lib/eal/common/eal_common_proc.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2018 Intel Corporation
   3  */
   4
   5 #include <dirent.h>
   6 #include <errno.h>
   7 #include <fcntl.h>
   8 #include <fnmatch.h>
   9 #include <inttypes.h>
  10 #include <libgen.h>
  11 #include <limits.h>
  12 #include <pthread.h>
  13 #include <stdio.h>
  14 #include <stdlib.h>
  15 #include <string.h>
  16 #include <sys/file.h>
  17 #include <sys/time.h>
  18 #include <sys/socket.h>
  19 #include <sys/un.h>
  20 #include <unistd.h>
  21
  22 #include <rte_alarm.h>
  23 #include <rte_common.h>
  24 #include <rte_cycles.h>
  25 #include <rte_eal.h>
  26 #include <rte_errno.h>
  27 #include <rte_lcore.h>
  28 #include <rte_log.h>
  29
  30 #include "eal_memcfg.h"
  31 #include "eal_private.h"
  32 #include "eal_filesystem.h"
  33 #include "eal_internal_cfg.h"
  34
  35 static int mp_fd = -1;
  36 static pthread_t mp_handle_tid;
  37 static char mp_filter[PATH_MAX];   /* Filter for secondary process sockets */
  38 static char mp_dir_path[PATH_MAX]; /* The directory path for all mp sockets */
  39 static pthread_mutex_t mp_mutex_action = PTHREAD_MUTEX_INITIALIZER;
  40 static char peer_name[PATH_MAX];
  41
  42 struct action_entry {
  43         TAILQ_ENTRY(action_entry) next;
  44         char action_name[RTE_MP_MAX_NAME_LEN];
  45         rte_mp_t action;
  46 };
  47
  48 /** Double linked list of actions. */
  49 TAILQ_HEAD(action_entry_list, action_entry);
  50
  51 static struct action_entry_list action_entry_list =
  52         TAILQ_HEAD_INITIALIZER(action_entry_list);
  53
  54 enum mp_type {
  55         MP_MSG, /* Share message with peers, will not block */
  56         MP_REQ, /* Request for information, Will block for a reply */
  57         MP_REP, /* Response to previously-received request */
  58         MP_IGN, /* Response telling requester to ignore this response */
  59 };
  60
  61 struct mp_msg_internal {
  62         int type;
  63         struct rte_mp_msg msg;
  64 };
  65
  66 struct async_request_param {
  67         rte_mp_async_reply_t clb;
  68         struct rte_mp_reply user_reply;
  69         struct timespec end;
  70         int n_responses_processed;
  71 };
  72
  73 struct pending_request {
  74         TAILQ_ENTRY(pending_request) next;
  75         enum {
  76                 REQUEST_TYPE_SYNC,
  77                 REQUEST_TYPE_ASYNC
  78         } type;
  79         char dst[PATH_MAX];
  80         struct rte_mp_msg *request;
  81         struct rte_mp_msg *reply;
  82         int reply_received;
  83         RTE_STD_C11
  84         union {
  85                 struct {
  86                         struct async_request_param *param;
  87                 } async;
  88                 struct {
  89                         pthread_cond_t cond;
  90                 } sync;
  91         };
  92 };
  93
  94 TAILQ_HEAD(pending_request_list, pending_request);
  95
  96 static struct {
  97         struct pending_request_list requests;
  98         pthread_mutex_t lock;
  99 } pending_requests = {
 100         .requests = TAILQ_HEAD_INITIALIZER(pending_requests.requests),
 101         .lock = PTHREAD_MUTEX_INITIALIZER,
 102         /**< used in async requests only */
 103 };
 104
 105 /* forward declarations */
 106 static int
 107 mp_send(struct rte_mp_msg *msg, const char *peer, int type);
 108
 109 /* for use with alarm callback */
 110 static void
 111 async_reply_handle(void *arg);
 112
 113 /* for use with process_msg */
 114 static struct pending_request *
 115 async_reply_handle_thread_unsafe(void *arg);
 116
 117 static void
 118 trigger_async_action(struct pending_request *req);
 119
 120 static struct pending_request *
 121 find_pending_request(const char *dst, const char *act_name)
 122 {
 123         struct pending_request *r;
 124
 125         TAILQ_FOREACH(r, &pending_requests.requests, next) {
 126                 if (!strcmp(r->dst, dst) &&
 127                     !strcmp(r->request->name, act_name))
 128                         break;
 129         }
 130
 131         return r;
 132 }
 133
 134 static void
 135 create_socket_path(const char *name, char *buf, int len)
 136 {
 137         const char *prefix = eal_mp_socket_path();
 138
 139         if (strlen(name) > 0)
 140                 snprintf(buf, len, "%s_%s", prefix, name);
 141         else
 142                 strlcpy(buf, prefix, len);
 143 }
 144
 145 int
 146 rte_eal_primary_proc_alive(const char *config_file_path)
 147 {
 148         int config_fd;
 149
 150         if (config_file_path)
 151                 config_fd = open(config_file_path, O_RDONLY);
 152         else {
 153                 const char *path;
 154
 155                 path = eal_runtime_config_path();
 156                 config_fd = open(path, O_RDONLY);
 157         }
 158         if (config_fd < 0)
 159                 return 0;
 160
 161         int ret = lockf(config_fd, F_TEST, 0);
 162         close(config_fd);
 163
 164         return !!ret;
 165 }
 166
 167 static struct action_entry *
 168 find_action_entry_by_name(const char *name)
 169 {
 170         struct action_entry *entry;
 171
 172         TAILQ_FOREACH(entry, &action_entry_list, next) {
 173                 if (strncmp(entry->action_name, name, RTE_MP_MAX_NAME_LEN) == 0)
 174                         break;
 175         }
 176
 177         return entry;
 178 }
 179
 180 static int
 181 validate_action_name(const char *name)
 182 {
 183         if (name == NULL) {
 184                 RTE_LOG(ERR, EAL, "Action name cannot be NULL\n");
 185                 rte_errno = EINVAL;
 186                 return -1;
 187         }
 188         if (strnlen(name, RTE_MP_MAX_NAME_LEN) == 0) {
 189                 RTE_LOG(ERR, EAL, "Length of action name is zero\n");
 190                 rte_errno = EINVAL;
 191                 return -1;
 192         }
 193         if (strnlen(name, RTE_MP_MAX_NAME_LEN) == RTE_MP_MAX_NAME_LEN) {
 194                 rte_errno = E2BIG;
 195                 return -1;
 196         }
 197         return 0;
 198 }
 199
 200 int
 201 rte_mp_action_register(const char *name, rte_mp_t action)
 202 {
 203         struct action_entry *entry;
 204         const struct internal_config *internal_conf =
 205                 eal_get_internal_configuration();
 206
 207         if (validate_action_name(name) != 0)
 208                 return -1;
 209
 210         if (internal_conf->no_shconf) {
 211                 RTE_LOG(DEBUG, EAL, "No shared files mode enabled, IPC is disabled\n");
