Test redirection logic. All supported and unsupported redirect combinations
are tested for success and failure respectively.

BPF_MAP_TYPE_SOCKMAP
BPF_MAP_TYPE_SOCKHASH
	x
sk_msg-to-egress
sk_msg-to-ingress
sk_skb-to-egress
sk_skb-to-ingress
	x
AF_INET, SOCK_STREAM
AF_INET6, SOCK_STREAM
AF_INET, SOCK_DGRAM
AF_INET6, SOCK_DGRAM
AF_UNIX, SOCK_STREAM
AF_UNIX, SOCK_DGRAM
AF_VSOCK, SOCK_STREAM
AF_VSOCK, SOCK_SEQPACKET

Suggested-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Michal Luczaj <mhal@rbox.co>
---
 .../selftests/bpf/prog_tests/sockmap_redir.c       | 461 +++++++++++++++++++++
 1 file changed, 461 insertions(+)

diff --git a/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c b/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c
new file mode 100644
index 0000000000000000000000000000000000000000..df550759c7e50d248322be3655b02b3a21267b4a
--- /dev/null
+++ b/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c
@@ -0,0 +1,461 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Test for sockmap/sockhash redirection.
+ *
+ * BPF_MAP_TYPE_SOCKMAP
+ * BPF_MAP_TYPE_SOCKHASH
+ *	x
+ * sk_msg-to-egress
+ * sk_msg-to-ingress
+ * sk_skb-to-egress
+ * sk_skb-to-ingress
+ *	x
+ * AF_INET, SOCK_STREAM
+ * AF_INET6, SOCK_STREAM
+ * AF_INET, SOCK_DGRAM
+ * AF_INET6, SOCK_DGRAM
+ * AF_UNIX, SOCK_STREAM
+ * AF_UNIX, SOCK_DGRAM
+ * AF_VSOCK, SOCK_STREAM
+ * AF_VSOCK, SOCK_SEQPACKET
+ */
+
+#include <errno.h>
+#include <error.h>
+#include <sched.h>
+#include <stdio.h>
+#include <unistd.h>
+
+#include <netinet/in.h>
+#include <sys/socket.h>
+#include <sys/types.h>
+#include <sys/un.h>
+#include <linux/string.h>
+#include <linux/vm_sockets.h>
+
+#include <bpf/bpf.h>
+#include <bpf/libbpf.h>
+
+#include "linux/const.h"
+#include "test_progs.h"
+#include "sockmap_helpers.h"
+#include "test_sockmap_listen.skel.h"
+
+/* The meaning of SUPPORTED is "will redirect packet as expected".
+ */
+#define SUPPORTED		_BITUL(0)
+
+/* Note on sk_skb-to-ingress ->af_vsock:
+ *
+ * Peer socket may receive the packet some time after the return from sendmsg().
+ * In a typical usage scenario, recvmsg() will block until the redirected packet
+ * appears in the destination queue, or timeout if the packet was dropped. By
+ * that point, the verdict map has already been updated to reflect what has
+ * happened.
+ *
+ * But sk_skb-to-ingress/af_vsock is an unsupported combination, so no recvmsg()
+ * takes place. Which means we may race the execution of the verdict logic and
+ * read map_verd before it has been updated, i.e. we might observe
+ * map_verd[SK_DROP]=0 instead of map_verd[SK_DROP]=1.
+ *
+ * This confuses the selftest logic: if there was no packet dropped, where's the
+ * packet? So here's a heuristic: on map_verd[SK_DROP]=map_verd[SK_PASS]=0
+ * (which implies the verdict program has not been ran) just re-read the verdict
+ * map again.
+ */
+#define UNSUPPORTED_RACY_VERD	_BITUL(1)
+
+enum prog_type {
+	SK_MSG_EGRESS,
+	SK_MSG_INGRESS,
+	SK_SKB_EGRESS,
+	SK_SKB_INGRESS,
+};
+
+enum {
+	SEND_INNER = 0,
+	SEND_OUTER,
+};
+
+enum {
+	RECV_INNER = 0,
+	RECV_OUTER,
+};
+
+struct maps {
+	int in;
+	int out;
+	int verd;
+};
+
+struct combo_spec {
+	enum prog_type prog_type;
+	const char *in, *out;
+};
+
+struct redir_spec {
+	const char *name;
+	int idx_send;
+	int idx_recv;
+	enum prog_type prog_type;
+};
+
+struct socket_spec {
+	int family;
+	int sotype;
+	int send_flags;
+	int in[2];
+	int out[2];
+};
+
+static int socket_spec_pairs(struct socket_spec *s)
+{
+	return create_socket_pairs(s->family, s->sotype,
+				   &s->in[0], &s->out[0],
+				   &s->in[1], &s->out[1]);
+}
+
+static void socket_spec_close(struct socket_spec *s)
+{
+	xclose(s->in[0]);
+	xclose(s->in[1]);
+	xclose(s->out[0]);
+	xclose(s->out[1]);
+}
+
+static void get_redir_params(struct redir_spec *redir,
+			     struct test_sockmap_listen *skel,
+			     int *prog_fd, enum bpf_attach_type *attach_type,
+			     bool *ingress_flag)
+{
+	enum prog_type type = redir->prog_type;
+	struct bpf_program *prog;
+	bool sk_msg;
+
+	sk_msg = type == SK_MSG_INGRESS || type == SK_MSG_EGRESS;
+	prog = sk_msg ? skel->progs.prog_msg_verdict : skel->progs.prog_skb_verdict;
+
+	*prog_fd = bpf_program__fd(prog);
+	*attach_type = sk_msg ? BPF_SK_MSG_VERDICT : BPF_SK_SKB_VERDICT;
+	*ingress_flag = type == SK_MSG_INGRESS || type == SK_SKB_INGRESS;
+}
+
+static void try_recv(const char *prefix, int fd, int flags, bool expect_success)
+{
+	ssize_t n;
+	char buf;
+
+	errno = 0;
+	n = recv(fd, &buf, 1, flags);
+	if (n < 0 && expect_success)
+		FAIL_ERRNO("%s: unexpected failure: retval=%zd", prefix, n);
+	if (!n && !expect_success)
+		FAIL("%s: expected failure: retval=%zd", prefix, n);
+}
+
+static void handle_unsupported(int sd_send, int sd_peer, int sd_in, int sd_out,
+			       int sd_recv, int map_verd, int status)
+{
+	unsigned int drop, pass;
+	char recv_buf;
+	ssize_t n;
+
+get_verdict:
+	if (xbpf_map_lookup_elem(map_verd, &u32(SK_DROP), &drop) ||
+	    xbpf_map_lookup_elem(map_verd, &u32(SK_PASS), &pass))
+		return;
+
+	if (pass == 0 && drop == 0 && (status & UNSUPPORTED_RACY_VERD)) {
+		sched_yield();
+		goto get_verdict;
+	}
+
+	if (pass != 0) {
+		FAIL("unsupported: wanted verdict pass 0, have %u", pass);
+		return;
+	}
+
+	/* If nothing was dropped, packet should have reached the peer */
+	if (drop == 0) {
+		errno = 0;
+		n = recv_timeout(sd_peer, &recv_buf, 1, 0, IO_TIMEOUT_SEC);
+		if (n != 1)
+			FAIL_ERRNO("unsupported: packet missing, retval=%zd", n);
+	}
+
+	/* Ensure queues are empty */
+	try_recv("bpf.recv(sd_send)", sd_send, MSG_DONTWAIT, false);
+	if (sd_in != sd_send)
+		try_recv("bpf.recv(sd_in)", sd_in, MSG_DONTWAIT, false);
+
+	try_recv("bpf.recv(sd_out)", sd_out, MSG_DONTWAIT, false);
+	if (sd_recv != sd_out)
+		try_recv("bpf.recv(sd_recv)", sd_recv, MSG_DONTWAIT, false);
+}
+
+static void test_send_redir_recv(int sd_send, int send_flags, int sd_peer,
+				 int sd_in, int sd_out, int sd_recv,
+				 struct maps *maps, int status)
+{
+	unsigned int drop, pass;
+	char *send_buf = "ab";
+	char recv_buf = '\0';
+	ssize_t n, len = 1;
+
+	/* Zero out the verdict map */
+	if (xbpf_map_update_elem(maps->verd, &u32(SK_DROP), &u32(0), BPF_ANY) ||
+	    xbpf_map_update_elem(maps->verd, &u32(SK_PASS), &u32(0), BPF_ANY))
+		return;
+
+	if (xbpf_map_update_elem(maps->in, &u32(0), &u64(sd_in), BPF_NOEXIST))
+		return;
+
+	if (xbpf_map_update_elem(maps->out, &u32(0), &u64(sd_out), BPF_NOEXIST))
+		goto del_in;
+
+	/* Last byte is OOB data when send_flags has MSG_OOB bit set */
+	if (send_flags & MSG_OOB)
+		len++;
+	n = send(sd_send, send_buf, len, send_flags);
+	if (n >= 0 && n < len)
+		FAIL("incomplete send");
