lib/netdev-vport.c

   1 /*
   2  * Copyright (c) 2010, 2011 Nicira Networks.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "netdev-vport.h"
  20
  21 #include <errno.h>
  22 #include <fcntl.h>
  23 #include <sys/socket.h>
  24 #include <linux/rtnetlink.h>
  25 #include <net/if.h>
  26 #include <sys/ioctl.h>
  27
  28 #include "byte-order.h"
  29 #include "daemon.h"
  30 #include "dirs.h"
  31 #include "dpif-linux.h"
  32 #include "hash.h"
  33 #include "hmap.h"
  34 #include "list.h"
  35 #include "netdev-linux.h"
  36 #include "netdev-provider.h"
  37 #include "netlink.h"
  38 #include "netlink-socket.h"
  39 #include "ofpbuf.h"
  40 #include "openvswitch/datapath-protocol.h"
  41 #include "openvswitch/tunnel.h"
  42 #include "packets.h"
  43 #include "route-table.h"
  44 #include "rtnetlink.h"
  45 #include "shash.h"
  46 #include "socket-util.h"
  47 #include "vlog.h"
  48
  49 VLOG_DEFINE_THIS_MODULE(netdev_vport);
  50
  51 struct netdev_vport_notifier {
  52     struct netdev_notifier notifier;
  53     struct list list_node;
  54     struct shash_node *shash_node;
  55 };
  56
  57 struct netdev_dev_vport {
  58     struct netdev_dev netdev_dev;
  59     struct ofpbuf *options;
  60     int dp_ifindex;             /* -1 if unknown. */
  61     uint32_t port_no;           /* UINT32_MAX if unknown. */
  62     unsigned int change_seq;
  63 };
  64
  65 struct netdev_vport {
  66     struct netdev netdev;
  67 };
  68
  69 struct vport_class {
  70     enum odp_vport_type type;
  71     struct netdev_class netdev_class;
  72     int (*parse_config)(const char *name, const char *type,
  73                         const struct shash *args, struct ofpbuf *options);
  74     int (*unparse_config)(const char *name, const char *type,
  75                           const struct nlattr *options, size_t options_len,
  76                           struct shash *args);
  77 };
  78
  79 static struct shash netdev_vport_notifiers =
  80                                     SHASH_INITIALIZER(&netdev_vport_notifiers);
  81
  82 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
  83
  84 static int netdev_vport_create(const struct netdev_class *, const char *,
  85                                const struct shash *, struct netdev_dev **);
  86 static void netdev_vport_poll_notify(const struct netdev *);
  87 static int tnl_port_config_from_nlattr(const struct nlattr *options,
  88                                        size_t options_len,
  89                                        struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1]);
  90
  91 static const char *netdev_vport_get_tnl_iface(const struct netdev *netdev);
  92
  93 static bool
  94 is_vport_class(const struct netdev_class *class)
  95 {
  96     return class->create == netdev_vport_create;
  97 }
  98
  99 static const struct vport_class *
 100 vport_class_cast(const struct netdev_class *class)
 101 {
 102     assert(is_vport_class(class));
 103     return CONTAINER_OF(class, struct vport_class, netdev_class);
 104 }
 105
 106 static struct netdev_dev_vport *
 107 netdev_dev_vport_cast(const struct netdev_dev *netdev_dev)
 108 {
 109     assert(is_vport_class(netdev_dev_get_class(netdev_dev)));
 110     return CONTAINER_OF(netdev_dev, struct netdev_dev_vport, netdev_dev);
 111 }
 112
 113 static struct netdev_vport *
 114 netdev_vport_cast(const struct netdev *netdev)
 115 {
 116     struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
 117     assert(is_vport_class(netdev_dev_get_class(netdev_dev)));
 118     return CONTAINER_OF(netdev, struct netdev_vport, netdev);
 119 }
 120
 121 /* If 'netdev' is a vport netdev, returns an ofpbuf that contains Netlink
 122  * options to include in ODP_VPORT_ATTR_OPTIONS for configuring that vport.
