lib/librte_vhost/socket.c

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