[PATCH bpf-next v3] selftests/bpf: Improve connect_force_port test reliability

[PATCH bpf-next v3] selftests/bpf: Improve connect_force_port test reliability

[Varun R Mallya]

The connect_force_port test fails intermittently in CI because the
hardcoded server ports (60123/60124) may already be in use by other
tests or processes [1].

Fix this by passing port 0 to start_server(), letting the kernel assign
a free port dynamically. The actual assigned port is then propagated to
the BPF programs by writing it into the .bss map's initial value (via
bpf_map__initial_value()) before loading, so the BPF programs use the
correct backend port at runtime.

[1] https://github.com/kernel-patches/bpf/actions/runs/22697676317/job/65808536038

Suggested-by: Jiayuan Chen <jiayuan.chen@linux.dev>
Signed-off-by: Varun R Mallya <varunrmallya@gmail.com>
Reviewed-by: Sun Jian <sun.jian.kdev@gmail.com>
---
This patch implements the approach suggested by Yonghong Song in his
review of Jiayuan's v1 patch [2]: fix only the hardcoded port issue
while keeping the existing bpf_object__*() loading model, skipping
the optional skeleton refactor.

[2] https://lore.kernel.org/bpf/6555dc32-b651-4c45-adaf-e3aecb013904@linux.dev/

Changes v2->v3: Changed port type from unsigned short to __u16 and
ASSERT_OK_PTR failures now directly jump to close_bpf_object

 .../bpf/prog_tests/connect_force_port.c       | 26 ++++++++++++++++---
 .../selftests/bpf/progs/connect_force_port4.c | 10 ++++---
 .../selftests/bpf/progs/connect_force_port6.c | 10 ++++---
 3 files changed, 34 insertions(+), 12 deletions(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/connect_force_port.c b/tools/testing/selftests/bpf/prog_tests/connect_force_port.c
index 24d553109f8d..dfb7f6cf3ee1 100644
--- a/tools/testing/selftests/bpf/prog_tests/connect_force_port.c
+++ b/tools/testing/selftests/bpf/prog_tests/connect_force_port.c
@@ -53,6 +53,9 @@ static int run_test(int cgroup_fd, int server_fd, int family, int type)
 	__u16 expected_peer_port = 60000;
 	struct bpf_program *prog;
 	struct bpf_object *obj;
+	struct bpf_map *map;
+	__u16 *port_ptr;
+	size_t port_size;
 	const char *obj_file = v4 ? "connect_force_port4.bpf.o" : "connect_force_port6.bpf.o";
 	int fd, err;
 	__u32 duration = 0;
@@ -61,6 +64,21 @@ static int run_test(int cgroup_fd, int server_fd, int family, int type)
 	if (!ASSERT_OK_PTR(obj, "bpf_obj_open"))
 		return -1;
 
+	map = bpf_object__find_map_by_name(obj, ".bss");
+	if (!ASSERT_OK_PTR(map, "find bss map")) {
+		err = -EIO;
+		goto close_bpf_object;
+	}
+
+	port_ptr = bpf_map__initial_value(map, &port_size);
+	if (!ASSERT_OK_PTR(port_ptr, "get bss initial value")) {
+		err = -EIO;
+		goto close_bpf_object;
+	}
+
+	/* Auto assigns the port according to availability */
+	*port_ptr = ntohs(get_socket_local_port(server_fd));
+
 	err = bpf_object__load(obj);
 	if (!ASSERT_OK(err, "bpf_obj_load")) {
 		err = -EIO;
@@ -138,25 +156,25 @@ void test_connect_force_port(void)
 	if (CHECK_FAIL(cgroup_fd < 0))
 		return;
 
-	server_fd = start_server(AF_INET, SOCK_STREAM, NULL, 60123, 0);
+	server_fd = start_server(AF_INET, SOCK_STREAM, NULL, 0, 0);
 	if (CHECK_FAIL(server_fd < 0))
 		goto close_cgroup_fd;
 	CHECK_FAIL(run_test(cgroup_fd, server_fd, AF_INET, SOCK_STREAM));
 	close(server_fd);
 
-	server_fd = start_server(AF_INET6, SOCK_STREAM, NULL, 60124, 0);
+	server_fd = start_server(AF_INET6, SOCK_STREAM, NULL, 0, 0);
 	if (CHECK_FAIL(server_fd < 0))
 		goto close_cgroup_fd;
 	CHECK_FAIL(run_test(cgroup_fd, server_fd, AF_INET6, SOCK_STREAM));
 	close(server_fd);
 