 123  * Otherwise returns NULL. */
 124 const struct ofpbuf *
 125 netdev_vport_get_options(const struct netdev *netdev)
 126 {
 127     const struct netdev_dev *dev = netdev_get_dev(netdev);
 128
 129     return (is_vport_class(netdev_dev_get_class(dev))
 130             ? netdev_dev_vport_cast(dev)->options
 131             : NULL);
 132 }
 133
 134 enum odp_vport_type
 135 netdev_vport_get_vport_type(const struct netdev *netdev)
 136 {
 137     const struct netdev_dev *dev = netdev_get_dev(netdev);
 138     const struct netdev_class *class = netdev_dev_get_class(dev);
 139
 140     return (is_vport_class(class) ? vport_class_cast(class)->type
 141             : class == &netdev_internal_class ? ODP_VPORT_TYPE_INTERNAL
 142             : class == &netdev_linux_class ? ODP_VPORT_TYPE_NETDEV
 143             : ODP_VPORT_TYPE_UNSPEC);
 144 }
 145
 146 const char *
 147 netdev_vport_get_netdev_type(const struct dpif_linux_vport *vport)
 148 {
 149     struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1];
 150
 151     switch (vport->type) {
 152     case ODP_VPORT_TYPE_UNSPEC:
 153         break;
 154
 155     case ODP_VPORT_TYPE_NETDEV:
 156         return "system";
 157
 158     case ODP_VPORT_TYPE_INTERNAL:
 159         return "internal";
 160
 161     case ODP_VPORT_TYPE_PATCH:
 162         return "patch";
 163
 164     case ODP_VPORT_TYPE_GRE:
 165         if (tnl_port_config_from_nlattr(vport->options, vport->options_len,
 166                                         a)) {
 167             break;
 168         }
 169         return (nl_attr_get_u32(a[ODP_TUNNEL_ATTR_FLAGS]) & TNL_F_IPSEC
 170                 ? "ipsec_gre" : "gre");
 171
 172     case ODP_VPORT_TYPE_CAPWAP:
 173         return "capwap";
 174
 175     case __ODP_VPORT_TYPE_MAX:
 176         break;
 177     }
 178
 179     VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u",
 180                  vport->dp_ifindex, vport->name, (unsigned int) vport->type);
 181     return "unknown";
 182 }
 183
 184 static int
 185 netdev_vport_create(const struct netdev_class *netdev_class, const char *name,
 186                     const struct shash *args,
 187                     struct netdev_dev **netdev_devp)
 188 {
 189     const struct vport_class *vport_class = vport_class_cast(netdev_class);
 190     struct ofpbuf *options = NULL;
 191     struct shash fetched_args;
 192     int dp_ifindex;
 193     uint32_t port_no;
 194     int error;
 195
 196     shash_init(&fetched_args);
 197
 198     dp_ifindex = -1;
 199     port_no = UINT32_MAX;
 200     if (!shash_is_empty(args)) {
 201         /* Parse the provided configuration. */
 202         options = ofpbuf_new(64);
 203         error = vport_class->parse_config(name, netdev_class->type,
 204                                           args, options);
 205     } else {
 206         /* Fetch an existing configuration from the kernel.
 207          *
 208          * This case could be ambiguous with initializing a new vport with an
 209          * empty configuration, but none of the existing vport classes accept
 210          * an empty configuration. */
 211         struct dpif_linux_vport reply;
 212         struct ofpbuf *buf;
 213
 214         error = dpif_linux_vport_get(name, &reply, &buf);
 215         if (!error) {
 216             /* XXX verify correct type */
 217             error = vport_class->unparse_config(name, netdev_class->type,
 218                                                 reply.options,
 219                                                 reply.options_len,
 220                                                 &fetched_args);
 221             if (error) {
 222                 VLOG_ERR_RL(&rl, "%s: failed to parse kernel config (%s)",
 223                             name, strerror(error));
 224             } else {
 225                 options = ofpbuf_clone_data(reply.options, reply.options_len);
 226                 dp_ifindex = reply.dp_ifindex;
 227                 port_no = reply.port_no;
 228             }
 229             ofpbuf_delete(buf);
 230         } else {
 231             VLOG_ERR_RL(&rl, "%s: vport query failed (%s)",
 232                         name, strerror(error));
 233         }
 234     }
 235
 236     if (!error) {
 237         struct netdev_dev_vport *dev;
 238
 239         dev = xmalloc(sizeof *dev);
 240         netdev_dev_init(&dev->netdev_dev, name,
 241                         shash_is_empty(&fetched_args) ? args : &fetched_args,
 242                         netdev_class);
 243         dev->options = options;
 244         dev->dp_ifindex = dp_ifindex;
 245         dev->port_no = port_no;
 246         dev->change_seq = 1;
 247
 248         *netdev_devp = &dev->netdev_dev;
 249         route_table_register();
 250     } else {
 251         ofpbuf_delete(options);
 252     }
 253
 254     shash_destroy(&fetched_args);
 255
 256     return error;
 257 }
 258
 259 static void
 260 netdev_vport_destroy(struct netdev_dev *netdev_dev_)
 261 {
