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 <poll.h>
  32 #include <stdlib.h>
  33 #include <strings.h>
  34 #include <sys/epoll.h>
  35 #include <sys/stat.h>
  36 #include <unistd.h>
  37
  38 #include "bitmap.h"
  39 #include "dpif-provider.h"
  40 #include "dynamic-string.h"
  41 #include "flow.h"
  42 #include "netdev.h"
  43 #include "netdev-linux.h"
  44 #include "netdev-vport.h"
  45 #include "netlink-notifier.h"
  46 #include "netlink-socket.h"
  47 #include "netlink.h"
  48 #include "odp-util.h"
  49 #include "ofpbuf.h"
  50 #include "openvswitch/datapath-compat.h"
  51 #include "openvswitch/tunnel.h"
  52 #include "packets.h"
  53 #include "poll-loop.h"
  54 #include "random.h"
  55 #include "shash.h"
  56 #include "sset.h"
  57 #include "unaligned.h"
  58 #include "util.h"
  59 #include "vlog.h"
  60
  61 VLOG_DEFINE_THIS_MODULE(dpif_linux);
  62 enum { MAX_PORTS = USHRT_MAX };
  63
  64 enum { N_UPCALL_SOCKS = 16 };
  65 BUILD_ASSERT_DECL(IS_POW2(N_UPCALL_SOCKS));
  66 BUILD_ASSERT_DECL(N_UPCALL_SOCKS <= 32); /* We use a 32-bit word as a mask. */
  67
  68 /* This ethtool flag was introduced in Linux 2.6.24, so it might be
  69  * missing if we have old headers. */
  70 #define ETH_FLAG_LRO      (1 << 15)    /* LRO is enabled */
  71
  72 struct dpif_linux_dp {
  73     /* Generic Netlink header. */
  74     uint8_t cmd;
  75
  76     /* struct ovs_header. */
  77     int dp_ifindex;
  78
  79     /* Attributes. */
  80     const char *name;                  /* OVS_DP_ATTR_NAME. */
  81     const uint32_t *upcall_pid;        /* OVS_DP_UPCALL_PID. */
  82     struct ovs_dp_stats stats;         /* OVS_DP_ATTR_STATS. */
  83 };
  84
  85 static void dpif_linux_dp_init(struct dpif_linux_dp *);
  86 static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
  87                                      const struct ofpbuf *);
  88 static void dpif_linux_dp_dump_start(struct nl_dump *);
  89 static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
  90                                   struct dpif_linux_dp *reply,
  91                                   struct ofpbuf **bufp);
  92 static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
  93                              struct ofpbuf **bufp);
  94
  95 struct dpif_linux_flow {
  96     /* Generic Netlink header. */
  97     uint8_t cmd;
  98
  99     /* struct ovs_header. */
 100     unsigned int nlmsg_flags;
 101     int dp_ifindex;
 102
 103     /* Attributes.
 104      *
 105      * The 'stats' member points to 64-bit data that might only be aligned on
 106      * 32-bit boundaries, so get_unaligned_u64() should be used to access its
 107      * values.
 108      *
 109      * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in
 110      * the Netlink version of the command, even if actions_len is zero. */
 111     const struct nlattr *key;           /* OVS_FLOW_ATTR_KEY. */
 112     size_t key_len;
 113     const struct nlattr *actions;       /* OVS_FLOW_ATTR_ACTIONS. */
 114     size_t actions_len;
 115     const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */
 116     const uint8_t *tcp_flags;           /* OVS_FLOW_ATTR_TCP_FLAGS. */
 117     const ovs_32aligned_u64 *used;      /* OVS_FLOW_ATTR_USED. */
 118     bool clear;                         /* OVS_FLOW_ATTR_CLEAR. */
 119 };
 120
 121 static void dpif_linux_flow_init(struct dpif_linux_flow *);
 122 static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *,
 123                                        const struct ofpbuf *);
 124 static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
 125                                       struct ofpbuf *);
 126 static int dpif_linux_flow_transact(struct dpif_linux_flow *request,
 127                                     struct dpif_linux_flow *reply,
 128                                     struct ofpbuf **bufp);
 129 static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
 130                                       struct dpif_flow_stats *);
 131
 132 /* Datapath interface for the openvswitch Linux kernel module. */
 133 struct dpif_linux {
 134     struct dpif dpif;
 135     int dp_ifindex;
 136
 137     /* Upcall messages. */
 138     struct nl_sock *upcall_socks[N_UPCALL_SOCKS];
 139     uint32_t ready_mask;        /* 1-bit for each sock with unread messages. */
 140     int epoll_fd;               /* epoll fd that includes the upcall socks. */
 141
 142     /* Change notification. */
 143     struct sset changed_ports;  /* Ports that have changed. */
 144     struct nln_notifier *port_notifier;
 145     bool change_error;
 146
 147     /* Port number allocation. */
 148     uint16_t alloc_port_no;
 149 };
 150
 151 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
 152
 153 /* Generic Netlink family numbers for OVS. */
 154 static int ovs_datapath_family;
 155 static int ovs_vport_family;
 156 static int ovs_flow_family;
 157 static int ovs_packet_family;
 158
 159 /* Generic Netlink socket. */
 160 static struct nl_sock *genl_sock;
 161 static struct nln *nln = NULL;
 162
 163 static int dpif_linux_init(void);
 164 static void open_dpif(const struct dpif_linux_dp *, struct dpif **);
 165 static bool dpif_linux_nln_parse(struct ofpbuf *, void *);
 166 static void dpif_linux_port_changed(const void *vport, void *dpif);
 167 static uint32_t dpif_linux_port_get_pid(const struct dpif *, uint16_t port_no);
 168
 169 static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
 170                                        struct ofpbuf *);
 171 static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *,
 172                                         const struct ofpbuf *);
 173
 174 static struct dpif_linux *
 175 dpif_linux_cast(const struct dpif *dpif)
 176 {
 177     dpif_assert_class(dpif, &dpif_linux_class);
 178     return CONTAINER_OF(dpif, struct dpif_linux, dpif);
 179 }
 180
 181 static int
 182 dpif_linux_enumerate(struct sset *all_dps)
 183 {
 184     struct nl_dump dump;
 185     struct ofpbuf msg;
 186     int error;
 187
 188     error = dpif_linux_init();
 189     if (error) {
 190         return error;
 191     }
 192
 193     dpif_linux_dp_dump_start(&dump);
 194     while (nl_dump_next(&dump, &msg)) {
 195         struct dpif_linux_dp dp;
 196
 197         if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
 198             sset_add(all_dps, dp.name);
 199         }
 200     }
 201     return nl_dump_done(&dump);
 202 }
 203
 204 static int
 205 dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
 206                 bool create, struct dpif **dpifp)
 207 {
 208     struct dpif_linux_dp dp_request, dp;
 209     struct ofpbuf *buf;
 210     uint32_t upcall_pid;
 211     int error;
 212
 213     error = dpif_linux_init();
 214     if (error) {
 215         return error;
 216     }
