lib/dpif.c

   1 /*
   2  * Copyright (c) 2008, 2009 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 #include "dpif.h"
  19
  20 #include <assert.h>
  21 #include <ctype.h>
  22 #include <errno.h>
  23 #include <fcntl.h>
  24 #include <inttypes.h>
  25 #include <net/if.h>
  26 #include <linux/rtnetlink.h>
  27 #include <linux/ethtool.h>
  28 #include <linux/sockios.h>
  29 #include <netinet/in.h>
  30 #include <stdlib.h>
  31 #include <string.h>
  32 #include <sys/ioctl.h>
  33 #include <sys/stat.h>
  34 #include <sys/sysmacros.h>
  35 #include <unistd.h>
  36
  37 #include "coverage.h"
  38 #include "dynamic-string.h"
  39 #include "flow.h"
  40 #include "netlink.h"
  41 #include "odp-util.h"
  42 #include "ofp-print.h"
  43 #include "ofpbuf.h"
  44 #include "packets.h"
  45 #include "poll-loop.h"
  46 #include "util.h"
  47 #include "valgrind.h"
  48
  49 #include "vlog.h"
  50 #define THIS_MODULE VLM_dpif
  51
  52 /* Rate limit for individual messages going to or from the datapath, output at
  53  * DBG level.  This is very high because, if these are enabled, it is because
  54  * we really need to see them. */
  55 static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600);
  56
  57 /* Not really much point in logging many dpif errors. */
  58 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
  59
  60 static int get_minor_from_name(const char *name, unsigned int *minor);
  61 static int name_to_minor(const char *name, unsigned int *minor);
  62 static int lookup_minor(const char *name, unsigned int *minor);
  63 static int open_by_minor(unsigned int minor, struct dpif *);
  64 static int make_openvswitch_device(unsigned int minor, char **fnp);
  65 static void check_rw_odp_flow(struct odp_flow *);
  66
  67 int
  68 dpif_open(const char *name, struct dpif *dpif)
  69 {
  70     int listen_mask;
  71     int error;
  72
  73     dpif->fd = -1;
  74
  75     error = name_to_minor(name, &dpif->minor);
  76     if (error) {
  77         return error;
  78     }
  79
  80     error = open_by_minor(dpif->minor, dpif);
  81     if (error) {
  82         return error;
  83     }
  84
  85     /* We can open the device, but that doesn't mean that it's been created.
  86      * If it hasn't been, then any command other than ODP_DP_CREATE will
  87      * return ENODEV.  Try something innocuous. */
  88     listen_mask = 0;            /* Make Valgrind happy. */
  89     if (ioctl(dpif->fd, ODP_GET_LISTEN_MASK, &listen_mask)) {
  90         error = errno;
  91         if (error != ENODEV) {
  92             VLOG_WARN("%s: probe returned unexpected error: %s",
  93                       dpif_name(dpif), strerror(error));
  94         }
  95         dpif_close(dpif);
  96         return error;
  97     }
  98     return 0;
  99 }
 100
 101 void
 102 dpif_close(struct dpif *dpif)
 103 {
 104     if (dpif) {
 105         free(dpif->name);
 106         dpif->name = NULL;
 107         close(dpif->fd);
 108         dpif->fd = -1;
 109     }
 110 }
 111
 112 static int
 113 do_ioctl(const struct dpif *dpif, int cmd, const char *cmd_name,
 114          const void *arg)
 115 {
 116     int error = ioctl(dpif->fd, cmd, arg) ? errno : 0;
 117     if (cmd_name) {
 118         if (error) {
 119             VLOG_WARN_RL(&error_rl, "%s: ioctl(%s) failed (%s)",
 120                          dpif_name(dpif), cmd_name, strerror(error));
 121         } else {
 122             VLOG_DBG_RL(&dpmsg_rl, "%s: ioctl(%s): success",
 123                         dpif_name(dpif), cmd_name);
 124         }
 125     }
 126     return error;
 127 }
 128
 129 int
 130 dpif_create(const char *name, struct dpif *dpif)
 131 {
 132     unsigned int minor;
 133     int error;
 134
 135     if (!get_minor_from_name(name, &minor)) {
 136         /* Minor was specified in 'name', go ahead and create it. */
 137         error = open_by_minor(minor, dpif);
 138         if (error) {
 139             return error;
 140         }
 141
 142         if (!strncmp(name, "nl:", 3)) {
 143             char devname[128];
 144             sprintf(devname, "of%u", minor);
 145             error = ioctl(dpif->fd, ODP_DP_CREATE, devname) < 0 ? errno : 0;
 146         } else {
 147             error = ioctl(dpif->fd, ODP_DP_CREATE, name) < 0 ? errno : 0;
 148         }
