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 "ofproto/ofproto-provider.h"
  21 #include "byte-order.h"
  22 #include "dynamic-string.h"
  23 #include "hash.h"
  24 #include "hmap.h"
  25 #include "netdev-vport.h"
  26 #include "odp-util.h"
  27 #include "packets.h"
  28 #include "smap.h"
  29 #include "socket-util.h"
  30 #include "tunnel.h"
  31 #include "vlog.h"
  32
  33 /* XXX:
  34  *
  35  * Ability to generate actions on input for ECN
  36  * Ability to generate metadata for packet-outs
  37  * IPsec using skb mark.
  38  * VXLAN.
  39  * Multicast group management (possibly).
  40  * Disallow netdevs with names like "gre64_system" to prevent collisions. */
  41
  42 VLOG_DEFINE_THIS_MODULE(tunnel);
  43
  44 struct tnl_match {
  45     ovs_be64 in_key;
  46     ovs_be32 ip_src;
  47     ovs_be32 ip_dst;
  48     uint32_t odp_port;
  49     bool in_key_flow;
  50 };
  51
  52 struct tnl_port {
  53     struct hmap_node match_node;
  54
  55     const struct ofport *ofport;
  56     unsigned int netdev_seq;
  57     struct tnl_match match;
  58 };
  59
  60 static struct hmap tnl_match_map = HMAP_INITIALIZER(&tnl_match_map);
  61
  62 /* Returned to callers when their ofport will never be used to receive or send
  63  * tunnel traffic. Alternatively, we could ask the caller to delete their
  64  * ofport, but this would be unclean in the reconfguration case.  For the first
  65  * time, an ofproto provider would have to call ofproto_port_del() on itself.*/
  66 static struct tnl_port void_tnl_port;
  67
  68 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
  69 static struct vlog_rate_limit dbg_rl = VLOG_RATE_LIMIT_INIT(60, 60);
  70
  71 static struct tnl_port *tnl_find(struct tnl_match *);
  72 static struct tnl_port *tnl_find_exact(struct tnl_match *);
  73 static uint32_t tnl_hash(struct tnl_match *);
  74 static void tnl_match_fmt(const struct tnl_match *, struct ds *);
  75 static char *tnl_port_fmt(const struct tnl_port *);
  76 static void tnl_port_mod_log(const struct tnl_port *, const char *action);
  77 static const char *tnl_port_get_name(const struct tnl_port *);
  78
  79 static struct tnl_port *
  80 tnl_port_add__(const struct ofport *ofport, uint32_t odp_port,
  81                bool warn)
  82 {
  83     const struct netdev_tunnel_config *cfg;
  84     struct tnl_port *existing_port;
  85     struct tnl_port *tnl_port;
  86
  87     cfg = netdev_get_tunnel_config(ofport->netdev);
  88     ovs_assert(cfg);
  89
  90     tnl_port = xzalloc(sizeof *tnl_port);
  91     tnl_port->ofport = ofport;
  92     tnl_port->netdev_seq = netdev_change_seq(tnl_port->ofport->netdev);
  93
  94     tnl_port->match.in_key = cfg->in_key;
  95     tnl_port->match.ip_src = cfg->ip_src;
  96     tnl_port->match.ip_dst = cfg->ip_dst;
  97     tnl_port->match.in_key_flow = cfg->in_key_flow;
  98     tnl_port->match.odp_port = odp_port;
  99
 100     existing_port = tnl_find_exact(&tnl_port->match);
 101     if (existing_port) {
 102         if (warn) {
 103             struct ds ds = DS_EMPTY_INITIALIZER;
 104             tnl_match_fmt(&tnl_port->match, &ds);
 105             VLOG_WARN("%s: attempting to add tunnel port with same config as "
 106                       "port '%s' (%s)", tnl_port_get_name(tnl_port),
 107                       tnl_port_get_name(existing_port), ds_cstr(&ds));
 108             ds_destroy(&ds);
 109             free(tnl_port);
 110         }
 111         return &void_tnl_port;
 112     }
 113
 114     hmap_insert(&tnl_match_map, &tnl_port->match_node,
 115                 tnl_hash(&tnl_port->match));
 116     tnl_port_mod_log(tnl_port, "adding");
 117     return tnl_port;
 118 }
 119
 120 /* Adds 'ofport' to the module with datapath port number 'odp_port'. 'ofport's
 121  * must be added before they can be used by the module. 'ofport' must be a
 122  * tunnel. */
 123 struct tnl_port *
 124 tnl_port_add(const struct ofport *ofport, uint32_t odp_port)
 125 {
 126     return tnl_port_add__(ofport, odp_port, true);
 127 }
 128
 129 /* Checks if the tnl_port pointed to by 'tnl_portp' needs reconfiguration due
 130  * to changes in its netdev_tunnel_config.  If it does, updates 'tnl_portp' to
 131  * point to a new tnl_port and returns true.  Otherwise, returns false.
