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 "svec.h"
  47 #include "util.h"
  48 #include "valgrind.h"
  49
  50 #include "vlog.h"
  51 #define THIS_MODULE VLM_dpif
  52
  53 /* Rate limit for individual messages going to or from the datapath, output at
  54  * DBG level.  This is very high because, if these are enabled, it is because
  55  * we really need to see them. */
  56 static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600);
  57
  58 /* Not really much point in logging many dpif errors. */
  59 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
  60
  61 static int get_minor_from_name(const char *name, unsigned int *minor);
  62 static int name_to_minor(const char *name, unsigned int *minor);
  63 static int lookup_minor(const char *name, unsigned int *minor);
  64 static int open_by_minor(unsigned int minor, struct dpif *);
  65 static int make_openvswitch_device(unsigned int minor, char **fnp);
  66 static void check_rw_odp_flow(struct odp_flow *);
  67
  68
  69 /* Clears 'all_dps' and enumerates the names of all known created
  70  * datapaths into it.  Returns 0 if successful, otherwise a positive
  71  * errno value. */
  72 int
  73 dp_enumerate(struct svec *all_dps)
  74 {
  75     int error;
  76     int i;
  77
  78     svec_clear(all_dps);
  79     error = 0;
  80     for (i = 0; i < ODP_MAX; i++) {
  81         struct dpif dpif;
  82         char devname[16];
  83         int retval;
  84
  85         sprintf(devname, "dp%d", i);
  86         retval = dpif_open(devname, &dpif);
  87         if (!retval) {
  88             svec_add(all_dps, devname);
  89             dpif_close(&dpif);
  90         } else if (retval != ENODEV && !error) {
  91             error = retval;
  92         }
  93     }
  94     return error;
  95 }
  96
  97 int
  98 dpif_open(const char *name, struct dpif *dpif)
  99 {
 100     int listen_mask;
 101     int error;
 102
 103     dpif->fd = -1;
 104
 105     error = name_to_minor(name, &dpif->minor);
 106     if (error) {
 107         return error;
 108     }
 109
 110     error = open_by_minor(dpif->minor, dpif);
 111     if (error) {
 112         return error;
 113     }
 114
 115     /* We can open the device, but that doesn't mean that it's been created.
 116      * If it hasn't been, then any command other than ODP_DP_CREATE will
 117      * return ENODEV.  Try something innocuous. */
 118     listen_mask = 0;            /* Make Valgrind happy. */
 119     if (ioctl(dpif->fd, ODP_GET_LISTEN_MASK, &listen_mask)) {
 120         error = errno;
 121         if (error != ENODEV) {
 122             VLOG_WARN("dp%u: probe returned unexpected error: %s",
 123                       dpif->minor, strerror(error));
 124         }
 125         dpif_close(dpif);
 126         return error;
 127     }
 128     return 0;
 129 }
 130
 131 void
 132 dpif_close(struct dpif *dpif)
 133 {
 134     if (dpif) {
 135         close(dpif->fd);
 136         dpif->fd = -1;
 137     }
 138 }
 139
 140 static int
 141 do_ioctl(const struct dpif *dpif, int cmd, const char *cmd_name,
 142          const void *arg)
 143 {
 144     int error = ioctl(dpif->fd, cmd, arg) ? errno : 0;
 145     if (cmd_name) {
 146         if (error) {
 147             VLOG_WARN_RL(&error_rl, "dp%u: ioctl(%s) failed (%s)",
 148                          dpif->minor, cmd_name, strerror(error));
 149         } else {
 150             VLOG_DBG_RL(&dpmsg_rl, "dp%u: ioctl(%s): success",
 151                         dpif->minor, cmd_name);
 152         }
 153     }
 154     return error;
 155 }
 156
 157 int
 158 dpif_create(const char *name, struct dpif *dpif)
 159 {
 160     unsigned int minor;
 161     int error;
 162
 163     if (!get_minor_from_name(name, &minor)) {
 164         /* Minor was specified in 'name', go ahead and create it. */
 165         error = open_by_minor(minor, dpif);
 166         if (error) {
 167             return error;
 168         }
 169
 170         if (!strncmp(name, "nl:", 3)) {
 171             char devname[128];
 172             sprintf(devname, "of%u", minor);
 173             error = ioctl(dpif->fd, ODP_DP_CREATE, devname) < 0 ? errno : 0;
 174         } else {
 175             error = ioctl(dpif->fd, ODP_DP_CREATE, name) < 0 ? errno : 0;
 176         }
 177         if (error) {
 178             dpif_close(dpif);
