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-ports.h"
38 #include "openvswitch/vlog.h"
39 #include "unaligned.h"
40 #include "ofproto-dpif.h"
42 VLOG_DEFINE_THIS_MODULE(tunnel);
44 /* skb mark used for IPsec tunnel packets */
49 struct in6_addr ipv6_src;
50 struct in6_addr ipv6_dst;
59 struct hmap_node ofport_node;
60 struct hmap_node match_node;
62 const struct ofport_dpif *ofport;
64 struct netdev *netdev;
66 struct tnl_match match;
69 static struct fat_rwlock rwlock;
73 * This module maps packets received over tunnel protocols to vports. The
74 * tunnel protocol and, for some protocols, tunnel-specific information (e.g.,
75 * for VXLAN, the UDP destination port number) are always use as part of the
76 * mapping. Which other fields are used for the mapping depends on the vports
77 * themselves (the parenthesized notations refer to "struct tnl_match" fields):
79 * - in_key: A vport may match a specific tunnel ID (in_key_flow == false)
80 * or arrange for the tunnel ID to be matched as tunnel.tun_id in the
81 * OpenFlow flow (in_key_flow == true).
83 * - ip_dst: A vport may match a specific destination IP address
84 * (ip_dst_flow == false) or arrange for the destination IP to be matched
85 * as tunnel.ip_dst in the OpenFlow flow (ip_dst_flow == true).
87 * - ip_src: A vport may match a specific IP source address (ip_src_flow ==
88 * false, ip_src != 0), wildcard all source addresses (ip_src_flow ==
89 * false, ip_src == 0), or arrange for the IP source address to be
90 * handled in the OpenFlow flow table (ip_src_flow == true).
92 * Thus, there are 2 * 2 * 3 == 12 possible ways a vport can match against a
93 * tunnel packet. We number the possibilities for each field in increasing
94 * order as listed in each bullet above. We order the 12 overall combinations
95 * in lexicographic order considering in_key first, then ip_dst, then
97 #define N_MATCH_TYPES (2 * 2 * 3)
99 /* The three possibilities (see above) for vport ip_src matches. */
101 IP_SRC_CFG, /* ip_src must equal configured address. */
102 IP_SRC_ANY, /* Any ip_src is acceptable. */
103 IP_SRC_FLOW /* ip_src is handled in flow table. */
106 /* Each hmap contains "struct tnl_port"s.
107 * The index is a combination of how each of the fields listed under "Tunnel
108 * matches" above matches, see the final paragraph for ordering. */
109 static struct hmap *tnl_match_maps[N_MATCH_TYPES] OVS_GUARDED_BY(rwlock);
110 static struct hmap **tnl_match_map(const struct tnl_match *);
112 static struct hmap ofport_map__ = HMAP_INITIALIZER(&ofport_map__);
113 static struct hmap *ofport_map OVS_GUARDED_BY(rwlock) = &ofport_map__;
115 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
116 static struct vlog_rate_limit dbg_rl = VLOG_RATE_LIMIT_INIT(60, 60);
118 static struct tnl_port *tnl_find(const struct flow *) OVS_REQ_RDLOCK(rwlock);
119 static struct tnl_port *tnl_find_exact(struct tnl_match *, struct hmap *)
