[PATCH net-next v2 1/2] tcp: Consider every port when connecting with IP_LOCAL_PORT_RANGE

[PATCH net-next v2 1/2] tcp: Consider every port when connecting with IP_LOCAL_PORT_RANGE

[Jakub Sitnicki]

Situation
---------

We observe the following scenario in production:

                                                  inet_bind_bucket
                                                state for port 54321
                                                --------------------

                                                (bucket doesn't exist)

// Process A opens a long-lived connection:
s1 = socket(AF_INET, SOCK_STREAM)
s1.setsockopt(IP_BIND_ADDRESS_NO_PORT)
s1.setsockopt(IP_LOCAL_PORT_RANGE, 54000..54500)
s1.bind(192.0.2.10, 0)
s1.connect(192.51.100.1, 443)
                                                tb->reuse = -1
                                                tb->reuseport = -1
s1.getsockname() -> 192.0.2.10:54321
s1.send()
s1.recv()
// ... s1 stays open.

// Process B opens a short-lived connection:
s2 = socket(AF_INET, SOCK_STREAM)
s2.setsockopt(SO_REUSEADDR)
s2.bind(192.0.2.20, 0)
                                                tb->reuse = 0
                                                tb->reuseport = 0
s2.connect(192.51.100.2, 53)
s2.getsockname() -> 192.0.2.20:54321
s2.send()
s2.recv()
s2.close()

                                                // bucket remains in this
                                                // state even though port
                                                // was released by s2
                                                tb->reuse = 0
                                                tb->reuseport = 0

// Process A attempts to open another connection
// when there is connection pressure from
// 192.0.2.30:54000..54500 to 192.51.100.1:443.
// Assume only port 54321 is still available.

s3 = socket(AF_INET, SOCK_STREAM)
s3.setsockopt(IP_BIND_ADDRESS_NO_PORT)
s3.setsockopt(IP_LOCAL_PORT_RANGE, 54000..54500)
s3.bind(192.0.2.30, 0)
s3.connect(192.51.100.1, 443) -> EADDRNOTAVAIL (99)

Problem
-------

We end up in a state where Process A can't reuse ephemeral port 54321 for
as long as there are sockets, like s1, that keep the bind bucket alive. The
bucket does not return to "reusable" state even when all sockets which
blocked it from reuse, like s2, are gone.

The ephemeral port becomes available for use again only after all sockets
bound to it are gone and the bind bucket is destroyed.

Programs which behave like Process B in this scenario - that is, binding to
an IP address without setting IP_BIND_ADDRESS_NO_PORT - might be considered
poorly written. However, the reality is that such implementation is not
actually uncommon. Trying to fix each and every such program is like
playing whack-a-mole.

For instance, it could be any software using Golang's net.Dialer with
LocalAddr provided:

        dialer := &net.Dialer{
                LocalAddr: &net.TCPAddr{IP: srcIP},
        }
        conn, err := dialer.Dial("tcp4", dialTarget)

Or even a ubiquitous tool like dig when using a specific local address:

        $ dig -b 127.1.1.1 +tcp +short example.com

Hence, we are proposing a systematic fix in the network stack itself.

Solution
--------

If there is no IP address conflict with any socket bound to a given local
port, then from the protocol's perspective, the port can be safely shared.

With that in mind, modify the port search during connect(), that is
__inet_hash_connect, to consider all bind buckets (ports) when looking for
a local port for egress.

To achieve this, add an extra walk over bhash2 buckets for the port to
check for IP conflicts. The additional walk is not free, so perform it only
once per port - during the second phase of conflict checking, when the
bhash bucket is locked.

We enable this changed behavior only if the IP_LOCAL_PORT_RANGE socket
option is set. The rationale is that users are likely to care about using
every possible local port only when they have deliberately constrained the
ephemeral port range.

Limitations
-----------

Sharing both the local IP and port with other established sockets, when the
remote address is unique is still not possible if the bucket is in
non-reusable state (tb->{fastreuse,fastreuseport} >= 0) because of a socket
explicitly bound to that port.

Alternatives
------------

* Update bind bucket state on port release

A valid solution to the described problem would also be to walk the bind
bucket owners when releasing the port and recalculate the
tb->{reuse,reuseport} state.

However, in comparison to the proposed solution, this alone would not allow
sharing the local port with other sockets bound to non-conflicting IPs for
as long as they exist.

Another downside is that we'd pay the extra cost on each unbind (more
frequent) rather than only when connecting with IP_LOCAL_PORT_RANGE
set (less frequent). Due to that we would also likely need to guard it
behind a sysctl (see below).

* Run your service in a dedicated netns

This would also solve the problem. While we don't rule out transitioning to
this setup in the future at a cost of shifting the complexity elsewhere.

Isolating your service in a netns requires assigning it dedicated IPs for
egress. If the egress IPs must be shared with other processes, as in our
case, then SNAT and connection tracking on egress are required - adding
complexity.

* Guard it behind a sysctl setting instead of a socket option

Users are unlikely to encounter this problem unless their workload connects
from a severely narrowed-down ephemeral port range. Hence, paying the bind
bucket walk cost for each connect() call doesn't seem sensible. Whereas
with a socket option, only a limited set of connections incur the
performance overhead.

Reported-by: Lee Valentine <lvalentine@cloudflare.com>
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
---
To: Eric Dumazet <edumazet@google.com>
To: Paolo Abeni <pabeni@redhat.com>
To: David S. Miller <davem@davemloft.net>
To: Jakub Kicinski <kuba@kernel.org>
To: Neal Cardwell <ncardwell@google.com>
To: Kuniyuki Iwashima <kuniyu@google.com>
Cc: netdev@vger.kernel.org
Cc: kernel-team@cloudflare.com
---
Changes in v2:
- Fix unused var warning when CONFIG_IPV6=n
- Convert INADDR_ANY to network byte order before comparison
- Link to v1: https://lore.kernel.org/r/20250626120247.1255223-1-jakub@cloudflare.com
---
 net/ipv4/inet_hashtables.c | 46 ++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 44 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
index ceeeec9b7290..b4c4caf3ff6c 100644
--- a/net/ipv4/inet_hashtables.c
+++ b/net/ipv4/inet_hashtables.c
@@ -1005,6 +1005,43 @@ EXPORT_IPV6_MOD(inet_bhash2_reset_saddr);
 #define INET_TABLE_PERTURB_SIZE (1 << CONFIG_INET_TABLE_PERTURB_ORDER)
 static u32 *table_perturb;
 
