failed to bind recv socket 1");
+
+	errno = 0;
+	if (!setsockopt(fd, SOL_SOCKET, SO_ATTACH_REUSEPORT_EBPF, &bpf_fd,
+			sizeof(bpf_fd)) || errno != EINVAL)
+		error(1, errno, "setsockopt should have returned EINVAL");
+
+	errno = 0;
+	if (!setsockopt(fd, SOL_SOCKET, SO_ATTACH_REUSEPORT_CBPF, &cprog,
+		       sizeof(cprog)) || errno != EINVAL)
+		error(1, errno, "setsockopt should have returned EINVAL");
+
+	free(addr);
+}
+
+static void test_filter_without_bind(void)
+{
+	int fd1, fd2;
+
+	fprintf(stderr, "Testing filter add without bind...\n");
+	fd1 = socket(AF_INET, SOCK_DGRAM, 0);
+	if (fd1 < 0)
+		error(1, errno, "failed to create socket 1");
+	fd2 = socket(AF_INET, SOCK_DGRAM, 0);
+	if (fd2 < 0)
+		error(1, errno, "failed to create socket 2");
+
+	attach_ebpf(fd1, 10);
+	attach_cbpf(fd2, 10);
+
+	close(fd1);
+	close(fd2);
+}
+
+
+int main(void)
+{
+	fprintf(stderr, "---- IPv4 UDP ----\n");
+	test_reuseport_ebpf((struct test_params) {
+		.recv_family = AF_INET,
+		.send_family = AF_INET,
+		.protocol = SOCK_DGRAM,
+		.recv_socks = 10,
+		.recv_port = 8000,
+		.send_port_min = 9000});
+	test_reuseport_cbpf((struct test_params) {
+		.recv_family = AF_INET,
+		.send_family = AF_INET,
+		.protocol = SOCK_DGRAM,
+		.recv_socks = 10,
+		.recv_port = 8001,
+		.send_port_min = 9020});
+	test_extra_filter((struct test_params) {
+		.recv_family = AF_INET,
+		.protocol = SOCK_DGRAM,
+		.recv_port = 8002});
+	test_filter_no_reuseport((struct test_params) {
+		.recv_family = AF_INET,
+		.protocol = SOCK_DGRAM,
+		.recv_port = 8008});
+
+	fprintf(stderr, "---- IPv6 UDP ----\n");
+	test_reuseport_ebpf((struct test_params) {
+		.recv_family = AF_INET6,
+		.send_family = AF_INET6,
+		.protocol = SOCK_DGRAM,
+		.recv_socks = 10,
+		.recv_port = 8003,
+		.send_port_min = 9040});
+	test_reuseport_cbpf((struct test_params) {
+		.recv_family = AF_INET6,
+		.send_family = AF_INET6,
+		.protocol = SOCK_DGRAM,
+		.recv_socks = 10,
+		.recv_port = 8004,
+		.send_port_min = 9060});
+	test_extra_filter((struct test_params) {
+		.recv_family = AF_INET6,
+		.protocol = SOCK_DGRAM,
+		.recv_port = 8005});
+	test_filter_no_reuseport((struct test_params) {
+		.recv_family = AF_INET6,
+		.protocol = SOCK_DGRAM,
+		.recv_port = 8009});
+
+	fprintf(stderr, "---- IPv6 UDP w/ mapped IPv4 ----\n");
+	test_reuseport_ebpf((struct test_params) {
+		.recv_family = AF_INET6,
+		.send_family = AF_INET,
+		.protocol = SOCK_DGRAM,
+		.recv_socks = 10,
+		.recv_port = 8006,
+		.send_port_min = 9080});
+	test_reuseport_cbpf((struct test_params) {
+		.recv_family = AF_INET6,
+		.send_family = AF_INET,
+		.protocol = SOCK_DGRAM,
+		.recv_socks = 10,
+		.recv_port = 8007,
+		.send_port_min = 9100});
+
+
+	test_filter_without_bind();
+
+	fprintf(stderr, "SUCCESS\n");
+	return 0;
+}
-- 
2.6.0.rc2.230.g3dd15c0

Re: [PATCH v4 net-next 4/4] soreuseport: BPF selection functional test

[Alexei Starovoitov]

On Mon, Jan 04, 2016 at 05:41:48PM -0500, Craig Gallek wrote:
> From: Craig Gallek <kraig@google.com>
> 
> This program will build classic and extended BPF programs and
> validate the socket selection logic when used with
> SO_ATTACH_REUSEPORT_CBPF and SO_ATTACH_REUSEPORT_EBPF.
> 
> It also validates the re-programing flow and several edge cases.
> 
> Signed-off-by: Craig Gallek <kraig@google.com>

Acked-by: Alexei Starovoitov <ast@kernel.org>

Re: [PATCH v4 net-next 0/4] Faster SO_REUSEPORT

[David Miller]

From: Craig Gallek <kraigatgoog@gmail.com>
Date: Mon,  4 Jan 2016 17:41:44 -0500

> This series contains two optimizations for the SO_REUSEPORT feature:
> Faster lookup when selecting a socket for an incoming packet and
> the ability to select the socket from the group using a BPF program.
> 
> This series only includes the UDP path.  I plan to submit a follow-up
> including the TCP path if the implementation in this series is
> acceptable.

This series looks great!

Applied, thanks Craig.