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