 149         if (error) {
 150             dpif_close(dpif);
 151         }
 152         return error;
 153     } else {
 154         for (minor = 0; minor < ODP_MAX; minor++) {
 155             error = open_by_minor(minor, dpif);
 156             if (error) {
 157                 return error;
 158             }
 159
 160             error = ioctl(dpif->fd, ODP_DP_CREATE, name) < 0 ? errno : 0;
 161             if (!error) {
 162                 return 0;
 163             }
 164             dpif_close(dpif);
 165             if (error != EBUSY) {
 166                 return error;
 167             }
 168         }
 169         return ENOBUFS;
 170     }
 171 }
 172
 173 const char *
 174 dpif_name(const struct dpif *dpif)
 175 {
 176     return dpif->name;
 177 }
 178
 179 int
 180 dpif_delete(struct dpif *dpif)
 181 {
 182     COVERAGE_INC(dpif_destroy);
 183     return do_ioctl(dpif, ODP_DP_DESTROY, "ODP_DP_DESTROY", NULL);
 184 }
 185
 186 int
 187 dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats)
 188 {
 189     memset(stats, 0, sizeof *stats);
 190     return do_ioctl(dpif, ODP_DP_STATS, "ODP_DP_STATS", stats);
 191 }
 192
 193 int
 194 dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags)
 195 {
 196     int tmp;
 197     int error = do_ioctl(dpif, ODP_GET_DROP_FRAGS, "ODP_GET_DROP_FRAGS", &tmp);
 198     *drop_frags = error ? tmp & 1 : false;
 199     return error;
 200 }
 201
 202 int
 203 dpif_set_drop_frags(struct dpif *dpif, bool drop_frags)
 204 {
 205     int tmp = drop_frags;
 206     return do_ioctl(dpif, ODP_SET_DROP_FRAGS, "ODP_SET_DROP_FRAGS", &tmp);
 207 }
 208
 209 int
 210 dpif_get_listen_mask(const struct dpif *dpif, int *listen_mask)
 211 {
 212     int error = do_ioctl(dpif, ODP_GET_LISTEN_MASK, "ODP_GET_LISTEN_MASK",
 213                          listen_mask);
 214     if (error) {
 215         *listen_mask = 0;
 216     }
 217     return error;
 218 }
 219
 220 int
 221 dpif_set_listen_mask(struct dpif *dpif, int listen_mask)
 222 {
 223     return do_ioctl(dpif, ODP_SET_LISTEN_MASK, "ODP_SET_LISTEN_MASK",
 224                     &listen_mask);
 225 }
 226
 227 int
 228 dpif_purge(struct dpif *dpif)
 229 {
 230     struct odp_stats stats;
 231     unsigned int i;
 232     int error;
 233
 234     COVERAGE_INC(dpif_purge);
 235
 236     error = dpif_get_dp_stats(dpif, &stats);
 237     if (error) {
 238         return error;
 239     }
 240
 241     for (i = 0; i < stats.max_miss_queue + stats.max_action_queue; i++) {
 242         struct ofpbuf *buf;
 243         error = dpif_recv(dpif, &buf);
 244         if (error) {
 245             return error == EAGAIN ? 0 : error;
 246         }
 247         ofpbuf_delete(buf);
 248     }
 249     return 0;
 250 }
 251
 252 int
 253 dpif_port_add(struct dpif *dpif, const char *devname, uint16_t port_no,
 254               uint16_t flags)
 255 {
 256     struct odp_port port;
 257
 258     COVERAGE_INC(dpif_port_add);
 259     memset(&port, 0, sizeof port);
 260     strncpy(port.devname, devname, sizeof port.devname);
 261     port.port = port_no;
 262     port.flags = flags;
 263     if (!ioctl(dpif->fd, ODP_PORT_ADD, &port)) {
 264         VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu16,
 265                     dpif_name(dpif), devname, port_no);
 266         return 0;
 267     } else {
 268         VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port %"PRIu16": %s",
 269                      dpif_name(dpif), devname, port_no, strerror(errno));
 270         return errno;
 271     }
 272 }
 273
 274 int
 275 dpif_port_del(struct dpif *dpif, uint16_t port_no)
 276 {
 277     int tmp = port_no;
 278     COVERAGE_INC(dpif_port_del);
 279     return do_ioctl(dpif, ODP_PORT_DEL, "ODP_PORT_DEL", &tmp);
 280 }
 281
 282 int
 283 dpif_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
 284                           struct odp_port *port)
 285 {
 286     memset(port, 0, sizeof *port);
 287     port->port = port_no;
 288     if (!ioctl(dpif->fd, ODP_PORT_QUERY, port)) {
 289         VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu16" is device %s",
 290                     dpif_name(dpif), port_no, port->devname);
 291         return 0;
 292     } else {
 293         VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu16": %s",