+	if (n < 0) {
+		/* sk_msg redirect combo not supported? */
+		if (status & SUPPORTED || errno != EACCES)
+			FAIL_ERRNO("send");
+		goto out;
+	}
+
+	if (!(status & SUPPORTED)) {
+		handle_unsupported(sd_send, sd_peer, sd_in, sd_out, sd_recv,
+				   maps->verd, status);
+		goto out;
+	}
+
+	errno = 0;
+	n = recv_timeout(sd_recv, &recv_buf, 1, 0, IO_TIMEOUT_SEC);
+	if (n != 1) {
+		FAIL_ERRNO("recv_timeout()");
+		goto out;
+	}
+
+	/* Check verdict _after_ recv(); af_vsock may need time to catch up */
+	if (xbpf_map_lookup_elem(maps->verd, &u32(SK_DROP), &drop) ||
+	    xbpf_map_lookup_elem(maps->verd, &u32(SK_PASS), &pass))
+		goto out;
+
+	if (drop != 0 || pass != 1)
+		FAIL("unexpected verdict drop/pass: wanted 0/1, have %u/%u",
+		     drop, pass);
+
+	if (recv_buf != send_buf[0])
+		FAIL("recv(): payload check, %02x != %02x", recv_buf, send_buf[0]);
+
+	if (send_flags & MSG_OOB) {
+		/* Fail reading OOB while in sockmap */
+		try_recv("bpf.recv(sd_out, MSG_OOB)", sd_out,
+			 MSG_OOB | MSG_DONTWAIT, false);
+
+		/* Remove sd_out from sockmap */
+		xbpf_map_delete_elem(maps->out, &u32(0));
+
+		/* Check that OOB was dropped on redirect */
+		try_recv("recv(sd_out, MSG_OOB)", sd_out,
+			 MSG_OOB | MSG_DONTWAIT, false);
+
+		goto del_in;
+	}
+out:
+	xbpf_map_delete_elem(maps->out, &u32(0));
+del_in:
+	xbpf_map_delete_elem(maps->in, &u32(0));
+}
+
+static int is_redir_supported(enum prog_type type, const char *in,
+			      const char *out)
+{
+	/* Matching based on strings returned by socket_kind_to_str():
+	 * tcp4, udp4, tcp6, udp6, u_str, u_dgr, v_str, v_seq
+	 * Plus a wildcard: any
+	 * Not in use: u_seq, v_dgr
+	 */
+	struct combo_spec *c, combos[] = {
+		/* Send to local: TCP -> any, but vsock */
+		{ SK_MSG_INGRESS,	"tcp",	"tcp"	},
+		{ SK_MSG_INGRESS,	"tcp",	"udp"	},
+		{ SK_MSG_INGRESS,	"tcp",	"u_str"	},
+		{ SK_MSG_INGRESS,	"tcp",	"u_dgr"	},
+
+		/* Send to egress: TCP -> TCP */
+		{ SK_MSG_EGRESS,	"tcp",	"tcp"	},
+
+		/* Ingress to egress: any -> any */
+		{ SK_SKB_EGRESS,	"any",	"any"	},
+
+		/* Ingress to local: any -> any, but vsock */
+		{ SK_SKB_INGRESS,	"any",	"tcp"	},
+		{ SK_SKB_INGRESS,	"any",	"udp"	},
+		{ SK_SKB_INGRESS,	"any",	"u_str"	},
+		{ SK_SKB_INGRESS,	"any",	"u_dgr"	},
+	};
+
+	for (c = combos; c < combos + ARRAY_SIZE(combos); c++) {
+		if (c->prog_type == type &&
+		    (!strcmp(c->in, "any") || strstarts(in, c->in)) &&
+		    (!strcmp(c->out, "any") || strstarts(out, c->out)))
+			return SUPPORTED;
+	}
+
+	return 0;
+}
+
+static int get_support_status(enum prog_type type, const char *in,
+			      const char *out)
+{
+	int status = is_redir_supported(type, in, out);
+
+	if (type == SK_SKB_INGRESS && strstarts(out, "v_"))
+		status |= UNSUPPORTED_RACY_VERD;
+
+	return status;
+}
+
+static void test_socket(enum bpf_map_type type, struct redir_spec *redir,
+			struct maps *maps, struct socket_spec *s_in,
+			struct socket_spec *s_out)
+{
+	int fd_in, fd_out, fd_send, fd_peer, fd_recv, flags, status;
+	const char *in_str, *out_str;
+	char s[MAX_TEST_NAME];
+
+	fd_in = s_in->in[0];
+	fd_out = s_out->out[0];
+	fd_send = s_in->in[redir->idx_send];
+	fd_peer = s_in->in[redir->idx_send ^ 1];
+	fd_recv = s_out->out[redir->idx_recv];
+	flags = s_in->send_flags;
+
+	in_str = socket_kind_to_str(fd_in);
+	out_str = socket_kind_to_str(fd_out);
+	status = get_support_status(redir->prog_type, in_str, out_str);
+
+	snprintf(s, sizeof(s),
+		 "%-4s %-17s %-5s %s %-5s%6s",
+		 /* hash sk_skb-to-ingress u_str → v_str (OOB) */
+		 type == BPF_MAP_TYPE_SOCKMAP ? "map" : "hash",
+		 redir->name,
+		 in_str,
+		 status & SUPPORTED ? "→" : " ",
+		 out_str,
+		 (flags & MSG_OOB) ? "(OOB)" : "");
+
+	if (!test__start_subtest(s))
+		return;
+
+	test_send_redir_recv(fd_send, flags, fd_peer, fd_in, fd_out, fd_recv,
+			     maps, status);
+}
+
+static void test_redir(enum bpf_map_type type, struct redir_spec *redir,
+		       struct maps *maps)
+{
+	struct socket_spec *s, sockets[] = {
+		{ AF_INET, SOCK_STREAM },
+		// { AF_INET, SOCK_STREAM, MSG_OOB }, /* Known to be broken */
+		{ AF_INET6, SOCK_STREAM },
+		{ AF_INET, SOCK_DGRAM },
+		{ AF_INET6, SOCK_DGRAM },
+		{ AF_UNIX, SOCK_STREAM },
+		{ AF_UNIX, SOCK_STREAM, MSG_OOB },
+		{ AF_UNIX, SOCK_DGRAM },
+		// { AF_UNIX, SOCK_SEQPACKET},	/* Unsupported BPF_MAP_UPDATE_ELEM */
+		{ AF_VSOCK, SOCK_STREAM },
+		// { AF_VSOCK, SOCK_DGRAM },	/* Unsupported socket() */
+		{ AF_VSOCK, SOCK_SEQPACKET },
+	};
+
+	for (s = sockets; s < sockets + ARRAY_SIZE(sockets); s++)
+		if (socket_spec_pairs(s))
+			goto out;
+
+	/* Intra-proto */
+	for (s = sockets; s < sockets + ARRAY_SIZE(sockets); s++)
+		test_socket(type, redir, maps, s, s);
+
+	/* Cross-proto */
+	for (int i = 0; i < ARRAY_SIZE(sockets); i++) {
+		for (int j = 0; j < ARRAY_SIZE(sockets); j++) {
+			struct socket_spec *out = &sockets[j];
+			struct socket_spec *in = &sockets[i];
+
+			/* Skip intra-proto and between variants */
+			if (out->send_flags ||
+			    (in->family == out->family &&
+			     in->sotype == out->sotype))
+				continue;
+
+			test_socket(type, redir, maps, in, out);
+		}
+	}
+out:
+	while (--s >= sockets)
+		socket_spec_close(s);
+}
+
+static void test_map(enum bpf_map_type type)
+{
+	struct redir_spec *r, redirs[] = {
+		{ "sk_msg-to-ingress", SEND_INNER, RECV_INNER, SK_MSG_INGRESS },
+		{ "sk_msg-to-egress", SEND_INNER, RECV_OUTER, SK_MSG_EGRESS },
+		{ "sk_skb-to-egress", SEND_OUTER, RECV_OUTER, SK_SKB_EGRESS },
+		{ "sk_skb-to-ingress", SEND_OUTER, RECV_INNER, SK_SKB_INGRESS },
+	};
+
+	for (r = redirs; r < redirs + ARRAY_SIZE(redirs); r++) {
+		enum bpf_attach_type attach_type;
+		struct test_sockmap_listen *skel;
+		struct maps maps;
+		int prog_fd;
+
+		skel = test_sockmap_listen__open_and_load();
+		if (!skel) {
+			FAIL("open_and_load");
+			return;
+		}
+
+		switch (type) {
+		case BPF_MAP_TYPE_SOCKMAP:
+			skel->bss->test_sockmap = true;
+			maps.out = bpf_map__fd(skel->maps.sock_map);
+			break;
+		case BPF_MAP_TYPE_SOCKHASH:
+			skel->bss->test_sockmap = false;
+			maps.out = bpf_map__fd(skel->maps.sock_hash);
+			break;
+		default:
+			FAIL("Unsupported bpf_map_type");
+			return;
+		}
+
+		maps.in = bpf_map__fd(skel->maps.nop_map);
+		maps.verd = bpf_map__fd(skel->maps.verdict_map);
+		get_redir_params(r, skel, &prog_fd, &attach_type,
+				 &skel->bss->test_ingress);
+
+		if (xbpf_prog_attach(prog_fd, maps.in, attach_type, 0))
+			return;
+
+		test_redir(type, r, &maps);
+
+		if (xbpf_prog_detach2(prog_fd, maps.in, attach_type))
+			return;
+
+		test_sockmap_listen__destroy(skel);
+	}
+}
+
+void serial_test_sockmap_redir(void)
+{
+	test_map(BPF_MAP_TYPE_SOCKMAP);
+	test_map(BPF_MAP_TYPE_SOCKHASH);
+}

