lib/dpif-linux.c

   1 /*
   2  * Copyright (c) 2008, 2009, 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 "dpif-linux.h"
  20
  21 #include <assert.h>
  22 #include <ctype.h>
  23 #include <errno.h>
  24 #include <fcntl.h>
  25 #include <inttypes.h>
  26 #include <net/if.h>
  27 #include <linux/types.h>
  28 #include <linux/pkt_sched.h>
  29 #include <linux/rtnetlink.h>
  30 #include <linux/sockios.h>
  31 #include <stdlib.h>
  32 #include <sys/stat.h>
  33 #include <unistd.h>
  34
  35 #include "bitmap.h"
  36 #include "dpif-provider.h"
  37 #include "netdev.h"
  38 #include "netdev-vport.h"
  39 #include "netlink-socket.h"
  40 #include "netlink.h"
  41 #include "odp-util.h"
  42 #include "ofpbuf.h"
  43 #include "openvswitch/tunnel.h"
  44 #include "packets.h"
  45 #include "poll-loop.h"
  46 #include "rtnetlink.h"
  47 #include "rtnetlink-link.h"
  48 #include "shash.h"
  49 #include "sset.h"
  50 #include "unaligned.h"
  51 #include "util.h"
  52 #include "vlog.h"
  53
  54 VLOG_DEFINE_THIS_MODULE(dpif_linux);
  55
  56 enum { LRU_MAX_PORTS = 1024 };
  57 enum { LRU_MASK = LRU_MAX_PORTS - 1};
  58 BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
  59
  60 struct dpif_linux_dp {
  61     /* Generic Netlink header. */
  62     uint8_t cmd;
  63
  64     /* struct odp_header. */
  65     int dp_ifindex;
  66
  67     /* Attributes. */
  68     const char *name;                  /* ODP_DP_ATTR_NAME. */
  69     struct odp_stats stats;            /* ODP_DP_ATTR_STATS. */
  70     enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */
  71     const uint32_t *sampling;          /* ODP_DP_ATTR_SAMPLING. */
  72     uint32_t mcgroups[DPIF_N_UC_TYPES]; /* ODP_DP_ATTR_MCGROUPS. */
  73 };
  74
  75 static void dpif_linux_dp_init(struct dpif_linux_dp *);
  76 static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
  77                                      const struct ofpbuf *);
  78 static void dpif_linux_dp_dump_start(struct nl_dump *);
  79 static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
  80                                   struct dpif_linux_dp *reply,
  81                                   struct ofpbuf **bufp);
  82 static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
  83                              struct ofpbuf **bufp);
  84
  85 struct dpif_linux_flow {
  86     /* Generic Netlink header. */
  87     uint8_t cmd;
  88
  89     /* struct odp_header. */
  90     unsigned int nlmsg_flags;
  91     int dp_ifindex;
  92
  93     /* Attributes.
  94      *
  95      * The 'stats' and 'used' members point to 64-bit data that might only be
  96      * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to
  97      * access their values.
  98      *
  99      * If 'actions' is nonnull then ODP_FLOW_ATTR_ACTIONS will be included in
 100      * the Netlink version of the command, even if actions_len is zero. */
 101     const struct nlattr *key;           /* ODP_FLOW_ATTR_KEY. */
 102     size_t key_len;
 103     const struct nlattr *actions;       /* ODP_FLOW_ATTR_ACTIONS. */
 104     size_t actions_len;
 105     const struct odp_flow_stats *stats; /* ODP_FLOW_ATTR_STATS. */
 106     const uint8_t *tcp_flags;           /* ODP_FLOW_ATTR_TCP_FLAGS. */
 107     const uint64_t *used;               /* ODP_FLOW_ATTR_USED. */
 108     bool clear;                         /* ODP_FLOW_ATTR_CLEAR. */
 109 };
 110
 111 static void dpif_linux_flow_init(struct dpif_linux_flow *);
 112 static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *,
 113                                        const struct ofpbuf *);
 114 static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
 115                                       struct ofpbuf *);
 116 static int dpif_linux_flow_transact(const struct dpif_linux_flow *request,
 117                                     struct dpif_linux_flow *reply,
 118                                     struct ofpbuf **bufp);
 119 static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
 120                                       struct dpif_flow_stats *);
 121
 122 /* Datapath interface for the openvswitch Linux kernel module. */
 123 struct dpif_linux {
 124     struct dpif dpif;
 125     int dp_ifindex;
 126
 127     /* Multicast group messages. */
 128     struct nl_sock *mc_sock;
 129     uint32_t mcgroups[DPIF_N_UC_TYPES];
 130     unsigned int listen_mask;
 131
 132     /* Change notification. */
 133     struct sset changed_ports;  /* Ports that have changed. */
 134     struct rtnetlink_notifier port_notifier;
 135     bool change_error;
 136
 137     /* Queue of unused ports. */
 138     unsigned long *lru_bitmap;
 139     uint16_t lru_ports[LRU_MAX_PORTS];
 140     size_t lru_head;
 141     size_t lru_tail;
 142 };
 143
 144 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
 145
 146 /* Generic Netlink family numbers for ODP. */
 147 static int odp_datapath_family;
 148 static int odp_vport_family;
 149 static int odp_flow_family;
 150 static int odp_packet_family;
 151
 152 /* Generic Netlink socket. */
 153 static struct nl_sock *genl_sock;
 154
 155 static int dpif_linux_init(void);
 156 static int open_dpif(const struct dpif_linux_dp *, struct dpif **);
 157 static void dpif_linux_port_changed(const struct rtnetlink_link_change *,
 158                                     void *dpif);
 159
 160 static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
 161                                        struct ofpbuf *);
 162 static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *,
 163                                         const struct ofpbuf *);
 164
 165 static struct dpif_linux *
 166 dpif_linux_cast(const struct dpif *dpif)
 167 {
 168     dpif_assert_class(dpif, &dpif_linux_class);
 169     return CONTAINER_OF(dpif, struct dpif_linux, dpif);
 170 }
 171
 172 static void
 173 dpif_linux_push_port(struct dpif_linux *dp, uint16_t port)
 174 {
 175     if (port < LRU_MAX_PORTS && !bitmap_is_set(dp->lru_bitmap, port)) {
 176         bitmap_set1(dp->lru_bitmap, port);
 177         dp->lru_ports[dp->lru_head++ & LRU_MASK] = port;
 178     }
 179 }
 180
 181 static uint32_t
 182 dpif_linux_pop_port(struct dpif_linux *dp)
 183 {
 184     uint16_t port;
 185
 186     if (dp->lru_head == dp->lru_tail) {
 187         return UINT32_MAX;
 188     }
 189
 190     port = dp->lru_ports[dp->lru_tail++ & LRU_MASK];
 191     bitmap_set0(dp->lru_bitmap, port);
 192     return port;
 193 }
 194
 195 static int
 196 dpif_linux_enumerate(struct sset *all_dps)
 197 {
 198     struct nl_dump dump;
 199     struct ofpbuf msg;
 200     int error;
 201
 202     error = dpif_linux_init();
 203     if (error) {
 204         return error;
 205     }
 206
 207     dpif_linux_dp_dump_start(&dump);
 208     while (nl_dump_next(&dump, &msg)) {
 209         struct dpif_linux_dp dp;
 210
 211         if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
 212             sset_add(all_dps, dp.name);
 213         }
 214     }
 215     return nl_dump_done(&dump);
 216 }
 217
 218 static int
 219 dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
 220                 bool create, struct dpif **dpifp)
 221 {
 222     struct dpif_linux_dp dp_request, dp;
 223     struct ofpbuf *buf;
 224     int error;
 225
 226     error = dpif_linux_init();
 227     if (error) {
 228         return error;
 229     }
 230