 217
 218     /* Create or look up datapath. */
 219     dpif_linux_dp_init(&dp_request);
 220     if (create) {
 221         dp_request.cmd = OVS_DP_CMD_NEW;
 222         upcall_pid = 0;
 223         dp_request.upcall_pid = &upcall_pid;
 224     } else {
 225         dp_request.cmd = OVS_DP_CMD_GET;
 226     }
 227     dp_request.name = name;
 228     error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
 229     if (error) {
 230         return error;
 231     }
 232
 233     open_dpif(&dp, dpifp);
 234     ofpbuf_delete(buf);
 235     return 0;
 236 }
 237
 238 static void
 239 open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
 240 {
 241     struct dpif_linux *dpif;
 242
 243     dpif = xzalloc(sizeof *dpif);
 244     dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed,
 245                                               dpif);
 246     dpif->epoll_fd = -1;
 247
 248     dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
 249               dp->dp_ifindex, dp->dp_ifindex);
 250
 251     dpif->dp_ifindex = dp->dp_ifindex;
 252     sset_init(&dpif->changed_ports);
 253     *dpifp = &dpif->dpif;
 254 }
 255
 256 static void
 257 destroy_upcall_socks(struct dpif_linux *dpif)
 258 {
 259     int i;
 260
 261     if (dpif->epoll_fd >= 0) {
 262         close(dpif->epoll_fd);
 263         dpif->epoll_fd = -1;
 264     }
 265     for (i = 0; i < N_UPCALL_SOCKS; i++) {
 266         nl_sock_destroy(dpif->upcall_socks[i]);
 267         dpif->upcall_socks[i] = NULL;
 268     }
 269 }
 270
 271 static void
 272 dpif_linux_close(struct dpif *dpif_)
 273 {
 274     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 275
 276     nln_notifier_destroy(dpif->port_notifier);
 277     destroy_upcall_socks(dpif);
 278     sset_destroy(&dpif->changed_ports);
 279     free(dpif);
 280 }
 281
 282 static int
 283 dpif_linux_destroy(struct dpif *dpif_)
 284 {
 285     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 286     struct dpif_linux_dp dp;
 287
 288     dpif_linux_dp_init(&dp);
 289     dp.cmd = OVS_DP_CMD_DEL;
 290     dp.dp_ifindex = dpif->dp_ifindex;
 291     return dpif_linux_dp_transact(&dp, NULL, NULL);
 292 }
 293
 294 static void
 295 dpif_linux_run(struct dpif *dpif OVS_UNUSED)
 296 {
 297     if (nln) {
 298         nln_run(nln);
 299     }
 300 }
 301
 302 static void
 303 dpif_linux_wait(struct dpif *dpif OVS_UNUSED)
 304 {
 305     if (nln) {
 306         nln_wait(nln);
 307     }
 308 }
 309
 310 static int
 311 dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats)
 312 {
 313     struct dpif_linux_dp dp;
 314     struct ofpbuf *buf;
 315     int error;
 316
 317     error = dpif_linux_dp_get(dpif_, &dp, &buf);
 318     if (!error) {
 319         stats->n_hit    = dp.stats.n_hit;
 320         stats->n_missed = dp.stats.n_missed;
 321         stats->n_lost   = dp.stats.n_lost;
 322         stats->n_flows  = dp.stats.n_flows;
 323         ofpbuf_delete(buf);
 324     }
 325     return error;
 326 }
 327
 328 static int
 329 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
 330                     uint16_t *port_nop)
 331 {
 332     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 333     const char *name = netdev_get_name(netdev);
 334     const char *type = netdev_get_type(netdev);
 335     struct dpif_linux_vport request, reply;
 336     const struct ofpbuf *options;
 337     struct ofpbuf *buf;
 338     int error, i = 0, max_ports = MAX_PORTS;
 339
 340     dpif_linux_vport_init(&request);
 341     request.cmd = OVS_VPORT_CMD_NEW;
 342     request.dp_ifindex = dpif->dp_ifindex;
 343     request.type = netdev_vport_get_vport_type(netdev);
 344     if (request.type == OVS_VPORT_TYPE_UNSPEC) {
 345         VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
 346                      "unsupported type `%s'",
 347                      dpif_name(dpif_), name, type);
 348         return EINVAL;
 349     }
 350     request.name = name;
 351
 352     options = netdev_vport_get_options(netdev);
 353     if (options && options->size) {
 354         request.options = options->data;
 355         request.options_len = options->size;
 356     }
 357
 358     if (request.type == OVS_VPORT_TYPE_NETDEV) {
 359         netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
 360     }
 361
 362     /* Loop until we find a port that isn't used. */
 363     do {
 364         uint32_t upcall_pid;
 365
 366         request.port_no = ++dpif->alloc_port_no;
 367         upcall_pid = dpif_linux_port_get_pid(dpif_, request.port_no);
 368         request.upcall_pid = &upcall_pid;
 369         error = dpif_linux_vport_transact(&request, &reply, &buf);
 370
 371         if (!error) {
 372             *port_nop = reply.port_no;
 373             VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
 374                      dpif_name(dpif_), request.port_no, upcall_pid);
 375         } else if (error == EFBIG) {
 376             /* Older datapath has lower limit. */
 377             max_ports = dpif->alloc_port_no;
 378             dpif->alloc_port_no = 0;
 379         }
 380
 381         ofpbuf_delete(buf);
 382     } while ((i++ < max_ports)
 383              && (error == EBUSY || error == EFBIG));
 384
 385     return error;
 386 }
 387
 388 static int
 389 dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
 390 {
 391     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 392     struct dpif_linux_vport vport;
 393     int error;
 394
 395     dpif_linux_vport_init(&vport);
 396     vport.cmd = OVS_VPORT_CMD_DEL;
 397     vport.dp_ifindex = dpif->dp_ifindex;
 398     vport.port_no = port_no;
 399     error = dpif_linux_vport_transact(&vport, NULL, NULL);
 400
 401     return error;
 402 }
 403
 404 static int
 405 dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
 406                         const char *port_name, struct dpif_port *dpif_port)
 407 {
 408     struct dpif_linux_vport request;
 409     struct dpif_linux_vport reply;
 410     struct ofpbuf *buf;
 411     int error;
 412
 413     dpif_linux_vport_init(&request);
 414     request.cmd = OVS_VPORT_CMD_GET;
 415     request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
 416     request.port_no = port_no;
 417     request.name = port_name;
 418
 419     error = dpif_linux_vport_transact(&request, &reply, &buf);
 420     if (!error) {
 421         dpif_port->name = xstrdup(reply.name);
 422         dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
 423         dpif_port->port_no = reply.port_no;
 424         ofpbuf_delete(buf);
 425     }
 426     return error;
 427 }
 428
 429 static int
 430 dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
 431                                 struct dpif_port *dpif_port)
 432 {
 433     return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