120 OVS_REQ_RDLOCK(rwlock);
121 static struct tnl_port *tnl_find_ofport(const struct ofport_dpif *)
122 OVS_REQ_RDLOCK(rwlock);
124 static uint32_t tnl_hash(struct tnl_match *);
125 static void tnl_match_fmt(const struct tnl_match *, struct ds *);
126 static char *tnl_port_fmt(const struct tnl_port *) OVS_REQ_RDLOCK(rwlock);
127 static void tnl_port_mod_log(const struct tnl_port *, const char *action)
128 OVS_REQ_RDLOCK(rwlock);
129 static const char *tnl_port_get_name(const struct tnl_port *)
130 OVS_REQ_RDLOCK(rwlock);
131 static void tnl_port_del__(const struct ofport_dpif *) OVS_REQ_WRLOCK(rwlock);
134 ofproto_tunnel_init(void)
136 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
138 if (ovsthread_once_start(&once)) {
139 fat_rwlock_init(&rwlock);
140 ovsthread_once_done(&once);
145 tnl_port_add__(const struct ofport_dpif *ofport, const struct netdev *netdev,
146 odp_port_t odp_port, bool warn, bool native_tnl, const char name[])
147 OVS_REQ_WRLOCK(rwlock)
149 const struct netdev_tunnel_config *cfg;
150 struct tnl_port *existing_port;
151 struct tnl_port *tnl_port;
154 cfg = netdev_get_tunnel_config(netdev);
157 tnl_port = xzalloc(sizeof *tnl_port);
158 tnl_port->ofport = ofport;
159 tnl_port->netdev = netdev_ref(netdev);
160 tnl_port->change_seq = netdev_get_change_seq(tnl_port->netdev);
162 tnl_port->match.in_key = cfg->in_key;
163 tnl_port->match.ipv6_src = cfg->ipv6_src;
164 tnl_port->match.ipv6_dst = cfg->ipv6_dst;
165 tnl_port->match.ip_src_flow = cfg->ip_src_flow;
166 tnl_port->match.ip_dst_flow = cfg->ip_dst_flow;
167 tnl_port->match.pkt_mark = cfg->ipsec ? IPSEC_MARK : 0;
168 tnl_port->match.in_key_flow = cfg->in_key_flow;
169 tnl_port->match.odp_port = odp_port;
171 map = tnl_match_map(&tnl_port->match);
172 existing_port = tnl_find_exact(&tnl_port->match, *map);
175 struct ds ds = DS_EMPTY_INITIALIZER;
176 tnl_match_fmt(&tnl_port->match, &ds);
177 VLOG_WARN("%s: attempting to add tunnel port with same config as "
178 "port '%s' (%s)", tnl_port_get_name(tnl_port),
179 tnl_port_get_name(existing_port), ds_cstr(&ds));
182 netdev_close(tnl_port->netdev);
187 hmap_insert(ofport_map, &tnl_port->ofport_node, hash_pointer(ofport, 0));
190 *map = xmalloc(sizeof **map);
193 hmap_insert(*map, &tnl_port->match_node, tnl_hash(&tnl_port->match));
194 tnl_port_mod_log(tnl_port, "adding");
197 tnl_port_map_insert(odp_port, cfg->dst_port, name);
202 /* Adds 'ofport' to the module with datapath port number 'odp_port'. 'ofport's
203 * must be added before they can be used by the module. 'ofport' must be a
206 * Returns 0 if successful, otherwise a positive errno value. */
208 tnl_port_add(const struct ofport_dpif *ofport, const struct netdev *netdev,
209 odp_port_t odp_port, bool native_tnl, const char name[]) OVS_EXCLUDED(rwlock)
213 fat_rwlock_wrlock(&rwlock);
214 ok = tnl_port_add__(ofport, netdev, odp_port, true, native_tnl, name);