 231     /* Create or look up datapath. */
 232     dpif_linux_dp_init(&dp_request);
 233     dp_request.cmd = create ? ODP_DP_CMD_NEW : ODP_DP_CMD_GET;
 234     dp_request.name = name;
 235     error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
 236     if (error) {
 237         return error;
 238     }
 239     error = open_dpif(&dp, dpifp);
 240     ofpbuf_delete(buf);
 241
 242     return error;
 243 }
 244
 245 static int
 246 open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
 247 {
 248     struct dpif_linux *dpif;
 249     int error;
 250     int i;
 251
 252     dpif = xmalloc(sizeof *dpif);
 253     error = rtnetlink_link_notifier_register(&dpif->port_notifier,
 254                                              dpif_linux_port_changed, dpif);
 255     if (error) {
 256         goto error_free;
 257     }
 258
 259     dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
 260               dp->dp_ifindex, dp->dp_ifindex);
 261
 262     dpif->mc_sock = NULL;
 263     for (i = 0; i < DPIF_N_UC_TYPES; i++) {
 264         dpif->mcgroups[i] = dp->mcgroups[i];
 265     }
 266     dpif->listen_mask = 0;
 267     dpif->dp_ifindex = dp->dp_ifindex;
 268     sset_init(&dpif->changed_ports);
 269     dpif->change_error = false;
 270     *dpifp = &dpif->dpif;
 271
 272     dpif->lru_head = dpif->lru_tail = 0;
 273     dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS);
 274     bitmap_set1(dpif->lru_bitmap, ODPP_LOCAL);
 275     for (i = 1; i < LRU_MAX_PORTS; i++) {
 276         dpif_linux_push_port(dpif, i);
 277     }
 278     return 0;
 279
 280 error_free:
 281     free(dpif);
 282     return error;
 283 }
 284
 285 static void
 286 dpif_linux_close(struct dpif *dpif_)
 287 {
 288     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 289     rtnetlink_link_notifier_unregister(&dpif->port_notifier);
 290     sset_destroy(&dpif->changed_ports);
 291     free(dpif->lru_bitmap);
 292     free(dpif);
 293 }
 294
 295 static int
 296 dpif_linux_destroy(struct dpif *dpif_)
 297 {
 298     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 299     struct dpif_linux_dp dp;
 300
 301     dpif_linux_dp_init(&dp);
 302     dp.cmd = ODP_DP_CMD_DEL;
 303     dp.dp_ifindex = dpif->dp_ifindex;
 304     return dpif_linux_dp_transact(&dp, NULL, NULL);
 305 }
 306
 307 static int
 308 dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats)
 309 {
 310     struct dpif_linux_dp dp;
 311     struct ofpbuf *buf;
 312     int error;
 313
 314     error = dpif_linux_dp_get(dpif_, &dp, &buf);
 315     if (!error) {
 316         *stats = dp.stats;
 317         ofpbuf_delete(buf);
 318     }
 319     return error;
 320 }
 321
 322 static int
 323 dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
 324 {
 325     struct dpif_linux_dp dp;
 326     struct ofpbuf *buf;
 327     int error;
 328
 329     error = dpif_linux_dp_get(dpif_, &dp, &buf);
 330     if (!error) {
 331         *drop_fragsp = dp.ipv4_frags == ODP_DP_FRAG_DROP;
 332         ofpbuf_delete(buf);
 333     }
 334     return error;
 335 }
 336
 337 static int
 338 dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
 339 {
 340     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 341     struct dpif_linux_dp dp;
 342
 343     dpif_linux_dp_init(&dp);
 344     dp.cmd = ODP_DP_CMD_SET;
 345     dp.dp_ifindex = dpif->dp_ifindex;
 346     dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO;
 347     return dpif_linux_dp_transact(&dp, NULL, NULL);
 348 }
 349
 350 static int
 351 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
 352                     uint16_t *port_nop)
 353 {
 354     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 355     const char *name = netdev_get_name(netdev);
 356     const char *type = netdev_get_type(netdev);
 357     struct dpif_linux_vport request, reply;
 358     const struct ofpbuf *options;
 359     struct ofpbuf *buf;
 360     int error;
 361
 362     dpif_linux_vport_init(&request);
 363     request.cmd = ODP_VPORT_CMD_NEW;
 364     request.dp_ifindex = dpif->dp_ifindex;
 365     request.type = netdev_vport_get_vport_type(netdev);
 366     if (request.type == ODP_VPORT_TYPE_UNSPEC) {
 367         VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
 368                      "unsupported type `%s'",
 369                      dpif_name(dpif_), name, type);
 370         return EINVAL;
 371     }
 372     request.name = name;
 373
 374     options = netdev_vport_get_options(netdev);
 375     if (options && options->size) {
 376         request.options = options->data;
 377         request.options_len = options->size;
 378     }
 379
 380     /* Loop until we find a port that isn't used. */
 381     do {
 382         request.port_no = dpif_linux_pop_port(dpif);
 383         error = dpif_linux_vport_transact(&request, &reply, &buf);
 384
 385         if (!error) {
 386             *port_nop = reply.port_no;
 387         }
 388         ofpbuf_delete(buf);
 389     } while (request.port_no != UINT32_MAX
 390              && (error == EBUSY || error == EFBIG));
 391
 392     return error;
 393 }
 394
 395 static int
 396 dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
 397 {
 398     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 399     struct dpif_linux_vport vport;
 400     int error;
 401
 402     dpif_linux_vport_init(&vport);
 403     vport.cmd = ODP_VPORT_CMD_DEL;
 404     vport.dp_ifindex = dpif->dp_ifindex;
 405     vport.port_no = port_no;
 406     error = dpif_linux_vport_transact(&vport, NULL, NULL);
 407
 408     if (!error) {
 409         dpif_linux_push_port(dpif, port_no);
 410     }
 411     return error;
 412 }
 413
 414 static int
 415 dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
 416                         const char *port_name, struct dpif_port *dpif_port)
 417 {
 418     struct dpif_linux_vport request;
 419     struct dpif_linux_vport reply;
 420     struct ofpbuf *buf;
 421     int error;
 422
 423     dpif_linux_vport_init(&request);
 424     request.cmd = ODP_VPORT_CMD_GET;
 425     request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
 426     request.port_no = port_no;
 427     request.name = port_name;
 428
 429     error = dpif_linux_vport_transact(&request, &reply, &buf);
 430     if (!error) {
 431         dpif_port->name = xstrdup(reply.name);
 432         dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
 433         dpif_port->port_no = reply.port_no;
 434         ofpbuf_delete(buf);
 435     }
 436     return error;
 437 }
 438
 439 static int
 440 dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
 441                                 struct dpif_port *dpif_port)
 442 {
 443     return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
 444 }
 445
 446 static int
 447 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
 448                               struct dpif_port *dpif_port)
 449 {
 450     return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
 451 }
 452
 453 static int
 454 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
 455 {
 456     /* If the datapath increases its range of supported ports, then it should
 457      * start reporting that. */
 458     return 1024;
 459 }
 460
 461 static int
 462 dpif_linux_flow_flush(struct dpif *dpif_)
 463 {
 464     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 465     struct dpif_linux_flow flow;