 434 }
 435
 436 static int
 437 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
 438                               struct dpif_port *dpif_port)
 439 {
 440     return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
 441 }
 442
 443 static int
 444 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
 445 {
 446     return MAX_PORTS;
 447 }
 448
 449 static uint32_t
 450 dpif_linux_port_get_pid(const struct dpif *dpif_, uint16_t port_no)
 451 {
 452     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 453
 454     if (dpif->epoll_fd < 0) {
 455         return 0;
 456     } else {
 457         int idx = port_no & (N_UPCALL_SOCKS - 1);
 458         return nl_sock_pid(dpif->upcall_socks[idx]);
 459     }
 460 }
 461
 462 static int
 463 dpif_linux_flow_flush(struct dpif *dpif_)
 464 {
 465     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 466     struct dpif_linux_flow flow;
 467
 468     dpif_linux_flow_init(&flow);
 469     flow.cmd = OVS_FLOW_CMD_DEL;
 470     flow.dp_ifindex = dpif->dp_ifindex;
 471     return dpif_linux_flow_transact(&flow, NULL, NULL);
 472 }
 473
 474 struct dpif_linux_port_state {
 475     struct nl_dump dump;
 476 };
 477
 478 static int
 479 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
 480 {
 481     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 482     struct dpif_linux_port_state *state;
 483     struct dpif_linux_vport request;
 484     struct ofpbuf *buf;
 485
 486     *statep = state = xmalloc(sizeof *state);
 487
 488     dpif_linux_vport_init(&request);
 489     request.cmd = OVS_DP_CMD_GET;
 490     request.dp_ifindex = dpif->dp_ifindex;
 491
 492     buf = ofpbuf_new(1024);
 493     dpif_linux_vport_to_ofpbuf(&request, buf);
 494     nl_dump_start(&state->dump, genl_sock, buf);
 495     ofpbuf_delete(buf);
 496
 497     return 0;
 498 }
 499
 500 static int
 501 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
 502                           struct dpif_port *dpif_port)
 503 {
 504     struct dpif_linux_port_state *state = state_;
 505     struct dpif_linux_vport vport;
 506     struct ofpbuf buf;
 507     int error;
 508
 509     if (!nl_dump_next(&state->dump, &buf)) {
 510         return EOF;
 511     }
 512
 513     error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
 514     if (error) {
 515         return error;
 516     }
 517
 518     dpif_port->name = (char *) vport.name;
 519     dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
 520     dpif_port->port_no = vport.port_no;
 521     return 0;
 522 }
 523
 524 static int
 525 dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
 526 {
 527     struct dpif_linux_port_state *state = state_;
 528     int error = nl_dump_done(&state->dump);
 529
 530     free(state);
 531     return error;
 532 }
 533
 534 static int
 535 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
 536 {
 537     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 538
 539     if (dpif->change_error) {
 540         dpif->change_error = false;
 541         sset_clear(&dpif->changed_ports);
 542         return ENOBUFS;
 543     } else if (!sset_is_empty(&dpif->changed_ports)) {
 544         *devnamep = sset_pop(&dpif->changed_ports);
 545         return 0;
 546     } else {
 547         return EAGAIN;
 548     }
 549 }
 550
 551 static void
 552 dpif_linux_port_poll_wait(const struct dpif *dpif_)
 553 {
 554     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 555     if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
 556         poll_immediate_wake();
 557     }
 558 }
 559
 560 static int
 561 dpif_linux_flow_get__(const struct dpif *dpif_,
 562                       const struct nlattr *key, size_t key_len,
 563                       struct dpif_linux_flow *reply, struct ofpbuf **bufp)
 564 {
 565     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 566     struct dpif_linux_flow request;
 567
 568     dpif_linux_flow_init(&request);
 569     request.cmd = OVS_FLOW_CMD_GET;
 570     request.dp_ifindex = dpif->dp_ifindex;
 571     request.key = key;
 572     request.key_len = key_len;
 573     return dpif_linux_flow_transact(&request, reply, bufp);
 574 }
 575
 576 static int
 577 dpif_linux_flow_get(const struct dpif *dpif_,
 578                     const struct nlattr *key, size_t key_len,
 579                     struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
 580 {
 581     struct dpif_linux_flow reply;
 582     struct ofpbuf *buf;
 583     int error;
 584
 585     error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
 586     if (!error) {
 587         if (stats) {
 588             dpif_linux_flow_get_stats(&reply, stats);
 589         }
 590         if (actionsp) {
 591             buf->data = (void *) reply.actions;
 592             buf->size = reply.actions_len;
 593             *actionsp = buf;
 594         } else {
 595             ofpbuf_delete(buf);
 596         }
 597     }
 598     return error;
 599 }
 600
 601 static void
 602 dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
 603                          struct dpif_linux_flow *request)
 604 {
 605     static struct nlattr dummy_action;
 606
 607     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 608
 609     dpif_linux_flow_init(request);
 610     request->cmd = (put->flags & DPIF_FP_CREATE
 611                     ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
 612     request->dp_ifindex = dpif->dp_ifindex;
 613     request->key = put->key;
 614     request->key_len = put->key_len;
 615     /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
 616     request->actions = put->actions ? put->actions : &dummy_action;
 617     request->actions_len = put->actions_len;
 618     if (put->flags & DPIF_FP_ZERO_STATS) {
 619         request->clear = true;
 620     }
 621     request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
 622 }
 623
 624 static int
 625 dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put)
 626 {
 627     struct dpif_linux_flow request, reply;
 628     struct ofpbuf *buf;
 629     int error;
 630
 631     dpif_linux_init_flow_put(dpif_, put, &request);
 632     error = dpif_linux_flow_transact(&request,
 633                                      put->stats ? &reply : NULL,
 634                                      put->stats ? &buf : NULL);
 635     if (!error && put->stats) {
 636         dpif_linux_flow_get_stats(&reply, put->stats);
 637         ofpbuf_delete(buf);
 638     }
 639     return error;
 640 }
 641
 642 static int
 643 dpif_linux_flow_del(struct dpif *dpif_,
 644                     const struct nlattr *key, size_t key_len,
 645                     struct dpif_flow_stats *stats)
 646 {
 647     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 648     struct dpif_linux_flow request, reply;
