lib/librte_vhost/socket.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <stdio.h>
   7 #include <stdbool.h>
   8 #include <limits.h>
   9 #include <stdlib.h>
  10 #include <unistd.h>
  11 #include <string.h>
  12 #include <sys/types.h>
  13 #include <sys/socket.h>
  14 #include <sys/un.h>
  15 #include <sys/queue.h>
  16 #include <errno.h>
  17 #include <fcntl.h>
  18 #include <pthread.h>
  19
  20 #include <rte_log.h>
  21
  22 #include "fd_man.h"
  23 #include "vhost.h"
  24 #include "vhost_user.h"
  25
  26
  27 TAILQ_HEAD(vhost_user_connection_list, vhost_user_connection);
  28
  29 /*
  30  * Every time rte_vhost_driver_register() is invoked, an associated
  31  * vhost_user_socket struct will be created.
  32  */
  33 struct vhost_user_socket {
  34         struct vhost_user_connection_list conn_list;
  35         pthread_mutex_t conn_mutex;
  36         char *path;
  37         int socket_fd;
  38         struct sockaddr_un un;
  39         bool is_server;
  40         bool reconnect;
  41         bool dequeue_zero_copy;
  42         bool iommu_support;
  43         bool use_builtin_virtio_net;
  44
  45         /*
  46          * The "supported_features" indicates the feature bits the
  47          * vhost driver supports. The "features" indicates the feature
  48          * bits after the rte_vhost_driver_features_disable/enable().
  49          * It is also the final feature bits used for vhost-user
  50          * features negotiation.
  51          */
  52         uint64_t supported_features;
  53         uint64_t features;
  54
  55         struct vhost_device_ops const *notify_ops;
  56 };
  57
  58 struct vhost_user_connection {
  59         struct vhost_user_socket *vsocket;
  60         int connfd;
  61         int vid;
  62
  63         TAILQ_ENTRY(vhost_user_connection) next;
  64 };
  65
  66 #define MAX_VHOST_SOCKET 1024
  67 struct vhost_user {
  68         struct vhost_user_socket *vsockets[MAX_VHOST_SOCKET];
  69         struct fdset fdset;
  70         int vsocket_cnt;
  71         pthread_mutex_t mutex;
  72 };
  73
  74 #define MAX_VIRTIO_BACKLOG 128
  75
  76 static void vhost_user_server_new_connection(int fd, void *data, int *remove);
  77 static void vhost_user_read_cb(int fd, void *dat, int *remove);
  78 static int create_unix_socket(struct vhost_user_socket *vsocket);
  79 static int vhost_user_start_client(struct vhost_user_socket *vsocket);
  80
  81 static struct vhost_user vhost_user = {
  82         .fdset = {
  83                 .fd = { [0 ... MAX_FDS - 1] = {-1, NULL, NULL, NULL, 0} },
  84                 .fd_mutex = PTHREAD_MUTEX_INITIALIZER,
  85                 .num = 0
  86         },
  87         .vsocket_cnt = 0,
  88         .mutex = PTHREAD_MUTEX_INITIALIZER,
  89 };
  90
  91 /* return bytes# of read on success or negative val on failure. */
  92 int
  93 read_fd_message(int sockfd, char *buf, int buflen, int *fds, int fd_num)
  94 {
  95         struct iovec iov;
  96         struct msghdr msgh;
  97         size_t fdsize = fd_num * sizeof(int);
  98         char control[CMSG_SPACE(fdsize)];
  99         struct cmsghdr *cmsg;
 100         int got_fds = 0;
 101         int ret;
 102
 103         memset(&msgh, 0, sizeof(msgh));
 104         iov.iov_base = buf;
 105         iov.iov_len  = buflen;
 106
 107         msgh.msg_iov = &iov;
 108         msgh.msg_iovlen = 1;
 109         msgh.msg_control = control;