-- 
2.49.0

On Fri, Apr 11, 2025 at 01:32:41PM +0200, Michal Luczaj wrote:
> +static void test_send_redir_recv(int sd_send, int send_flags, int sd_peer,
> +				 int sd_in, int sd_out, int sd_recv,
> +				 struct maps *maps, int status)
> +{
> +	unsigned int drop, pass;
> +	char *send_buf = "ab";
> +	char recv_buf = '\0';
> +	ssize_t n, len = 1;
> +	/* Zero out the verdict map */
> +	if (xbpf_map_update_elem(maps->verd, &u32(SK_DROP), &u32(0), BPF_ANY) ||
> +	    xbpf_map_update_elem(maps->verd, &u32(SK_PASS), &u32(0), BPF_ANY))
> +		return;
> +
> +	if (xbpf_map_update_elem(maps->in, &u32(0), &u64(sd_in), BPF_NOEXIST))
> +		return;
> +
> +	if (xbpf_map_update_elem(maps->out, &u32(0), &u64(sd_out), BPF_NOEXIST))
> +		goto del_in;
> +
> +	/* Last byte is OOB data when send_flags has MSG_OOB bit set */
> +	if (send_flags & MSG_OOB)
> +		len++;
> +	n = send(sd_send, send_buf, len, send_flags);
> +	if (n >= 0 && n < len)
> +		FAIL("incomplete send");
> +	if (n < 0) {
> +		/* sk_msg redirect combo not supported? */
> +		if (status & SUPPORTED || errno != EACCES)
> +			FAIL_ERRNO("send");
> +		goto out;
> +	}
> +
> +	if (!(status & SUPPORTED)) {
> +		handle_unsupported(sd_send, sd_peer, sd_in, sd_out, sd_recv,
> +				   maps->verd, status);
> +		goto out;
> +	}
> +
> +	errno = 0;
> +	n = recv_timeout(sd_recv, &recv_buf, 1, 0, IO_TIMEOUT_SEC);
> +	if (n != 1) {
> +		FAIL_ERRNO("recv_timeout()");
> +		goto out;
> +	}
I prefer multiple send and receive operations, or implementing a loop at
the outer level.

Thanks.

On 4/11/25 16:31, Jiayuan Chen wrote:
> On Fri, Apr 11, 2025 at 01:32:41PM +0200, Michal Luczaj wrote:
>> +static void test_send_redir_recv(int sd_send, int send_flags, int sd_peer,
>> +				 int sd_in, int sd_out, int sd_recv,
>> +				 struct maps *maps, int status)
>> +{
>> +	unsigned int drop, pass;
>> +	char *send_buf = "ab";
>> +	char recv_buf = '\0';
>> +	ssize_t n, len = 1;
>> +	/* Zero out the verdict map */
>> +	if (xbpf_map_update_elem(maps->verd, &u32(SK_DROP), &u32(0), BPF_ANY) ||
>> +	    xbpf_map_update_elem(maps->verd, &u32(SK_PASS), &u32(0), BPF_ANY))
>> +		return;
>> +
>> +	if (xbpf_map_update_elem(maps->in, &u32(0), &u64(sd_in), BPF_NOEXIST))
>> +		return;
>> +
>> +	if (xbpf_map_update_elem(maps->out, &u32(0), &u64(sd_out), BPF_NOEXIST))
>> +		goto del_in;
>> +
>> +	/* Last byte is OOB data when send_flags has MSG_OOB bit set */
>> +	if (send_flags & MSG_OOB)
>> +		len++;
>> +	n = send(sd_send, send_buf, len, send_flags);
>> +	if (n >= 0 && n < len)
>> +		FAIL("incomplete send");
>> +	if (n < 0) {
>> +		/* sk_msg redirect combo not supported? */
>> +		if (status & SUPPORTED || errno != EACCES)
>> +			FAIL_ERRNO("send");
>> +		goto out;
>> +	}
>> +
>> +	if (!(status & SUPPORTED)) {
>> +		handle_unsupported(sd_send, sd_peer, sd_in, sd_out, sd_recv,
>> +				   maps->verd, status);
>> +		goto out;
>> +	}
>> +
>> +	errno = 0;
>> +	n = recv_timeout(sd_recv, &recv_buf, 1, 0, IO_TIMEOUT_SEC);
>> +	if (n != 1) {
>> +		FAIL_ERRNO("recv_timeout()");
>> +		goto out;
>> +	}
> I prefer multiple send and receive operations, or implementing a loop at
> the outer level.

If you referring to MSG_OOB that "emulates" multiple send(), that's quite
deliberate: to exercise what is actually happening on
unix_stream_sendmsg(, MSG_OOB). And, well, to follow the selftests logic
in sockmap_listen.c:pairs_redir_to_connected().

Would you rather have it split anyway?