 649     struct ofpbuf *buf;
 650     int error;
 651
 652     dpif_linux_flow_init(&request);
 653     request.cmd = OVS_FLOW_CMD_DEL;
 654     request.dp_ifindex = dpif->dp_ifindex;
 655     request.key = key;
 656     request.key_len = key_len;
 657     error = dpif_linux_flow_transact(&request,
 658                                      stats ? &reply : NULL,
 659                                      stats ? &buf : NULL);
 660     if (!error && stats) {
 661         dpif_linux_flow_get_stats(&reply, stats);
 662         ofpbuf_delete(buf);
 663     }
 664     return error;
 665 }
 666
 667 struct dpif_linux_flow_state {
 668     struct nl_dump dump;
 669     struct dpif_linux_flow flow;
 670     struct dpif_flow_stats stats;
 671     struct ofpbuf *buf;
 672 };
 673
 674 static int
 675 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
 676 {
 677     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 678     struct dpif_linux_flow_state *state;
 679     struct dpif_linux_flow request;
 680     struct ofpbuf *buf;
 681
 682     *statep = state = xmalloc(sizeof *state);
 683
 684     dpif_linux_flow_init(&request);
 685     request.cmd = OVS_DP_CMD_GET;
 686     request.dp_ifindex = dpif->dp_ifindex;
 687
 688     buf = ofpbuf_new(1024);
 689     dpif_linux_flow_to_ofpbuf(&request, buf);
 690     nl_dump_start(&state->dump, genl_sock, buf);
 691     ofpbuf_delete(buf);
 692
 693     state->buf = NULL;
 694
 695     return 0;
 696 }
 697
 698 static int
 699 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
 700                           const struct nlattr **key, size_t *key_len,
 701                           const struct nlattr **actions, size_t *actions_len,
 702                           const struct dpif_flow_stats **stats)
 703 {
 704     struct dpif_linux_flow_state *state = state_;
 705     struct ofpbuf buf;
 706     int error;
 707
 708     do {
 709         ofpbuf_delete(state->buf);
 710         state->buf = NULL;
 711
 712         if (!nl_dump_next(&state->dump, &buf)) {
 713             return EOF;
 714         }
 715
 716         error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
 717         if (error) {
 718             return error;
 719         }
 720
 721         if (actions && !state->flow.actions) {
 722             error = dpif_linux_flow_get__(dpif_, state->flow.key,
 723                                           state->flow.key_len,
 724                                           &state->flow, &state->buf);
 725             if (error == ENOENT) {
 726                 VLOG_DBG("dumped flow disappeared on get");
 727             } else if (error) {
 728                 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
 729             }
 730         }
 731     } while (error);
 732
 733     if (actions) {
 734         *actions = state->flow.actions;
 735         *actions_len = state->flow.actions_len;
 736     }
 737     if (key) {
 738         *key = state->flow.key;
 739         *key_len = state->flow.key_len;
 740     }
 741     if (stats) {
 742         dpif_linux_flow_get_stats(&state->flow, &state->stats);
 743         *stats = &state->stats;
 744     }
 745     return error;
 746 }
 747
 748 static int
 749 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
 750 {
 751     struct dpif_linux_flow_state *state = state_;
 752     int error = nl_dump_done(&state->dump);
 753     ofpbuf_delete(state->buf);
 754     free(state);
 755     return error;
 756 }
 757
 758 static struct ofpbuf *
 759 dpif_linux_encode_execute(int dp_ifindex,
 760                           const struct dpif_execute *d_exec)
 761 {
 762     struct ovs_header *k_exec;
 763     struct ofpbuf *buf;
 764
 765     buf = ofpbuf_new(128 + d_exec->actions_len + d_exec->packet->size);
 766
 767     nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
 768                           OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
 769
 770     k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
 771     k_exec->dp_ifindex = dp_ifindex;
 772
 773     nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET,
 774                       d_exec->packet->data, d_exec->packet->size);
 775     nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len);
 776     nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS,
 777                       d_exec->actions, d_exec->actions_len);
 778
 779     return buf;
 780 }
 781
 782 static int
 783 dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
 784 {
 785     struct ofpbuf *request;
 786     int error;
 787
 788     request = dpif_linux_encode_execute(dp_ifindex, execute);
 789     error = nl_sock_transact(genl_sock, request, NULL);
 790     ofpbuf_delete(request);
 791
 792     return error;
 793 }
 794
 795 static int
 796 dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
 797 {
 798     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 799
 800     return dpif_linux_execute__(dpif->dp_ifindex, execute);
 801 }
 802
 803 static void
 804 dpif_linux_operate(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
 805 {
 806     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 807     struct nl_transaction **txnsp;
 808     struct nl_transaction *txns;
 809     size_t i;
 810
 811     txns = xmalloc(n_ops * sizeof *txns);
 812     for (i = 0; i < n_ops; i++) {
 813         struct nl_transaction *txn = &txns[i];
 814         struct dpif_op *op = ops[i];
 815
 816         if (op->type == DPIF_OP_FLOW_PUT) {
 817             struct dpif_flow_put *put = &op->u.flow_put;
 818             struct dpif_linux_flow request;
 819
 820             dpif_linux_init_flow_put(dpif_, put, &request);
 821             if (put->stats) {
 822                 request.nlmsg_flags |= NLM_F_ECHO;
 823             }
 824             txn->request = ofpbuf_new(1024);
 825             dpif_linux_flow_to_ofpbuf(&request, txn->request);
 826         } else if (op->type == DPIF_OP_EXECUTE) {
 827             struct dpif_execute *execute = &op->u.execute;
 828
 829             txn->request = dpif_linux_encode_execute(dpif->dp_ifindex,
 830                                                      execute);
 831         } else {
 832             NOT_REACHED();
 833         }
 834     }
 835
 836     txnsp = xmalloc(n_ops * sizeof *txnsp);
 837     for (i = 0; i < n_ops; i++) {
 838         txnsp[i] = &txns[i];
 839     }
 840
 841     nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
 842
 843     free(txnsp);
 844
 845     for (i = 0; i < n_ops; i++) {
 846         struct nl_transaction *txn = &txns[i];
 847         struct dpif_op *op = ops[i];
 848
 849         if (op->type == DPIF_OP_FLOW_PUT) {
 850             struct dpif_flow_put *put = &op->u.flow_put;
 851             int error = txn->error;
 852
 853             if (!error && put->stats) {