 212                 rte_errno = ENOTSUP;
 213                 return -1;
 214         }
 215
 216         entry = malloc(sizeof(struct action_entry));
 217         if (entry == NULL) {
 218                 rte_errno = ENOMEM;
 219                 return -1;
 220         }
 221         strlcpy(entry->action_name, name, sizeof(entry->action_name));
 222         entry->action = action;
 223
 224         pthread_mutex_lock(&mp_mutex_action);
 225         if (find_action_entry_by_name(name) != NULL) {
 226                 pthread_mutex_unlock(&mp_mutex_action);
 227                 rte_errno = EEXIST;
 228                 free(entry);
 229                 return -1;
 230         }
 231         TAILQ_INSERT_TAIL(&action_entry_list, entry, next);
 232         pthread_mutex_unlock(&mp_mutex_action);
 233         return 0;
 234 }
 235
 236 void
 237 rte_mp_action_unregister(const char *name)
 238 {
 239         struct action_entry *entry;
 240         const struct internal_config *internal_conf =
 241                 eal_get_internal_configuration();
 242
 243         if (validate_action_name(name) != 0)
 244                 return;
 245
 246         if (internal_conf->no_shconf) {
 247                 RTE_LOG(DEBUG, EAL, "No shared files mode enabled, IPC is disabled\n");
 248                 return;
 249         }
 250
 251         pthread_mutex_lock(&mp_mutex_action);
 252         entry = find_action_entry_by_name(name);
 253         if (entry == NULL) {
 254                 pthread_mutex_unlock(&mp_mutex_action);
 255                 return;
 256         }
 257         TAILQ_REMOVE(&action_entry_list, entry, next);
 258         pthread_mutex_unlock(&mp_mutex_action);
 259         free(entry);
 260 }
 261
 262 static int
 263 read_msg(struct mp_msg_internal *m, struct sockaddr_un *s)
 264 {
 265         int msglen;
 266         struct iovec iov;
 267         struct msghdr msgh;
 268         char control[CMSG_SPACE(sizeof(m->msg.fds))];
 269         struct cmsghdr *cmsg;
 270         int buflen = sizeof(*m) - sizeof(m->msg.fds);
 271
 272         memset(&msgh, 0, sizeof(msgh));
 273         iov.iov_base = m;
 274         iov.iov_len  = buflen;
 275
 276         msgh.msg_name = s;
 277         msgh.msg_namelen = sizeof(*s);
 278         msgh.msg_iov = &iov;
 279         msgh.msg_iovlen = 1;
 280         msgh.msg_control = control;
 281         msgh.msg_controllen = sizeof(control);
 282
 283 retry:
 284         msglen = recvmsg(mp_fd, &msgh, 0);
 285
 286         /* zero length message means socket was closed */
 287         if (msglen == 0)
 288                 return 0;
 289
 290         if (msglen < 0) {
 291                 if (errno == EINTR)
 292                         goto retry;
 293
 294                 RTE_LOG(ERR, EAL, "recvmsg failed, %s\n", strerror(errno));
 295                 return -1;
 296         }
 297
 298         if (msglen != buflen || (msgh.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
 299                 RTE_LOG(ERR, EAL, "truncated msg\n");
 300                 return -1;
 301         }
 302
 303         /* read auxiliary FDs if any */
 304         for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg != NULL;
 305                 cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
 306                 if ((cmsg->cmsg_level == SOL_SOCKET) &&
 307                         (cmsg->cmsg_type == SCM_RIGHTS)) {
 308                         memcpy(m->msg.fds, CMSG_DATA(cmsg), sizeof(m->msg.fds));
 309                         break;
 310                 }
 311         }
 312         /* sanity-check the response */
 313         if (m->msg.num_fds < 0 || m->msg.num_fds > RTE_MP_MAX_FD_NUM) {
 314                 RTE_LOG(ERR, EAL, "invalid number of fd's received\n");
 315                 return -1;
 316         }
 317         if (m->msg.len_param < 0 || m->msg.len_param > RTE_MP_MAX_PARAM_LEN) {
 318                 RTE_LOG(ERR, EAL, "invalid received data length\n");
 319                 return -1;
 320         }
 321         return msglen;
 322 }
 323
 324 static void
 325 process_msg(struct mp_msg_internal *m, struct sockaddr_un *s)
 326 {
 327         struct pending_request *pending_req;
 328         struct action_entry *entry;
 329         struct rte_mp_msg *msg = &m->msg;
 330         rte_mp_t action = NULL;
 331         const struct internal_config *internal_conf =
 332                 eal_get_internal_configuration();
 333
 334         RTE_LOG(DEBUG, EAL, "msg: %s\n", msg->name);
 335
 336         if (m->type == MP_REP || m->type == MP_IGN) {
 337                 struct pending_request *req = NULL;
 338
 339                 pthread_mutex_lock(&pending_requests.lock);
 340                 pending_req = find_pending_request(s->sun_path, msg->name);
 341                 if (pending_req) {
 342                         memcpy(pending_req->reply, msg, sizeof(*msg));
 343                         /* -1 indicates that we've been asked to ignore */
 344                         pending_req->reply_received =
 345                                 m->type == MP_REP ? 1 : -1;
 346
 347                         if (pending_req->type == REQUEST_TYPE_SYNC)
 348                                 pthread_cond_signal(&pending_req->sync.cond);
 349                         else if (pending_req->type == REQUEST_TYPE_ASYNC)
 350                                 req = async_reply_handle_thread_unsafe(
 351                                                 pending_req);
 352                 } else
 353                         RTE_LOG(ERR, EAL, "Drop mp reply: %s\n", msg->name);
 354                 pthread_mutex_unlock(&pending_requests.lock);
 355
 356                 if (req != NULL)
 357                         trigger_async_action(req);
 358                 return;
 359         }
 360
 361         pthread_mutex_lock(&mp_mutex_action);
 362         entry = find_action_entry_by_name(msg->name);
 363         if (entry != NULL)
 364                 action = entry->action;
 365         pthread_mutex_unlock(&mp_mutex_action);
 366
 367         if (!action) {
 368                 if (m->type == MP_REQ && !internal_conf->init_complete) {
 369                         /* if this is a request, and init is not yet complete,
 370                          * and callback wasn't registered, we should tell the
 371                          * requester to ignore our existence because we're not
 372                          * yet ready to process this request.
