selftests: fcnal: Test SO_DONTROUTE socket option.

The objective is to cover kernel paths that use the RTO_ONLINK flag in .flowi4_tos. This way we'll be able to safely remove this flag in the future by properly setting .flowi4_scope instead. With these selftests in place, we can make sure this won't introduce regressions.

For more context, the final objective is to convert .flowi4_tos to dscp_t, to ensure that ECN bits don't influence route and fib-rule lookups (see commit a410a0cf9885 ("ipv6: Define dscp_t and stop taking ECN bits into account in fib6-rules")).

These selftests only cover IPv4, as SO_DONTROUTE has no effect on IPv6 sockets.

v2:
- Use two different nettest options for setting SO_DONTROUTE either on the server or on the client socket.

- Use the above feature to run a single 'nettest -B' instance per test (instead of having two nettest processes for server and client).

Guillaume Nault (4):
selftests: Add SO_DONTROUTE option to nettest.
selftests: fcnal: Test SO_DONTROUTE on TCP sockets.
selftests: fcnal: Test SO_DONTROUTE on UDP sockets.
selftests: fcnal: Test SO_DONTROUTE on raw and ping sockets.

 tools/testing/selftests/net/fcnal-test.sh | 87 +++++++++++++++++++++++
 tools/testing/selftests/net/nettest.c     | 46 +++++++++++-
 2 files changed, 132 insertions(+), 1 deletion(-)

2.30.2

Add --client-dontroute and --server-dontroute options to nettest. They allow to set the SO_DONTROUTE option to the client and server sockets respectively. This will be used by the following patches to test the SO_DONTROUTE kernel behaviour with TCP and UDP.

---
v2: Use two different options for setting SO_DONTROUTE either on the
client or on the server socket.

 tools/testing/selftests/net/nettest.c | 46 ++++++++++++++++++++++++++-
 1 file changed, 45 insertions(+), 1 deletion(-)

index ee9a72982705..39a0e01f8554 100644
@@ -76,7 +76,9 @@ struct sock_args {
 		     has_grp:1,
 		     has_expected_laddr:1,
 		     has_expected_raddr:1,
-		     bind_test_only:1;
+		     bind_test_only:1,
+		     client_dontroute:1,
+		     server_dontroute:1;
 
 	unsigned short port;
 

@@ -611,6 +613,18 @@ static int set_dsfield(int sd, int version, int dsfield)
 	return 0;
 }
 
+static int set_dontroute(int sd)
+{
+	unsigned int one = 1;
+
+	if (setsockopt(sd, SOL_SOCKET, SO_DONTROUTE, &one, sizeof(one)) < 0) {
+		log_err_errno("setsockopt(SO_DONTROUTE)");
+		return -1;
+	}
+
+	return 0;
+}
+
 static int str_to_uint(const char *str, int min, int max, unsigned int *value)
 {
 	int number;

@@ -1351,6 +1365,14 @@ static int msock_init(struct sock_args *args, int server)
 	if (set_dsfield(sd, AF_INET, args->dsfield) != 0)
 		goto out_err;
 
+	if (server) {
+		if (args->server_dontroute && set_dontroute(sd) != 0)
+			goto out_err;
+	} else {
+		if (args->client_dontroute && set_dontroute(sd) != 0)
+			goto out_err;
+	}
+
 	if (args->dev && bind_to_device(sd, args->dev) != 0)
 		goto out_err;
 	else if (args->use_setsockopt &&

@@ -1482,6 +1504,9 @@ static int lsock_init(struct sock_args *args)
 	if (set_dsfield(sd, args->version, args->dsfield) != 0)
 		goto err;
 
+	if (args->server_dontroute && set_dontroute(sd) != 0)
+		goto err;
+
 	if (args->dev && bind_to_device(sd, args->dev) != 0)
 		goto err;
 	else if (args->use_setsockopt &&

@@ -1698,6 +1723,9 @@ static int connectsock(void *addr, socklen_t alen, struct sock_args *args)
 	if (set_dsfield(sd, args->version, args->dsfield) != 0)
 		goto err;
 
+	if (args->client_dontroute && set_dontroute(sd) != 0)
+		goto err;
+
 	if (args->dev && bind_to_device(sd, args->dev) != 0)
 		goto err;
 	else if (args->use_setsockopt &&

