ofproto/tunnel.c

   1 /* Copyright (c) 2013 Nicira, Inc.
   2  *
   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:
   6  *
   7  *     http://www.apache.org/licenses/LICENSE-2.0
   8  *
   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. */
  14
  15 #include <config.h>
  16 #include "tunnel.h"
  17
  18 #include <errno.h>
  19
  20 #include "byte-order.h"
  21 #include "dynamic-string.h"
  22 #include "hash.h"
  23 #include "hmap.h"
  24 #include "netdev.h"
  25 #include "odp-util.h"
  26 #include "packets.h"
  27 #include "smap.h"
  28 #include "socket-util.h"
  29 #include "tunnel.h"
  30 #include "vlog.h"
  31
  32 VLOG_DEFINE_THIS_MODULE(tunnel);
  33
  34 /* skb mark used for IPsec tunnel packets */
  35 #define IPSEC_MARK 1
  36
  37 struct tnl_match {
  38     ovs_be64 in_key;
  39     ovs_be32 ip_src;
  40     ovs_be32 ip_dst;
  41     odp_port_t odp_port;
  42     uint32_t pkt_mark;
  43     bool in_key_flow;
  44     bool ip_src_flow;
  45     bool ip_dst_flow;
  46 };
  47
  48 struct tnl_port {
  49     struct hmap_node ofport_node;
  50     struct hmap_node match_node;
  51
  52     const struct ofport_dpif *ofport;
  53     unsigned int netdev_seq;
  54     struct netdev *netdev;
  55
  56     struct tnl_match match;
  57 };
  58
  59 static struct ovs_rwlock rwlock = OVS_RWLOCK_INITIALIZER;
  60
  61 static struct hmap tnl_match_map__ = HMAP_INITIALIZER(&tnl_match_map__);
  62 static struct hmap *tnl_match_map OVS_GUARDED_BY(rwlock) = &tnl_match_map__;
  63
  64 static struct hmap ofport_map__ = HMAP_INITIALIZER(&ofport_map__);
  65 static struct hmap *ofport_map OVS_GUARDED_BY(rwlock) = &ofport_map__;
  66
  67 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
  68 static struct vlog_rate_limit dbg_rl = VLOG_RATE_LIMIT_INIT(60, 60);
  69
  70 static struct tnl_port *tnl_find(struct tnl_match *) OVS_REQ_RDLOCK(rwlock);
  71 static struct tnl_port *tnl_find_exact(struct tnl_match *)
  72     OVS_REQ_RDLOCK(rwlock);
  73 static struct tnl_port *tnl_find_ofport(const struct ofport_dpif *)
  74     OVS_REQ_RDLOCK(rwlock);
  75
  76 static uint32_t tnl_hash(struct tnl_match *);
  77 static void tnl_match_fmt(const struct tnl_match *, struct ds *);
  78 static char *tnl_port_fmt(const struct tnl_port *) OVS_REQ_RDLOCK(rwlock);
  79 static void tnl_port_mod_log(const struct tnl_port *, const char *action)
  80     OVS_REQ_RDLOCK(rwlock);
  81 static const char *tnl_port_get_name(const struct tnl_port *)
  82     OVS_REQ_RDLOCK(rwlock);
  83 static void tnl_port_del__(const struct ofport_dpif *) OVS_REQ_WRLOCK(rwlock);
  84
  85 static bool
  86 tnl_port_add__(const struct ofport_dpif *ofport, const struct netdev *netdev,
  87                odp_port_t odp_port, bool warn)
  88     OVS_REQ_WRLOCK(rwlock)
  89 {
  90     const struct netdev_tunnel_config *cfg;
  91     struct tnl_port *existing_port;
  92     struct tnl_port *tnl_port;
  93
  94     cfg = netdev_get_tunnel_config(netdev);
  95     ovs_assert(cfg);
  96
  97     tnl_port = xzalloc(sizeof *tnl_port);
  98     tnl_port->ofport = ofport;
  99     tnl_port->netdev = netdev_ref(netdev);
 100     tnl_port->netdev_seq = netdev_change_seq(tnl_port->netdev);
 101
 102     tnl_port->match.in_key = cfg->in_key;
 103     tnl_port->match.ip_src = cfg->ip_src;
 104     tnl_port->match.ip_dst = cfg->ip_dst;
 105     tnl_port->match.ip_src_flow = cfg->ip_src_flow;
 106     tnl_port->match.ip_dst_flow = cfg->ip_dst_flow;
 107     tnl_port->match.pkt_mark = cfg->ipsec ? IPSEC_MARK : 0;
 108     tnl_port->match.in_key_flow = cfg->in_key_flow;
