selftests/net/openvswitch: add SCTP flow key support and test
Subsystems (4)
PATCH v8 1/1 selftests/net/openvswitch: add SCTP flow key support and test
The ovskey flow-string parser has no OVS_KEY_ATTR_SCTP entry, so a flow string containing sctp(src=.../dst=...) parses without error but silently drops the L4 key. The resulting flow carries only ipv4(proto=132), and the kernel rejects it: match_validate() in flow_netlink.c requires OVS_KEY_ATTR_SCTP when the IP protocol is IPPROTO_SCTP and returns -EINVAL for the missing key.
Register OVS_KEY_ATTR_SCTP in the parse table and add a matching selftest that verifies SCTP flow key matching (sctp src/dst port).
Also enable CONFIG_IP_SCTP in the selftest kernel config.
---
Changes from v7:
- Drop all unit tests per Ilya's feedback (tests for test code feels excessive); merge patches back into one.
- Enable CONFIG_IP_SCTP in the selftest kernel config.
- Rebase onto latest net-next; add base-commit header.
v7: https://lore.kernel.org/netdev/[email protected]/
v6: https://lore.kernel.org/netdev/[email protected]/
v5: https://lore.kernel.org/netdev/[email protected]/
v4: https://lore.kernel.org/netdev/[email protected]/
v3: https://lore.kernel.org/netdev/[email protected]/
v2: https://lore.kernel.org/netdev/[email protected]/
v1: https://lore.kernel.org/netdev/[email protected]/
.../testing/selftests/net/openvswitch/config | 1 + .../selftests/net/openvswitch/openvswitch.sh | 121 ++++++++++++++++++ .../selftests/net/openvswitch/ovs-dpctl.py | 5 + 3 files changed, 127 insertions(+)
tools/testing/selftests/net/openvswitch/config +1 -0
@@ -1,6 +1,7 @@ CONFIG_GENEVE=m CONFIG_INET_DIAG=y CONFIG_IPV6=y +CONFIG_IP_SCTP=y CONFIG_NETFILTER=y CONFIG_NET_IPGRE=m CONFIG_NET_IPGRE_DEMUX=m
tools/testing/selftests/net/openvswitch/openvswitch.sh +121 -0
133 lines not on this page · raw message
tools/testing/selftests/net/openvswitch/ovs-dpctl.py +5 -0
@@ -1984,6 +1984,11 @@ class ovskey(nla): "udp", ovskey.ovs_key_udp, ), + ( + "OVS_KEY_ATTR_SCTP", + "sctp", + ovskey.ovs_key_sctp, + ), ( "OVS_KEY_ATTR_ICMP", "icmp",
base-commit: 2fbade66245059c78daeaccfce13ecf499fffb51
Signature · 1 line
2.55.0
Aaron Conole
maintainerHi Minxi,
Quoted 66 lines · Minxi Hou
The ovskey flow-string parser has no
OVS_KEY_ATTR_SCTPentry, so a flow string containing sctp(src=.../dst=...) parses without error but silently drops the L4 key. The resulting flow carries only ipv4(proto=132), and the kernel rejects it:match_validate()inflow_netlink.crequiresOVS_KEY_ATTR_SCTPwhen the IP protocol isIPPROTO_SCTPand returns-EINVALfor the missing key.Register
OVS_KEY_ATTR_SCTPin the parse table and add a matching selftest that verifies SCTP flow key matching (sctp src/dst port).Also enable
CONFIG_IP_SCTPin the selftest kernel config.Signed-off-by Minxi Hou <[email protected]>---
Changes from v7:
- Drop all unit tests per Ilya's feedback (tests for test code feels excessive); merge patches back into one.
- EnableCONFIG_IP_SCTPin the selftest kernel config.
- Rebase onto latest net-next; add base-commit header.
