1 /* Copyright (c) 2013, 2014, 2015 Nicira, Inc.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License. */
20 #include "byte-order.h"
21 #include "connectivity.h"
24 #include "dynamic-string.h"
25 #include "fat-rwlock.h"
32 #include "route-table.h"
35 #include "socket-util.h"
36 #include "tnl-arp-cache.h"
37 #include "tnl-ports.h"
39 #include "openvswitch/vlog.h"
40 #include "unaligned.h"
41 #include "ofproto-dpif.h"
43 VLOG_DEFINE_THIS_MODULE(tunnel);
45 /* skb mark used for IPsec tunnel packets */
60 struct hmap_node ofport_node;
61 struct hmap_node match_node;
63 const struct ofport_dpif *ofport;
65 struct netdev *netdev;
67 struct tnl_match match;
70 static struct fat_rwlock rwlock;
74 * This module maps packets received over tunnel protocols to vports. The
75 * tunnel protocol and, for some protocols, tunnel-specific information (e.g.,
76 * for VXLAN, the UDP destination port number) are always use as part of the
77 * mapping. Which other fields are used for the mapping depends on the vports
78 * themselves (the parenthesized notations refer to "struct tnl_match" fields):
80 * - in_key: A vport may match a specific tunnel ID (in_key_flow == false)
81 * or arrange for the tunnel ID to be matched as tunnel.tun_id in the
82 * OpenFlow flow (in_key_flow == true).
84 * - ip_dst: A vport may match a specific destination IP address
85 * (ip_dst_flow == false) or arrange for the destination IP to be matched
86 * as tunnel.ip_dst in the OpenFlow flow (ip_dst_flow == true).
88 * - ip_src: A vport may match a specific IP source address (ip_src_flow ==
89 * false, ip_src != 0), wildcard all source addresses (ip_src_flow ==
90 * false, ip_src == 0), or arrange for the IP source address to be
91 * handled in the OpenFlow flow table (ip_src_flow == true).
93 * Thus, there are 2 * 2 * 3 == 12 possible ways a vport can match against a
94 * tunnel packet. We number the possibilities for each field in increasing
95 * order as listed in each bullet above. We order the 12 overall combinations
96 * in lexicographic order considering in_key first, then ip_dst, then
98 #define N_MATCH_TYPES (2 * 2 * 3)
100 /* The three possibilities (see above) for vport ip_src matches. */
102 IP_SRC_CFG, /* ip_src must equal configured address. */
103 IP_SRC_ANY, /* Any ip_src is acceptable. */
104 IP_SRC_FLOW /* ip_src is handled in flow table. */
107 /* Each hmap contains "struct tnl_port"s.
108 * The index is a combination of how each of the fields listed under "Tunnel
109 * matches" above matches, see the final paragraph for ordering. */
110 static struct hmap *tnl_match_maps[N_MATCH_TYPES] OVS_GUARDED_BY(rwlock);
111 static struct hmap **tnl_match_map(const struct tnl_match *);
113 static struct hmap ofport_map__ = HMAP_INITIALIZER(&ofport_map__);
114 static struct hmap *ofport_map OVS_GUARDED_BY(rwlock) = &ofport_map__;
116 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
117 static struct vlog_rate_limit dbg_rl = VLOG_RATE_LIMIT_INIT(60, 60);
119 static struct tnl_port *tnl_find(const struct flow *) OVS_REQ_RDLOCK(rwlock);
120 static struct tnl_port *tnl_find_exact(struct tnl_match *, struct hmap *)