215 fat_rwlock_unlock(&rwlock);
217 return ok ? 0 : EEXIST;
220 /* Checks if the tunnel represented by 'ofport' reconfiguration due to changes
221 * in its netdev_tunnel_config. If it does, returns true. Otherwise, returns
222 * false. 'ofport' and 'odp_port' should be the same as would be passed to
225 tnl_port_reconfigure(const struct ofport_dpif *ofport,
226 const struct netdev *netdev, odp_port_t odp_port,
227 bool native_tnl, const char name[])
230 struct tnl_port *tnl_port;
231 bool changed = false;
233 fat_rwlock_wrlock(&rwlock);
234 tnl_port = tnl_find_ofport(ofport);
236 changed = tnl_port_add__(ofport, netdev, odp_port, false, native_tnl, name);
237 } else if (tnl_port->netdev != netdev
238 || tnl_port->match.odp_port != odp_port
239 || tnl_port->change_seq != netdev_get_change_seq(tnl_port->netdev)) {
240 VLOG_DBG("reconfiguring %s", tnl_port_get_name(tnl_port));
241 tnl_port_del__(ofport);
242 tnl_port_add__(ofport, netdev, odp_port, true, native_tnl, name);
245 fat_rwlock_unlock(&rwlock);
250 tnl_port_del__(const struct ofport_dpif *ofport) OVS_REQ_WRLOCK(rwlock)
252 struct tnl_port *tnl_port;
258 tnl_port = tnl_find_ofport(ofport);
260 const struct netdev_tunnel_config *cfg =
261 netdev_get_tunnel_config(tnl_port->netdev);
264 tnl_port_map_delete(cfg->dst_port);
265 tnl_port_mod_log(tnl_port, "removing");
266 map = tnl_match_map(&tnl_port->match);
267 hmap_remove(*map, &tnl_port->match_node);
268 if (hmap_is_empty(*map)) {
273 hmap_remove(ofport_map, &tnl_port->ofport_node);
274 netdev_close(tnl_port->netdev);
279 /* Removes 'ofport' from the module. */
281 tnl_port_del(const struct ofport_dpif *ofport) OVS_EXCLUDED(rwlock)
283 fat_rwlock_wrlock(&rwlock);
284 tnl_port_del__(ofport);
285 fat_rwlock_unlock(&rwlock);
288 /* Looks in the table of tunnels for a tunnel matching the metadata in 'flow'.
289 * Returns the 'ofport' corresponding to the new in_port, or a null pointer if
292 * Callers should verify that 'flow' needs to be received by calling
293 * tnl_port_should_receive() before this function. */
294 const struct ofport_dpif *
295 tnl_port_receive(const struct flow *flow) OVS_EXCLUDED(rwlock)
297 char *pre_flow_str = NULL;
298 const struct ofport_dpif *ofport;
299 struct tnl_port *tnl_port;
301 fat_rwlock_rdlock(&rwlock);
302 tnl_port = tnl_find(flow);
303 ofport = tnl_port ? tnl_port->ofport : NULL;
305 char *flow_str = flow_to_string(flow);
307 VLOG_WARN_RL(&rl, "receive tunnel port not found (%s)", flow_str);
312 if (!VLOG_DROP_DBG(&dbg_rl)) {
313 pre_flow_str = flow_to_string(flow);
317 char *post_flow_str = flow_to_string(flow);
318 char *tnl_str = tnl_port_fmt(tnl_port);
319 VLOG_DBG("flow received\n"
323 tnl_str, pre_flow_str, post_flow_str);
330 fat_rwlock_unlock(&rwlock);
334 /* Should be called at the beginning of action translation to initialize
335 * wildcards and perform any actions based on receiving on tunnel port.
