soreuseport: pass skb to secondary UDP socket lookup
authorCraig Gallek <kraig@google.com>
Tue, 5 Jan 2016 20:08:07 +0000 (15:08 -0500)
committerDavid S. Miller <davem@davemloft.net>
Wed, 6 Jan 2016 06:28:04 +0000 (01:28 -0500)
This socket-lookup path did not pass along the skb in question
in my original BPF-based socket selection patch.  The skb in the
udpN_lib_lookup2 path can be used for BPF-based socket selection just
like it is in the 'traditional' udpN_lib_lookup path.

udpN_lib_lookup2 kicks in when there are greater than 10 sockets in
the same hlist slot.  Coincidentally, I chose 10 sockets per
reuseport group in my functional test, so the lookup2 path was not
excersised. This adds an additional set of tests with 20 sockets.

Fixes: 538950a1b752 ("soreuseport: setsockopt SO_ATTACH_REUSEPORT_[CE]BPF")
Fixes: 3ca8e4029969 ("soreuseport: BPF selection functional test")
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Craig Gallek <kraig@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/udp.c
net/ipv6/udp.c
tools/testing/selftests/net/reuseport_bpf.c

index 835378365f259811c471d04acf2bb2a8466d0ec9..3a66731e3af6d083f3b3b2fecc25bd09d183aab7 100644 (file)
@@ -493,7 +493,8 @@ static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
 static struct sock *udp4_lib_lookup2(struct net *net,
                __be32 saddr, __be16 sport,
                __be32 daddr, unsigned int hnum, int dif,
-               struct udp_hslot *hslot2, unsigned int slot2)
+               struct udp_hslot *hslot2, unsigned int slot2,
+               struct sk_buff *skb)
 {
        struct sock *sk, *result;
        struct hlist_nulls_node *node;
@@ -514,7 +515,8 @@ begin:
                                struct sock *sk2;
                                hash = udp_ehashfn(net, daddr, hnum,
                                                   saddr, sport);
-                               sk2 = reuseport_select_sock(sk, hash, NULL, 0);
+                               sk2 = reuseport_select_sock(sk, hash, skb,
+                                                           sizeof(struct udphdr));
                                if (sk2) {
                                        result = sk2;
                                        goto found;
@@ -573,7 +575,7 @@ struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 
                result = udp4_lib_lookup2(net, saddr, sport,
                                          daddr, hnum, dif,
-                                         hslot2, slot2);
+                                         hslot2, slot2, skb);
                if (!result) {
                        hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
                        slot2 = hash2 & udptable->mask;
@@ -583,7 +585,7 @@ struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 
                        result = udp4_lib_lookup2(net, saddr, sport,
                                                  htonl(INADDR_ANY), hnum, dif,
-                                                 hslot2, slot2);
+                                                 hslot2, slot2, skb);
                }
                rcu_read_unlock();
                return result;
index 56fcb55fda31a8c15e23ed6051f1ab87e121d13d..5d2c2afffe7b2c0790bc00505c518d21eb4b589b 100644 (file)
@@ -251,7 +251,8 @@ static inline int compute_score2(struct sock *sk, struct net *net,
 static struct sock *udp6_lib_lookup2(struct net *net,
                const struct in6_addr *saddr, __be16 sport,
                const struct in6_addr *daddr, unsigned int hnum, int dif,
-               struct udp_hslot *hslot2, unsigned int slot2)
+               struct udp_hslot *hslot2, unsigned int slot2,
+               struct sk_buff *skb)
 {
        struct sock *sk, *result;
        struct hlist_nulls_node *node;
@@ -272,7 +273,8 @@ begin:
                                struct sock *sk2;
                                hash = udp6_ehashfn(net, daddr, hnum,
                                                    saddr, sport);
-                               sk2 = reuseport_select_sock(sk, hash, NULL, 0);
+                               sk2 = reuseport_select_sock(sk, hash, skb,
+                                                           sizeof(struct udphdr));
                                if (sk2) {
                                        result = sk2;
                                        goto found;
@@ -331,7 +333,7 @@ struct sock *__udp6_lib_lookup(struct net *net,
 
                result = udp6_lib_lookup2(net, saddr, sport,
                                          daddr, hnum, dif,
-                                         hslot2, slot2);
+                                         hslot2, slot2, skb);
                if (!result) {
                        hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
                        slot2 = hash2 & udptable->mask;
@@ -341,7 +343,7 @@ struct sock *__udp6_lib_lookup(struct net *net,
 