 466
 467     dpif_linux_flow_init(&flow);
 468     flow.cmd = ODP_FLOW_CMD_DEL;
 469     flow.dp_ifindex = dpif->dp_ifindex;
 470     return dpif_linux_flow_transact(&flow, NULL, NULL);
 471 }
 472
 473 struct dpif_linux_port_state {
 474     struct nl_dump dump;
 475 };
 476
 477 static int
 478 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
 479 {
 480     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 481     struct dpif_linux_port_state *state;
 482     struct dpif_linux_vport request;
 483     struct ofpbuf *buf;
 484
 485     *statep = state = xmalloc(sizeof *state);
 486
 487     dpif_linux_vport_init(&request);
 488     request.cmd = ODP_DP_CMD_GET;
 489     request.dp_ifindex = dpif->dp_ifindex;
 490
 491     buf = ofpbuf_new(1024);
 492     dpif_linux_vport_to_ofpbuf(&request, buf);
 493     nl_dump_start(&state->dump, genl_sock, buf);
 494     ofpbuf_delete(buf);
 495
 496     return 0;
 497 }
 498
 499 static int
 500 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
 501                           struct dpif_port *dpif_port)
 502 {
 503     struct dpif_linux_port_state *state = state_;
 504     struct dpif_linux_vport vport;
 505     struct ofpbuf buf;
 506     int error;
 507
 508     if (!nl_dump_next(&state->dump, &buf)) {
 509         return EOF;
 510     }
 511
 512     error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
 513     if (error) {
 514         return error;
 515     }
 516
 517     dpif_port->name = (char *) vport.name;
 518     dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
 519     dpif_port->port_no = vport.port_no;
 520     return 0;
 521 }
 522
 523 static int
 524 dpif_linux_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
 525 {
 526     struct dpif_linux_port_state *state = state_;
 527     int error = nl_dump_done(&state->dump);
 528     free(state);
 529     return error;
 530 }
 531
 532 static int
 533 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
 534 {
 535     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 536
 537     if (dpif->change_error) {
 538         dpif->change_error = false;
 539         sset_clear(&dpif->changed_ports);
 540         return ENOBUFS;
 541     } else if (!sset_is_empty(&dpif->changed_ports)) {
 542         *devnamep = sset_pop(&dpif->changed_ports);
 543         return 0;
 544     } else {
 545         return EAGAIN;
 546     }
 547 }
 548
 549 static void
 550 dpif_linux_port_poll_wait(const struct dpif *dpif_)
 551 {
 552     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 553     if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
 554         poll_immediate_wake();
 555     } else {
 556         rtnetlink_link_notifier_wait();
 557     }
 558 }
 559
 560 static int
 561 dpif_linux_flow_get__(const struct dpif *dpif_,
 562                       const struct nlattr *key, size_t key_len,
 563                       struct dpif_linux_flow *reply, struct ofpbuf **bufp)
 564 {
 565     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 566     struct dpif_linux_flow request;
 567
 568     dpif_linux_flow_init(&request);
 569     request.cmd = ODP_FLOW_CMD_GET;
 570     request.dp_ifindex = dpif->dp_ifindex;
 571     request.key = key;
 572     request.key_len = key_len;
 573     return dpif_linux_flow_transact(&request, reply, bufp);
 574 }
 575
 576 static int
 577 dpif_linux_flow_get(const struct dpif *dpif_,
 578                     const struct nlattr *key, size_t key_len,
 579                     struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
 580 {
 581     struct dpif_linux_flow reply;
 582     struct ofpbuf *buf;
 583     int error;
 584
 585     error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
 586     if (!error) {
 587         if (stats) {
 588             dpif_linux_flow_get_stats(&reply, stats);
 589         }
 590         if (actionsp) {
 591             buf->data = (void *) reply.actions;
 592             buf->size = reply.actions_len;
 593             *actionsp = buf;
 594         } else {
 595             ofpbuf_delete(buf);
 596         }
 597     }
 598     return error;
 599 }
 600
 601 static int
 602 dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags,
 603                     const struct nlattr *key, size_t key_len,
 604                     const struct nlattr *actions, size_t actions_len,
 605                     struct dpif_flow_stats *stats)
 606 {
 607     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 608     struct dpif_linux_flow request, reply;
 609     struct nlattr dummy_action;
 610     struct ofpbuf *buf;
 611     int error;
 612
 613     dpif_linux_flow_init(&request);
 614     request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET;
 615     request.dp_ifindex = dpif->dp_ifindex;
 616     request.key = key;
 617     request.key_len = key_len;
 618     /* Ensure that ODP_FLOW_ATTR_ACTIONS will always be included. */
 619     request.actions = actions ? actions : &dummy_action;
 620     request.actions_len = actions_len;
 621     if (flags & DPIF_FP_ZERO_STATS) {
 622         request.clear = true;
 623     }
 624     request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
 625     error = dpif_linux_flow_transact(&request,
 626                                      stats ? &reply : NULL,
 627                                      stats ? &buf : NULL);
 628     if (!error && stats) {
 629         dpif_linux_flow_get_stats(&reply, stats);
 630         ofpbuf_delete(buf);
 631     }
 632     return error;
 633 }
 634
 635 static int
 636 dpif_linux_flow_del(struct dpif *dpif_,
 637                     const struct nlattr *key, size_t key_len,
 638                     struct dpif_flow_stats *stats)
 639 {
 640     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 641     struct dpif_linux_flow request, reply;
 642     struct ofpbuf *buf;
 643     int error;
 644
 645     dpif_linux_flow_init(&request);
 646     request.cmd = ODP_FLOW_CMD_DEL;
 647     request.dp_ifindex = dpif->dp_ifindex;
 648     request.key = key;
 649     request.key_len = key_len;
 650     error = dpif_linux_flow_transact(&request,
 651                                      stats ? &reply : NULL,
 652                                      stats ? &buf : NULL);
 653     if (!error && stats) {
 654         dpif_linux_flow_get_stats(&reply, stats);
 655         ofpbuf_delete(buf);
 656     }
 657     return error;
 658 }
 659
 660 struct dpif_linux_flow_state {
 661     struct nl_dump dump;
 662     struct dpif_linux_flow flow;
 663     struct dpif_flow_stats stats;
 664     struct ofpbuf *buf;
 665 };
 666
 667 static int
 668 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
 669 {
 670     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 671     struct dpif_linux_flow_state *state;
 672     struct dpif_linux_flow request;
 673     struct ofpbuf *buf;
 674
 675     *statep = state = xmalloc(sizeof *state);
 676
 677     dpif_linux_flow_init(&request);
 678     request.cmd = ODP_DP_CMD_GET;
 679     request.dp_ifindex = dpif->dp_ifindex;
 680
 681     buf = ofpbuf_new(1024);
 682     dpif_linux_flow_to_ofpbuf(&request, buf);
 683     nl_dump_start(&state->dump, genl_sock, buf);
 684     ofpbuf_delete(buf);
 685
 686     state->buf = NULL;
 687
 688     return 0;
 689 }
 690
 691 static int
 692 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
