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 <limits.h>
   8 #include <stdlib.h>
   9 #include <unistd.h>
  10 #include <string.h>
  11 #include <sys/types.h>
  12 #include <sys/socket.h>
  13 #include <sys/un.h>
  14 #include <sys/queue.h>
  15 #include <errno.h>
  16 #include <fcntl.h>
  17 #include <pthread.h>
  18
  19 #include <rte_log.h>
  20
  21 #include "fd_man.h"
  22 #include "vhost.h"
  23 #include "vhost_user.h"
  24
  25
  26 TAILQ_HEAD(vhost_user_connection_list, vhost_user_connection);
  27
  28 /*
  29  * Every time rte_vhost_driver_register() is invoked, an associated
  30  * vhost_user_socket struct will be created.
  31  */
  32 struct vhost_user_socket {
  33         struct vhost_user_connection_list conn_list;
  34         pthread_mutex_t conn_mutex;
  35         char *path;
  36         int socket_fd;
  37         struct sockaddr_un un;
  38         bool is_server;
  39         bool reconnect;
  40         bool dequeue_zero_copy;
  41         bool iommu_support;
  42         bool use_builtin_virtio_net;
  43
  44         /*
  45          * The "supported_features" indicates the feature bits the
  46          * vhost driver supports. The "features" indicates the feature
  47          * bits after the rte_vhost_driver_features_disable/enable().
  48          * It is also the final feature bits used for vhost-user
  49          * features negotiation.
  50          */
  51         uint64_t supported_features;
  52         uint64_t features;
  53
  54         struct vhost_device_ops const *notify_ops;
  55 };
  56
  57 struct vhost_user_connection {
  58         struct vhost_user_socket *vsocket;
  59         int connfd;
  60         int vid;
  61
  62         TAILQ_ENTRY(vhost_user_connection) next;
  63 };
  64
  65 #define MAX_VHOST_SOCKET 1024
  66 struct vhost_user {
  67         struct vhost_user_socket *vsockets[MAX_VHOST_SOCKET];
  68         struct fdset fdset;
  69         int vsocket_cnt;
  70         pthread_mutex_t mutex;
  71 };
  72
  73 #define MAX_VIRTIO_BACKLOG 128
  74
  75 static void vhost_user_server_new_connection(int fd, void *data, int *remove);
  76 static void vhost_user_read_cb(int fd, void *dat, int *remove);
  77 static int create_unix_socket(struct vhost_user_socket *vsocket);
  78 static int vhost_user_start_client(struct vhost_user_socket *vsocket);
  79
  80 static struct vhost_user vhost_user = {
  81         .fdset = {
  82                 .fd = { [0 ... MAX_FDS - 1] = {-1, NULL, NULL, NULL, 0} },
  83                 .fd_mutex = PTHREAD_MUTEX_INITIALIZER,
  84                 .num = 0
  85         },
  86         .vsocket_cnt = 0,
  87         .mutex = PTHREAD_MUTEX_INITIALIZER,
  88 };
  89
  90 /* return bytes# of read on success or negative val on failure. */
  91 int
  92 read_fd_message(int sockfd, char *buf, int buflen, int *fds, int fd_num)
  93 {
  94         struct iovec iov;
  95         struct msghdr msgh;
  96         size_t fdsize = fd_num * sizeof(int);
  97         char control[CMSG_SPACE(fdsize)];
  98         struct cmsghdr *cmsg;
  99         int got_fds = 0;
 100         int ret;
 101
 102         memset(&msgh, 0, sizeof(msgh));
 103         iov.iov_base = buf;
 104         iov.iov_len  = buflen;
 105
 106         msgh.msg_iov = &iov;
 107         msgh.msg_iovlen = 1;
 108         msgh.msg_control = control;
 109         msgh.msg_controllen = sizeof(control);
 110
 111         ret = recvmsg(sockfd, &msgh, 0);
 112         if (ret <= 0) {
 113                 RTE_LOG(ERR, VHOST_CONFIG, "recvmsg failed\n");
 114                 return ret;
 115         }
 116
 117         if (msgh.msg_flags & (MSG_TRUNC | MSG_CTRUNC)) {
 118                 RTE_LOG(ERR, VHOST_CONFIG, "truncted msg\n");
 119                 return -1;
 120         }
 121
 122         for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg != NULL;