 854                 struct dpif_linux_flow reply;
 855
 856                 error = dpif_linux_flow_from_ofpbuf(&reply, txn->reply);
 857                 if (!error) {
 858                     dpif_linux_flow_get_stats(&reply, put->stats);
 859                 }
 860             }
 861             op->error = error;
 862         } else if (op->type == DPIF_OP_EXECUTE) {
 863             op->error = txn->error;
 864         } else {
 865             NOT_REACHED();
 866         }
 867
 868         ofpbuf_delete(txn->request);
 869         ofpbuf_delete(txn->reply);
 870     }
 871     free(txns);
 872 }
 873
 874 static void
 875 set_upcall_pids(struct dpif *dpif_)
 876 {
 877     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 878     struct dpif_port_dump port_dump;
 879     struct dpif_port port;
 880     int error;
 881
 882     DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
 883         uint32_t upcall_pid = dpif_linux_port_get_pid(dpif_, port.port_no);
 884         struct dpif_linux_vport vport_request;
 885
 886         dpif_linux_vport_init(&vport_request);
 887         vport_request.cmd = OVS_VPORT_CMD_SET;
 888         vport_request.dp_ifindex = dpif->dp_ifindex;
 889         vport_request.port_no = port.port_no;
 890         vport_request.upcall_pid = &upcall_pid;
 891         error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
 892         if (!error) {
 893             VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
 894                      dpif_name(&dpif->dpif), vport_request.port_no,
 895                      upcall_pid);
 896         } else {
 897             VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on port: %s",
 898                          dpif_name(&dpif->dpif), strerror(error));
 899         }
 900     }
 901 }
 902
 903 static int
 904 dpif_linux_recv_set(struct dpif *dpif_, bool enable)
 905 {
 906     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 907
 908     if ((dpif->epoll_fd >= 0) == enable) {
 909         return 0;
 910     }
 911
 912     if (!enable) {
 913         destroy_upcall_socks(dpif);
 914     } else {
 915         int i;
 916         int error;
 917
 918         dpif->epoll_fd = epoll_create(N_UPCALL_SOCKS);
 919         if (dpif->epoll_fd < 0) {
 920             return errno;
 921         }
 922
 923         for (i = 0; i < N_UPCALL_SOCKS; i++) {
 924             struct epoll_event event;
 925
 926             error = nl_sock_create(NETLINK_GENERIC, &dpif->upcall_socks[i]);
 927             if (error) {
 928                 destroy_upcall_socks(dpif);
 929                 return error;
 930             }
 931
 932             memset(&event, 0, sizeof event);
 933             event.events = EPOLLIN;
 934             event.data.u32 = i;
 935             if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD,
 936                           nl_sock_fd(dpif->upcall_socks[i]), &event) < 0) {
 937                 error = errno;
 938                 destroy_upcall_socks(dpif);
 939                 return error;
 940             }
 941         }
 942
 943         dpif->ready_mask = 0;
 944     }
 945
 946     set_upcall_pids(dpif_);
 947
 948     return 0;
 949 }
 950
 951 static int
 952 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
 953                              uint32_t queue_id, uint32_t *priority)
 954 {
 955     if (queue_id < 0xf000) {
 956         *priority = TC_H_MAKE(1 << 16, queue_id + 1);
 957         return 0;
 958     } else {
 959         return EINVAL;
 960     }
 961 }
 962
 963 static int
 964 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
 965                  int *dp_ifindex)
 966 {
 967     static const struct nl_policy ovs_packet_policy[] = {
 968         /* Always present. */
 969         [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
 970                                      .min_len = ETH_HEADER_LEN },
 971         [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
 972
 973         /* OVS_PACKET_CMD_ACTION only. */
 974         [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
 975     };
 976
 977     struct ovs_header *ovs_header;
 978     struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
 979     struct nlmsghdr *nlmsg;
 980     struct genlmsghdr *genl;
 981     struct ofpbuf b;
 982     int type;
 983
 984     ofpbuf_use_const(&b, buf->data, buf->size);
 985
 986     nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
 987     genl = ofpbuf_try_pull(&b, sizeof *genl);
 988     ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
 989     if (!nlmsg || !genl || !ovs_header
 990         || nlmsg->nlmsg_type != ovs_packet_family
 991         || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
 992                             ARRAY_SIZE(ovs_packet_policy))) {
 993         return EINVAL;
 994     }
 995
 996     type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
 997             : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
 998             : -1);
 999     if (type < 0) {
1000         return EINVAL;
1001     }
1002
1003     memset(upcall, 0, sizeof *upcall);
1004     upcall->type = type;
1005     upcall->packet = buf;
1006     upcall->packet->data = (void *) nl_attr_get(a[OVS_PACKET_ATTR_PACKET]);
1007     upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
1008     upcall->key = (void *) nl_attr_get(a[OVS_PACKET_ATTR_KEY]);
1009     upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
1010     upcall->userdata = (a[OVS_PACKET_ATTR_USERDATA]
1011                         ? nl_attr_get_u64(a[OVS_PACKET_ATTR_USERDATA])
1012                         : 0);
1013     *dp_ifindex = ovs_header->dp_ifindex;
1014
1015     return 0;
1016 }
1017
1018 static int
1019 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
1020 {
1021     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1022     int read_tries = 0;
1023
1024     if (dpif->epoll_fd < 0) {
1025        return EAGAIN;
1026     }
1027
1028     if (!dpif->ready_mask) {
1029         struct epoll_event events[N_UPCALL_SOCKS];
1030         int retval;
1031         int i;
1032
1033         do {
1034             retval = epoll_wait(dpif->epoll_fd, events, N_UPCALL_SOCKS, 0);
1035         } while (retval < 0 && errno == EINTR);
1036         if (retval < 0) {
1037             static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1038             VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", strerror(errno));
1039         }
1040
1041         for (i = 0; i < retval; i++) {
1042             dpif->ready_mask |= 1u << events[i].data.u32;
1043         }
1044     }
1045
1046     while (dpif->ready_mask) {
1047         int indx = ffs(dpif->ready_mask) - 1;
1048         struct nl_sock *upcall_sock = dpif->upcall_socks[indx];
1049
1050         dpif->ready_mask &= ~(1u << indx);