 693                           const struct nlattr **key, size_t *key_len,
 694                           const struct nlattr **actions, size_t *actions_len,
 695                           const struct dpif_flow_stats **stats)
 696 {
 697     struct dpif_linux_flow_state *state = state_;
 698     struct ofpbuf buf;
 699     int error;
 700
 701     do {
 702         ofpbuf_delete(state->buf);
 703         state->buf = NULL;
 704
 705         if (!nl_dump_next(&state->dump, &buf)) {
 706             return EOF;
 707         }
 708
 709         error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
 710         if (error) {
 711             return error;
 712         }
 713
 714         if (actions && !state->flow.actions) {
 715             error = dpif_linux_flow_get__(dpif_, state->flow.key,
 716                                           state->flow.key_len,
 717                                           &state->flow, &state->buf);
 718             if (error == ENOENT) {
 719                 VLOG_DBG("dumped flow disappeared on get");
 720             } else if (error) {
 721                 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
 722             }
 723         }
 724     } while (error);
 725
 726     if (actions) {
 727         *actions = state->flow.actions;
 728         *actions_len = state->flow.actions_len;
 729     }
 730     if (key) {
 731         *key = state->flow.key;
 732         *key_len = state->flow.key_len;
 733     }
 734     if (stats) {
 735         dpif_linux_flow_get_stats(&state->flow, &state->stats);
 736         *stats = &state->stats;
 737     }
 738     return error;
 739 }
 740
 741 static int
 742 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
 743 {
 744     struct dpif_linux_flow_state *state = state_;
 745     int error = nl_dump_done(&state->dump);
 746     ofpbuf_delete(state->buf);
 747     free(state);
 748     return error;
 749 }
 750
 751 static int
 752 dpif_linux_execute(struct dpif *dpif_,
 753                    const struct nlattr *actions, size_t actions_len,
 754                    const struct ofpbuf *packet)
 755 {
 756     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 757     struct odp_header *execute;
 758     struct ofpbuf *buf;
 759     int error;
 760
 761     buf = ofpbuf_new(128 + actions_len + packet->size);
 762
 763     nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
 764                           ODP_PACKET_CMD_EXECUTE, 1);
 765
 766     execute = ofpbuf_put_uninit(buf, sizeof *execute);
 767     execute->dp_ifindex = dpif->dp_ifindex;
 768
 769     nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size);
 770     nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len);
 771
 772     error = nl_sock_transact(genl_sock, buf, NULL);
 773     ofpbuf_delete(buf);
 774     return error;
 775 }
 776
 777 static int
 778 dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
 779 {
 780     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 781     *listen_mask = dpif->listen_mask;
 782     return 0;
 783 }
 784
 785 static int
 786 dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
 787 {
 788     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 789     int error;
 790     int i;
 791
 792     if (listen_mask == dpif->listen_mask) {
 793         return 0;
 794     } else if (!listen_mask) {
 795         nl_sock_destroy(dpif->mc_sock);
 796         dpif->mc_sock = NULL;
 797         dpif->listen_mask = 0;
 798         return 0;
 799     } else if (!dpif->mc_sock) {
 800         error = nl_sock_create(NETLINK_GENERIC, &dpif->mc_sock);
 801         if (error) {
 802             return error;
 803         }
 804     }
 805
 806     /* Unsubscribe from old groups. */
 807     for (i = 0; i < DPIF_N_UC_TYPES; i++) {
 808         if (dpif->listen_mask & (1u << i)) {
 809             nl_sock_leave_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
 810         }
 811     }
 812
 813     /* Update listen_mask. */
 814     dpif->listen_mask = listen_mask;
 815
 816     /* Subscribe to new groups. */
 817     error = 0;
 818     for (i = 0; i < DPIF_N_UC_TYPES; i++) {
 819         if (dpif->listen_mask & (1u << i)) {
 820             int retval;
 821
 822             retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
 823             if (retval) {
 824                 error = retval;
 825             }
 826         }
 827     }
 828     return error;
 829 }
 830
 831 static int
 832 dpif_linux_get_sflow_probability(const struct dpif *dpif_,
 833                                  uint32_t *probability)
 834 {
 835     struct dpif_linux_dp dp;
 836     struct ofpbuf *buf;
 837     int error;
 838
 839     error = dpif_linux_dp_get(dpif_, &dp, &buf);
 840     if (!error) {
 841         *probability = dp.sampling ? *dp.sampling : 0;
 842         ofpbuf_delete(buf);
 843     }
 844     return error;
 845 }
 846
 847 static int
 848 dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
 849 {
 850     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 851     struct dpif_linux_dp dp;
 852
 853     dpif_linux_dp_init(&dp);
 854     dp.cmd = ODP_DP_CMD_SET;
 855     dp.dp_ifindex = dpif->dp_ifindex;
 856     dp.sampling = &probability;
 857     return dpif_linux_dp_transact(&dp, NULL, NULL);
 858 }
 859
 860 static int
 861 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
 862                              uint32_t queue_id, uint32_t *priority)
 863 {
 864     if (queue_id < 0xf000) {
 865         *priority = TC_H_MAKE(1 << 16, queue_id + 1);
 866         return 0;
 867     } else {
 868         return EINVAL;
 869     }
 870 }
 871
 872 static int
 873 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
 874                  int *dp_ifindex)
 875 {
 876     static const struct nl_policy odp_packet_policy[] = {
 877         /* Always present. */
 878         [ODP_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
 879                                      .min_len = ETH_HEADER_LEN },
 880         [ODP_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
 881
 882         /* ODP_PACKET_CMD_ACTION only. */
 883         [ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
 884
 885         /* ODP_PACKET_CMD_SAMPLE only. */
 886         [ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
 887         [ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
 888     };
 889
 890     struct odp_header *odp_header;
 891     struct nlattr *a[ARRAY_SIZE(odp_packet_policy)];
 892     struct nlmsghdr *nlmsg;
 893     struct genlmsghdr *genl;
 894     struct ofpbuf b;
 895     int type;
 896
 897     ofpbuf_use_const(&b, buf->data, buf->size);
 898
 899     nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
 900     genl = ofpbuf_try_pull(&b, sizeof *genl);
 901     odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
 902     if (!nlmsg || !genl || !odp_header
 903         || nlmsg->nlmsg_type != odp_packet_family
 904         || !nl_policy_parse(&b, 0, odp_packet_policy, a,
 905                             ARRAY_SIZE(odp_packet_policy))) {
 906         return EINVAL;
 907     }
 908
 909     type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS
 910             : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION
 911             : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE
 912             : -1);