Thanks,
Michal

On Fri, Apr 11, 2025 at 01:32:41PM +0200, Michal Luczaj wrote:
> Test redirection logic. All supported and unsupported redirect combinations
> are tested for success and failure respectively.
> 
> BPF_MAP_TYPE_SOCKMAP
> BPF_MAP_TYPE_SOCKHASH
> 	x
> sk_msg-to-egress
> sk_msg-to-ingress
> sk_skb-to-egress
> sk_skb-to-ingress
> 	x
> AF_INET, SOCK_STREAM
> AF_INET6, SOCK_STREAM
> AF_INET, SOCK_DGRAM
> AF_INET6, SOCK_DGRAM
> AF_UNIX, SOCK_STREAM
> AF_UNIX, SOCK_DGRAM
> AF_VSOCK, SOCK_STREAM
> AF_VSOCK, SOCK_SEQPACKET
> 
> Suggested-by: Jakub Sitnicki <jakub@cloudflare.com>
> Signed-off-by: Michal Luczaj <mhal@rbox.co>
> ---
>  .../selftests/bpf/prog_tests/sockmap_redir.c       | 461 +++++++++++++++++++++
>  1 file changed, 461 insertions(+)
> 
> diff --git a/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c b/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c
> new file mode 100644
> index 0000000000000000000000000000000000000000..df550759c7e50d248322be3655b02b3a21267b4a
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c
> @@ -0,0 +1,461 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Test for sockmap/sockhash redirection.
> + *
> + * BPF_MAP_TYPE_SOCKMAP
> + * BPF_MAP_TYPE_SOCKHASH
> + *	x
> + * sk_msg-to-egress
> + * sk_msg-to-ingress
> + * sk_skb-to-egress
> + * sk_skb-to-ingress
> + *	x
> + * AF_INET, SOCK_STREAM
> + * AF_INET6, SOCK_STREAM
> + * AF_INET, SOCK_DGRAM
> + * AF_INET6, SOCK_DGRAM
> + * AF_UNIX, SOCK_STREAM
> + * AF_UNIX, SOCK_DGRAM
> + * AF_VSOCK, SOCK_STREAM
> + * AF_VSOCK, SOCK_SEQPACKET
> + */
> +
> +#include <errno.h>
> +#include <error.h>
> +#include <sched.h>
> +#include <stdio.h>
> +#include <unistd.h>
> +
> +#include <netinet/in.h>
> +#include <sys/socket.h>
> +#include <sys/types.h>
> +#include <sys/un.h>
> +#include <linux/string.h>
> +#include <linux/vm_sockets.h>
> +
> +#include <bpf/bpf.h>
> +#include <bpf/libbpf.h>
> +
> +#include "linux/const.h"
> +#include "test_progs.h"
> +#include "sockmap_helpers.h"
> +#include "test_sockmap_listen.skel.h"
> +
> +/* The meaning of SUPPORTED is "will redirect packet as expected".
> + */
> +#define SUPPORTED		_BITUL(0)
> +
> +/* Note on sk_skb-to-ingress ->af_vsock:
> + *
> + * Peer socket may receive the packet some time after the return from sendmsg().
> + * In a typical usage scenario, recvmsg() will block until the redirected packet
> + * appears in the destination queue, or timeout if the packet was dropped. By
> + * that point, the verdict map has already been updated to reflect what has
> + * happened.
> + *
> + * But sk_skb-to-ingress/af_vsock is an unsupported combination, so no recvmsg()
> + * takes place. Which means we may race the execution of the verdict logic and
> + * read map_verd before it has been updated, i.e. we might observe
> + * map_verd[SK_DROP]=0 instead of map_verd[SK_DROP]=1.
> + *
> + * This confuses the selftest logic: if there was no packet dropped, where's the
> + * packet? So here's a heuristic: on map_verd[SK_DROP]=map_verd[SK_PASS]=0
> + * (which implies the verdict program has not been ran) just re-read the verdict
> + * map again.
> + */
> +#define UNSUPPORTED_RACY_VERD	_BITUL(1)
> +
> +enum prog_type {
> +	SK_MSG_EGRESS,
> +	SK_MSG_INGRESS,
> +	SK_SKB_EGRESS,
> +	SK_SKB_INGRESS,
> +};
> +
> +enum {
> +	SEND_INNER = 0,
> +	SEND_OUTER,
> +};
> +
> +enum {
> +	RECV_INNER = 0,
> +	RECV_OUTER,
> +};
> +
> +struct maps {
> +	int in;
> +	int out;
> +	int verd;
> +};
> +
> +struct combo_spec {
> +	enum prog_type prog_type;
> +	const char *in, *out;
> +};
> +
> +struct redir_spec {
> +	const char *name;
> +	int idx_send;
> +	int idx_recv;
> +	enum prog_type prog_type;
> +};
> +
> +struct socket_spec {
> +	int family;
> +	int sotype;
> +	int send_flags;
> +	int in[2];
> +	int out[2];
> +};
> +
> +static int socket_spec_pairs(struct socket_spec *s)
> +{
> +	return create_socket_pairs(s->family, s->sotype,
> +				   &s->in[0], &s->out[0],
> +				   &s->in[1], &s->out[1]);
> +}
> +
> +static void socket_spec_close(struct socket_spec *s)
> +{
> +	xclose(s->in[0]);
> +	xclose(s->in[1]);
> +	xclose(s->out[0]);
> +	xclose(s->out[1]);
> +}
> +
> +static void get_redir_params(struct redir_spec *redir,
> +			     struct test_sockmap_listen *skel,
> +			     int *prog_fd, enum bpf_attach_type *attach_type,
> +			     bool *ingress_flag)
> +{
> +	enum prog_type type = redir->prog_type;
> +	struct bpf_program *prog;
> +	bool sk_msg;
> +
> +	sk_msg = type == SK_MSG_INGRESS || type == SK_MSG_EGRESS;
> +	prog = sk_msg ? skel->progs.prog_msg_verdict : skel->progs.prog_skb_verdict;
> +
> +	*prog_fd = bpf_program__fd(prog);
> +	*attach_type = sk_msg ? BPF_SK_MSG_VERDICT : BPF_SK_SKB_VERDICT;
> +	*ingress_flag = type == SK_MSG_INGRESS || type == SK_SKB_INGRESS;
> +}
> +
> +static void try_recv(const char *prefix, int fd, int flags, bool expect_success)
> +{
> +	ssize_t n;
> +	char buf;
> +
> +	errno = 0;
> +	n = recv(fd, &buf, 1, flags);
> +	if (n < 0 && expect_success)
> +		FAIL_ERRNO("%s: unexpected failure: retval=%zd", prefix, n);
> +	if (!n && !expect_success)
> +		FAIL("%s: expected failure: retval=%zd", prefix, n);
> +}
> +
> +static void handle_unsupported(int sd_send, int sd_peer, int sd_in, int sd_out,
> +			       int sd_recv, int map_verd, int status)
> +{
> +	unsigned int drop, pass;
> +	char recv_buf;
> +	ssize_t n;
> +
> +get_verdict:
> +	if (xbpf_map_lookup_elem(map_verd, &u32(SK_DROP), &drop) ||
> +	    xbpf_map_lookup_elem(map_verd, &u32(SK_PASS), &pass))
> +		return;
> +
> +	if (pass == 0 && drop == 0 && (status & UNSUPPORTED_RACY_VERD)) {
> +		sched_yield();
> +		goto get_verdict;
> +	}
> +
> +	if (pass != 0) {
> +		FAIL("unsupported: wanted verdict pass 0, have %u", pass);
> +		return;
> +	}
> +
> +	/* If nothing was dropped, packet should have reached the peer */
> +	if (drop == 0) {
> +		errno = 0;
> +		n = recv_timeout(sd_peer, &recv_buf, 1, 0, IO_TIMEOUT_SEC);
> +		if (n != 1)
> +			FAIL_ERRNO("unsupported: packet missing, retval=%zd", n);
> +	}
> +
> +	/* Ensure queues are empty */
> +	try_recv("bpf.recv(sd_send)", sd_send, MSG_DONTWAIT, false);
> +	if (sd_in != sd_send)
> +		try_recv("bpf.recv(sd_in)", sd_in, MSG_DONTWAIT, false);
> +
> +	try_recv("bpf.recv(sd_out)", sd_out, MSG_DONTWAIT, false);
> +	if (sd_recv != sd_out)
> +		try_recv("bpf.recv(sd_recv)", sd_recv, MSG_DONTWAIT, false);
> +}
> +
> +static void test_send_redir_recv(int sd_send, int send_flags, int sd_peer,
> +				 int sd_in, int sd_out, int sd_recv,
> +				 struct maps *maps, int status)
> +{
> +	unsigned int drop, pass;
> +	char *send_buf = "ab";
> +	char recv_buf = '\0';
> +	ssize_t n, len = 1;
> +
> +	/* Zero out the verdict map */
> +	if (xbpf_map_update_elem(maps->verd, &u32(SK_DROP), &u32(0), BPF_ANY) ||
> +	    xbpf_map_update_elem(maps->verd, &u32(SK_PASS), &u32(0), BPF_ANY))
> +		return;
> +
> +	if (xbpf_map_update_elem(maps->in, &u32(0), &u64(sd_in), BPF_NOEXIST))
> +		return;
> +
> +	if (xbpf_map_update_elem(maps->out, &u32(0), &u64(sd_out), BPF_NOEXIST))
> +		goto del_in;
> +
> +	/* Last byte is OOB data when send_flags has MSG_OOB bit set */
> +	if (send_flags & MSG_OOB)
> +		len++;
> +	n = send(sd_send, send_buf, len, send_flags);
> +	if (n >= 0 && n < len)
> +		FAIL("incomplete send");
> +	if (n < 0) {
> +		/* sk_msg redirect combo not supported? */
> +		if (status & SUPPORTED || errno != EACCES)
> +			FAIL_ERRNO("send");
> +		goto out;
> +	}
> +
> +	if (!(status & SUPPORTED)) {
> +		handle_unsupported(sd_send, sd_peer, sd_in, sd_out, sd_recv,
> +				   maps->verd, status);
> +		goto out;
> +	}
> +
> +	errno = 0;
> +	n = recv_timeout(sd_recv, &recv_buf, 1, 0, IO_TIMEOUT_SEC);
> +	if (n != 1) {
> +		FAIL_ERRNO("recv_timeout()");
> +		goto out;
> +	}
> +
> +	/* Check verdict _after_ recv(); af_vsock may need time to catch up */
> +	if (xbpf_map_lookup_elem(maps->verd, &u32(SK_DROP), &drop) ||
> +	    xbpf_map_lookup_elem(maps->verd, &u32(SK_PASS), &pass))
> +		goto out;
> +
> +	if (drop != 0 || pass != 1)
> +		FAIL("unexpected verdict drop/pass: wanted 0/1, have %u/%u",
> +		     drop, pass);
> +
> +	if (recv_buf != send_buf[0])
> +		FAIL("recv(): payload check, %02x != %02x", recv_buf, send_buf[0]);
> +
> +	if (send_flags & MSG_OOB) {
> +		/* Fail reading OOB while in sockmap */
> +		try_recv("bpf.recv(sd_out, MSG_OOB)", sd_out,
> +			 MSG_OOB | MSG_DONTWAIT, false);
> +
> +		/* Remove sd_out from sockmap */
> +		xbpf_map_delete_elem(maps->out, &u32(0));
> +
> +		/* Check that OOB was dropped on redirect */
> +		try_recv("recv(sd_out, MSG_OOB)", sd_out,
> +			 MSG_OOB | MSG_DONTWAIT, false);
> +
> +		goto del_in;
> +	}
> +out:
> +	xbpf_map_delete_elem(maps->out, &u32(0));
> +del_in:
> +	xbpf_map_delete_elem(maps->in, &u32(0));
> +}
> +
> +static int is_redir_supported(enum prog_type type, const char *in,
> +			      const char *out)
> +{
> +	/* Matching based on strings returned by socket_kind_to_str():
> +	 * tcp4, udp4, tcp6, udp6, u_str, u_dgr, v_str, v_seq
> +	 * Plus a wildcard: any
> +	 * Not in use: u_seq, v_dgr
> +	 */
> +	struct combo_spec *c, combos[] = {
> +		/* Send to local: TCP -> any, but vsock */
> +		{ SK_MSG_INGRESS,	"tcp",	"tcp"	},
> +		{ SK_MSG_INGRESS,	"tcp",	"udp"	},
> +		{ SK_MSG_INGRESS,	"tcp",	"u_str"	},
> +		{ SK_MSG_INGRESS,	"tcp",	"u_dgr"	},
> +
> +		/* Send to egress: TCP -> TCP */
> +		{ SK_MSG_EGRESS,	"tcp",	"tcp"	},
> +
> +		/* Ingress to egress: any -> any */
> +		{ SK_SKB_EGRESS,	"any",	"any"	},
> +
> +		/* Ingress to local: any -> any, but vsock */
> +		{ SK_SKB_INGRESS,	"any",	"tcp"	},
> +		{ SK_SKB_INGRESS,	"any",	"udp"	},
> +		{ SK_SKB_INGRESS,	"any",	"u_str"	},
> +		{ SK_SKB_INGRESS,	"any",	"u_dgr"	},
> +	};
> +
> +	for (c = combos; c < combos + ARRAY_SIZE(combos); c++) {
> +		if (c->prog_type == type &&
> +		    (!strcmp(c->in, "any") || strstarts(in, c->in)) &&
> +		    (!strcmp(c->out, "any") || strstarts(out, c->out)))
> +			return SUPPORTED;
> +	}
> +
> +	return 0;
> +}
> +
> +static int get_support_status(enum prog_type type, const char *in,
> +			      const char *out)
> +{
> +	int status = is_redir_supported(type, in, out);
> +
> +	if (type == SK_SKB_INGRESS && strstarts(out, "v_"))
> +		status |= UNSUPPORTED_RACY_VERD;
> +
> +	return status;
> +}
> +
> +static void test_socket(enum bpf_map_type type, struct redir_spec *redir,
> +			struct maps *maps, struct socket_spec *s_in,
> +			struct socket_spec *s_out)
> +{
> +	int fd_in, fd_out, fd_send, fd_peer, fd_recv, flags, status;
> +	const char *in_str, *out_str;
> +	char s[MAX_TEST_NAME];
> +
> +	fd_in = s_in->in[0];
> +	fd_out = s_out->out[0];
> +	fd_send = s_in->in[redir->idx_send];
> +	fd_peer = s_in->in[redir->idx_send ^ 1];
> +	fd_recv = s_out->out[redir->idx_recv];
> +	flags = s_in->send_flags;
> +
> +	in_str = socket_kind_to_str(fd_in);
> +	out_str = socket_kind_to_str(fd_out);
> +	status = get_support_status(redir->prog_type, in_str, out_str);
> +
> +	snprintf(s, sizeof(s),
> +		 "%-4s %-17s %-5s %s %-5s%6s",
> +		 /* hash sk_skb-to-ingress u_str → v_str (OOB) */
> +		 type == BPF_MAP_TYPE_SOCKMAP ? "map" : "hash",
> +		 redir->name,
> +		 in_str,
> +		 status & SUPPORTED ? "→" : " ",
> +		 out_str,
> +		 (flags & MSG_OOB) ? "(OOB)" : "");
> +
> +	if (!test__start_subtest(s))
> +		return;
> +
> +	test_send_redir_recv(fd_send, flags, fd_peer, fd_in, fd_out, fd_recv,
> +			     maps, status);
> +}
> +
> +static void test_redir(enum bpf_map_type type, struct redir_spec *redir,
> +		       struct maps *maps)
> +{
> +	struct socket_spec *s, sockets[] = {
> +		{ AF_INET, SOCK_STREAM },
> +		// { AF_INET, SOCK_STREAM, MSG_OOB }, /* Known to be broken */
> +		{ AF_INET6, SOCK_STREAM },
> +		{ AF_INET, SOCK_DGRAM },
> +		{ AF_INET6, SOCK_DGRAM },
> +		{ AF_UNIX, SOCK_STREAM },
> +		{ AF_UNIX, SOCK_STREAM, MSG_OOB },
> +		{ AF_UNIX, SOCK_DGRAM },
> +		// { AF_UNIX, SOCK_SEQPACKET},	/* Unsupported BPF_MAP_UPDATE_ELEM */
> +		{ AF_VSOCK, SOCK_STREAM },
> +		// { AF_VSOCK, SOCK_DGRAM },	/* Unsupported socket() */
> +		{ AF_VSOCK, SOCK_SEQPACKET },
> +	};
> +
> +	for (s = sockets; s < sockets + ARRAY_SIZE(sockets); s++)
> +		if (socket_spec_pairs(s))
> +			goto out;
> +
> +	/* Intra-proto */
> +	for (s = sockets; s < sockets + ARRAY_SIZE(sockets); s++)
> +		test_socket(type, redir, maps, s, s);
> +
> +	/* Cross-proto */
> +	for (int i = 0; i < ARRAY_SIZE(sockets); i++) {
> +		for (int j = 0; j < ARRAY_SIZE(sockets); j++) {
> +			struct socket_spec *out = &sockets[j];
> +			struct socket_spec *in = &sockets[i];
> +
> +			/* Skip intra-proto and between variants */
> +			if (out->send_flags ||
> +			    (in->family == out->family &&
> +			     in->sotype == out->sotype))
> +				continue;
> +
> +			test_socket(type, redir, maps, in, out);
> +		}
> +	}
> +out:
> +	while (--s >= sockets)
> +		socket_spec_close(s);
> +}
> +
> +static void test_map(enum bpf_map_type type)
> +{
> +	struct redir_spec *r, redirs[] = {
> +		{ "sk_msg-to-ingress", SEND_INNER, RECV_INNER, SK_MSG_INGRESS },
> +		{ "sk_msg-to-egress", SEND_INNER, RECV_OUTER, SK_MSG_EGRESS },
> +		{ "sk_skb-to-egress", SEND_OUTER, RECV_OUTER, SK_SKB_EGRESS },
> +		{ "sk_skb-to-ingress", SEND_OUTER, RECV_INNER, SK_SKB_INGRESS },
> +	};
> +
> +	for (r = redirs; r < redirs + ARRAY_SIZE(redirs); r++) {
> +		enum bpf_attach_type attach_type;
> +		struct test_sockmap_listen *skel;