 123                 cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
 124                 if ((cmsg->cmsg_level == SOL_SOCKET) &&
 125                         (cmsg->cmsg_type == SCM_RIGHTS)) {
 126                         got_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
 127                         memcpy(fds, CMSG_DATA(cmsg), got_fds * sizeof(int));
 128                         break;
 129                 }
 130         }
 131
 132         /* Clear out unused file descriptors */
 133         while (got_fds < fd_num)
 134                 fds[got_fds++] = -1;
 135
 136         return ret;
 137 }
 138
 139 int
 140 send_fd_message(int sockfd, char *buf, int buflen, int *fds, int fd_num)
 141 {
 142
 143         struct iovec iov;
 144         struct msghdr msgh;
 145         size_t fdsize = fd_num * sizeof(int);
 146         char control[CMSG_SPACE(fdsize)];
 147         struct cmsghdr *cmsg;
 148         int ret;
 149
 150         memset(&msgh, 0, sizeof(msgh));
 151         iov.iov_base = buf;
 152         iov.iov_len = buflen;
 153
 154         msgh.msg_iov = &iov;
 155         msgh.msg_iovlen = 1;
 156
 157         if (fds && fd_num > 0) {
 158                 msgh.msg_control = control;
 159                 msgh.msg_controllen = sizeof(control);
 160                 cmsg = CMSG_FIRSTHDR(&msgh);
 161                 if (cmsg == NULL) {
 162                         RTE_LOG(ERR, VHOST_CONFIG, "cmsg == NULL\n");
 163                         errno = EINVAL;
 164                         return -1;
 165                 }
 166                 cmsg->cmsg_len = CMSG_LEN(fdsize);
 167                 cmsg->cmsg_level = SOL_SOCKET;
 168                 cmsg->cmsg_type = SCM_RIGHTS;
 169                 memcpy(CMSG_DATA(cmsg), fds, fdsize);
 170         } else {
 171                 msgh.msg_control = NULL;
 172                 msgh.msg_controllen = 0;
 173         }
 174
 175         do {
 176                 ret = sendmsg(sockfd, &msgh, MSG_NOSIGNAL);
 177         } while (ret < 0 && errno == EINTR);
 178
 179         if (ret < 0) {
 180                 RTE_LOG(ERR, VHOST_CONFIG,  "sendmsg error\n");
 181                 return ret;
 182         }
 183
 184         return ret;
 185 }
 186
 187 static void
 188 vhost_user_add_connection(int fd, struct vhost_user_socket *vsocket)
 189 {
 190         int vid;
 191         size_t size;
 192         struct vhost_user_connection *conn;
 193         int ret;
 194
 195         conn = malloc(sizeof(*conn));
 196         if (conn == NULL) {
 197                 close(fd);
 198                 return;
 199         }
 200
 201         vid = vhost_new_device();
 202         if (vid == -1) {
 203                 goto err;
 204         }
 205
 206         size = strnlen(vsocket->path, PATH_MAX);
 207         vhost_set_ifname(vid, vsocket->path, size);
 208
 209         vhost_set_builtin_virtio_net(vid, vsocket->use_builtin_virtio_net);
 210
 211         if (vsocket->dequeue_zero_copy)
 212                 vhost_enable_dequeue_zero_copy(vid);
 213
 214         RTE_LOG(INFO, VHOST_CONFIG, "new device, handle is %d\n", vid);
 215
 216         if (vsocket->notify_ops->new_connection) {
 217                 ret = vsocket->notify_ops->new_connection(vid);
 218                 if (ret < 0) {
 219                         RTE_LOG(ERR, VHOST_CONFIG,
 220                                 "failed to add vhost user connection with fd %d\n",
 221                                 fd);
 222                         goto err;
 223                 }
 224         }
 225
 226         conn->connfd = fd;
 227         conn->vsocket = vsocket;
 228         conn->vid = vid;
 229         ret = fdset_add(&vhost_user.fdset, fd, vhost_user_read_cb,
 230                         NULL, conn);
 231         if (ret < 0) {
 232                 RTE_LOG(ERR, VHOST_CONFIG,