 913     if (type < 0) {
 914         return EINVAL;
 915     }
 916
 917     memset(upcall, 0, sizeof *upcall);
 918     upcall->type = type;
 919     upcall->packet = buf;
 920     upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]);
 921     upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]);
 922     upcall->key = (void *) nl_attr_get(a[ODP_PACKET_ATTR_KEY]);
 923     upcall->key_len = nl_attr_get_size(a[ODP_PACKET_ATTR_KEY]);
 924     upcall->userdata = (a[ODP_PACKET_ATTR_USERDATA]
 925                         ? nl_attr_get_u64(a[ODP_PACKET_ATTR_USERDATA])
 926                         : 0);
 927     upcall->sample_pool = (a[ODP_PACKET_ATTR_SAMPLE_POOL]
 928                         ? nl_attr_get_u32(a[ODP_PACKET_ATTR_SAMPLE_POOL])
 929                            : 0);
 930     if (a[ODP_PACKET_ATTR_ACTIONS]) {
 931         upcall->actions = (void *) nl_attr_get(a[ODP_PACKET_ATTR_ACTIONS]);
 932         upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]);
 933     }
 934
 935     *dp_ifindex = odp_header->dp_ifindex;
 936
 937     return 0;
 938 }
 939
 940 static int
 941 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
 942 {
 943     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 944     struct ofpbuf *buf;
 945     int error;
 946     int i;
 947
 948     if (!dpif->mc_sock) {
 949         return EAGAIN;
 950     }
 951
 952     for (i = 0; i < 50; i++) {
 953         int dp_ifindex;
 954
 955         error = nl_sock_recv(dpif->mc_sock, &buf, false);
 956         if (error) {
 957             return error;
 958         }
 959
 960         error = parse_odp_packet(buf, upcall, &dp_ifindex);
 961         if (!error
 962             && dp_ifindex == dpif->dp_ifindex
 963             && dpif->listen_mask & (1u << upcall->type)) {
 964             return 0;
 965         }
 966
 967         ofpbuf_delete(buf);
 968         if (error) {
 969             return error;
 970         }
 971     }
 972
 973     return EAGAIN;
 974 }
 975
 976 static void
 977 dpif_linux_recv_wait(struct dpif *dpif_)
 978 {
 979     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 980     if (dpif->mc_sock) {
 981         nl_sock_wait(dpif->mc_sock, POLLIN);
 982     }
 983 }
 984
 985 static void
 986 dpif_linux_recv_purge(struct dpif *dpif_)
 987 {
 988     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 989
 990     if (dpif->mc_sock) {
 991         nl_sock_drain(dpif->mc_sock);
 992     }
 993 }
 994
 995 const struct dpif_class dpif_linux_class = {
 996     "system",
 997     dpif_linux_enumerate,
 998     dpif_linux_open,
 999     dpif_linux_close,
1000     dpif_linux_destroy,
1001     NULL,                       /* run */
1002     NULL,                       /* wait */
1003     dpif_linux_get_stats,
1004     dpif_linux_get_drop_frags,
1005     dpif_linux_set_drop_frags,
1006     dpif_linux_port_add,
1007     dpif_linux_port_del,
1008     dpif_linux_port_query_by_number,
1009     dpif_linux_port_query_by_name,
1010     dpif_linux_get_max_ports,
1011     dpif_linux_port_dump_start,
1012     dpif_linux_port_dump_next,
1013     dpif_linux_port_dump_done,
1014     dpif_linux_port_poll,
1015     dpif_linux_port_poll_wait,
1016     dpif_linux_flow_get,
1017     dpif_linux_flow_put,
1018     dpif_linux_flow_del,
1019     dpif_linux_flow_flush,
1020     dpif_linux_flow_dump_start,
1021     dpif_linux_flow_dump_next,
1022     dpif_linux_flow_dump_done,
1023     dpif_linux_execute,
1024     dpif_linux_recv_get_mask,
1025     dpif_linux_recv_set_mask,
1026     dpif_linux_get_sflow_probability,
1027     dpif_linux_set_sflow_probability,
1028     dpif_linux_queue_to_priority,
1029     dpif_linux_recv,
1030     dpif_linux_recv_wait,
1031     dpif_linux_recv_purge,
1032 };
1033 \f
1034 static int
1035 dpif_linux_init(void)
1036 {
1037     static int error = -1;
1038
1039     if (error < 0) {
1040         error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY,
1041                                       &odp_datapath_family);
1042         if (error) {
1043             VLOG_ERR("Generic Netlink family '%s' does not exist. "
1044                      "The Open vSwitch kernel module is probably not loaded.",
1045                      ODP_DATAPATH_FAMILY);
1046         }
1047         if (!error) {
1048             error = nl_lookup_genl_family(ODP_VPORT_FAMILY, &odp_vport_family);
1049         }
1050         if (!error) {
1051             error = nl_lookup_genl_family(ODP_FLOW_FAMILY, &odp_flow_family);
1052         }
1053         if (!error) {
1054             error = nl_lookup_genl_family(ODP_PACKET_FAMILY,
1055                                           &odp_packet_family);
1056         }
1057         if (!error) {
1058             error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1059         }
1060     }
1061
1062     return error;
1063 }
1064
1065 bool
1066 dpif_linux_is_internal_device(const char *name)
1067 {
1068     struct dpif_linux_vport reply;
1069     struct ofpbuf *buf;
1070     int error;
1071
1072     error = dpif_linux_vport_get(name, &reply, &buf);
1073     if (!error) {
1074         ofpbuf_delete(buf);
1075     } else if (error != ENODEV && error != ENOENT) {
1076         VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1077                      name, strerror(error));
1078     }
1079
1080     return reply.type == ODP_VPORT_TYPE_INTERNAL;
1081 }
1082
1083 int
1084 dpif_linux_vport_send(int dp_ifindex, uint32_t port_no,
1085                       const void *data, size_t size)
1086 {
1087     struct odp_header *execute;
1088     struct ofpbuf *buf;
1089     size_t actions_ofs;
1090     int error;
1091
1092     buf = ofpbuf_new(128 + size);
1093
1094     nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
1095                           ODP_PACKET_CMD_EXECUTE, 1);
1096
1097     execute = ofpbuf_put_uninit(buf, sizeof *execute);
1098     execute->dp_ifindex = dp_ifindex;
1099
1100     nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, data, size);
1101
1102     actions_ofs = nl_msg_start_nested(buf, ODP_PACKET_ATTR_ACTIONS);
1103     nl_msg_put_u32(buf, ODP_ACTION_ATTR_OUTPUT, port_no);
1104     nl_msg_end_nested(buf, actions_ofs);
1105
1106     error = nl_sock_transact(genl_sock, buf, NULL);
1107     ofpbuf_delete(buf);
1108     return error;
1109 }
1110
1111 static void
1112 dpif_linux_port_changed(const struct rtnetlink_link_change *change,
1113                         void *dpif_)
1114 {
1115     struct dpif_linux *dpif = dpif_;
1116
1117     if (change) {
1118         if (change->master_ifindex == dpif->dp_ifindex
1119             && (change->nlmsg_type == RTM_NEWLINK
1120                 || change->nlmsg_type == RTM_DELLINK))
1121         {
1122             /* Our datapath changed, either adding a new port or deleting an
1123              * existing one. */
1124             sset_add(&dpif->changed_ports, change->ifname);
1125         }
1126     } else {
1127         dpif->change_error = true;
1128     }
1129 }
1130 \f
1131 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1132  * by Netlink attributes, into 'vport'.  Returns 0 if successful, otherwise a
1133  * positive errno value.
1134  *
1135  * 'vport' will contain pointers into 'buf', so the caller should not free
1136  * 'buf' while 'vport' is still in use. */
1137 static int
1138 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1139                              const struct ofpbuf *buf)
