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