 262     struct netdev_dev_vport *netdev_dev = netdev_dev_vport_cast(netdev_dev_);
 263
 264     route_table_unregister();
 265     free(netdev_dev);
 266 }
 267
 268 static int
 269 netdev_vport_open(struct netdev_dev *netdev_dev_, int ethertype OVS_UNUSED,
 270                 struct netdev **netdevp)
 271 {
 272     struct netdev_vport *netdev;
 273
 274     netdev = xmalloc(sizeof *netdev);
 275     netdev_init(&netdev->netdev, netdev_dev_);
 276
 277     *netdevp = &netdev->netdev;
 278     return 0;
 279 }
 280
 281 static void
 282 netdev_vport_close(struct netdev *netdev_)
 283 {
 284     struct netdev_vport *netdev = netdev_vport_cast(netdev_);
 285     free(netdev);
 286 }
 287
 288 static int
 289 netdev_vport_set_config(struct netdev_dev *dev_, const struct shash *args)
 290 {
 291     const struct netdev_class *netdev_class = netdev_dev_get_class(dev_);
 292     const struct vport_class *vport_class = vport_class_cast(netdev_class);
 293     struct netdev_dev_vport *dev = netdev_dev_vport_cast(dev_);
 294     const char *name = netdev_dev_get_name(dev_);
 295     struct ofpbuf *options;
 296     int error;
 297
 298     options = ofpbuf_new(64);
 299     error = vport_class->parse_config(name, netdev_dev_get_type(dev_),
 300                                       args, options);
 301     if (!error
 302         && (options->size != dev->options->size
 303             || memcmp(options->data, dev->options->data, options->size))) {
 304         struct dpif_linux_vport vport;
 305
 306         dpif_linux_vport_init(&vport);
 307         vport.cmd = ODP_VPORT_CMD_SET;
 308         vport.name = name;
 309         vport.options = options->data;
 310         vport.options_len = options->size;
 311         error = dpif_linux_vport_transact(&vport, NULL, NULL);
 312         if (!error || error == ENODEV) {
 313             /* Either reconfiguration succeeded or this vport is not installed
 314              * in the kernel (e.g. it hasn't been added to a dpif yet with
 315              * dpif_port_add()). */
 316             ofpbuf_delete(dev->options);
 317             dev->options = options;
 318             options = NULL;
 319             error = 0;
 320         }
 321     }
 322     ofpbuf_delete(options);
 323
 324     return error;
 325 }
 326
 327 static int
 328 netdev_vport_send(struct netdev *netdev, const void *data, size_t size)
 329 {
 330     struct netdev_dev *dev_ = netdev_get_dev(netdev);
 331     struct netdev_dev_vport *dev = netdev_dev_vport_cast(dev_);
 332
 333     if (dev->dp_ifindex == -1) {
 334         const char *name = netdev_get_name(netdev);
 335         struct dpif_linux_vport reply;
 336         struct ofpbuf *buf;
 337         int error;
 338
 339         error = dpif_linux_vport_get(name, &reply, &buf);
 340         if (error) {
 341             VLOG_ERR_RL(&rl, "%s: failed to query vport for send (%s)",
 342                         name, strerror(error));
 343             return error;
 344         }
 345         dev->dp_ifindex = reply.dp_ifindex;
 346         dev->port_no = reply.port_no;
 347         ofpbuf_delete(buf);
 348     }
 349
 350     return dpif_linux_vport_send(dev->dp_ifindex, dev->port_no, data, size);
 351 }
 352
 353 static int
 354 netdev_vport_set_etheraddr(struct netdev *netdev,
 355                            const uint8_t mac[ETH_ADDR_LEN])
 356 {
 357     struct dpif_linux_vport vport;
 358     int error;
 359
 360     dpif_linux_vport_init(&vport);
 361     vport.cmd = ODP_VPORT_CMD_SET;
 362     vport.name = netdev_get_name(netdev);
 363     vport.address = mac;
 364
 365     error = dpif_linux_vport_transact(&vport, NULL, NULL);
 366     if (!error) {
 367         netdev_vport_poll_notify(netdev);
 368     }
 369     return error;
 370 }
 371
 372 static int
 373 netdev_vport_get_etheraddr(const struct netdev *netdev,
 374                            uint8_t mac[ETH_ADDR_LEN])
 375 {
 376     struct dpif_linux_vport reply;
 377     struct ofpbuf *buf;
 378     int error;
 379
 380     error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf);
 381     if (!error) {
 382         if (reply.address) {
 383             memcpy(mac, reply.address, ETH_ADDR_LEN);
 384         } else {
 385             error = EOPNOTSUPP;
 386         }
 387         ofpbuf_delete(buf);
 388     }
 389     return error;
 390 }
 391
 392 static int
 393 netdev_vport_get_mtu(const struct netdev *netdev, int *mtup)
 394 {
 395     struct dpif_linux_vport reply;
 396     struct ofpbuf *buf;
 397     int error;
 398
 399     error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf);
 400     if (!error) {
 401         *mtup = reply.mtu;
 402         ofpbuf_delete(buf);
 403     }
 404     return error;
 405 }
 406
 407 int
 408 netdev_vport_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
 409 {
 410     struct dpif_linux_vport reply;
 411     struct ofpbuf *buf;
 412     int error;
 413
 414     error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf);
 415     if (error) {