 110         msgh.msg_controllen = sizeof(control);
 111
 112         ret = recvmsg(sockfd, &msgh, 0);
 113         if (ret <= 0) {
 114                 RTE_LOG(ERR, VHOST_CONFIG, "recvmsg failed\n");
 115                 return ret;
 116         }
 117
 118         if (msgh.msg_flags & (MSG_TRUNC | MSG_CTRUNC)) {
 119                 RTE_LOG(ERR, VHOST_CONFIG, "truncted msg\n");
 120                 return -1;
 121         }
 122
 123         for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg != NULL;
 124                 cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
 125                 if ((cmsg->cmsg_level == SOL_SOCKET) &&
 126                         (cmsg->cmsg_type == SCM_RIGHTS)) {
 127                         got_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
 128                         memcpy(fds, CMSG_DATA(cmsg), got_fds * sizeof(int));
 129                         break;
 130                 }
 131         }
 132
 133         /* Clear out unused file descriptors */
 134         while (got_fds < fd_num)
 135                 fds[got_fds++] = -1;
 136
 137         return ret;
 138 }
 139
 140 int
 141 send_fd_message(int sockfd, char *buf, int buflen, int *fds, int fd_num)
 142 {
 143
 144         struct iovec iov;
 145         struct msghdr msgh;
 146         size_t fdsize = fd_num * sizeof(int);
 147         char control[CMSG_SPACE(fdsize)];
 148         struct cmsghdr *cmsg;
 149         int ret;
 150
 151         memset(&msgh, 0, sizeof(msgh));
 152         iov.iov_base = buf;
 153         iov.iov_len = buflen;
 154
 155         msgh.msg_iov = &iov;
 156         msgh.msg_iovlen = 1;
 157
 158         if (fds && fd_num > 0) {
 159                 msgh.msg_control = control;
 160                 msgh.msg_controllen = sizeof(control);
 161                 cmsg = CMSG_FIRSTHDR(&msgh);
 162                 cmsg->cmsg_len = CMSG_LEN(fdsize);
 163                 cmsg->cmsg_level = SOL_SOCKET;
 164                 cmsg->cmsg_type = SCM_RIGHTS;
 165                 memcpy(CMSG_DATA(cmsg), fds, fdsize);
 166         } else {
 167                 msgh.msg_control = NULL;
 168                 msgh.msg_controllen = 0;
 169         }
 170
 171         do {
 172                 ret = sendmsg(sockfd, &msgh, 0);
 173         } while (ret < 0 && errno == EINTR);
 174
 175         if (ret < 0) {
 176                 RTE_LOG(ERR, VHOST_CONFIG,  "sendmsg error\n");
 177                 return ret;
 178         }
 179
 180         return ret;
 181 }
 182
 183 static void
 184 vhost_user_add_connection(int fd, struct vhost_user_socket *vsocket)
 185 {
 186         int vid;
 187         size_t size;
 188         struct vhost_user_connection *conn;
 189         int ret;
 190
 191         conn = malloc(sizeof(*conn));
 192         if (conn == NULL) {
 193                 close(fd);
 194                 return;
 195         }
 196
 197         vid = vhost_new_device();
 198         if (vid == -1) {
 199                 goto err;
 200         }
 201
 202         size = strnlen(vsocket->path, PATH_MAX);
 203         vhost_set_ifname(vid, vsocket->path, size);
 204
 205         vhost_set_builtin_virtio_net(vid, vsocket->use_builtin_virtio_net);
 206
 207         if (vsocket->dequeue_zero_copy)
 208                 vhost_enable_dequeue_zero_copy(vid);
 209
 210         RTE_LOG(INFO, VHOST_CONFIG, "new device, handle is %d\n", vid);
 211