-	server_fd = start_server(AF_INET, SOCK_DGRAM, NULL, 60123, 0);
+	server_fd = start_server(AF_INET, SOCK_DGRAM, NULL, 0, 0);
 	if (CHECK_FAIL(server_fd < 0))
 		goto close_cgroup_fd;
 	CHECK_FAIL(run_test(cgroup_fd, server_fd, AF_INET, SOCK_DGRAM));
 	close(server_fd);
 
-	server_fd = start_server(AF_INET6, SOCK_DGRAM, NULL, 60124, 0);
+	server_fd = start_server(AF_INET6, SOCK_DGRAM, NULL, 0, 0);
 	if (CHECK_FAIL(server_fd < 0))
 		goto close_cgroup_fd;
 	CHECK_FAIL(run_test(cgroup_fd, server_fd, AF_INET6, SOCK_DGRAM));
diff --git a/tools/testing/selftests/bpf/progs/connect_force_port4.c b/tools/testing/selftests/bpf/progs/connect_force_port4.c
index 27a632dd382e..d5be6a559d6a 100644
--- a/tools/testing/selftests/bpf/progs/connect_force_port4.c
+++ b/tools/testing/selftests/bpf/progs/connect_force_port4.c
@@ -14,6 +14,8 @@
 
 char _license[] SEC("license") = "GPL";
 
+__u16 port = 0;
+
 struct svc_addr {
 	__be32 addr;
 	__be16 port;
@@ -40,7 +42,7 @@ int connect4(struct bpf_sock_addr *ctx)
 	if (bpf_bind(ctx, (struct sockaddr *)&sa, sizeof(sa)) != 0)
 		return 0;
 
-	/* Rewire service 1.2.3.4:60000 to backend 127.0.0.1:60123. */
+	/* Rewire service 1.2.3.4:60000 to backend 127.0.0.1:port. */
 	if (ctx->user_port == bpf_htons(60000)) {
 		orig = bpf_sk_storage_get(&service_mapping, ctx->sk, 0,
 					  BPF_SK_STORAGE_GET_F_CREATE);
@@ -51,7 +53,7 @@ int connect4(struct bpf_sock_addr *ctx)
 		orig->port = ctx->user_port;
 
 		ctx->user_ip4 = bpf_htonl(0x7f000001);
-		ctx->user_port = bpf_htons(60123);
+		ctx->user_port = bpf_htons(port);
 	}
 	return 1;
 }
@@ -63,7 +65,7 @@ int getsockname4(struct bpf_sock_addr *ctx)
 		return 1;
 
 	/* Expose local server as 1.2.3.4:60000 to client. */
-	if (ctx->user_port == bpf_htons(60123)) {
+	if (ctx->user_port == bpf_htons(port)) {
 		ctx->user_ip4 = bpf_htonl(0x01020304);
 		ctx->user_port = bpf_htons(60000);
 	}
@@ -79,7 +81,7 @@ int getpeername4(struct bpf_sock_addr *ctx)
 		return 1;
 
 	/* Expose service 1.2.3.4:60000 as peer instead of backend. */
-	if (ctx->user_port == bpf_htons(60123)) {
+	if (ctx->user_port == bpf_htons(port)) {
 		orig = bpf_sk_storage_get(&service_mapping, ctx->sk, 0, 0);
 		if (orig) {
 			ctx->user_ip4 = orig->addr;
diff --git a/tools/testing/selftests/bpf/progs/connect_force_port6.c b/tools/testing/selftests/bpf/progs/connect_force_port6.c
index 19cad93e612f..a1a671b39083 100644
--- a/tools/testing/selftests/bpf/progs/connect_force_port6.c
+++ b/tools/testing/selftests/bpf/progs/connect_force_port6.c
@@ -13,6 +13,8 @@
 
 char _license[] SEC("license") = "GPL";
 
+__u16 port = 0;
+
 struct svc_addr {
 	__be32 addr[4];
 	__be16 port;
@@ -39,7 +41,7 @@ int connect6(struct bpf_sock_addr *ctx)
 	if (bpf_bind(ctx, (struct sockaddr *)&sa, sizeof(sa)) != 0)
 		return 0;
 