Could we create a new BPF program file, such as test_sockmap_redir.c,
instead of reusing existing ones? This file would be specifically
dedicated to serving the sockmap_redir.c functionality.

I understand there are already some duplicate programs, but isn’t this
exactly the goal of our current work—to make the test examples more
comprehensive and cleaner?


Thanks.

On 4/11/25 15:17, Jiayuan Chen wrote:
> On Fri, Apr 11, 2025 at 01:32:41PM +0200, Michal Luczaj wrote:
>> Test redirection logic. All supported and unsupported redirect combinations
>> are tested for success and failure respectively.
>> ...
>> +	for (r = redirs; r < redirs + ARRAY_SIZE(redirs); r++) {
>> +		enum bpf_attach_type attach_type;
>> +		struct test_sockmap_listen *skel;
> 
> Could we create a new BPF program file, such as test_sockmap_redir.c,
> instead of reusing existing ones? This file would be specifically
> dedicated to serving the sockmap_redir.c functionality.
> 
> I understand there are already some duplicate programs, but isn’t this
> exactly the goal of our current work—to make the test examples more
> comprehensive and cleaner?

Sure, I get your point. I'll do that.

Thanks,
Michal

On Fri, Apr 11, 2025 at 01:32:41PM +0200, Michal Luczaj wrote:
> Test redirection logic. All supported and unsupported redirect combinations
> are tested for success and failure respectively.
> 
> BPF_MAP_TYPE_SOCKMAP
> BPF_MAP_TYPE_SOCKHASH
> 	x
> sk_msg-to-egress
> sk_msg-to-ingress
> sk_skb-to-egress
> sk_skb-to-ingress

Could we also add test cases for SK_PASS (and even SK_DROP)?
Previously, we encountered deadlocks and incorrect sequence issues when
the program returned SK_PASS, so explicit testing for these cases would
be helpful.