 233                         "failed to add fd %d into vhost server fdset\n",
 234                         fd);
 235
 236                 if (vsocket->notify_ops->destroy_connection)
 237                         vsocket->notify_ops->destroy_connection(conn->vid);
 238
 239                 goto err;
 240         }
 241
 242         pthread_mutex_lock(&vsocket->conn_mutex);
 243         TAILQ_INSERT_TAIL(&vsocket->conn_list, conn, next);
 244         pthread_mutex_unlock(&vsocket->conn_mutex);
 245         return;
 246
 247 err:
 248         free(conn);
 249         close(fd);
 250 }
 251
 252 /* call back when there is new vhost-user connection from client  */
 253 static void
 254 vhost_user_server_new_connection(int fd, void *dat, int *remove __rte_unused)
 255 {
 256         struct vhost_user_socket *vsocket = dat;
 257
 258         fd = accept(fd, NULL, NULL);
 259         if (fd < 0)
 260                 return;
 261
 262         RTE_LOG(INFO, VHOST_CONFIG, "new vhost user connection is %d\n", fd);
 263         vhost_user_add_connection(fd, vsocket);
 264 }
 265
 266 static void
 267 vhost_user_read_cb(int connfd, void *dat, int *remove)
 268 {
 269         struct vhost_user_connection *conn = dat;
 270         struct vhost_user_socket *vsocket = conn->vsocket;
 271         int ret;
 272
 273         ret = vhost_user_msg_handler(conn->vid, connfd);
 274         if (ret < 0) {
 275                 close(connfd);
 276                 *remove = 1;
 277                 vhost_destroy_device(conn->vid);
 278
 279                 if (vsocket->notify_ops->destroy_connection)
 280                         vsocket->notify_ops->destroy_connection(conn->vid);
 281
 282                 pthread_mutex_lock(&vsocket->conn_mutex);
 283                 TAILQ_REMOVE(&vsocket->conn_list, conn, next);
 284                 pthread_mutex_unlock(&vsocket->conn_mutex);
 285
 286                 free(conn);
 287
 288                 if (vsocket->reconnect) {
 289                         create_unix_socket(vsocket);
 290                         vhost_user_start_client(vsocket);
 291                 }
 292         }
 293 }
 294
 295 static int
 296 create_unix_socket(struct vhost_user_socket *vsocket)
 297 {
 298         int fd;
 299         struct sockaddr_un *un = &vsocket->un;
 300
 301         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 302         if (fd < 0)
 303                 return -1;
 304         RTE_LOG(INFO, VHOST_CONFIG, "vhost-user %s: socket created, fd: %d\n",
 305                 vsocket->is_server ? "server" : "client", fd);
 306
 307         if (!vsocket->is_server && fcntl(fd, F_SETFL, O_NONBLOCK)) {
 308                 RTE_LOG(ERR, VHOST_CONFIG,
 309                         "vhost-user: can't set nonblocking mode for socket, fd: "
 310                         "%d (%s)\n", fd, strerror(errno));
 311                 close(fd);
 312                 return -1;
 313         }
 314
 315         memset(un, 0, sizeof(*un));
 316         un->sun_family = AF_UNIX;
 317         strncpy(un->sun_path, vsocket->path, sizeof(un->sun_path));
 318         un->sun_path[sizeof(un->sun_path) - 1] = '\0';
 319
 320         vsocket->socket_fd = fd;
 321         return 0;
 322 }
 323
 324 static int
 325 vhost_user_start_server(struct vhost_user_socket *vsocket)
 326 {
 327         int ret;
 328         int fd = vsocket->socket_fd;
 329         const char *path = vsocket->path;
 330
 331         /*
 332          * bind () may fail if the socket file with the same name already
 333          * exists. But the library obviously should not delete the file
 334          * provided by the user, since we can not be sure that it is not
 335          * being used by other applications. Moreover, many applications form
 336          * socket names based on user input, which is prone to errors.
