ceph/src/spdk/lib/rte_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                 if (ret)
 141                         RTE_LOG(ERR, VHOST_CONFIG, "recvmsg failed, %s\n", strerror(errno));
 142                 else
 143                         RTE_LOG(INFO, VHOST_CONFIG, "peer closed\n");
 144                 return ret;
 145         }
 146
 147         if (msgh.msg_flags & (MSG_TRUNC | MSG_CTRUNC)) {
 148                 RTE_LOG(ERR, VHOST_CONFIG, "truncted msg\n");
 149                 return -1;
 150         }
 151
 152         for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg != NULL;
 153                 cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
 154                 if ((cmsg->cmsg_level == SOL_SOCKET) &&
 155                         (cmsg->cmsg_type == SCM_RIGHTS)) {
 156                         memcpy(fds, CMSG_DATA(cmsg), fdsize);
 157                         break;
 158                 }
 159         }
 160
 161         return ret;
 162 }
 163
 164 int
 165 send_fd_message(int sockfd, char *buf, int buflen, int *fds, int fd_num)
 166 {
 167
 168         struct iovec iov;
 169         struct msghdr msgh;
 170         size_t fdsize = fd_num * sizeof(int);
 171         char control[CMSG_SPACE(fdsize)];
 172         struct cmsghdr *cmsg;
 173         int ret;
 174
 175         memset(&msgh, 0, sizeof(msgh));
 176         iov.iov_base = buf;
 177         iov.iov_len = buflen;
 178
 179         msgh.msg_iov = &iov;
 180         msgh.msg_iovlen = 1;
 181
 182         if (fds && fd_num > 0) {
 183                 msgh.msg_control = control;
 184                 msgh.msg_controllen = sizeof(control);
 185                 cmsg = CMSG_FIRSTHDR(&msgh);
 186                 if (cmsg == NULL) {
 187                         RTE_LOG(ERR, VHOST_CONFIG,  "cmsg == NULL\n");
 188                         errno = EINVAL;
 189                         return -1;
 190                 }
 191                 cmsg->cmsg_len = CMSG_LEN(fdsize);
 192                 cmsg->cmsg_level = SOL_SOCKET;
 193                 cmsg->cmsg_type = SCM_RIGHTS;
 194                 memcpy(CMSG_DATA(cmsg), fds, fdsize);
 195         } else {
 196                 msgh.msg_control = NULL;
 197                 msgh.msg_controllen = 0;
 198         }
 199
 200         do {
 201                 ret = sendmsg(sockfd, &msgh, 0);
 202         } while (ret < 0 && errno == EINTR);
 203
 204         if (ret < 0) {
 205                 RTE_LOG(ERR, VHOST_CONFIG,  "sendmsg error\n");
 206                 return ret;
 207         }
 208
 209         return ret;
 210 }
 211
 212 static void
 213 vhost_user_add_connection(int fd, struct vhost_user_socket *vsocket)
 214 {
 215         int vid;
 216         size_t size;
 217         struct vhost_user_connection *conn;
 218         int ret;
 219
 220         conn = malloc(sizeof(*conn));
 221         if (conn == NULL) {
 222                 close(fd);
 223                 return;
 224         }
 225
 226         vid = vhost_new_device(vsocket->features, vsocket->notify_ops);
 227         if (vid == -1) {
 228                 goto err;
 229         }
 230
 231         size = strnlen(vsocket->path, PATH_MAX);
 232         vhost_set_ifname(vid, vsocket->path, size);
 233
 234         if (vsocket->dequeue_zero_copy)
 235                 vhost_enable_dequeue_zero_copy(vid);
 236
 237         RTE_LOG(INFO, VHOST_CONFIG, "new device, handle is %d\n", vid);
 238
 239         if (vsocket->notify_ops->new_connection) {