1051
1052         for (;;) {
1053             struct ofpbuf *buf;
1054             int dp_ifindex;
1055             int error;
1056
1057             if (++read_tries > 50) {
1058                 return EAGAIN;
1059             }
1060
1061             error = nl_sock_recv(upcall_sock, &buf, false);
1062             if (error == EAGAIN) {
1063                 break;
1064             } else if (error) {
1065                 return error;
1066             }
1067
1068             error = parse_odp_packet(buf, upcall, &dp_ifindex);
1069             if (!error && dp_ifindex == dpif->dp_ifindex) {
1070                 return 0;
1071             }
1072
1073             ofpbuf_delete(buf);
1074             if (error) {
1075                 return error;
1076             }
1077         }
1078     }
1079
1080     return EAGAIN;
1081 }
1082
1083 static void
1084 dpif_linux_recv_wait(struct dpif *dpif_)
1085 {
1086     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1087
1088     if (dpif->epoll_fd < 0) {
1089        return;
1090     }
1091
1092     poll_fd_wait(dpif->epoll_fd, POLLIN);
1093 }
1094
1095 static void
1096 dpif_linux_recv_purge(struct dpif *dpif_)
1097 {
1098     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1099     int i;
1100
1101     if (dpif->epoll_fd < 0) {
1102        return;
1103     }
1104
1105     for (i = 0; i < N_UPCALL_SOCKS; i++) {
1106         nl_sock_drain(dpif->upcall_socks[i]);
1107     }
1108 }
1109
1110 const struct dpif_class dpif_linux_class = {
1111     "system",
1112     dpif_linux_enumerate,
1113     dpif_linux_open,
1114     dpif_linux_close,
1115     dpif_linux_destroy,
1116     dpif_linux_run,
1117     dpif_linux_wait,
1118     dpif_linux_get_stats,
1119     dpif_linux_port_add,
1120     dpif_linux_port_del,
1121     dpif_linux_port_query_by_number,
1122     dpif_linux_port_query_by_name,
1123     dpif_linux_get_max_ports,
1124     dpif_linux_port_get_pid,
1125     dpif_linux_port_dump_start,
1126     dpif_linux_port_dump_next,
1127     dpif_linux_port_dump_done,
1128     dpif_linux_port_poll,
1129     dpif_linux_port_poll_wait,
1130     dpif_linux_flow_get,
1131     dpif_linux_flow_put,
1132     dpif_linux_flow_del,
1133     dpif_linux_flow_flush,
1134     dpif_linux_flow_dump_start,
1135     dpif_linux_flow_dump_next,
1136     dpif_linux_flow_dump_done,
1137     dpif_linux_execute,
1138     dpif_linux_operate,
1139     dpif_linux_recv_set,
1140     dpif_linux_queue_to_priority,
1141     dpif_linux_recv,
1142     dpif_linux_recv_wait,
1143     dpif_linux_recv_purge,
1144 };
1145 \f
1146 static int
1147 dpif_linux_init(void)
1148 {
1149     static int error = -1;
1150
1151     if (error < 0) {
1152         unsigned int ovs_vport_mcgroup;
1153
1154         error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
1155                                       &ovs_datapath_family);
1156         if (error) {
1157             VLOG_ERR("Generic Netlink family '%s' does not exist. "
1158                      "The Open vSwitch kernel module is probably not loaded.",
1159                      OVS_DATAPATH_FAMILY);
1160         }
1161         if (!error) {
1162             error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
1163         }
1164         if (!error) {
1165             error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
1166         }
1167         if (!error) {
1168             error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
1169                                           &ovs_packet_family);
1170         }
1171         if (!error) {
1172             error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1173         }
1174         if (!error) {
1175             error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
1176                                            &ovs_vport_mcgroup,
1177                                            OVS_VPORT_MCGROUP_FALLBACK_ID);
1178         }
1179         if (!error) {
1180             static struct dpif_linux_vport vport;
1181             nln = nln_create(NETLINK_GENERIC, ovs_vport_mcgroup,
1182                              dpif_linux_nln_parse, &vport);
1183         }
1184     }
1185
1186     return error;
1187 }
1188
1189 bool
1190 dpif_linux_is_internal_device(const char *name)
1191 {
1192     struct dpif_linux_vport reply;
1193     struct ofpbuf *buf;
1194     int error;
1195
1196     error = dpif_linux_vport_get(name, &reply, &buf);
1197     if (!error) {
1198         ofpbuf_delete(buf);
1199     } else if (error != ENODEV && error != ENOENT) {
1200         VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1201                      name, strerror(error));
1202     }
1203
1204     return reply.type == OVS_VPORT_TYPE_INTERNAL;
1205 }
1206
1207 int
1208 dpif_linux_vport_send(int dp_ifindex, uint32_t port_no,
1209                       const void *data, size_t size)
1210 {
1211     struct ofpbuf actions, key, packet;
1212     struct odputil_keybuf keybuf;
1213     struct dpif_execute execute;
1214     struct flow flow;
1215     uint64_t action;
1216
1217     ofpbuf_use_const(&packet, data, size);
1218     flow_extract(&packet, 0, htonll(0), 0, &flow);
1219
1220     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1221     odp_flow_key_from_flow(&key, &flow);
1222
1223     ofpbuf_use_stack(&actions, &action, sizeof action);
1224     nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, port_no);
1225
1226     execute.key = key.data;
1227     execute.key_len = key.size;
1228     execute.actions = actions.data;
1229     execute.actions_len = actions.size;
1230     execute.packet = &packet;
1231     return dpif_linux_execute__(dp_ifindex, &execute);
1232 }
1233
1234 static bool
1235 dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_)
1236 {
1237     struct dpif_linux_vport *vport = vport_;
1238     return dpif_linux_vport_from_ofpbuf(vport, buf) == 0;
1239 }
1240
1241 static void
1242 dpif_linux_port_changed(const void *vport_, void *dpif_)
1243 {
1244     const struct dpif_linux_vport *vport = vport_;
1245     struct dpif_linux *dpif = dpif_;
1246
1247     if (vport) {
1248         if (vport->dp_ifindex == dpif->dp_ifindex
1249             && (vport->cmd == OVS_VPORT_CMD_NEW
1250                 || vport->cmd == OVS_VPORT_CMD_DEL
1251                 || vport->cmd == OVS_VPORT_CMD_SET)) {
1252             VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
1253                      dpif->dpif.full_name, vport->name, vport->cmd);
1254             sset_add(&dpif->changed_ports, vport->name);
1255         }
1256     } else {
1257         dpif->change_error = true;
1258     }
1259 }
1260 \f
1261 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1262  * by Netlink attributes, into 'vport'.  Returns 0 if successful, otherwise a
1263  * positive errno value.
