1 /* Copyright (c) 2013, 2014 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"
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, tnl_port->match.ip_dst,
199 cfg->dst_port, name);
204 /* Adds 'ofport' to the module with datapath port number 'odp_port'. 'ofport's
205 * must be added before they can be used by the module. 'ofport' must be a
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)
211 fat_rwlock_wrlock(&rwlock);
212 tnl_port_add__(ofport, netdev, odp_port, true, native_tnl, name);
213 fat_rwlock_unlock(&rwlock);
216 /* Checks if the tunnel represented by 'ofport' reconfiguration due to changes
217 * in its netdev_tunnel_config. If it does, returns true. Otherwise, returns
218 * false. 'ofport' and 'odp_port' should be the same as would be passed to
221 tnl_port_reconfigure(const struct ofport_dpif *ofport,
222 const struct netdev *netdev, odp_port_t odp_port,
223 bool native_tnl, const char name[])
226 struct tnl_port *tnl_port;
227 bool changed = false;
229 fat_rwlock_wrlock(&rwlock);
230 tnl_port = tnl_find_ofport(ofport);
232 changed = tnl_port_add__(ofport, netdev, odp_port, false, native_tnl, name);
233 } else if (tnl_port->netdev != netdev
234 || tnl_port->match.odp_port != odp_port
235 || tnl_port->change_seq != netdev_get_change_seq(tnl_port->netdev)) {
236 VLOG_DBG("reconfiguring %s", tnl_port_get_name(tnl_port));
237 tnl_port_del__(ofport);
238 tnl_port_add__(ofport, netdev, odp_port, true, native_tnl, name);
241 fat_rwlock_unlock(&rwlock);
246 tnl_port_del__(const struct ofport_dpif *ofport) OVS_REQ_WRLOCK(rwlock)
248 struct tnl_port *tnl_port;
254 tnl_port = tnl_find_ofport(ofport);
256 const struct netdev_tunnel_config *cfg =
257 netdev_get_tunnel_config(tnl_port->netdev);
260 tnl_port_map_delete(tnl_port->match.ip_dst, cfg->dst_port);
261 tnl_port_mod_log(tnl_port, "removing");
262 map = tnl_match_map(&tnl_port->match);
263 hmap_remove(*map, &tnl_port->match_node);
264 if (hmap_is_empty(*map)) {
269 hmap_remove(ofport_map, &tnl_port->ofport_node);
270 netdev_close(tnl_port->netdev);
275 /* Removes 'ofport' from the module. */
277 tnl_port_del(const struct ofport_dpif *ofport) OVS_EXCLUDED(rwlock)
279 fat_rwlock_wrlock(&rwlock);
280 tnl_port_del__(ofport);
281 fat_rwlock_unlock(&rwlock);
284 /* Looks in the table of tunnels for a tunnel matching the metadata in 'flow'.
285 * Returns the 'ofport' corresponding to the new in_port, or a null pointer if
288 * Callers should verify that 'flow' needs to be received by calling
289 * tnl_port_should_receive() before this function. */
290 const struct ofport_dpif *
291 tnl_port_receive(const struct flow *flow) OVS_EXCLUDED(rwlock)
293 char *pre_flow_str = NULL;
294 const struct ofport_dpif *ofport;
295 struct tnl_port *tnl_port;
297 fat_rwlock_rdlock(&rwlock);
298 tnl_port = tnl_find(flow);
299 ofport = tnl_port ? tnl_port->ofport : NULL;
301 char *flow_str = flow_to_string(flow);
303 VLOG_WARN_RL(&rl, "receive tunnel port not found (%s)", flow_str);
308 if (!VLOG_DROP_DBG(&dbg_rl)) {
309 pre_flow_str = flow_to_string(flow);
313 char *post_flow_str = flow_to_string(flow);
314 char *tnl_str = tnl_port_fmt(tnl_port);
315 VLOG_DBG("flow received\n"
319 tnl_str, pre_flow_str, post_flow_str);
326 fat_rwlock_unlock(&rwlock);
331 tnl_ecn_ok(const struct flow *base_flow, struct flow *flow,
332 struct flow_wildcards *wc)
334 if (is_ip_any(base_flow)) {
335 if ((flow->tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
336 wc->masks.nw_tos |= IP_ECN_MASK;
337 if ((base_flow->nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
338 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
339 " but is not ECN capable");
342 /* Set the ECN CE value in the tunneled packet. */
343 flow->nw_tos |= IP_ECN_CE;
351 /* Should be called at the beginning of action translation to initialize
352 * wildcards and perform any actions based on receiving on tunnel port.