 373                          */
 374                         struct rte_mp_msg dummy;
 375
 376                         memset(&dummy, 0, sizeof(dummy));
 377                         strlcpy(dummy.name, msg->name, sizeof(dummy.name));
 378                         mp_send(&dummy, s->sun_path, MP_IGN);
 379                 } else {
 380                         RTE_LOG(ERR, EAL, "Cannot find action: %s\n",
 381                                 msg->name);
 382                 }
 383         } else if (action(msg, s->sun_path) < 0) {
 384                 RTE_LOG(ERR, EAL, "Fail to handle message: %s\n", msg->name);
 385         }
 386 }
 387
 388 static void *
 389 mp_handle(void *arg __rte_unused)
 390 {
 391         struct mp_msg_internal msg;
 392         struct sockaddr_un sa;
 393
 394         while (mp_fd >= 0) {
 395                 int ret;
 396
 397                 ret = read_msg(&msg, &sa);
 398                 if (ret <= 0)
 399                         break;
 400
 401                 process_msg(&msg, &sa);
 402         }
 403
 404         return NULL;
 405 }
 406
 407 static int
 408 timespec_cmp(const struct timespec *a, const struct timespec *b)
 409 {
 410         if (a->tv_sec < b->tv_sec)
 411                 return -1;
 412         if (a->tv_sec > b->tv_sec)
 413                 return 1;
 414         if (a->tv_nsec < b->tv_nsec)
 415                 return -1;
 416         if (a->tv_nsec > b->tv_nsec)
 417                 return 1;
 418         return 0;
 419 }
 420
 421 enum async_action {
 422         ACTION_FREE, /**< free the action entry, but don't trigger callback */
 423         ACTION_TRIGGER /**< trigger callback, then free action entry */
 424 };
 425
 426 static enum async_action
 427 process_async_request(struct pending_request *sr, const struct timespec *now)
 428 {
 429         struct async_request_param *param;
 430         struct rte_mp_reply *reply;
 431         bool timeout, last_msg;
 432
 433         param = sr->async.param;
 434         reply = &param->user_reply;
 435
 436         /* did we timeout? */
 437         timeout = timespec_cmp(&param->end, now) <= 0;
 438
 439         /* if we received a response, adjust relevant data and copy message. */
 440         if (sr->reply_received == 1 && sr->reply) {
 441                 struct rte_mp_msg *msg, *user_msgs, *tmp;
 442
 443                 msg = sr->reply;
 444                 user_msgs = reply->msgs;
 445
 446                 tmp = realloc(user_msgs, sizeof(*msg) *
 447                                 (reply->nb_received + 1));
 448                 if (!tmp) {
 449                         RTE_LOG(ERR, EAL, "Fail to alloc reply for request %s:%s\n",
 450                                 sr->dst, sr->request->name);
 451                         /* this entry is going to be removed and its message
 452                          * dropped, but we don't want to leak memory, so
 453                          * continue.
 454                          */
 455                 } else {
 456                         user_msgs = tmp;
 457                         reply->msgs = user_msgs;
 458                         memcpy(&user_msgs[reply->nb_received],
 459                                         msg, sizeof(*msg));
 460                         reply->nb_received++;
 461                 }
 462
 463                 /* mark this request as processed */
 464                 param->n_responses_processed++;
 465         } else if (sr->reply_received == -1) {
 466                 /* we were asked to ignore this process */
 467                 reply->nb_sent--;
 468         } else if (timeout) {
 469                 /* count it as processed response, but don't increment
 470                  * nb_received.
 471                  */
 472                 param->n_responses_processed++;
 473         }
 474
 475         free(sr->reply);
 476
 477         last_msg = param->n_responses_processed == reply->nb_sent;
 478
 479         return last_msg ? ACTION_TRIGGER : ACTION_FREE;
 480 }
 481
 482 static void
 483 trigger_async_action(struct pending_request *sr)
 484 {
 485         struct async_request_param *param;
 486         struct rte_mp_reply *reply;
 487
 488         param = sr->async.param;
 489         reply = &param->user_reply;
 490
 491         param->clb(sr->request, reply);
 492
 493         /* clean up */
 494         free(sr->async.param->user_reply.msgs);
 495         free(sr->async.param);
 496         free(sr->request);
 497         free(sr);
 498 }
 499
 500 static struct pending_request *
 501 async_reply_handle_thread_unsafe(void *arg)
 502 {
 503         struct pending_request *req = (struct pending_request *)arg;
 504         enum async_action action;
 505         struct timespec ts_now;
 506
 507         if (clock_gettime(CLOCK_MONOTONIC, &ts_now) < 0) {
 508                 RTE_LOG(ERR, EAL, "Cannot get current time\n");
 509                 goto no_trigger;
 510         }
 511
 512         action = process_async_request(req, &ts_now);
 513
 514         TAILQ_REMOVE(&pending_requests.requests, req, next);
 515
 516         if (rte_eal_alarm_cancel(async_reply_handle, req) < 0) {
 517                 /* if we failed to cancel the alarm because it's already in
 518                  * progress, don't proceed because otherwise we will end up
 519                  * handling the same message twice.