 240                 ret = vsocket->notify_ops->new_connection(vid);
 241                 if (ret < 0) {
 242                         RTE_LOG(ERR, VHOST_CONFIG,
 243                                 "failed to add vhost user connection with fd %d\n",
 244                                 fd);
 245                         goto err;
 246                 }
 247         }
 248
 249         conn->connfd = fd;
 250         conn->vsocket = vsocket;
 251         conn->vid = vid;
 252         ret = fdset_add(&vhost_user.fdset, fd, vhost_user_read_cb,
 253                         NULL, conn);
 254         if (ret < 0) {
 255                 RTE_LOG(ERR, VHOST_CONFIG,
 256                         "failed to add fd %d into vhost server fdset\n",
 257                         fd);
 258
 259                 if (vsocket->notify_ops->destroy_connection)
 260                         vsocket->notify_ops->destroy_connection(conn->vid);
 261
 262                 goto err;
 263         }
 264
 265         pthread_mutex_lock(&vsocket->conn_mutex);
 266         TAILQ_INSERT_TAIL(&vsocket->conn_list, conn, next);
 267         pthread_mutex_unlock(&vsocket->conn_mutex);
 268         return;
 269
 270 err:
 271         free(conn);
 272         close(fd);
 273 }
 274
 275 /* call back when there is new vhost-user connection from client  */
 276 static void
 277 vhost_user_server_new_connection(int fd, void *dat, int *remove __rte_unused)
 278 {
 279         struct vhost_user_socket *vsocket = dat;
 280
 281         fd = accept(fd, NULL, NULL);
 282         if (fd < 0)
 283                 return;
 284
 285         RTE_LOG(INFO, VHOST_CONFIG, "new vhost user connection is %d\n", fd);
 286         vhost_user_add_connection(fd, vsocket);
 287 }
 288
 289 static void
 290 vhost_user_read_cb(int connfd, void *dat, int *remove)
 291 {
 292         struct vhost_user_connection *conn = dat;
 293         struct vhost_user_socket *vsocket = conn->vsocket;
 294         int ret;
 295
 296         ret = vhost_user_msg_handler(conn->vid, connfd);
 297         if (ret < 0) {
 298                 *remove = 1;
 299                 vhost_destroy_device(conn->vid);
 300
 301                 if (vsocket->notify_ops->destroy_connection)
 302                         vsocket->notify_ops->destroy_connection(conn->vid);
 303
 304                 pthread_mutex_lock(&vsocket->conn_mutex);
 305                 TAILQ_REMOVE(&vsocket->conn_list, conn, next);
 306                 if (conn->connfd != -1) {
 307                         close(conn->connfd);
 308                         conn->connfd = -1;
 309                 }
 310                 pthread_mutex_unlock(&vsocket->conn_mutex);
 311
 312                 free(conn);
 313
 314                 if (vsocket->reconnect) {
 315                         create_unix_socket(vsocket);
 316                         vhost_user_start_client(vsocket);
 317                 }
 318         }
 319 }
 320
 321 static int
 322 create_unix_socket(struct vhost_user_socket *vsocket)
 323 {
 324         int fd;
 325         struct sockaddr_un *un = &vsocket->un;
 326
 327         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 328         if (fd < 0)
 329                 return -1;
 330         RTE_LOG(INFO, VHOST_CONFIG, "vhost-user %s: socket created, fd: %d\n",
 331                 vsocket->is_server ? "server" : "client", fd);
 332
 333         if (!vsocket->is_server && fcntl(fd, F_SETFL, O_NONBLOCK)) {