+/* True on source address conflict with another socket. False otherwise.
+ * Caller must hold hashbucket lock for this tb.
+ */
+static inline bool check_bound(const struct sock *sk,
+			       const struct inet_bind_bucket *tb)
+{
+	const struct inet_bind2_bucket *tb2;
+
+	hlist_for_each_entry(tb2, &tb->bhash2, bhash_node) {
+#if IS_ENABLED(CONFIG_IPV6)
+		const struct sock *sk2;
+
+		if (sk->sk_family == AF_INET6) {
+			if (tb2->addr_type == IPV6_ADDR_ANY ||
+			    ipv6_addr_equal(&tb2->v6_rcv_saddr,
+					    &sk->sk_v6_rcv_saddr))
+				return true;
+			continue;
+		}
+
+		/* Check for ipv6 non-v6only wildcard sockets */
+		if (tb2->addr_type == IPV6_ADDR_ANY)
+			sk_for_each_bound(sk2, &tb2->owners)
+				if (!sk2->sk_ipv6only)
+					return true;
+
+		if (tb2->addr_type != IPV6_ADDR_MAPPED)
+			continue;
+#endif
+		if (tb2->rcv_saddr == htonl(INADDR_ANY) ||
+		    tb2->rcv_saddr == sk->sk_rcv_saddr)
+			return true;
+	}
+
+	return false;
+}
+
 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 		struct sock *sk, u64 port_offset,
 		u32 hash_port0,
@@ -1070,6 +1107,8 @@ int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 			if (!inet_bind_bucket_match(tb, net, port, l3mdev))
 				continue;
 			if (tb->fastreuse >= 0 || tb->fastreuseport >= 0) {
+				if (unlikely(local_ports))
+					break; /* optimistic assumption */
 				rcu_read_unlock();
 				goto next_port;
 			}
@@ -1088,9 +1127,12 @@ int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 		 */
 		inet_bind_bucket_for_each(tb, &head->chain) {
 			if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
-				if (tb->fastreuse >= 0 ||
-				    tb->fastreuseport >= 0)
+				if (tb->fastreuse >= 0 || tb->fastreuseport >= 0) {
+					if (unlikely(local_ports &&
+						     !check_bound(sk, tb)))
+						goto ok;
 					goto next_port_unlock;
+				}
 				WARN_ON(hlist_empty(&tb->bhash2));
 				if (!check_established(death_row, sk,
 						       port, &tw, false,

-- 
2.43.0

[PATCH net-next v2 2/2] selftests/net: Cover port sharing scenarios with IP_LOCAL_PORT_RANGE

[Jakub Sitnicki]

Expand the ip_local_port_range tests to check that when using
IP_LOCAL_PORT_RANGE socket option:

1) We can share the local port as long as there is no IP address conflict
   with any other socket. Covered by tcp_port_reuse__no_ip_conflict* tests.

2) We cannot share the local port with wildcard sockets or when there is a
   local IP conflict. Covered by tcp_port_reuse__ip_conflict* tests.

3) We cannot share the local IP and port to connect to different remote IPs
   if the port bucket is in non-reuseable state, Corner case covered by
   tcp_port_reuse__ip_port_conflict_with_unique_dst_after_bind test.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
---
 tools/testing/selftests/net/ip_local_port_range.c  | 525 +++++++++++++++++++++
 tools/testing/selftests/net/ip_local_port_range.sh |  14 +-
 2 files changed, 534 insertions(+), 5 deletions(-)

diff --git a/tools/testing/selftests/net/ip_local_port_range.c b/tools/testing/selftests/net/ip_local_port_range.c
index 29451d2244b7..33288732dc1b 100644
--- a/tools/testing/selftests/net/ip_local_port_range.c
+++ b/tools/testing/selftests/net/ip_local_port_range.c
@@ -9,6 +9,7 @@
 
 #include <fcntl.h>
 #include <netinet/ip.h>
+#include <arpa/inet.h>
 
 #include "../kselftest_harness.h"
 
@@ -20,6 +21,15 @@
 #define IPPROTO_MPTCP 262
 #endif
 
+static const int ONE = 1;
+
+__attribute__((nonnull)) static inline void close_fd(int *fd)
+{
+	close(*fd);
+}
+
+#define __close_fd __attribute__((cleanup(close_fd)))
+
 static __u32 pack_port_range(__u16 lo, __u16 hi)
 {
 	return (hi << 16) | (lo << 0);
@@ -116,6 +126,81 @@ static int get_ip_local_port_range(int fd, __u32 *range)
 	return 0;
 }
 
+struct sockaddr_inet {
+	union {
+		struct sockaddr_storage ss;
+		struct sockaddr_in6 v6;
+		struct sockaddr_in v4;
+		struct sockaddr sa;
+	};
+	socklen_t len;
+};
+
+static void make_inet_addr(int af, const char *ip, __u16 port,
+			   struct sockaddr_inet *addr)
+{
+	memset(addr, 0, sizeof(*addr));
+
+	switch (af) {
+	case AF_INET:
+		addr->len = sizeof(addr->v4);
+		addr->v4.sin_family = af;
+		addr->v4.sin_port = htons(port);
+		inet_pton(af, ip, &addr->v4.sin_addr);
+		break;
+	case AF_INET6:
+		addr->len = sizeof(addr->v6);
+		addr->v6.sin6_family = af;
+		addr->v6.sin6_port = htons(port);
+		inet_pton(af, ip, &addr->v6.sin6_addr);
+		break;
+	}
+}
+
+static bool is_v4mapped(const struct sockaddr_inet *a)
+{
+	return (a->sa.sa_family == AF_INET6 &&
+		IN6_IS_ADDR_V4MAPPED(&a->v6.sin6_addr));
+}
+
+static void v4mapped_to_ipv4(struct sockaddr_inet *a)
+{
+	in_port_t port = a->v6.sin6_port;
+	in_addr_t ip4 = *(in_addr_t *)&a->v6.sin6_addr.s6_addr[12];
+
+	memset(a, 0, sizeof(*a));
+	a->len = sizeof(a->v4);
+	a->v4.sin_family = AF_INET;
+	a->v4.sin_port = port;
+	a->v4.sin_addr.s_addr = ip4;
+}
+
+static void ipv4_to_v4mapped(struct sockaddr_inet *a)
+{
+	in_port_t port = a->v4.sin_port;
+	in_addr_t ip4 = a->v4.sin_addr.s_addr;
+
+	memset(a, 0, sizeof(*a));
+	a->len = sizeof(a->v6);
+	a->v6.sin6_family = AF_INET6;
+	a->v6.sin6_port = port;
+	a->v6.sin6_addr.s6_addr[10] = 0xff;
+	a->v6.sin6_addr.s6_addr[11] = 0xff;
+	memcpy(&a->v6.sin6_addr.s6_addr[12], &ip4, sizeof(ip4));
+}
+
+static __u16 inet_port(const struct sockaddr_inet *a)
+{
+	switch (a->sa.sa_family) {
+	case AF_INET:
+		return ntohs(a->v4.sin_port);
+	case AF_INET6:
+		return ntohs(a->v6.sin6_port);
+	default:
+		return 0;
+	}
+}
+
 FIXTURE(ip_local_port_range) {};
 