 212         if (vsocket->notify_ops->new_connection) {
 213                 ret = vsocket->notify_ops->new_connection(vid);
 214                 if (ret < 0) {
 215                         RTE_LOG(ERR, VHOST_CONFIG,
 216                                 "failed to add vhost user connection with fd %d\n",
 217                                 fd);
 218                         goto err;
 219                 }
 220         }
 221
 222         conn->connfd = fd;
 223         conn->vsocket = vsocket;
 224         conn->vid = vid;
 225         ret = fdset_add(&vhost_user.fdset, fd, vhost_user_read_cb,
 226                         NULL, conn);
 227         if (ret < 0) {
 228                 RTE_LOG(ERR, VHOST_CONFIG,
 229                         "failed to add fd %d into vhost server fdset\n",
 230                         fd);
 231
 232                 if (vsocket->notify_ops->destroy_connection)
 233                         vsocket->notify_ops->destroy_connection(conn->vid);
 234
 235                 goto err;
 236         }
 237
 238         pthread_mutex_lock(&vsocket->conn_mutex);
 239         TAILQ_INSERT_TAIL(&vsocket->conn_list, conn, next);
 240         pthread_mutex_unlock(&vsocket->conn_mutex);
 241         return;
 242
 243 err:
 244         free(conn);
 245         close(fd);
 246 }
 247
 248 /* call back when there is new vhost-user connection from client  */
 249 static void
 250 vhost_user_server_new_connection(int fd, void *dat, int *remove __rte_unused)
 251 {
 252         struct vhost_user_socket *vsocket = dat;
 253
 254         fd = accept(fd, NULL, NULL);
 255         if (fd < 0)
 256                 return;
 257
 258         RTE_LOG(INFO, VHOST_CONFIG, "new vhost user connection is %d\n", fd);
 259         vhost_user_add_connection(fd, vsocket);
 260 }
 261
 262 static void
 263 vhost_user_read_cb(int connfd, void *dat, int *remove)
 264 {
 265         struct vhost_user_connection *conn = dat;
 266         struct vhost_user_socket *vsocket = conn->vsocket;
 267         int ret;
 268
 269         ret = vhost_user_msg_handler(conn->vid, connfd);
 270         if (ret < 0) {
 271                 close(connfd);
 272                 *remove = 1;
 273                 vhost_destroy_device(conn->vid);
 274
 275                 if (vsocket->notify_ops->destroy_connection)
 276                         vsocket->notify_ops->destroy_connection(conn->vid);
 277
 278                 pthread_mutex_lock(&vsocket->conn_mutex);
 279                 TAILQ_REMOVE(&vsocket->conn_list, conn, next);
 280                 pthread_mutex_unlock(&vsocket->conn_mutex);
 281
 282                 free(conn);
 283
 284                 if (vsocket->reconnect) {
 285                         create_unix_socket(vsocket);
 286                         vhost_user_start_client(vsocket);
 287                 }
 288         }
 289 }
 290
 291 static int
 292 create_unix_socket(struct vhost_user_socket *vsocket)
 293 {
 294         int fd;
 295         struct sockaddr_un *un = &vsocket->un;
 296
 297         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 298         if (fd < 0)
 299                 return -1;
 300         RTE_LOG(INFO, VHOST_CONFIG, "vhost-user %s: socket created, fd: %d\n",
 301                 vsocket->is_server ? "server" : "client", fd);
 302
 303         if (!vsocket->is_server && fcntl(fd, F_SETFL, O_NONBLOCK)) {
 304                 RTE_LOG(ERR, VHOST_CONFIG,