 337          *
 338          * The user must ensure that the socket does not exist before
 339          * registering the vhost driver in server mode.
 340          */
 341         ret = bind(fd, (struct sockaddr *)&vsocket->un, sizeof(vsocket->un));
 342         if (ret < 0) {
 343                 RTE_LOG(ERR, VHOST_CONFIG,
 344                         "failed to bind to %s: %s; remove it and try again\n",
 345                         path, strerror(errno));
 346                 goto err;
 347         }
 348         RTE_LOG(INFO, VHOST_CONFIG, "bind to %s\n", path);
 349
 350         ret = listen(fd, MAX_VIRTIO_BACKLOG);
 351         if (ret < 0)
 352                 goto err;
 353
 354         ret = fdset_add(&vhost_user.fdset, fd, vhost_user_server_new_connection,
 355                   NULL, vsocket);
 356         if (ret < 0) {
 357                 RTE_LOG(ERR, VHOST_CONFIG,
 358                         "failed to add listen fd %d to vhost server fdset\n",
 359                         fd);
 360                 goto err;
 361         }
 362
 363         return 0;
 364
 365 err:
 366         close(fd);
 367         return -1;
 368 }
 369
 370 struct vhost_user_reconnect {
 371         struct sockaddr_un un;
 372         int fd;
 373         struct vhost_user_socket *vsocket;
 374
 375         TAILQ_ENTRY(vhost_user_reconnect) next;
 376 };
 377
 378 TAILQ_HEAD(vhost_user_reconnect_tailq_list, vhost_user_reconnect);
 379 struct vhost_user_reconnect_list {
 380         struct vhost_user_reconnect_tailq_list head;
 381         pthread_mutex_t mutex;
 382 };
 383
 384 static struct vhost_user_reconnect_list reconn_list;
 385 static pthread_t reconn_tid;
 386
 387 static int
 388 vhost_user_connect_nonblock(int fd, struct sockaddr *un, size_t sz)
 389 {
 390         int ret, flags;
 391
 392         ret = connect(fd, un, sz);
 393         if (ret < 0 && errno != EISCONN)
 394                 return -1;
 395
 396         flags = fcntl(fd, F_GETFL, 0);
 397         if (flags < 0) {
 398                 RTE_LOG(ERR, VHOST_CONFIG,
 399                         "can't get flags for connfd %d\n", fd);
 400                 return -2;
 401         }
 402         if ((flags & O_NONBLOCK) && fcntl(fd, F_SETFL, flags & ~O_NONBLOCK)) {
 403                 RTE_LOG(ERR, VHOST_CONFIG,
 404                                 "can't disable nonblocking on fd %d\n", fd);
 405                 return -2;
 406         }
 407         return 0;
 408 }
 409
 410 static void *
 411 vhost_user_client_reconnect(void *arg __rte_unused)
 412 {
 413         int ret;
 414         struct vhost_user_reconnect *reconn, *next;
 415
 416         while (1) {
 417                 pthread_mutex_lock(&reconn_list.mutex);
 418
 419                 /*
 420                  * An equal implementation of TAILQ_FOREACH_SAFE,
 421                  * which does not exist on all platforms.