If implemented, this test would fully exercise all code paths and
demonstrate a complete example that covers every aspect of
sockmap's packet steering and connection management capabilities.
> 	x
> AF_INET, SOCK_STREAM
> AF_INET6, SOCK_STREAM
> AF_INET, SOCK_DGRAM
> AF_INET6, SOCK_DGRAM
> AF_UNIX, SOCK_STREAM
> AF_UNIX, SOCK_DGRAM
> AF_VSOCK, SOCK_STREAM
> AF_VSOCK, SOCK_SEQPACKET
> 
> Suggested-by: Jakub Sitnicki <jakub@cloudflare.com>
> Signed-off-by: Michal Luczaj <mhal@rbox.co>
> ---
>  .../selftests/bpf/prog_tests/sockmap_redir.c       | 461 +++++++++++++++++++++
>  1 file changed, 461 insertions(+)
> 
> diff --git a/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c b/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c
> new file mode 100644
> index 0000000000000000000000000000000000000000..df550759c7e50d248322be3655b02b3a21267b4a
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c
> @@ -0,0 +1,461 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Test for sockmap/sockhash redirection.
> + *
> + * BPF_MAP_TYPE_SOCKMAP
> + * BPF_MAP_TYPE_SOCKHASH
> + *	x
> + * sk_msg-to-egress
> + * sk_msg-to-ingress
> + * sk_skb-to-egress
> + * sk_skb-to-ingress
> + *	x
> + * AF_INET, SOCK_STREAM
> + * AF_INET6, SOCK_STREAM
> + * AF_INET, SOCK_DGRAM
> + * AF_INET6, SOCK_DGRAM
> + * AF_UNIX, SOCK_STREAM
> + * AF_UNIX, SOCK_DGRAM
> + * AF_VSOCK, SOCK_STREAM
> + * AF_VSOCK, SOCK_SEQPACKET
> + */
> +
> +#include <errno.h>
> +#include <error.h>
> +#include <sched.h>
> +#include <stdio.h>
> +#include <unistd.h>
> +
> +#include <netinet/in.h>
> +#include <sys/socket.h>
> +#include <sys/types.h>
> +#include <sys/un.h>
> +#include <linux/string.h>
> +#include <linux/vm_sockets.h>
> +
> +#include <bpf/bpf.h>
> +#include <bpf/libbpf.h>
> +
> +#include "linux/const.h"
> +#include "test_progs.h"
> +#include "sockmap_helpers.h"
> +#include "test_sockmap_listen.skel.h"
> +
> +/* The meaning of SUPPORTED is "will redirect packet as expected".
> + */
> +#define SUPPORTED		_BITUL(0)
> +
> +/* Note on sk_skb-to-ingress ->af_vsock:
> + *
> + * Peer socket may receive the packet some time after the return from sendmsg().
> + * In a typical usage scenario, recvmsg() will block until the redirected packet
> + * appears in the destination queue, or timeout if the packet was dropped. By
> + * that point, the verdict map has already been updated to reflect what has
> + * happened.
> + *
> + * But sk_skb-to-ingress/af_vsock is an unsupported combination, so no recvmsg()
> + * takes place. Which means we may race the execution of the verdict logic and
> + * read map_verd before it has been updated, i.e. we might observe
> + * map_verd[SK_DROP]=0 instead of map_verd[SK_DROP]=1.
> + *
> + * This confuses the selftest logic: if there was no packet dropped, where's the
> + * packet? So here's a heuristic: on map_verd[SK_DROP]=map_verd[SK_PASS]=0
> + * (which implies the verdict program has not been ran) just re-read the verdict
> + * map again.
> + */
> +#define UNSUPPORTED_RACY_VERD	_BITUL(1)
> +
> +enum prog_type {
> +	SK_MSG_EGRESS,
> +	SK_MSG_INGRESS,
> +	SK_SKB_EGRESS,
> +	SK_SKB_INGRESS,
> +};
> +
> +enum {
> +	SEND_INNER = 0,
> +	SEND_OUTER,
> +};
> +
> +enum {
> +	RECV_INNER = 0,
> +	RECV_OUTER,
> +};
> +
> +struct maps {
> +	int in;
> +	int out;
> +	int verd;
> +};
> +
> +struct combo_spec {
> +	enum prog_type prog_type;
> +	const char *in, *out;
> +};
> +
> +struct redir_spec {
> +	const char *name;
> +	int idx_send;
> +	int idx_recv;
> +	enum prog_type prog_type;
> +};
> +
> +struct socket_spec {
> +	int family;
> +	int sotype;
> +	int send_flags;
> +	int in[2];
> +	int out[2];
> +};
> +
> +static int socket_spec_pairs(struct socket_spec *s)
> +{
> +	return create_socket_pairs(s->family, s->sotype,
> +				   &s->in[0], &s->out[0],
> +				   &s->in[1], &s->out[1]);
> +}
> +
> +static void socket_spec_close(struct socket_spec *s)
> +{
> +	xclose(s->in[0]);
> +	xclose(s->in[1]);
> +	xclose(s->out[0]);
> +	xclose(s->out[1]);
> +}
> +
> +static void get_redir_params(struct redir_spec *redir,
> +			     struct test_sockmap_listen *skel,
> +			     int *prog_fd, enum bpf_attach_type *attach_type,
> +			     bool *ingress_flag)
> +{
> +	enum prog_type type = redir->prog_type;
> +	struct bpf_program *prog;
> +	bool sk_msg;
> +
> +	sk_msg = type == SK_MSG_INGRESS || type == SK_MSG_EGRESS;
> +	prog = sk_msg ? skel->progs.prog_msg_verdict : skel->progs.prog_skb_verdict;
> +
> +	*prog_fd = bpf_program__fd(prog);
> +	*attach_type = sk_msg ? BPF_SK_MSG_VERDICT : BPF_SK_SKB_VERDICT;
> +	*ingress_flag = type == SK_MSG_INGRESS || type == SK_SKB_INGRESS;
> +}
> +
> +static void try_recv(const char *prefix, int fd, int flags, bool expect_success)
> +{
> +	ssize_t n;
> +	char buf;
> +
> +	errno = 0;
> +	n = recv(fd, &buf, 1, flags);
> +	if (n < 0 && expect_success)
> +		FAIL_ERRNO("%s: unexpected failure: retval=%zd", prefix, n);
> +	if (!n && !expect_success)
> +		FAIL("%s: expected failure: retval=%zd", prefix, n);
> +}
> +
> +static void handle_unsupported(int sd_send, int sd_peer, int sd_in, int sd_out,
> +			       int sd_recv, int map_verd, int status)
> +{
> +	unsigned int drop, pass;
> +	char recv_buf;
> +	ssize_t n;
> +
> +get_verdict:
> +	if (xbpf_map_lookup_elem(map_verd, &u32(SK_DROP), &drop) ||
> +	    xbpf_map_lookup_elem(map_verd, &u32(SK_PASS), &pass))
> +		return;
> +
> +	if (pass == 0 && drop == 0 && (status & UNSUPPORTED_RACY_VERD)) {
> +		sched_yield();
> +		goto get_verdict;
> +	}
> +
> +	if (pass != 0) {
> +		FAIL("unsupported: wanted verdict pass 0, have %u", pass);
> +		return;
> +	}
> +
> +	/* If nothing was dropped, packet should have reached the peer */
> +	if (drop == 0) {
> +		errno = 0;
> +		n = recv_timeout(sd_peer, &recv_buf, 1, 0, IO_TIMEOUT_SEC);
> +		if (n != 1)
> +			FAIL_ERRNO("unsupported: packet missing, retval=%zd", n);
> +	}
> +
> +	/* Ensure queues are empty */
> +	try_recv("bpf.recv(sd_send)", sd_send, MSG_DONTWAIT, false);
> +	if (sd_in != sd_send)
> +		try_recv("bpf.recv(sd_in)", sd_in, MSG_DONTWAIT, false);
> +
> +	try_recv("bpf.recv(sd_out)", sd_out, MSG_DONTWAIT, false);
> +	if (sd_recv != sd_out)
> +		try_recv("bpf.recv(sd_recv)", sd_recv, MSG_DONTWAIT, false);
> +}
> +
> +static void test_send_redir_recv(int sd_send, int send_flags, int sd_peer,
> +				 int sd_in, int sd_out, int sd_recv,
> +				 struct maps *maps, int status)
> +{
> +	unsigned int drop, pass;
> +	char *send_buf = "ab";
> +	char recv_buf = '\0';
> +	ssize_t n, len = 1;
> +
> +	/* Zero out the verdict map */
> +	if (xbpf_map_update_elem(maps->verd, &u32(SK_DROP), &u32(0), BPF_ANY) ||
> +	    xbpf_map_update_elem(maps->verd, &u32(SK_PASS), &u32(0), BPF_ANY))
> +		return;
> +
> +	if (xbpf_map_update_elem(maps->in, &u32(0), &u64(sd_in), BPF_NOEXIST))
> +		return;
> +
> +	if (xbpf_map_update_elem(maps->out, &u32(0), &u64(sd_out), BPF_NOEXIST))
> +		goto del_in;
> +
> +	/* Last byte is OOB data when send_flags has MSG_OOB bit set */
> +	if (send_flags & MSG_OOB)
> +		len++;
> +	n = send(sd_send, send_buf, len, send_flags);
> +	if (n >= 0 && n < len)
> +		FAIL("incomplete send");
> +	if (n < 0) {
> +		/* sk_msg redirect combo not supported? */