 520                  */
 521                 if (rte_errno == EINPROGRESS) {
 522                         RTE_LOG(DEBUG, EAL, "Request handling is already in progress\n");
 523                         goto no_trigger;
 524                 }
 525                 RTE_LOG(ERR, EAL, "Failed to cancel alarm\n");
 526         }
 527
 528         if (action == ACTION_TRIGGER)
 529                 return req;
 530 no_trigger:
 531         free(req);
 532         return NULL;
 533 }
 534
 535 static void
 536 async_reply_handle(void *arg)
 537 {
 538         struct pending_request *req;
 539
 540         pthread_mutex_lock(&pending_requests.lock);
 541         req = async_reply_handle_thread_unsafe(arg);
 542         pthread_mutex_unlock(&pending_requests.lock);
 543
 544         if (req != NULL)
 545                 trigger_async_action(req);
 546 }
 547
 548 static int
 549 open_socket_fd(void)
 550 {
 551         struct sockaddr_un un;
 552
 553         peer_name[0] = '\0';
 554         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
 555                 snprintf(peer_name, sizeof(peer_name),
 556                                 "%d_%"PRIx64, getpid(), rte_rdtsc());
 557
 558         mp_fd = socket(AF_UNIX, SOCK_DGRAM, 0);
 559         if (mp_fd < 0) {
 560                 RTE_LOG(ERR, EAL, "failed to create unix socket\n");
 561                 return -1;
 562         }
 563
 564         memset(&un, 0, sizeof(un));
 565         un.sun_family = AF_UNIX;
 566
 567         create_socket_path(peer_name, un.sun_path, sizeof(un.sun_path));
 568
 569         unlink(un.sun_path); /* May still exist since last run */
 570
 571         if (bind(mp_fd, (struct sockaddr *)&un, sizeof(un)) < 0) {
 572                 RTE_LOG(ERR, EAL, "failed to bind %s: %s\n",
 573                         un.sun_path, strerror(errno));
 574                 close(mp_fd);
 575                 return -1;
 576         }
 577
 578         RTE_LOG(INFO, EAL, "Multi-process socket %s\n", un.sun_path);
 579         return mp_fd;
 580 }
 581
 582 static void
 583 close_socket_fd(int fd)
 584 {
 585         char path[PATH_MAX];
 586
 587         close(fd);
 588         create_socket_path(peer_name, path, sizeof(path));
 589         unlink(path);
 590 }
 591
 592 int
 593 rte_mp_channel_init(void)
 594 {
 595         char path[PATH_MAX];
 596         int dir_fd;
 597         const struct internal_config *internal_conf =
 598                 eal_get_internal_configuration();
 599
 600         /* in no shared files mode, we do not have secondary processes support,
 601          * so no need to initialize IPC.
 602          */
 603         if (internal_conf->no_shconf) {
 604                 RTE_LOG(DEBUG, EAL, "No shared files mode enabled, IPC will be disabled\n");
 605                 rte_errno = ENOTSUP;
 606                 return -1;
 607         }
 608
 609         /* create filter path */
 610         create_socket_path("*", path, sizeof(path));
 611         strlcpy(mp_filter, basename(path), sizeof(mp_filter));
 612
 613         /* path may have been modified, so recreate it */
 614         create_socket_path("*", path, sizeof(path));
 615         strlcpy(mp_dir_path, dirname(path), sizeof(mp_dir_path));
 616
 617         /* lock the directory */
 618         dir_fd = open(mp_dir_path, O_RDONLY);
 619         if (dir_fd < 0) {
 620                 RTE_LOG(ERR, EAL, "failed to open %s: %s\n",
 621                         mp_dir_path, strerror(errno));
 622                 return -1;
 623         }
 624
 625         if (flock(dir_fd, LOCK_EX)) {
 626                 RTE_LOG(ERR, EAL, "failed to lock %s: %s\n",
 627                         mp_dir_path, strerror(errno));
 628                 close(dir_fd);
 629                 return -1;
 630         }
 631
 632         if (open_socket_fd() < 0) {
 633                 close(dir_fd);
 634                 return -1;
 635         }
 636
 637         if (rte_ctrl_thread_create(&mp_handle_tid, "rte_mp_handle",
 638                         NULL, mp_handle, NULL) < 0) {
 639                 RTE_LOG(ERR, EAL, "failed to create mp thread: %s\n",
 640                         strerror(errno));
 641                 close(mp_fd);
 642                 close(dir_fd);
 643                 mp_fd = -1;
 644                 return -1;
 645         }
 646
 647         /* unlock the directory */
 648         flock(dir_fd, LOCK_UN);
 649         close(dir_fd);
 650
 651         return 0;
 652 }
 653
 654 void
 655 rte_mp_channel_cleanup(void)
 656 {
 657         int fd;
 658
 659         if (mp_fd < 0)
 660                 return;
 661
 662         fd = mp_fd;
 663         mp_fd = -1;
 664         pthread_cancel(mp_handle_tid);
 665         pthread_join(mp_handle_tid, NULL);
 666         close_socket_fd(fd);
 667 }
 668
 669 /**
 670  * Return -1, as fail to send message and it's caused by the local side.