 422                  */
 423                 for (reconn = TAILQ_FIRST(&reconn_list.head);
 424                      reconn != NULL; reconn = next) {
 425                         next = TAILQ_NEXT(reconn, next);
 426
 427                         ret = vhost_user_connect_nonblock(reconn->fd,
 428                                                 (struct sockaddr *)&reconn->un,
 429                                                 sizeof(reconn->un));
 430                         if (ret == -2) {
 431                                 close(reconn->fd);
 432                                 RTE_LOG(ERR, VHOST_CONFIG,
 433                                         "reconnection for fd %d failed\n",
 434                                         reconn->fd);
 435                                 goto remove_fd;
 436                         }
 437                         if (ret == -1)
 438                                 continue;
 439
 440                         RTE_LOG(INFO, VHOST_CONFIG,
 441                                 "%s: connected\n", reconn->vsocket->path);
 442                         vhost_user_add_connection(reconn->fd, reconn->vsocket);
 443 remove_fd:
 444                         TAILQ_REMOVE(&reconn_list.head, reconn, next);
 445                         free(reconn);
 446                 }
 447
 448                 pthread_mutex_unlock(&reconn_list.mutex);
 449                 sleep(1);
 450         }
 451
 452         return NULL;
 453 }
 454
 455 static int
 456 vhost_user_reconnect_init(void)
 457 {
 458         int ret;
 459         char thread_name[RTE_MAX_THREAD_NAME_LEN];
 460
 461         ret = pthread_mutex_init(&reconn_list.mutex, NULL);
 462         if (ret < 0) {
 463                 RTE_LOG(ERR, VHOST_CONFIG, "failed to initialize mutex");
 464                 return ret;
 465         }
 466         TAILQ_INIT(&reconn_list.head);
 467
 468         ret = pthread_create(&reconn_tid, NULL,
 469                              vhost_user_client_reconnect, NULL);
 470         if (ret != 0) {
 471                 RTE_LOG(ERR, VHOST_CONFIG, "failed to create reconnect thread");
 472                 if (pthread_mutex_destroy(&reconn_list.mutex)) {
 473                         RTE_LOG(ERR, VHOST_CONFIG,
 474                                 "failed to destroy reconnect mutex");
 475                 }
 476         } else {
 477                 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
 478                          "vhost-reconn");
 479
 480                 if (rte_thread_setname(reconn_tid, thread_name)) {
 481                         RTE_LOG(DEBUG, VHOST_CONFIG,
 482                                 "failed to set reconnect thread name");
 483                 }
 484         }
 485
 486         return ret;
 487 }
 488
 489 static int
 490 vhost_user_start_client(struct vhost_user_socket *vsocket)
 491 {
 492         int ret;
 493         int fd = vsocket->socket_fd;
 494         const char *path = vsocket->path;
 495         struct vhost_user_reconnect *reconn;
 496
 497         ret = vhost_user_connect_nonblock(fd, (struct sockaddr *)&vsocket->un,
 498                                           sizeof(vsocket->un));
 499         if (ret == 0) {
 500                 vhost_user_add_connection(fd, vsocket);
 501                 return 0;
 502         }
 503
 504         RTE_LOG(WARNING, VHOST_CONFIG,
 505                 "failed to connect to %s: %s\n",
 506                 path, strerror(errno));
 507
 508         if (ret == -2 || !vsocket->reconnect) {
 509                 close(fd);
 510                 return -1;
 511         }
 512
 513         RTE_LOG(INFO, VHOST_CONFIG, "%s: reconnecting...\n", path);
 514         reconn = malloc(sizeof(*reconn));
 515         if (reconn == NULL) {
 516                 RTE_LOG(ERR, VHOST_CONFIG,
 517                         "failed to allocate memory for reconnect\n");
 518                 close(fd);
 519                 return -1;
 520         }
 521         reconn->un = vsocket->un;
 522         reconn->fd = fd;
 523         reconn->vsocket = vsocket;
 524         pthread_mutex_lock(&reconn_list.mutex);
 525         TAILQ_INSERT_TAIL(&reconn_list.head, reconn, next);
 526         pthread_mutex_unlock(&reconn_list.mutex);
 527
 528         return 0;
 529 }
 530
 531 static struct vhost_user_socket *
 532 find_vhost_user_socket(const char *path)
 533 {
 534         int i;
 535
 536         for (i = 0; i < vhost_user.vsocket_cnt; i++) {
 537                 struct vhost_user_socket *vsocket = vhost_user.vsockets[i];
 538
 539                 if (!strcmp(vsocket->path, path))
 540                         return vsocket;
 541         }
 542
 543         return NULL;
 544 }
 545
 546 int
 547 rte_vhost_driver_disable_features(const char *path, uint64_t features)
 548 {
 549         struct vhost_user_socket *vsocket;
 550
 551         pthread_mutex_lock(&vhost_user.mutex);
 552         vsocket = find_vhost_user_socket(path);
 553
 554         /* Note that use_builtin_virtio_net is not affected by this function
 555          * since callers may want to selectively disable features of the
 556          * built-in vhost net device backend.