 294                      dpif_name(dpif), port_no, strerror(errno));
 295         return errno;
 296     }
 297 }
 298
 299 int
 300 dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
 301                         struct odp_port *port)
 302 {
 303     memset(port, 0, sizeof *port);
 304     strncpy(port->devname, devname, sizeof port->devname);
 305     if (!ioctl(dpif->fd, ODP_PORT_QUERY, port)) {
 306         VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu16,
 307                     dpif_name(dpif), devname, port->port);
 308         return 0;
 309     } else {
 310         /* Log level is DBG here because all the current callers are interested
 311          * in whether 'dpif' actually has a port 'devname', so that it's not an
 312          * issue worth logging if it doesn't. */
 313         VLOG_DBG_RL(&error_rl, "%s: failed to query port %s: %s",
 314                     dpif_name(dpif), devname, strerror(errno));
 315         return errno;
 316     }
 317 }
 318
 319 int
 320 dpif_port_get_name(struct dpif *dpif, uint16_t port_no,
 321                    char *name, size_t name_size)
 322 {
 323     struct odp_port port;
 324     int error;
 325
 326     assert(name_size > 0);
 327
 328     error = dpif_port_query_by_number(dpif, port_no, &port);
 329     if (!error) {
 330         ovs_strlcpy(name, port.devname, name_size);
 331     } else {
 332         *name = '\0';
 333     }
 334     return error;
 335 }
 336
 337 int
 338 dpif_port_list(const struct dpif *dpif,
 339                struct odp_port **ports, size_t *n_ports)
 340 {
 341     struct odp_portvec pv;
 342     struct odp_stats stats;
 343     int error;
 344
 345     do {
 346         error = dpif_get_dp_stats(dpif, &stats);
 347         if (error) {
 348             goto error;
 349         }
 350
 351         *ports = xcalloc(1, stats.n_ports * sizeof **ports);
 352         pv.ports = *ports;
 353         pv.n_ports = stats.n_ports;
 354         error = do_ioctl(dpif, ODP_PORT_LIST, "ODP_PORT_LIST", &pv);
 355         if (error) {
 356             free(*ports);
 357             goto error;
 358         }
 359     } while (pv.n_ports != stats.n_ports);
 360     *n_ports = pv.n_ports;
 361     return 0;
 362
 363 error:
 364     *ports = NULL;
 365     *n_ports = 0;
 366     return error;
 367 }
 368
 369 int
 370 dpif_port_group_set(struct dpif *dpif, uint16_t group,
 371                     const uint16_t ports[], size_t n_ports)
 372 {
 373     struct odp_port_group pg;
 374
 375     COVERAGE_INC(dpif_port_group_set);
 376     assert(n_ports <= UINT16_MAX);
 377     pg.group = group;
 378     pg.ports = (uint16_t *) ports;
 379     pg.n_ports = n_ports;
 380     return do_ioctl(dpif, ODP_PORT_GROUP_SET, "ODP_PORT_GROUP_SET", &pg);
 381 }
 382
 383 /* Careful: '*n_out' can be greater than 'n_ports' on return, if 'n_ports' is
 384  * less than the number of ports in 'group'. */
 385 int
 386 dpif_port_group_get(const struct dpif *dpif, uint16_t group,
 387                     uint16_t ports[], size_t n_ports, size_t *n_out)
 388 {
 389     struct odp_port_group pg;
 390     int error;
 391
 392     assert(n_ports <= UINT16_MAX);
 393     pg.group = group;
 394     pg.ports = ports;
 395     pg.n_ports = n_ports;
 396     error = do_ioctl(dpif, ODP_PORT_GROUP_GET, "ODP_PORT_GROUP_GET", &pg);
 397     *n_out = error ? 0 : pg.n_ports;
 398     return error;
 399 }
 400
 401 int
 402 dpif_flow_flush(struct dpif *dpif)
 403 {
 404     COVERAGE_INC(dpif_flow_flush);
 405     return do_ioctl(dpif, ODP_FLOW_FLUSH, "ODP_FLOW_FLUSH", NULL);
 406 }
 407
 408 static enum vlog_level
 409 flow_message_log_level(int error)
 410 {
 411     return error ? VLL_WARN : VLL_DBG;
 412 }
 413
 414 static bool
 415 should_log_flow_message(int error)
 416 {
 417     return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
 418                              error ? &error_rl : &dpmsg_rl);
 419 }
 420
 421 static void
 422 log_flow_message(const struct dpif *dpif, int error,
 423                  const char *operation,
 424                  const flow_t *flow, const struct odp_flow_stats *stats,
 425                  const union odp_action *actions, size_t n_actions)
 426 {
 427     struct ds ds = DS_EMPTY_INITIALIZER;
 428     ds_put_format(&ds, "%s: ", dpif_name(dpif));