 305                         "vhost-user: can't set nonblocking mode for socket, fd: "
 306                         "%d (%s)\n", fd, strerror(errno));
 307                 close(fd);
 308                 return -1;
 309         }
 310
 311         memset(un, 0, sizeof(*un));
 312         un->sun_family = AF_UNIX;
 313         strncpy(un->sun_path, vsocket->path, sizeof(un->sun_path));
 314         un->sun_path[sizeof(un->sun_path) - 1] = '\0';
 315
 316         vsocket->socket_fd = fd;
 317         return 0;
 318 }
 319
 320 static int
 321 vhost_user_start_server(struct vhost_user_socket *vsocket)
 322 {
 323         int ret;
 324         int fd = vsocket->socket_fd;
 325         const char *path = vsocket->path;
 326
 327         ret = bind(fd, (struct sockaddr *)&vsocket->un, sizeof(vsocket->un));
 328         if (ret < 0) {
 329                 RTE_LOG(ERR, VHOST_CONFIG,
 330                         "failed to bind to %s: %s; remove it and try again\n",
 331                         path, strerror(errno));
 332                 goto err;
 333         }
 334         RTE_LOG(INFO, VHOST_CONFIG, "bind to %s\n", path);
 335
 336         ret = listen(fd, MAX_VIRTIO_BACKLOG);
 337         if (ret < 0)
 338                 goto err;
 339
 340         ret = fdset_add(&vhost_user.fdset, fd, vhost_user_server_new_connection,
 341                   NULL, vsocket);
 342         if (ret < 0) {
 343                 RTE_LOG(ERR, VHOST_CONFIG,
 344                         "failed to add listen fd %d to vhost server fdset\n",
 345                         fd);
 346                 goto err;
 347         }
 348
 349         return 0;
 350
 351 err:
 352         close(fd);
 353         return -1;
 354 }
 355
 356 struct vhost_user_reconnect {
 357         struct sockaddr_un un;
 358         int fd;
 359         struct vhost_user_socket *vsocket;
 360
 361         TAILQ_ENTRY(vhost_user_reconnect) next;
 362 };
 363
 364 TAILQ_HEAD(vhost_user_reconnect_tailq_list, vhost_user_reconnect);
 365 struct vhost_user_reconnect_list {
 366         struct vhost_user_reconnect_tailq_list head;
 367         pthread_mutex_t mutex;
 368 };
 369
 370 static struct vhost_user_reconnect_list reconn_list;
 371 static pthread_t reconn_tid;
 372
 373 static int
 374 vhost_user_connect_nonblock(int fd, struct sockaddr *un, size_t sz)
 375 {
 376         int ret, flags;
 377
 378         ret = connect(fd, un, sz);
 379         if (ret < 0 && errno != EISCONN)
 380                 return -1;
 381
 382         flags = fcntl(fd, F_GETFL, 0);
 383         if (flags < 0) {
 384                 RTE_LOG(ERR, VHOST_CONFIG,
 385                         "can't get flags for connfd %d\n", fd);
 386                 return -2;
 387         }
 388         if ((flags & O_NONBLOCK) && fcntl(fd, F_SETFL, flags & ~O_NONBLOCK)) {
 389                 RTE_LOG(ERR, VHOST_CONFIG,
 390                                 "can't disable nonblocking on fd %d\n", fd);
 391                 return -2;
 392         }
 393         return 0;
 394 }
 395
 396 static void *
 397 vhost_user_client_reconnect(void *arg __rte_unused)
 398 {
 399         int ret;
 400         struct vhost_user_reconnect *reconn, *next;
 401
 402         while (1) {
 403                 pthread_mutex_lock(&reconn_list.mutex);
 404
 405                 /*
 406                  * An equal implementation of TAILQ_FOREACH_SAFE,
 407                  * which does not exist on all platforms.