 334                 RTE_LOG(ERR, VHOST_CONFIG,
 335                         "vhost-user: can't set nonblocking mode for socket, fd: "
 336                         "%d (%s)\n", fd, strerror(errno));
 337                 close(fd);
 338                 return -1;
 339         }
 340
 341         memset(un, 0, sizeof(*un));
 342         un->sun_family = AF_UNIX;
 343         strncpy(un->sun_path, vsocket->path, sizeof(un->sun_path));
 344         un->sun_path[sizeof(un->sun_path) - 1] = '\0';
 345
 346         vsocket->socket_fd = fd;
 347         return 0;
 348 }
 349
 350 static int
 351 vhost_user_start_server(struct vhost_user_socket *vsocket)
 352 {
 353         int ret;
 354         int fd = vsocket->socket_fd;
 355         const char *path = vsocket->path;
 356
 357         ret = bind(fd, (struct sockaddr *)&vsocket->un, sizeof(vsocket->un));
 358         if (ret < 0) {
 359                 RTE_LOG(ERR, VHOST_CONFIG,
 360                         "failed to bind to %s: %s; remove it and try again\n",
 361                         path, strerror(errno));
 362                 goto err;
 363         }
 364         RTE_LOG(INFO, VHOST_CONFIG, "bind to %s\n", path);
 365
 366         ret = listen(fd, MAX_VIRTIO_BACKLOG);
 367         if (ret < 0)
 368                 goto err;
 369
 370         ret = fdset_add(&vhost_user.fdset, fd, vhost_user_server_new_connection,
 371                   NULL, vsocket);
 372         if (ret < 0) {
 373                 RTE_LOG(ERR, VHOST_CONFIG,
 374                         "failed to add listen fd %d to vhost server fdset\n",
 375                         fd);
 376                 goto err;
 377         }
 378
 379         return 0;
 380
 381 err:
 382         close(fd);
 383         return -1;
 384 }
 385
 386 struct vhost_user_reconnect {
 387         struct sockaddr_un un;
 388         int fd;
 389         struct vhost_user_socket *vsocket;
 390
 391         TAILQ_ENTRY(vhost_user_reconnect) next;
 392 };
 393
 394 TAILQ_HEAD(vhost_user_reconnect_tailq_list, vhost_user_reconnect);
 395 struct vhost_user_reconnect_list {
 396         struct vhost_user_reconnect_tailq_list head;
 397         pthread_mutex_t mutex;
 398 };
 399
 400 static struct vhost_user_reconnect_list reconn_list;
 401 static pthread_t reconn_tid;
 402
 403 static int
 404 vhost_user_connect_nonblock(int fd, struct sockaddr *un, size_t sz)
 405 {
 406         int ret, flags;
 407
 408         ret = connect(fd, un, sz);
 409         if (ret < 0 && errno != EISCONN)
 410                 return -1;
 411
 412         flags = fcntl(fd, F_GETFL, 0);
 413         if (flags < 0) {
 414                 RTE_LOG(ERR, VHOST_CONFIG,
 415                         "can't get flags for connfd %d\n", fd);
 416                 return -2;
 417         }
 418         if ((flags & O_NONBLOCK) && fcntl(fd, F_SETFL, flags & ~O_NONBLOCK)) {
 419                 RTE_LOG(ERR, VHOST_CONFIG,
 420                                 "can't disable nonblocking on fd %d\n", fd);
 421                 return -2;
 422         }
 423         return 0;
 424 }
 425
 426 static void *
 427 vhost_user_client_reconnect(void *arg __rte_unused)
 428 {
 429         int ret;
 430         struct vhost_user_reconnect *reconn, *next;
 431
 432         while (1) {
 433                 pthread_mutex_lock(&reconn_list.mutex);
 434
 435                 /*
 436                  * An equal implementation of TAILQ_FOREACH_SAFE,
 437                  * which does not exist on all platforms.