-	/* Rewire service [fc00::1]:60000 to backend [::1]:60124. */
+	/* Rewire service [fc00::1]:60000 to backend [::1]:port. */
 	if (ctx->user_port == bpf_htons(60000)) {
 		orig = bpf_sk_storage_get(&service_mapping, ctx->sk, 0,
 					  BPF_SK_STORAGE_GET_F_CREATE);
@@ -56,7 +58,7 @@ int connect6(struct bpf_sock_addr *ctx)
 		ctx->user_ip6[1] = 0;
 		ctx->user_ip6[2] = 0;
 		ctx->user_ip6[3] = bpf_htonl(1);
-		ctx->user_port = bpf_htons(60124);
+		ctx->user_port = bpf_htons(port);
 	}
 	return 1;
 }
@@ -68,7 +70,7 @@ int getsockname6(struct bpf_sock_addr *ctx)
 		return 1;
 
 	/* Expose local server as [fc00::1]:60000 to client. */
-	if (ctx->user_port == bpf_htons(60124)) {
+	if (ctx->user_port == bpf_htons(port)) {
 		ctx->user_ip6[0] = bpf_htonl(0xfc000000);
 		ctx->user_ip6[1] = 0;
 		ctx->user_ip6[2] = 0;
@@ -87,7 +89,7 @@ int getpeername6(struct bpf_sock_addr *ctx)
 		return 1;
 
 	/* Expose service [fc00::1]:60000 as peer instead of backend. */
-	if (ctx->user_port == bpf_htons(60124)) {
+	if (ctx->user_port == bpf_htons(port)) {
 		orig = bpf_sk_storage_get(&service_mapping, ctx->sk, 0, 0);
 		if (orig) {
 			ctx->user_ip6[0] = orig->addr[0];
-- 
2.53.0

Re: [PATCH bpf-next v3] selftests/bpf: Improve connect_force_port test reliability

[Jiayuan Chen]

On 3/23/26 4:11 PM, Varun R Mallya wrote:
> The connect_force_port test fails intermittently in CI because the
> hardcoded server ports (60123/60124) may already be in use by other
> tests or processes [1].
>
> Fix this by passing port 0 to start_server(), letting the kernel assign
> a free port dynamically. The actual assigned port is then propagated to
> the BPF programs by writing it into the .bss map's initial value (via
> bpf_map__initial_value()) before loading, so the BPF programs use the
> correct backend port at runtime.
>
> [1] https://github.com/kernel-patches/bpf/actions/runs/22697676317/job/65808536038
>
> Suggested-by: Jiayuan Chen <jiayuan.chen@linux.dev>
> Signed-off-by: Varun R Mallya <varunrmallya@gmail.com>
> Reviewed-by: Sun Jian <sun.jian.kdev@gmail.com>
> ---
> This patch implements the approach suggested by Yonghong Song in his
> review of Jiayuan's v1 patch [2]: fix only the hardcoded port issue
> while keeping the existing bpf_object__*() loading model, skipping
> the optional skeleton refactor.
>
> [2] https://lore.kernel.org/bpf/6555dc32-b651-4c45-adaf-e3aecb013904@linux.dev/
Thanks for picking this up.

Reviewed-by: Jiayuan Chen <jiayuan.chen@linux.dev>

Re: [PATCH bpf-next v3] selftests/bpf: Improve connect_force_port test reliability

[Varun R Mallya]

On Mon, Mar 23, 2026 at 06:40:08PM +0800, Jiayuan Chen wrote:
> Thanks for picking this up.
You're welcome :) Would love to pick up any other thing put on the backburner
as well.
> Reviewed-by: Jiayuan Chen <jiayuan.chen@linux.dev>

Re: [PATCH bpf-next v3] selftests/bpf: Improve connect_force_port test reliability

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to bpf/bpf-next.git (master)
by Martin KaFai Lau <martin.lau@kernel.org>:

On Mon, 23 Mar 2026 13:41:31 +0530 you wrote:
> The connect_force_port test fails intermittently in CI because the
> hardcoded server ports (60123/60124) may already be in use by other
> tests or processes [1].
> 
> Fix this by passing port 0 to start_server(), letting the kernel assign
> a free port dynamically. The actual assigned port is then propagated to
> the BPF programs by writing it into the .bss map's initial value (via
> bpf_map__initial_value()) before loading, so the BPF programs use the
> correct backend port at runtime.
> 
> [...]

Here is the summary with links:
  - [bpf-next,v3] selftests/bpf: Improve connect_force_port test reliability
    https://git.kernel.org/bpf/bpf-next/c/bb6da652c585

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html