2 * Copyright (c) 2008, 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include <arpa/inet.h>
28 #include "mac-learning.h"
31 #include "ofp-print.h"
34 #include "openflow/openflow.h"
35 #include "openvswitch/datapath-protocol.h"
37 #include "poll-loop.h"
43 #define THIS_MODULE VLM_in_band
46 /* In-band control allows a single network to be used for OpenFlow
47 * traffic and other data traffic. Refer to ovs-vswitchd.conf(5) and
48 * secchan(8) for a description of configuring in-band control.
50 * This comment is an attempt to describe how in-band control works at a
51 * wire- and implementation-level. Correctly implementing in-band
52 * control has proven difficult due to its many subtleties, and has thus
53 * gone through many iterations. Please read through and understand the
54 * reasoning behind the chosen rules before making modifications.
56 * In Open vSwitch, in-band control is implemented as "hidden" flows (in
57 * that they are not visible through OpenFlow) and at a higher priority
58 * than wildcarded flows can be set up by the controller. This is done
59 * so that the controller cannot interfere with them and possibly break
60 * connectivity with its switches. It is possible to see all flows,
61 * including in-band ones, with the ovs-appctl "bridge/dump-flows"
64 * The following rules are always enabled with the "normal" action by a
65 * switch with in-band control:
67 * a. DHCP requests sent from the local port.
68 * b. ARP replies to the local port's MAC address.
69 * c. ARP requests from the local port's MAC address.
70 * d. ARP replies to the remote side's MAC address. Note that the
71 * remote side is either the controller or the gateway to reach
73 * e. ARP requests from the remote side's MAC address. Note that
74 * like (d), the MAC is either for the controller or gateway.
75 * f. ARP replies containing the controller's IP address as a target.
76 * g. ARP requests containing the controller's IP address as a source.
77 * h. OpenFlow (6633/tcp) traffic to the controller's IP.
78 * i. OpenFlow (6633/tcp) traffic from the controller's IP.
80 * The goal of these rules is to be as narrow as possible to allow a
81 * switch to join a network and be able to communicate with a
82 * controller. As mentioned earlier, these rules have higher priority
83 * than the controller's rules, so if they are too broad, they may
84 * prevent the controller from implementing its policy. As such,
85 * in-band actively monitors some aspects of flow and packet processing
86 * so that the rules can be made more precise.
88 * In-band control monitors attempts to add flows into the datapath that
89 * could interfere with its duties. The datapath only allows exact
90 * match entries, so in-band control is able to be very precise about
91 * the flows it prevents. Flows that miss in the datapath are sent to
92 * userspace to be processed, so preventing these flows from being
93 * cached in the "fast path" does not affect correctness. The only type
94 * of flow that is currently prevented is one that would prevent DHCP
95 * replies from being seen by the local port. For example, a rule that
96 * forwarded all DHCP traffic to the controller would not be allowed,
97 * but one that forwarded to all ports (including the local port) would.
99 * As mentioned earlier, packets that miss in the datapath are sent to
100 * the userspace for processing. The userspace has its own flow table,
101 * the "classifier", so in-band checks whether any special processing
102 * is needed before the classifier is consulted. If a packet is a DHCP
103 * response to a request from the local port, the packet is forwarded to
104 * the local port, regardless of the flow table. Note that this requires
105 * L7 processing of DHCP replies to determine whether the 'chaddr' field
106 * matches the MAC address of the local port.
108 * It is interesting to note that for an L3-based in-band control
109 * mechanism, the majority of rules are devoted to ARP traffic. At first
110 * glance, some of these rules appear redundant. However, each serves an
111 * important role. First, in order to determine the MAC address of the
112 * remote side (controller or gateway) for other ARP rules, we must allow
113 * ARP traffic for our local port with rules (b) and (c). If we are
114 * between a switch and its connection to the controller, we have to
115 * allow the other switch's ARP traffic to through. This is done with
116 * rules (d) and (e), since we do not know the addresses of the other
117 * switches a priori, but do know the controller's or gateway's. Finally,
118 * if the controller is running in a local guest VM that is not reached
119 * through the local port, the switch that is connected to the VM must
120 * allow ARP traffic based on the controller's IP address, since it will
121 * not know the MAC address of the local port that is sending the traffic
122 * or the MAC address of the controller in the guest VM.