 438                  */
 439                 for (reconn = TAILQ_FIRST(&reconn_list.head);
 440                      reconn != NULL; reconn = next) {
 441                         next = TAILQ_NEXT(reconn, next);
 442
 443                         ret = vhost_user_connect_nonblock(reconn->fd,
 444                                                 (struct sockaddr *)&reconn->un,
 445                                                 sizeof(reconn->un));
 446                         if (ret == -2) {
 447                                 close(reconn->fd);
 448                                 RTE_LOG(ERR, VHOST_CONFIG,
 449                                         "reconnection for fd %d failed\n",
 450                                         reconn->fd);
 451                                 goto remove_fd;
 452                         }
 453                         if (ret == -1)
 454                                 continue;
 455
 456                         RTE_LOG(INFO, VHOST_CONFIG,
 457                                 "%s: connected\n", reconn->vsocket->path);
 458                         vhost_user_add_connection(reconn->fd, reconn->vsocket);
 459 remove_fd:
 460                         TAILQ_REMOVE(&reconn_list.head, reconn, next);
 461                         free(reconn);
 462                 }
 463
 464                 pthread_mutex_unlock(&reconn_list.mutex);
 465                 sleep(1);
 466         }
 467
 468         return NULL;
 469 }
 470
 471 static int
 472 vhost_user_reconnect_init(void)
 473 {
 474         int ret;
 475
 476         pthread_mutex_init(&reconn_list.mutex, NULL);
 477         TAILQ_INIT(&reconn_list.head);
 478
 479         ret = pthread_create(&reconn_tid, NULL,
 480                              vhost_user_client_reconnect, NULL);
 481         if (ret < 0)
 482                 RTE_LOG(ERR, VHOST_CONFIG, "failed to create reconnect thread");
 483
 484         return ret;
 485 }
 486
 487 static int
 488 vhost_user_start_client(struct vhost_user_socket *vsocket)
 489 {
 490         int ret;
 491         int fd = vsocket->socket_fd;
 492         const char *path = vsocket->path;
 493         struct vhost_user_reconnect *reconn;
 494
 495         ret = vhost_user_connect_nonblock(fd, (struct sockaddr *)&vsocket->un,
 496                                           sizeof(vsocket->un));
 497         if (ret == 0) {
 498                 vhost_user_add_connection(fd, vsocket);
 499                 return 0;
 500         }
 501
 502         RTE_LOG(WARNING, VHOST_CONFIG,
 503                 "failed to connect to %s: %s\n",
 504                 path, strerror(errno));
 505
 506         if (ret == -2 || !vsocket->reconnect) {
 507                 close(fd);
 508                 return -1;
 509         }
 510
 511         RTE_LOG(INFO, VHOST_CONFIG, "%s: reconnecting...\n", path);
 512         reconn = malloc(sizeof(*reconn));
 513         if (reconn == NULL) {
 514                 RTE_LOG(ERR, VHOST_CONFIG,
 515                         "failed to allocate memory for reconnect\n");
 516                 close(fd);
 517                 return -1;
 518         }
 519         reconn->un = vsocket->un;
 520         reconn->fd = fd;
 521         reconn->vsocket = vsocket;
 522         pthread_mutex_lock(&reconn_list.mutex);
 523         TAILQ_INSERT_TAIL(&reconn_list.head, reconn, next);
 524         pthread_mutex_unlock(&reconn_list.mutex);
 525
 526         return 0;
 527 }
 528
 529 static struct vhost_user_socket *
 530 find_vhost_user_socket(const char *path)
 531 {
 532         int i;
 533
 534         for (i = 0; i < vhost_user.vsocket_cnt; i++) {
 535                 struct vhost_user_socket *vsocket = vhost_user.vsockets[i];
 536
 537                 if (!strcmp(vsocket->path, path))
 538                         return vsocket;
 539         }
 540
 541         return NULL;
 542 }
 543
 544 int
 545 rte_vhost_driver_disable_features(const char *path, uint64_t features)
 546 {
 547         struct vhost_user_socket *vsocket;
 548
 549         pthread_mutex_lock(&vhost_user.mutex);
 550         vsocket = find_vhost_user_socket(path);
 551         if (vsocket)
 552                 vsocket->features &= ~features;
 553         pthread_mutex_unlock(&vhost_user.mutex);
 554
 555         return vsocket ? 0 : -1;
 556 }
 557
 558 int
 559 rte_vhost_driver_enable_features(const char *path, uint64_t features)
 560 {
 561         struct vhost_user_socket *vsocket;
 562
 563         pthread_mutex_lock(&vhost_user.mutex);
 564         vsocket = find_vhost_user_socket(path);
 565         if (vsocket) {
 566                 if ((vsocket->supported_features & features) != features) {
 567                         /*
 568                          * trying to enable features the driver doesn't
 569                          * support.
