Merge branch 'soreuseport-mixed-v4-v6-fixes'

Craig Gallek says:

====================
Fixes for SO_REUSEPORT and mixed v4/v6 sockets

Recent changes to the datastructures associated with SO_REUSEPORT broke
an existing behavior when equivalent SO_REUSEPORT sockets are created
using both AF_INET and AF_INET6.  This patch series restores the previous
behavior and includes a test to validate it.

This series should be a trivial merge to stable kernels (if deemed
necessary), but will have conflicts in net-next.  The following patches
recently replaced the use of hlist_nulls with hlists for UDP and TCP
socket lists:
ca065d0 ("udp: no longer use SLAB_DESTROY_BY_RCU")
3b24d85 ("tcp/dccp: do not touch listener sk_refcnt under synflood")

If this series is accepted, I will send an RFC for the net-next change
to assist with the merge.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

Author: davem330

include/linux/rculist_nulls.h

@@ -98,6 +98,45 @@ static inline void hlist_nulls_add_head_rcu(struct hlist_nulls_node *n,
 	if (!is_a_nulls(first))
 		first->pprev = &n->next;
 }
+
+/**
+ * hlist_nulls_add_tail_rcu
+ * @n: the element to add to the hash list.
+ * @h: the list to add to.
+ *
+ * Description:
+ * Adds the specified element to the end of the specified hlist_nulls,
+ * while permitting racing traversals.  NOTE: tail insertion requires
+ * list traversal.
+ *
+ * The caller must take whatever precautions are necessary
+ * (such as holding appropriate locks) to avoid racing
+ * with another list-mutation primitive, such as hlist_nulls_add_head_rcu()
+ * or hlist_nulls_del_rcu(), running on this same list.
+ * However, it is perfectly legal to run concurrently with
+ * the _rcu list-traversal primitives, such as
+ * hlist_nulls_for_each_entry_rcu(), used to prevent memory-consistency
+ * problems on Alpha CPUs.  Regardless of the type of CPU, the
+ * list-traversal primitive must be guarded by rcu_read_lock().
+ */
+static inline void hlist_nulls_add_tail_rcu(struct hlist_nulls_node *n,
+					struct hlist_nulls_head *h)
+{
+	struct hlist_nulls_node *i, *last = NULL;
+
+	for (i = hlist_nulls_first_rcu(h); !is_a_nulls(i);
+	     i = hlist_nulls_next_rcu(i))
+		last = i;
+
+	if (last) {
+		n->next = last->next;
+		n->pprev = &last->next;
+		rcu_assign_pointer(hlist_nulls_next_rcu(last), n);
+	} else {
+		hlist_nulls_add_head_rcu(n, h);
+	}
+}
+
 /**
  * hlist_nulls_for_each_entry_rcu - iterate over rcu list of given type
  * @tpos:	the type * to use as a loop cursor.

include/net/sock.h

@@ -630,7 +630,11 @@ static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
 
 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 {
-	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
+	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
+	    sk->sk_family == AF_INET6)
+		hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
+	else
+		hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
 }
 
 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)

net/ipv4/udp.c

@@ -339,8 +339,13 @@ int udp_lib_get_port(struct sock *sk, unsigned short snum,
 
 		hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 		spin_lock(&hslot2->lock);
-		hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
-					 &hslot2->head);
+		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
+			sk->sk_family == AF_INET6)
+			hlist_nulls_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
+						 &hslot2->head);
+		else
+			hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
+						 &hslot2->head);
 		hslot2->count++;
 		spin_unlock(&hslot2->lock);
 	}

tools/testing/selftests/net/.gitignore

@@ -3,3 +3,4 @@ psock_fanout
 psock_tpacket
 reuseport_bpf
 reuseport_bpf_cpu
+reuseport_dualstack

tools/testing/selftests/net/Makefile

@@ -4,7 +4,7 @@ CFLAGS = -Wall -O2 -g
 
 CFLAGS += -I../../../../usr/include/
 
-NET_PROGS = socket psock_fanout psock_tpacket reuseport_bpf reuseport_bpf_cpu
+NET_PROGS = socket psock_fanout psock_tpacket reuseport_bpf reuseport_bpf_cpu reuseport_dualstack
 