124 * With a few notable exceptions below, in-band should work in most
125 * network setups. The following are considered "supported' in the
126 * current implementation:
128 * - Locally Connected. The switch and controller are on the same
129 * subnet. This uses rules (a), (b), (c), (h), and (i).
131 * - Reached through Gateway. The switch and controller are on
132 * different subnets and must go through a gateway. This uses
133 * rules (a), (b), (c), (h), and (i).
135 * - Between Switch and Controller. This switch is between another
136 * switch and the controller, and we want to allow the other
137 * switch's traffic through. This uses rules (d), (e), (h), and
138 * (i). It uses (b) and (c) indirectly in order to know the MAC
139 * address for rules (d) and (e). Note that DHCP for the other
140 * switch will not work unless the controller explicitly lets this
141 * switch pass the traffic.
143 * - Between Switch and Gateway. This switch is between another
144 * switch and the gateway, and we want to allow the other switch's
145 * traffic through. This uses the same rules and logic as the
146 * "Between Switch and Controller" configuration described earlier.
148 * - Controller on Local VM. The controller is a guest VM on the
149 * system running in-band control. This uses rules (a), (b), (c),
152 * - Controller on Local VM with Different Networks. The controller
153 * is a guest VM on the system running in-band control, but the
154 * local port is not used to connect to the controller. For
155 * example, an IP address is configured on eth0 of the switch. The
156 * controller's VM is connected through eth1 of the switch, but an
157 * IP address has not been configured for that port on the switch.
158 * As such, the switch will use eth0 to connect to the controller,
159 * and eth1's rules about the local port will not work. In the
160 * example, the switch attached to eth0 would use rules (a), (b),
161 * (c), (h), and (i) on eth0. The switch attached to eth1 would use
162 * rules (f), (g), (h), and (i).
164 * The following are explicitly *not* supported by in-band control:
166 * - Specify Controller by Name. Currently, the controller must be
167 * identified by IP address. A naive approach would be to permit
168 * all DNS traffic. Unfortunately, this would prevent the
169 * controller from defining any policy over DNS. Since switches
170 * that are located behind us need to connect to the controller,
171 * in-band cannot simply add a rule that allows DNS traffic from
172 * the local port. The "correct" way to support this is to parse
173 * DNS requests to allow all traffic related to a request for the
174 * controller's name through. Due to the potential security
175 * problems and amount of processing, we decided to hold off for
178 * - Multiple Controllers. There is nothing intrinsic in the high-
179 * level design that prevents using multiple (known) controllers,
180 * however, the current implementation's data structures assume
183 * - Differing Controllers for Switches. All switches must know
184 * the L3 addresses for all the controllers that other switches
185 * may use, since rules need to be set up to allow traffic related
186 * to those controllers through. See rules (f), (g), (h), and (i).
188 * - Differing Routes for Switches. In order for the switch to
189 * allow other switches to connect to a controller through a
190 * gateway, it allows the gateway's traffic through with rules (d)
191 * and (e). If the routes to the controller differ for the two
192 * switches, we will not know the MAC address of the alternate
196 #define IB_BASE_PRIORITY 18181800
199 IBR_FROM_LOCAL_DHCP, /* (a) From local port, DHCP. */
200 IBR_TO_LOCAL_ARP, /* (b) To local port, ARP. */
201 IBR_FROM_LOCAL_ARP, /* (c) From local port, ARP. */
202 IBR_TO_REMOTE_ARP, /* (d) To remote MAC, ARP. */
203 IBR_FROM_REMOTE_ARP, /* (e) From remote MAC, ARP. */
204 IBR_TO_CTL_ARP, /* (f) To controller IP, ARP. */
205 IBR_FROM_CTL_ARP, /* (g) From controller IP, ARP. */
206 IBR_TO_CTL_OFP, /* (h) To controller, OpenFlow port. */
207 IBR_FROM_CTL_OFP, /* (i) From controller, OpenFlow port. */
208 #if OFP_TCP_PORT != OFP_SSL_PORT
209 #error Need to support separate TCP and SSL flows.