 416         return error;
 417     } else if (!reply.stats) {
 418         ofpbuf_delete(buf);
 419         return EOPNOTSUPP;
 420     }
 421
 422     netdev_stats_from_rtnl_link_stats64(stats, reply.stats);
 423
 424     ofpbuf_delete(buf);
 425
 426     return 0;
 427 }
 428
 429 int
 430 netdev_vport_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
 431 {
 432     struct rtnl_link_stats64 rtnl_stats;
 433     struct dpif_linux_vport vport;
 434     int err;
 435
 436     netdev_stats_to_rtnl_link_stats64(&rtnl_stats, stats);
 437
 438     dpif_linux_vport_init(&vport);
 439     vport.cmd = ODP_VPORT_CMD_SET;
 440     vport.name = netdev_get_name(netdev);
 441     vport.stats = &rtnl_stats;
 442
 443     err = dpif_linux_vport_transact(&vport, NULL, NULL);
 444
 445     /* If the vport layer doesn't know about the device, that doesn't mean it
 446      * doesn't exist (after all were able to open it when netdev_open() was
 447      * called), it just means that it isn't attached and we'll be getting
 448      * stats a different way. */
 449     if (err == ENODEV) {
 450         err = EOPNOTSUPP;
 451     }
 452
 453     return err;
 454 }
 455
 456 static int
 457 netdev_vport_get_status(const struct netdev *netdev, struct shash *sh)
 458 {
 459     const char *iface = netdev_vport_get_tnl_iface(netdev);
 460
 461     if (iface) {
 462         struct netdev *egress_netdev;
 463
 464         shash_add(sh, "tunnel_egress_iface", xstrdup(iface));
 465
 466         if (!netdev_open_default(iface, &egress_netdev)) {
 467             shash_add(sh, "tunnel_egress_iface_carrier",
 468                       xstrdup(netdev_get_carrier(egress_netdev)
 469                               ? "up" : "down"));
 470             netdev_close(egress_netdev);
 471         }
 472     }
 473
 474     return 0;
 475 }
 476
 477 static int
 478 netdev_vport_update_flags(struct netdev *netdev OVS_UNUSED,
 479                         enum netdev_flags off, enum netdev_flags on OVS_UNUSED,
 480                         enum netdev_flags *old_flagsp)
 481 {
 482     if (off & (NETDEV_UP | NETDEV_PROMISC)) {
 483         return EOPNOTSUPP;
 484     }
 485
 486     *old_flagsp = NETDEV_UP | NETDEV_PROMISC;
 487     return 0;
 488 }
 489
 490 static char *
 491 make_poll_name(const struct netdev *netdev)
 492 {
 493     return xasprintf("%s:%s", netdev_get_type(netdev), netdev_get_name(netdev));
 494 }
 495
 496 static int
 497 netdev_vport_poll_add(struct netdev *netdev,
 498                       void (*cb)(struct netdev_notifier *), void *aux,
 499                       struct netdev_notifier **notifierp)
 500 {
 501     char *poll_name = make_poll_name(netdev);
 502     struct netdev_vport_notifier *notifier;
 503     struct list *list;
 504     struct shash_node *shash_node;
 505
 506     shash_node = shash_find(&netdev_vport_notifiers, poll_name);
 507     if (!shash_node) {
 508         list = xmalloc(sizeof *list);
 509         list_init(list);
 510         shash_node = shash_add(&netdev_vport_notifiers, poll_name, list);
 511     } else {
 512         list = shash_node->data;
 513     }
 514
 515     notifier = xmalloc(sizeof *notifier);
 516     netdev_notifier_init(&notifier->notifier, netdev, cb, aux);
 517     list_push_back(list, &notifier->list_node);
 518     notifier->shash_node = shash_node;
 519
 520     *notifierp = &notifier->notifier;
 521     free(poll_name);
 522
 523     return 0;
 524 }
 525
 526 static void
 527 netdev_vport_poll_remove(struct netdev_notifier *notifier_)
 528 {
 529     struct netdev_vport_notifier *notifier =
 530                 CONTAINER_OF(notifier_, struct netdev_vport_notifier, notifier);
 531
 532     struct list *list;
 533
 534     list = list_remove(&notifier->list_node);
 535     if (list_is_empty(list)) {
 536         shash_delete(&netdev_vport_notifiers, notifier->shash_node);
 537         free(list);
 538     }
 539
 540     free(notifier);
 541 }
 542
 543 static unsigned int
 544 netdev_vport_change_seq(const struct netdev *netdev)
 545 {
 546     return netdev_dev_vport_cast(netdev_get_dev(netdev))->change_seq;
 547 }
 548
 549 static void
 550 netdev_vport_run(void)
 551 {
 552     route_table_run();
 553 }
 554
 555 static void
 556 netdev_vport_wait(void)
 557 {
 558     route_table_wait();
 559 }
 560 \f
 561 /* get_tnl_iface() implementation. */
 562 static const char *
 563 netdev_vport_get_tnl_iface(const struct netdev *netdev)
 564 {
 565     struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1];
 566     ovs_be32 route;
 567     struct netdev_dev_vport *ndv;
 568     static char name[IFNAMSIZ];
 569
 570     ndv = netdev_dev_vport_cast(netdev_get_dev(netdev));
 571     if (tnl_port_config_from_nlattr(ndv->options->data, ndv->options->size,
 572                                     a)) {
 573         return NULL;
 574     }
 575     route = nl_attr_get_be32(a[ODP_TUNNEL_ATTR_DST_IPV4]);
 576
 577     if (route_table_get_name(route, name)) {
 578         return name;