 429     if (error) {
 430         ds_put_cstr(&ds, "failed to ");
 431     }
 432     ds_put_format(&ds, "%s ", operation);
 433     if (error) {
 434         ds_put_format(&ds, "(%s) ", strerror(error));
 435     }
 436     flow_format(&ds, flow);
 437     if (stats) {
 438         ds_put_cstr(&ds, ", ");
 439         format_odp_flow_stats(&ds, stats);
 440     }
 441     if (actions || n_actions) {
 442         ds_put_cstr(&ds, ", actions:");
 443         format_odp_actions(&ds, actions, n_actions);
 444     }
 445     vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
 446     ds_destroy(&ds);
 447 }
 448
 449 static int
 450 do_flow_ioctl(const struct dpif *dpif, int cmd, struct odp_flow *flow,
 451               const char *operation, bool show_stats)
 452 {
 453     int error = do_ioctl(dpif, cmd, NULL, flow);
 454     if (error && show_stats) {
 455         flow->n_actions = 0;
 456     }
 457     if (should_log_flow_message(error)) {
 458         log_flow_message(dpif, error, operation, &flow->key,
 459                          show_stats && !error ? &flow->stats : NULL,
 460                          flow->actions, flow->n_actions);
 461     }
 462     return error;
 463 }
 464
 465 int
 466 dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put)
 467 {
 468     int error = do_ioctl(dpif, ODP_FLOW_PUT, NULL, put);
 469     COVERAGE_INC(dpif_flow_put);
 470     if (should_log_flow_message(error)) {
 471         struct ds operation = DS_EMPTY_INITIALIZER;
 472         ds_put_cstr(&operation, "put");
 473         if (put->flags & ODPPF_CREATE) {
 474             ds_put_cstr(&operation, "[create]");
 475         }
 476         if (put->flags & ODPPF_MODIFY) {
 477             ds_put_cstr(&operation, "[modify]");
 478         }
 479         if (put->flags & ODPPF_ZERO_STATS) {
 480             ds_put_cstr(&operation, "[zero]");
 481         }
 482 #define ODPPF_ALL (ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS)
 483         if (put->flags & ~ODPPF_ALL) {
 484             ds_put_format(&operation, "[%x]", put->flags & ~ODPPF_ALL);
 485         }
 486         log_flow_message(dpif, error, ds_cstr(&operation), &put->flow.key,
 487                          !error ? &put->flow.stats : NULL,
 488                          put->flow.actions, put->flow.n_actions);
 489         ds_destroy(&operation);
 490     }
 491     return error;
 492 }
 493
 494 int
 495 dpif_flow_del(struct dpif *dpif, struct odp_flow *flow)
 496 {
 497     COVERAGE_INC(dpif_flow_del);
 498     check_rw_odp_flow(flow);
 499     memset(&flow->stats, 0, sizeof flow->stats);
 500     return do_flow_ioctl(dpif, ODP_FLOW_DEL, flow, "delete flow", true);
 501 }
 502
 503 int
 504 dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow)
 505 {
 506     COVERAGE_INC(dpif_flow_query);
 507     check_rw_odp_flow(flow);
 508     memset(&flow->stats, 0, sizeof flow->stats);
 509     return do_flow_ioctl(dpif, ODP_FLOW_GET, flow, "get flow", true);
 510 }
 511
 512 int
 513 dpif_flow_get_multiple(const struct dpif *dpif,
 514                        struct odp_flow flows[], size_t n)
 515 {
 516     struct odp_flowvec fv;
 517     size_t i;
 518
 519     COVERAGE_ADD(dpif_flow_query_multiple, n);
 520     fv.flows = flows;
 521     fv.n_flows = n;
 522     for (i = 0; i < n; i++) {
 523         check_rw_odp_flow(&flows[i]);
 524     }
 525     return do_ioctl(dpif, ODP_FLOW_GET_MULTIPLE, "ODP_FLOW_GET_MULTIPLE",
 526                     &fv);
 527 }
 528
 529 int
 530 dpif_flow_list(const struct dpif *dpif, struct odp_flow flows[], size_t n,
 531                size_t *n_out)
 532 {
 533     struct odp_flowvec fv;
 534     uint32_t i;
 535     int error;
 536
 537     COVERAGE_INC(dpif_flow_query_list);
 538     fv.flows = flows;
 539     fv.n_flows = n;
 540     if (RUNNING_ON_VALGRIND) {
 541         memset(flows, 0, n * sizeof *flows);
 542     } else {
 543         for (i = 0; i < n; i++) {
 544             flows[i].actions = NULL;
 545             flows[i].n_actions = 0;
 546         }
 547     }
 548     error = do_ioctl(dpif, ODP_FLOW_LIST, NULL, &fv);
 549     if (error) {
 550         *n_out = 0;
 551         VLOG_WARN_RL(&error_rl, "%s: flow list failed (%s)",
 552                      dpif_name(dpif), strerror(error));
 553     } else {