121 OVS_REQ_RDLOCK(rwlock);
122 static struct tnl_port *tnl_find_ofport(const struct ofport_dpif *)
123 OVS_REQ_RDLOCK(rwlock);
125 static uint32_t tnl_hash(struct tnl_match *);
126 static void tnl_match_fmt(const struct tnl_match *, struct ds *);
127 static char *tnl_port_fmt(const struct tnl_port *) OVS_REQ_RDLOCK(rwlock);
128 static void tnl_port_mod_log(const struct tnl_port *, const char *action)
129 OVS_REQ_RDLOCK(rwlock);
130 static const char *tnl_port_get_name(const struct tnl_port *)
131 OVS_REQ_RDLOCK(rwlock);
132 static void tnl_port_del__(const struct ofport_dpif *) OVS_REQ_WRLOCK(rwlock);
135 ofproto_tunnel_init(void)
137 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
139 if (ovsthread_once_start(&once)) {
140 fat_rwlock_init(&rwlock);
141 ovsthread_once_done(&once);
146 tnl_port_add__(const struct ofport_dpif *ofport, const struct netdev *netdev,
147 odp_port_t odp_port, bool warn, bool native_tnl, const char name[])
148 OVS_REQ_WRLOCK(rwlock)
150 const struct netdev_tunnel_config *cfg;
151 struct tnl_port *existing_port;
152 struct tnl_port *tnl_port;
155 cfg = netdev_get_tunnel_config(netdev);
158 tnl_port = xzalloc(sizeof *tnl_port);
159 tnl_port->ofport = ofport;
160 tnl_port->netdev = netdev_ref(netdev);
161 tnl_port->change_seq = netdev_get_change_seq(tnl_port->netdev);
163 tnl_port->match.in_key = cfg->in_key;
164 tnl_port->match.ip_src = cfg->ip_src;
165 tnl_port->match.ip_dst = cfg->ip_dst;
166 tnl_port->match.ip_src_flow = cfg->ip_src_flow;
167 tnl_port->match.ip_dst_flow = cfg->ip_dst_flow;
168 tnl_port->match.pkt_mark = cfg->ipsec ? IPSEC_MARK : 0;
169 tnl_port->match.in_key_flow = cfg->in_key_flow;
170 tnl_port->match.odp_port = odp_port;
172 map = tnl_match_map(&tnl_port->match);
173 existing_port = tnl_find_exact(&tnl_port->match, *map);
176 struct ds ds = DS_EMPTY_INITIALIZER;
177 tnl_match_fmt(&tnl_port->match, &ds);
178 VLOG_WARN("%s: attempting to add tunnel port with same config as "
179 "port '%s' (%s)", tnl_port_get_name(tnl_port),
180 tnl_port_get_name(existing_port), ds_cstr(&ds));
183 netdev_close(tnl_port->netdev);
188 hmap_insert(ofport_map, &tnl_port->ofport_node, hash_pointer(ofport, 0));
191 *map = xmalloc(sizeof **map);
194 hmap_insert(*map, &tnl_port->match_node, tnl_hash(&tnl_port->match));
195 tnl_port_mod_log(tnl_port, "adding");
198 tnl_port_map_insert(odp_port, cfg->dst_port, name);
203 /* Adds 'ofport' to the module with datapath port number 'odp_port'. 'ofport's
204 * must be added before they can be used by the module. 'ofport' must be a
207 * Returns 0 if successful, otherwise a positive errno value. */
209 tnl_port_add(const struct ofport_dpif *ofport, const struct netdev *netdev,
210 odp_port_t odp_port, bool native_tnl, const char name[]) OVS_EXCLUDED(rwlock)
214 fat_rwlock_wrlock(&rwlock);
215 ok = tnl_port_add__(ofport, netdev, odp_port, true, native_tnl, name);
216 fat_rwlock_unlock(&rwlock);
218 return ok ? 0 : EEXIST;
221 /* Checks if the tunnel represented by 'ofport' reconfiguration due to changes
222 * in its netdev_tunnel_config. If it does, returns true. Otherwise, returns