@@ -1905,10 +1933,14 @@ static int ipc_parent(int cpid, int fd, struct sock_args *args)
 #define GETOPT_STR  "sr:l:c:Q:p:t:g:P:DRn:M:X:m:d:I:BN:O:SUCi6xL:0:1:2:3:Fbqf"
 #define OPT_FORCE_BIND_KEY_IFINDEX 1001
 #define OPT_NO_BIND_KEY_IFINDEX 1002
+#define OPT_CLIENT_DONTROUTE 1003
+#define OPT_SERVER_DONTROUTE 1004
 
 static struct option long_opts[] = {
 	{"force-bind-key-ifindex", 0, 0, OPT_FORCE_BIND_KEY_IFINDEX},
 	{"no-bind-key-ifindex", 0, 0, OPT_NO_BIND_KEY_IFINDEX},
+	{"client-dontroute", 0, 0, OPT_CLIENT_DONTROUTE},
+	{"server-dontroute", 0, 0, OPT_SERVER_DONTROUTE},
 	{0, 0, 0, 0}
 };
 

@@ -1954,6 +1986,12 @@ static void print_usage(char *prog)
 	"    --no-bind-key-ifindex: Force TCP_MD5SIG_FLAG_IFINDEX off\n"
 	"    --force-bind-key-ifindex: Force TCP_MD5SIG_FLAG_IFINDEX on\n"
 	"        (default: only if -I is passed)\n"
+	"    --client-dontroute: don't use gateways for client socket: send\n"
+	"                        packets only if destination is on link (see\n"
+	"                        SO_DONTROUTE in socket(7))\n"
+	"    --server-dontroute: don't use gateways for server socket: send\n"
+	"                        packets only if destination is on link (see\n"
+	"                        SO_DONTROUTE in socket(7))\n"
 	"\n"
 	"    -g grp        multicast group (e.g., 239.1.1.1)\n"
 	"    -i            interactive mode (default is echo and terminate)\n"

@@ -2076,6 +2114,12 @@ int main(int argc, char *argv[])
 		case OPT_NO_BIND_KEY_IFINDEX:
 			args.bind_key_ifindex = -1;
 			break;
+		case OPT_CLIENT_DONTROUTE:
+			args.client_dontroute = 1;
+			break;
+		case OPT_SERVER_DONTROUTE:
+			args.server_dontroute = 1;
+			break;
 		case 'X':
 			args.client_pw = optarg;
 			break;

2.30.2

Use nettest --{client,server}-dontroute to test the kernel behaviour with TCP sockets having the SO_DONTROUTE option. Sending packets to a neighbour (on link) host, should work. When the host is behind a router, sending should fail.

Client and server sockets are tested independently, so that we can cover different TCP kernel paths.

SO_DONTROUTE also affects the syncookies path. So ipv4_tcp_dontroute() is made to work with or without syncookies, to cover both paths.

---
v2: Use 'nettest -B' instead of invoking two nettest instances for
client and server.

 tools/testing/selftests/net/fcnal-test.sh | 56 +++++++++++++++++++++++
 1 file changed, 56 insertions(+)

index 21ca91473c09..3a1f3051321f 100755
@@ -1098,6 +1098,59 @@ test_ipv4_md5_vrf__global_server__bind_ifindex0()
 	set_sysctl net.ipv4.tcp_l3mdev_accept="$old_tcp_l3mdev_accept"
 }
 
+ipv4_tcp_dontroute()
+{
+	local syncookies=$1
+	local nsa_syncookies
+	local nsb_syncookies
+	local a
+
+	#
+	# Link local connection tests (SO_DONTROUTE).
+	# Connections should succeed only when the remote IP address is
+	# on link (doesn't need to be routed through a gateway).
+	#
+
+	nsa_syncookies=$(ip netns exec "${NSA}" sysctl -n net.ipv4.tcp_syncookies)
+	nsb_syncookies=$(ip netns exec "${NSB}" sysctl -n net.ipv4.tcp_syncookies)
+	ip netns exec "${NSA}" sysctl -wq net.ipv4.tcp_syncookies=${syncookies}
+	ip netns exec "${NSB}" sysctl -wq net.ipv4.tcp_syncookies=${syncookies}
+
+	# Test with eth1 address (on link).
+
+	a=${NSB_IP}
+	log_start
+	do_run_cmd nettest -B -N "${NSA}" -O "${NSB}" -r ${a} --client-dontroute
+	log_test_addr ${a} $? 0 "SO_DONTROUTE client, syncookies=${syncookies}"
+
+	a=${NSB_IP}
+	log_start
+	do_run_cmd nettest -B -N "${NSA}" -O "${NSB}" -r ${a} --server-dontroute
+	log_test_addr ${a} $? 0 "SO_DONTROUTE server, syncookies=${syncookies}"
+
+	# Test with loopback address (routed).
+	#
+	# The client would use the eth1 address as source IP by default.
+	# Therefore, we need to use the -c option here, to force the use of the
+	# routed (loopback) address as source IP (so that the server will try
+	# to respond to a routed address and not a link local one).
+
+	a=${NSB_LO_IP}
+	log_start
+	show_hint "Should fail 'Network is unreachable' since server is not on link"
+	do_run_cmd nettest -B -N "${NSA}" -O "${NSB}" -c "${NSA_LO_IP}" -r ${a} --client-dontroute
+	log_test_addr ${a} $? 1 "SO_DONTROUTE client, syncookies=${syncookies}"
+
+	a=${NSB_LO_IP}
+	log_start
+	show_hint "Should timeout since server cannot respond (client is not on link)"
+	do_run_cmd nettest -B -N "${NSA}" -O "${NSB}" -c "${NSA_LO_IP}" -r ${a} --server-dontroute
+	log_test_addr ${a} $? 2 "SO_DONTROUTE server, syncookies=${syncookies}"
+
+	ip netns exec "${NSB}" sysctl -wq net.ipv4.tcp_syncookies=${nsb_syncookies}
+	ip netns exec "${NSA}" sysctl -wq net.ipv4.tcp_syncookies=${nsa_syncookies}
+}
+
 ipv4_tcp_novrf()
 {
 	local a