 132  * 'ofport' and 'odp_port' should be the same as would be passed to
 133  * tnl_port_add(). */
 134 bool
 135 tnl_port_reconfigure(const struct ofport *ofport, uint32_t odp_port,
 136                      struct tnl_port **tnl_portp)
 137 {
 138     struct tnl_port *tnl_port = *tnl_portp;
 139
 140     if (tnl_port == &void_tnl_port) {
 141         *tnl_portp = tnl_port_add__(ofport, odp_port, false);
 142         return *tnl_portp != &void_tnl_port;
 143     } else if (tnl_port->ofport != ofport
 144                || tnl_port->match.odp_port != odp_port
 145                || tnl_port->netdev_seq != netdev_change_seq(ofport->netdev)) {
 146         VLOG_DBG("reconfiguring %s", tnl_port_get_name(tnl_port));
 147         tnl_port_del(tnl_port);
 148         *tnl_portp = tnl_port_add(ofport, odp_port);
 149         return true;
 150     }
 151     return false;
 152 }
 153
 154 /* Removes 'tnl_port' from the module. */
 155 void
 156 tnl_port_del(struct tnl_port *tnl_port)
 157 {
 158     if (tnl_port && tnl_port != &void_tnl_port) {
 159         tnl_port_mod_log(tnl_port, "removing");
 160         hmap_remove(&tnl_match_map, &tnl_port->match_node);
 161         free(tnl_port);
 162     }
 163 }
 164
 165 /* Transforms 'flow' so that it appears to have been received by a tunnel
 166  * OpenFlow port controlled by this module instead of the datapath port it
 167  * actually came in on.  Sets 'flow''s in_port to the appropriate OpenFlow port
 168  * number.  Returns the 'ofport' corresponding to the new in_port.
 169  *
 170  * Callers should verify that 'flow' needs to be received by calling
 171  * tnl_port_should_receive() before this function.
 172  *
 173  * Leaves 'flow' untouched and returns null if unsuccessful. */
 174 const struct ofport *
 175 tnl_port_receive(struct flow *flow)
 176 {
 177     char *pre_flow_str = NULL;
 178     struct tnl_port *tnl_port;
 179     struct tnl_match match;
 180
 181     memset(&match, 0, sizeof match);
 182     match.odp_port = flow->in_port;
 183     match.ip_src = flow->tunnel.ip_dst;
 184     match.ip_dst = flow->tunnel.ip_src;
 185     match.in_key = flow->tunnel.tun_id;
 186
 187     tnl_port = tnl_find(&match);
 188     if (!tnl_port) {
 189         struct ds ds = DS_EMPTY_INITIALIZER;
 190
 191         tnl_match_fmt(&match, &ds);
 192         VLOG_WARN_RL(&rl, "receive tunnel port not found (%s)", ds_cstr(&ds));
 193         ds_destroy(&ds);
 194         return NULL;
 195     }
 196
 197     if (is_ip_any(flow)
 198         && ((flow->tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE)
 199         && (flow->nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
 200         VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE but is not ECN"
 201                      " capable");
 202         return NULL;
 203     }
 204
 205     if (!VLOG_DROP_DBG(&dbg_rl)) {
 206         pre_flow_str = flow_to_string(flow);
 207     }
 208
 209     flow->in_port = tnl_port->ofport->ofp_port;
 210     memset(&flow->tunnel, 0, sizeof flow->tunnel);
 211     flow->tunnel.tun_id = match.in_key;
 212
 213     if (pre_flow_str) {
 214         char *post_flow_str = flow_to_string(flow);
 215         char *tnl_str = tnl_port_fmt(tnl_port);
 216         VLOG_DBG("flow received\n"
 217                  "%s"
 218                  " pre: %s\n"
 219                  "post: %s",
 220                  tnl_str, pre_flow_str, post_flow_str);
 221         free(tnl_str);
 222         free(pre_flow_str);
 223         free(post_flow_str);
 224     }
 225     return tnl_port->ofport;
 226 }
 227
 228 /* Given that 'flow' should be output to the ofport corresponding to