223 * false. 'ofport' and 'odp_port' should be the same as would be passed to
226 tnl_port_reconfigure(const struct ofport_dpif *ofport,
227 const struct netdev *netdev, odp_port_t odp_port,
228 bool native_tnl, const char name[])
231 struct tnl_port *tnl_port;
232 bool changed = false;
234 fat_rwlock_wrlock(&rwlock);
235 tnl_port = tnl_find_ofport(ofport);
237 changed = tnl_port_add__(ofport, netdev, odp_port, false, native_tnl, name);
238 } else if (tnl_port->netdev != netdev
239 || tnl_port->match.odp_port != odp_port
240 || tnl_port->change_seq != netdev_get_change_seq(tnl_port->netdev)) {
241 VLOG_DBG("reconfiguring %s", tnl_port_get_name(tnl_port));
242 tnl_port_del__(ofport);
243 tnl_port_add__(ofport, netdev, odp_port, true, native_tnl, name);
246 fat_rwlock_unlock(&rwlock);
251 tnl_port_del__(const struct ofport_dpif *ofport) OVS_REQ_WRLOCK(rwlock)
253 struct tnl_port *tnl_port;
259 tnl_port = tnl_find_ofport(ofport);
261 const struct netdev_tunnel_config *cfg =
262 netdev_get_tunnel_config(tnl_port->netdev);
265 tnl_port_map_delete(cfg->dst_port);
266 tnl_port_mod_log(tnl_port, "removing");
267 map = tnl_match_map(&tnl_port->match);
268 hmap_remove(*map, &tnl_port->match_node);
269 if (hmap_is_empty(*map)) {
274 hmap_remove(ofport_map, &tnl_port->ofport_node);
275 netdev_close(tnl_port->netdev);
280 /* Removes 'ofport' from the module. */
282 tnl_port_del(const struct ofport_dpif *ofport) OVS_EXCLUDED(rwlock)
284 fat_rwlock_wrlock(&rwlock);
285 tnl_port_del__(ofport);
286 fat_rwlock_unlock(&rwlock);
289 /* Looks in the table of tunnels for a tunnel matching the metadata in 'flow'.
290 * Returns the 'ofport' corresponding to the new in_port, or a null pointer if
293 * Callers should verify that 'flow' needs to be received by calling
294 * tnl_port_should_receive() before this function. */
295 const struct ofport_dpif *
296 tnl_port_receive(const struct flow *flow) OVS_EXCLUDED(rwlock)
298 char *pre_flow_str = NULL;
299 const struct ofport_dpif *ofport;
300 struct tnl_port *tnl_port;
302 fat_rwlock_rdlock(&rwlock);
303 tnl_port = tnl_find(flow);
304 ofport = tnl_port ? tnl_port->ofport : NULL;
306 char *flow_str = flow_to_string(flow);
308 VLOG_WARN_RL(&rl, "receive tunnel port not found (%s)", flow_str);
313 if (!VLOG_DROP_DBG(&dbg_rl)) {
314 pre_flow_str = flow_to_string(flow);
318 char *post_flow_str = flow_to_string(flow);
319 char *tnl_str = tnl_port_fmt(tnl_port);
320 VLOG_DBG("flow received\n"
324 tnl_str, pre_flow_str, post_flow_str);
331 fat_rwlock_unlock(&rwlock);
336 tnl_ecn_ok(const struct flow *base_flow, struct flow *flow,
337 struct flow_wildcards *wc)
339 if (is_ip_any(base_flow)) {
340 if ((flow->tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
342 wc->masks.nw_tos |= IP_ECN_MASK;
344 if ((base_flow->nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
345 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
346 " but is not ECN capable");
349 /* Set the ECN CE value in the tunneled packet. */
350 flow->nw_tos |= IP_ECN_CE;
358 /* Should be called at the beginning of action translation to initialize
359 * wildcards and perform any actions based on receiving on tunnel port.