 554         COVERAGE_ADD(dpif_flow_query_list_n, fv.n_flows);
 555         *n_out = fv.n_flows;
 556         VLOG_DBG_RL(&dpmsg_rl, "%s: listed %zu flows",
 557                     dpif_name(dpif), *n_out);
 558     }
 559     return error;
 560 }
 561
 562 int
 563 dpif_flow_list_all(const struct dpif *dpif,
 564                    struct odp_flow **flowsp, size_t *np)
 565 {
 566     struct odp_stats stats;
 567     struct odp_flow *flows;
 568     size_t n_flows;
 569     int error;
 570
 571     *flowsp = NULL;
 572     *np = 0;
 573
 574     error = dpif_get_dp_stats(dpif, &stats);
 575     if (error) {
 576         return error;
 577     }
 578
 579     flows = xmalloc(sizeof *flows * stats.n_flows);
 580     error = dpif_flow_list(dpif, flows, stats.n_flows, &n_flows);
 581     if (error) {
 582         free(flows);
 583         return error;
 584     }
 585
 586     if (stats.n_flows != n_flows) {
 587         VLOG_WARN_RL(&error_rl, "%s: datapath stats reported %"PRIu32" "
 588                      "flows but flow listing reported %zu",
 589                      dpif_name(dpif), stats.n_flows, n_flows);
 590     }
 591     *flowsp = flows;
 592     *np = n_flows;
 593     return 0;
 594 }
 595
 596 int
 597 dpif_execute(struct dpif *dpif, uint16_t in_port,
 598              const union odp_action actions[], size_t n_actions,
 599              const struct ofpbuf *buf)
 600 {
 601     int error;
 602
 603     COVERAGE_INC(dpif_execute);
 604     if (n_actions > 0) {
 605         struct odp_execute execute;
 606         memset(&execute, 0, sizeof execute);
 607         execute.in_port = in_port;
 608         execute.actions = (union odp_action *) actions;
 609         execute.n_actions = n_actions;
 610         execute.data = buf->data;
 611         execute.length = buf->size;
 612         error = do_ioctl(dpif, ODP_EXECUTE, NULL, &execute);
 613     } else {
 614         error = 0;
 615     }
 616
 617     if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
 618         struct ds ds = DS_EMPTY_INITIALIZER;
 619         char *packet = ofp_packet_to_string(buf->data, buf->size, buf->size);
 620         ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
 621         format_odp_actions(&ds, actions, n_actions);
 622         if (error) {
 623             ds_put_format(&ds, " failed (%s)", strerror(error));
 624         }
 625         ds_put_format(&ds, " on packet %s", packet);
 626         vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
 627         ds_destroy(&ds);
 628         free(packet);
 629     }
 630     return error;
 631 }
 632
 633 int
 634 dpif_recv(struct dpif *dpif, struct ofpbuf **bufp)
 635 {
 636     struct ofpbuf *buf;
 637     int retval;
 638     int error;
 639
 640     buf = ofpbuf_new(65536);
 641     retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf));
 642     if (retval < 0) {
 643         error = errno;
 644         if (error != EAGAIN) {
 645             VLOG_WARN_RL(&error_rl, "%s: read failed: %s",
 646                          dpif_name(dpif), strerror(error));
 647         }
 648     } else if (retval >= sizeof(struct odp_msg)) {
 649         struct odp_msg *msg = buf->data;
 650         if (msg->length <= retval) {
 651             buf->size += retval;
 652             if (VLOG_IS_DBG_ENABLED()) {
 653                 void *payload = msg + 1;
 654                 size_t length = buf->size - sizeof *msg;
 655                 char *s = ofp_packet_to_string(payload, length, length);
 656                 VLOG_DBG_RL(&dpmsg_rl, "%s: received %s message of length "
 657                             "%zu on port %"PRIu16": %s", dpif_name(dpif),
 658                             (msg->type == _ODPL_MISS_NR ? "miss"
 659                              : msg->type == _ODPL_ACTION_NR ? "action"
 660                              : "<unknown>"),
 661                             msg->length - sizeof(struct odp_msg),
 662                             msg->port, s);
 663                 free(s);
 664             }
 665             *bufp = buf;
 666             COVERAGE_INC(dpif_recv);
 667             return 0;
 668         } else {
 669             VLOG_WARN_RL(&error_rl, "%s: discarding message truncated "
 670                          "from %zu bytes to %d",
 671                          dpif_name(dpif), msg->length, retval);