 179         }
 180         return error;
 181     } else {
 182         for (minor = 0; minor < ODP_MAX; minor++) {
 183             error = open_by_minor(minor, dpif);
 184             if (error) {
 185                 return error;
 186             }
 187
 188             error = ioctl(dpif->fd, ODP_DP_CREATE, name) < 0 ? errno : 0;
 189             if (!error) {
 190                 return 0;
 191             }
 192             dpif_close(dpif);
 193             if (error != EBUSY) {
 194                 return error;
 195             }
 196         }
 197         return ENOBUFS;
 198     }
 199 }
 200
 201 int
 202 dpif_get_name(struct dpif *dpif, char *name, size_t name_size)
 203 {
 204     struct odp_port port;
 205     int error;
 206
 207     assert(name_size > 0);
 208     *name = '\0';
 209
 210     error = dpif_port_query_by_number(dpif, ODPP_LOCAL, &port);
 211     if (!error) {
 212         ovs_strlcpy(name, port.devname, name_size);
 213     }
 214     return error;
 215 }
 216
 217 int
 218 dpif_delete(struct dpif *dpif)
 219 {
 220     COVERAGE_INC(dpif_destroy);
 221     return do_ioctl(dpif, ODP_DP_DESTROY, "ODP_DP_DESTROY", NULL);
 222 }
 223
 224 int
 225 dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats)
 226 {
 227     memset(stats, 0, sizeof *stats);
 228     return do_ioctl(dpif, ODP_DP_STATS, "ODP_DP_STATS", stats);
 229 }
 230
 231 int
 232 dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags)
 233 {
 234     int tmp;
 235     int error = do_ioctl(dpif, ODP_GET_DROP_FRAGS, "ODP_GET_DROP_FRAGS", &tmp);
 236     *drop_frags = error ? tmp & 1 : false;
 237     return error;
 238 }
 239
 240 int
 241 dpif_set_drop_frags(struct dpif *dpif, bool drop_frags)
 242 {
 243     int tmp = drop_frags;
 244     return do_ioctl(dpif, ODP_SET_DROP_FRAGS, "ODP_SET_DROP_FRAGS", &tmp);
 245 }
 246
 247 int
 248 dpif_get_listen_mask(const struct dpif *dpif, int *listen_mask)
 249 {
 250     int error = do_ioctl(dpif, ODP_GET_LISTEN_MASK, "ODP_GET_LISTEN_MASK",
 251                          listen_mask);
 252     if (error) {
 253         *listen_mask = 0;
 254     }
 255     return error;
 256 }
 257
 258 int
 259 dpif_set_listen_mask(struct dpif *dpif, int listen_mask)
 260 {
 261     return do_ioctl(dpif, ODP_SET_LISTEN_MASK, "ODP_SET_LISTEN_MASK",
 262                     &listen_mask);
 263 }
 264
 265 int
 266 dpif_purge(struct dpif *dpif)
 267 {
 268     struct odp_stats stats;
 269     unsigned int i;
 270     int error;
 271
 272     COVERAGE_INC(dpif_purge);
 273
 274     error = dpif_get_dp_stats(dpif, &stats);
 275     if (error) {
 276         return error;
 277     }
 278
 279     for (i = 0; i < stats.max_miss_queue + stats.max_action_queue; i++) {
 280         struct ofpbuf *buf;
 281         error = dpif_recv(dpif, &buf);
 282         if (error) {
 283             return error == EAGAIN ? 0 : error;
 284         }
 285         ofpbuf_delete(buf);
 286     }
 287     return 0;
 288 }
 289
 290 int
 291 dpif_port_add(struct dpif *dpif, const char *devname, uint16_t port_no,
 292               uint16_t flags)
 293 {
 294     struct odp_port port;
 295
 296     COVERAGE_INC(dpif_port_add);
 297     memset(&port, 0, sizeof port);
 298     strncpy(port.devname, devname, sizeof port.devname);
 299     port.port = port_no;
 300     port.flags = flags;
 301     if (!ioctl(dpif->fd, ODP_PORT_ADD, &port)) {
 302         VLOG_DBG_RL(&dpmsg_rl, "dp%u: added %s as port %"PRIu16,
 303                     dpif->minor, devname, port_no);
 304         return 0;
 305     } else {
 306         VLOG_WARN_RL(&error_rl, "dp%u: failed to add %s as port "
 307                      "%"PRIu16": %s", dpif->minor, devname, port_no,
 308                      strerror(errno));
 309         return errno;
 310     }
 311 }
 312
 313 int
 314 dpif_port_del(struct dpif *dpif, uint16_t port_no)
 315 {
 316     int tmp = port_no;
 317     COVERAGE_INC(dpif_port_del);
 318     return do_ioctl(dpif, ODP_PORT_DEL, "ODP_PORT_DEL", &tmp);
 319 }
 320
 321 int
 322 dpif_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
 323                           struct odp_port *port)
 324 {
 325     memset(port, 0, sizeof *port);
 326     port->port = port_no;
 327     if (!ioctl(dpif->fd, ODP_PORT_QUERY, port)) {
 328         VLOG_DBG_RL(&dpmsg_rl, "dp%u: port %"PRIu16" is device %s",