 408                  */
 409                 for (reconn = TAILQ_FIRST(&reconn_list.head);
 410                      reconn != NULL; reconn = next) {
 411                         next = TAILQ_NEXT(reconn, next);
 412
 413                         ret = vhost_user_connect_nonblock(reconn->fd,
 414                                                 (struct sockaddr *)&reconn->un,
 415                                                 sizeof(reconn->un));
 416                         if (ret == -2) {
 417                                 close(reconn->fd);
 418                                 RTE_LOG(ERR, VHOST_CONFIG,
 419                                         "reconnection for fd %d failed\n",
 420                                         reconn->fd);
 421                                 goto remove_fd;
 422                         }
 423                         if (ret == -1)
 424                                 continue;
 425
 426                         RTE_LOG(INFO, VHOST_CONFIG,
 427                                 "%s: connected\n", reconn->vsocket->path);
 428                         vhost_user_add_connection(reconn->fd, reconn->vsocket);
 429 remove_fd:
 430                         TAILQ_REMOVE(&reconn_list.head, reconn, next);
 431                         free(reconn);
 432                 }
 433
 434                 pthread_mutex_unlock(&reconn_list.mutex);
 435                 sleep(1);
 436         }
 437
 438         return NULL;
 439 }
 440
 441 static int
 442 vhost_user_reconnect_init(void)
 443 {
 444         int ret;
 445         char thread_name[RTE_MAX_THREAD_NAME_LEN];
 446
 447         ret = pthread_mutex_init(&reconn_list.mutex, NULL);
 448         if (ret < 0) {
 449                 RTE_LOG(ERR, VHOST_CONFIG, "failed to initialize mutex");
 450                 return ret;
 451         }
 452         TAILQ_INIT(&reconn_list.head);
 453
 454         ret = pthread_create(&reconn_tid, NULL,
 455                              vhost_user_client_reconnect, NULL);
 456         if (ret != 0) {
 457                 RTE_LOG(ERR, VHOST_CONFIG, "failed to create reconnect thread");
 458                 if (pthread_mutex_destroy(&reconn_list.mutex)) {
 459                         RTE_LOG(ERR, VHOST_CONFIG,
 460                                 "failed to destroy reconnect mutex");
 461                 }
 462         } else {
 463                 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
 464                          "vhost-reconn");
 465
 466                 if (rte_thread_setname(reconn_tid, thread_name)) {
 467                         RTE_LOG(DEBUG, VHOST_CONFIG,
 468                                 "failed to set reconnect thread name");
 469                 }
 470         }
 471
 472         return ret;
 473 }
 474
 475 static int
 476 vhost_user_start_client(struct vhost_user_socket *vsocket)
 477 {
 478         int ret;
 479         int fd = vsocket->socket_fd;
 480         const char *path = vsocket->path;
 481         struct vhost_user_reconnect *reconn;
 482
 483         ret = vhost_user_connect_nonblock(fd, (struct sockaddr *)&vsocket->un,
 484                                           sizeof(vsocket->un));
 485         if (ret == 0) {
 486                 vhost_user_add_connection(fd, vsocket);
 487                 return 0;
 488         }
 489
 490         RTE_LOG(WARNING, VHOST_CONFIG,
 491                 "failed to connect to %s: %s\n",
 492                 path, strerror(errno));
 493
 494         if (ret == -2 || !vsocket->reconnect) {
 495                 close(fd);
 496                 return -1;
 497         }
 498
 499         RTE_LOG(INFO, VHOST_CONFIG, "%s: reconnecting...\n", path);
 500         reconn = malloc(sizeof(*reconn));
 501         if (reconn == NULL) {
 502                 RTE_LOG(ERR, VHOST_CONFIG,
 503                         "failed to allocate memory for reconnect\n");
 504                 close(fd);
 505                 return -1;
 506         }
 507         reconn->un = vsocket->un;
 508         reconn->fd = fd;
 509         reconn->vsocket = vsocket;
 510         pthread_mutex_lock(&reconn_list.mutex);
 511         TAILQ_INSERT_TAIL(&reconn_list.head, reconn, next);
 512         pthread_mutex_unlock(&reconn_list.mutex);
 513
 514         return 0;
 515 }
 516
 517 static struct vhost_user_socket *
 518 find_vhost_user_socket(const char *path)
 519 {
 520         int i;
 521
 522         for (i = 0; i < vhost_user.vsocket_cnt; i++) {
 523                 struct vhost_user_socket *vsocket = vhost_user.vsockets[i];
 524
 525                 if (!strcmp(vsocket->path, path))
 526                         return vsocket;
 527         }
 528
 529         return NULL;
 530 }
 531
 532 int
 533 rte_vhost_driver_disable_features(const char *path, uint64_t features)
 534 {
 535         struct vhost_user_socket *vsocket;
 536
 537         pthread_mutex_lock(&vhost_user.mutex);
 538         vsocket = find_vhost_user_socket(path);
 539
 540         /* Note that use_builtin_virtio_net is not affected by this function
 541          * since callers may want to selectively disable features of the
 542          * built-in vhost net device backend.