v7: https://lore.kernel.org/netdev/[email protected]/
v6: https://lore.kernel.org/netdev/[email protected]/
v5: https://lore.kernel.org/netdev/[email protected]/
v4: https://lore.kernel.org/netdev/[email protected]/
v3: https://lore.kernel.org/netdev/[email protected]/
v2: https://lore.kernel.org/netdev/[email protected]/
v1: https://lore.kernel.org/netdev/[email protected]/.../testing/selftests/net/openvswitch/config | 1 + .../selftests/net/openvswitch/openvswitch.sh | 121 ++++++++++++++++++ .../selftests/net/openvswitch/ovs-dpctl.py | 5 + 3 files changed, 127 insertions(+)tools/testing/selftests/net/openvswitch/config +1
-0@@ -1,6 +1,7 @@ CONFIG_GENEVE=m CONFIG_INET_DIAG=y CONFIG_IPV6=y +CONFIG_IP_SCTP=y CONFIG_NETFILTER=y CONFIG_NET_IPGRE=m CONFIG_NET_IPGRE_DEMUX=mtools/testing/selftests/net/openvswitch/openvswitch.sh +6
-0@@ -34,6 +34,7 @@ tests=" action_set set: SET action rewrites fields trunc trunc: output truncation icmpv6 icmpv6: ICMPv6 echo type match + sctp_connect_v4 sctp: SCTP flow key matching psample psample: Sampling packets with psample" info() {@@ -611,6 +612,126 @@ test_icmpv6() { return 0 } +# Check for an SCTP endpoint via /proc, which works without sctp_diag. +sctp_eps_has() { + ip netns exec "$1" awk -v p="$2" '$6==p' /proc/net/sctp/eps | grep -q . +} +
tools/testing/selftests/net/openvswitch/openvswitch.sh test_icmpv6() { · line 629
Quoted 10 lines
tools/testing/selftests/net/openvswitch/openvswitch.sh +10
-0+# sctp_connect_v4 test +# - sctp(dst=4443) matches client-to-server INIT +# - sctp(src=4443) matches server-to-client INIT-ACK +# - remove flows and verify connection fails, reinstall and recover +test_sctp_connect_v4() { + local t="test_sctp_connect_v4" + local srv_ip=172.31.110.20 + + modprobe -q sctp 2>/dev/null || return "$ksft_skip" + socat -V 2>&1 | grep -q "define WITH_SCTP" || return "$ksft_skip"
socat has the option to run the listener with ',fork' option, which I'll talk about below.
Quoted 40 lines
tools/testing/selftests/net/openvswitch/openvswitch.sh +40
-0+ sbx_add "$t" || return $? + ovs_add_dp "$t" sctp4 || return 1 + + info "create namespaces" + for ns in client server; do + ovs_add_netns_and_veths "$t" "sctp4" "$ns" \ + "${ns:0:1}0" "${ns:0:1}1" || return 1 + done + + ip netns exec client ip addr add 172.31.110.10/24 dev c1 + ip netns exec client ip link set c1 up + ip netns exec server ip addr add "${srv_ip}/24" dev s1 + ip netns exec server ip link set s1 up + + # Probe: check if kernel supports sctp flow key. + ovs_add_flow "$t" sctp4 \ + 'in_port(1),eth(),eth_type(0x0800),ipv4(proto=132),sctp(dst=4443)' \ + '2' &>/dev/null + if [ $? -ne 0 ]; then + info "no support for sctp key - skipping" + ovs_exit_sig + return $ksft_skip + fi + ovs_del_flows "$t" sctp4 + + # ARP forwarding + ovs_add_flow "$t" sctp4 \ + 'in_port(1),eth(),eth_type(0x0806),arp()' \ + '2' || return 1 + ovs_add_flow "$t" sctp4 \ + 'in_port(2),eth(),eth_type(0x0806),arp()' \ + '1' || return 1 + + # SCTP port matching: dst for request, src for reply + ovs_add_flow "$t" sctp4 \ + 'in_port(1),eth(),eth_type(0x0800),ipv4(proto=132),sctp(dst=4443)' \ + '2' || return 1 + ovs_add_flow "$t" sctp4 \ + 'in_port(2),eth(),eth_type(0x0800),ipv4(proto=132),sctp(src=4443)' \ + '1' || return 1
tools/testing/selftests/net/openvswitch/openvswitch.sh test_icmpv6() { · line 680
Quoted 10 lines
tools/testing/selftests/net/openvswitch/openvswitch.sh +10
-0+ + ovs_netns_spawn_daemon "$t" "server" \ + socat -u SCTP4-LISTEN:4443 STDOUT + local server_pid="$pid" + ovs_wait sctp_eps_has server 4443 || return 1 + + info "verify SCTP association with port-keyed flows" + ovs_sbx "$t" ip netns exec client \ + timeout 3 socat -u STDIN "SCTP4-CONNECT:${srv_ip}:4443" </dev/null \ + || return 1
We spawn a socat instance to listen, and then after we determine the connection has succeeded, we kill everything, and create new flows. BUT, as sashiko points out, we do this multiple times, and don't even check whether the listen socket successfully comes up in the second / third ones.