 579     }
 580
 581     return NULL;
 582 }
 583 \f
 584 /* Helper functions. */
 585
 586 static void
 587 netdev_vport_poll_notify(const struct netdev *netdev)
 588 {
 589     char *poll_name = make_poll_name(netdev);
 590     struct list *list = shash_find_data(&netdev_vport_notifiers,
 591                                         poll_name);
 592     struct netdev_dev_vport *ndv;
 593
 594     ndv = netdev_dev_vport_cast(netdev_get_dev(netdev));
 595
 596     if (list) {
 597         struct netdev_vport_notifier *notifier;
 598
 599         LIST_FOR_EACH (notifier, list_node, list) {
 600             struct netdev_notifier *n = &notifier->notifier;
 601             n->cb(n);
 602         }
 603     }
 604
 605     ndv->change_seq++;
 606     if (!ndv->change_seq) {
 607         ndv->change_seq++;
 608     }
 609
 610     free(poll_name);
 611 }
 612 \f
 613 /* Code specific to individual vport types. */
 614
 615 static void
 616 set_key(const struct shash *args, const char *name, uint16_t type,
 617         struct ofpbuf *options)
 618 {
 619     const char *s;
 620
 621     s = shash_find_data(args, name);
 622     if (!s) {
 623         s = shash_find_data(args, "key");
 624         if (!s) {
 625             s = "0";
 626         }
 627     }
 628
 629     if (!strcmp(s, "flow")) {
 630         /* This is the default if no attribute is present. */
 631     } else {
 632         nl_msg_put_be64(options, type, htonll(strtoull(s, NULL, 0)));
 633     }
 634 }
 635
 636 static int
 637 parse_tunnel_config(const char *name, const char *type,
 638                     const struct shash *args, struct ofpbuf *options)
 639 {
 640     bool is_gre = false;
 641     bool is_ipsec = false;
 642     struct shash_node *node;
 643     bool ipsec_mech_set = false;
 644     ovs_be32 daddr = htonl(0);
 645     uint32_t flags;
 646
 647     flags = TNL_F_DF_DEFAULT | TNL_F_PMTUD | TNL_F_HDR_CACHE;
 648     if (!strcmp(type, "gre")) {
 649         is_gre = true;
 650     } else if (!strcmp(type, "ipsec_gre")) {
 651         is_gre = true;
 652         is_ipsec = true;
 653         flags |= TNL_F_IPSEC;
 654         flags &= ~TNL_F_HDR_CACHE;
 655     }
 656
 657     SHASH_FOR_EACH (node, args) {
 658         if (!strcmp(node->name, "remote_ip")) {
 659             struct in_addr in_addr;
 660             if (lookup_ip(node->data, &in_addr)) {
 661                 VLOG_WARN("%s: bad %s 'remote_ip'", name, type);
 662             } else {
 663                 daddr = in_addr.s_addr;
 664             }
 665         } else if (!strcmp(node->name, "local_ip")) {
 666             struct in_addr in_addr;
 667             if (lookup_ip(node->data, &in_addr)) {
 668                 VLOG_WARN("%s: bad %s 'local_ip'", name, type);
 669             } else {
 670                 nl_msg_put_be32(options, ODP_TUNNEL_ATTR_SRC_IPV4,
 671                                 in_addr.s_addr);
 672             }
 673         } else if (!strcmp(node->name, "tos")) {
 674             if (!strcmp(node->data, "inherit")) {
 675                 flags |= TNL_F_TOS_INHERIT;
 676             } else {
 677                 nl_msg_put_u8(options, ODP_TUNNEL_ATTR_TOS, atoi(node->data));
 678             }
 679         } else if (!strcmp(node->name, "ttl")) {
 680             if (!strcmp(node->data, "inherit")) {
 681                 flags |= TNL_F_TTL_INHERIT;
 682             } else {
 683                 nl_msg_put_u8(options, ODP_TUNNEL_ATTR_TTL, atoi(node->data));
 684             }
 685         } else if (!strcmp(node->name, "csum") && is_gre) {
 686             if (!strcmp(node->data, "true")) {
 687                 flags |= TNL_F_CSUM;
 688             }
 689         } else if (!strcmp(node->name, "df_inherit")) {
 690             if (!strcmp(node->data, "true")) {
 691                 flags |= TNL_F_DF_INHERIT;
 692             }
 693         } else if (!strcmp(node->name, "df_default")) {
 694             if (!strcmp(node->data, "false")) {
 695                 flags &= ~TNL_F_DF_DEFAULT;
 696             }
 697         } else if (!strcmp(node->name, "pmtud")) {
 698             if (!strcmp(node->data, "false")) {
 699                 flags &= ~TNL_F_PMTUD;
 700             }
 701         } else if (!strcmp(node->name, "header_cache")) {
 702             if (!strcmp(node->data, "false")) {
 703                 flags &= ~TNL_F_HDR_CACHE;
 704             }
 705         } else if (!strcmp(node->name, "peer_cert") && is_ipsec) {
 706             if (shash_find(args, "certificate")) {
 707                 ipsec_mech_set = true;
 708             } else {
 709                 const char *use_ssl_cert;
 710
 711                 /* If the "use_ssl_cert" is true, then "certificate" and
 712                  * "private_key" will be pulled from the SSL table.  The
 713                  * use of this option is strongly discouraged, since it
 714                  * will like be removed when multiple SSL configurations
 715                  * are supported by OVS.