337 * Returns false if the packet must be dropped. */
339 tnl_process_ecn(struct flow *flow)
341 if (!tnl_port_should_receive(flow)) {
345 if (is_ip_any(flow) && (flow->tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
346 if ((flow->nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
347 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
348 " but is not ECN capable");
352 /* Set the ECN CE value in the tunneled packet. */
353 flow->nw_tos |= IP_ECN_CE;
356 flow->pkt_mark &= ~IPSEC_MARK;
361 tnl_wc_init(struct flow *flow, struct flow_wildcards *wc)
363 if (tnl_port_should_receive(flow)) {
364 wc->masks.tunnel.tun_id = OVS_BE64_MAX;
365 wc->masks.tunnel.ip_src = OVS_BE32_MAX;
366 wc->masks.tunnel.ip_dst = OVS_BE32_MAX;
367 wc->masks.tunnel.flags = (FLOW_TNL_F_DONT_FRAGMENT |
370 wc->masks.tunnel.ip_tos = UINT8_MAX;
371 wc->masks.tunnel.ip_ttl = UINT8_MAX;
372 /* The tp_src and tp_dst members in flow_tnl are set to be always
373 * wildcarded, not to unwildcard them here. */
374 wc->masks.tunnel.tp_src = 0;
375 wc->masks.tunnel.tp_dst = 0;
377 memset(&wc->masks.pkt_mark, 0xff, sizeof wc->masks.pkt_mark);
380 && (flow->tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
381 wc->masks.nw_tos |= IP_ECN_MASK;
386 /* Given that 'flow' should be output to the ofport corresponding to
387 * 'tnl_port', updates 'flow''s tunnel headers and returns the actual datapath
388 * port that the output should happen on. May return ODPP_NONE if the output
389 * shouldn't occur. */
391 tnl_port_send(const struct ofport_dpif *ofport, struct flow *flow,
392 struct flow_wildcards *wc) OVS_EXCLUDED(rwlock)
394 const struct netdev_tunnel_config *cfg;
395 struct tnl_port *tnl_port;
396 char *pre_flow_str = NULL;
399 fat_rwlock_rdlock(&rwlock);
400 tnl_port = tnl_find_ofport(ofport);
401 out_port = tnl_port ? tnl_port->match.odp_port : ODPP_NONE;
406 cfg = netdev_get_tunnel_config(tnl_port->netdev);
409 if (!VLOG_DROP_DBG(&dbg_rl)) {
410 pre_flow_str = flow_to_string(flow);
413 if (!cfg->ip_src_flow) {
414 flow->tunnel.ip_src = in6_addr_get_mapped_ipv4(&tnl_port->match.ipv6_src);
416 if (!cfg->ip_dst_flow) {
417 flow->tunnel.ip_dst = in6_addr_get_mapped_ipv4(&tnl_port->match.ipv6_dst);
419 flow->pkt_mark = tnl_port->match.pkt_mark;
421 if (!cfg->out_key_flow) {
422 flow->tunnel.tun_id = cfg->out_key;
425 if (cfg->ttl_inherit && is_ip_any(flow)) {
426 wc->masks.nw_ttl = 0xff;
427 flow->tunnel.ip_ttl = flow->nw_ttl;
429 flow->tunnel.ip_ttl = cfg->ttl;
432 if (cfg->tos_inherit && is_ip_any(flow)) {
433 wc->masks.nw_tos |= IP_DSCP_MASK;
434 flow->tunnel.ip_tos = flow->nw_tos & IP_DSCP_MASK;
436 flow->tunnel.ip_tos = cfg->tos;
439 /* ECN fields are always inherited. */
440 if (is_ip_any(flow)) {
441 wc->masks.nw_tos |= IP_ECN_MASK;
443 if ((flow->nw_tos & IP_ECN_MASK) == IP_ECN_CE) {
444 flow->tunnel.ip_tos |= IP_ECN_ECT_0;
446 flow->tunnel.ip_tos |= flow->nw_tos & IP_ECN_MASK;
450 flow->tunnel.flags |= (cfg->dont_fragment ? FLOW_TNL_F_DONT_FRAGMENT : 0)