 FIXTURE_SETUP(ip_local_port_range)
@@ -460,4 +545,444 @@ TEST_F(ip_local_port_range, get_port_range)
 	ASSERT_TRUE(!err) TH_LOG("close failed");
 }
 
+FIXTURE(tcp_port_reuse__no_ip_conflict) {};
+FIXTURE_SETUP(tcp_port_reuse__no_ip_conflict) {}
+FIXTURE_TEARDOWN(tcp_port_reuse__no_ip_conflict) {}
+
+FIXTURE_VARIANT(tcp_port_reuse__no_ip_conflict) {
+	int af_one;
+	const char *ip_one;
+	int af_two;
+	const char *ip_two;
+};
+
+FIXTURE_VARIANT_ADD(tcp_port_reuse__no_ip_conflict, ipv4) {
+	.af_one = AF_INET,
+	.ip_one = "127.0.0.1",
+	.af_two = AF_INET,
+	.ip_two = "127.0.0.2",
+};
+
+FIXTURE_VARIANT_ADD(tcp_port_reuse__no_ip_conflict, ipv6_v4mapped) {
+	.af_one = AF_INET6,
+	.ip_one = "::ffff:127.0.0.1",
+	.af_two = AF_INET,
+	.ip_two = "127.0.0.2",
+};
+
+FIXTURE_VARIANT_ADD(tcp_port_reuse__no_ip_conflict, ipv6) {
+	.af_one = AF_INET6,
+	.ip_one = "2001:db8::1",
+	.af_two = AF_INET6,
+	.ip_two = "2001:db8::2",
+};
+
+/* Check that a connected socket, which is using IP_LOCAL_PORT_RANGE to relax
+ * port search restrictions at connect() time, can share a local port with a
+ * listening socket bound to a different IP.
+ */
+TEST_F(tcp_port_reuse__no_ip_conflict, share_port_with_listening_socket)
+{
+	const typeof(variant) v = variant;
+	struct sockaddr_inet addr;
+	__close_fd int ln = -1;
+	__close_fd int c = -1;
+	__close_fd int p = -1;
+	__u32 range;
+	int r;
+
+	/* Listen on <ip one>:40000 */
+	ln = socket(v->af_one, SOCK_STREAM, 0);
+	ASSERT_GE(ln, 0) TH_LOG("socket");
+
+	r = setsockopt(ln, SOL_SOCKET, SO_REUSEADDR, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(SO_REUSEADDR)");
+
+	make_inet_addr(v->af_one, v->ip_one, 40000, &addr);
+	r = bind(ln, &addr.sa, addr.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind(<ip_one>:40000)");
+
+	r = listen(ln, 1);
+	ASSERT_EQ(r, 0) TH_LOG("listen");
+
+	/* Connect from <ip two>:40000 to <ip one>:40000 */
+	c = socket(v->af_two, SOCK_STREAM, 0);
+	ASSERT_GE(c, 0) TH_LOG("socket");
+
+	r = setsockopt(c, SOL_IP, IP_BIND_ADDRESS_NO_PORT, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_BIND_ADDRESS_NO_PORT)");
+
+	range = pack_port_range(40000, 40000);
+	r = setsockopt(c, SOL_IP, IP_LOCAL_PORT_RANGE, &range, sizeof(range));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_LOCAL_PORT_RANGE)");
+
+	make_inet_addr(v->af_two, v->ip_two, 0, &addr);
+	r = bind(c, &addr.sa, addr.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind(<ip_two>:0)");
+
+	make_inet_addr(v->af_one, v->ip_one, 40000, &addr);
+	if (is_v4mapped(&addr))
+		v4mapped_to_ipv4(&addr);
+	r = connect(c, &addr.sa, addr.len);
+	EXPECT_EQ(r, 0) TH_LOG("connect(<ip_one>:40000)");
+	EXPECT_EQ(get_sock_port(c), 40000);
+}
+
+/* Check that a connected socket, which is using IP_LOCAL_PORT_RANGE to relax
+ * port search restrictions at connect() time, can share a local port with
+ * another connected socket bound to a different IP without
+ * IP_BIND_ADDRESS_NO_PORT enabled.
+ */
+TEST_F(tcp_port_reuse__no_ip_conflict, share_port_with_connected_socket)
+{
+	const typeof(variant) v = variant;
+	struct sockaddr_inet dst = {};
+	struct sockaddr_inet src = {};
+	__close_fd int ln = -1;
+	__close_fd int c1 = -1;
+	__close_fd int c2 = -1;
+	__u32 range;
+	__u16 port;
+	int r;
+
+	/* Listen on wildcard. Same family as <ip_two>. */
+	ln = socket(v->af_two, SOCK_STREAM, 0);
+	ASSERT_GE(ln, 0) TH_LOG("socket");
+
+	r = setsockopt(ln, SOL_SOCKET, SO_REUSEADDR, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(SO_REUSEADDR");
+
+	r = listen(ln, 2);
+	ASSERT_EQ(r, 0) TH_LOG("listen");
+
+	dst.len = sizeof(dst.ss);
+	r = getsockname(ln, &dst.sa, &dst.len);
+	ASSERT_EQ(r, 0) TH_LOG("getsockname");
+
+	/* Connect from <ip one> but without IP_BIND_ADDRESS_NO_PORT */
+	c1 = socket(v->af_one, SOCK_STREAM, 0);
+	ASSERT_GE(c1, 0) TH_LOG("socket");
+
+	make_inet_addr(v->af_one, v->ip_one, 0, &src);
+	r = bind(c1, &src.sa, src.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind");
+
+	if (src.sa.sa_family == AF_INET6 && dst.sa.sa_family == AF_INET)
+		ipv4_to_v4mapped(&dst);
+	r = connect(c1, &dst.sa, dst.len);
+	ASSERT_EQ(r, 0) TH_LOG("connect");
+
+	src.len = sizeof(src.ss);
+	r = getsockname(c1, &src.sa, &src.len);
+	ASSERT_EQ(r, 0) TH_LOG("getsockname");
+
+	/* Connect from <ip two>:<c1 port> with IP_BIND_ADDRESS_NO_PORT */
+	c2 = socket(v->af_two, SOCK_STREAM, 0);
+	ASSERT_GE(c2, 0) TH_LOG("socket");
+
+	r = setsockopt(c2, SOL_IP, IP_BIND_ADDRESS_NO_PORT, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_BIND_ADDRESS_NO_PORT)");
+
+	port = inet_port(&src);
+	range = pack_port_range(port, port);
+	r = setsockopt(c2, SOL_IP, IP_LOCAL_PORT_RANGE, &range, sizeof(range));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_LOCAL_PORT_RANGE)");
+
+	make_inet_addr(v->af_two, v->ip_two, 0, &src);