361 * Returns false if the packet must be dropped. */
363 tnl_xlate_init(const struct flow *base_flow, struct flow *flow,
364 struct flow_wildcards *wc)
366 /* tnl_port_should_receive() examines the 'tunnel.ip_dst' field to
367 * determine the presence of the tunnel metadata. However, since tunnels'
368 * datapath port numbers are different from the non-tunnel ports, and we
369 * always unwildcard the 'in_port', we do not need to unwildcard
370 * the 'tunnel.ip_dst' for non-tunneled packets. */
371 if (tnl_port_should_receive(flow)) {
373 wc->masks.tunnel.tun_id = OVS_BE64_MAX;
374 wc->masks.tunnel.ip_src = OVS_BE32_MAX;
375 wc->masks.tunnel.ip_dst = OVS_BE32_MAX;
376 wc->masks.tunnel.flags = (FLOW_TNL_F_DONT_FRAGMENT |
379 wc->masks.tunnel.ip_tos = UINT8_MAX;
380 wc->masks.tunnel.ip_ttl = UINT8_MAX;
381 /* The tp_src and tp_dst members in flow_tnl are set to be always
382 * wildcarded, not to unwildcard them here. */
383 wc->masks.tunnel.tp_src = 0;
384 wc->masks.tunnel.tp_dst = 0;
386 memset(&wc->masks.pkt_mark, 0xff, sizeof wc->masks.pkt_mark);
388 if (!tnl_ecn_ok(base_flow, flow, wc)) {
392 flow->pkt_mark &= ~IPSEC_MARK;
398 /* Given that 'flow' should be output to the ofport corresponding to
399 * 'tnl_port', updates 'flow''s tunnel headers and returns the actual datapath
400 * port that the output should happen on. May return ODPP_NONE if the output
401 * shouldn't occur. */
403 tnl_port_send(const struct ofport_dpif *ofport, struct flow *flow,
404 struct flow_wildcards *wc) OVS_EXCLUDED(rwlock)
406 const struct netdev_tunnel_config *cfg;
407 struct tnl_port *tnl_port;
408 char *pre_flow_str = NULL;
411 fat_rwlock_rdlock(&rwlock);
412 tnl_port = tnl_find_ofport(ofport);
413 out_port = tnl_port ? tnl_port->match.odp_port : ODPP_NONE;
418 cfg = netdev_get_tunnel_config(tnl_port->netdev);
421 if (!VLOG_DROP_DBG(&dbg_rl)) {
422 pre_flow_str = flow_to_string(flow);
425 if (!cfg->ip_src_flow) {
426 flow->tunnel.ip_src = tnl_port->match.ip_src;
428 if (!cfg->ip_dst_flow) {
429 flow->tunnel.ip_dst = tnl_port->match.ip_dst;
431 flow->pkt_mark = tnl_port->match.pkt_mark;
433 if (!cfg->out_key_flow) {
434 flow->tunnel.tun_id = cfg->out_key;
437 if (cfg->ttl_inherit && is_ip_any(flow)) {
438 wc->masks.nw_ttl = 0xff;
439 flow->tunnel.ip_ttl = flow->nw_ttl;
441 flow->tunnel.ip_ttl = cfg->ttl;
444 if (cfg->tos_inherit && is_ip_any(flow)) {
445 wc->masks.nw_tos |= IP_DSCP_MASK;
446 flow->tunnel.ip_tos = flow->nw_tos & IP_DSCP_MASK;
448 flow->tunnel.ip_tos = cfg->tos;
451 /* ECN fields are always inherited. */
452 if (is_ip_any(flow)) {
453 wc->masks.nw_tos |= IP_ECN_MASK;
455 if ((flow->nw_tos & IP_ECN_MASK) == IP_ECN_CE) {
456 flow->tunnel.ip_tos |= IP_ECN_ECT_0;
458 flow->tunnel.ip_tos |= flow->nw_tos & IP_ECN_MASK;
462 flow->tunnel.flags |= (cfg->dont_fragment ? FLOW_TNL_F_DONT_FRAGMENT : 0)
463 | (cfg->csum ? FLOW_TNL_F_CSUM : 0)
464 | (cfg->out_key_present ? FLOW_TNL_F_KEY : 0);