 672             error = ERANGE;
 673         }
 674     } else if (!retval) {
 675         VLOG_WARN_RL(&error_rl, "%s: unexpected end of file", dpif_name(dpif));
 676         error = EPROTO;
 677     } else {
 678         VLOG_WARN_RL(&error_rl,
 679                      "%s: discarding too-short message (%d bytes)",
 680                      dpif_name(dpif), retval);
 681         error = ERANGE;
 682     }
 683
 684     *bufp = NULL;
 685     ofpbuf_delete(buf);
 686     return error;
 687 }
 688
 689 void
 690 dpif_recv_wait(struct dpif *dpif)
 691 {
 692     poll_fd_wait(dpif->fd, POLLIN);
 693 }
 694
 695 void
 696 dpif_get_netflow_ids(const struct dpif *dpif,
 697                      uint8_t *engine_type, uint8_t *engine_id)
 698 {
 699     *engine_type = *engine_id = dpif->minor;
 700 }
 701 \f
 702 struct dpifmon {
 703     struct dpif dpif;
 704     struct nl_sock *sock;
 705     int local_ifindex;
 706 };
 707
 708 int
 709 dpifmon_create(const char *datapath_name, struct dpifmon **monp)
 710 {
 711     struct dpifmon *mon;
 712     char local_name[IFNAMSIZ];
 713     int error;
 714
 715     mon = *monp = xmalloc(sizeof *mon);
 716
 717     error = dpif_open(datapath_name, &mon->dpif);
 718     if (error) {
 719         goto error;
 720     }
 721     error = dpif_port_get_name(&mon->dpif, ODPP_LOCAL,
 722                                local_name, sizeof local_name);
 723     if (error) {
 724         goto error_close_dpif;
 725     }
 726
 727     mon->local_ifindex = if_nametoindex(local_name);
 728     if (!mon->local_ifindex) {
 729         error = errno;
 730         VLOG_WARN("could not get ifindex of %s device: %s",
 731                   local_name, strerror(errno));
 732         goto error_close_dpif;
 733     }
 734
 735     error = nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0, &mon->sock);
 736     if (error) {
 737         VLOG_WARN("could not create rtnetlink socket: %s", strerror(error));
 738         goto error_close_dpif;
 739     }
 740
 741     return 0;
 742
 743 error_close_dpif:
 744     dpif_close(&mon->dpif);
 745 error:
 746     free(mon);
 747     *monp = NULL;
 748     return error;
 749 }
 750
 751 void
 752 dpifmon_destroy(struct dpifmon *mon)
 753 {
 754     if (mon) {
 755         dpif_close(&mon->dpif);
 756         nl_sock_destroy(mon->sock);
 757     }
 758 }
 759
 760 int
 761 dpifmon_poll(struct dpifmon *mon, char **devnamep)
 762 {
 763     static struct vlog_rate_limit slow_rl = VLOG_RATE_LIMIT_INIT(1, 5);
 764     static const struct nl_policy rtnlgrp_link_policy[] = {
 765         [IFLA_IFNAME] = { .type = NL_A_STRING },
 766         [IFLA_MASTER] = { .type = NL_A_U32, .optional = true },
 767     };
 768     struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
 769     struct ofpbuf *buf;
 770     int error;
 771
 772     *devnamep = NULL;
 773 again:
 774     error = nl_sock_recv(mon->sock, &buf, false);
 775     switch (error) {
 776     case 0:
 777         if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
 778                              rtnlgrp_link_policy,
 779                              attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
 780             VLOG_WARN_RL(&slow_rl, "received bad rtnl message");
 781             error = ENOBUFS;
 782         } else {
 783             const char *devname = nl_attr_get_string(attrs[IFLA_IFNAME]);
 784             bool for_us;
 785
 786             if (attrs[IFLA_MASTER]) {
 787                 uint32_t master_ifindex = nl_attr_get_u32(attrs[IFLA_MASTER]);
 788                 for_us = master_ifindex == mon->local_ifindex;
 789             } else {
 790                 /* It's for us if that device is one of our ports. */
 791                 struct odp_port port;
 792                 for_us = !dpif_port_query_by_name(mon->dpif, devname, &port);
 793             }
 794
 795             if (!for_us) {
 796                 /* Not for us, try again. */
 797                 ofpbuf_delete(buf);
 798                 COVERAGE_INC(dpifmon_poll_false_wakeup);
 799                 goto again;
 800             }
 801             COVERAGE_INC(dpifmon_poll_changed);
 802             *devnamep = xstrdup(devname);
 803         }
 804         ofpbuf_delete(buf);
 805         break;
 806