+	r = bind(c2, &src.sa, src.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind");
+
+	if (is_v4mapped(&dst))
+		v4mapped_to_ipv4(&dst);
+	r = connect(c2, &dst.sa, dst.len);
+	EXPECT_EQ(r, 0) TH_LOG("connect");
+	EXPECT_EQ(get_sock_port(c2), port);
+}
+
+/* Check that a connected socket, which is using IP_LOCAL_PORT_RANGE to relax
+ * port search restrictions at connect() time, can share a local port with an
+ * IPv6 wildcard socket which is not dualstack (v6-only).
+ */
+TEST(tcp_port_reuse__no_ip_conflict_wildcard_v6only)
+{
+	struct sockaddr_inet addr;
+	__close_fd int ln4 = -1;
+	__close_fd int ln6 = -1;
+	__close_fd int c = -1;
+	__u32 range;
+	int r;
+
+	/* Listen on [::]:40000 (v6only) */
+	ln6 = socket(AF_INET6, SOCK_STREAM, 0);
+	ASSERT_GE(ln6, 0) TH_LOG("socket");
+
+	r = setsockopt(ln6, SOL_SOCKET, SO_REUSEADDR, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(SO_REUSEADDR)");
+
+	r = setsockopt(ln6, IPPROTO_IPV6, IPV6_V6ONLY, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IPV6_V6ONLY)");
+
+	make_inet_addr(AF_INET6, "::", 40000, &addr);
+	r = bind(ln6, &addr.sa, addr.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind([::]:40000)");
+
+	r = listen(ln6, 1);
+	ASSERT_EQ(r, 0) TH_LOG("listen");
+
+	/* Listen on 127.0.0.1:30000 */
+	ln4 = socket(AF_INET, SOCK_STREAM, 0);
+	ASSERT_GE(ln4, 0) TH_LOG("socket");
+
+	r = setsockopt(ln4, SOL_SOCKET, SO_REUSEADDR, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(SO_REUSEADDR)");
+
+	make_inet_addr(AF_INET, "127.0.0.1", 30000, &addr);
+	r = bind(ln4, &addr.sa, addr.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind(127.0.0.1:30000)");
+
+	r = listen(ln4, 1);
+	ASSERT_EQ(r, 0) TH_LOG("listen");
+
+	/* Connect from 127.0.0.1:40000 to 127.0.0.1:30000*/
+	c = socket(AF_INET, SOCK_STREAM, 0);
+	ASSERT_GE(c, 0) TH_LOG("socket");
+
+	range = pack_port_range(40000, 40000);
+	r = setsockopt(c, SOL_IP, IP_LOCAL_PORT_RANGE, &range, sizeof(range));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_LOCAL_PORT_RANGE)");
+
+	r = connect(c, &addr.sa, addr.len);
+	EXPECT_EQ(r, 0) TH_LOG("connect(127.0.0.1:30000)");
+	EXPECT_EQ(get_sock_port(c), 40000);
+}
+
+/* Check that two sockets can share the local IP and the ephemeral port when the
+ * destination address differs.
+ */
+TEST(tcp_port_reuse__no_ip_conflict_with_unique_dst)
+{
+	struct sockaddr_inet addr;
+	__close_fd int ln = -1;
+	__close_fd int c1 = -1;
+	__close_fd int c2 = -1;
+	__u32 range;
+	int r;
+
+	/* Listen on 0.0.0.0:30000 */
+	ln = socket(AF_INET, SOCK_STREAM, 0);
+	ASSERT_GE(ln, 0) TH_LOG("socket");
+
+	r = setsockopt(ln, SOL_SOCKET, SO_REUSEADDR, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(SO_REUSEADDR)");
+
+	make_inet_addr(AF_INET, "0.0.0.0", 30000, &addr);
+	r = bind(ln, &addr.sa, addr.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind");
+
+	r = listen(ln, 2);
+	ASSERT_EQ(r, 0) TH_LOG("listen");
+
+	/* Connect from 127.0.0.1:40000 to 127.1.1.1:30000 */
+	c1 = socket(AF_INET, SOCK_STREAM, 0);
+	ASSERT_GE(c1, 0) TH_LOG("socket");
+
+	range = pack_port_range(40000, 40000);
+	r = setsockopt(c1, SOL_IP, IP_LOCAL_PORT_RANGE, &range, sizeof(range));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_LOCAL_PORT_RANGE)");
+
+	make_inet_addr(AF_INET, "127.1.1.1", 30000, &addr);
+	r = connect(c1, &addr.sa, addr.len);
+	ASSERT_EQ(r, 0) TH_LOG("connect(127.1.1.1:30000)");
+	ASSERT_EQ(get_sock_port(c1), 40000);
+
+	/* Connect from 127.0.0.1:40000 to 127.2.2.2:30000 */
+	c2 = socket(AF_INET, SOCK_STREAM, 0);
+	ASSERT_GE(c2, 0) TH_LOG("socket");
+
+	range = pack_port_range(40000, 40000);
+	r = setsockopt(c2, SOL_IP, IP_LOCAL_PORT_RANGE, &range, sizeof(range));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_LOCAL_PORT_RANGE)");
+
+	make_inet_addr(AF_INET, "127.2.2.2", 30000, &addr);
+	r = connect(c2, &addr.sa, addr.len);
+	EXPECT_EQ(r, 0) TH_LOG("connect(127.1.1.1:30000)");
+	EXPECT_EQ(get_sock_port(c2), 40000);
+}
+
+FIXTURE(tcp_port_reuse__ip_conflict) {};
+FIXTURE_SETUP(tcp_port_reuse__ip_conflict) {}
+FIXTURE_TEARDOWN(tcp_port_reuse__ip_conflict) {}
+
+FIXTURE_VARIANT(tcp_port_reuse__ip_conflict) {
+	int af_one;
+	const char *ip_one;
+	int af_two;
+	const char *ip_two;
+};
+
+FIXTURE_VARIANT_ADD(tcp_port_reuse__ip_conflict, ipv4) {
+	.af_one = AF_INET,
+	.ip_one = "127.0.0.1",
+	.af_two = AF_INET,
+	.ip_two = "127.0.0.1",
+};
+
+FIXTURE_VARIANT_ADD(tcp_port_reuse__ip_conflict, ipv6_v4mapped) {
+	.af_one = AF_INET6,
+	.ip_one = "::ffff:127.0.0.1",
+	.af_two = AF_INET,
+	.ip_two = "127.0.0.1",
+};
+
+FIXTURE_VARIANT_ADD(tcp_port_reuse__ip_conflict, ipv6) {
+	.af_one = AF_INET6,
+	.ip_one = "2001:db8::1",
+	.af_two = AF_INET6,
+	.ip_two = "2001:db8::1",
+};
+
+FIXTURE_VARIANT_ADD(tcp_port_reuse__ip_conflict, ipv4_wildcard) {
+	.af_one = AF_INET,
+	.ip_one = "0.0.0.0",
+	.af_two = AF_INET,
+	.ip_two = "127.0.0.1",
+};
+
