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 int
 356 dpif_port_list(const struct dpif *dpif,
 357                struct odp_port **ports, size_t *n_ports)
 358 {
 359     struct odp_portvec pv;
 360     struct odp_stats stats;
 361     int error;
 362
 363     do {
 364         error = dpif_get_dp_stats(dpif, &stats);
 365         if (error) {
 366             goto error;
 367         }
 368
 369         *ports = xcalloc(1, stats.n_ports * sizeof **ports);
 370         pv.ports = *ports;
 371         pv.n_ports = stats.n_ports;
 372         error = do_ioctl(dpif, ODP_PORT_LIST, "ODP_PORT_LIST", &pv);
 373         if (error) {
 374             free(*ports);
 375             goto error;
 376         }
 377     } while (pv.n_ports != stats.n_ports);
 378     *n_ports = pv.n_ports;
 379     return 0;
 380
 381 error:
 382     *ports = NULL;
 383     *n_ports = 0;
 384     return error;
 385 }
 386
 387 int
 388 dpif_port_group_set(struct dpif *dpif, uint16_t group,
 389                     const uint16_t ports[], size_t n_ports)
 390 {
 391     struct odp_port_group pg;
 392
 393     COVERAGE_INC(dpif_port_group_set);
 394     assert(n_ports <= UINT16_MAX);
 395     pg.group = group;
 396     pg.ports = (uint16_t *) ports;
 397     pg.n_ports = n_ports;
 398     return do_ioctl(dpif, ODP_PORT_GROUP_SET, "ODP_PORT_GROUP_SET", &pg);
 399 }
 400
 401 /* Careful: '*n_out' can be greater than 'n_ports' on return, if 'n_ports' is
 402  * less than the number of ports in 'group'. */
 403 int
 404 dpif_port_group_get(const struct dpif *dpif, uint16_t group,
 405                     uint16_t ports[], size_t n_ports, size_t *n_out)
 406 {
 407     struct odp_port_group pg;
 408     int error;
 409
 410     assert(n_ports <= UINT16_MAX);
 411     pg.group = group;
 412     pg.ports = ports;
 413     pg.n_ports = n_ports;
 414     error = do_ioctl(dpif, ODP_PORT_GROUP_GET, "ODP_PORT_GROUP_GET", &pg);
 415     *n_out = error ? 0 : pg.n_ports;
 416     return error;
 417 }
 418
 419 int
 420 dpif_flow_flush(struct dpif *dpif)
 421 {
 422     COVERAGE_INC(dpif_flow_flush);
 423     return do_ioctl(dpif, ODP_FLOW_FLUSH, "ODP_FLOW_FLUSH", NULL);
 424 }
 425
 426 static enum vlog_level
 427 flow_message_log_level(int error)
 428 {
 429     return error ? VLL_WARN : VLL_DBG;
 430 }
 431
 432 static bool
 433 should_log_flow_message(int error)
 434 {
 435     return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
 436                              error ? &error_rl : &dpmsg_rl);
 437 }
 438
 439 static void
 440 log_flow_message(const struct dpif *dpif, int error,
 441                  const char *operation,
 442                  const flow_t *flow, const struct odp_flow_stats *stats,
 443                  const union odp_action *actions, size_t n_actions)
 444 {
 445     struct ds ds = DS_EMPTY_INITIALIZER;
 446     ds_put_format(&ds, "dp%u: ", dpif->minor);
 447     if (error) {
 448         ds_put_cstr(&ds, "failed to ");
 449     }
 450     ds_put_format(&ds, "%s ", operation);
 451     if (error) {
 452         ds_put_format(&ds, "(%s) ", strerror(error));
 453     }
 454     flow_format(&ds, flow);
 455     if (stats) {
 456         ds_put_cstr(&ds, ", ");
 457         format_odp_flow_stats(&ds, stats);
 458     }
 459     if (actions || n_actions) {
 460         ds_put_cstr(&ds, ", actions:");
 461         format_odp_actions(&ds, actions, n_actions);
 462     }
 463     vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
 464     ds_destroy(&ds);
 465 }
 466
 467 static int
 468 do_flow_ioctl(const struct dpif *dpif, int cmd, struct odp_flow *flow,
 469               const char *operation, bool show_stats)
 470 {
 471     int error = do_ioctl(dpif, cmd, NULL, flow);
 472     if (error && show_stats) {
 473         flow->n_actions = 0;
 474     }
 475     if (should_log_flow_message(error)) {
 476         log_flow_message(dpif, error, operation, &flow->key,
 477                          show_stats && !error ? &flow->stats : NULL,
 478                          flow->actions, flow->n_actions);