 109     tnl_port->match.odp_port = odp_port;
 110
 111     existing_port = tnl_find_exact(&tnl_port->match);
 112     if (existing_port) {
 113         if (warn) {
 114             struct ds ds = DS_EMPTY_INITIALIZER;
 115             tnl_match_fmt(&tnl_port->match, &ds);
 116             VLOG_WARN("%s: attempting to add tunnel port with same config as "
 117                       "port '%s' (%s)", tnl_port_get_name(tnl_port),
 118                       tnl_port_get_name(existing_port), ds_cstr(&ds));
 119             ds_destroy(&ds);
 120             free(tnl_port);
 121         }
 122         return false;
 123     }
 124
 125     hmap_insert(ofport_map, &tnl_port->ofport_node, hash_pointer(ofport, 0));
 126     hmap_insert(tnl_match_map, &tnl_port->match_node,
 127                 tnl_hash(&tnl_port->match));
 128     tnl_port_mod_log(tnl_port, "adding");
 129     return true;
 130 }
 131
 132 /* Adds 'ofport' to the module with datapath port number 'odp_port'. 'ofport's
 133  * must be added before they can be used by the module. 'ofport' must be a
 134  * tunnel. */
 135 void
 136 tnl_port_add(const struct ofport_dpif *ofport, const struct netdev *netdev,
 137              odp_port_t odp_port) OVS_EXCLUDED(rwlock)
 138 {
 139     ovs_rwlock_wrlock(&rwlock);
 140     tnl_port_add__(ofport, netdev, odp_port, true);
 141     ovs_rwlock_unlock(&rwlock);
 142 }
 143
 144 /* Checks if the tunnel represented by 'ofport' reconfiguration due to changes
 145  * in its netdev_tunnel_config.  If it does, returns true. Otherwise, returns
 146  * false.  'ofport' and 'odp_port' should be the same as would be passed to
 147  * tnl_port_add(). */
 148 bool
 149 tnl_port_reconfigure(const struct ofport_dpif *ofport,
 150                      const struct netdev *netdev, odp_port_t odp_port)
 151     OVS_EXCLUDED(rwlock)
 152 {
 153     struct tnl_port *tnl_port;
 154     bool changed = false;
 155
 156     ovs_rwlock_wrlock(&rwlock);
 157     tnl_port = tnl_find_ofport(ofport);
 158     if (!tnl_port) {
 159         changed = tnl_port_add__(ofport, netdev, odp_port, false);
 160     } else if (tnl_port->netdev != netdev
 161                || tnl_port->match.odp_port != odp_port
 162                || tnl_port->netdev_seq != netdev_change_seq(netdev)) {
 163         VLOG_DBG("reconfiguring %s", tnl_port_get_name(tnl_port));
 164         tnl_port_del__(ofport);
 165         tnl_port_add__(ofport, netdev, odp_port, true);
 166         changed = true;
 167     }
 168     ovs_rwlock_unlock(&rwlock);
 169     return changed;
 170 }
 171
 172 static void
 173 tnl_port_del__(const struct ofport_dpif *ofport) OVS_REQ_WRLOCK(rwlock)
 174 {
 175     struct tnl_port *tnl_port;
 176
 177     if (!ofport) {
 178         return;
 179     }
 180
 181     tnl_port = tnl_find_ofport(ofport);
 182     if (tnl_port) {
 183         tnl_port_mod_log(tnl_port, "removing");
 184         hmap_remove(tnl_match_map, &tnl_port->match_node);
 185         hmap_remove(ofport_map, &tnl_port->ofport_node);
 186         netdev_close(tnl_port->netdev);
 187         free(tnl_port);
 188     }
 189 }
 190
 191 /* Removes 'ofport' from the module. */
 192 void
 193 tnl_port_del(const struct ofport_dpif *ofport) OVS_EXCLUDED(rwlock)
 194 {
 195     ovs_rwlock_wrlock(&rwlock);
 196     tnl_port_del__(ofport);
 197     ovs_rwlock_unlock(&rwlock);
 198 }
 199
 200 /* Looks in the table of tunnels for a tunnel matching the metadata in 'flow'.
 201  * Returns the 'ofport' corresponding to the new in_port, or a null pointer if
 202  * none is found.