1264  *
1265  * 'vport' will contain pointers into 'buf', so the caller should not free
1266  * 'buf' while 'vport' is still in use. */
1267 static int
1268 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1269                              const struct ofpbuf *buf)
1270 {
1271     static const struct nl_policy ovs_vport_policy[] = {
1272         [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1273         [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1274         [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1275         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
1276         [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats),
1277                                    .optional = true },
1278         [OVS_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
1279                                      .min_len = ETH_ADDR_LEN,
1280                                      .max_len = ETH_ADDR_LEN,
1281                                      .optional = true },
1282         [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1283     };
1284
1285     struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
1286     struct ovs_header *ovs_header;
1287     struct nlmsghdr *nlmsg;
1288     struct genlmsghdr *genl;
1289     struct ofpbuf b;
1290
1291     dpif_linux_vport_init(vport);
1292
1293     ofpbuf_use_const(&b, buf->data, buf->size);
1294     nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1295     genl = ofpbuf_try_pull(&b, sizeof *genl);
1296     ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1297     if (!nlmsg || !genl || !ovs_header
1298         || nlmsg->nlmsg_type != ovs_vport_family
1299         || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
1300                             ARRAY_SIZE(ovs_vport_policy))) {
1301         return EINVAL;
1302     }
1303
1304     vport->cmd = genl->cmd;
1305     vport->dp_ifindex = ovs_header->dp_ifindex;
1306     vport->port_no = nl_attr_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1307     vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1308     vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
1309     if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1310         vport->upcall_pid = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]);
1311     }
1312     if (a[OVS_VPORT_ATTR_STATS]) {
1313         vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
1314     }
1315     if (a[OVS_VPORT_ATTR_ADDRESS]) {
1316         vport->address = nl_attr_get(a[OVS_VPORT_ATTR_ADDRESS]);
1317     }
1318     if (a[OVS_VPORT_ATTR_OPTIONS]) {
1319         vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
1320         vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
1321     }
1322     return 0;
1323 }
1324
1325 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1326  * followed by Netlink attributes corresponding to 'vport'. */
1327 static void
1328 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1329                            struct ofpbuf *buf)
1330 {
1331     struct ovs_header *ovs_header;
1332
1333     nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1334                           vport->cmd, OVS_VPORT_VERSION);
1335
1336     ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1337     ovs_header->dp_ifindex = vport->dp_ifindex;
1338
1339     if (vport->port_no != UINT32_MAX) {
1340         nl_msg_put_u32(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1341     }
1342
1343     if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
1344         nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
1345     }
1346
1347     if (vport->name) {
1348         nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
1349     }
1350
1351     if (vport->upcall_pid) {
1352         nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid);
1353     }
1354
1355     if (vport->stats) {
1356         nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
1357                           vport->stats, sizeof *vport->stats);
1358     }
1359
1360     if (vport->address) {
1361         nl_msg_put_unspec(buf, OVS_VPORT_ATTR_ADDRESS,
1362                           vport->address, ETH_ADDR_LEN);
1363     }
1364
1365     if (vport->options) {
1366         nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
1367                           vport->options, vport->options_len);
1368     }
1369 }
1370
1371 /* Clears 'vport' to "empty" values. */
1372 void
1373 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1374 {
1375     memset(vport, 0, sizeof *vport);
1376     vport->port_no = UINT32_MAX;
1377 }
1378
1379 /* Executes 'request' in the kernel datapath.  If the command fails, returns a
1380  * positive errno value.  Otherwise, if 'reply' and 'bufp' are null, returns 0
1381  * without doing anything else.  If 'reply' and 'bufp' are nonnull, then the
1382  * result of the command is expected to be an ovs_vport also, which is decoded
1383  * and stored in '*reply' and '*bufp'.  The caller must free '*bufp' when the
1384  * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1385 int
1386 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1387                           struct dpif_linux_vport *reply,
1388                           struct ofpbuf **bufp)
1389 {
1390     struct ofpbuf *request_buf;
1391     int error;
1392
1393     assert((reply != NULL) == (bufp != NULL));
1394
1395     error = dpif_linux_init();
1396     if (error) {
1397         if (reply) {
1398             *bufp = NULL;
1399             dpif_linux_vport_init(reply);
1400         }
1401         return error;
1402     }
1403
1404     request_buf = ofpbuf_new(1024);
1405     dpif_linux_vport_to_ofpbuf(request, request_buf);
1406     error = nl_sock_transact(genl_sock, request_buf, bufp);
1407     ofpbuf_delete(request_buf);
1408
1409     if (reply) {
1410         if (!error) {
1411             error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1412         }
1413         if (error) {
1414             dpif_linux_vport_init(reply);
1415             ofpbuf_delete(*bufp);
1416             *bufp = NULL;
1417         }
1418     }
1419     return error;
1420 }
1421
1422 /* Obtains information about the kernel vport named 'name' and stores it into
1423  * '*reply' and '*bufp'.  The caller must free '*bufp' when the reply is no
1424  * longer needed ('reply' will contain pointers into '*bufp').  */
1425 int
1426 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1427                      struct ofpbuf **bufp)
1428 {
1429     struct dpif_linux_vport request;
1430
1431     dpif_linux_vport_init(&request);
1432     request.cmd = OVS_VPORT_CMD_GET;
1433     request.name = name;
1434
1435     return dpif_linux_vport_transact(&request, reply, bufp);
1436 }
1437 \f
1438 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1439  * by Netlink attributes, into 'dp'.  Returns 0 if successful, otherwise a
1440  * positive errno value.
1441  *
1442  * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1443  * while 'dp' is still in use. */
1444 static int
1445 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1446 {
1447     static const struct nl_policy ovs_datapath_policy[] = {
1448         [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1449         [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
1450                                 .optional = true },
1451     };
1452
1453     struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
1454     struct ovs_header *ovs_header;
1455     struct nlmsghdr *nlmsg;
1456     struct genlmsghdr *genl;
1457     struct ofpbuf b;
1458
1459     dpif_linux_dp_init(dp);
1460
1461     ofpbuf_use_const(&b, buf->data, buf->size);
1462     nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1463     genl = ofpbuf_try_pull(&b, sizeof *genl);
1464     ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1465     if (!nlmsg || !genl || !ovs_header
1466         || nlmsg->nlmsg_type != ovs_datapath_family
1467         || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
1468                             ARRAY_SIZE(ovs_datapath_policy))) {
1469         return EINVAL;
1470     }
1471
1472     dp->cmd = genl->cmd;
1473     dp->dp_ifindex = ovs_header->dp_ifindex;
1474     dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
1475     if (a[OVS_DP_ATTR_STATS]) {
1476         /* Can't use structure assignment because Netlink doesn't ensure
1477          * sufficient alignment for 64-bit members. */
1478         memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
1479                sizeof dp->stats);
1480     }
1481
1482     return 0;
1483 }
1484
1485 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1486 static void
1487 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1488 {
1489     struct ovs_header *ovs_header;
1490
1491     nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
1492                           NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
