On 08.08.2024 18:45, Dumitru Ceara wrote:
On 8/8/24 11:14, Vladislav Odintsov wrote:On 08.08.2024 15:51,[email protected] wrote:On Thu, 2024-08-08 at 10:46 +0200, Dumitru Ceara wrote:On 8/8/24 10:42, Vladislav Odintsov wrote:On 08.08.2024 03:16, Dumitru Ceara wrote:Re-adding the dev list.On 8/7/24 18:12, Vladislav Odintsov wrote:Hi Dumitru,Hi Vladislav,I'd like to add some thoughts to your inputs.Thanks for that, I added some more of my own below. :)On 06.08.2024 19:23, Dumitru Ceara wrote:Hi Martin, Sorry, I was reviewing V3 but I was slow in actually sending out the email. On 8/6/24 13:17, Martin Kalcok wrote:This change adds a 'bgp-redirect' option to LRP that allows redirecting of BGP control plane traffic to an arbitrary LSP in its peer LS. The option expects a string with a LSP name. When set, any traffic entering LS that's destined for any of the LRP's IP addresses (including IPv6 LLA) is redirected to the LSP specified in the option's value. This enables external BGP daemons to listen on an interface bound to a LSP and effectively act as if they were listening on (and speaking from) LRP's IP address. Signed-off-by: Martin Kalcok<[email protected]> ---Strictly on this version of the patch, and with the thought in mind that we might want to consider this feature experimental [0] and maybe change it (NB too) in the future, I left a few comments inline. With those addressed I think the patch looks OK to me. [0] https://patchwork.ozlabs.org/project/ovn/patch/[email protected]/ Now, for the point when we'd remove the "experimental" tag: In general, it makes me a bit uneasy that we have to share the MAC and IP between the LRP and the VIF of logical switch port that's connected to the same switch as the LRP. I was thinking of alternatives for the future and the only things I could come up with until now are: 1. Create a separate, "disconnected" logical switch: ovn-nbctl ls-add bgp Add the bgp-daemon port to it: ovn-nbctl lsp-add bgp bgp-daemon. Then we don't need "unknown" either, I think. But it's not possible today to move packets from the ingress pipeline of a logical switch ("public" in this test) to the egress pipeline of a different logical switch ("bgp" in my example). It also feels icky to implement that. 2. Add the ability to bind an OVS port directly to a logical router port: then we could do the same type of redirect you do here but instead in the logical router pipeline. The advantage is we don't have to drop any non-bgp traffic in the switch pipeline. The switch keeps functioning as it does today. Maybe an advantage of this second alternative would be that we can easily attach a filtering option to this LRP (e.g., LRP.options:control-traffic-filter) to allow us to be more flexible in what kind of traffic we forward to the actuall routing protocol daemon that runs behind that OVS port - Vladislav also mentioned during the meeting it might be interesting to forward BFD packets to the FRR (or whatever application implements the routing protocol) instance too.The idea to be able to bind LRP to OVS port sounds very interesting to me. But with a note that it's not a pure "bind", but a partial: as you wrote with some "filter" applied to pass not all the traffic. The main idea here is to pass only control plane traffic destined to LRP'sAs we're discussing on the other thread (Martin pointed it out) we also probably need to involve conntrack and allow replies to any kind of traffic initiated by the entity behind the LRP's VIF (e.g., BGP sessions initiated from the host).addresses. Other than that seems odd. Transit traffic should be remained in LR pipeline otherwise it will have no difference with a regular VIF LSP.Definitely, traffic that needs to be DNATed (DNAT/unSNAT rules or LB rules that will change the