 229  * 'tnl_port', updates 'flow''s tunnel headers and returns the actual datapath
 230  * port that the output should happen on.  May return OVSP_NONE if the output
 231  * shouldn't occur. */
 232 uint32_t
 233 tnl_port_send(const struct tnl_port *tnl_port, struct flow *flow)
 234 {
 235     const struct netdev_tunnel_config *cfg;
 236     char *pre_flow_str = NULL;
 237
 238     if (tnl_port == &void_tnl_port) {
 239         return OVSP_NONE;
 240     }
 241
 242     cfg = netdev_get_tunnel_config(tnl_port->ofport->netdev);
 243     ovs_assert(cfg);
 244
 245     if (!VLOG_DROP_DBG(&dbg_rl)) {
 246         pre_flow_str = flow_to_string(flow);
 247     }
 248
 249     flow->tunnel.ip_src = tnl_port->match.ip_src;
 250     flow->tunnel.ip_dst = tnl_port->match.ip_dst;
 251
 252     if (!cfg->out_key_flow) {
 253         flow->tunnel.tun_id = cfg->out_key;
 254     }
 255
 256     if (cfg->ttl_inherit && is_ip_any(flow)) {
 257         flow->tunnel.ip_ttl = flow->nw_ttl;
 258     } else {
 259         flow->tunnel.ip_ttl = cfg->ttl;
 260     }
 261
 262     if (cfg->tos_inherit && is_ip_any(flow)) {
 263         flow->tunnel.ip_tos = flow->nw_tos & IP_DSCP_MASK;
 264     } else {
 265         flow->tunnel.ip_tos = cfg->tos;
 266     }
 267
 268     if ((flow->nw_tos & IP_ECN_MASK) == IP_ECN_CE) {
 269         flow->tunnel.ip_tos |= IP_ECN_ECT_0;
 270     } else {
 271         flow->tunnel.ip_tos |= flow->nw_tos & IP_ECN_MASK;
 272     }
 273
 274     flow->tunnel.flags = (cfg->dont_fragment ? FLOW_TNL_F_DONT_FRAGMENT : 0)
 275         | (cfg->csum ? FLOW_TNL_F_CSUM : 0)
 276         | (cfg->out_key_present ? FLOW_TNL_F_KEY : 0);
 277
 278     if (pre_flow_str) {
 279         char *post_flow_str = flow_to_string(flow);
 280         char *tnl_str = tnl_port_fmt(tnl_port);
 281         VLOG_DBG("flow sent\n"
 282                  "%s"
 283                  " pre: %s\n"
 284                  "post: %s",
 285                  tnl_str, pre_flow_str, post_flow_str);
 286         free(tnl_str);
 287         free(pre_flow_str);
 288         free(post_flow_str);
 289     }
 290
 291     return tnl_port->match.odp_port;
 292 }
 293
 294 static uint32_t
 295 tnl_hash(struct tnl_match *match)
 296 {
 297     BUILD_ASSERT_DECL(sizeof *match % sizeof(uint32_t) == 0);
 298     return hash_words((uint32_t *) match, sizeof *match / sizeof(uint32_t), 0);
 299 }
 300
 301 static struct tnl_port *
 302 tnl_find_exact(struct tnl_match *match)
 303 {
 304     struct tnl_port *tnl_port;
 305
 306     HMAP_FOR_EACH_WITH_HASH (tnl_port, match_node, tnl_hash(match),
 307                              &tnl_match_map) {
 308         if (!memcmp(match, &tnl_port->match, sizeof *match)) {
 309             return tnl_port;
 310         }
 311     }
 312     return NULL;
 313 }
 314
 315 static struct tnl_port *
 316 tnl_find(struct tnl_match *match_)
 317 {
 318     struct tnl_match match = *match_;
 319     bool is_multicast = ip_is_multicast(match.ip_src);
 320     struct tnl_port *tnl_port;
 321
 322     /* remote_ip, local_ip, in_key */
 323     if (!is_multicast) {
 324         tnl_port = tnl_find_exact(&match);
 325         if (tnl_port) {
 326             return tnl_port;
 327         }
 328     }
 329
 330     /* remote_ip, in_key */
 331     match.ip_src = 0;
 332     tnl_port = tnl_find_exact(&match);
 333     if (tnl_port) {
 334         return tnl_port;
 335     }
 336     match.ip_src = match_->ip_src;
 337
 338     /* remote_ip, local_ip */
 339     if (!is_multicast) {
 340         match.in_key = 0;
 341         match.in_key_flow = true;
 342         tnl_port = tnl_find_exact(&match);
 343         if (tnl_port) {
 344             return tnl_port;
 345         }