 716                  */
 717                 use_ssl_cert = shash_find_data(args, "use_ssl_cert");
 718                 if (!use_ssl_cert || strcmp(use_ssl_cert, "true")) {
 719                     VLOG_ERR("%s: 'peer_cert' requires 'certificate' argument",
 720                              name);
 721                     return EINVAL;
 722                 }
 723                 ipsec_mech_set = true;
 724             }
 725         } else if (!strcmp(node->name, "psk") && is_ipsec) {
 726             ipsec_mech_set = true;
 727         } else if (is_ipsec
 728                 && (!strcmp(node->name, "certificate")
 729                     || !strcmp(node->name, "private_key")
 730                     || !strcmp(node->name, "use_ssl_cert"))) {
 731             /* Ignore options not used by the netdev. */
 732         } else if (is_gre && (!strcmp(node->name, "key") ||
 733                               !strcmp(node->name, "in_key") ||
 734                               !strcmp(node->name, "out_key"))) {
 735             /* Handled separately below. */
 736         } else {
 737             VLOG_WARN("%s: unknown %s argument '%s'", name, type, node->name);
 738         }
 739     }
 740
 741     if (is_ipsec) {
 742         char *file_name = xasprintf("%s/%s", ovs_rundir(),
 743                 "ovs-monitor-ipsec.pid");
 744         pid_t pid = read_pidfile(file_name);
 745         free(file_name);
 746         if (pid < 0) {
 747             VLOG_ERR("%s: IPsec requires the ovs-monitor-ipsec daemon",
 748                      name);
 749             return EINVAL;
 750         }
 751
 752         if (shash_find(args, "peer_cert") && shash_find(args, "psk")) {
 753             VLOG_ERR("%s: cannot define both 'peer_cert' and 'psk'", name);
 754             return EINVAL;
 755         }
 756
 757         if (!ipsec_mech_set) {
 758             VLOG_ERR("%s: IPsec requires an 'peer_cert' or psk' argument",
 759                      name);
 760             return EINVAL;
 761         }
 762     }
 763
 764     if (is_gre) {
 765         set_key(args, "in_key", ODP_TUNNEL_ATTR_IN_KEY, options);
 766         set_key(args, "out_key", ODP_TUNNEL_ATTR_OUT_KEY, options);
 767     }
 768
 769     if (!daddr) {
 770         VLOG_ERR("%s: %s type requires valid 'remote_ip' argument",
 771                  name, type);
 772         return EINVAL;
 773     }
 774     nl_msg_put_be32(options, ODP_TUNNEL_ATTR_DST_IPV4, daddr);
 775
 776     nl_msg_put_u32(options, ODP_TUNNEL_ATTR_FLAGS, flags);
 777
 778     return 0;
 779 }
 780
 781 static int
 782 tnl_port_config_from_nlattr(const struct nlattr *options, size_t options_len,
 783                             struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1])
 784 {
 785     static const struct nl_policy odp_tunnel_policy[] = {
 786         [ODP_TUNNEL_ATTR_FLAGS] = { .type = NL_A_U32 },
 787         [ODP_TUNNEL_ATTR_DST_IPV4] = { .type = NL_A_BE32 },
 788         [ODP_TUNNEL_ATTR_SRC_IPV4] = { .type = NL_A_BE32, .optional = true },
 789         [ODP_TUNNEL_ATTR_IN_KEY] = { .type = NL_A_BE64, .optional = true },
 790         [ODP_TUNNEL_ATTR_OUT_KEY] = { .type = NL_A_BE64, .optional = true },
 791         [ODP_TUNNEL_ATTR_TOS] = { .type = NL_A_U8, .optional = true },
 792         [ODP_TUNNEL_ATTR_TTL] = { .type = NL_A_U8, .optional = true },
 793     };
 794     struct ofpbuf buf;
 795
 796     ofpbuf_use_const(&buf, options, options_len);
 797     if (!nl_policy_parse(&buf, 0, odp_tunnel_policy,
 798                          a, ARRAY_SIZE(odp_tunnel_policy))) {
 799         return EINVAL;
 800     }
 801     return 0;
 802 }
 803
 804 static uint64_t
 805 get_be64_or_zero(const struct nlattr *a)
 806 {
 807     return a ? ntohll(nl_attr_get_be64(a)) : 0;
 808 }
 809
 810 static int
 811 unparse_tunnel_config(const char *name OVS_UNUSED, const char *type OVS_UNUSED,
 812                       const struct nlattr *options, size_t options_len,
 813                       struct shash *args)
 814 {
 815     struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1];
 816     ovs_be32 daddr;
 817     uint32_t flags;
 818     int error;
 819
 820     error = tnl_port_config_from_nlattr(options, options_len, a);
 821     if (error) {
 822         return error;
 823     }
 824
 825     flags = nl_attr_get_u32(a[ODP_TUNNEL_ATTR_FLAGS]);
 826     if (!(flags & TNL_F_HDR_CACHE) == !(flags & TNL_F_IPSEC)) {
 827         smap_add(args, "header_cache",
 828                  flags & TNL_F_HDR_CACHE ? "true" : "false");