 557          */
 558
 559         if (vsocket)
 560                 vsocket->features &= ~features;
 561         pthread_mutex_unlock(&vhost_user.mutex);
 562
 563         return vsocket ? 0 : -1;
 564 }
 565
 566 int
 567 rte_vhost_driver_enable_features(const char *path, uint64_t features)
 568 {
 569         struct vhost_user_socket *vsocket;
 570
 571         pthread_mutex_lock(&vhost_user.mutex);
 572         vsocket = find_vhost_user_socket(path);
 573         if (vsocket) {
 574                 if ((vsocket->supported_features & features) != features) {
 575                         /*
 576                          * trying to enable features the driver doesn't
 577                          * support.
 578                          */
 579                         pthread_mutex_unlock(&vhost_user.mutex);
 580                         return -1;
 581                 }
 582                 vsocket->features |= features;
 583         }
 584         pthread_mutex_unlock(&vhost_user.mutex);
 585
 586         return vsocket ? 0 : -1;
 587 }
 588
 589 int
 590 rte_vhost_driver_set_features(const char *path, uint64_t features)
 591 {
 592         struct vhost_user_socket *vsocket;
 593
 594         pthread_mutex_lock(&vhost_user.mutex);
 595         vsocket = find_vhost_user_socket(path);
 596         if (vsocket) {
 597                 vsocket->supported_features = features;
 598                 vsocket->features = features;
 599
 600                 /* Anyone setting feature bits is implementing their own vhost
 601                  * device backend.
 602                  */
 603                 vsocket->use_builtin_virtio_net = false;
 604         }
 605         pthread_mutex_unlock(&vhost_user.mutex);
 606
 607         return vsocket ? 0 : -1;
 608 }
 609
 610 int
 611 rte_vhost_driver_get_features(const char *path, uint64_t *features)
 612 {
 613         struct vhost_user_socket *vsocket;
 614
 615         pthread_mutex_lock(&vhost_user.mutex);
 616         vsocket = find_vhost_user_socket(path);
 617         if (vsocket)
 618                 *features = vsocket->features;
 619         pthread_mutex_unlock(&vhost_user.mutex);
 620
 621         if (!vsocket) {
 622                 RTE_LOG(ERR, VHOST_CONFIG,
 623                         "socket file %s is not registered yet.\n", path);
 624                 return -1;
 625         } else {
 626                 return 0;
 627         }
 628 }
 629
 630 /*
 631  * Register a new vhost-user socket; here we could act as server
 632  * (the default case), or client (when RTE_VHOST_USER_CLIENT) flag
 633  * is set.
 634  */
 635 int
 636 rte_vhost_driver_register(const char *path, uint64_t flags)
 637 {
 638         int ret = -1;
 639         struct vhost_user_socket *vsocket;
 640
 641         if (!path)
 642                 return -1;
 643
 644         pthread_mutex_lock(&vhost_user.mutex);
 645
 646         if (vhost_user.vsocket_cnt == MAX_VHOST_SOCKET) {
 647                 RTE_LOG(ERR, VHOST_CONFIG,
 648                         "error: the number of vhost sockets reaches maximum\n");
 649                 goto out;
 650         }
 651
 652         vsocket = malloc(sizeof(struct vhost_user_socket));
 653         if (!vsocket)
 654                 goto out;
 655         memset(vsocket, 0, sizeof(struct vhost_user_socket));
 656         vsocket->path = strdup(path);
 657         if (vsocket->path == NULL) {
 658                 RTE_LOG(ERR, VHOST_CONFIG,
 659                         "error: failed to copy socket path string\n");
 660                 free(vsocket);
 661                 goto out;
 662         }
 663         TAILQ_INIT(&vsocket->conn_list);
 664         ret = pthread_mutex_init(&vsocket->conn_mutex, NULL);
 665         if (ret) {
 666                 RTE_LOG(ERR, VHOST_CONFIG,
 667                         "error: failed to init connection mutex\n");
 668                 goto out_free;
 669         }
 670         vsocket->dequeue_zero_copy = flags & RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
 671
 672         /*
 673          * Set the supported features correctly for the builtin vhost-user
 674          * net driver.