 543          */
 544
 545         if (vsocket)
 546                 vsocket->features &= ~features;
 547         pthread_mutex_unlock(&vhost_user.mutex);
 548
 549         return vsocket ? 0 : -1;
 550 }
 551
 552 int
 553 rte_vhost_driver_enable_features(const char *path, uint64_t features)
 554 {
 555         struct vhost_user_socket *vsocket;
 556
 557         pthread_mutex_lock(&vhost_user.mutex);
 558         vsocket = find_vhost_user_socket(path);
 559         if (vsocket) {
 560                 if ((vsocket->supported_features & features) != features) {
 561                         /*
 562                          * trying to enable features the driver doesn't
 563                          * support.
 564                          */
 565                         pthread_mutex_unlock(&vhost_user.mutex);
 566                         return -1;
 567                 }
 568                 vsocket->features |= features;
 569         }
 570         pthread_mutex_unlock(&vhost_user.mutex);
 571
 572         return vsocket ? 0 : -1;
 573 }
 574
 575 int
 576 rte_vhost_driver_set_features(const char *path, uint64_t features)
 577 {
 578         struct vhost_user_socket *vsocket;
 579
 580         pthread_mutex_lock(&vhost_user.mutex);
 581         vsocket = find_vhost_user_socket(path);
 582         if (vsocket) {
 583                 vsocket->supported_features = features;
 584                 vsocket->features = features;
 585
 586                 /* Anyone setting feature bits is implementing their own vhost
 587                  * device backend.
 588                  */
 589                 vsocket->use_builtin_virtio_net = false;
 590         }
 591         pthread_mutex_unlock(&vhost_user.mutex);
 592
 593         return vsocket ? 0 : -1;
 594 }
 595
 596 int
 597 rte_vhost_driver_get_features(const char *path, uint64_t *features)
 598 {
 599         struct vhost_user_socket *vsocket;
 600
 601         pthread_mutex_lock(&vhost_user.mutex);
 602         vsocket = find_vhost_user_socket(path);
 603         if (vsocket)
 604                 *features = vsocket->features;
 605         pthread_mutex_unlock(&vhost_user.mutex);
 606
 607         if (!vsocket) {
 608                 RTE_LOG(ERR, VHOST_CONFIG,
 609                         "socket file %s is not registered yet.\n", path);
 610                 return -1;
 611         } else {
 612                 return 0;
 613         }
 614 }
 615
 616 /*
 617  * Register a new vhost-user socket; here we could act as server
 618  * (the default case), or client (when RTE_VHOST_USER_CLIENT) flag
 619  * is set.
 620  */
 621 int
 622 rte_vhost_driver_register(const char *path, uint64_t flags)
 623 {
 624         int ret = -1;
 625         struct vhost_user_socket *vsocket;
 626
 627         if (!path)
 628                 return -1;
 629
 630         pthread_mutex_lock(&vhost_user.mutex);
 631
 632         if (vhost_user.vsocket_cnt == MAX_VHOST_SOCKET) {
 633                 RTE_LOG(ERR, VHOST_CONFIG,
 634                         "error: the number of vhost sockets reaches maximum\n");
 635                 goto out;
 636         }
 637
 638         vsocket = malloc(sizeof(struct vhost_user_socket));
 639         if (!vsocket)
 640                 goto out;
 641         memset(vsocket, 0, sizeof(struct vhost_user_socket));
 642         vsocket->path = strdup(path);
 643         if (vsocket->path == NULL) {
 644                 RTE_LOG(ERR, VHOST_CONFIG,
 645                         "error: failed to copy socket path string\n");
 646                 free(vsocket);
 647                 goto out;
 648         }
 649         TAILQ_INIT(&vsocket->conn_list);
 650         ret = pthread_mutex_init(&vsocket->conn_mutex, NULL);
 651         if (ret) {
 652                 RTE_LOG(ERR, VHOST_CONFIG,
 653                         "error: failed to init connection mutex\n");
 654                 goto out_free;
 655         }
 656         vsocket->dequeue_zero_copy = flags & RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
 657
 658         /*
 659          * Set the supported features correctly for the builtin vhost-user
 660          * net driver.