 829     }
 830
 831     daddr = nl_attr_get_be32(a[ODP_TUNNEL_ATTR_DST_IPV4]);
 832     shash_add(args, "remote_ip", xasprintf(IP_FMT, IP_ARGS(&daddr)));
 833
 834     if (a[ODP_TUNNEL_ATTR_SRC_IPV4]) {
 835         ovs_be32 saddr = nl_attr_get_be32(a[ODP_TUNNEL_ATTR_SRC_IPV4]);
 836         shash_add(args, "local_ip", xasprintf(IP_FMT, IP_ARGS(&saddr)));
 837     }
 838
 839     if (!a[ODP_TUNNEL_ATTR_IN_KEY] && !a[ODP_TUNNEL_ATTR_OUT_KEY]) {
 840         smap_add(args, "key", "flow");
 841     } else {
 842         uint64_t in_key = get_be64_or_zero(a[ODP_TUNNEL_ATTR_IN_KEY]);
 843         uint64_t out_key = get_be64_or_zero(a[ODP_TUNNEL_ATTR_OUT_KEY]);
 844
 845         if (in_key && in_key == out_key) {
 846             shash_add(args, "key", xasprintf("%"PRIu64, in_key));
 847         } else {
 848             if (!a[ODP_TUNNEL_ATTR_IN_KEY]) {
 849                 smap_add(args, "in_key", "flow");
 850             } else if (in_key) {
 851                 shash_add(args, "in_key", xasprintf("%"PRIu64, in_key));
 852             }
 853
 854             if (!a[ODP_TUNNEL_ATTR_OUT_KEY]) {
 855                 smap_add(args, "out_key", "flow");
 856             } else if (out_key) {
 857                 shash_add(args, "out_key", xasprintf("%"PRIu64, out_key));
 858             }
 859         }
 860     }
 861
 862     if (flags & TNL_F_TTL_INHERIT) {
 863         smap_add(args, "tos", "inherit");
 864     } else if (a[ODP_TUNNEL_ATTR_TTL]) {
 865         int ttl = nl_attr_get_u8(a[ODP_TUNNEL_ATTR_TTL]);
 866         shash_add(args, "tos", xasprintf("%d", ttl));
 867     }
 868
 869     if (flags & TNL_F_TOS_INHERIT) {
 870         smap_add(args, "tos", "inherit");
 871     } else if (a[ODP_TUNNEL_ATTR_TOS]) {
 872         int tos = nl_attr_get_u8(a[ODP_TUNNEL_ATTR_TOS]);
 873         shash_add(args, "tos", xasprintf("%d", tos));
 874     }
 875
 876     if (flags & TNL_F_CSUM) {
 877         smap_add(args, "csum", "true");
 878     }
 879     if (flags & TNL_F_DF_INHERIT) {
 880         smap_add(args, "df_inherit", "true");
 881     }
 882     if (!(flags & TNL_F_DF_DEFAULT)) {
 883         smap_add(args, "df_default", "false");
 884     }
 885     if (!(flags & TNL_F_PMTUD)) {
 886         smap_add(args, "pmtud", "false");
 887     }
 888
 889     return 0;
 890 }
 891
 892 static int
 893 parse_patch_config(const char *name, const char *type OVS_UNUSED,
 894                    const struct shash *args, struct ofpbuf *options)
 895 {
 896     const char *peer;
 897
 898     peer = shash_find_data(args, "peer");
 899     if (!peer) {
 900         VLOG_ERR("%s: patch type requires valid 'peer' argument", name);
 901         return EINVAL;
 902     }
 903
 904     if (shash_count(args) > 1) {
 905         VLOG_ERR("%s: patch type takes only a 'peer' argument", name);
 906         return EINVAL;
 907     }
 908
 909     if (strlen(peer) >= IFNAMSIZ) {
 910         VLOG_ERR("%s: patch 'peer' arg too long", name);
 911         return EINVAL;
 912     }
 913
 914     if (!strcmp(name, peer)) {
 915         VLOG_ERR("%s: patch peer must not be self", name);
 916         return EINVAL;
 917     }
 918
 919     nl_msg_put_string(options, ODP_PATCH_ATTR_PEER, peer);
 920
 921     return 0;
 922 }
 923
 924 static int
 925 unparse_patch_config(const char *name OVS_UNUSED, const char *type OVS_UNUSED,
 926                      const struct nlattr *options, size_t options_len,
 927                      struct shash *args)
 928 {
 929     static const struct nl_policy odp_patch_policy[] = {
 930         [ODP_PATCH_ATTR_PEER] = { .type = NL_A_STRING,
 931                                .max_len = IFNAMSIZ,
 932                                .optional = false }
 933     };
 934
 935     struct nlattr *a[ARRAY_SIZE(odp_patch_policy)];
 936     struct ofpbuf buf;
 937
 938     ofpbuf_use_const(&buf, options, options_len);
 939     if (!nl_policy_parse(&buf, 0, odp_patch_policy,
 940                          a, ARRAY_SIZE(odp_patch_policy))) {
 941         return EINVAL;
 942     }
 943
 944     smap_add(args, "peer", nl_attr_get_string(a[ODP_PATCH_ATTR_PEER]));
 945     return 0;
 946 }
 947 \f
 948 #define VPORT_FUNCTIONS(GET_STATUS)                         \
 949     NULL,                                                   \
 950     netdev_vport_run,                                       \
 951     netdev_vport_wait,                                      \
 952                                                             \