 675          *
 676          * Applications know nothing about features the builtin virtio net
 677          * driver (virtio_net.c) supports, thus it's not possible for them
 678          * to invoke rte_vhost_driver_set_features(). To workaround it, here
 679          * we set it unconditionally. If the application want to implement
 680          * another vhost-user driver (say SCSI), it should call the
 681          * rte_vhost_driver_set_features(), which will overwrite following
 682          * two values.
 683          */
 684         vsocket->use_builtin_virtio_net = true;
 685         vsocket->supported_features = VIRTIO_NET_SUPPORTED_FEATURES;
 686         vsocket->features           = VIRTIO_NET_SUPPORTED_FEATURES;
 687
 688         if (!(flags & RTE_VHOST_USER_IOMMU_SUPPORT)) {
 689                 vsocket->supported_features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
 690                 vsocket->features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
 691         }
 692
 693         if ((flags & RTE_VHOST_USER_CLIENT) != 0) {
 694                 vsocket->reconnect = !(flags & RTE_VHOST_USER_NO_RECONNECT);
 695                 if (vsocket->reconnect && reconn_tid == 0) {
 696                         if (vhost_user_reconnect_init() != 0)
 697                                 goto out_mutex;
 698                 }
 699         } else {
 700                 vsocket->is_server = true;
 701         }
 702         ret = create_unix_socket(vsocket);
 703         if (ret < 0) {
 704                 goto out_mutex;
 705         }
 706
 707         vhost_user.vsockets[vhost_user.vsocket_cnt++] = vsocket;
 708
 709         pthread_mutex_unlock(&vhost_user.mutex);
 710         return ret;
 711
 712 out_mutex:
 713         if (pthread_mutex_destroy(&vsocket->conn_mutex)) {
 714                 RTE_LOG(ERR, VHOST_CONFIG,
 715                         "error: failed to destroy connection mutex\n");
 716         }
 717 out_free:
 718         free(vsocket->path);
 719         free(vsocket);
 720 out:
 721         pthread_mutex_unlock(&vhost_user.mutex);
 722
 723         return ret;
 724 }
 725
 726 static bool
 727 vhost_user_remove_reconnect(struct vhost_user_socket *vsocket)
 728 {
 729         int found = false;
 730         struct vhost_user_reconnect *reconn, *next;
 731
 732         pthread_mutex_lock(&reconn_list.mutex);
 733
 734         for (reconn = TAILQ_FIRST(&reconn_list.head);
 735              reconn != NULL; reconn = next) {
 736                 next = TAILQ_NEXT(reconn, next);
 737
 738                 if (reconn->vsocket == vsocket) {
 739                         TAILQ_REMOVE(&reconn_list.head, reconn, next);
 740                         close(reconn->fd);
 741                         free(reconn);
 742                         found = true;
 743                         break;
 744                 }
 745         }
 746         pthread_mutex_unlock(&reconn_list.mutex);
 747         return found;
 748 }
 749
 750 /**
 751  * Unregister the specified vhost socket
 752  */
 753 int
 754 rte_vhost_driver_unregister(const char *path)
 755 {
 756         int i;
 757         int count;
 758         struct vhost_user_connection *conn, *next;
 759
 760         pthread_mutex_lock(&vhost_user.mutex);
 761
 762         for (i = 0; i < vhost_user.vsocket_cnt; i++) {
 763                 struct vhost_user_socket *vsocket = vhost_user.vsockets[i];
 764
 765                 if (!strcmp(vsocket->path, path)) {
 766                         if (vsocket->is_server) {
 767                                 fdset_del(&vhost_user.fdset, vsocket->socket_fd);
 768                                 close(vsocket->socket_fd);
 769                                 unlink(path);
 770                         } else if (vsocket->reconnect) {
 771                                 vhost_user_remove_reconnect(vsocket);
 772                         }
 773
 774                         pthread_mutex_lock(&vsocket->conn_mutex);