467 char *post_flow_str = flow_to_string(flow);
468 char *tnl_str = tnl_port_fmt(tnl_port);
469 VLOG_DBG("flow sent\n"
473 tnl_str, pre_flow_str, post_flow_str);
480 fat_rwlock_unlock(&rwlock);
485 tnl_hash(struct tnl_match *match)
487 BUILD_ASSERT_DECL(sizeof *match % sizeof(uint32_t) == 0);
488 return hash_words((uint32_t *) match, sizeof *match / sizeof(uint32_t), 0);
491 static struct tnl_port *
492 tnl_find_ofport(const struct ofport_dpif *ofport) OVS_REQ_RDLOCK(rwlock)
494 struct tnl_port *tnl_port;
496 HMAP_FOR_EACH_IN_BUCKET (tnl_port, ofport_node, hash_pointer(ofport, 0),
498 if (tnl_port->ofport == ofport) {
505 static struct tnl_port *
506 tnl_find_exact(struct tnl_match *match, struct hmap *map)
507 OVS_REQ_RDLOCK(rwlock)
510 struct tnl_port *tnl_port;
512 HMAP_FOR_EACH_WITH_HASH (tnl_port, match_node, tnl_hash(match), map) {
513 if (!memcmp(match, &tnl_port->match, sizeof *match)) {
521 /* Returns the tnl_port that is the best match for the tunnel data in 'flow',
522 * or NULL if no tnl_port matches 'flow'. */
523 static struct tnl_port *
524 tnl_find(const struct flow *flow) OVS_REQ_RDLOCK(rwlock)
526 enum ip_src_type ip_src;
532 for (in_key_flow = 0; in_key_flow < 2; in_key_flow++) {
533 for (ip_dst_flow = 0; ip_dst_flow < 2; ip_dst_flow++) {
534 for (ip_src = 0; ip_src < 3; ip_src++) {
535 struct hmap *map = tnl_match_maps[i];
538 struct tnl_port *tnl_port;
539 struct tnl_match match;
541 memset(&match, 0, sizeof match);
543 /* The apparent mix-up of 'ip_dst' and 'ip_src' below is
544 * correct, because "struct tnl_match" is expressed in
545 * terms of packets being sent out, but we are using it
546 * here as a description of how to treat received
548 match.in_key = in_key_flow ? 0 : flow->tunnel.tun_id;
549 match.ip_src = (ip_src == IP_SRC_CFG
550 ? flow->tunnel.ip_dst
552 match.ip_dst = ip_dst_flow ? 0 : flow->tunnel.ip_src;
553 match.odp_port = flow->in_port.odp_port;
554 match.pkt_mark = flow->pkt_mark;
555 match.in_key_flow = in_key_flow;
556 match.ip_dst_flow = ip_dst_flow;
557 match.ip_src_flow = ip_src == IP_SRC_FLOW;
559 tnl_port = tnl_find_exact(&match, map);
573 /* Returns a pointer to the 'tnl_match_maps' element corresponding to 'm''s
574 * matching criteria. */
575 static struct hmap **
576 tnl_match_map(const struct tnl_match *m)
578 enum ip_src_type ip_src;
580 ip_src = (m->ip_src_flow ? IP_SRC_FLOW
581 : m->ip_src ? IP_SRC_CFG
584 return &tnl_match_maps[6 * m->in_key_flow + 3 * m->ip_dst_flow + ip_src];
588 tnl_match_fmt(const struct tnl_match *match, struct ds *ds)
589 OVS_REQ_RDLOCK(rwlock)
591 if (!match->ip_dst_flow) {
592 ds_put_format(ds, IP_FMT"->"IP_FMT, IP_ARGS(match->ip_src),
593 IP_ARGS(match->ip_dst));
594 } else if (!match->ip_src_flow) {
595 ds_put_format(ds, IP_FMT"->flow", IP_ARGS(match->ip_src));
597 ds_put_cstr(ds, "flow->flow");
600 if (match->in_key_flow) {
601 ds_put_cstr(ds, ", key=flow");
603 ds_put_format(ds, ", key=%#"PRIx64, ntohll(match->in_key));