218 unsigned int priority;
222 struct ofproto *ofproto;
223 struct rconn *controller;
224 struct status_category *ss_cat;
226 /* Keep track of local port's information. */
227 uint8_t local_mac[ETH_ADDR_LEN]; /* Current MAC. */
228 struct netdev *local_netdev; /* Local port's network device. */
229 time_t next_local_refresh;
231 /* Keep track of controller and next hop's information. */
232 uint32_t controller_ip; /* Controller IP, 0 if unknown. */
233 uint8_t remote_mac[ETH_ADDR_LEN]; /* Remote MAC. */
234 struct netdev *remote_netdev;
235 uint8_t last_remote_mac[ETH_ADDR_LEN]; /* Previous remote MAC. */
236 time_t next_remote_refresh;
238 /* Rules that we set up. */
239 struct ib_rule rules[N_IB_RULES];
242 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60);
244 static const uint8_t *
245 get_remote_mac(struct in_band *ib)
249 struct in_addr c_in4; /* Controller's IP address. */
250 struct in_addr r_in4; /* Next hop IP address. */
252 time_t now = time_now();
254 if (now >= ib->next_remote_refresh) {
255 /* Find the next-hop IP address. */
256 c_in4.s_addr = ib->controller_ip;
257 memset(ib->remote_mac, 0, sizeof ib->remote_mac);
258 retval = netdev_get_next_hop(ib->local_netdev,
259 &c_in4, &r_in4, &next_hop_dev);
261 VLOG_WARN("cannot find route for controller ("IP_FMT"): %s",
262 IP_ARGS(&ib->controller_ip), strerror(retval));
263 ib->next_remote_refresh = now + 1;
267 r_in4.s_addr = c_in4.s_addr;
270 /* Get the next-hop IP and network device. */
271 if (!ib->remote_netdev
272 || strcmp(netdev_get_name(ib->remote_netdev), next_hop_dev))
274 netdev_close(ib->remote_netdev);
276 retval = netdev_open_default(next_hop_dev, &ib->remote_netdev);
278 VLOG_WARN_RL(&rl, "cannot open netdev %s (next hop "
279 "to controller "IP_FMT"): %s",
280 next_hop_dev, IP_ARGS(&ib->controller_ip),
282 ib->next_remote_refresh = now + 1;
287 /* Look up the MAC address of the next-hop IP address. */
288 retval = netdev_arp_lookup(ib->remote_netdev, r_in4.s_addr,
291 VLOG_DBG_RL(&rl, "cannot look up remote MAC address ("IP_FMT"): %s",
292 IP_ARGS(&r_in4.s_addr), strerror(retval));
294 have_mac = !eth_addr_is_zero(ib->remote_mac);
297 && !eth_addr_equals(ib->last_remote_mac, ib->remote_mac)) {
298 VLOG_DBG("remote MAC address changed from "ETH_ADDR_FMT" to "
300 ETH_ADDR_ARGS(ib->last_remote_mac),
301 ETH_ADDR_ARGS(ib->remote_mac));
302 memcpy(ib->last_remote_mac, ib->remote_mac, ETH_ADDR_LEN);
305 /* Schedule next refresh.
307 * If we have an IP address but not a MAC address, then refresh
308 * quickly, since we probably will get a MAC address soon (via ARP).
