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