destination IP from LRP IP to something else) should not be affected. All other "transit" traffic doesn't have LRP IP, does it?You're right. @Dumitru, @Martin, what if we just redirect all traffic destined to LRP IPs to redirect port? Is there any drawbacks? For security it is possible to optionally use ACLs with or without conntrack. It's up to user/CMS. This seems quite simple in the code and very flexible. So no additional flows seems to be needed in future to support any other routing protocols (or for another possible non-routing usecases).Won't this break all "transit" traffic that has destination IP the LRP IP (DNAT, LB), etc?I'm afraid so. I guess I was just tired yesterday when I made that proposal for general redirect. Given that the redirect is implemented in the LS pipeline, it would eat up all the traffic for LRP's IP before the DNAT/unSNAT occurs in the LR pipeline. I'll give it a quick test in case I'm wrong here again.Oops, this will definitely break that cases. For some reason I thought only about routing cases.If that's the case, I'll try the conntrack approach and update you all asap. Martin.Having a filter (or match like in ACL or Logical_Router_Policies) looks more flexible in terms of used protocols. This can coexist with proposal from current patch, where the flow is not programmable from user.I think so too. Regards, DumitruBut again, if we consider the current feature "experimental", we can spend more time during the next release cycle to figure out what's best.v4 of this patch renames the feature from "bgp-mirror" to "bgp-redirect" as discussed during community meeting. northd/northd.c | 108 ++++++++++++++++++++++++++++++++++++++++ northd/ovn-northd.8.xml | 23 +++++++++ ovn-nb.xml | 14 ++++++ tests/ovn-northd.at | 58 +++++++++++++++++++++ tests/system-ovn.at | 86 ++++++++++++++++++++++++++++++++ 5 files changed, 289 insertions(+) diff --git a/northd/northd.c b/northd/northd.c index 4353df07d..088104f25 100644 --- a/northd/northd.c +++ b/northd/northd.c @@ -13048,6 +13048,113 @@ build_arp_resolve_flows_for_lrp(struct ovn_port *op, } }+static void+build_bgp_redirect_rule__( + const char *s_addr, const char *bgp_port_name, bool is_ipv6, + struct ovn_port *ls_peer, struct lflow_table *lflows, + struct ds *match, struct ds *actions) +{ + int ip_ver = is_ipv6 ? 6 : 4; + /* Redirect packets in the input pipeline destined for LR's IP to + * the port specified in 'bgp-redirect' option. + */ + ds_clear(match); + ds_clear(actions); + ds_put_format(match, "ip%d.dst == %s && tcp.dst == 179", ip_ver, s_addr); + ds_put_format(actions, "outport = \"%s\"; output;", bgp_port_name); + ovn_lflow_add(lflows, ls_peer->od, S_SWITCH_IN_L2_LKUP, 100, + ds_cstr(match), + ds_cstr(actions), + ls_peer->lflow_ref); + + + /* Drop any traffic originating from 'bgp-redirect' port that does + * not originate from BGP daemon port. This blocks unnecessary + * traffic like ARP broadcasts or IPv6 router solicitation packets + * from the dummy 'bgp-redirect' port. + */ + ds_clear(match); + ds_put_format(match, "inport == \"%s\"", bgp_port_name); + ovn_lflow_add(lflows, ls_peer->od, S_SWITCH_IN_CHECK_PORT_SEC, 80, + ds_cstr(match), + REGBIT_PORT_SEC_DROP " = 1; next;", + ls_peer->lflow_ref);Not a huge deal but this logical flow gets regenerated multiple times for each bgp_port_name (for each LRP address). In the end we'll only install one in the database so there's no impact on SB. But maybe we should consider moving this part out of this function.