+FIXTURE_VARIANT_ADD(tcp_port_reuse__ip_conflict, ipv6_v4mapped_wildcard) {
+	.af_one = AF_INET6,
+	.ip_one = "::ffff:0.0.0.0",
+	.af_two = AF_INET,
+	.ip_two = "127.0.0.1",
+};
+
+FIXTURE_VARIANT_ADD(tcp_port_reuse__ip_conflict, ipv6_wildcard) {
+	.af_one = AF_INET6,
+	.ip_one = "::",
+	.af_two = AF_INET6,
+	.ip_two = "2001:db8::1",
+};
+
+FIXTURE_VARIANT_ADD(tcp_port_reuse__ip_conflict, dualstack_wildcard) {
+	.af_one = AF_INET6,
+	.ip_one = "::",
+	.af_two = AF_INET6,
+	.ip_two = "127.0.0.1",
+};
+
+/* Check that a socket, which using IP_LOCAL_PORT_RANGE to relax local port
+ * search restrictions at connect() time, can't share a local port with a
+ * listening socket when there is IP address conflict.
+ */
+TEST_F(tcp_port_reuse__ip_conflict, cannot_share_port)
+{
+	const typeof(variant) v = variant;
+	struct sockaddr_inet dst, src;
+	__close_fd int ln = -1;
+	__close_fd int c = -1;
+	__u32 range;
+	int r;
+
+	/* Listen on <ip_one>:40000 */
+	ln = socket(v->af_one, SOCK_STREAM, 0);
+	ASSERT_GE(ln, 0) TH_LOG("socket");
+
+	r = setsockopt(ln, SOL_SOCKET, SO_REUSEADDR, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(SO_REUSEADDR)");
+
+	make_inet_addr(v->af_one, v->ip_one, 40000, &dst);
+	r = bind(ln, &dst.sa, dst.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind(<ip_one>:40000)");
+
+	r = listen(ln, 1);
+	ASSERT_EQ(r, 0) TH_LOG("listen");
+
+	/* Attempt to connect from <ip two>:40000 */
+	c = socket(v->af_two, SOCK_STREAM, 0);
+	ASSERT_GE(c, 0) TH_LOG("socket");
+
+	r = setsockopt(c, SOL_IP, IP_BIND_ADDRESS_NO_PORT, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_BIND_ADDRESS_NO_PORT)");
+
+	range = pack_port_range(40000, 40000);
+	r = setsockopt(c, SOL_IP, IP_LOCAL_PORT_RANGE, &range, sizeof(range));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_LOCAL_PORT_RANGE)");
+
+	make_inet_addr(v->af_two, v->ip_two, 0, &src);
+	r = bind(c, &src.sa, src.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind(<ip_two>:40000)");
+
+	if (is_v4mapped(&dst))
+		v4mapped_to_ipv4(&dst);
+	r = connect(c, &dst.sa, dst.len);
+	EXPECT_EQ(r, -1) TH_LOG("connect(*:40000)");
+	EXPECT_EQ(errno, EADDRNOTAVAIL);
+}
+
+/* Demonstrate that a local IP and port can't be shared any more, even when the
+ * remote address is unique, after explicitly binding to that port.
+ */
+TEST(tcp_port_reuse__ip_port_conflict_with_unique_dst_after_bind)
+{
+	struct sockaddr_inet addr;
+	__close_fd int ln = -1;
+	__close_fd int c1 = -1;
+	__close_fd int c2 = -1;
+	__u32 range;
+	int s, r;
+
+	/* Listen on 0.0.0.0:30000 */
+	ln = socket(AF_INET, SOCK_STREAM, 0);
+	ASSERT_GE(ln, 0) TH_LOG("socket");
+
+	r = setsockopt(ln, SOL_SOCKET, SO_REUSEADDR, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(SO_REUSEADDR)");
+
+	make_inet_addr(AF_INET, "0.0.0.0", 30000, &addr);
+	r = bind(ln, &addr.sa, addr.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind(0.0.0.0:30000)");
+
+	r = listen(ln, 2);
+	ASSERT_EQ(r, 0) TH_LOG("listen");
+
+	/* Connect from 127.0.0.1:40000 to 127.1.1.1:30000 */
+	c1 = socket(AF_INET, SOCK_STREAM, 0);
+	ASSERT_GE(c1, 0) TH_LOG("socket");
+
+	range = pack_port_range(40000, 40000);
+	r = setsockopt(c1, SOL_IP, IP_LOCAL_PORT_RANGE, &range, sizeof(range));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_LOCAL_PORT_RANGE)");
+
+	make_inet_addr(AF_INET, "127.1.1.1", 30000, &addr);
+	r = connect(c1, &addr.sa, addr.len);
+	ASSERT_EQ(r, 0) TH_LOG("connect(127.1.1.1:30000)");
+	ASSERT_EQ(get_sock_port(c1), 40000);
+
+	/* Block the port. Bind to 127.9.9.9:40000 and unbind immediately */
+	s = socket(AF_INET, SOCK_STREAM, 0);
+	ASSERT_GE(s, 0) TH_LOG("socket");
+
+	r = setsockopt(s, SOL_SOCKET, SO_REUSEADDR, &ONE, sizeof(ONE));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(SO_REUSEADDR)");
+
+	make_inet_addr(AF_INET, "127.9.9.9", 40000, &addr);
+	r = bind(s, &addr.sa, addr.len);
+	ASSERT_EQ(r, 0) TH_LOG("bind(127.9.9.9:40000)");
+
+	r = close(s);
+	ASSERT_EQ(r, 0) TH_LOG("close");
+
+	/* Connect from 127.0.0.1:40000 to 127.2.2.2:30000 */
+	c2 = socket(AF_INET, SOCK_STREAM, 0);
+	ASSERT_GE(c2, 0) TH_LOG("socket");
+
+	range = pack_port_range(40000, 40000);
+	r = setsockopt(c2, SOL_IP, IP_LOCAL_PORT_RANGE, &range, sizeof(range));
+	ASSERT_EQ(r, 0) TH_LOG("setsockopt(IP_LOCAL_PORT_RANGE)");
+
+	make_inet_addr(AF_INET, "127.2.2.2", 30000, &addr);
+	r = connect(c2, &addr.sa, addr.len);
+	EXPECT_EQ(r, -1) TH_LOG("connect(127.1.1.1:30000)");
+	EXPECT_EQ(errno, EADDRNOTAVAIL);
+}
+
 TEST_HARNESS_MAIN
diff --git a/tools/testing/selftests/net/ip_local_port_range.sh b/tools/testing/selftests/net/ip_local_port_range.sh
index 4ff746db1256..3fc151545b2d 100755
--- a/tools/testing/selftests/net/ip_local_port_range.sh
+++ b/tools/testing/selftests/net/ip_local_port_range.sh
@@ -1,7 +1,11 @@
-#!/bin/sh
+#!/bin/bash
 # SPDX-License-Identifier: GPL-2.0
 