I think we could use a single SCTP listen instance with the fork option as an optimization. In that case, it would better test that the flows are working versus just colliding on some possible listener race issue. What I mean is the pattern:
-- setup flows
-- connect
-- tear down flows
seems fine to me, but the respawning of the listen side feels excessive.
Sashiko seems to imply that we could also do the wait-test with longer timeout, in the case that we respawn the listener. There again I'd argue that it's better just not restarting all of the listening side each time.
Quoted 74 lines
tools/testing/selftests/net/openvswitch/openvswitch.sh +53
-0+ ovs_del_flows "$t" sctp4 + + info "verify connection fails without flows" + ovs_add_flow "$t" sctp4 \ + 'in_port(1),eth(),eth_type(0x0806),arp()' \ + '2' || return 1 + ovs_add_flow "$t" sctp4 \ + 'in_port(2),eth(),eth_type(0x0806),arp()' \ + '1' || return 1 + + kill -TERM "$server_pid" 2>/dev/null + local i=0 + while kill -0 "$server_pid" 2>/dev/null && [ "$i" -lt 5 ]; do + sleep 0.2 + i=$((i + 1)) + done + ovs_netns_spawn_daemon "$t" "server" \ + socat -u SCTP4-LISTEN:4443 STDOUT + server_pid="$pid" + ovs_wait sctp_eps_has server 4443 || return 1 + + ovs_sbx "$t" ip netns exec client \ + timeout 3 socat -u STDIN "SCTP4-CONNECT:${srv_ip}:4443" </dev/null \ + >/dev/null 2>&1 \ + && { info "connection should fail without flows" + return 1; } + + info "reinstall flows and verify recovery" + ovs_add_flow "$t" sctp4 \ + 'in_port(1),eth(),eth_type(0x0800),ipv4(proto=132),sctp(dst=4443)' \ + '2' || return 1 + ovs_add_flow "$t" sctp4 \ + 'in_port(2),eth(),eth_type(0x0800),ipv4(proto=132),sctp(src=4443)' \ + '1' || return 1 + + kill -TERM "$server_pid" 2>/dev/null + i=0 + while kill -0 "$server_pid" 2>/dev/null && [ "$i" -lt 5 ]; do + sleep 0.2 + i=$((i + 1)) + done + ovs_netns_spawn_daemon "$t" "server" \ + socat -u SCTP4-LISTEN:4443 STDOUT + server_pid="$pid" + ovs_wait sctp_eps_has server 4443 || return 1 + + ovs_sbx "$t" ip netns exec client \ + timeout 3 socat -u STDIN "SCTP4-CONNECT:${srv_ip}:4443" </dev/null \ + || return 1 + + return 0 +} + # psample test # - use psample to observe packets test_psample() {tools/testing/selftests/net/openvswitch/ovs-dpctl.py +5
-0@@ -1984,6 +1984,11 @@ class ovskey(nla): "udp", ovskey.ovs_key_udp, ), + ( + "OVS_KEY_ATTR_SCTP", + "sctp", + ovskey.ovs_key_sctp, + ), ( "OVS_KEY_ATTR_ICMP", "icmp",base-commit:
2fbade66245059c78daeaccfce13ecf499fffb51
Agreed. v9 uses a single
socat -u SCTP4-LISTEN:4443,fork STDOUT
for the whole test, so the flow rules are the only thing that changes between the three phases. Both respawn blocks go away.
Quoted 3 lines
BUT, as sashiko points out, we do this multiple times, and don't even check whether the listen socket successfully comes up in the second / third ones.
The check is there -- each respawn is followed by
ovs_wait sctp_eps_has server 4443
For reference, the sashiko run on v8 reports no open concerns; the listener restart and that wait both appear under dismissed concerns.
What is fair to say is that the check is weaker than it looks. sctp_eps_has matches on the local port alone, so across a restart it can assert only that some listener holds 4443, not that it is the new one. With a single forking listener that question does not arise, which is a better reason to make the change than the line count.
Quoted 2 lines
Sashiko seems to imply that we could also do the wait-test with longer timeout, in the case that we respawn the listener.
Not needed either way. Two of those dismissed concerns are about the restart itself, so dropping it removes the code they were about rather than widening a window around it.
That failure is unrelated to the listener. The selftest config check strips underscores before comparing keys, so CONFIG_IP_SCTP has to sort before CONFIG_IPV6 even though a plain sort puts it after. Fixed in v9, and validate_config_format.py is clean on the new file.
Thanks for the review.