1140 {
1141     static const struct nl_policy odp_vport_policy[] = {
1142         [ODP_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1143         [ODP_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1144         [ODP_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1145         [ODP_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC,
1146                                    .min_len = sizeof(struct rtnl_link_stats64),
1147                                    .max_len = sizeof(struct rtnl_link_stats64),
1148                                    .optional = true },
1149         [ODP_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
1150                                      .min_len = ETH_ADDR_LEN,
1151                                      .max_len = ETH_ADDR_LEN,
1152                                      .optional = true },
1153         [ODP_VPORT_ATTR_MTU] = { .type = NL_A_U32, .optional = true },
1154         [ODP_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1155         [ODP_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true },
1156         [ODP_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true },
1157     };
1158
1159     struct nlattr *a[ARRAY_SIZE(odp_vport_policy)];
1160     struct odp_header *odp_header;
1161     struct nlmsghdr *nlmsg;
1162     struct genlmsghdr *genl;
1163     struct ofpbuf b;
1164
1165     dpif_linux_vport_init(vport);
1166
1167     ofpbuf_use_const(&b, buf->data, buf->size);
1168     nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1169     genl = ofpbuf_try_pull(&b, sizeof *genl);
1170     odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1171     if (!nlmsg || !genl || !odp_header
1172         || nlmsg->nlmsg_type != odp_vport_family
1173         || !nl_policy_parse(&b, 0, odp_vport_policy, a,
1174                             ARRAY_SIZE(odp_vport_policy))) {
1175         return EINVAL;
1176     }
1177
1178     vport->cmd = genl->cmd;
1179     vport->dp_ifindex = odp_header->dp_ifindex;
1180     vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1181     vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1182     vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]);
1183     if (a[ODP_VPORT_ATTR_STATS]) {
1184         vport->stats = nl_attr_get(a[ODP_VPORT_ATTR_STATS]);
1185     }
1186     if (a[ODP_VPORT_ATTR_ADDRESS]) {
1187         vport->address = nl_attr_get(a[ODP_VPORT_ATTR_ADDRESS]);
1188     }
1189     if (a[ODP_VPORT_ATTR_MTU]) {
1190         vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]);
1191     } else {
1192         vport->mtu = INT_MAX;
1193     }
1194     if (a[ODP_VPORT_ATTR_OPTIONS]) {
1195         vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]);
1196         vport->options_len = nl_attr_get_size(a[ODP_VPORT_ATTR_OPTIONS]);
1197     }
1198     if (a[ODP_VPORT_ATTR_IFINDEX]) {
1199         vport->ifindex = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFINDEX]);
1200     }
1201     if (a[ODP_VPORT_ATTR_IFLINK]) {
1202         vport->iflink = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFLINK]);
1203     }
1204     return 0;
1205 }
1206
1207 /* Appends to 'buf' (which must initially be empty) a "struct odp_header"
1208  * followed by Netlink attributes corresponding to 'vport'. */
1209 static void
1210 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1211                            struct ofpbuf *buf)
1212 {
1213     struct odp_header *odp_header;
1214
1215     nl_msg_put_genlmsghdr(buf, 0, odp_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1216                           vport->cmd, 1);
1217
1218     odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1219     odp_header->dp_ifindex = vport->dp_ifindex;
1220
1221     if (vport->port_no != UINT32_MAX) {
1222         nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1223     }
1224
1225     if (vport->type != ODP_VPORT_TYPE_UNSPEC) {
1226         nl_msg_put_u32(buf, ODP_VPORT_ATTR_TYPE, vport->type);
1227     }
1228
1229     if (vport->name) {
1230         nl_msg_put_string(buf, ODP_VPORT_ATTR_NAME, vport->name);
1231     }
1232
1233     if (vport->stats) {
1234         nl_msg_put_unspec(buf, ODP_VPORT_ATTR_STATS,
1235                           vport->stats, sizeof *vport->stats);
1236     }
1237
1238     if (vport->address) {
1239         nl_msg_put_unspec(buf, ODP_VPORT_ATTR_ADDRESS,
1240                           vport->address, ETH_ADDR_LEN);
1241     }
1242
1243     if (vport->mtu && vport->mtu != INT_MAX) {
1244         nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu);
1245     }
1246
1247     if (vport->options) {
1248         nl_msg_put_nested(buf, ODP_VPORT_ATTR_OPTIONS,
1249                           vport->options, vport->options_len);
1250     }
1251
1252     if (vport->ifindex) {
1253         nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFINDEX, vport->ifindex);
1254     }
1255
1256     if (vport->iflink) {
1257         nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFLINK, vport->iflink);
1258     }
1259 }
1260
1261 /* Clears 'vport' to "empty" values. */
1262 void
1263 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1264 {
1265     memset(vport, 0, sizeof *vport);
1266     vport->port_no = UINT32_MAX;
1267 }
1268
1269 /* Executes 'request' in the kernel datapath.  If the command fails, returns a
1270  * positive errno value.  Otherwise, if 'reply' and 'bufp' are null, returns 0
1271  * without doing anything else.  If 'reply' and 'bufp' are nonnull, then the
1272  * result of the command is expected to be an odp_vport also, which is decoded
1273  * and stored in '*reply' and '*bufp'.  The caller must free '*bufp' when the
1274  * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1275 int
1276 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1277                           struct dpif_linux_vport *reply,
1278                           struct ofpbuf **bufp)
1279 {
1280     struct ofpbuf *request_buf;
1281     int error;
1282
1283     assert((reply != NULL) == (bufp != NULL));
1284
1285     error = dpif_linux_init();
1286     if (error) {
1287         if (reply) {
1288             *bufp = NULL;
1289             dpif_linux_vport_init(reply);
1290         }
1291         return error;
1292     }
1293
1294     request_buf = ofpbuf_new(1024);
1295     dpif_linux_vport_to_ofpbuf(request, request_buf);
1296     error = nl_sock_transact(genl_sock, request_buf, bufp);
1297     ofpbuf_delete(request_buf);
1298
1299     if (reply) {
1300         if (!error) {
1301             error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1302         }
1303         if (error) {
1304             dpif_linux_vport_init(reply);
1305             ofpbuf_delete(*bufp);
1306             *bufp = NULL;
1307         }
1308     }
1309     return error;
1310 }
1311
1312 /* Obtains information about the kernel vport named 'name' and stores it into
1313  * '*reply' and '*bufp'.  The caller must free '*bufp' when the reply is no
1314  * longer needed ('reply' will contain pointers into '*bufp').  */
1315 int
1316 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1317                      struct ofpbuf **bufp)
1318 {
1319     struct dpif_linux_vport request;
1320
1321     dpif_linux_vport_init(&request);
1322     request.cmd = ODP_VPORT_CMD_GET;
1323     request.name = name;