 329                     dpif->minor, port_no, port->devname);
 330         return 0;
 331     } else {
 332         VLOG_WARN_RL(&error_rl, "dp%u: failed to query port %"PRIu16": %s",
 333                      dpif->minor, port_no, strerror(errno));
 334         return errno;
 335     }
 336 }
 337
 338 int
 339 dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
 340                         struct odp_port *port)
 341 {
 342     memset(port, 0, sizeof *port);
 343     strncpy(port->devname, devname, sizeof port->devname);
 344     if (!ioctl(dpif->fd, ODP_PORT_QUERY, port)) {
 345         VLOG_DBG_RL(&dpmsg_rl, "dp%u: device %s is on port %"PRIu16,
 346                     dpif->minor, devname, port->port);
 347         return 0;
 348     } else {
 349         VLOG_WARN_RL(&error_rl, "dp%u: failed to query port %s: %s",
 350                      dpif->minor, devname, strerror(errno));
 351         return errno;
 352     }
 353 }
 354
 355 /* Looks up port number 'port_no' in 'dpif'.  On success, returns 0 and copies
 356  * the port's name into the 'name_size' bytes in 'name', ensuring that the
 357  * result is null-terminated.  On failure, returns a positive errno value and
 358  * makes 'name' the empty string. */
 359 int
 360 dpif_port_get_name(struct dpif *dpif, uint16_t port_no,
 361                    char *name, size_t name_size)
 362 {
 363     struct odp_port port;
 364     int error;
 365
 366     assert(name_size > 0);
 367
 368     error = dpif_port_query_by_number(dpif, port_no, &port);
 369     if (!error) {
 370         ovs_strlcpy(name, port.devname, name_size);
 371     } else {
 372         *name = '\0';
 373     }
 374     return error;
 375 }
 376
 377 int
 378 dpif_port_list(const struct dpif *dpif,
 379                struct odp_port **ports, size_t *n_ports)
 380 {
 381     struct odp_portvec pv;
 382     struct odp_stats stats;
 383     int error;
 384
 385     do {
 386         error = dpif_get_dp_stats(dpif, &stats);
 387         if (error) {
 388             goto error;
 389         }
 390
 391         *ports = xcalloc(1, stats.n_ports * sizeof **ports);
 392         pv.ports = *ports;
 393         pv.n_ports = stats.n_ports;
 394         error = do_ioctl(dpif, ODP_PORT_LIST, "ODP_PORT_LIST", &pv);
 395         if (error) {
 396             free(*ports);
 397             goto error;
 398         }
 399     } while (pv.n_ports != stats.n_ports);
 400     *n_ports = pv.n_ports;
 401     return 0;
 402
 403 error:
 404     *ports = NULL;
 405     *n_ports = 0;
 406     return error;
 407 }
 408
 409 int
 410 dpif_port_group_set(struct dpif *dpif, uint16_t group,
 411                     const uint16_t ports[], size_t n_ports)
 412 {
 413     struct odp_port_group pg;
 414
 415     COVERAGE_INC(dpif_port_group_set);
 416     assert(n_ports <= UINT16_MAX);
 417     pg.group = group;
 418     pg.ports = (uint16_t *) ports;
 419     pg.n_ports = n_ports;
 420     return do_ioctl(dpif, ODP_PORT_GROUP_SET, "ODP_PORT_GROUP_SET", &pg);
 421 }
 422
 423 /* Careful: '*n_out' can be greater than 'n_ports' on return, if 'n_ports' is
 424  * less than the number of ports in 'group'. */
 425 int
 426 dpif_port_group_get(const struct dpif *dpif, uint16_t group,
 427                     uint16_t ports[], size_t n_ports, size_t *n_out)
 428 {
 429     struct odp_port_group pg;
 430     int error;
 431
 432     assert(n_ports <= UINT16_MAX);
 433     pg.group = group;
 434     pg.ports = ports;
 435     pg.n_ports = n_ports;
 436     error = do_ioctl(dpif, ODP_PORT_GROUP_GET, "ODP_PORT_GROUP_GET", &pg);
 437     *n_out = error ? 0 : pg.n_ports;
 438     return error;
 439 }
 440
 441 int
 442 dpif_flow_flush(struct dpif *dpif)
 443 {
 444     COVERAGE_INC(dpif_flow_flush);
 445     return do_ioctl(dpif, ODP_FLOW_FLUSH, "ODP_FLOW_FLUSH", NULL);
 446 }
 447
 448 static enum vlog_level
 449 flow_message_log_level(int error)
 450 {
 451     return error ? VLL_WARN : VLL_DBG;
 452 }
 453
 454 static bool
 455 should_log_flow_message(int error)
 456 {
 457     return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
 458                              error ? &error_rl : &dpmsg_rl);
 459 }
 460
 461 static void
 462 log_flow_message(const struct dpif *dpif, int error,
 463                  const char *operation,