 479     }
 480     return error;
 481 }
 482
 483 int
 484 dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put)
 485 {
 486     int error = do_ioctl(dpif, ODP_FLOW_PUT, NULL, put);
 487     COVERAGE_INC(dpif_flow_put);
 488     if (should_log_flow_message(error)) {
 489         struct ds operation = DS_EMPTY_INITIALIZER;
 490         ds_put_cstr(&operation, "put");
 491         if (put->flags & ODPPF_CREATE) {
 492             ds_put_cstr(&operation, "[create]");
 493         }
 494         if (put->flags & ODPPF_MODIFY) {
 495             ds_put_cstr(&operation, "[modify]");
 496         }
 497         if (put->flags & ODPPF_ZERO_STATS) {
 498             ds_put_cstr(&operation, "[zero]");
 499         }
 500 #define ODPPF_ALL (ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS)
 501         if (put->flags & ~ODPPF_ALL) {
 502             ds_put_format(&operation, "[%x]", put->flags & ~ODPPF_ALL);
 503         }
 504         log_flow_message(dpif, error, ds_cstr(&operation), &put->flow.key,
 505                          !error ? &put->flow.stats : NULL,
 506                          put->flow.actions, put->flow.n_actions);
 507         ds_destroy(&operation);
 508     }
 509     return error;
 510 }
 511
 512 int
 513 dpif_flow_del(struct dpif *dpif, struct odp_flow *flow)
 514 {
 515     COVERAGE_INC(dpif_flow_del);
 516     check_rw_odp_flow(flow);
 517     memset(&flow->stats, 0, sizeof flow->stats);
 518     return do_flow_ioctl(dpif, ODP_FLOW_DEL, flow, "delete flow", true);
 519 }
 520
 521 int
 522 dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow)
 523 {
 524     COVERAGE_INC(dpif_flow_query);
 525     check_rw_odp_flow(flow);
 526     memset(&flow->stats, 0, sizeof flow->stats);
 527     return do_flow_ioctl(dpif, ODP_FLOW_GET, flow, "get flow", true);
 528 }
 529
 530 int
 531 dpif_flow_get_multiple(const struct dpif *dpif,
 532                        struct odp_flow flows[], size_t n)
 533 {
 534     struct odp_flowvec fv;
 535     size_t i;
 536
 537     COVERAGE_ADD(dpif_flow_query_multiple, n);
 538     fv.flows = flows;
 539     fv.n_flows = n;
 540     for (i = 0; i < n; i++) {
 541         check_rw_odp_flow(&flows[i]);
 542     }
 543     return do_ioctl(dpif, ODP_FLOW_GET_MULTIPLE, "ODP_FLOW_GET_MULTIPLE",
 544                     &fv);
 545 }
 546
 547 int
 548 dpif_flow_list(const struct dpif *dpif, struct odp_flow flows[], size_t n,
 549                size_t *n_out)
 550 {
 551     struct odp_flowvec fv;
 552     uint32_t i;
 553     int error;
 554
 555     COVERAGE_INC(dpif_flow_query_list);
 556     fv.flows = flows;
 557     fv.n_flows = n;
 558     if (RUNNING_ON_VALGRIND) {
 559         memset(flows, 0, n * sizeof *flows);
 560     } else {
 561         for (i = 0; i < n; i++) {
 562             flows[i].actions = NULL;
 563             flows[i].n_actions = 0;
 564         }
 565     }
 566     error = do_ioctl(dpif, ODP_FLOW_LIST, NULL, &fv);
 567     if (error) {
 568         *n_out = 0;
 569         VLOG_WARN_RL(&error_rl, "dp%u: flow list failed (%s)",
 570                      dpif->minor, strerror(error));
 571     } else {
 572         COVERAGE_ADD(dpif_flow_query_list_n, fv.n_flows);
 573         *n_out = fv.n_flows;
 574         VLOG_DBG_RL(&dpmsg_rl, "dp%u: listed %zu flows", dpif->minor, *n_out);
 575     }
 576     return error;
 577 }
 578
 579 int
 580 dpif_flow_list_all(const struct dpif *dpif,
 581                    struct odp_flow **flowsp, size_t *np)
 582 {
 583     struct odp_stats stats;
 584     struct odp_flow *flows;
 585     size_t n_flows;
 586     int error;
 587
 588     *flowsp = NULL;
 589     *np = 0;
 590
 591     error = dpif_get_dp_stats(dpif, &stats);
 592     if (error) {
 593         return error;
 594     }
 595
 596     flows = xmalloc(sizeof *flows * stats.n_flows);
 597     error = dpif_flow_list(dpif, flows, stats.n_flows, &n_flows);
 598     if (error) {
 599         free(flows);
 600         return error;
 601     }
 602
 603     if (stats.n_flows != n_flows) {