+ + ds_put_format(match, + " && ip%d.src == %s && tcp.src == 179", + ip_ver, + s_addr); + ovn_lflow_add(lflows, ls_peer->od, S_SWITCH_IN_CHECK_PORT_SEC, 81, + ds_cstr(match), + REGBIT_PORT_SEC_DROP " = check_in_port_sec(); next;", + ls_peer->lflow_ref);In your system test we set 'addresses = "unknown"' for 'bgp_port_name'. And I guess we kind of have to do it like that all the time when using this feature. I'm trying to think of scenarios in which users set LSP.addresses = 'unknown' and LSP.port-security = '<some- addresses>'. I'm not sure I saw those until now but maybe I'm wrong.I'm also interested whether this LSP port with unknown addresses will get all the broadcast traffic? @Martin, didn't you look for ARP behavior? If we have one LS with three LSPs: - 1 router type LSP with MAC 00:00:00:00:00:01 and IP 169.254.0.1/30 - one "vif" LSP with "unknown" addresses set and configured underlying OVS port with same MAC and IP set - one localnet port for external connectivity. If some party from localnet side sends broadcast ARP request "Who has 169.254.0.1? Answer <...>", will this ARP request got flooded to both LRP and LSP with "unknown" addresses? If yes, will the sender receive 2 ARP replies?+} + +static void +build_lrouter_bgp_redirect( + struct ovn_port *op, struct lflow_table *lflows, + struct ds *match, struct ds *actions) +{ + /* LRP has to have a peer.*/ + if (op->peer == NULL) { + return; + } + /* LRP has to have NB record.*/ + if (op->nbrp == NULL) { + return; + } + + /* Proceed only for LRPs that have 'bgp-redirect' option set. Value of this + * option is the name of LSP to which a BGP traffic will be redirected. + */ + const char *bgp_port = smap_get(&op->nbrp->options, "bgp-redirect"); + if (bgp_port == NULL) { + return; + } + + /* Ensure that LSP, to which the BGP traffic is redirected, exists.*/ + struct ovn_port *peer_lsp; + bool bgp_port_exists = false; + HMAP_FOR_EACH (peer_lsp, dp_node, &op->peer->od-ports) {+ size_t peer_lsp_s = strlen(peer_lsp->key); + if (peer_lsp_s == strlen(bgp_port) + && !strncmp(peer_lsp->key, bgp_port, peer_lsp_s)){ + bgp_port_exists = true; + break; + } + } + + if (!bgp_port_exists) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "Option 'bgp-redirect' set on Logical Router Port " + "'%s' refers to non-existent Logical Switch Port. " + "BGP redirecting won't be configured.", + op->key); + return; + } + + if (op->cr_port != NULL) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "Option 'bgp-redirect' is not supported on" + " Distributed Gateway Port '%s'", op->key); + return; + } + + /* Mirror traffic destined for LRP's IPs and default BGP port + * to the port defined in 'bgp-redirect' option. + */ + for (size_t i = 0; i < op->lrp_networks.n_ipv4_addrs; i++) { + const char *ip_s = op-lrp_networks.ipv4_addrs[i].addr_s;+ build_bgp_redirect_rule__(ip_s, bgp_port, false, op->peer, lflows, + match, actions); + }I'm thinking more and more that having a "match" expression in a "NB.Logical_Router_Port.options:redirect-match" option it a good choice.Would the match be enforced only for redirected traffic (from outside to bgp target LSP)?
I guess I didn't get your question, sorry.
LRP can have multiple IP addresses and enabling redirect for all of them could be implicit for the user and may contain security concerns. As an example: LR can have configured LB with tcp/179. If user configured peering IP-address (which can be in a link-local address range) and also added an IP address for Load Balancer in a `networks` list of a LRP. Enabling BGP redirect for all IPs of LRP will break load balancing for tcp/179. While it is not needed for load balancers to assign their VIP to networks, there is no validations, which prohibit doing so. It isI think this is actually something that ovn-kubernetes does in their deployments for cluster-IP services. And I suspect that today there's nothing preventing users from creating cluster-IP services for TCP/179.
So, this proofs that this is a possible problem if enabled redirect and if load balancer's IP is configured to be in networks list.