 464                  const flow_t *flow, const struct odp_flow_stats *stats,
 465                  const union odp_action *actions, size_t n_actions)
 466 {
 467     struct ds ds = DS_EMPTY_INITIALIZER;
 468     ds_put_format(&ds, "dp%u: ", dpif->minor);
 469     if (error) {
 470         ds_put_cstr(&ds, "failed to ");
 471     }
 472     ds_put_format(&ds, "%s ", operation);
 473     if (error) {
 474         ds_put_format(&ds, "(%s) ", strerror(error));
 475     }
 476     flow_format(&ds, flow);
 477     if (stats) {
 478         ds_put_cstr(&ds, ", ");
 479         format_odp_flow_stats(&ds, stats);
 480     }
 481     if (actions || n_actions) {
 482         ds_put_cstr(&ds, ", actions:");
 483         format_odp_actions(&ds, actions, n_actions);
 484     }
 485     vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
 486     ds_destroy(&ds);
 487 }
 488
 489 static int
 490 do_flow_ioctl(const struct dpif *dpif, int cmd, struct odp_flow *flow,
 491               const char *operation, bool show_stats)
 492 {
 493     int error = do_ioctl(dpif, cmd, NULL, flow);
 494     if (error && show_stats) {
 495         flow->n_actions = 0;
 496     }
 497     if (should_log_flow_message(error)) {
 498         log_flow_message(dpif, error, operation, &flow->key,
 499                          show_stats && !error ? &flow->stats : NULL,
 500                          flow->actions, flow->n_actions);
 501     }
 502     return error;
 503 }
 504
 505 int
 506 dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put)
 507 {
 508     int error = do_ioctl(dpif, ODP_FLOW_PUT, NULL, put);
 509     COVERAGE_INC(dpif_flow_put);
 510     if (should_log_flow_message(error)) {
 511         struct ds operation = DS_EMPTY_INITIALIZER;
 512         ds_put_cstr(&operation, "put");
 513         if (put->flags & ODPPF_CREATE) {
 514             ds_put_cstr(&operation, "[create]");
 515         }
 516         if (put->flags & ODPPF_MODIFY) {
 517             ds_put_cstr(&operation, "[modify]");
 518         }
 519         if (put->flags & ODPPF_ZERO_STATS) {
 520             ds_put_cstr(&operation, "[zero]");
 521         }
 522 #define ODPPF_ALL (ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS)
 523         if (put->flags & ~ODPPF_ALL) {
 524             ds_put_format(&operation, "[%x]", put->flags & ~ODPPF_ALL);
 525         }
 526         log_flow_message(dpif, error, ds_cstr(&operation), &put->flow.key,
 527                          !error ? &put->flow.stats : NULL,
 528                          put->flow.actions, put->flow.n_actions);
 529         ds_destroy(&operation);
 530     }
 531     return error;
 532 }
 533
 534 int
 535 dpif_flow_del(struct dpif *dpif, struct odp_flow *flow)
 536 {
 537     COVERAGE_INC(dpif_flow_del);
 538     check_rw_odp_flow(flow);
 539     memset(&flow->stats, 0, sizeof flow->stats);
 540     return do_flow_ioctl(dpif, ODP_FLOW_DEL, flow, "delete flow", true);
 541 }
 542
 543 int
 544 dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow)
 545 {
 546     COVERAGE_INC(dpif_flow_query);
 547     check_rw_odp_flow(flow);
 548     memset(&flow->stats, 0, sizeof flow->stats);
 549     return do_flow_ioctl(dpif, ODP_FLOW_GET, flow, "get flow", true);
 550 }
 551
 552 int
 553 dpif_flow_get_multiple(const struct dpif *dpif,
 554                        struct odp_flow flows[], size_t n)
 555 {
 556     struct odp_flowvec fv;
 557     size_t i;
 558
 559     COVERAGE_ADD(dpif_flow_query_multiple, n);
 560     fv.flows = flows;
 561     fv.n_flows = n;
 562     for (i = 0; i < n; i++) {
 563         check_rw_odp_flow(&flows[i]);
 564     }
 565     return do_ioctl(dpif, ODP_FLOW_GET_MULTIPLE, "ODP_FLOW_GET_MULTIPLE",
 566                     &fv);
 567 }
 568
 569 int
 570 dpif_flow_list(const struct dpif *dpif, struct odp_flow flows[], size_t n,
 571                size_t *n_out)
 572 {
 573     struct odp_flowvec fv;
 574     uint32_t i;
 575     int error;
 576
 577     COVERAGE_INC(dpif_flow_query_list);
 578     fv.flows = flows;
 579     fv.n_flows = n;
 580     if (RUNNING_ON_VALGRIND) {
 581         memset(flows, 0, n * sizeof *flows);
 582     } else {
 583         for (i = 0; i < n; i++) {
 584             flows[i].actions = NULL;
 585             flows[i].n_actions = 0;
 586         }
 587     }
 588     error = do_ioctl(dpif, ODP_FLOW_LIST, NULL, &fv);