 807     case EAGAIN:
 808         /* Nothing to do. */
 809         break;
 810
 811     case ENOBUFS:
 812         VLOG_WARN_RL(&slow_rl, "dpifmon socket overflowed");
 813         break;
 814
 815     default:
 816         VLOG_WARN_RL(&slow_rl, "error on dpifmon socket: %s", strerror(error));
 817         break;
 818     }
 819     return error;
 820 }
 821
 822 void
 823 dpifmon_run(struct dpifmon *mon UNUSED)
 824 {
 825     /* Nothing to do in this implementation. */
 826 }
 827
 828 void
 829 dpifmon_wait(struct dpifmon *mon)
 830 {
 831     nl_sock_wait(mon->sock, POLLIN);
 832 }
 833 \f
 834 static int get_openvswitch_major(void);
 835 static int get_major(const char *target, int default_major);
 836
 837 static int
 838 lookup_minor(const char *name, unsigned int *minor)
 839 {
 840     struct ethtool_drvinfo drvinfo;
 841     struct ifreq ifr;
 842     int error;
 843     int sock;
 844
 845     *minor = -1;
 846     sock = socket(AF_INET, SOCK_DGRAM, 0);
 847     if (sock < 0) {
 848         VLOG_WARN("socket(AF_INET) failed: %s", strerror(errno));
 849         error = errno;
 850         goto error;
 851     }
 852
 853     memset(&ifr, 0, sizeof ifr);
 854     strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
 855     ifr.ifr_data = (caddr_t) &drvinfo;
 856
 857     memset(&drvinfo, 0, sizeof drvinfo);
 858     drvinfo.cmd = ETHTOOL_GDRVINFO;
 859     if (ioctl(sock, SIOCETHTOOL, &ifr)) {
 860         VLOG_WARN("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
 861         error = errno;
 862         goto error_close_sock;
 863     }
 864
 865     if (strcmp(drvinfo.driver, "openvswitch")) {
 866         VLOG_WARN("%s is not an openvswitch device", name);
 867         error = EOPNOTSUPP;
 868         goto error_close_sock;
 869     }
 870
 871     if (!isdigit(drvinfo.bus_info[0])) {
 872         VLOG_WARN("%s ethtool info does not contain an openvswitch minor",
 873                   name);
 874         error = EPROTOTYPE;
 875         goto error_close_sock;
 876     }
 877
 878     *minor = atoi(drvinfo.bus_info);
 879     close(sock);
 880     return 0;
 881
 882 error_close_sock:
 883     close(sock);
 884 error:
 885     return error;
 886 }
 887
 888 static int
 889 make_openvswitch_device(unsigned int minor, char **fnp)
 890 {
 891     dev_t dev = makedev(get_openvswitch_major(), minor);
 892     const char dirname[] = "/dev/net";
 893     struct stat s;
 894     char fn[128];
 895
 896     *fnp = NULL;
 897     sprintf(fn, "%s/dp%d", dirname, minor);
 898     if (!stat(fn, &s)) {
 899         if (!S_ISCHR(s.st_mode)) {
 900             VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing",
 901                          fn);
 902         } else if (s.st_rdev != dev) {
 903             VLOG_WARN_RL(&error_rl,
 904                          "%s is device %u:%u instead of %u:%u, fixing",
 905                          fn, major(s.st_rdev), minor(s.st_rdev),
 906                          major(dev), minor(dev));
 907         } else {
 908             goto success;
 909         }
 910         if (unlink(fn)) {
 911             VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)",
 912                          fn, strerror(errno));
 913             return errno;
 914         }
 915     } else if (errno == ENOENT) {
 916         if (stat(dirname, &s)) {
 917             if (errno == ENOENT) {
 918                 if (mkdir(dirname, 0755)) {
 919                     VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)",
 920                                  dirname, strerror(errno));
 921                     return errno;
 922                 }
 923             } else {
 924                 VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)",
 925                              dirname, strerror(errno));
 926                 return errno;
 927             }
 928         }
 929     } else {
 930         VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno));
 931         return errno;
 932     }
 933
 934     /* The device needs to be created. */
 935     if (mknod(fn, S_IFCHR | 0700, dev)) {
 936         VLOG_WARN_RL(&error_rl,
 937                      "%s: creating character device %u:%u failed (%s)",
 938                      fn, major(dev), minor(dev), strerror(errno));
 939         return errno;