 671  * Return 0, as fail to send message and it's caused by the remote side.
 672  * Return 1, as succeed to send message.
 673  *
 674  */
 675 static int
 676 send_msg(const char *dst_path, struct rte_mp_msg *msg, int type)
 677 {
 678         int snd;
 679         struct iovec iov;
 680         struct msghdr msgh;
 681         struct cmsghdr *cmsg;
 682         struct sockaddr_un dst;
 683         struct mp_msg_internal m;
 684         int fd_size = msg->num_fds * sizeof(int);
 685         char control[CMSG_SPACE(fd_size)];
 686
 687         m.type = type;
 688         memcpy(&m.msg, msg, sizeof(*msg));
 689
 690         memset(&dst, 0, sizeof(dst));
 691         dst.sun_family = AF_UNIX;
 692         strlcpy(dst.sun_path, dst_path, sizeof(dst.sun_path));
 693
 694         memset(&msgh, 0, sizeof(msgh));
 695         memset(control, 0, sizeof(control));
 696
 697         iov.iov_base = &m;
 698         iov.iov_len = sizeof(m) - sizeof(msg->fds);
 699
 700         msgh.msg_name = &dst;
 701         msgh.msg_namelen = sizeof(dst);
 702         msgh.msg_iov = &iov;
 703         msgh.msg_iovlen = 1;
 704         msgh.msg_control = control;
 705         msgh.msg_controllen = sizeof(control);
 706
 707         cmsg = CMSG_FIRSTHDR(&msgh);
 708         cmsg->cmsg_len = CMSG_LEN(fd_size);
 709         cmsg->cmsg_level = SOL_SOCKET;
 710         cmsg->cmsg_type = SCM_RIGHTS;
 711         memcpy(CMSG_DATA(cmsg), msg->fds, fd_size);
 712
 713         do {
 714                 snd = sendmsg(mp_fd, &msgh, 0);
 715         } while (snd < 0 && errno == EINTR);
 716
 717         if (snd < 0) {
 718                 rte_errno = errno;
 719                 /* Check if it caused by peer process exits */
 720                 if (errno == ECONNREFUSED &&
 721                                 rte_eal_process_type() == RTE_PROC_PRIMARY) {
 722                         unlink(dst_path);
 723                         return 0;
 724                 }
 725                 RTE_LOG(ERR, EAL, "failed to send to (%s) due to %s\n",
 726                         dst_path, strerror(errno));
 727                 return -1;
 728         }
 729
 730         return 1;
 731 }
 732
 733 static int
 734 mp_send(struct rte_mp_msg *msg, const char *peer, int type)
 735 {
 736         int dir_fd, ret = 0;
 737         DIR *mp_dir;
 738         struct dirent *ent;
 739
 740         if (!peer && (rte_eal_process_type() == RTE_PROC_SECONDARY))
 741                 peer = eal_mp_socket_path();
 742
 743         if (peer) {
 744                 if (send_msg(peer, msg, type) < 0)
 745                         return -1;
 746                 else
 747                         return 0;
 748         }
 749
 750         /* broadcast to all secondary processes */
 751         mp_dir = opendir(mp_dir_path);
 752         if (!mp_dir) {
 753                 RTE_LOG(ERR, EAL, "Unable to open directory %s\n",
 754                                 mp_dir_path);
 755                 rte_errno = errno;
 756                 return -1;
 757         }
 758
 759         dir_fd = dirfd(mp_dir);
 760         /* lock the directory to prevent processes spinning up while we send */
 761         if (flock(dir_fd, LOCK_SH)) {
 762                 RTE_LOG(ERR, EAL, "Unable to lock directory %s\n",
 763                         mp_dir_path);
 764                 rte_errno = errno;
 765                 closedir(mp_dir);
 766                 return -1;
 767         }
 768
 769         while ((ent = readdir(mp_dir))) {
 770                 char path[PATH_MAX];
 771
 772                 if (fnmatch(mp_filter, ent->d_name, 0) != 0)
 773                         continue;
 774
 775                 snprintf(path, sizeof(path), "%s/%s", mp_dir_path,
 776                          ent->d_name);
 777                 if (send_msg(path, msg, type) < 0)
 778                         ret = -1;
 779         }
 780         /* unlock the dir */
 781         flock(dir_fd, LOCK_UN);
 782
 783         /* dir_fd automatically closed on closedir */
 784         closedir(mp_dir);
 785         return ret;
 786 }
 787
 788 static int
 789 check_input(const struct rte_mp_msg *msg)
 790 {
 791         if (msg == NULL) {
 792                 RTE_LOG(ERR, EAL, "Msg cannot be NULL\n");
 793                 rte_errno = EINVAL;
 794                 return -1;
 795         }
 796
 797         if (validate_action_name(msg->name) != 0)
 798                 return -1;
 799
 800         if (msg->len_param < 0) {
 801                 RTE_LOG(ERR, EAL, "Message data length is negative\n");
 802                 rte_errno = EINVAL;
 803                 return -1;
 804         }
 805
 806         if (msg->num_fds < 0) {
 807                 RTE_LOG(ERR, EAL, "Number of fd's is negative\n");
 808                 rte_errno = EINVAL;
 809                 return -1;
 810         }
 811
 812         if (msg->len_param > RTE_MP_MAX_PARAM_LEN) {
 813                 RTE_LOG(ERR, EAL, "Message data is too long\n");
 814                 rte_errno = E2BIG;
 815                 return -1;
 816         }
 817
 818         if (msg->num_fds > RTE_MP_MAX_FD_NUM) {
 819                 RTE_LOG(ERR, EAL, "Cannot send more than %d FDs\n",
 820                         RTE_MP_MAX_FD_NUM);
 821                 rte_errno = E2BIG;
 822                 return -1;
 823         }
 824
 825         return 0;
 826 }
 827
 828 int
 829 rte_mp_sendmsg(struct rte_mp_msg *msg)
 830 {
 831         const struct internal_config *internal_conf =
 832                 eal_get_internal_configuration();
 833
 834         if (check_input(msg) != 0)
 835                 return -1;
 836
 837         if (internal_conf->no_shconf) {