In any case this doesn't prevent from current implementation to merged as "experimental". I'm just started discussion about a GA solution.
still possible and can be difficult to catch such behavior for redirected traffic with Load Balancers.+ for (size_t i = 0; i < op->lrp_networks.n_ipv6_addrs; i++) { + const char *ip_s = op-lrp_networks.ipv6_addrs[i].addr_s;+ build_bgp_redirect_rule__(ip_s, bgp_port, true, op->peer, lflows, + match, actions); + } +} + /* This function adds ARP resolve flows related to a LSP. */ static void build_arp_resolve_flows_for_lsp( @@ -16003,6 +16110,7 @@ build_lswitch_and_lrouter_iterate_by_lrp(struct ovn_port *op, lsi->meter_groups, op-lflow_ref);build_lrouter_icmp_packet_toobig_admin_flows(op, lsi-lflows, &lsi->match,&lsi-actions, op->lflow_ref);+ build_lrouter_bgp_redirect(op, lsi->lflows, &lsi-match, &lsi->actions);}static void *diff --git a/northd/ovn-northd.8.xml b/northd/ovn- northd.8.xml index b06b09ac5..4cbdf8f74 100644 --- a/northd/ovn-northd.8.xml +++ b/northd/ovn-northd.8.xml @@ -284,6 +284,21 @@ dropped in the next stage. </li>+ <li>+ For each logical port that's defined as a target of BGP redirecting + (via <code>bgp-redirect</code> option set on Logical Router Port), a + filter is set in place that disallows any traffic entering this port + that does not originate from Logical Router Port's IPs and default BGP + port (TCP 179). This filtering is achieved by two rules. First rule has + priority 81, it matches on <code>inport == <var>BGP_MIRROR_PORT</var> + && ip.src == <var>LRP_IP</var> && tcp.src == 179</code> + and allows traffic further with <code>REGBIT_PORT_SEC_DROP" = + check_in_port_sec(); next;</code>. Second rule with priority 80 matches + the rest of the traffic from that port and sets + <code>REGBIT_PORT_SEC_DROP" = 1; next;"</code> so that the packets are + dropped in the next stage. + </li> + <li> For each (enabled) vtep logical port, a priority 70 flow is added which matches on all packets and applies the action @@ -1949,6 +1964,14 @@ output; on the logical switch. </li>+ <li>+ For any logical port that's defined as a target of BGP redirecting (via + <code>bgp-redirect</code> option set on Logical Router Port), a + priority-100 flow is added that redirects traffic for Logical Router + Port IPs (including IPv6 LLA) and TCP port 179, to the targeted + logical port. + </li> + <li> Priority-90 flows for transit switches that forward registered IP multicast traffic to their corresponding multicast group , which diff --git a/ovn-nb.xml b/ovn-nb.xml index 9552534f6..2d2f5c9e0 100644 --- a/ovn-nb.xml +++ b/ovn-nb.xml @@ -3440,6 +3440,20 @@ or </p> </column>+ <column name="options" key="bgp-redirect"type='{"type": "string"}'> + <p> + This option expects a name of a Logical Switch Port that's present + in the peer's Logical Switch. If set, it causes any traffic + that's destined for Logical Router Port's IP addresses (including + its IPv6 LLA) and the default BGP port (TCP 179), to be redirected + to the specified Logical Switch Port. + + This allows external BGP daemon to be bound to a port in OVN's + Logical Switch and act as if it was listening on Logical Router + Port's IP address. + </p> + </column> + <column name="options" key="gateway_mtu_bypass"> <p> When configured, represents a match expression, in