 604         VLOG_WARN_RL(&error_rl, "dp%u: datapath stats reported %"PRIu32" "
 605                      "flows but flow listing reported %zu",
 606                      dpif->minor, stats.n_flows, n_flows);
 607     }
 608     *flowsp = flows;
 609     *np = n_flows;
 610     return 0;
 611 }
 612
 613 int
 614 dpif_execute(struct dpif *dpif, uint16_t in_port,
 615              const union odp_action actions[], size_t n_actions,
 616              const struct ofpbuf *buf)
 617 {
 618     int error;
 619
 620     COVERAGE_INC(dpif_execute);
 621     if (n_actions > 0) {
 622         struct odp_execute execute;
 623         memset(&execute, 0, sizeof execute);
 624         execute.in_port = in_port;
 625         execute.actions = (union odp_action *) actions;
 626         execute.n_actions = n_actions;
 627         execute.data = buf->data;
 628         execute.length = buf->size;
 629         error = do_ioctl(dpif, ODP_EXECUTE, NULL, &execute);
 630     } else {
 631         error = 0;
 632     }
 633
 634     if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
 635         struct ds ds = DS_EMPTY_INITIALIZER;
 636         char *packet = ofp_packet_to_string(buf->data, buf->size, buf->size);
 637         ds_put_format(&ds, "dp%u: execute ", dpif->minor);
 638         format_odp_actions(&ds, actions, n_actions);
 639         if (error) {
 640             ds_put_format(&ds, " failed (%s)", strerror(error));
 641         }
 642         ds_put_format(&ds, " on packet %s", packet);
 643         vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
 644         ds_destroy(&ds);
 645         free(packet);
 646     }
 647     return error;
 648 }
 649
 650 int
 651 dpif_recv(struct dpif *dpif, struct ofpbuf **bufp)
 652 {
 653     struct ofpbuf *buf;
 654     int retval;
 655     int error;
 656
 657     buf = ofpbuf_new(65536);
 658     retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf));
 659     if (retval < 0) {
 660         error = errno;
 661         if (error != EAGAIN) {
 662             VLOG_WARN_RL(&error_rl, "dp%u: read failed: %s",
 663                          dpif->minor, strerror(error));
 664         }
 665     } else if (retval >= sizeof(struct odp_msg)) {
 666         struct odp_msg *msg = buf->data;
 667         if (msg->length <= retval) {
 668             buf->size += retval;
 669             if (VLOG_IS_DBG_ENABLED()) {
 670                 void *payload = msg + 1;
 671                 size_t length = buf->size - sizeof *msg;
 672                 char *s = ofp_packet_to_string(payload, length, length);
 673                 VLOG_DBG_RL(&dpmsg_rl, "dp%u: received %s message of length "
 674                             "%zu on port %"PRIu16": %s", dpif->minor,
 675                             (msg->type == _ODPL_MISS_NR ? "miss"
 676                              : msg->type == _ODPL_ACTION_NR ? "action"
 677                              : "<unknown>"),
 678                             msg->length - sizeof(struct odp_msg),
 679                             msg->port, s);
 680                 free(s);
 681             }
 682             *bufp = buf;
 683             COVERAGE_INC(dpif_recv);
 684             return 0;
 685         } else {
 686             VLOG_WARN_RL(&error_rl, "dp%u: discarding message truncated "
 687                          "from %zu bytes to %d",
 688                          dpif->minor, msg->length, retval);
 689             error = ERANGE;
 690         }
 691     } else if (!retval) {
 692         VLOG_WARN_RL(&error_rl, "dp%u: unexpected end of file", dpif->minor);
 693         error = EPROTO;
 694     } else {
 695         VLOG_WARN_RL(&error_rl,
 696                      "dp%u: discarding too-short message (%d bytes)",
 697                      dpif->minor, retval);
 698         error = ERANGE;
 699     }
 700
 701     *bufp = NULL;
 702     ofpbuf_delete(buf);
 703     return error;
 704 }
 705
 706 void
 707 dpif_recv_wait(struct dpif *dpif)
 708 {
 709     poll_fd_wait(dpif->fd, POLLIN);
 710 }
 711 \f
 712 struct dpifmon {
 713     struct dpif dpif;
 714     struct nl_sock *sock;