 all: $(NET_PROGS)
 %: %.c

tools/testing/selftests/net/reuseport_dualstack.c

@@ -0,0 +1,208 @@
+/*
+ * It is possible to use SO_REUSEPORT to open multiple sockets bound to
+ * equivalent local addresses using AF_INET and AF_INET6 at the same time.  If
+ * the AF_INET6 socket has IPV6_V6ONLY set, it's clear which socket should
+ * receive a given incoming packet.  However, when it is not set, incoming v4
+ * packets should prefer the AF_INET socket(s).  This behavior was defined with
+ * the original SO_REUSEPORT implementation, but broke with
+ * e32ea7e74727 ("soreuseport: fast reuseport UDP socket selection")
+ * This test creates these mixed AF_INET/AF_INET6 sockets and asserts the
+ * AF_INET preference for v4 packets.
+ */
+
+#define _GNU_SOURCE
+
+#include <arpa/inet.h>
+#include <errno.h>
+#include <error.h>
+#include <linux/in.h>
+#include <linux/unistd.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>
+
+static const int PORT = 8888;
+
+static void build_rcv_fd(int family, int proto, int *rcv_fds, int count)
+{
+	struct sockaddr_storage addr;
+	struct sockaddr_in  *addr4;
+	struct sockaddr_in6 *addr6;
+	int opt, i;
+
+	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);
+	}
+
+	for (i = 0; i < count; ++i) {
+		rcv_fds[i] = socket(family, proto, 0);
+		if (rcv_fds[i] < 0)
+			error(1, errno, "failed to create receive socket");
+
+		opt = 1;
+		if (setsockopt(rcv_fds[i], SOL_SOCKET, SO_REUSEPORT, &opt,
+			       sizeof(opt)))
+			error(1, errno, "failed to set SO_REUSEPORT");
+
+		if (bind(rcv_fds[i], (struct sockaddr *)&addr, sizeof(addr)))
+			error(1, errno, "failed to bind receive socket");
+
+		if (proto == SOCK_STREAM && listen(rcv_fds[i], 10))
+			error(1, errno, "failed to listen on receive port");
+	}
+}
+
+static void send_from_v4(int proto)
+{
+	struct sockaddr_in  saddr, daddr;
+	int fd;
+
+	saddr.sin_family = AF_INET;
+	saddr.sin_addr.s_addr = htonl(INADDR_ANY);
+	saddr.sin_port = 0;
+
+	daddr.sin_family = AF_INET;
+	daddr.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
+	daddr.sin_port = htons(PORT);
+
+	fd = socket(AF_INET, proto, 0);
+	if (fd < 0)
+		error(1, errno, "failed to create send socket");
+
+	if (bind(fd, (struct sockaddr *)&saddr, sizeof(saddr)))
+		error(1, errno, "failed to bind send socket");
+
+	if (connect(fd, (struct sockaddr *)&daddr, sizeof(daddr)))
+		error(1, errno, "failed to connect send socket");
+
+	if (send(fd, "a", 1, 0) < 0)
+		error(1, errno, "failed to send message");
+
+	close(fd);
+}
+
+static int receive_once(int epfd, int proto)
+{
+	struct epoll_event ev;
+	int i, fd;
+	char buf[8];
+
+	i = epoll_wait(epfd, &ev, 1, -1);
+	if (i < 0)
+		error(1, errno, "epoll_wait failed");
+
+	if (proto == SOCK_STREAM) {
+		fd = accept(ev.data.fd, NULL, NULL);