the same diff --git a/tests/ovn-northd.at b/tests/ovn-northd.at index a389d1988..df1e924ab 100644 --- a/tests/ovn-northd.at +++ b/tests/ovn-northd.at @@ -12721,3 +12721,61 @@ AT_CHECK([ovn-sbctl dump-flows lr | grep lr_in_dnat | ovn_strip_lflows], [0], [dAT_CLEANUP]) + +OVN_FOR_EACH_NORTHD_NO_HV([ +AT_SETUP([BGP control plane redirect]) +ovn_start + +check ovn-sbctl chassis-add hv1 geneve 127.0.0.1 + +check ovn-nbctl lr-add lr -- \ + lrp-add lr lr-ls 02:ac:10:01:00:01 172.16.1.1/24 +check ovn-nbctl --wait=sb set logical_router lr options:chassis=hv1 + +check ovn-nbctl ls-add ls -- \ + lsp-add ls ls-lr -- \ + lsp-set-type ls-lr router -- \ + lsp-set-addresses ls-lr router -- \ + lsp-set-options ls-lr router-port=lr-ls + +check ovn-nbctl lsp-add ls lsp-bgp -- \ + lsp-set-addresses lsp-bgp unknown + +# Function that ensures that no BGP redirect ruels are installed. +check_no_bgp_redirect() { + AT_CHECK([ovn-sbctl dump-flows ls | grep ls_in_l2_lkup | grep "tcp.dst == 179" | wc -l], [0], [0 +]) + + AT_CHECK([ovn-sbctl dump-flows ls | grep ls_in_check_port_sec | grep -E "priority=80|priority=81" | wc -l], [0], [0 +]) +} + +# By default, no rules related to BGP redirect are present +check_no_bgp_redirect + +# Set "lsp-bgp" port as target of BGP control plane redirected traffic +check ovn-nbctl --wait=sb set logical_router_port lr-ls options:bgp-redirect=lsp-bgp + +# Check that BGP control plane traffic is redirected to "lsp-bgp" +AT_CHECK([ovn-sbctl dump-flows ls | grep ls_in_l2_lkup | grep "tcp.dst == 179" | ovn_strip_lflows], [0], [dnl + table=??(ls_in_l2_lkup ), priority=100 , match=(ip4.dst == 172.16.1.1 && tcp.dst == 179), action=(outport = "lsp-bgp"; output;) + table=??(ls_in_l2_lkup ), priority=100 , match=(ip6.dst == fe80::ac:10ff:fe01:1 && tcp.dst == 179), action=(outport = "lsp-bgp"; output;) +]) + +# Check that only BGP-related traffic is accepted on "lsp- bgp" port +AT_CHECK([ovn-sbctl dump-flows ls | grep ls_in_check_port_sec | grep -E "priority=80|priority=81" | ovn_strip_lflows], [0], [dnl + table=??(ls_in_check_port_sec), priority=80 , match=(inport == "lsp-bgp"), action=(reg0[[15]] = 1; next;) + table=??(ls_in_check_port_sec), priority=81 , match=(inport == "lsp-bgp" && ip4.src == 172.16.1.1 && tcp.src == 179), action=(reg0[[15]] = check_in_port_sec(); next;) + table=??(ls_in_check_port_sec), priority=81 , match=(inport == "lsp-bgp" && ip6.src == fe80::ac:10ff:fe01:1 && tcp.src == 179), action=(reg0[[15]] = check_in_port_sec(); next;) +]) + +# Remove 'bgp-redirect' option from LRP and check that rules are removed +check ovn-nbctl --wait=sb remove logical_router_port lr-ls options bgp-redirect +check_no_bgp_redirect + +# Set non-existent LSP as target of 'bgp-redirect' and check that no rules are added +check ovn-nbctl --wait=sb set logical_router_port lr-ls options:bgp-redirect=lsp-foo +check_no_bgp_redirect + +AT_CLEANUP +]) diff --git a/tests/system-ovn.at b/tests/system-ovn.at index c24ede7c5..c09c1136e 100644 --- a/tests/system-ovn.at +++ b/tests/system-ovn.at @@ -13027,3 +13027,89 @@ OVS_TRAFFIC_VSWITCHD_STOP(["/failed to query port patch- .*/d /connection dropped.*/d"]) AT_CLEANUP ]) + +OVN_FOR_EACH_NORTHD([ +AT_SETUP([BGP control plane redirect])Nit: probably best to add AT_SKIP_IF([test $HAVE_NC = no]) here.