 715     int local_ifindex;
 716 };
 717
 718 int
 719 dpifmon_create(const char *datapath_name, struct dpifmon **monp)
 720 {
 721     struct dpifmon *mon;
 722     char local_name[IFNAMSIZ];
 723     int error;
 724
 725     mon = *monp = xmalloc(sizeof *mon);
 726
 727     error = dpif_open(datapath_name, &mon->dpif);
 728     if (error) {
 729         goto error;
 730     }
 731     error = dpif_get_name(&mon->dpif, local_name, sizeof local_name);
 732     if (error) {
 733         goto error_close_dpif;
 734     }
 735
 736     mon->local_ifindex = if_nametoindex(local_name);
 737     if (!mon->local_ifindex) {
 738         error = errno;
 739         VLOG_WARN("could not get ifindex of %s device: %s",
 740                   local_name, strerror(errno));
 741         goto error_close_dpif;
 742     }
 743
 744     error = nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0, &mon->sock);
 745     if (error) {
 746         VLOG_WARN("could not create rtnetlink socket: %s", strerror(error));
 747         goto error_close_dpif;
 748     }
 749
 750     return 0;
 751
 752 error_close_dpif:
 753     dpif_close(&mon->dpif);
 754 error:
 755     free(mon);
 756     *monp = NULL;
 757     return error;
 758 }
 759
 760 void
 761 dpifmon_destroy(struct dpifmon *mon)
 762 {
 763     if (mon) {
 764         dpif_close(&mon->dpif);
 765         nl_sock_destroy(mon->sock);
 766     }
 767 }
 768
 769 int
 770 dpifmon_poll(struct dpifmon *mon, char **devnamep)
 771 {
 772     static struct vlog_rate_limit slow_rl = VLOG_RATE_LIMIT_INIT(1, 5);
 773     static const struct nl_policy rtnlgrp_link_policy[] = {
 774         [IFLA_IFNAME] = { .type = NL_A_STRING },
 775         [IFLA_MASTER] = { .type = NL_A_U32, .optional = true },
 776     };
 777     struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
 778     struct ofpbuf *buf;
 779     int error;
 780
 781     *devnamep = NULL;
 782 again:
 783     error = nl_sock_recv(mon->sock, &buf, false);
 784     switch (error) {
 785     case 0:
 786         if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
 787                              rtnlgrp_link_policy,
 788                              attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
 789             VLOG_WARN_RL(&slow_rl, "received bad rtnl message");
 790             error = ENOBUFS;
 791         } else {
 792             const char *devname = nl_attr_get_string(attrs[IFLA_IFNAME]);
 793             bool for_us;
 794
 795             if (attrs[IFLA_MASTER]) {
 796                 uint32_t master_ifindex = nl_attr_get_u32(attrs[IFLA_MASTER]);
 797                 for_us = master_ifindex == mon->local_ifindex;
 798             } else {
 799                 /* It's for us if that device is one of our ports.  This is
 800                  * open-coded instead of using dpif_port_query_by_name() to
 801                  * avoid logging a warning on failure. */
 802                 struct odp_port port;
 803                 memset(&port, 0, sizeof port);
 804                 strncpy(port.devname, devname, sizeof port.devname);
 805                 for_us = !ioctl(mon->dpif.fd, ODP_PORT_QUERY, &port);
 806             }
 807
 808             if (!for_us) {
 809                 /* Not for us, try again. */
 810                 ofpbuf_delete(buf);
 811                 COVERAGE_INC(dpifmon_poll_false_wakeup);
 812                 goto again;
 813             }
 814             COVERAGE_INC(dpifmon_poll_changed);
 815             *devnamep = xstrdup(devname);
 816         }
 817         ofpbuf_delete(buf);
 818         break;
 819
 820     case EAGAIN:
 821         /* Nothing to do. */
 822         break;
 823
 824     case ENOBUFS:
 825         VLOG_WARN_RL(&slow_rl, "dpifmon socket overflowed");
 826         break;
 827
 828     default:
 829         VLOG_WARN_RL(&slow_rl, "error on dpifmon socket: %s", strerror(error));
 830         break;
 831     }
 832     return error;
 833 }
 834
 835 void
 836 dpifmon_run(struct dpifmon *mon UNUSED)
 837 {