 775                         for (conn = TAILQ_FIRST(&vsocket->conn_list);
 776                              conn != NULL;
 777                              conn = next) {
 778                                 next = TAILQ_NEXT(conn, next);
 779
 780                                 fdset_del(&vhost_user.fdset, conn->connfd);
 781                                 RTE_LOG(INFO, VHOST_CONFIG,
 782                                         "free connfd = %d for device '%s'\n",
 783                                         conn->connfd, path);
 784                                 close(conn->connfd);
 785                                 vhost_destroy_device(conn->vid);
 786                                 TAILQ_REMOVE(&vsocket->conn_list, conn, next);
 787                                 free(conn);
 788                         }
 789                         pthread_mutex_unlock(&vsocket->conn_mutex);
 790
 791                         pthread_mutex_destroy(&vsocket->conn_mutex);
 792                         free(vsocket->path);
 793                         free(vsocket);
 794
 795                         count = --vhost_user.vsocket_cnt;
 796                         vhost_user.vsockets[i] = vhost_user.vsockets[count];
 797                         vhost_user.vsockets[count] = NULL;
 798                         pthread_mutex_unlock(&vhost_user.mutex);
 799
 800                         return 0;
 801                 }
 802         }
 803         pthread_mutex_unlock(&vhost_user.mutex);
 804
 805         return -1;
 806 }
 807
 808 /*
 809  * Register ops so that we can add/remove device to data core.
 810  */
 811 int
 812 rte_vhost_driver_callback_register(const char *path,
 813         struct vhost_device_ops const * const ops)
 814 {
 815         struct vhost_user_socket *vsocket;
 816
 817         pthread_mutex_lock(&vhost_user.mutex);
 818         vsocket = find_vhost_user_socket(path);
 819         if (vsocket)
 820                 vsocket->notify_ops = ops;
 821         pthread_mutex_unlock(&vhost_user.mutex);
 822
 823         return vsocket ? 0 : -1;
 824 }
 825
 826 struct vhost_device_ops const *
 827 vhost_driver_callback_get(const char *path)
 828 {
 829         struct vhost_user_socket *vsocket;
 830
 831         pthread_mutex_lock(&vhost_user.mutex);
 832         vsocket = find_vhost_user_socket(path);
 833         pthread_mutex_unlock(&vhost_user.mutex);
 834
 835         return vsocket ? vsocket->notify_ops : NULL;
 836 }
 837
 838 int
 839 rte_vhost_driver_start(const char *path)
 840 {
 841         struct vhost_user_socket *vsocket;
 842         static pthread_t fdset_tid;
 843         char thread_name[RTE_MAX_THREAD_NAME_LEN];
 844
 845         pthread_mutex_lock(&vhost_user.mutex);
 846         vsocket = find_vhost_user_socket(path);
 847         pthread_mutex_unlock(&vhost_user.mutex);
 848
 849         if (!vsocket)
 850                 return -1;
 851
 852         if (fdset_tid == 0) {
 853                 int ret = pthread_create(&fdset_tid, NULL, fdset_event_dispatch,
 854                                      &vhost_user.fdset);
 855                 if (ret != 0) {
 856                         RTE_LOG(ERR, VHOST_CONFIG,
 857                                 "failed to create fdset handling thread");
 858                         return -1;
 859                 } else {
 860                         snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
 861                                  "vhost-events");
 862
 863                         if (rte_thread_setname(fdset_tid, thread_name)) {
 864                                 RTE_LOG(DEBUG, VHOST_CONFIG,
 865                                         "failed to set vhost-event thread name");
 866                         }
 867                 }
 868         }
 869
 870         if (vsocket->is_server)
 871                 return vhost_user_start_server(vsocket);
 872         else
 873                 return vhost_user_start_client(vsocket);
 874 }