 661          *
 662          * Applications know nothing about features the builtin virtio net
 663          * driver (virtio_net.c) supports, thus it's not possible for them
 664          * to invoke rte_vhost_driver_set_features(). To workaround it, here
 665          * we set it unconditionally. If the application want to implement
 666          * another vhost-user driver (say SCSI), it should call the
 667          * rte_vhost_driver_set_features(), which will overwrite following
 668          * two values.
 669          */
 670         vsocket->use_builtin_virtio_net = true;
 671         vsocket->supported_features = VIRTIO_NET_SUPPORTED_FEATURES;
 672         vsocket->features           = VIRTIO_NET_SUPPORTED_FEATURES;
 673
 674         if (!(flags & RTE_VHOST_USER_IOMMU_SUPPORT)) {
 675                 vsocket->supported_features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
 676                 vsocket->features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
 677         }
 678
 679         if ((flags & RTE_VHOST_USER_CLIENT) != 0) {
 680                 vsocket->reconnect = !(flags & RTE_VHOST_USER_NO_RECONNECT);
 681                 if (vsocket->reconnect && reconn_tid == 0) {
 682                         if (vhost_user_reconnect_init() != 0)
 683                                 goto out_mutex;
 684                 }
 685         } else {
 686                 vsocket->is_server = true;
 687         }
 688         ret = create_unix_socket(vsocket);
 689         if (ret < 0) {
 690                 goto out_mutex;
 691         }
 692
 693         vhost_user.vsockets[vhost_user.vsocket_cnt++] = vsocket;
 694
 695         pthread_mutex_unlock(&vhost_user.mutex);
 696         return ret;
 697
 698 out_mutex:
 699         if (pthread_mutex_destroy(&vsocket->conn_mutex)) {
 700                 RTE_LOG(ERR, VHOST_CONFIG,
 701                         "error: failed to destroy connection mutex\n");
 702         }
 703 out_free:
 704         free(vsocket->path);
 705         free(vsocket);
 706 out:
 707         pthread_mutex_unlock(&vhost_user.mutex);
 708
 709         return ret;
 710 }
 711
 712 static bool
 713 vhost_user_remove_reconnect(struct vhost_user_socket *vsocket)
 714 {
 715         int found = false;
 716         struct vhost_user_reconnect *reconn, *next;
 717
 718         pthread_mutex_lock(&reconn_list.mutex);
 719
 720         for (reconn = TAILQ_FIRST(&reconn_list.head);
 721              reconn != NULL; reconn = next) {
 722                 next = TAILQ_NEXT(reconn, next);
 723
 724                 if (reconn->vsocket == vsocket) {
 725                         TAILQ_REMOVE(&reconn_list.head, reconn, next);
 726                         close(reconn->fd);
 727                         free(reconn);
 728                         found = true;
 729                         break;
 730                 }
 731         }
 732         pthread_mutex_unlock(&reconn_list.mutex);
 733         return found;
 734 }
 735
 736 /**
 737  * Unregister the specified vhost socket
 738  */
 739 int
 740 rte_vhost_driver_unregister(const char *path)
 741 {
 742         int i;
 743         int count;
 744         struct vhost_user_connection *conn, *next;
 745
 746         pthread_mutex_lock(&vhost_user.mutex);
 747
 748         for (i = 0; i < vhost_user.vsocket_cnt; i++) {
 749                 struct vhost_user_socket *vsocket = vhost_user.vsockets[i];
 750
 751                 if (!strcmp(vsocket->path, path)) {
 752                         if (vsocket->is_server) {
 753                                 fdset_del(&vhost_user.fdset, vsocket->socket_fd);
 754                                 close(vsocket->socket_fd);