309 * Otherwise, we can afford to wait a little while. */
310 ib->next_remote_refresh
311 = now + (!ib->controller_ip || have_mac ? 10 : 1);
314 return !eth_addr_is_zero(ib->remote_mac) ? ib->remote_mac : NULL;
317 static const uint8_t *
318 get_local_mac(struct in_band *ib)
320 time_t now = time_now();
321 if (now >= ib->next_local_refresh) {
322 uint8_t ea[ETH_ADDR_LEN];
323 if (ib->local_netdev && !netdev_get_etheraddr(ib->local_netdev, ea)) {
324 memcpy(ib->local_mac, ea, ETH_ADDR_LEN);
326 ib->next_local_refresh = now + 1;
328 return !eth_addr_is_zero(ib->local_mac) ? ib->local_mac : NULL;
332 in_band_status_cb(struct status_reply *sr, void *in_band_)
334 struct in_band *in_band = in_band_;
336 if (!eth_addr_is_zero(in_band->local_mac)) {
337 status_reply_put(sr, "local-mac="ETH_ADDR_FMT,
338 ETH_ADDR_ARGS(in_band->local_mac));
341 if (!eth_addr_is_zero(in_band->remote_mac)) {
342 status_reply_put(sr, "remote-mac="ETH_ADDR_FMT,
343 ETH_ADDR_ARGS(in_band->remote_mac));
348 drop_flow(struct in_band *in_band, int rule_idx)
350 struct ib_rule *rule = &in_band->rules[rule_idx];
352 if (rule->installed) {
353 rule->installed = false;
354 ofproto_delete_flow(in_band->ofproto, &rule->flow, rule->wildcards,
359 /* out_port and fixed_fields are assumed never to change. */
361 set_up_flow(struct in_band *in_band, int rule_idx, const flow_t *flow,
362 uint32_t fixed_fields, uint16_t out_port)
364 struct ib_rule *rule = &in_band->rules[rule_idx];
366 if (!rule->installed || memcmp(flow, &rule->flow, sizeof *flow)) {
367 union ofp_action action;
369 drop_flow(in_band, rule_idx);
371 rule->installed = true;
373 rule->wildcards = OFPFW_ALL & ~fixed_fields;
374 rule->priority = IB_BASE_PRIORITY + (N_IB_RULES - rule_idx);
376 action.type = htons(OFPAT_OUTPUT);
377 action.output.len = htons(sizeof action);
378 action.output.port = htons(out_port);
379 action.output.max_len = htons(0);
380 ofproto_add_flow(in_band->ofproto, &rule->flow, rule->wildcards,
381 rule->priority, &action, 1, 0);
385 /* Returns true if 'packet' should be sent to the local port regardless
386 * of the flow table. */
388 in_band_msg_in_hook(struct in_band *in_band, const flow_t *flow,
389 const struct ofpbuf *packet)
395 /* Regardless of how the flow table is configured, we want to be
396 * able to see replies to our DHCP requests. */
397 if (flow->dl_type == htons(ETH_TYPE_IP)
398 && flow->nw_proto == IP_TYPE_UDP
399 && flow->tp_src == htons(DHCP_SERVER_PORT)
400 && flow->tp_dst == htons(DHCP_CLIENT_PORT)
402 struct dhcp_header *dhcp;
403 const uint8_t *local_mac;
405 dhcp = ofpbuf_at(packet, (char *)packet->l7 - (char *)packet->data,
411 local_mac = get_local_mac(in_band);
412 if (eth_addr_equals(dhcp->chaddr, local_mac)) {
420 /* Returns true if the rule that would match 'flow' with 'actions' is
421 * allowed to be set up in the datapath. */
423 in_band_rule_check(struct in_band *in_band, const flow_t *flow,
424 const struct odp_actions *actions)
430 /* Don't allow flows that would prevent DHCP replies from being seen
431 * by the local port. */
432 if (flow->dl_type == htons(ETH_TYPE_IP)
433 && flow->nw_proto == IP_TYPE_UDP
434 && flow->tp_src == htons(DHCP_SERVER_PORT)
435 && flow->tp_dst == htons(DHCP_CLIENT_PORT)) {
438 for (i=0; i<actions->n_actions; i++) {
439 if (actions->actions[i].output.type == ODPAT_OUTPUT