> +		if (status & SUPPORTED || errno != EACCES)
> +			FAIL_ERRNO("send");
> +		goto out;
> +	}
> +
> +	if (!(status & SUPPORTED)) {
> +		handle_unsupported(sd_send, sd_peer, sd_in, sd_out, sd_recv,
> +				   maps->verd, status);
> +		goto out;
> +	}
> +
> +	errno = 0;
> +	n = recv_timeout(sd_recv, &recv_buf, 1, 0, IO_TIMEOUT_SEC);
> +	if (n != 1) {
> +		FAIL_ERRNO("recv_timeout()");
> +		goto out;
> +	}
> +
> +	/* Check verdict _after_ recv(); af_vsock may need time to catch up */
> +	if (xbpf_map_lookup_elem(maps->verd, &u32(SK_DROP), &drop) ||
> +	    xbpf_map_lookup_elem(maps->verd, &u32(SK_PASS), &pass))
> +		goto out;
> +
> +	if (drop != 0 || pass != 1)
> +		FAIL("unexpected verdict drop/pass: wanted 0/1, have %u/%u",
> +		     drop, pass);
> +
> +	if (recv_buf != send_buf[0])
> +		FAIL("recv(): payload check, %02x != %02x", recv_buf, send_buf[0]);
> +
> +	if (send_flags & MSG_OOB) {
> +		/* Fail reading OOB while in sockmap */
> +		try_recv("bpf.recv(sd_out, MSG_OOB)", sd_out,
> +			 MSG_OOB | MSG_DONTWAIT, false);
> +
> +		/* Remove sd_out from sockmap */
> +		xbpf_map_delete_elem(maps->out, &u32(0));
> +
> +		/* Check that OOB was dropped on redirect */
> +		try_recv("recv(sd_out, MSG_OOB)", sd_out,
> +			 MSG_OOB | MSG_DONTWAIT, false);
> +
> +		goto del_in;
> +	}
> +out:
> +	xbpf_map_delete_elem(maps->out, &u32(0));
> +del_in:
> +	xbpf_map_delete_elem(maps->in, &u32(0));
> +}
> +
> +static int is_redir_supported(enum prog_type type, const char *in,
> +			      const char *out)
> +{
> +	/* Matching based on strings returned by socket_kind_to_str():
> +	 * tcp4, udp4, tcp6, udp6, u_str, u_dgr, v_str, v_seq
> +	 * Plus a wildcard: any
> +	 * Not in use: u_seq, v_dgr
> +	 */
> +	struct combo_spec *c, combos[] = {
> +		/* Send to local: TCP -> any, but vsock */
> +		{ SK_MSG_INGRESS,	"tcp",	"tcp"	},
> +		{ SK_MSG_INGRESS,	"tcp",	"udp"	},
> +		{ SK_MSG_INGRESS,	"tcp",	"u_str"	},
> +		{ SK_MSG_INGRESS,	"tcp",	"u_dgr"	},
> +
> +		/* Send to egress: TCP -> TCP */
> +		{ SK_MSG_EGRESS,	"tcp",	"tcp"	},
> +
> +		/* Ingress to egress: any -> any */
> +		{ SK_SKB_EGRESS,	"any",	"any"	},
> +
> +		/* Ingress to local: any -> any, but vsock */
> +		{ SK_SKB_INGRESS,	"any",	"tcp"	},
> +		{ SK_SKB_INGRESS,	"any",	"udp"	},
> +		{ SK_SKB_INGRESS,	"any",	"u_str"	},
> +		{ SK_SKB_INGRESS,	"any",	"u_dgr"	},
> +	};
> +
> +	for (c = combos; c < combos + ARRAY_SIZE(combos); c++) {
> +		if (c->prog_type == type &&
> +		    (!strcmp(c->in, "any") || strstarts(in, c->in)) &&
> +		    (!strcmp(c->out, "any") || strstarts(out, c->out)))
> +			return SUPPORTED;
> +	}
> +
> +	return 0;
> +}
> +
> +static int get_support_status(enum prog_type type, const char *in,
> +			      const char *out)
> +{
> +	int status = is_redir_supported(type, in, out);
> +
> +	if (type == SK_SKB_INGRESS && strstarts(out, "v_"))
> +		status |= UNSUPPORTED_RACY_VERD;
> +
> +	return status;
> +}
> +
> +static void test_socket(enum bpf_map_type type, struct redir_spec *redir,
> +			struct maps *maps, struct socket_spec *s_in,
> +			struct socket_spec *s_out)
> +{
> +	int fd_in, fd_out, fd_send, fd_peer, fd_recv, flags, status;
> +	const char *in_str, *out_str;
> +	char s[MAX_TEST_NAME];
> +
> +	fd_in = s_in->in[0];
> +	fd_out = s_out->out[0];
> +	fd_send = s_in->in[redir->idx_send];
> +	fd_peer = s_in->in[redir->idx_send ^ 1];
> +	fd_recv = s_out->out[redir->idx_recv];
> +	flags = s_in->send_flags;
> +
> +	in_str = socket_kind_to_str(fd_in);
> +	out_str = socket_kind_to_str(fd_out);
> +	status = get_support_status(redir->prog_type, in_str, out_str);
> +
> +	snprintf(s, sizeof(s),
> +		 "%-4s %-17s %-5s %s %-5s%6s",
> +		 /* hash sk_skb-to-ingress u_str → v_str (OOB) */
> +		 type == BPF_MAP_TYPE_SOCKMAP ? "map" : "hash",
> +		 redir->name,
> +		 in_str,
> +		 status & SUPPORTED ? "→" : " ",
> +		 out_str,
> +		 (flags & MSG_OOB) ? "(OOB)" : "");
> +
> +	if (!test__start_subtest(s))
> +		return;
> +
> +	test_send_redir_recv(fd_send, flags, fd_peer, fd_in, fd_out, fd_recv,
> +			     maps, status);
> +}
> +
> +static void test_redir(enum bpf_map_type type, struct redir_spec *redir,
> +		       struct maps *maps)
> +{
> +	struct socket_spec *s, sockets[] = {
> +		{ AF_INET, SOCK_STREAM },
> +		// { AF_INET, SOCK_STREAM, MSG_OOB }, /* Known to be broken */
> +		{ AF_INET6, SOCK_STREAM },
> +		{ AF_INET, SOCK_DGRAM },
> +		{ AF_INET6, SOCK_DGRAM },
> +		{ AF_UNIX, SOCK_STREAM },
> +		{ AF_UNIX, SOCK_STREAM, MSG_OOB },
> +		{ AF_UNIX, SOCK_DGRAM },
> +		// { AF_UNIX, SOCK_SEQPACKET},	/* Unsupported BPF_MAP_UPDATE_ELEM */
> +		{ AF_VSOCK, SOCK_STREAM },
> +		// { AF_VSOCK, SOCK_DGRAM },	/* Unsupported socket() */
> +		{ AF_VSOCK, SOCK_SEQPACKET },
> +	};
> +
> +	for (s = sockets; s < sockets + ARRAY_SIZE(sockets); s++)
> +		if (socket_spec_pairs(s))
> +			goto out;
> +
> +	/* Intra-proto */
> +	for (s = sockets; s < sockets + ARRAY_SIZE(sockets); s++)
> +		test_socket(type, redir, maps, s, s);
> +
> +	/* Cross-proto */
> +	for (int i = 0; i < ARRAY_SIZE(sockets); i++) {
> +		for (int j = 0; j < ARRAY_SIZE(sockets); j++) {
> +			struct socket_spec *out = &sockets[j];
> +			struct socket_spec *in = &sockets[i];
> +
> +			/* Skip intra-proto and between variants */
> +			if (out->send_flags ||
> +			    (in->family == out->family &&
> +			     in->sotype == out->sotype))
> +				continue;
> +
> +			test_socket(type, redir, maps, in, out);
> +		}
> +	}
> +out:
> +	while (--s >= sockets)
> +		socket_spec_close(s);
> +}
> +
> +static void test_map(enum bpf_map_type type)
> +{
> +	struct redir_spec *r, redirs[] = {
> +		{ "sk_msg-to-ingress", SEND_INNER, RECV_INNER, SK_MSG_INGRESS },
> +		{ "sk_msg-to-egress", SEND_INNER, RECV_OUTER, SK_MSG_EGRESS },
> +		{ "sk_skb-to-egress", SEND_OUTER, RECV_OUTER, SK_SKB_EGRESS },
> +		{ "sk_skb-to-ingress", SEND_OUTER, RECV_INNER, SK_SKB_INGRESS },
> +	};
> +
> +	for (r = redirs; r < redirs + ARRAY_SIZE(redirs); r++) {
> +		enum bpf_attach_type attach_type;
> +		struct test_sockmap_listen *skel;
> +		struct maps maps;
> +		int prog_fd;
> +
> +		skel = test_sockmap_listen__open_and_load();
> +		if (!skel) {
> +			FAIL("open_and_load");
> +			return;
> +		}
> +
> +		switch (type) {
> +		case BPF_MAP_TYPE_SOCKMAP:
> +			skel->bss->test_sockmap = true;
> +			maps.out = bpf_map__fd(skel->maps.sock_map);
> +			break;
> +		case BPF_MAP_TYPE_SOCKHASH:
> +			skel->bss->test_sockmap = false;
> +			maps.out = bpf_map__fd(skel->maps.sock_hash);
> +			break;
> +		default:
> +			FAIL("Unsupported bpf_map_type");
> +			return;
> +		}
> +
> +		maps.in = bpf_map__fd(skel->maps.nop_map);
> +		maps.verd = bpf_map__fd(skel->maps.verdict_map);
> +		get_redir_params(r, skel, &prog_fd, &attach_type,
> +				 &skel->bss->test_ingress);
> +
> +		if (xbpf_prog_attach(prog_fd, maps.in, attach_type, 0))
> +			return;
> +
> +		test_redir(type, r, &maps);
> +
> +		if (xbpf_prog_detach2(prog_fd, maps.in, attach_type))
> +			return;
> +
> +		test_sockmap_listen__destroy(skel);
> +	}
> +}
> +
> +void serial_test_sockmap_redir(void)
> +{
> +	test_map(BPF_MAP_TYPE_SOCKMAP);
> +	test_map(BPF_MAP_TYPE_SOCKHASH);
> +}
> 
> -- 
> 2.49.0
> 