 203  *
 204  * Callers should verify that 'flow' needs to be received by calling
 205  * tnl_port_should_receive() before this function. */
 206 const struct ofport_dpif *
 207 tnl_port_receive(const struct flow *flow) OVS_EXCLUDED(rwlock)
 208 {
 209     char *pre_flow_str = NULL;
 210     const struct ofport_dpif *ofport;
 211     struct tnl_port *tnl_port;
 212     struct tnl_match match;
 213
 214     memset(&match, 0, sizeof match);
 215     match.odp_port = flow->in_port.odp_port;
 216     match.ip_src = flow->tunnel.ip_dst;
 217     match.ip_dst = flow->tunnel.ip_src;
 218     match.in_key = flow->tunnel.tun_id;
 219     match.pkt_mark = flow->pkt_mark;
 220
 221     ovs_rwlock_rdlock(&rwlock);
 222     tnl_port = tnl_find(&match);
 223     ofport = tnl_port ? tnl_port->ofport : NULL;
 224     if (!tnl_port) {
 225         struct ds ds = DS_EMPTY_INITIALIZER;
 226
 227         tnl_match_fmt(&match, &ds);
 228         VLOG_WARN_RL(&rl, "receive tunnel port not found (%s)", ds_cstr(&ds));
 229         ds_destroy(&ds);
 230         goto out;
 231     }
 232
 233     if (!VLOG_DROP_DBG(&dbg_rl)) {
 234         pre_flow_str = flow_to_string(flow);
 235     }
 236
 237     if (pre_flow_str) {
 238         char *post_flow_str = flow_to_string(flow);
 239         char *tnl_str = tnl_port_fmt(tnl_port);
 240         VLOG_DBG("flow received\n"
 241                  "%s"
 242                  " pre: %s\n"
 243                  "post: %s",
 244                  tnl_str, pre_flow_str, post_flow_str);
 245         free(tnl_str);
 246         free(pre_flow_str);
 247         free(post_flow_str);
 248     }
 249
 250 out:
 251     ovs_rwlock_unlock(&rwlock);
 252     return ofport;
 253 }
 254
 255 static bool
 256 tnl_ecn_ok(const struct flow *base_flow, struct flow *flow)
 257 {
 258     if (is_ip_any(base_flow)
 259         && (flow->tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
 260         if ((base_flow->nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
 261             VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
 262                          " but is not ECN capable");
 263             return false;
 264         } else {
 265             /* Set the ECN CE value in the tunneled packet. */
 266             flow->nw_tos |= IP_ECN_CE;
 267         }
 268     }
 269
 270     return true;
 271 }
 272
 273 /* Should be called at the beginning of action translation to initialize
 274  * wildcards and perform any actions based on receiving on tunnel port.
 275  *
 276  * Returns false if the packet must be dropped. */
 277 bool
 278 tnl_xlate_init(const struct flow *base_flow, struct flow *flow,
 279                struct flow_wildcards *wc)
 280 {
 281     if (tnl_port_should_receive(flow)) {
 282         memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
 283         memset(&wc->masks.pkt_mark, 0xff, sizeof wc->masks.pkt_mark);
 284
 285         if (!tnl_ecn_ok(base_flow, flow)) {
 286             return false;
 287         }
 288
 289         flow->pkt_mark &= ~IPSEC_MARK;
 290     }
 291
 292     return true;
 293 }
 294
 295 /* Given that 'flow' should be output to the ofport corresponding to
 296  * 'tnl_port', updates 'flow''s tunnel headers and returns the actual datapath
 297  * port that the output should happen on.  May return ODPP_NONE if the output
 298  * shouldn't occur. */
 299 odp_port_t
 300 tnl_port_send(const struct ofport_dpif *ofport, struct flow *flow,
 301               struct flow_wildcards *wc) OVS_EXCLUDED(rwlock)
 302 {
 303     const struct netdev_tunnel_config *cfg;
 304     struct tnl_port *tnl_port;
 305     char *pre_flow_str = NULL;
 306     odp_port_t out_port;
 307
 308     ovs_rwlock_rdlock(&rwlock);
 309     tnl_port = tnl_find_ofport(ofport);
 310     out_port = tnl_port ? tnl_port->match.odp_port : ODPP_NONE;