 589     if (error) {
 590         *n_out = 0;
 591         VLOG_WARN_RL(&error_rl, "dp%u: flow list failed (%s)",
 592                      dpif->minor, strerror(error));
 593     } else {
 594         COVERAGE_ADD(dpif_flow_query_list_n, fv.n_flows);
 595         *n_out = fv.n_flows;
 596         VLOG_DBG_RL(&dpmsg_rl, "dp%u: listed %zu flows", dpif->minor, *n_out);
 597     }
 598     return error;
 599 }
 600
 601 int
 602 dpif_flow_list_all(const struct dpif *dpif,
 603                    struct odp_flow **flowsp, size_t *np)
 604 {
 605     struct odp_stats stats;
 606     struct odp_flow *flows;
 607     size_t n_flows;
 608     int error;
 609
 610     *flowsp = NULL;
 611     *np = 0;
 612
 613     error = dpif_get_dp_stats(dpif, &stats);
 614     if (error) {
 615         return error;
 616     }
 617
 618     flows = xmalloc(sizeof *flows * stats.n_flows);
 619     error = dpif_flow_list(dpif, flows, stats.n_flows, &n_flows);
 620     if (error) {
 621         free(flows);
 622         return error;
 623     }
 624
 625     if (stats.n_flows != n_flows) {
 626         VLOG_WARN_RL(&error_rl, "dp%u: datapath stats reported %"PRIu32" "
 627                      "flows but flow listing reported %zu",
 628                      dpif->minor, stats.n_flows, n_flows);
 629     }
 630     *flowsp = flows;
 631     *np = n_flows;
 632     return 0;
 633 }
 634
 635 int
 636 dpif_execute(struct dpif *dpif, uint16_t in_port,
 637              const union odp_action actions[], size_t n_actions,
 638              const struct ofpbuf *buf)
 639 {
 640     int error;
 641
 642     COVERAGE_INC(dpif_execute);
 643     if (n_actions > 0) {
 644         struct odp_execute execute;
 645         memset(&execute, 0, sizeof execute);
 646         execute.in_port = in_port;
 647         execute.actions = (union odp_action *) actions;
 648         execute.n_actions = n_actions;
 649         execute.data = buf->data;
 650         execute.length = buf->size;
 651         error = do_ioctl(dpif, ODP_EXECUTE, NULL, &execute);
 652     } else {
 653         error = 0;
 654     }
 655
 656     if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
 657         struct ds ds = DS_EMPTY_INITIALIZER;
 658         char *packet = ofp_packet_to_string(buf->data, buf->size, buf->size);
 659         ds_put_format(&ds, "dp%u: execute ", dpif->minor);
 660         format_odp_actions(&ds, actions, n_actions);
 661         if (error) {
 662             ds_put_format(&ds, " failed (%s)", strerror(error));
 663         }
 664         ds_put_format(&ds, " on packet %s", packet);
 665         vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
 666         ds_destroy(&ds);
 667         free(packet);
 668     }
 669     return error;
 670 }
 671
 672 int
 673 dpif_recv(struct dpif *dpif, struct ofpbuf **bufp)
 674 {
 675     struct ofpbuf *buf;
 676     int retval;
 677     int error;
 678
 679     buf = ofpbuf_new(65536);
 680     retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf));
 681     if (retval < 0) {
 682         error = errno;
 683         if (error != EAGAIN) {
 684             VLOG_WARN_RL(&error_rl, "dp%u: read failed: %s",
 685                          dpif->minor, strerror(error));
 686         }
 687     } else if (retval >= sizeof(struct odp_msg)) {
 688         struct odp_msg *msg = buf->data;
 689         if (msg->length <= retval) {
 690             buf->size += retval;
 691             if (VLOG_IS_DBG_ENABLED()) {
 692                 void *payload = msg + 1;
 693                 size_t length = buf->size - sizeof *msg;
 694                 char *s = ofp_packet_to_string(payload, length, length);
 695                 VLOG_DBG_RL(&dpmsg_rl, "dp%u: received %s message of length "
 696                             "%zu on port %"PRIu16": %s", dpif->minor,
 697                             (msg->type == _ODPL_MISS_NR ? "miss"
 698                              : msg->type == _ODPL_ACTION_NR ? "action"
 699                              : "<unknown>"),
 700                             msg->length - sizeof(struct odp_msg),
 701                             msg->port, s);
 702                 free(s);
 703             }
 704             *bufp = buf;
 705             COVERAGE_INC(dpif_recv);
 706             return 0;
 707         } else {
 708             VLOG_WARN_RL(&error_rl, "dp%u: discarding message truncated "