451 | (cfg->csum ? FLOW_TNL_F_CSUM : 0)
452 | (cfg->out_key_present ? FLOW_TNL_F_KEY : 0);
455 char *post_flow_str = flow_to_string(flow);
456 char *tnl_str = tnl_port_fmt(tnl_port);
457 VLOG_DBG("flow sent\n"
461 tnl_str, pre_flow_str, post_flow_str);
468 fat_rwlock_unlock(&rwlock);
473 tnl_hash(struct tnl_match *match)
475 BUILD_ASSERT_DECL(sizeof *match % sizeof(uint32_t) == 0);
476 return hash_words((uint32_t *) match, sizeof *match / sizeof(uint32_t), 0);
479 static struct tnl_port *
480 tnl_find_ofport(const struct ofport_dpif *ofport) OVS_REQ_RDLOCK(rwlock)
482 struct tnl_port *tnl_port;
484 HMAP_FOR_EACH_IN_BUCKET (tnl_port, ofport_node, hash_pointer(ofport, 0),
486 if (tnl_port->ofport == ofport) {
493 static struct tnl_port *
494 tnl_find_exact(struct tnl_match *match, struct hmap *map)
495 OVS_REQ_RDLOCK(rwlock)
498 struct tnl_port *tnl_port;
500 HMAP_FOR_EACH_WITH_HASH (tnl_port, match_node, tnl_hash(match), map) {
501 if (!memcmp(match, &tnl_port->match, sizeof *match)) {
509 /* Returns the tnl_port that is the best match for the tunnel data in 'flow',
510 * or NULL if no tnl_port matches 'flow'. */
511 static struct tnl_port *
512 tnl_find(const struct flow *flow) OVS_REQ_RDLOCK(rwlock)
514 enum ip_src_type ip_src;
520 for (in_key_flow = 0; in_key_flow < 2; in_key_flow++) {
521 for (ip_dst_flow = 0; ip_dst_flow < 2; ip_dst_flow++) {
522 for (ip_src = 0; ip_src < 3; ip_src++) {
523 struct hmap *map = tnl_match_maps[i];
526 struct tnl_port *tnl_port;
527 struct tnl_match match;
529 memset(&match, 0, sizeof match);
531 /* The apparent mix-up of 'ip_dst' and 'ip_src' below is
532 * correct, because "struct tnl_match" is expressed in
533 * terms of packets being sent out, but we are using it
534 * here as a description of how to treat received
536 match.in_key = in_key_flow ? 0 : flow->tunnel.tun_id;
537 if (ip_src == IP_SRC_CFG && flow->tunnel.ip_dst) {
538 in6_addr_set_mapped_ipv4(&match.ipv6_src, flow->tunnel.ip_dst);
540 if (!ip_dst_flow && flow->tunnel.ip_src) {
541 in6_addr_set_mapped_ipv4(&match.ipv6_dst, flow->tunnel.ip_src);
543 match.odp_port = flow->in_port.odp_port;
544 match.pkt_mark = flow->pkt_mark;
545 match.in_key_flow = in_key_flow;
546 match.ip_dst_flow = ip_dst_flow;
547 match.ip_src_flow = ip_src == IP_SRC_FLOW;
549 tnl_port = tnl_find_exact(&match, map);
563 /* Returns a pointer to the 'tnl_match_maps' element corresponding to 'm''s
564 * matching criteria. */
565 static struct hmap **
566 tnl_match_map(const struct tnl_match *m)
568 enum ip_src_type ip_src;
570 ip_src = (m->ip_src_flow ? IP_SRC_FLOW
571 : ipv6_addr_is_set(&m->ipv6_src) ? IP_SRC_CFG
574 return &tnl_match_maps[6 * m->in_key_flow + 3 * m->ip_dst_flow + ip_src];
578 tnl_match_fmt(const struct tnl_match *match, struct ds *ds)
579 OVS_REQ_RDLOCK(rwlock)
581 if (!match->ip_dst_flow) {
582 ipv6_format_mapped(&match->ipv6_src, ds);
583 ds_put_cstr(ds, "->");
584 ipv6_format_mapped(&match->ipv6_dst, ds);
585 } else if (!match->ip_src_flow) {
586 ipv6_format_mapped(&match->ipv6_src, ds);