 838                 RTE_LOG(DEBUG, EAL, "No shared files mode enabled, IPC is disabled\n");
 839                 rte_errno = ENOTSUP;
 840                 return -1;
 841         }
 842
 843         RTE_LOG(DEBUG, EAL, "sendmsg: %s\n", msg->name);
 844         return mp_send(msg, NULL, MP_MSG);
 845 }
 846
 847 static int
 848 mp_request_async(const char *dst, struct rte_mp_msg *req,
 849                 struct async_request_param *param, const struct timespec *ts)
 850 {
 851         struct rte_mp_msg *reply_msg;
 852         struct pending_request *pending_req, *exist;
 853         int ret = -1;
 854
 855         pending_req = calloc(1, sizeof(*pending_req));
 856         reply_msg = calloc(1, sizeof(*reply_msg));
 857         if (pending_req == NULL || reply_msg == NULL) {
 858                 RTE_LOG(ERR, EAL, "Could not allocate space for sync request\n");
 859                 rte_errno = ENOMEM;
 860                 ret = -1;
 861                 goto fail;
 862         }
 863
 864         pending_req->type = REQUEST_TYPE_ASYNC;
 865         strlcpy(pending_req->dst, dst, sizeof(pending_req->dst));
 866         pending_req->request = req;
 867         pending_req->reply = reply_msg;
 868         pending_req->async.param = param;
 869
 870         /* queue already locked by caller */
 871
 872         exist = find_pending_request(dst, req->name);
 873         if (exist) {
 874                 RTE_LOG(ERR, EAL, "A pending request %s:%s\n", dst, req->name);
 875                 rte_errno = EEXIST;
 876                 ret = -1;
 877                 goto fail;
 878         }
 879
 880         ret = send_msg(dst, req, MP_REQ);
 881         if (ret < 0) {
 882                 RTE_LOG(ERR, EAL, "Fail to send request %s:%s\n",
 883                         dst, req->name);
 884                 ret = -1;
 885                 goto fail;
 886         } else if (ret == 0) {
 887                 ret = 0;
 888                 goto fail;
 889         }
 890         param->user_reply.nb_sent++;
 891
 892         /* if alarm set fails, we simply ignore the reply */
 893         if (rte_eal_alarm_set(ts->tv_sec * 1000000 + ts->tv_nsec / 1000,
 894                               async_reply_handle, pending_req) < 0) {
 895                 RTE_LOG(ERR, EAL, "Fail to set alarm for request %s:%s\n",
 896                         dst, req->name);
 897                 ret = -1;
 898                 goto fail;
 899         }
 900         TAILQ_INSERT_TAIL(&pending_requests.requests, pending_req, next);
 901
 902         return 0;
 903 fail:
 904         free(pending_req);
 905         free(reply_msg);
 906         return ret;
 907 }
 908
 909 static int
 910 mp_request_sync(const char *dst, struct rte_mp_msg *req,
 911                struct rte_mp_reply *reply, const struct timespec *ts)
 912 {
 913         int ret;
 914         pthread_condattr_t attr;
 915         struct rte_mp_msg msg, *tmp;
 916         struct pending_request pending_req, *exist;
 917
 918         pending_req.type = REQUEST_TYPE_SYNC;
 919         pending_req.reply_received = 0;
 920         strlcpy(pending_req.dst, dst, sizeof(pending_req.dst));
 921         pending_req.request = req;
 922         pending_req.reply = &msg;
 923         pthread_condattr_init(&attr);
 924         pthread_condattr_setclock(&attr, CLOCK_MONOTONIC);
 925         pthread_cond_init(&pending_req.sync.cond, &attr);
 926
 927         exist = find_pending_request(dst, req->name);
 928         if (exist) {
 929                 RTE_LOG(ERR, EAL, "A pending request %s:%s\n", dst, req->name);
 930                 rte_errno = EEXIST;
 931                 return -1;
 932         }
 933
 934         ret = send_msg(dst, req, MP_REQ);
 935         if (ret < 0) {
 936                 RTE_LOG(ERR, EAL, "Fail to send request %s:%s\n",
 937                         dst, req->name);
 938                 return -1;
 939         } else if (ret == 0)
 940                 return 0;
 941
 942         TAILQ_INSERT_TAIL(&pending_requests.requests, &pending_req, next);
 943
 944         reply->nb_sent++;
 945
 946         do {
 947                 ret = pthread_cond_timedwait(&pending_req.sync.cond,
 948                                 &pending_requests.lock, ts);
 949         } while (ret != 0 && ret != ETIMEDOUT);
 950
 951         TAILQ_REMOVE(&pending_requests.requests, &pending_req, next);
 952
 953         if (pending_req.reply_received == 0) {
 954                 RTE_LOG(ERR, EAL, "Fail to recv reply for request %s:%s\n",
 955                         dst, req->name);
 956                 rte_errno = ETIMEDOUT;
 957                 return -1;
 958         }
 959         if (pending_req.reply_received == -1) {
 960                 RTE_LOG(DEBUG, EAL, "Asked to ignore response\n");
 961                 /* not receiving this message is not an error, so decrement
 962                  * number of sent messages
 963                  */
 964                 reply->nb_sent--;
 965                 return 0;
 966         }
 967
 968         tmp = realloc(reply->msgs, sizeof(msg) * (reply->nb_received + 1));
 969         if (!tmp) {
 970                 RTE_LOG(ERR, EAL, "Fail to alloc reply for request %s:%s\n",
 971                         dst, req->name);
 972                 rte_errno = ENOMEM;
 973                 return -1;
 974         }
 975         memcpy(&tmp[reply->nb_received], &msg, sizeof(msg));
 976         reply->msgs = tmp;
 977         reply->nb_received++;
 978         return 0;
 979 }
 980
 981 int
 982 rte_mp_request_sync(struct rte_mp_msg *req, struct rte_mp_reply *reply,
 983                 const struct timespec *ts)
 984 {
 985         int dir_fd, ret = -1;
 986         DIR *mp_dir;
 987         struct dirent *ent;