 755                                 unlink(path);
 756                         } else if (vsocket->reconnect) {
 757                                 vhost_user_remove_reconnect(vsocket);
 758                         }
 759
 760                         pthread_mutex_lock(&vsocket->conn_mutex);
 761                         for (conn = TAILQ_FIRST(&vsocket->conn_list);
 762                              conn != NULL;
 763                              conn = next) {
 764                                 next = TAILQ_NEXT(conn, next);
 765
 766                                 fdset_del(&vhost_user.fdset, conn->connfd);
 767                                 RTE_LOG(INFO, VHOST_CONFIG,
 768                                         "free connfd = %d for device '%s'\n",
 769                                         conn->connfd, path);
 770                                 close(conn->connfd);
 771                                 vhost_destroy_device(conn->vid);
 772                                 TAILQ_REMOVE(&vsocket->conn_list, conn, next);
 773                                 free(conn);
 774                         }
 775                         pthread_mutex_unlock(&vsocket->conn_mutex);
 776
 777                         pthread_mutex_destroy(&vsocket->conn_mutex);
 778                         free(vsocket->path);
 779                         free(vsocket);
 780
 781                         count = --vhost_user.vsocket_cnt;
 782                         vhost_user.vsockets[i] = vhost_user.vsockets[count];
 783                         vhost_user.vsockets[count] = NULL;
 784                         pthread_mutex_unlock(&vhost_user.mutex);
 785
 786                         return 0;
 787                 }
 788         }
 789         pthread_mutex_unlock(&vhost_user.mutex);
 790
 791         return -1;
 792 }
 793
 794 /*
 795  * Register ops so that we can add/remove device to data core.
 796  */
 797 int
 798 rte_vhost_driver_callback_register(const char *path,
 799         struct vhost_device_ops const * const ops)
 800 {
 801         struct vhost_user_socket *vsocket;
 802
 803         pthread_mutex_lock(&vhost_user.mutex);
 804         vsocket = find_vhost_user_socket(path);
 805         if (vsocket)
 806                 vsocket->notify_ops = ops;
 807         pthread_mutex_unlock(&vhost_user.mutex);
 808
 809         return vsocket ? 0 : -1;
 810 }
 811
 812 struct vhost_device_ops const *
 813 vhost_driver_callback_get(const char *path)
 814 {
 815         struct vhost_user_socket *vsocket;
 816
 817         pthread_mutex_lock(&vhost_user.mutex);
 818         vsocket = find_vhost_user_socket(path);
 819         pthread_mutex_unlock(&vhost_user.mutex);
 820
 821         return vsocket ? vsocket->notify_ops : NULL;
 822 }
 823
 824 int
 825 rte_vhost_driver_start(const char *path)
 826 {
 827         struct vhost_user_socket *vsocket;
 828         static pthread_t fdset_tid;
 829
 830         pthread_mutex_lock(&vhost_user.mutex);
 831         vsocket = find_vhost_user_socket(path);
 832         pthread_mutex_unlock(&vhost_user.mutex);
 833
 834         if (!vsocket)
 835                 return -1;
 836
 837         if (fdset_tid == 0) {
 838                 int ret = pthread_create(&fdset_tid, NULL, fdset_event_dispatch,
 839                                      &vhost_user.fdset);
 840                 if (ret != 0)
 841                         RTE_LOG(ERR, VHOST_CONFIG,
 842                                 "failed to create fdset handling thread");
 843         }
 844
 845         if (vsocket->is_server)
 846                 return vhost_user_start_server(vsocket);
 847         else
 848                 return vhost_user_start_client(vsocket);
 849 }