354 * Returns false if the packet must be dropped. */
356 tnl_xlate_init(const struct flow *base_flow, struct flow *flow,
357 struct flow_wildcards *wc)
359 /* tnl_port_should_receive() examines the 'tunnel.ip_dst' field to
360 * determine the presence of the tunnel metadata. However, since tunnels'
361 * datapath port numbers are different from the non-tunnel ports, and we
362 * always unwildcard the 'in_port', we do not need to unwildcard
363 * the 'tunnel.ip_dst' for non-tunneled packets. */
364 if (tnl_port_should_receive(flow)) {
365 wc->masks.tunnel.tun_id = OVS_BE64_MAX;
366 wc->masks.tunnel.ip_src = OVS_BE32_MAX;
367 wc->masks.tunnel.ip_dst = OVS_BE32_MAX;
368 wc->masks.tunnel.flags = (FLOW_TNL_F_DONT_FRAGMENT |
371 wc->masks.tunnel.ip_tos = UINT8_MAX;
372 wc->masks.tunnel.ip_ttl = UINT8_MAX;
373 /* The tp_src and tp_dst members in flow_tnl are set to be always
374 * wildcarded, not to unwildcard them here. */
375 wc->masks.tunnel.tp_src = 0;
376 wc->masks.tunnel.tp_dst = 0;
378 memset(&wc->masks.pkt_mark, 0xff, sizeof wc->masks.pkt_mark);
380 if (!tnl_ecn_ok(base_flow, flow, wc)) {
384 flow->pkt_mark &= ~IPSEC_MARK;
390 /* Given that 'flow' should be output to the ofport corresponding to
391 * 'tnl_port', updates 'flow''s tunnel headers and returns the actual datapath
392 * port that the output should happen on. May return ODPP_NONE if the output
393 * shouldn't occur. */
395 tnl_port_send(const struct ofport_dpif *ofport, struct flow *flow,
396 struct flow_wildcards *wc) OVS_EXCLUDED(rwlock)
398 const struct netdev_tunnel_config *cfg;
399 struct tnl_port *tnl_port;
400 char *pre_flow_str = NULL;
403 fat_rwlock_rdlock(&rwlock);
404 tnl_port = tnl_find_ofport(ofport);
405 out_port = tnl_port ? tnl_port->match.odp_port : ODPP_NONE;
410 cfg = netdev_get_tunnel_config(tnl_port->netdev);
413 if (!VLOG_DROP_DBG(&dbg_rl)) {
414 pre_flow_str = flow_to_string(flow);
417 if (!cfg->ip_src_flow) {
418 flow->tunnel.ip_src = tnl_port->match.ip_src;
420 if (!cfg->ip_dst_flow) {
421 flow->tunnel.ip_dst = tnl_port->match.ip_dst;
423 flow->pkt_mark = tnl_port->match.pkt_mark;
425 if (!cfg->out_key_flow) {
426 flow->tunnel.tun_id = cfg->out_key;
429 if (cfg->ttl_inherit && is_ip_any(flow)) {
430 wc->masks.nw_ttl = 0xff;
431 flow->tunnel.ip_ttl = flow->nw_ttl;
433 flow->tunnel.ip_ttl = cfg->ttl;
436 if (cfg->tos_inherit && is_ip_any(flow)) {
437 wc->masks.nw_tos |= IP_DSCP_MASK;
438 flow->tunnel.ip_tos = flow->nw_tos & IP_DSCP_MASK;
440 flow->tunnel.ip_tos = cfg->tos;
443 /* ECN fields are always inherited. */
444 if (is_ip_any(flow)) {
445 wc->masks.nw_tos |= IP_ECN_MASK;
447 if ((flow->nw_tos & IP_ECN_MASK) == IP_ECN_CE) {
448 flow->tunnel.ip_tos |= IP_ECN_ECT_0;
450 flow->tunnel.ip_tos |= flow->nw_tos & IP_ECN_MASK;
454 flow->tunnel.flags = (cfg->dont_fragment ? FLOW_TNL_F_DONT_FRAGMENT : 0)
455 | (cfg->csum ? FLOW_TNL_F_CSUM : 0)
456 | (cfg->out_key_present ? FLOW_TNL_F_KEY : 0);
459 char *post_flow_str = flow_to_string(flow);
460 char *tnl_str = tnl_port_fmt(tnl_port);
461 VLOG_DBG("flow sent\n"