-./in_netns.sh \
-  sh -c 'sysctl -q -w net.mptcp.enabled=1 && \
-         sysctl -q -w net.ipv4.ip_local_port_range="40000 49999" && \
-         ./ip_local_port_range'
+./in_netns.sh sh <(cat <<-EOF
+        sysctl -q -w net.mptcp.enabled=1
+        sysctl -q -w net.ipv4.ip_local_port_range="40000 49999"
+        ip -6 addr add dev lo 2001:db8::1/32 nodad
+        ip -6 addr add dev lo 2001:db8::2/32 nodad
+        exec ./ip_local_port_range
+EOF
+)

-- 
2.43.0

Re: [PATCH net-next v2 1/2] tcp: Consider every port when connecting with IP_LOCAL_PORT_RANGE

[Jakub Sitnicki]

On Thu, Jul 03, 2025 at 05:59 PM +02, Jakub Sitnicki wrote:
> Situation
> ---------
>
> We observe the following scenario in production:
>
>                                                   inet_bind_bucket
>                                                 state for port 54321
>                                                 --------------------
>
>                                                 (bucket doesn't exist)
>
> // Process A opens a long-lived connection:
> s1 = socket(AF_INET, SOCK_STREAM)
> s1.setsockopt(IP_BIND_ADDRESS_NO_PORT)
> s1.setsockopt(IP_LOCAL_PORT_RANGE, 54000..54500)
> s1.bind(192.0.2.10, 0)
> s1.connect(192.51.100.1, 443)
>                                                 tb->reuse = -1
>                                                 tb->reuseport = -1
> s1.getsockname() -> 192.0.2.10:54321
> s1.send()
> s1.recv()
> // ... s1 stays open.
>
> // Process B opens a short-lived connection:
> s2 = socket(AF_INET, SOCK_STREAM)
> s2.setsockopt(SO_REUSEADDR)
> s2.bind(192.0.2.20, 0)
>                                                 tb->reuse = 0
>                                                 tb->reuseport = 0
> s2.connect(192.51.100.2, 53)
> s2.getsockname() -> 192.0.2.20:54321
> s2.send()
> s2.recv()
> s2.close()
>
>                                                 // bucket remains in this
>                                                 // state even though port
>                                                 // was released by s2
>                                                 tb->reuse = 0
>                                                 tb->reuseport = 0
>
> // Process A attempts to open another connection
> // when there is connection pressure from
> // 192.0.2.30:54000..54500 to 192.51.100.1:443.
> // Assume only port 54321 is still available.
>
> s3 = socket(AF_INET, SOCK_STREAM)
> s3.setsockopt(IP_BIND_ADDRESS_NO_PORT)
> s3.setsockopt(IP_LOCAL_PORT_RANGE, 54000..54500)
> s3.bind(192.0.2.30, 0)
> s3.connect(192.51.100.1, 443) -> EADDRNOTAVAIL (99)
>
> Problem
> -------
>
> We end up in a state where Process A can't reuse ephemeral port 54321 for
> as long as there are sockets, like s1, that keep the bind bucket alive. The
> bucket does not return to "reusable" state even when all sockets which
> blocked it from reuse, like s2, are gone.
>
> The ephemeral port becomes available for use again only after all sockets
> bound to it are gone and the bind bucket is destroyed.
>
> Programs which behave like Process B in this scenario - that is, binding to
> an IP address without setting IP_BIND_ADDRESS_NO_PORT - might be considered
> poorly written. However, the reality is that such implementation is not
> actually uncommon. Trying to fix each and every such program is like
> playing whack-a-mole.
>
> For instance, it could be any software using Golang's net.Dialer with
> LocalAddr provided:
>
>         dialer := &net.Dialer{
>                 LocalAddr: &net.TCPAddr{IP: srcIP},
>         }
>         conn, err := dialer.Dial("tcp4", dialTarget)
>
> Or even a ubiquitous tool like dig when using a specific local address:
>
>         $ dig -b 127.1.1.1 +tcp +short example.com
>
> Hence, we are proposing a systematic fix in the network stack itself.
>
> Solution
> --------
>
> If there is no IP address conflict with any socket bound to a given local
> port, then from the protocol's perspective, the port can be safely shared.
>
> With that in mind, modify the port search during connect(), that is
> __inet_hash_connect, to consider all bind buckets (ports) when looking for
> a local port for egress.
>
> To achieve this, add an extra walk over bhash2 buckets for the port to
> check for IP conflicts. The additional walk is not free, so perform it only
> once per port - during the second phase of conflict checking, when the
> bhash bucket is locked.
>
> We enable this changed behavior only if the IP_LOCAL_PORT_RANGE socket
> option is set. The rationale is that users are likely to care about using
> every possible local port only when they have deliberately constrained the
> ephemeral port range.