 346         match.in_key = match_->in_key;
 347         match.in_key_flow = false;
 348     }
 349
 350     /* remote_ip */
 351     match.ip_src = 0;
 352     match.in_key = 0;
 353     match.in_key_flow = true;
 354     tnl_port = tnl_find_exact(&match);
 355     if (tnl_port) {
 356         return tnl_port;
 357     }
 358     match.ip_src = match_->ip_src;
 359     match.in_key = match_->in_key;
 360     match.in_key_flow = false;
 361
 362     if (is_multicast) {
 363         match.ip_src = 0;
 364         match.ip_dst = match_->ip_src;
 365
 366         /* multicast remote_ip, in_key */
 367         tnl_port = tnl_find_exact(&match);
 368         if (tnl_port) {
 369             return tnl_port;
 370         }
 371
 372         /* multicast remote_ip */
 373         match.in_key = 0;
 374         match.in_key_flow = true;
 375         tnl_port = tnl_find_exact(&match);
 376         if (tnl_port) {
 377             return tnl_port;
 378         }
 379     }
 380     return NULL;
 381 }
 382
 383 static void
 384 tnl_match_fmt(const struct tnl_match *match, struct ds *ds)
 385 {
 386     ds_put_format(ds, IP_FMT"->"IP_FMT, IP_ARGS(match->ip_src),
 387                   IP_ARGS(match->ip_dst));
 388
 389     if (match->in_key_flow) {
 390         ds_put_cstr(ds, ", key=flow");
 391     } else {
 392         ds_put_format(ds, ", key=%#"PRIx64, ntohll(match->in_key));
 393     }
 394
 395     ds_put_format(ds, ", dp port=%"PRIu32, match->odp_port);
 396 }
 397
 398 static void
 399 tnl_port_mod_log(const struct tnl_port *tnl_port, const char *action)
 400 {
 401     if (VLOG_IS_DBG_ENABLED()) {
 402         struct ds ds = DS_EMPTY_INITIALIZER;
 403
 404         tnl_match_fmt(&tnl_port->match, &ds);
 405         VLOG_INFO("%s tunnel port %s (%s)", action,
 406                   tnl_port_get_name(tnl_port), ds_cstr(&ds));
 407         ds_destroy(&ds);
 408     }
 409 }
 410
 411 static char *
 412 tnl_port_fmt(const struct tnl_port *tnl_port)
 413 {
 414     const struct netdev_tunnel_config *cfg =
 415         netdev_get_tunnel_config(tnl_port->ofport->netdev);
 416     struct ds ds = DS_EMPTY_INITIALIZER;
 417
 418     ds_put_format(&ds, "port %"PRIu32": %s (%s: ", tnl_port->match.odp_port,
 419                   tnl_port_get_name(tnl_port),
 420                   netdev_get_type(tnl_port->ofport->netdev));
 421     tnl_match_fmt(&tnl_port->match, &ds);
 422
 423     if (cfg->out_key != cfg->in_key ||
 424         cfg->out_key_present != cfg->in_key_present ||
 425         cfg->out_key_flow != cfg->in_key_flow) {
 426         ds_put_cstr(&ds, ", out_key=");
 427         if (!cfg->out_key_present) {
 428             ds_put_cstr(&ds, "none");
 429         } else if (cfg->out_key_flow) {
 430             ds_put_cstr(&ds, "flow");
 431         } else {
 432             ds_put_format(&ds, "%#"PRIx64, ntohll(cfg->out_key));
 433         }
 434     }
 435
 436     if (cfg->ttl_inherit) {
 437         ds_put_cstr(&ds, ", ttl=inherit");
 438     } else {
 439         ds_put_format(&ds, ", ttl=%"PRIu8, cfg->ttl);
 440     }
 441
 442     if (cfg->tos_inherit) {
 443         ds_put_cstr(&ds, ", tos=inherit");
 444     } else if (cfg->tos) {
 445         ds_put_format(&ds, ", tos=%#"PRIx8, cfg->tos);
 446     }
 447
 448     if (!cfg->dont_fragment) {
 449         ds_put_cstr(&ds, ", df=false");
 450     }
 451
 452     if (cfg->csum) {
 453         ds_put_cstr(&ds, ", csum=true");
 454     }
 455
 456     ds_put_cstr(&ds, ")\n");
 457
 458     return ds_steal_cstr(&ds);
 459 }
 460
 461 static const char *
 462 tnl_port_get_name(const struct tnl_port *tnl_port)
 463 {
 464     return netdev_get_name(tnl_port->ofport->netdev);
 465 }