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