                        result = udp6_lib_lookup2(net, saddr, sport,
                                                  &in6addr_any, hnum, dif,
-                                                 hslot2, slot2);
+                                                 hslot2, slot2, skb);
                }
                rcu_read_unlock();
                return result;
index 74ff0998895812850137675082ffaf8ec09f5bc0..bec1b5dd2530591543e688785674c4bd4178024a 100644 (file)
@@ -123,6 +123,8 @@ static void attach_ebpf(int fd, uint16_t mod)
        if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_REUSEPORT_EBPF, &bpf_fd,
                        sizeof(bpf_fd)))
                error(1, errno, "failed to set SO_ATTACH_REUSEPORT_EBPF");
+
+       close(bpf_fd);
 }
 
 static void attach_cbpf(int fd, uint16_t mod)
@@ -396,6 +398,9 @@ static void test_filter_without_bind(void)
 int main(void)
 {
        fprintf(stderr, "---- IPv4 UDP ----\n");
+       /* NOTE: UDP socket lookups traverse a different code path when there
+        * are > 10 sockets in a group.  Run the bpf test through both paths.
+        */
        test_reuseport_ebpf((struct test_params) {
                .recv_family = AF_INET,
                .send_family = AF_INET,
@@ -403,6 +408,13 @@ int main(void)
                .recv_socks = 10,
                .recv_port = 8000,
                .send_port_min = 9000});
+       test_reuseport_ebpf((struct test_params) {
+               .recv_family = AF_INET,
+               .send_family = AF_INET,
+               .protocol = SOCK_DGRAM,
+               .recv_socks = 20,
+               .recv_port = 8000,
+               .send_port_min = 9000});
        test_reuseport_cbpf((struct test_params) {
                .recv_family = AF_INET,
                .send_family = AF_INET,
@@ -410,6 +422,13 @@ int main(void)
                .recv_socks = 10,
                .recv_port = 8001,
                .send_port_min = 9020});
+       test_reuseport_cbpf((struct test_params) {
+               .recv_family = AF_INET,
+               .send_family = AF_INET,
+               .protocol = SOCK_DGRAM,
+               .recv_socks = 20,
+               .recv_port = 8001,
+               .send_port_min = 9020});
        test_extra_filter((struct test_params) {
                .recv_family = AF_INET,
                .protocol = SOCK_DGRAM,
@@ -427,6 +446,13 @@ int main(void)
                .recv_socks = 10,
                .recv_port = 8003,
                .send_port_min = 9040});
+       test_reuseport_ebpf((struct test_params) {
+               .recv_family = AF_INET6,
+               .send_family = AF_INET6,
+               .protocol = SOCK_DGRAM,
+               .recv_socks = 20,
+               .recv_port = 8003,
+               .send_port_min = 9040});
        test_reuseport_cbpf((struct test_params) {
                .recv_family = AF_INET6,
                .send_family = AF_INET6,
@@ -434,6 +460,13 @@ int main(void)
                .recv_socks = 10,
                .recv_port = 8004,
                .send_port_min = 9060});
+       test_reuseport_cbpf((struct test_params) {
+               .recv_family = AF_INET6,
+               .send_family = AF_INET6,
+               .protocol = SOCK_DGRAM,
+               .recv_socks = 20,
+               .recv_port = 8004,
+               .send_port_min = 9060});
        test_extra_filter((struct test_params) {
                .recv_family = AF_INET6,
                .protocol = SOCK_DGRAM,
@@ -444,6 +477,13 @@ int main(void)
                .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 = 20,
+               .recv_port = 8006,
+               .send_port_min = 9080});
        test_reuseport_ebpf((struct test_params) {
                .recv_family = AF_INET6,
                .send_family = AF_INET,
@@ -458,6 +498,13 @@ int main(void)
                .recv_socks = 10,
                .recv_port = 8007,
                .send_port_min = 9100});
+       test_reuseport_cbpf((struct test_params) {
+               .recv_family = AF_INET6,
+               .send_family = AF_INET,
+               .protocol = SOCK_DGRAM,
+               .recv_socks = 20,
+               .recv_port = 8007,
+               .send_port_min = 9100});
 
 
        test_filter_without_bind();