 311     if (!tnl_port) {
 312         goto out;
 313     }
 314
 315     cfg = netdev_get_tunnel_config(tnl_port->netdev);
 316     ovs_assert(cfg);
 317
 318     if (!VLOG_DROP_DBG(&dbg_rl)) {
 319         pre_flow_str = flow_to_string(flow);
 320     }
 321
 322     if (!cfg->ip_src_flow) {
 323         flow->tunnel.ip_src = tnl_port->match.ip_src;
 324     }
 325     if (!cfg->ip_dst_flow) {
 326         flow->tunnel.ip_dst = tnl_port->match.ip_dst;
 327     }
 328     flow->pkt_mark = tnl_port->match.pkt_mark;
 329
 330     if (!cfg->out_key_flow) {
 331         flow->tunnel.tun_id = cfg->out_key;
 332     }
 333
 334     if (cfg->ttl_inherit && is_ip_any(flow)) {
 335         wc->masks.nw_ttl = 0xff;
 336         flow->tunnel.ip_ttl = flow->nw_ttl;
 337     } else {
 338         flow->tunnel.ip_ttl = cfg->ttl;
 339     }
 340
 341     if (cfg->tos_inherit && is_ip_any(flow)) {
 342         wc->masks.nw_tos = 0xff;
 343         flow->tunnel.ip_tos = flow->nw_tos & IP_DSCP_MASK;
 344     } else {
 345         flow->tunnel.ip_tos = cfg->tos;
 346     }
 347
 348     /* ECN fields are always inherited. */
 349     if (is_ip_any(flow)) {
 350         wc->masks.nw_tos |= IP_ECN_MASK;
 351     }
 352
 353     if ((flow->nw_tos & IP_ECN_MASK) == IP_ECN_CE) {
 354         flow->tunnel.ip_tos |= IP_ECN_ECT_0;
 355     } else {
 356         flow->tunnel.ip_tos |= flow->nw_tos & IP_ECN_MASK;
 357     }
 358
 359     flow->tunnel.flags = (cfg->dont_fragment ? FLOW_TNL_F_DONT_FRAGMENT : 0)
 360         | (cfg->csum ? FLOW_TNL_F_CSUM : 0)
 361         | (cfg->out_key_present ? FLOW_TNL_F_KEY : 0);
 362
 363     if (pre_flow_str) {
 364         char *post_flow_str = flow_to_string(flow);
 365         char *tnl_str = tnl_port_fmt(tnl_port);
 366         VLOG_DBG("flow sent\n"
 367                  "%s"
 368                  " pre: %s\n"
 369                  "post: %s",
 370                  tnl_str, pre_flow_str, post_flow_str);
 371         free(tnl_str);
 372         free(pre_flow_str);
 373         free(post_flow_str);
 374     }
 375
 376 out:
 377     ovs_rwlock_unlock(&rwlock);
 378     return out_port;
 379 }
 380
 381 static uint32_t
 382 tnl_hash(struct tnl_match *match)
 383 {
 384     BUILD_ASSERT_DECL(sizeof *match % sizeof(uint32_t) == 0);
 385     return hash_words((uint32_t *) match, sizeof *match / sizeof(uint32_t), 0);
 386 }
 387
 388 static struct tnl_port *
 389 tnl_find_ofport(const struct ofport_dpif *ofport) OVS_REQ_RDLOCK(rwlock)
 390 {
 391     struct tnl_port *tnl_port;
 392
 393     HMAP_FOR_EACH_IN_BUCKET (tnl_port, ofport_node, hash_pointer(ofport, 0),
 394                              ofport_map) {
 395         if (tnl_port->ofport == ofport) {
 396             return tnl_port;
 397         }
 398     }
 399     return NULL;
 400 }
 401
 402 static struct tnl_port *
 403 tnl_find_exact(struct tnl_match *match) OVS_REQ_RDLOCK(rwlock)
 404 {
 405     struct tnl_port *tnl_port;
 406
 407     HMAP_FOR_EACH_WITH_HASH (tnl_port, match_node, tnl_hash(match),
 408                              tnl_match_map) {
 409         if (!memcmp(match, &tnl_port->match, sizeof *match)) {
 410             return tnl_port;
 411         }
 412     }
 413     return NULL;
 414 }
 415
 416 static struct tnl_port *
 417 tnl_find(struct tnl_match *match_) OVS_REQ_RDLOCK(rwlock)
 418 {
 419     struct tnl_match match = *match_;
 420     struct tnl_port *tnl_port;
 421
 422     /* remote_ip, local_ip, in_key */
 423     tnl_port = tnl_find_exact(&match);
 424     if (tnl_port) {
 425         return tnl_port;
 426     }
 427
 428     /* remote_ip, in_key */
 429     match.ip_src = 0;
 430     tnl_port = tnl_find_exact(&match);
 431     if (tnl_port) {
 432         return tnl_port;
 433     }
 434     match.ip_src = match_->ip_src;
 435
 436     /* remote_ip, local_ip */