1324
1325     return dpif_linux_vport_transact(&request, reply, bufp);
1326 }
1327 \f
1328 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1329  * by Netlink attributes, into 'dp'.  Returns 0 if successful, otherwise a
1330  * positive errno value.
1331  *
1332  * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1333  * while 'dp' is still in use. */
1334 static int
1335 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1336 {
1337     static const struct nl_policy odp_datapath_policy[] = {
1338         [ODP_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1339         [ODP_DP_ATTR_STATS] = { .type = NL_A_UNSPEC,
1340                                 .min_len = sizeof(struct odp_stats),
1341                                 .max_len = sizeof(struct odp_stats),
1342                                 .optional = true },
1343         [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
1344         [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
1345         [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
1346     };
1347
1348     struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)];
1349     struct odp_header *odp_header;
1350     struct nlmsghdr *nlmsg;
1351     struct genlmsghdr *genl;
1352     struct ofpbuf b;
1353
1354     dpif_linux_dp_init(dp);
1355
1356     ofpbuf_use_const(&b, buf->data, buf->size);
1357     nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1358     genl = ofpbuf_try_pull(&b, sizeof *genl);
1359     odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1360     if (!nlmsg || !genl || !odp_header
1361         || nlmsg->nlmsg_type != odp_datapath_family
1362         || !nl_policy_parse(&b, 0, odp_datapath_policy, a,
1363                             ARRAY_SIZE(odp_datapath_policy))) {
1364         return EINVAL;
1365     }
1366
1367     dp->cmd = genl->cmd;
1368     dp->dp_ifindex = odp_header->dp_ifindex;
1369     dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]);
1370     if (a[ODP_DP_ATTR_STATS]) {
1371         /* Can't use structure assignment because Netlink doesn't ensure
1372          * sufficient alignment for 64-bit members. */
1373         memcpy(&dp->stats, nl_attr_get(a[ODP_DP_ATTR_STATS]),
1374                sizeof dp->stats);
1375     }
1376     if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1377         dp->ipv4_frags = nl_attr_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1378     }
1379     if (a[ODP_DP_ATTR_SAMPLING]) {
1380         dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]);
1381     }
1382
1383     if (a[ODP_DP_ATTR_MCGROUPS]) {
1384         static const struct nl_policy odp_mcgroup_policy[] = {
1385             [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
1386             [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
1387             [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
1388         };
1389
1390         struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)];
1391
1392         if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy,
1393                              mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) {
1394             return EINVAL;
1395         }
1396
1397         if (mcgroups[ODP_PACKET_CMD_MISS]) {
1398             dp->mcgroups[DPIF_UC_MISS]
1399                 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]);
1400         }
1401         if (mcgroups[ODP_PACKET_CMD_ACTION]) {
1402             dp->mcgroups[DPIF_UC_ACTION]
1403                 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]);
1404         }
1405         if (mcgroups[ODP_PACKET_CMD_SAMPLE]) {
1406             dp->mcgroups[DPIF_UC_SAMPLE]
1407                 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]);
1408         }
1409     }
1410
1411     return 0;
1412 }
1413
1414 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1415 static void
1416 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1417 {
1418     struct odp_header *odp_header;
1419
1420     nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family,
1421                           NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1);
1422
1423     odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1424     odp_header->dp_ifindex = dp->dp_ifindex;
1425
1426     if (dp->name) {
1427         nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name);
1428     }
1429
1430     /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */
1431
1432     if (dp->ipv4_frags) {
1433         nl_msg_put_u32(buf, ODP_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags);
1434     }
1435
1436     if (dp->sampling) {
1437         nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling);
1438     }
1439 }
1440
1441 /* Clears 'dp' to "empty" values. */
1442 void
1443 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1444 {
1445     memset(dp, 0, sizeof *dp);
1446 }
1447
1448 static void
1449 dpif_linux_dp_dump_start(struct nl_dump *dump)
1450 {
1451     struct dpif_linux_dp request;
1452     struct ofpbuf *buf;
1453
1454     dpif_linux_dp_init(&request);
1455     request.cmd = ODP_DP_CMD_GET;
1456
1457     buf = ofpbuf_new(1024);
1458     dpif_linux_dp_to_ofpbuf(&request, buf);
1459     nl_dump_start(dump, genl_sock, buf);
1460     ofpbuf_delete(buf);
1461 }
1462
1463 /* Executes 'request' in the kernel datapath.  If the command fails, returns a
1464  * positive errno value.  Otherwise, if 'reply' and 'bufp' are null, returns 0
1465  * without doing anything else.  If 'reply' and 'bufp' are nonnull, then the
1466  * result of the command is expected to be of the same form, which is decoded
1467  * and stored in '*reply' and '*bufp'.  The caller must free '*bufp' when the
1468  * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1469 int
1470 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1471                        struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1472 {
1473     struct ofpbuf *request_buf;
1474     int error;
1475
1476     assert((reply != NULL) == (bufp != NULL));
1477
1478     request_buf = ofpbuf_new(1024);
1479     dpif_linux_dp_to_ofpbuf(request, request_buf);
1480     error = nl_sock_transact(genl_sock, request_buf, bufp);
1481     ofpbuf_delete(request_buf);
1482
1483     if (reply) {
1484         if (!error) {
1485             error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1486         }
1487         if (error) {
1488             dpif_linux_dp_init(reply);
1489             ofpbuf_delete(*bufp);
1490             *bufp = NULL;
1491         }
1492     }
1493     return error;
1494 }
1495
1496 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1497  * The caller must free '*bufp' when the reply is no longer needed ('reply'
1498  * will contain pointers into '*bufp').  */
1499 int
1500 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1501                   struct ofpbuf **bufp)
1502 {
1503     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1504     struct dpif_linux_dp request;
1505
1506     dpif_linux_dp_init(&request);
1507     request.cmd = ODP_DP_CMD_GET;
1508     request.dp_ifindex = dpif->dp_ifindex;
1509
1510     return dpif_linux_dp_transact(&request, reply, bufp);
1511 }
1512 \f
1513 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1514  * by Netlink attributes, into 'flow'.  Returns 0 if successful, otherwise a
1515  * positive errno value.