 940     }
 941
 942 success:
 943     *fnp = xstrdup(fn);
 944     return 0;
 945 }
 946
 947
 948 static int
 949 get_openvswitch_major(void)
 950 {
 951     static unsigned int openvswitch_major;
 952     if (!openvswitch_major) {
 953         enum { DEFAULT_MAJOR = 248 };
 954         openvswitch_major = get_major("openvswitch", DEFAULT_MAJOR);
 955     }
 956     return openvswitch_major;
 957 }
 958
 959 static int
 960 get_major(const char *target, int default_major)
 961 {
 962     const char fn[] = "/proc/devices";
 963     char line[128];
 964     FILE *file;
 965     int ln;
 966
 967     file = fopen(fn, "r");
 968     if (!file) {
 969         VLOG_ERR("opening %s failed (%s)", fn, strerror(errno));
 970         goto error;
 971     }
 972
 973     for (ln = 1; fgets(line, sizeof line, file); ln++) {
 974         char name[64];
 975         int major;
 976
 977         if (!strncmp(line, "Character", 9) || line[0] == '\0') {
 978             /* Nothing to do. */
 979         } else if (!strncmp(line, "Block", 5)) {
 980             /* We only want character devices, so skip the rest of the file. */
 981             break;
 982         } else if (sscanf(line, "%d %63s", &major, name)) {
 983             if (!strcmp(name, target)) {
 984                 fclose(file);
 985                 return major;
 986             }
 987         } else {
 988             static bool warned;
 989             if (!warned) {
 990                 VLOG_WARN("%s:%d: syntax error", fn, ln);
 991             }
 992             warned = true;
 993         }
 994     }
 995
 996     VLOG_ERR("%s: %s major not found (is the module loaded?), using "
 997              "default major %d", fn, target, default_major);
 998 error:
 999     VLOG_INFO("using default major %d for %s", default_major, target);
1000     return default_major;
1001 }
1002
1003 static int
1004 name_to_minor(const char *name, unsigned int *minor)
1005 {
1006     if (!get_minor_from_name(name, minor)) {
1007         return 0;
1008     }
1009     return lookup_minor(name, minor);
1010 }
1011
1012 static int
1013 get_minor_from_name(const char *name, unsigned int *minor)
1014 {
1015     if (!strncmp(name, "dp", 2) && isdigit(name[2])) {
1016         *minor = atoi(name + 2);
1017         return 0;
1018     } else if (!strncmp(name, "nl:", 3) && isdigit(name[3])) {
1019         /* This is for compatibility only and will be dropped. */
1020         *minor = atoi(name + 3);
1021         return 0;
1022     } else {
1023         return EINVAL;
1024     }
1025 }
1026
1027 static int
1028 open_by_minor(unsigned int minor, struct dpif *dpif)
1029 {
1030     int error;
1031     char *fn;
1032     int fd;
1033
1034     dpif->minor = -1;
1035     dpif->fd = -1;
1036     error = make_openvswitch_device(minor, &fn);
1037     if (error) {
1038         return error;
1039     }
1040
1041     fd = open(fn, O_RDONLY | O_NONBLOCK);
1042     if (fd < 0) {
1043         error = errno;
1044         VLOG_WARN("%s: open failed (%s)", fn, strerror(error));
1045         free(fn);
1046         return error;
1047     }
1048
1049     free(fn);
1050     dpif->name = xasprintf("dp%u", dpif->minor);
1051     dpif->minor = minor;
1052     dpif->fd = fd;
1053     return 0;
1054 }
1055 \f
1056 /* There is a tendency to construct odp_flow objects on the stack and to
1057  * forget to properly initialize their "actions" and "n_actions" members.
1058  * When this happens, we get memory corruption because the kernel
1059  * writes through the random pointer that is in the "actions" member.
1060  *
1061  * This function attempts to combat the problem by:
1062  *
1063  *      - Forcing a segfault if "actions" points to an invalid region (instead
1064  *        of just getting back EFAULT, which can be easily missed in the log).
1065  *
1066  *      - Storing a distinctive value that is likely to cause an
1067  *        easy-to-identify error later if it is dereferenced, etc.
1068  *
1069  *      - Triggering a warning on uninitialized memory from Valgrind if
1070  *        "actions" or "n_actions" was not initialized.
1071  */
1072 static void
1073 check_rw_odp_flow(struct odp_flow *flow)
1074 {
1075     if (flow->n_actions) {
1076         memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]);
1077     }
1078 }