 838     /* Nothing to do in this implementation. */
 839 }
 840
 841 void
 842 dpifmon_wait(struct dpifmon *mon)
 843 {
 844     nl_sock_wait(mon->sock, POLLIN);
 845 }
 846 \f
 847 static int get_openvswitch_major(void);
 848 static int get_major(const char *target, int default_major);
 849
 850 static int
 851 lookup_minor(const char *name, unsigned int *minor)
 852 {
 853     struct ethtool_drvinfo drvinfo;
 854     struct ifreq ifr;
 855     int error;
 856     int sock;
 857
 858     *minor = -1;
 859     sock = socket(AF_INET, SOCK_DGRAM, 0);
 860     if (sock < 0) {
 861         VLOG_WARN("socket(AF_INET) failed: %s", strerror(errno));
 862         error = errno;
 863         goto error;
 864     }
 865
 866     memset(&ifr, 0, sizeof ifr);
 867     strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
 868     ifr.ifr_data = (caddr_t) &drvinfo;
 869
 870     memset(&drvinfo, 0, sizeof drvinfo);
 871     drvinfo.cmd = ETHTOOL_GDRVINFO;
 872     if (ioctl(sock, SIOCETHTOOL, &ifr)) {
 873         VLOG_WARN("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
 874         error = errno;
 875         goto error_close_sock;
 876     }
 877
 878     if (strcmp(drvinfo.driver, "openvswitch")) {
 879         VLOG_WARN("%s is not an openvswitch device", name);
 880         error = EOPNOTSUPP;
 881         goto error_close_sock;
 882     }
 883
 884     if (!isdigit(drvinfo.bus_info[0])) {
 885         VLOG_WARN("%s ethtool info does not contain an openvswitch minor",
 886                   name);
 887         error = EPROTOTYPE;
 888         goto error_close_sock;
 889     }
 890
 891     *minor = atoi(drvinfo.bus_info);
 892     close(sock);
 893     return 0;
 894
 895 error_close_sock:
 896     close(sock);
 897 error:
 898     return error;
 899 }
 900
 901 static int
 902 make_openvswitch_device(unsigned int minor, char **fnp)
 903 {
 904     dev_t dev = makedev(get_openvswitch_major(), minor);
 905     const char dirname[] = "/dev/net";
 906     struct stat s;
 907     char fn[128];
 908
 909     *fnp = NULL;
 910     sprintf(fn, "%s/dp%d", dirname, minor);
 911     if (!stat(fn, &s)) {
 912         if (!S_ISCHR(s.st_mode)) {
 913             VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing",
 914                          fn);
 915         } else if (s.st_rdev != dev) {
 916             VLOG_WARN_RL(&error_rl,
 917                          "%s is device %u:%u instead of %u:%u, fixing",
 918                          fn, major(s.st_rdev), minor(s.st_rdev),
 919                          major(dev), minor(dev));
 920         } else {
 921             goto success;
 922         }
 923         if (unlink(fn)) {
 924             VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)",
 925                          fn, strerror(errno));
 926             return errno;
 927         }
 928     } else if (errno == ENOENT) {
 929         if (stat(dirname, &s)) {
 930             if (errno == ENOENT) {
 931                 if (mkdir(dirname, 0755)) {
 932                     VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)",
 933                                  dirname, strerror(errno));
 934                     return errno;
 935                 }
 936             } else {
 937                 VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)",
 938                              dirname, strerror(errno));
 939                 return errno;
 940             }
 941         }
 942     } else {
 943         VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno));
 944         return errno;
 945     }
 946
 947     /* The device needs to be created. */
 948     if (mknod(fn, S_IFCHR | 0700, dev)) {
 949         VLOG_WARN_RL(&error_rl,
 950                      "%s: creating character device %u:%u failed (%s)",
 951                      fn, major(dev), minor(dev), strerror(errno));
 952         return errno;
 953     }
 954
 955 success:
 956     *fnp = xstrdup(fn);
 957     return 0;
 958 }
 959
 960
 961 static int
 962 get_openvswitch_major(void)
 963 {
 964     static unsigned int openvswitch_major;