 988         struct timespec now, end;
 989         const struct internal_config *internal_conf =
 990                 eal_get_internal_configuration();
 991
 992         RTE_LOG(DEBUG, EAL, "request: %s\n", req->name);
 993
 994         reply->nb_sent = 0;
 995         reply->nb_received = 0;
 996         reply->msgs = NULL;
 997
 998         if (check_input(req) != 0)
 999                 goto end;
1000
1001         if (internal_conf->no_shconf) {
1002                 RTE_LOG(DEBUG, EAL, "No shared files mode enabled, IPC is disabled\n");
1003                 rte_errno = ENOTSUP;
1004                 return -1;
1005         }
1006
1007         if (clock_gettime(CLOCK_MONOTONIC, &now) < 0) {
1008                 RTE_LOG(ERR, EAL, "Failed to get current time\n");
1009                 rte_errno = errno;
1010                 goto end;
1011         }
1012
1013         end.tv_nsec = (now.tv_nsec + ts->tv_nsec) % 1000000000;
1014         end.tv_sec = now.tv_sec + ts->tv_sec +
1015                         (now.tv_nsec + ts->tv_nsec) / 1000000000;
1016
1017         /* for secondary process, send request to the primary process only */
1018         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1019                 pthread_mutex_lock(&pending_requests.lock);
1020                 ret = mp_request_sync(eal_mp_socket_path(), req, reply, &end);
1021                 pthread_mutex_unlock(&pending_requests.lock);
1022                 goto end;
1023         }
1024
1025         /* for primary process, broadcast request, and collect reply 1 by 1 */
1026         mp_dir = opendir(mp_dir_path);
1027         if (!mp_dir) {
1028                 RTE_LOG(ERR, EAL, "Unable to open directory %s\n", mp_dir_path);
1029                 rte_errno = errno;
1030                 goto end;
1031         }
1032
1033         dir_fd = dirfd(mp_dir);
1034         /* lock the directory to prevent processes spinning up while we send */
1035         if (flock(dir_fd, LOCK_SH)) {
1036                 RTE_LOG(ERR, EAL, "Unable to lock directory %s\n",
1037                         mp_dir_path);
1038                 rte_errno = errno;
1039                 goto close_end;
1040         }
1041
1042         pthread_mutex_lock(&pending_requests.lock);
1043         while ((ent = readdir(mp_dir))) {
1044                 char path[PATH_MAX];
1045
1046                 if (fnmatch(mp_filter, ent->d_name, 0) != 0)
1047                         continue;
1048
1049                 snprintf(path, sizeof(path), "%s/%s", mp_dir_path,
1050                          ent->d_name);
1051
1052                 /* unlocks the mutex while waiting for response,
1053                  * locks on receive
1054                  */
1055                 if (mp_request_sync(path, req, reply, &end))
1056                         goto unlock_end;
1057         }
1058         ret = 0;
1059
1060 unlock_end:
1061         pthread_mutex_unlock(&pending_requests.lock);
1062         /* unlock the directory */
1063         flock(dir_fd, LOCK_UN);
1064
1065 close_end:
1066         /* dir_fd automatically closed on closedir */
1067         closedir(mp_dir);
1068
1069 end:
1070         if (ret) {
1071                 free(reply->msgs);
1072                 reply->nb_received = 0;
1073                 reply->msgs = NULL;
1074         }
1075         return ret;
1076 }
1077
1078 int
1079 rte_mp_request_async(struct rte_mp_msg *req, const struct timespec *ts,
1080                 rte_mp_async_reply_t clb)
1081 {
1082         struct rte_mp_msg *copy;
1083         struct pending_request *dummy;
1084         struct async_request_param *param;
1085         struct rte_mp_reply *reply;
1086         int dir_fd, ret = 0;
1087         DIR *mp_dir;
1088         struct dirent *ent;
1089         struct timespec now;
1090         struct timespec *end;
1091         bool dummy_used = false;
1092         const struct internal_config *internal_conf =
1093                 eal_get_internal_configuration();
1094
1095         RTE_LOG(DEBUG, EAL, "request: %s\n", req->name);
1096
1097         if (check_input(req) != 0)
1098                 return -1;
1099
1100         if (internal_conf->no_shconf) {
1101                 RTE_LOG(DEBUG, EAL, "No shared files mode enabled, IPC is disabled\n");
1102                 rte_errno = ENOTSUP;
1103                 return -1;
1104         }
1105
1106         if (clock_gettime(CLOCK_MONOTONIC, &now) < 0) {
1107                 RTE_LOG(ERR, EAL, "Failed to get current time\n");
1108                 rte_errno = errno;
1109                 return -1;
1110         }
1111         copy = calloc(1, sizeof(*copy));
1112         dummy = calloc(1, sizeof(*dummy));
1113         param = calloc(1, sizeof(*param));
1114         if (copy == NULL || dummy == NULL || param == NULL) {
1115                 RTE_LOG(ERR, EAL, "Failed to allocate memory for async reply\n");
1116                 rte_errno = ENOMEM;
1117                 goto fail;
1118         }
1119
1120         /* copy message */
1121         memcpy(copy, req, sizeof(*copy));
1122
1123         param->n_responses_processed = 0;
1124         param->clb = clb;
1125         end = &param->end;
1126         reply = &param->user_reply;
1127
1128         end->tv_nsec = (now.tv_nsec + ts->tv_nsec) % 1000000000;
1129         end->tv_sec = now.tv_sec + ts->tv_sec +
1130                         (now.tv_nsec + ts->tv_nsec) / 1000000000;
1131         reply->nb_sent = 0;
1132         reply->nb_received = 0;
1133         reply->msgs = NULL;
1134
1135         /* we have to lock the request queue here, as we will be adding a bunch
1136          * of requests to the queue at once, and some of the replies may arrive
1137          * before we add all of the requests to the queue.