440 && actions->actions[i].output.port == ODPP_LOCAL) {
451 in_band_run(struct in_band *in_band)
453 time_t now = time_now();
454 uint32_t controller_ip;
455 const uint8_t *remote_mac;
456 const uint8_t *local_mac;
459 if (now < in_band->next_remote_refresh
460 && now < in_band->next_local_refresh) {
464 controller_ip = rconn_get_remote_ip(in_band->controller);
465 if (in_band->controller_ip && controller_ip != in_band->controller_ip) {
466 VLOG_DBG("controller IP address changed from "IP_FMT" to "IP_FMT,
467 IP_ARGS(&in_band->controller_ip),
468 IP_ARGS(&controller_ip));
470 in_band->controller_ip = controller_ip;
472 remote_mac = get_remote_mac(in_band);
473 local_mac = get_local_mac(in_band);
476 /* Allow DHCP requests to be sent from the local port. */
477 memset(&flow, 0, sizeof flow);
478 flow.in_port = ODPP_LOCAL;
479 flow.dl_type = htons(ETH_TYPE_IP);
480 memcpy(flow.dl_src, local_mac, ETH_ADDR_LEN);
481 flow.nw_proto = IP_TYPE_UDP;
482 flow.tp_src = htons(DHCP_CLIENT_PORT);
483 flow.tp_dst = htons(DHCP_SERVER_PORT);
484 set_up_flow(in_band, IBR_FROM_LOCAL_DHCP, &flow,
485 (OFPFW_IN_PORT | OFPFW_DL_TYPE | OFPFW_DL_SRC
486 | OFPFW_NW_PROTO | OFPFW_TP_SRC | OFPFW_TP_DST),
489 /* Allow the connection's interface to receive directed ARP traffic. */
490 memset(&flow, 0, sizeof flow);
491 flow.dl_type = htons(ETH_TYPE_ARP);
492 memcpy(flow.dl_dst, local_mac, ETH_ADDR_LEN);
493 flow.nw_proto = ARP_OP_REPLY;
494 set_up_flow(in_band, IBR_TO_LOCAL_ARP, &flow,
495 (OFPFW_DL_TYPE | OFPFW_DL_DST | OFPFW_NW_PROTO),
498 /* Allow the connection's interface to be the source of ARP traffic. */
499 memset(&flow, 0, sizeof flow);
500 flow.dl_type = htons(ETH_TYPE_ARP);
501 memcpy(flow.dl_src, local_mac, ETH_ADDR_LEN);
502 flow.nw_proto = ARP_OP_REQUEST;
503 set_up_flow(in_band, IBR_FROM_LOCAL_ARP, &flow,
504 (OFPFW_DL_TYPE | OFPFW_DL_SRC | OFPFW_NW_PROTO),
507 drop_flow(in_band, IBR_TO_LOCAL_ARP);
508 drop_flow(in_band, IBR_FROM_LOCAL_ARP);
512 /* Allow ARP replies to the remote side's MAC. */
513 memset(&flow, 0, sizeof flow);
514 flow.dl_type = htons(ETH_TYPE_ARP);
515 memcpy(flow.dl_dst, remote_mac, ETH_ADDR_LEN);
516 flow.nw_proto = ARP_OP_REPLY;
517 set_up_flow(in_band, IBR_TO_REMOTE_ARP, &flow,
518 (OFPFW_DL_TYPE | OFPFW_DL_DST | OFPFW_NW_PROTO),
521 /* Allow ARP requests from the remote side's MAC. */
522 memset(&flow, 0, sizeof flow);
523 flow.dl_type = htons(ETH_TYPE_ARP);
524 memcpy(flow.dl_src, remote_mac, ETH_ADDR_LEN);
525 flow.nw_proto = ARP_OP_REQUEST;
526 set_up_flow(in_band, IBR_FROM_REMOTE_ARP, &flow,
527 (OFPFW_DL_TYPE | OFPFW_DL_SRC | OFPFW_NW_PROTO),
530 drop_flow(in_band, IBR_TO_REMOTE_ARP);
531 drop_flow(in_band, IBR_FROM_REMOTE_ARP);
535 /* Allow ARP replies to the controller's IP. */
536 memset(&flow, 0, sizeof flow);
537 flow.dl_type = htons(ETH_TYPE_ARP);
538 flow.nw_proto = ARP_OP_REPLY;
539 flow.nw_dst = controller_ip;
540 set_up_flow(in_band, IBR_TO_CTL_ARP, &flow,
541 (OFPFW_DL_TYPE | OFPFW_NW_PROTO | OFPFW_NW_DST_MASK),
544 /* Allow ARP requests from the controller's IP. */
545 memset(&flow, 0, sizeof flow);
546 flow.dl_type = htons(ETH_TYPE_ARP);
547 flow.nw_proto = ARP_OP_REQUEST;
548 flow.nw_src = controller_ip;
549 set_up_flow(in_band, IBR_FROM_CTL_ARP, &flow,
550 (OFPFW_DL_TYPE | OFPFW_NW_PROTO | OFPFW_NW_SRC_MASK),
553 /* OpenFlow traffic to or from the controller.