 953     netdev_vport_create,                                    \
 954     netdev_vport_destroy,                                   \
 955     netdev_vport_set_config,                                \
 956                                                             \
 957     netdev_vport_open,                                      \
 958     netdev_vport_close,                                     \
 959                                                             \
 960     NULL,                       /* enumerate */             \
 961                                                             \
 962     NULL,                       /* recv */                  \
 963     NULL,                       /* recv_wait */             \
 964     NULL,                       /* drain */                 \
 965                                                             \
 966     netdev_vport_send,          /* send */                  \
 967     NULL,                       /* send_wait */             \
 968                                                             \
 969     netdev_vport_set_etheraddr,                             \
 970     netdev_vport_get_etheraddr,                             \
 971     netdev_vport_get_mtu,                                   \
 972     NULL,                       /* get_ifindex */           \
 973     NULL,                       /* get_carrier */           \
 974     NULL,                       /* get_miimon */            \
 975     netdev_vport_get_stats,                                 \
 976     netdev_vport_set_stats,                                 \
 977                                                             \
 978     NULL,                       /* get_features */          \
 979     NULL,                       /* set_advertisements */    \
 980     NULL,                       /* get_vlan_vid */          \
 981                                                             \
 982     NULL,                       /* set_policing */          \
 983     NULL,                       /* get_qos_types */         \
 984     NULL,                       /* get_qos_capabilities */  \
 985     NULL,                       /* get_qos */               \
 986     NULL,                       /* set_qos */               \
 987     NULL,                       /* get_queue */             \
 988     NULL,                       /* set_queue */             \
 989     NULL,                       /* delete_queue */          \
 990     NULL,                       /* get_queue_stats */       \
 991     NULL,                       /* dump_queues */           \
 992     NULL,                       /* dump_queue_stats */      \
 993                                                             \
 994     NULL,                       /* get_in4 */               \
 995     NULL,                       /* set_in4 */               \
 996     NULL,                       /* get_in6 */               \
 997     NULL,                       /* add_router */            \
 998     NULL,                       /* get_next_hop */          \
 999     GET_STATUS,                                             \
1000     NULL,                       /* arp_lookup */            \
1001                                                             \
1002     netdev_vport_update_flags,                              \
1003                                                             \
1004     netdev_vport_poll_add,                                  \
1005     netdev_vport_poll_remove,                               \
1006     netdev_vport_change_seq
1007
1008 void
1009 netdev_vport_register(void)
1010 {
1011     static const struct vport_class vport_classes[] = {
1012         { ODP_VPORT_TYPE_GRE,
1013           { "gre", VPORT_FUNCTIONS(netdev_vport_get_status) },
1014           parse_tunnel_config, unparse_tunnel_config },
1015
1016         { ODP_VPORT_TYPE_GRE,
1017           { "ipsec_gre", VPORT_FUNCTIONS(netdev_vport_get_status) },
1018           parse_tunnel_config, unparse_tunnel_config },
1019
1020         { ODP_VPORT_TYPE_CAPWAP,
1021           { "capwap", VPORT_FUNCTIONS(netdev_vport_get_status) },
1022           parse_tunnel_config, unparse_tunnel_config },
1023
1024         { ODP_VPORT_TYPE_PATCH,
1025           { "patch", VPORT_FUNCTIONS(NULL) },
1026           parse_patch_config, unparse_patch_config }
1027     };
1028
1029     int i;
1030
1031     for (i = 0; i < ARRAY_SIZE(vport_classes); i++) {
1032         netdev_register_provider(&vport_classes[i].netdev_class);
1033     }
1034 }