I was wondering what the maintainers' outlook on this change is:

Does this sound acceptable?
Or should we start looking in a different direction?

Thanks,
-jkbs

[...]

Re: [PATCH net-next v2 1/2] tcp: Consider every port when connecting with IP_LOCAL_PORT_RANGE

[Eric Dumazet]

On Tue, Jul 8, 2025 at 2:13 AM Jakub Sitnicki <jakub@cloudflare.com> wrote:

> I was wondering what the maintainers' outlook on this change is:
>
> Does this sound acceptable?
> Or should we start looking in a different direction?
>


Long weekend, and a big number of submissions while we were away.

Let me take a look.

Re: [PATCH net-next v2 1/2] tcp: Consider every port when connecting with IP_LOCAL_PORT_RANGE

[Eric Dumazet]

On Thu, Jul 3, 2025 at 8:59 AM Jakub Sitnicki <jakub@cloudflare.com> wrote:
>
> Situation
> ---------
>
> We observe the following scenario in production:
>
>                                                   inet_bind_bucket
>                                                 state for port 54321
>                                                 --------------------
>
>                                                 (bucket doesn't exist)
>
> // Process A opens a long-lived connection:
> s1 = socket(AF_INET, SOCK_STREAM)
> s1.setsockopt(IP_BIND_ADDRESS_NO_PORT)
> s1.setsockopt(IP_LOCAL_PORT_RANGE, 54000..54500)
> s1.bind(192.0.2.10, 0)
> s1.connect(192.51.100.1, 443)
>                                                 tb->reuse = -1
>                                                 tb->reuseport = -1
> s1.getsockname() -> 192.0.2.10:54321
> s1.send()
> s1.recv()
> // ... s1 stays open.
>
> // Process B opens a short-lived connection:
> s2 = socket(AF_INET, SOCK_STREAM)
> s2.setsockopt(SO_REUSEADDR)
> s2.bind(192.0.2.20, 0)
>                                                 tb->reuse = 0
>                                                 tb->reuseport = 0
> s2.connect(192.51.100.2, 53)
> s2.getsockname() -> 192.0.2.20:54321
> s2.send()
> s2.recv()
> s2.close()
>
>                                                 // bucket remains in this
>                                                 // state even though port
>                                                 // was released by s2
>                                                 tb->reuse = 0
>                                                 tb->reuseport = 0
>
> // Process A attempts to open another connection
> // when there is connection pressure from
> // 192.0.2.30:54000..54500 to 192.51.100.1:443.
> // Assume only port 54321 is still available.
>
> s3 = socket(AF_INET, SOCK_STREAM)
> s3.setsockopt(IP_BIND_ADDRESS_NO_PORT)
> s3.setsockopt(IP_LOCAL_PORT_RANGE, 54000..54500)
> s3.bind(192.0.2.30, 0)
> s3.connect(192.51.100.1, 443) -> EADDRNOTAVAIL (99)
>
> Problem
> -------
>
> We end up in a state where Process A can't reuse ephemeral port 54321 for
> as long as there are sockets, like s1, that keep the bind bucket alive. The
> bucket does not return to "reusable" state even when all sockets which
> blocked it from reuse, like s2, are gone.
>
> The ephemeral port becomes available for use again only after all sockets
> bound to it are gone and the bind bucket is destroyed.
>
> Programs which behave like Process B in this scenario - that is, binding to
> an IP address without setting IP_BIND_ADDRESS_NO_PORT - might be considered
> poorly written. However, the reality is that such implementation is not
> actually uncommon. Trying to fix each and every such program is like
> playing whack-a-mole.
>
> For instance, it could be any software using Golang's net.Dialer with
> LocalAddr provided:
>
>         dialer := &net.Dialer{
>                 LocalAddr: &net.TCPAddr{IP: srcIP},
>         }
>         conn, err := dialer.Dial("tcp4", dialTarget)
>
> Or even a ubiquitous tool like dig when using a specific local address:
>
>         $ dig -b 127.1.1.1 +tcp +short example.com
>
> Hence, we are proposing a systematic fix in the network stack itself.
>
> Solution
> --------
>
> If there is no IP address conflict with any socket bound to a given local
> port, then from the protocol's perspective, the port can be safely shared.
>
> With that in mind, modify the port search during connect(), that is
> __inet_hash_connect, to consider all bind buckets (ports) when looking for
> a local port for egress.
>
> To achieve this, add an extra walk over bhash2 buckets for the port to
> check for IP conflicts. The additional walk is not free, so perform it only
> once per port - during the second phase of conflict checking, when the
> bhash bucket is locked.
>
> We enable this changed behavior only if the IP_LOCAL_PORT_RANGE socket
> option is set. The rationale is that users are likely to care about using
> every possible local port only when they have deliberately constrained the
> ephemeral port range.
>
> Limitations
> -----------
>
> Sharing both the local IP and port with other established sockets, when the
> remote address is unique is still not possible if the bucket is in
> non-reusable state (tb->{fastreuse,fastreuseport} >= 0) because of a socket
> explicitly bound to that port.
>
> Alternatives
> ------------
>
> * Update bind bucket state on port release
>
> A valid solution to the described problem would also be to walk the bind
> bucket owners when releasing the port and recalculate the
> tb->{reuse,reuseport} state.
>
> However, in comparison to the proposed solution, this alone would not allow
> sharing the local port with other sockets bound to non-conflicting IPs for
> as long as they exist.
>
> Another downside is that we'd pay the extra cost on each unbind (more
> frequent) rather than only when connecting with IP_LOCAL_PORT_RANGE
> set (less frequent). Due to that we would also likely need to guard it
> behind a sysctl (see below).
>
> * Run your service in a dedicated netns
>
> This would also solve the problem. While we don't rule out transitioning to
> this setup in the future at a cost of shifting the complexity elsewhere.
>
> Isolating your service in a netns requires assigning it dedicated IPs for
> egress. If the egress IPs must be shared with other processes, as in our
> case, then SNAT and connection tracking on egress are required - adding
> complexity.
>
> * Guard it behind a sysctl setting instead of a socket option
>
> Users are unlikely to encounter this problem unless their workload connects
> from a severely narrowed-down ephemeral port range. Hence, paying the bind
> bucket walk cost for each connect() call doesn't seem sensible. Whereas
> with a socket option, only a limited set of connections incur the
> performance overhead.
>
> Reported-by: Lee Valentine <lvalentine@cloudflare.com>
> Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
> ---
> To: Eric Dumazet <edumazet@google.com>
> To: Paolo Abeni <pabeni@redhat.com>
> To: David S. Miller <davem@davemloft.net>
> To: Jakub Kicinski <kuba@kernel.org>
> To: Neal Cardwell <ncardwell@google.com>
> To: Kuniyuki Iwashima <kuniyu@google.com>
> Cc: netdev@vger.kernel.org
> Cc: kernel-team@cloudflare.com
> ---
> Changes in v2:
> - Fix unused var warning when CONFIG_IPV6=n
> - Convert INADDR_ANY to network byte order before comparison
> - Link to v1: https://lore.kernel.org/r/20250626120247.1255223-1-jakub@cloudflare.com
> ---
>  net/ipv4/inet_hashtables.c | 46 ++++++++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 44 insertions(+), 2 deletions(-)
>
> diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
> index ceeeec9b7290..b4c4caf3ff6c 100644
> --- a/net/ipv4/inet_hashtables.c
> +++ b/net/ipv4/inet_hashtables.c
> @@ -1005,6 +1005,43 @@ EXPORT_IPV6_MOD(inet_bhash2_reset_saddr);
>  #define INET_TABLE_PERTURB_SIZE (1 << CONFIG_INET_TABLE_PERTURB_ORDER)
>  static u32 *table_perturb;
>
> +/* True on source address conflict with another socket. False otherwise.
> + * Caller must hold hashbucket lock for this tb.
> + */
> +static inline bool check_bound(const struct sock *sk,
> +                              const struct inet_bind_bucket *tb)
> +{
> +       const struct inet_bind2_bucket *tb2;
> +
> +       hlist_for_each_entry(tb2, &tb->bhash2, bhash_node) {
> +#if IS_ENABLED(CONFIG_IPV6)
> +               const struct sock *sk2;
> +
> +               if (sk->sk_family == AF_INET6) {
> +                       if (tb2->addr_type == IPV6_ADDR_ANY ||
> +                           ipv6_addr_equal(&tb2->v6_rcv_saddr,
> +                                           &sk->sk_v6_rcv_saddr))
> +                               return true;
> +                       continue;
> +               }
> +
> +               /* Check for ipv6 non-v6only wildcard sockets */
> +               if (tb2->addr_type == IPV6_ADDR_ANY)
> +                       sk_for_each_bound(sk2, &tb2->owners)
> +                               if (!sk2->sk_ipv6only)
> +                                       return true;
> +
> +               if (tb2->addr_type != IPV6_ADDR_MAPPED)
> +                       continue;
> +#endif
> +               if (tb2->rcv_saddr == htonl(INADDR_ANY) ||
> +                   tb2->rcv_saddr == sk->sk_rcv_saddr)
> +                       return true;
> +       }
> +
> +       return false;
> +}
> +
>  int __inet_hash_connect(struct inet_timewait_death_row *death_row,
>                 struct sock *sk, u64 port_offset,
>                 u32 hash_port0,
> @@ -1070,6 +1107,8 @@ int __inet_hash_connect(struct inet_timewait_death_row *death_row,
>                         if (!inet_bind_bucket_match(tb, net, port, l3mdev))
>                                 continue;
>                         if (tb->fastreuse >= 0 || tb->fastreuseport >= 0) {
> +                               if (unlikely(local_ports))
> +                                       break; /* optimistic assumption */