+		if (fd < 0)
+			error(1, errno, "failed to accept");
+		i = recv(fd, buf, sizeof(buf), 0);
+		close(fd);
+	} else {
+		i = recv(ev.data.fd, buf, sizeof(buf), 0);
+	}
+
+	if (i < 0)
+		error(1, errno, "failed to recv");
+
+	return ev.data.fd;
+}
+
+static void test(int *rcv_fds, int count, int proto)
+{
+	struct epoll_event ev;
+	int epfd, i, test_fd;
+	uint16_t test_family;
+	socklen_t len;
+
+	epfd = epoll_create(1);
+	if (epfd < 0)
+		error(1, errno, "failed to create epoll");
+
+	ev.events = EPOLLIN;
+	for (i = 0; i < count; ++i) {
+		ev.data.fd = rcv_fds[i];
+		if (epoll_ctl(epfd, EPOLL_CTL_ADD, rcv_fds[i], &ev))
+			error(1, errno, "failed to register sock epoll");
+	}
+
+	send_from_v4(proto);
+
+	test_fd = receive_once(epfd, proto);
+	if (getsockopt(test_fd, SOL_SOCKET, SO_DOMAIN, &test_family, &len))
+		error(1, errno, "failed to read socket domain");
+	if (test_family != AF_INET)
+		error(1, 0, "expected to receive on v4 socket but got v6 (%d)",
+		      test_family);
+
+	close(epfd);
+}
+
+int main(void)
+{
+	int rcv_fds[32], i;
+
+	fprintf(stderr, "---- UDP IPv4 created before IPv6 ----\n");
+	build_rcv_fd(AF_INET, SOCK_DGRAM, rcv_fds, 5);
+	build_rcv_fd(AF_INET6, SOCK_DGRAM, &(rcv_fds[5]), 5);
+	test(rcv_fds, 10, SOCK_DGRAM);
+	for (i = 0; i < 10; ++i)
+		close(rcv_fds[i]);
+
+	fprintf(stderr, "---- UDP IPv6 created before IPv4 ----\n");
+	build_rcv_fd(AF_INET6, SOCK_DGRAM, rcv_fds, 5);
+	build_rcv_fd(AF_INET, SOCK_DGRAM, &(rcv_fds[5]), 5);
+	test(rcv_fds, 10, SOCK_DGRAM);
+	for (i = 0; i < 10; ++i)
+		close(rcv_fds[i]);
+
+	/* NOTE: UDP socket lookups traverse a different code path when there
+	 * are > 10 sockets in a group.
+	 */
+	fprintf(stderr, "---- UDP IPv4 created before IPv6 (large) ----\n");
+	build_rcv_fd(AF_INET, SOCK_DGRAM, rcv_fds, 16);
+	build_rcv_fd(AF_INET6, SOCK_DGRAM, &(rcv_fds[16]), 16);
+	test(rcv_fds, 32, SOCK_DGRAM);
+	for (i = 0; i < 32; ++i)
+		close(rcv_fds[i]);
+
+	fprintf(stderr, "---- UDP IPv6 created before IPv4 (large) ----\n");
+	build_rcv_fd(AF_INET6, SOCK_DGRAM, rcv_fds, 16);
+	build_rcv_fd(AF_INET, SOCK_DGRAM, &(rcv_fds[16]), 16);
+	test(rcv_fds, 32, SOCK_DGRAM);
+	for (i = 0; i < 32; ++i)
+		close(rcv_fds[i]);
+
+	fprintf(stderr, "---- TCP IPv4 created before IPv6 ----\n");
+	build_rcv_fd(AF_INET, SOCK_STREAM, rcv_fds, 5);
+	build_rcv_fd(AF_INET6, SOCK_STREAM, &(rcv_fds[5]), 5);
+	test(rcv_fds, 10, SOCK_STREAM);
+	for (i = 0; i < 10; ++i)
+		close(rcv_fds[i]);
+
+	fprintf(stderr, "---- TCP IPv6 created before IPv4 ----\n");
+	build_rcv_fd(AF_INET6, SOCK_STREAM, rcv_fds, 5);
+	build_rcv_fd(AF_INET, SOCK_STREAM, &(rcv_fds[5]), 5);
+	test(rcv_fds, 10, SOCK_STREAM);
+	for (i = 0; i < 10; ++i)
+		close(rcv_fds[i]);
+
+	fprintf(stderr, "SUCCESS\n");
+	return 0;
+}