@@ -1217,6 +1270,9 @@ ipv4_tcp_novrf()
 	log_test_addr ${a} $? 1 "No server, device client, local conn"
 
 	ipv4_tcp_md5_novrf
+
+	ipv4_tcp_dontroute 0
+	ipv4_tcp_dontroute 2
 }
 
 ipv4_tcp_vrf()

2.30.2

Use nettest --client-dontroute to test the kernel behaviour with UDP sockets having the SO_DONTROUTE option. Sending packets to a neighbour (on link) host, should work. When the host is behind a router, sending should fail.

---
v2: Use 'nettest -B' instead of invoking two nettest instances for
client and server.

 tools/testing/selftests/net/fcnal-test.sh | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

index 3a1f3051321f..08b4b96cbd63 100755
@@ -1641,6 +1641,23 @@ ipv4_udp_novrf()
 	log_start
 	run_cmd nettest -D -d ${NSA_DEV} -r ${a}
 	log_test_addr ${a} $? 2 "No server, device client, local conn"
+
+	#
+	# Link local connection tests (SO_DONTROUTE).
+	# Connections should succeed only when the remote IP address is
+	# on link (doesn't need to be routed through a gateway).
+	#
+
+	a=${NSB_IP}
+	log_start
+	do_run_cmd nettest -B -D -N "${NSA}" -O "${NSB}" -r ${a} --client-dontroute
+	log_test_addr ${a} $? 0 "SO_DONTROUTE client"
+
+	a=${NSB_LO_IP}
+	log_start
+	show_hint "Should fail 'Network is unreachable' since server is not on link"
+	do_run_cmd nettest -B -D -N "${NSA}" -O "${NSB}" -r ${a} --client-dontroute
+	log_test_addr ${a} $? 1 "SO_DONTROUTE client"
 }
 
 ipv4_udp_vrf()

2.30.2

I haven't looked at the VRF cases. I don't really see what kernel path I could cover that isn't already covered by the non-vrf cases:

* From ns-B point of view, VRF_IP is just a routed IP, like NSA_LO_IP. So testing SO_DONTROUTE on a socket belonging to ns-B wouldn't cover any new kernel code path.

* From ns-A point of view, the routing table content is the same. It just has to use table $VRF_TABLE instead of main. So the test would just ensure that we jump to the right routing table, something that SO_DONTROUTE doesn't influences. But the route lookup itself is actually the same as for the non-vrf test (just using a different table). Therefore I feel that adding vrf tests for SO_DONTROUTE wouldn't bring much value.

But I may very well be missing some interesting points. If you have any VRF test case in mind, I can add them.

Use ping -r to test the kernel behaviour with raw and ping sockets having the SO_DONTROUTE option.

Since ipv4_ping_novrf() is called with different values of net.ipv4.ping_group_range, then it tests both raw and ping sockets (ping uses ping sockets if its user ID belongs to ping_group_range and raw sockets otherwise).

With both socket types, sending packets to a neighbour (on link) host, should work. When the host is behind a router, sending should fail.

---
 tools/testing/selftests/net/fcnal-test.sh | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

index 08b4b96cbd63..05b5c4af7a08 100755
@@ -584,6 +584,20 @@ ipv4_ping_novrf()
 		log_test_addr ${a} $? 0 "ping out, address bind"
 	done
 
+	#
+	# out, but don't use gateway if peer is not on link
+	#
+	a=${NSB_IP}
+	log_start
+	run_cmd ping -c 1 -w 1 -r ${a}
+	log_test_addr ${a} $? 0 "ping out (don't route), peer on link"
+
+	a=${NSB_LO_IP}
+	log_start
+	show_hint "Fails since peer is not on link"
+	run_cmd ping -c 1 -w 1 -r ${a}
+	log_test_addr ${a} $? 1 "ping out (don't route), peer not on link"
+
 	#
 	# in
 	#

2.30.2