555 * (A given field's value is completely ignored if it is wildcarded,
556 * which is why we can get away with using a single 'flow' in each
558 memset(&flow, 0, sizeof flow);
559 flow.dl_type = htons(ETH_TYPE_IP);
560 flow.nw_proto = IP_TYPE_TCP;
561 flow.nw_src = controller_ip;
562 flow.nw_dst = controller_ip;
563 flow.tp_src = htons(OFP_TCP_PORT);
564 flow.tp_dst = htons(OFP_TCP_PORT);
565 set_up_flow(in_band, IBR_TO_CTL_OFP, &flow,
566 (OFPFW_DL_TYPE | OFPFW_NW_PROTO | OFPFW_NW_DST_MASK
567 | OFPFW_TP_DST), OFPP_NORMAL);
568 set_up_flow(in_band, IBR_FROM_CTL_OFP, &flow,
569 (OFPFW_DL_TYPE | OFPFW_NW_PROTO | OFPFW_NW_SRC_MASK
570 | OFPFW_TP_SRC), OFPP_NORMAL);
572 drop_flow(in_band, IBR_TO_CTL_ARP);
573 drop_flow(in_band, IBR_FROM_CTL_ARP);
574 drop_flow(in_band, IBR_TO_CTL_OFP);
575 drop_flow(in_band, IBR_FROM_CTL_OFP);
580 in_band_wait(struct in_band *in_band)
582 time_t now = time_now();
584 = MIN(in_band->next_remote_refresh, in_band->next_local_refresh);
586 poll_timer_wait((wakeup - now) * 1000);
588 poll_immediate_wake();
593 in_band_flushed(struct in_band *in_band)
597 for (i = 0; i < N_IB_RULES; i++) {
598 in_band->rules[i].installed = false;
603 in_band_create(struct ofproto *ofproto, struct dpif *dpif,
604 struct switch_status *ss, struct rconn *controller,
605 struct in_band **in_bandp)
607 struct in_band *in_band;
608 char local_name[IF_NAMESIZE];
609 struct netdev *local_netdev;
612 error = dpif_port_get_name(dpif, ODPP_LOCAL,
613 local_name, sizeof local_name);
615 VLOG_ERR("failed to initialize in-band control: cannot get name "
616 "of datapath local port (%s)", strerror(error));
620 error = netdev_open_default(local_name, &local_netdev);
622 VLOG_ERR("failed to initialize in-band control: cannot open "
623 "datapath local port %s (%s)", local_name, strerror(error));
627 in_band = xzalloc(sizeof *in_band);
628 in_band->ofproto = ofproto;
629 in_band->controller = controller;
630 in_band->ss_cat = switch_status_register(ss, "in-band",
631 in_band_status_cb, in_band);
632 in_band->local_netdev = local_netdev;
633 in_band->next_local_refresh = TIME_MIN;
634 in_band->remote_netdev = NULL;
635 in_band->next_remote_refresh = TIME_MIN;
643 in_band_destroy(struct in_band *in_band)
646 switch_status_unregister(in_band->ss_cat);
647 netdev_close(in_band->local_netdev);
648 netdev_close(in_band->remote_netdev);
649 /* We don't own the rconn. */