 437     match.in_key = 0;
 438     match.in_key_flow = true;
 439     tnl_port = tnl_find_exact(&match);
 440     if (tnl_port) {
 441         return tnl_port;
 442     }
 443
 444     /* remote_ip */
 445     match.ip_src = 0;
 446     tnl_port = tnl_find_exact(&match);
 447     if (tnl_port) {
 448         return tnl_port;
 449     }
 450
 451     /* Flow-based remote */
 452     match.ip_dst = 0;
 453     match.ip_dst_flow = true;
 454     tnl_port = tnl_find_exact(&match);
 455     if (tnl_port) {
 456         return tnl_port;
 457     }
 458
 459     /* Flow-based everything */
 460     match.ip_src_flow = true;
 461     tnl_port = tnl_find_exact(&match);
 462     if (tnl_port) {
 463         return tnl_port;
 464     }
 465
 466     return NULL;
 467 }
 468
 469 static void
 470 tnl_match_fmt(const struct tnl_match *match, struct ds *ds)
 471     OVS_REQ_RDLOCK(rwlock)
 472 {
 473     if (!match->ip_dst_flow) {
 474         ds_put_format(ds, IP_FMT"->"IP_FMT, IP_ARGS(match->ip_src),
 475                       IP_ARGS(match->ip_dst));
 476     } else if (!match->ip_src_flow) {
 477         ds_put_format(ds, IP_FMT"->flow", IP_ARGS(match->ip_src));
 478     } else {
 479         ds_put_cstr(ds, "flow->flow");
 480     }
 481
 482     if (match->in_key_flow) {
 483         ds_put_cstr(ds, ", key=flow");
 484     } else {
 485         ds_put_format(ds, ", key=%#"PRIx64, ntohll(match->in_key));
 486     }
 487
 488     ds_put_format(ds, ", dp port=%"PRIu32, match->odp_port);
 489     ds_put_format(ds, ", pkt mark=%"PRIu32, match->pkt_mark);
 490 }
 491
 492 static void
 493 tnl_port_mod_log(const struct tnl_port *tnl_port, const char *action)
 494     OVS_REQ_RDLOCK(rwlock)
 495 {
 496     if (VLOG_IS_DBG_ENABLED()) {
 497         struct ds ds = DS_EMPTY_INITIALIZER;
 498
 499         tnl_match_fmt(&tnl_port->match, &ds);
 500         VLOG_INFO("%s tunnel port %s (%s)", action,
 501                   tnl_port_get_name(tnl_port), ds_cstr(&ds));
 502         ds_destroy(&ds);
 503     }
 504 }
 505
 506 static char *
 507 tnl_port_fmt(const struct tnl_port *tnl_port) OVS_REQ_RDLOCK(rwlock)
 508 {
 509     const struct netdev_tunnel_config *cfg =
 510         netdev_get_tunnel_config(tnl_port->netdev);
 511     struct ds ds = DS_EMPTY_INITIALIZER;
 512
 513     ds_put_format(&ds, "port %"PRIu32": %s (%s: ", tnl_port->match.odp_port,
 514                   tnl_port_get_name(tnl_port),
 515                   netdev_get_type(tnl_port->netdev));
 516     tnl_match_fmt(&tnl_port->match, &ds);
 517
 518     if (cfg->out_key != cfg->in_key ||
 519         cfg->out_key_present != cfg->in_key_present ||
 520         cfg->out_key_flow != cfg->in_key_flow) {
 521         ds_put_cstr(&ds, ", out_key=");
 522         if (!cfg->out_key_present) {
 523             ds_put_cstr(&ds, "none");
 524         } else if (cfg->out_key_flow) {
 525             ds_put_cstr(&ds, "flow");
 526         } else {
 527             ds_put_format(&ds, "%#"PRIx64, ntohll(cfg->out_key));
 528         }
 529     }
 530
 531     if (cfg->ttl_inherit) {
 532         ds_put_cstr(&ds, ", ttl=inherit");
 533     } else {
 534         ds_put_format(&ds, ", ttl=%"PRIu8, cfg->ttl);
 535     }
 536
 537     if (cfg->tos_inherit) {
 538         ds_put_cstr(&ds, ", tos=inherit");
 539     } else if (cfg->tos) {
 540         ds_put_format(&ds, ", tos=%#"PRIx8, cfg->tos);
 541     }
 542
 543     if (!cfg->dont_fragment) {
 544         ds_put_cstr(&ds, ", df=false");
 545     }
 546
 547     if (cfg->csum) {
 548         ds_put_cstr(&ds, ", csum=true");
 549     }
 550
 551     ds_put_cstr(&ds, ")\n");
 552
 553     return ds_steal_cstr(&ds);
 554 }
 555
 556 static const char *
 557 tnl_port_get_name(const struct tnl_port *tnl_port) OVS_REQ_RDLOCK(rwlock)
 558 {
 559     return netdev_get_name(tnl_port->netdev);
 560 }