1516  *
1517  * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1518  * while 'flow' is still in use. */
1519 static int
1520 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1521                             const struct ofpbuf *buf)
1522 {
1523     static const struct nl_policy odp_flow_policy[] = {
1524         [ODP_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1525         [ODP_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1526         [ODP_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC,
1527                                   .min_len = sizeof(struct odp_flow_stats),
1528                                   .max_len = sizeof(struct odp_flow_stats),
1529                                   .optional = true },
1530         [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1531         [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1532         /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */
1533     };
1534
1535     struct nlattr *a[ARRAY_SIZE(odp_flow_policy)];
1536     struct odp_header *odp_header;
1537     struct nlmsghdr *nlmsg;
1538     struct genlmsghdr *genl;
1539     struct ofpbuf b;
1540
1541     dpif_linux_flow_init(flow);
1542
1543     ofpbuf_use_const(&b, buf->data, buf->size);
1544     nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1545     genl = ofpbuf_try_pull(&b, sizeof *genl);
1546     odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1547     if (!nlmsg || !genl || !odp_header
1548         || nlmsg->nlmsg_type != odp_flow_family
1549         || !nl_policy_parse(&b, 0, odp_flow_policy, a,
1550                             ARRAY_SIZE(odp_flow_policy))) {
1551         return EINVAL;
1552     }
1553
1554     flow->nlmsg_flags = nlmsg->nlmsg_flags;
1555     flow->dp_ifindex = odp_header->dp_ifindex;
1556     flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]);
1557     flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]);
1558     if (a[ODP_FLOW_ATTR_ACTIONS]) {
1559         flow->actions = nl_attr_get(a[ODP_FLOW_ATTR_ACTIONS]);
1560         flow->actions_len = nl_attr_get_size(a[ODP_FLOW_ATTR_ACTIONS]);
1561     }
1562     if (a[ODP_FLOW_ATTR_STATS]) {
1563         flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]);
1564     }
1565     if (a[ODP_FLOW_ATTR_TCP_FLAGS]) {
1566         flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]);
1567     }
1568     if (a[ODP_FLOW_ATTR_USED]) {
1569         flow->used = nl_attr_get(a[ODP_FLOW_ATTR_USED]);
1570     }
1571     return 0;
1572 }
1573
1574 /* Appends to 'buf' (which must initially be empty) a "struct odp_header"
1575  * followed by Netlink attributes corresponding to 'flow'. */
1576 static void
1577 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1578                           struct ofpbuf *buf)
1579 {
1580     struct odp_header *odp_header;
1581
1582     nl_msg_put_genlmsghdr(buf, 0, odp_flow_family,
1583                           NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags,
1584                           flow->cmd, 1);
1585
1586     odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1587     odp_header->dp_ifindex = flow->dp_ifindex;
1588
1589     if (flow->key_len) {
1590         nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len);
1591     }
1592
1593     if (flow->actions || flow->actions_len) {
1594         nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS,
1595                           flow->actions, flow->actions_len);
1596     }
1597
1598     /* We never need to send these to the kernel. */
1599     assert(!flow->stats);
1600     assert(!flow->tcp_flags);
1601     assert(!flow->used);
1602
1603     if (flow->clear) {
1604         nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR);
1605     }
1606 }
1607
1608 /* Clears 'flow' to "empty" values. */
1609 void
1610 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1611 {
1612     memset(flow, 0, sizeof *flow);
1613 }
1614
1615 /* Executes 'request' in the kernel datapath.  If the command fails, returns a
1616  * positive errno value.  Otherwise, if 'reply' and 'bufp' are null, returns 0
1617  * without doing anything else.  If 'reply' and 'bufp' are nonnull, then the
1618  * result of the command is expected to be a flow also, which is decoded and
1619  * stored in '*reply' and '*bufp'.  The caller must free '*bufp' when the reply
1620  * is no longer needed ('reply' will contain pointers into '*bufp'). */
1621 int
1622 dpif_linux_flow_transact(const struct dpif_linux_flow *request,
1623                          struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1624 {
1625     struct ofpbuf *request_buf;
1626     int error;
1627
1628     assert((reply != NULL) == (bufp != NULL));
1629
1630     request_buf = ofpbuf_new(1024);
1631     dpif_linux_flow_to_ofpbuf(request, request_buf);
1632     error = nl_sock_transact(genl_sock, request_buf, bufp);
1633     ofpbuf_delete(request_buf);
1634
1635     if (reply) {
1636         if (!error) {
1637             error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1638         }
1639         if (error) {
1640             dpif_linux_flow_init(reply);
1641             ofpbuf_delete(*bufp);
1642             *bufp = NULL;
1643         }
1644     }
1645     return error;
1646 }
1647
1648 static void
1649 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1650                           struct dpif_flow_stats *stats)
1651 {
1652     if (flow->stats) {
1653         stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1654         stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1655     } else {
1656         stats->n_packets = 0;
1657         stats->n_bytes = 0;
1658     }
1659     stats->used = flow->used ? get_unaligned_u64(flow->used) : 0;
1660     stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
1661 }
1662