1138          */
1139         pthread_mutex_lock(&pending_requests.lock);
1140
1141         /* we have to ensure that callback gets triggered even if we don't send
1142          * anything, therefore earlier we have allocated a dummy request. fill
1143          * it, and put it on the queue if we don't send any requests.
1144          */
1145         dummy->type = REQUEST_TYPE_ASYNC;
1146         dummy->request = copy;
1147         dummy->reply = NULL;
1148         dummy->async.param = param;
1149         dummy->reply_received = 1; /* short-circuit the timeout */
1150
1151         /* for secondary process, send request to the primary process only */
1152         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1153                 ret = mp_request_async(eal_mp_socket_path(), copy, param, ts);
1154
1155                 /* if we didn't send anything, put dummy request on the queue */
1156                 if (ret == 0 && reply->nb_sent == 0) {
1157                         TAILQ_INSERT_TAIL(&pending_requests.requests, dummy,
1158                                         next);
1159                         dummy_used = true;
1160                 }
1161
1162                 pthread_mutex_unlock(&pending_requests.lock);
1163
1164                 /* if we couldn't send anything, clean up */
1165                 if (ret != 0)
1166                         goto fail;
1167                 return 0;
1168         }
1169
1170         /* for primary process, broadcast request */
1171         mp_dir = opendir(mp_dir_path);
1172         if (!mp_dir) {
1173                 RTE_LOG(ERR, EAL, "Unable to open directory %s\n", mp_dir_path);
1174                 rte_errno = errno;
1175                 goto unlock_fail;
1176         }
1177         dir_fd = dirfd(mp_dir);
1178
1179         /* lock the directory to prevent processes spinning up while we send */
1180         if (flock(dir_fd, LOCK_SH)) {
1181                 RTE_LOG(ERR, EAL, "Unable to lock directory %s\n",
1182                         mp_dir_path);
1183                 rte_errno = errno;
1184                 goto closedir_fail;
1185         }
1186
1187         while ((ent = readdir(mp_dir))) {
1188                 char path[PATH_MAX];
1189
1190                 if (fnmatch(mp_filter, ent->d_name, 0) != 0)
1191                         continue;
1192
1193                 snprintf(path, sizeof(path), "%s/%s", mp_dir_path,
1194                          ent->d_name);
1195
1196                 if (mp_request_async(path, copy, param, ts))
1197                         ret = -1;
1198         }
1199         /* if we didn't send anything, put dummy request on the queue */
1200         if (ret == 0 && reply->nb_sent == 0) {
1201                 TAILQ_INSERT_HEAD(&pending_requests.requests, dummy, next);
1202                 dummy_used = true;
1203         }
1204
1205         /* finally, unlock the queue */
1206         pthread_mutex_unlock(&pending_requests.lock);
1207
1208         /* unlock the directory */
1209         flock(dir_fd, LOCK_UN);
1210
1211         /* dir_fd automatically closed on closedir */
1212         closedir(mp_dir);
1213
1214         /* if dummy was unused, free it */
1215         if (!dummy_used)
1216                 free(dummy);
1217
1218         return ret;
1219 closedir_fail:
1220         closedir(mp_dir);
1221 unlock_fail:
1222         pthread_mutex_unlock(&pending_requests.lock);
1223 fail:
1224         free(dummy);
1225         free(param);
1226         free(copy);
1227         return -1;
1228 }
1229
1230 int
1231 rte_mp_reply(struct rte_mp_msg *msg, const char *peer)
1232 {
1233         RTE_LOG(DEBUG, EAL, "reply: %s\n", msg->name);
1234         const struct internal_config *internal_conf =
1235                 eal_get_internal_configuration();
1236
1237         if (check_input(msg) != 0)
1238                 return -1;
1239
1240         if (peer == NULL) {
1241                 RTE_LOG(ERR, EAL, "peer is not specified\n");
1242                 rte_errno = EINVAL;
1243                 return -1;
1244         }
1245
1246         if (internal_conf->no_shconf) {
1247                 RTE_LOG(DEBUG, EAL, "No shared files mode enabled, IPC is disabled\n");
1248                 return 0;
1249         }
1250
1251         return mp_send(msg, peer, MP_REP);
1252 }
1253
1254 /* Internally, the status of the mp feature is represented as a three-state:
1255  * - "unknown" as long as no secondary process attached to a primary process
1256  *   and there was no call to rte_mp_disable yet,
1257  * - "enabled" as soon as a secondary process attaches to a primary process,
1258  * - "disabled" when a primary process successfully called rte_mp_disable,
1259  */
1260 enum mp_status {
1261         MP_STATUS_UNKNOWN,
1262         MP_STATUS_DISABLED,
1263         MP_STATUS_ENABLED,
1264 };
1265
1266 static bool
1267 set_mp_status(enum mp_status status)
1268 {
1269         struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
1270         uint8_t expected;
1271         uint8_t desired;
1272
1273         expected = MP_STATUS_UNKNOWN;
1274         desired = status;
1275         if (__atomic_compare_exchange_n(&mcfg->mp_status, &expected, desired,
1276                         false, __ATOMIC_RELAXED, __ATOMIC_RELAXED))
1277                 return true;
1278
1279         return __atomic_load_n(&mcfg->mp_status, __ATOMIC_RELAXED) == desired;
1280 }
1281
1282 bool
1283 rte_mp_disable(void)
1284 {
1285         return set_mp_status(MP_STATUS_DISABLED);
1286 }
1287
1288 bool
1289 __rte_mp_enable(void)
1290 {
1291         return set_mp_status(MP_STATUS_ENABLED);
1292 }