I find this quite pessimistic :/

It seems you had some internal code before my recent change (86c2bc293b8130
"tcp: use RCU lookup in __inet_hash_connect()") ?

Instead, make the RCU changes so that check_bound() can be called from RCU,
and call it here before taking the decision to break off this loop.

>                                 rcu_read_unlock();
>                                 goto next_port;
>                         }
> @@ -1088,9 +1127,12 @@ int __inet_hash_connect(struct inet_timewait_death_row *death_row,
>                  */
>                 inet_bind_bucket_for_each(tb, &head->chain) {
>                         if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
> -                               if (tb->fastreuse >= 0 ||
> -                                   tb->fastreuseport >= 0)
> +                               if (tb->fastreuse >= 0 || tb->fastreuseport >= 0) {
> +                                       if (unlikely(local_ports &&
> +                                                    !check_bound(sk, tb)))
> +                                               goto ok;
>                                         goto next_port_unlock;
> +                               }
>                                 WARN_ON(hlist_empty(&tb->bhash2));
>                                 if (!check_established(death_row, sk,
>                                                        port, &tw, false,
>
> --
> 2.43.0
>

Re: [PATCH net-next v2 1/2] tcp: Consider every port when connecting with IP_LOCAL_PORT_RANGE

[Jakub Sitnicki]

On Tue, Jul 08, 2025 at 04:38 AM -07, Eric Dumazet wrote:
> On Thu, Jul 3, 2025 at 8:59 AM Jakub Sitnicki <jakub@cloudflare.com> wrote:

[...]

>> @@ -1070,6 +1107,8 @@ int __inet_hash_connect(struct inet_timewait_death_row *death_row,
>>                         if (!inet_bind_bucket_match(tb, net, port, l3mdev))
>>                                 continue;
>>                         if (tb->fastreuse >= 0 || tb->fastreuseport >= 0) {
>> +                               if (unlikely(local_ports))
>> +                                       break; /* optimistic assumption */
>
> I find this quite pessimistic :/
>
> It seems you had some internal code before my recent change (86c2bc293b8130
> "tcp: use RCU lookup in __inet_hash_connect()") ?
>
> Instead, make the RCU changes so that check_bound() can be called from RCU,
> and call it here before taking the decision to break off this loop.

Thanks for taking a look. I appreciate it.

That was intentional. Perhaps a bad call on my side.

My thinking here was that since we're already short on ephemeral ports
when these "blocked buckets" become an issue, then I wouldn't want to
dismiss a potential port in case the socket blocking it from reuse due
to src (IP, port) conflict goes away before we get to the second
(locked) phase of the search.

But then again, in hindsight, if we run into ephemeral port depletion,
then we must be under pressure from outgoing connections, so it seems
like a bad idea to put more stress on the bucket lock.

I will rework it as suggested.

>>                                 rcu_read_unlock();
>>                                 goto next_port;
>>                         }

[...]