 709                          "from %"PRIu32" bytes to %d",
 710                          dpif->minor, msg->length, retval);
 711             error = ERANGE;
 712         }
 713     } else if (!retval) {
 714         VLOG_WARN_RL(&error_rl, "dp%u: unexpected end of file", dpif->minor);
 715         error = EPROTO;
 716     } else {
 717         VLOG_WARN_RL(&error_rl,
 718                      "dp%u: discarding too-short message (%d bytes)",
 719                      dpif->minor, retval);
 720         error = ERANGE;
 721     }
 722
 723     *bufp = NULL;
 724     ofpbuf_delete(buf);
 725     return error;
 726 }
 727
 728 void
 729 dpif_recv_wait(struct dpif *dpif)
 730 {
 731     poll_fd_wait(dpif->fd, POLLIN);
 732 }
 733 \f
 734 struct dpifmon {
 735     struct dpif dpif;
 736     struct nl_sock *sock;
 737     int local_ifindex;
 738 };
 739
 740 int
 741 dpifmon_create(const char *datapath_name, struct dpifmon **monp)
 742 {
 743     struct dpifmon *mon;
 744     char local_name[IFNAMSIZ];
 745     int error;
 746
 747     mon = *monp = xmalloc(sizeof *mon);
 748
 749     error = dpif_open(datapath_name, &mon->dpif);
 750     if (error) {
 751         goto error;
 752     }
 753     error = dpif_get_name(&mon->dpif, local_name, sizeof local_name);
 754     if (error) {
 755         goto error_close_dpif;
 756     }
 757
 758     mon->local_ifindex = if_nametoindex(local_name);
 759     if (!mon->local_ifindex) {
 760         error = errno;
 761         VLOG_WARN("could not get ifindex of %s device: %s",
 762                   local_name, strerror(errno));
 763         goto error_close_dpif;
 764     }
 765
 766     error = nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0, &mon->sock);
 767     if (error) {
 768         VLOG_WARN("could not create rtnetlink socket: %s", strerror(error));
 769         goto error_close_dpif;
 770     }
 771
 772     return 0;
 773
 774 error_close_dpif:
 775     dpif_close(&mon->dpif);
 776 error:
 777     free(mon);
 778     *monp = NULL;
 779     return error;
 780 }
 781
 782 void
 783 dpifmon_destroy(struct dpifmon *mon)
 784 {
 785     if (mon) {
 786         dpif_close(&mon->dpif);
 787         nl_sock_destroy(mon->sock);
 788     }
 789 }
 790
 791 int
 792 dpifmon_poll(struct dpifmon *mon, char **devnamep)
 793 {
 794     static struct vlog_rate_limit slow_rl = VLOG_RATE_LIMIT_INIT(1, 5);
 795     static const struct nl_policy rtnlgrp_link_policy[] = {
 796         [IFLA_IFNAME] = { .type = NL_A_STRING },
 797         [IFLA_MASTER] = { .type = NL_A_U32, .optional = true },
 798     };
 799     struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
 800     struct ofpbuf *buf;
 801     int error;
 802
 803     *devnamep = NULL;
 804 again:
 805     error = nl_sock_recv(mon->sock, &buf, false);
 806     switch (error) {
 807     case 0:
 808         if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
 809                              rtnlgrp_link_policy,
 810                              attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
 811             VLOG_WARN_RL(&slow_rl, "received bad rtnl message");
 812             error = ENOBUFS;
 813         } else {
 814             const char *devname = nl_attr_get_string(attrs[IFLA_IFNAME]);
 815             bool for_us;
 816
 817             if (attrs[IFLA_MASTER]) {
 818                 uint32_t master_ifindex = nl_attr_get_u32(attrs[IFLA_MASTER]);
 819                 for_us = master_ifindex == mon->local_ifindex;
 820             } else {
 821                 /* It's for us if that device is one of our ports.  This is
 822                  * open-coded instead of using dpif_port_query_by_name() to
 823                  * avoid logging a warning on failure. */
 824                 struct odp_port port;
 825                 memset(&port, 0, sizeof port);
 826                 strncpy(port.devname, devname, sizeof port.devname);
 827                 for_us = !ioctl(mon->dpif.fd, ODP_PORT_QUERY, &port);
 828             }
 829
 830             if (!for_us) {
 831                 /* Not for us, try again. */
 832                 ofpbuf_delete(buf);
 833                 COVERAGE_INC(dpifmon_poll_false_wakeup);
 834                 goto again;
 835             }
 836             COVERAGE_INC(dpifmon_poll_changed);