587 ds_put_cstr(ds, "->flow");
589 ds_put_cstr(ds, "flow->flow");
592 if (match->in_key_flow) {
593 ds_put_cstr(ds, ", key=flow");
595 ds_put_format(ds, ", key=%#"PRIx64, ntohll(match->in_key));
598 ds_put_format(ds, ", dp port=%"PRIu32, match->odp_port);
599 ds_put_format(ds, ", pkt mark=%"PRIu32, match->pkt_mark);
603 tnl_port_mod_log(const struct tnl_port *tnl_port, const char *action)
604 OVS_REQ_RDLOCK(rwlock)
606 if (VLOG_IS_DBG_ENABLED()) {
607 struct ds ds = DS_EMPTY_INITIALIZER;
609 tnl_match_fmt(&tnl_port->match, &ds);
610 VLOG_INFO("%s tunnel port %s (%s)", action,
611 tnl_port_get_name(tnl_port), ds_cstr(&ds));
617 tnl_port_fmt(const struct tnl_port *tnl_port) OVS_REQ_RDLOCK(rwlock)
619 const struct netdev_tunnel_config *cfg =
620 netdev_get_tunnel_config(tnl_port->netdev);
621 struct ds ds = DS_EMPTY_INITIALIZER;
623 ds_put_format(&ds, "port %"PRIu32": %s (%s: ", tnl_port->match.odp_port,
624 tnl_port_get_name(tnl_port),
625 netdev_get_type(tnl_port->netdev));
626 tnl_match_fmt(&tnl_port->match, &ds);
628 if (cfg->out_key != cfg->in_key ||
629 cfg->out_key_present != cfg->in_key_present ||
630 cfg->out_key_flow != cfg->in_key_flow) {
631 ds_put_cstr(&ds, ", out_key=");
632 if (!cfg->out_key_present) {
633 ds_put_cstr(&ds, "none");
634 } else if (cfg->out_key_flow) {
635 ds_put_cstr(&ds, "flow");
637 ds_put_format(&ds, "%#"PRIx64, ntohll(cfg->out_key));
641 if (cfg->ttl_inherit) {
642 ds_put_cstr(&ds, ", ttl=inherit");
644 ds_put_format(&ds, ", ttl=%"PRIu8, cfg->ttl);
647 if (cfg->tos_inherit) {
648 ds_put_cstr(&ds, ", tos=inherit");
649 } else if (cfg->tos) {
650 ds_put_format(&ds, ", tos=%#"PRIx8, cfg->tos);
653 if (!cfg->dont_fragment) {
654 ds_put_cstr(&ds, ", df=false");
658 ds_put_cstr(&ds, ", csum=true");
661 ds_put_cstr(&ds, ")\n");
663 return ds_steal_cstr(&ds);
667 tnl_port_get_name(const struct tnl_port *tnl_port) OVS_REQ_RDLOCK(rwlock)
669 return netdev_get_name(tnl_port->netdev);
673 tnl_port_build_header(const struct ofport_dpif *ofport,
674 const struct flow *tnl_flow,
675 const struct eth_addr dmac,
676 const struct eth_addr smac,
677 ovs_be32 ip_src, struct ovs_action_push_tnl *data)
679 struct tnl_port *tnl_port;
680 struct eth_header *eth;
681 struct ip_header *ip;
685 fat_rwlock_rdlock(&rwlock);
686 tnl_port = tnl_find_ofport(ofport);
687 ovs_assert(tnl_port);
689 /* Build Ethernet and IP headers. */
690 memset(data->header, 0, sizeof data->header);
692 eth = (struct eth_header *)data->header;
695 eth->eth_type = htons(ETH_TYPE_IP);
698 ip = (struct ip_header *) l3;
700 ip->ip_ihl_ver = IP_IHL_VER(5, 4);
701 ip->ip_tos = tnl_flow->tunnel.ip_tos;
702 ip->ip_ttl = tnl_flow->tunnel.ip_ttl;
703 ip->ip_frag_off = (tnl_flow->tunnel.flags & FLOW_TNL_F_DONT_FRAGMENT) ?
706 put_16aligned_be32(&ip->ip_src, ip_src);
707 put_16aligned_be32(&ip->ip_dst, tnl_flow->tunnel.ip_dst);
709 res = netdev_build_header(tnl_port->netdev, data, tnl_flow);
710 ip->ip_csum = csum(ip, sizeof *ip);
711 fat_rwlock_unlock(&rwlock);