606 ds_put_format(ds, ", dp port=%"PRIu32, match->odp_port);
607 ds_put_format(ds, ", pkt mark=%"PRIu32, match->pkt_mark);
611 tnl_port_mod_log(const struct tnl_port *tnl_port, const char *action)
612 OVS_REQ_RDLOCK(rwlock)
614 if (VLOG_IS_DBG_ENABLED()) {
615 struct ds ds = DS_EMPTY_INITIALIZER;
617 tnl_match_fmt(&tnl_port->match, &ds);
618 VLOG_INFO("%s tunnel port %s (%s)", action,
619 tnl_port_get_name(tnl_port), ds_cstr(&ds));
625 tnl_port_fmt(const struct tnl_port *tnl_port) OVS_REQ_RDLOCK(rwlock)
627 const struct netdev_tunnel_config *cfg =
628 netdev_get_tunnel_config(tnl_port->netdev);
629 struct ds ds = DS_EMPTY_INITIALIZER;
631 ds_put_format(&ds, "port %"PRIu32": %s (%s: ", tnl_port->match.odp_port,
632 tnl_port_get_name(tnl_port),
633 netdev_get_type(tnl_port->netdev));
634 tnl_match_fmt(&tnl_port->match, &ds);
636 if (cfg->out_key != cfg->in_key ||
637 cfg->out_key_present != cfg->in_key_present ||
638 cfg->out_key_flow != cfg->in_key_flow) {
639 ds_put_cstr(&ds, ", out_key=");
640 if (!cfg->out_key_present) {
641 ds_put_cstr(&ds, "none");
642 } else if (cfg->out_key_flow) {
643 ds_put_cstr(&ds, "flow");
645 ds_put_format(&ds, "%#"PRIx64, ntohll(cfg->out_key));
649 if (cfg->ttl_inherit) {
650 ds_put_cstr(&ds, ", ttl=inherit");
652 ds_put_format(&ds, ", ttl=%"PRIu8, cfg->ttl);
655 if (cfg->tos_inherit) {
656 ds_put_cstr(&ds, ", tos=inherit");
657 } else if (cfg->tos) {
658 ds_put_format(&ds, ", tos=%#"PRIx8, cfg->tos);
661 if (!cfg->dont_fragment) {
662 ds_put_cstr(&ds, ", df=false");
666 ds_put_cstr(&ds, ", csum=true");
669 ds_put_cstr(&ds, ")\n");
671 return ds_steal_cstr(&ds);
675 tnl_port_get_name(const struct tnl_port *tnl_port) OVS_REQ_RDLOCK(rwlock)
677 return netdev_get_name(tnl_port->netdev);
681 tnl_port_build_header(const struct ofport_dpif *ofport,
682 const struct flow *tnl_flow,
683 uint8_t dmac[ETH_ADDR_LEN],
684 uint8_t smac[ETH_ADDR_LEN],
685 ovs_be32 ip_src, struct ovs_action_push_tnl *data)
687 struct tnl_port *tnl_port;
688 struct eth_header *eth;
689 struct ip_header *ip;
693 fat_rwlock_rdlock(&rwlock);
694 tnl_port = tnl_find_ofport(ofport);
695 ovs_assert(tnl_port);
697 /* Build Ethernet and IP headers. */
698 memset(data->header, 0, sizeof data->header);
700 eth = (struct eth_header *)data->header;
701 memcpy(eth->eth_dst, dmac, ETH_ADDR_LEN);
702 memcpy(eth->eth_src, smac, ETH_ADDR_LEN);
703 eth->eth_type = htons(ETH_TYPE_IP);
706 ip = (struct ip_header *) l3;
708 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
709 ip->ip_tos = tnl_flow->tunnel.ip_tos;
710 ip->ip_ttl = tnl_flow->tunnel.ip_ttl;
711 ip->ip_frag_off = (tnl_flow->tunnel.flags & FLOW_TNL_F_DONT_FRAGMENT) ?
714 put_16aligned_be32(&ip->ip_src, ip_src);
715 put_16aligned_be32(&ip->ip_dst, tnl_flow->tunnel.ip_dst);
717 res = netdev_build_header(tnl_port->netdev, data, tnl_flow);
718 ip->ip_csum = csum(ip, sizeof *ip);
719 fat_rwlock_unlock(&rwlock);