1493                           OVS_DATAPATH_VERSION);
1494
1495     ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1496     ovs_header->dp_ifindex = dp->dp_ifindex;
1497
1498     if (dp->name) {
1499         nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
1500     }
1501
1502     if (dp->upcall_pid) {
1503         nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
1504     }
1505
1506     /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
1507 }
1508
1509 /* Clears 'dp' to "empty" values. */
1510 static void
1511 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1512 {
1513     memset(dp, 0, sizeof *dp);
1514 }
1515
1516 static void
1517 dpif_linux_dp_dump_start(struct nl_dump *dump)
1518 {
1519     struct dpif_linux_dp request;
1520     struct ofpbuf *buf;
1521
1522     dpif_linux_dp_init(&request);
1523     request.cmd = OVS_DP_CMD_GET;
1524
1525     buf = ofpbuf_new(1024);
1526     dpif_linux_dp_to_ofpbuf(&request, buf);
1527     nl_dump_start(dump, genl_sock, buf);
1528     ofpbuf_delete(buf);
1529 }
1530
1531 /* Executes 'request' in the kernel datapath.  If the command fails, returns a
1532  * positive errno value.  Otherwise, if 'reply' and 'bufp' are null, returns 0
1533  * without doing anything else.  If 'reply' and 'bufp' are nonnull, then the
1534  * result of the command is expected to be of the same form, which is decoded
1535  * and stored in '*reply' and '*bufp'.  The caller must free '*bufp' when the
1536  * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1537 static int
1538 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1539                        struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1540 {
1541     struct ofpbuf *request_buf;
1542     int error;
1543
1544     assert((reply != NULL) == (bufp != NULL));
1545
1546     request_buf = ofpbuf_new(1024);
1547     dpif_linux_dp_to_ofpbuf(request, request_buf);
1548     error = nl_sock_transact(genl_sock, request_buf, bufp);
1549     ofpbuf_delete(request_buf);
1550
1551     if (reply) {
1552         if (!error) {
1553             error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1554         }
1555         if (error) {
1556             dpif_linux_dp_init(reply);
1557             ofpbuf_delete(*bufp);
1558             *bufp = NULL;
1559         }
1560     }
1561     return error;
1562 }
1563
1564 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1565  * The caller must free '*bufp' when the reply is no longer needed ('reply'
1566  * will contain pointers into '*bufp').  */
1567 static int
1568 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1569                   struct ofpbuf **bufp)
1570 {
1571     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1572     struct dpif_linux_dp request;
1573
1574     dpif_linux_dp_init(&request);
1575     request.cmd = OVS_DP_CMD_GET;
1576     request.dp_ifindex = dpif->dp_ifindex;
1577
1578     return dpif_linux_dp_transact(&request, reply, bufp);
1579 }
1580 \f
1581 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1582  * by Netlink attributes, into 'flow'.  Returns 0 if successful, otherwise a
1583  * positive errno value.
1584  *
1585  * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1586  * while 'flow' is still in use. */
1587 static int
1588 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1589                             const struct ofpbuf *buf)
1590 {
1591     static const struct nl_policy ovs_flow_policy[] = {
1592         [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1593         [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1594         [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
1595                                   .optional = true },
1596         [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1597         [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1598         /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
1599     };
1600
1601     struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
1602     struct ovs_header *ovs_header;
1603     struct nlmsghdr *nlmsg;
1604     struct genlmsghdr *genl;
1605     struct ofpbuf b;
1606
1607     dpif_linux_flow_init(flow);
1608
1609     ofpbuf_use_const(&b, buf->data, buf->size);
1610     nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1611     genl = ofpbuf_try_pull(&b, sizeof *genl);
1612     ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1613     if (!nlmsg || !genl || !ovs_header
1614         || nlmsg->nlmsg_type != ovs_flow_family
1615         || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
1616                             ARRAY_SIZE(ovs_flow_policy))) {
1617         return EINVAL;
1618     }
1619
1620     flow->nlmsg_flags = nlmsg->nlmsg_flags;
1621     flow->dp_ifindex = ovs_header->dp_ifindex;
1622     flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
1623     flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
1624     if (a[OVS_FLOW_ATTR_ACTIONS]) {
1625         flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
1626         flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
1627     }
1628     if (a[OVS_FLOW_ATTR_STATS]) {
1629         flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
1630     }
1631     if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
1632         flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
1633     }
1634     if (a[OVS_FLOW_ATTR_USED]) {
1635         flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
1636     }
1637     return 0;
1638 }
1639
1640 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1641  * followed by Netlink attributes corresponding to 'flow'. */
1642 static void
1643 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1644                           struct ofpbuf *buf)
1645 {
1646     struct ovs_header *ovs_header;
1647
1648     nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
1649                           NLM_F_REQUEST | flow->nlmsg_flags,
1650                           flow->cmd, OVS_FLOW_VERSION);
1651
1652     ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1653     ovs_header->dp_ifindex = flow->dp_ifindex;
1654
1655     if (flow->key_len) {
1656         nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
1657     }
1658
1659     if (flow->actions || flow->actions_len) {
1660         nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
1661                           flow->actions, flow->actions_len);
1662     }
1663
1664     /* We never need to send these to the kernel. */
1665     assert(!flow->stats);
1666     assert(!flow->tcp_flags);
1667     assert(!flow->used);
1668
1669     if (flow->clear) {
1670         nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
1671     }
1672 }
1673
1674 /* Clears 'flow' to "empty" values. */
1675 static void
1676 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1677 {
1678     memset(flow, 0, sizeof *flow);
1679 }
1680
1681 /* Executes 'request' in the kernel datapath.  If the command fails, returns a
1682  * positive errno value.  Otherwise, if 'reply' and 'bufp' are null, returns 0
1683  * without doing anything else.  If 'reply' and 'bufp' are nonnull, then the
1684  * result of the command is expected to be a flow also, which is decoded and
1685  * stored in '*reply' and '*bufp'.  The caller must free '*bufp' when the reply
1686  * is no longer needed ('reply' will contain pointers into '*bufp'). */
1687 static int
1688 dpif_linux_flow_transact(struct dpif_linux_flow *request,
1689                          struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1690 {
1691     struct ofpbuf *request_buf;
1692     int error;
1693
1694     assert((reply != NULL) == (bufp != NULL));
1695
1696     if (reply) {
1697         request->nlmsg_flags |= NLM_F_ECHO;
1698     }
1699
1700     request_buf = ofpbuf_new(1024);
1701     dpif_linux_flow_to_ofpbuf(request, request_buf);
1702     error = nl_sock_transact(genl_sock, request_buf, bufp);
1703     ofpbuf_delete(request_buf);
1704
1705     if (reply) {
1706         if (!error) {
1707             error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1708         }
1709         if (error) {
1710             dpif_linux_flow_init(reply);
1711             ofpbuf_delete(*bufp);
1712             *bufp = NULL;
1713         }
1714     }
1715     return error;
1716 }
1717
1718 static void
1719 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1720                           struct dpif_flow_stats *stats)
1721 {
1722     if (flow->stats) {
1723         stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1724         stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1725     } else {
1726         stats->n_packets = 0;
1727         stats->n_bytes = 0;
1728     }
1729     stats->used = flow->used ? get_32aligned_u64(flow->used) : 0;
1730     stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
1731 }