+ovn_start +OVS_TRAFFIC_VSWITCHD_START() + +ADD_BR([br-int]) +ADD_BR([br-ext]) + +ovs-ofctl add-flow br-ext action=normal +# Set external-ids in br-int needed for ovn-controller +ovs-vsctl \ + -- set Open_vSwitch . external-ids:system-id=hv1 \ + -- set Open_vSwitch . external-ids:ovn- remote=unix:$ovs_base/ovn-sb/ovn-sb.sock \ + -- set Open_vSwitch . external-ids:ovn-encap- type=geneve \ + -- set Open_vSwitch . external-ids:ovn-encap- ip=169.0.0.1 \ + -- set bridge br-int fail-mode=secure other- config:disable-in-band=trueMost ov*-ctl commands should be prefixed with "check" in this test.+ +# Start ovn-controller +start_daemon ovn-controller + +ovn-nbctl lr-add R1 \ + -- set Logical_Router R1 options:chassis=hv1 + +ovn-nbctl ls-add public + +ovn-nbctl lrp-add R1 rp-public 00:00:02:01:02:03 172.16.1.1/24 + +ovn-nbctl lsp-add public public-rp -- set Logical_Switch_Port public-rp \ + type=router options:router-port=rp-public \ + -- lsp-set-addresses public-rp router + +ovn-nbctl lsp-add public bgp-daemon \ + -- lsp-set-addresses bgp-daemon unknown + +AT_CHECK([ovs-vsctl set Open_vSwitch . external-ids:ovn- bridge-mappings=phynet:br-ext]) +ovn-nbctl lsp-add public public1 \ + -- lsp-set-addresses public1 unknown \ + -- lsp-set-type public1 localnet \ + -- lsp-set-options public1 network_name=phynet + +ovn-nbctl --wait=hv sync + +# Set option that redirects BGP control plane traffic to a LSP "bgp-daemon" +ovn-nbctl --wait=sb set logical_router_port rp-public options:bgp-redirect=bgp-daemon + +# Create "bgp-daemon" interface in a namespace with IP and MAC matching LRP "rp-public" +ADD_NAMESPACES(bgp-daemon) +ADD_VETH(bgp-daemon, bgp-daemon, br-int, "172.16.1.1/24", "00:00:02:01:02:03") + +ADD_NAMESPACES(ext-foo) +ADD_VETH(ext-foo, ext-foo, br-ext, "172.16.1.100/24", "00:10:10:01:02:13", \ + "172.16.1.1") + +# Flip the interface down/up to get proper IPv6 LLA +NS_EXEC([bgp-daemon], [ip link set down bgp-daemon]) +NS_EXEC([bgp-daemon], [ip link set up bgp-daemon]) + +# Wait until IPv6 LLA loses the "tentative" flag otherwise it can't be bound to. +OVS_WAIT_UNTIL([NS_EXEC([bgp-daemon], [ip a show dev bgp- daemon | grep "fe80::" | grep -v tentative])]) +OVS_WAIT_UNTIL([NS_EXEC([ext-foo], [ip a show dev ext-foo | grep "fe80::" | grep -v tentative])]) + +# Verify that BGP control plane traffic is delivered to the "bgp-daemon" +# interface on both IPv4 and IPv6 LLA addresses +NETNS_DAEMONIZE([bgp-daemon], [nc -l -k 172.16.1.1 179], [bgp_v4.pid]) +NS_CHECK_EXEC([ext-foo], [echo "TCP test" | nc --send-only 172.16.1.1 179]) + +NETNS_DAEMONIZE([bgp-daemon], [nc -l -6 -k fe80::200:2ff:fe01:203%bgp-daemon 179], [bgp_v6.pid]) +NS_CHECK_EXEC([ext-foo], [echo "TCP test" | nc --send-only -6 fe80::200:2ff:fe01:203%ext-foo 179]) + +OVS_APP_EXIT_AND_WAIT([ovn-controller]) + +as ovn-sb +OVS_APP_EXIT_AND_WAIT([ovsdb-server]) + +as ovn-nb +OVS_APP_EXIT_AND_WAIT([ovsdb-server]) + +as northd +OVS_APP_EXIT_AND_WAIT([ovn-northd]) + +as +OVS_TRAFFIC_VSWITCHD_STOP(["/.*error receiving.*/d +/.*terminating with signal 15.*/d"]) +AT_CLEANUP +])Regards, Dumitru_______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev_______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev_______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
_______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
