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