465 tnl_str, pre_flow_str, post_flow_str);
472 fat_rwlock_unlock(&rwlock);
477 tnl_hash(struct tnl_match *match)
479 BUILD_ASSERT_DECL(sizeof *match % sizeof(uint32_t) == 0);
480 return hash_words((uint32_t *) match, sizeof *match / sizeof(uint32_t), 0);
483 static struct tnl_port *
484 tnl_find_ofport(const struct ofport_dpif *ofport) OVS_REQ_RDLOCK(rwlock)
486 struct tnl_port *tnl_port;
488 HMAP_FOR_EACH_IN_BUCKET (tnl_port, ofport_node, hash_pointer(ofport, 0),
490 if (tnl_port->ofport == ofport) {
497 static struct tnl_port *
498 tnl_find_exact(struct tnl_match *match, struct hmap *map)
499 OVS_REQ_RDLOCK(rwlock)
502 struct tnl_port *tnl_port;
504 HMAP_FOR_EACH_WITH_HASH (tnl_port, match_node, tnl_hash(match), map) {
505 if (!memcmp(match, &tnl_port->match, sizeof *match)) {
513 /* Returns the tnl_port that is the best match for the tunnel data in 'flow',
514 * or NULL if no tnl_port matches 'flow'. */
515 static struct tnl_port *
516 tnl_find(const struct flow *flow) OVS_REQ_RDLOCK(rwlock)
518 enum ip_src_type ip_src;
524 for (in_key_flow = 0; in_key_flow < 2; in_key_flow++) {
525 for (ip_dst_flow = 0; ip_dst_flow < 2; ip_dst_flow++) {
526 for (ip_src = 0; ip_src < 3; ip_src++) {
527 struct hmap *map = tnl_match_maps[i];
530 struct tnl_port *tnl_port;
531 struct tnl_match match;
533 memset(&match, 0, sizeof match);
535 /* The apparent mix-up of 'ip_dst' and 'ip_src' below is
536 * correct, because "struct tnl_match" is expressed in
537 * terms of packets being sent out, but we are using it
538 * here as a description of how to treat received
540 match.in_key = in_key_flow ? 0 : flow->tunnel.tun_id;
541 match.ip_src = (ip_src == IP_SRC_CFG
542 ? flow->tunnel.ip_dst
544 match.ip_dst = ip_dst_flow ? 0 : flow->tunnel.ip_src;
545 match.odp_port = flow->in_port.odp_port;
546 match.pkt_mark = flow->pkt_mark;
547 match.in_key_flow = in_key_flow;
548 match.ip_dst_flow = ip_dst_flow;
549 match.ip_src_flow = ip_src == IP_SRC_FLOW;
551 tnl_port = tnl_find_exact(&match, map);
565 /* Returns a pointer to the 'tnl_match_maps' element corresponding to 'm''s
566 * matching criteria. */
567 static struct hmap **
568 tnl_match_map(const struct tnl_match *m)
570 enum ip_src_type ip_src;
572 ip_src = (m->ip_src_flow ? IP_SRC_FLOW
573 : m->ip_src ? IP_SRC_CFG
576 return &tnl_match_maps[6 * m->in_key_flow + 3 * m->ip_dst_flow + ip_src];
580 tnl_match_fmt(const struct tnl_match *match, struct ds *ds)
581 OVS_REQ_RDLOCK(rwlock)
583 if (!match->ip_dst_flow) {
584 ds_put_format(ds, IP_FMT"->"IP_FMT, IP_ARGS(match->ip_src),
585 IP_ARGS(match->ip_dst));
586 } else if (!match->ip_src_flow) {
587 ds_put_format(ds, IP_FMT"->flow", IP_ARGS(match->ip_src));
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 uint8_t dmac[ETH_ADDR_LEN],
676 uint8_t smac[ETH_ADDR_LEN],
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;
693 memcpy(eth->eth_dst, dmac, ETH_ADDR_LEN);
694 memcpy(eth->eth_src, smac, ETH_ADDR_LEN);
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);
710 ip->ip_csum = csum(ip, sizeof *ip);
711 fat_rwlock_unlock(&rwlock);