 837             *devnamep = xstrdup(devname);
 838         }
 839         ofpbuf_delete(buf);
 840         break;
 841
 842     case EAGAIN:
 843         /* Nothing to do. */
 844         break;
 845
 846     case ENOBUFS:
 847         VLOG_WARN_RL(&slow_rl, "dpifmon socket overflowed");
 848         break;
 849
 850     default:
 851         VLOG_WARN_RL(&slow_rl, "error on dpifmon socket: %s", strerror(error));
 852         break;
 853     }
 854     return error;
 855 }
 856
 857 void
 858 dpifmon_run(struct dpifmon *mon UNUSED)
 859 {
 860     /* Nothing to do in this implementation. */
 861 }
 862
 863 void
 864 dpifmon_wait(struct dpifmon *mon)
 865 {
 866     nl_sock_wait(mon->sock, POLLIN);
 867 }
 868 \f
 869 static int get_openvswitch_major(void);
 870 static int get_major(const char *target, int default_major);
 871
 872 static int
 873 lookup_minor(const char *name, unsigned int *minor)
 874 {
 875     struct ethtool_drvinfo drvinfo;
 876     struct ifreq ifr;
 877     int error;
 878     int sock;
 879
 880     *minor = -1;
 881     sock = socket(AF_INET, SOCK_DGRAM, 0);
 882     if (sock < 0) {
 883         VLOG_WARN("socket(AF_INET) failed: %s", strerror(errno));
 884         error = errno;
 885         goto error;
 886     }
 887
 888     memset(&ifr, 0, sizeof ifr);
 889     strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
 890     ifr.ifr_data = (caddr_t) &drvinfo;
 891
 892     memset(&drvinfo, 0, sizeof drvinfo);
 893     drvinfo.cmd = ETHTOOL_GDRVINFO;
 894     if (ioctl(sock, SIOCETHTOOL, &ifr)) {
 895         VLOG_WARN("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
 896         error = errno;
 897         goto error_close_sock;
 898     }
 899
 900     if (strcmp(drvinfo.driver, "openvswitch")) {
 901         VLOG_WARN("%s is not an openvswitch device", name);
 902         error = EOPNOTSUPP;
 903         goto error_close_sock;
 904     }
 905
 906     if (!isdigit(drvinfo.bus_info[0])) {
 907         VLOG_WARN("%s ethtool info does not contain an openvswitch minor",
 908                   name);
 909         error = EPROTOTYPE;
 910         goto error_close_sock;
 911     }
 912
 913     *minor = atoi(drvinfo.bus_info);
 914     close(sock);
 915     return 0;
 916
 917 error_close_sock:
 918     close(sock);
 919 error:
 920     return error;
 921 }
 922
 923 static int
 924 make_openvswitch_device(unsigned int minor, char **fnp)
 925 {
 926     dev_t dev = makedev(get_openvswitch_major(), minor);
 927     const char dirname[] = "/dev/net";
 928     struct stat s;
 929     char fn[128];
 930
 931     *fnp = NULL;
 932     sprintf(fn, "%s/dp%d", dirname, minor);
 933     if (!stat(fn, &s)) {
 934         if (!S_ISCHR(s.st_mode)) {
 935             VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing",
 936                          fn);
 937         } else if (s.st_rdev != dev) {
 938             VLOG_WARN_RL(&error_rl,
 939                          "%s is device %u:%u instead of %u:%u, fixing",
 940                          fn, major(s.st_rdev), minor(s.st_rdev),
 941                          major(dev), minor(dev));
 942         } else {
 943             goto success;
 944         }
 945         if (unlink(fn)) {
 946             VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)",
 947                          fn, strerror(errno));
 948             return errno;
 949         }
 950     } else if (errno == ENOENT) {
 951         if (stat(dirname, &s)) {
 952             if (errno == ENOENT) {
 953                 if (mkdir(dirname, 0755)) {
 954                     VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)",
 955                                  dirname, strerror(errno));
 956                     return errno;
 957                 }
 958             } else {
 959                 VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)",
 960                              dirname, strerror(errno));
 961                 return errno;
 962             }
 963         }
 964     } else {
 965         VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno));
 966         return errno;
 967     }
 968
 969     /* The device needs to be created. */
 970     if (mknod(fn, S_IFCHR | 0700, dev)) {
 971         VLOG_WARN_RL(&error_rl,
 972                      "%s: creating character device %u:%u failed (%s)",