 965     if (!openvswitch_major) {
 966         enum { DEFAULT_MAJOR = 248 };
 967         openvswitch_major = get_major("openvswitch", DEFAULT_MAJOR);
 968     }
 969     return openvswitch_major;
 970 }
 971
 972 static int
 973 get_major(const char *target, int default_major)
 974 {
 975     const char fn[] = "/proc/devices";
 976     char line[128];
 977     FILE *file;
 978     int ln;
 979
 980     file = fopen(fn, "r");
 981     if (!file) {
 982         VLOG_ERR("opening %s failed (%s)", fn, strerror(errno));
 983         goto error;
 984     }
 985
 986     for (ln = 1; fgets(line, sizeof line, file); ln++) {
 987         char name[64];
 988         int major;
 989
 990         if (!strncmp(line, "Character", 9) || line[0] == '\0') {
 991             /* Nothing to do. */
 992         } else if (!strncmp(line, "Block", 5)) {
 993             /* We only want character devices, so skip the rest of the file. */
 994             break;
 995         } else if (sscanf(line, "%d %63s", &major, name)) {
 996             if (!strcmp(name, target)) {
 997                 fclose(file);
 998                 return major;
 999             }
1000         } else {
1001             static bool warned;
1002             if (!warned) {
1003                 VLOG_WARN("%s:%d: syntax error", fn, ln);
1004             }
1005             warned = true;
1006         }
1007     }
1008
1009     VLOG_ERR("%s: %s major not found (is the module loaded?), using "
1010              "default major %d", fn, target, default_major);
1011 error:
1012     VLOG_INFO("using default major %d for %s", default_major, target);
1013     return default_major;
1014 }
1015
1016 static int
1017 name_to_minor(const char *name, unsigned int *minor)
1018 {
1019     if (!get_minor_from_name(name, minor)) {
1020         return 0;
1021     }
1022     return lookup_minor(name, minor);
1023 }
1024
1025 static int
1026 get_minor_from_name(const char *name, unsigned int *minor)
1027 {
1028     if (!strncmp(name, "dp", 2) && isdigit(name[2])) {
1029         *minor = atoi(name + 2);
1030         return 0;
1031     } else if (!strncmp(name, "nl:", 3) && isdigit(name[3])) {
1032         /* This is for compatibility only and will be dropped. */
1033         *minor = atoi(name + 3);
1034         return 0;
1035     } else {
1036         return EINVAL;
1037     }
1038 }
1039
1040 static int
1041 open_by_minor(unsigned int minor, struct dpif *dpif)
1042 {
1043     int error;
1044     char *fn;
1045     int fd;
1046
1047     dpif->minor = -1;
1048     dpif->fd = -1;
1049     error = make_openvswitch_device(minor, &fn);
1050     if (error) {
1051         return error;
1052     }
1053
1054     fd = open(fn, O_RDONLY | O_NONBLOCK);
1055     if (fd < 0) {
1056         error = errno;
1057         VLOG_WARN("%s: open failed (%s)", fn, strerror(error));
1058         free(fn);
1059         return error;
1060     }
1061
1062     free(fn);
1063     dpif->minor = minor;
1064     dpif->fd = fd;
1065     return 0;
1066 }
1067 \f
1068 /* There is a tendency to construct odp_flow objects on the stack and to
1069  * forget to properly initialize their "actions" and "n_actions" members.
1070  * When this happens, we get memory corruption because the kernel
1071  * writes through the random pointer that is in the "actions" member.
1072  *
1073  * This function attempts to combat the problem by:
1074  *
1075  *      - Forcing a segfault if "actions" points to an invalid region (instead
1076  *        of just getting back EFAULT, which can be easily missed in the log).
1077  *
1078  *      - Storing a distinctive value that is likely to cause an
1079  *        easy-to-identify error later if it is dereferenced, etc.
1080  *
1081  *      - Triggering a warning on uninitialized memory from Valgrind if
1082  *        "actions" or "n_actions" was not initialized.
1083  */
1084 static void
1085 check_rw_odp_flow(struct odp_flow *flow)
1086 {
1087     if (flow->n_actions) {
1088         memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]);
1089     }
1090 }