On Fri, Apr 11, 2025 at 09:09 PM +08, Jiayuan Chen wrote:
> On Fri, Apr 11, 2025 at 01:32:41PM +0200, Michal Luczaj wrote:
>> Test redirection logic. All supported and unsupported redirect combinations
>> are tested for success and failure respectively.
>> 
>> BPF_MAP_TYPE_SOCKMAP
>> BPF_MAP_TYPE_SOCKHASH
>> 	x
>> sk_msg-to-egress
>> sk_msg-to-ingress
>> sk_skb-to-egress
>> sk_skb-to-ingress
>
> Could we also add test cases for SK_PASS (and even SK_DROP)?
> Previously, we encountered deadlocks and incorrect sequence issues when
> the program returned SK_PASS, so explicit testing for these cases would
> be helpful.
>
> If implemented, this test would fully exercise all code paths and
> demonstrate a complete example that covers every aspect of
> sockmap's packet steering and connection management capabilities.

This could easily be a follow up in my mind.

[...]

On 4/11/25 19:54, Jakub Sitnicki wrote:
> On Fri, Apr 11, 2025 at 09:09 PM +08, Jiayuan Chen wrote:
>> On Fri, Apr 11, 2025 at 01:32:41PM +0200, Michal Luczaj wrote:
>>> Test redirection logic. All supported and unsupported redirect combinations
>>> are tested for success and failure respectively.
>>>
>>> BPF_MAP_TYPE_SOCKMAP
>>> BPF_MAP_TYPE_SOCKHASH
>>> 	x
>>> sk_msg-to-egress
>>> sk_msg-to-ingress
>>> sk_skb-to-egress
>>> sk_skb-to-ingress
>>
>> Could we also add test cases for SK_PASS (and even SK_DROP)?
>> Previously, we encountered deadlocks and incorrect sequence issues when
>> the program returned SK_PASS, so explicit testing for these cases would
>> be helpful.
>>
>> If implemented, this test would fully exercise all code paths and
>> demonstrate a complete example that covers every aspect of
>> sockmap's packet steering and connection management capabilities.
> 
> This could easily be a follow up in my mind.
> 
> [...]

Yeah, I wouldn't mind doing this in multiple steps.

That said, with SK_PASS/SK_DROP involved, are we sticking with the name
"sockmap_redir"?