 973                      fn, major(dev), minor(dev), strerror(errno));
 974         return errno;
 975     }
 976
 977 success:
 978     *fnp = xstrdup(fn);
 979     return 0;
 980 }
 981
 982
 983 static int
 984 get_openvswitch_major(void)
 985 {
 986     static unsigned int openvswitch_major;
 987     if (!openvswitch_major) {
 988         enum { DEFAULT_MAJOR = 248 };
 989         openvswitch_major = get_major("openvswitch", DEFAULT_MAJOR);
 990     }
 991     return openvswitch_major;
 992 }
 993
 994 static int
 995 get_major(const char *target, int default_major)
 996 {
 997     const char fn[] = "/proc/devices";
 998     char line[128];
 999     FILE *file;
1000     int ln;
1001
1002     file = fopen(fn, "r");
1003     if (!file) {
1004         VLOG_ERR("opening %s failed (%s)", fn, strerror(errno));
1005         goto error;
1006     }
1007
1008     for (ln = 1; fgets(line, sizeof line, file); ln++) {
1009         char name[64];
1010         int major;
1011
1012         if (!strncmp(line, "Character", 9) || line[0] == '\0') {
1013             /* Nothing to do. */
1014         } else if (!strncmp(line, "Block", 5)) {
1015             /* We only want character devices, so skip the rest of the file. */
1016             break;
1017         } else if (sscanf(line, "%d %63s", &major, name)) {
1018             if (!strcmp(name, target)) {
1019                 fclose(file);
1020                 return major;
1021             }
1022         } else {
1023             static bool warned;
1024             if (!warned) {
1025                 VLOG_WARN("%s:%d: syntax error", fn, ln);
1026             }
1027             warned = true;
1028         }
1029     }
1030
1031     VLOG_ERR("%s: %s major not found (is the module loaded?), using "
1032              "default major %d", fn, target, default_major);
1033 error:
1034     VLOG_INFO("using default major %d for %s", default_major, target);
1035     return default_major;
1036 }
1037
1038 static int
1039 name_to_minor(const char *name, unsigned int *minor)
1040 {
1041     if (!get_minor_from_name(name, minor)) {
1042         return 0;
1043     }
1044     return lookup_minor(name, minor);
1045 }
1046
1047 static int
1048 get_minor_from_name(const char *name, unsigned int *minor)
1049 {
1050     if (!strncmp(name, "dp", 2) && isdigit(name[2])) {
1051         *minor = atoi(name + 2);
1052         return 0;
1053     } else if (!strncmp(name, "nl:", 3) && isdigit(name[3])) {
1054         /* This is for compatibility only and will be dropped. */
1055         *minor = atoi(name + 3);
1056         return 0;
1057     } else {
1058         return EINVAL;
1059     }
1060 }
1061
1062 static int
1063 open_by_minor(unsigned int minor, struct dpif *dpif)
1064 {
1065     int error;
1066     char *fn;
1067     int fd;
1068
1069     dpif->minor = -1;
1070     dpif->fd = -1;
1071     error = make_openvswitch_device(minor, &fn);
1072     if (error) {
1073         return error;
1074     }
1075
1076     fd = open(fn, O_RDONLY | O_NONBLOCK);
1077     if (fd < 0) {
1078         error = errno;
1079         VLOG_WARN("%s: open failed (%s)", fn, strerror(error));
1080         free(fn);
1081         return error;
1082     }
1083
1084     free(fn);
1085     dpif->minor = minor;
1086     dpif->fd = fd;
1087     return 0;
1088 }
1089 \f
1090 /* There is a tendency to construct odp_flow objects on the stack and to
1091  * forget to properly initialize their "actions" and "n_actions" members.
1092  * When this happens, we get memory corruption because the kernel
1093  * writes through the random pointer that is in the "actions" member.
1094  *
1095  * This function attempts to combat the problem by:
1096  *
1097  *      - Forcing a segfault if "actions" points to an invalid region (instead
1098  *        of just getting back EFAULT, which can be easily missed in the log).
1099  *
1100  *      - Storing a distinctive value that is likely to cause an
1101  *        easy-to-identify error later if it is dereferenced, etc.
1102  *
1103  *      - Triggering a warning on uninitialized memory from Valgrind if
1104  *        "actions" or "n_actions" was not initialized.
1105  */
1106 static void
1107 check_rw_odp_flow(struct odp_flow *flow)
1108 {
1109     if (flow->n_actions) {
1110         memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]);
1111     }
1112 }