 570                          */
 571                         pthread_mutex_unlock(&vhost_user.mutex);
 572                         return -1;
 573                 }
 574                 vsocket->features |= features;
 575         }
 576         pthread_mutex_unlock(&vhost_user.mutex);
 577
 578         return vsocket ? 0 : -1;
 579 }
 580
 581 int
 582 rte_vhost_driver_set_features(const char *path, uint64_t features)
 583 {
 584         struct vhost_user_socket *vsocket;
 585
 586         pthread_mutex_lock(&vhost_user.mutex);
 587         vsocket = find_vhost_user_socket(path);
 588         if (vsocket) {
 589                 vsocket->supported_features = features;
 590                 vsocket->features = features;
 591         }
 592         pthread_mutex_unlock(&vhost_user.mutex);
 593
 594         return vsocket ? 0 : -1;
 595 }
 596
 597 int
 598 rte_vhost_driver_get_features(const char *path, uint64_t *features)
 599 {
 600         struct vhost_user_socket *vsocket;
 601
 602         pthread_mutex_lock(&vhost_user.mutex);
 603         vsocket = find_vhost_user_socket(path);
 604         if (vsocket)
 605                 *features = vsocket->features;
 606         pthread_mutex_unlock(&vhost_user.mutex);
 607
 608         if (!vsocket) {
 609                 RTE_LOG(ERR, VHOST_CONFIG,
 610                         "socket file %s is not registered yet.\n", path);
 611                 return -1;
 612         } else {
 613                 return 0;
 614         }
 615 }
 616
 617 /*
 618  * Register a new vhost-user socket; here we could act as server
 619  * (the default case), or client (when RTE_VHOST_USER_CLIENT) flag
 620  * is set.
 621  */
 622 int
 623 rte_vhost_driver_register(const char *path, uint64_t flags)
 624 {
 625         int ret = -1;
 626         struct vhost_user_socket *vsocket;
 627
 628         if (!path)
 629                 return -1;
 630
 631         pthread_mutex_lock(&vhost_user.mutex);
 632
 633         if (vhost_user.vsocket_cnt == MAX_VHOST_SOCKET) {
 634                 RTE_LOG(ERR, VHOST_CONFIG,
 635                         "error: the number of vhost sockets reaches maximum\n");
 636                 goto out;
 637         }
 638
 639         vsocket = malloc(sizeof(struct vhost_user_socket));
 640         if (!vsocket)
 641                 goto out;
 642         memset(vsocket, 0, sizeof(struct vhost_user_socket));
 643         vsocket->path = strdup(path);
 644         if (!vsocket->path) {
 645                 free(vsocket);
 646                 goto out;
 647         }
 648         TAILQ_INIT(&vsocket->conn_list);
 649         vsocket->dequeue_zero_copy = flags & RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
 650
 651         /*
 652          * Set the supported features correctly for the builtin vhost-user
 653          * net driver.
 654          *
 655          * Applications know nothing about features the builtin virtio net
 656          * driver (virtio_net.c) supports, thus it's not possible for them
 657          * to invoke rte_vhost_driver_set_features(). To workaround it, here
 658          * we set it unconditionally. If the application want to implement
 659          * another vhost-user driver (say SCSI), it should call the
 660          * rte_vhost_driver_set_features(), which will overwrite following
 661          * two values.
 662          */
 663         vsocket->supported_features = VIRTIO_NET_SUPPORTED_FEATURES;
 664         vsocket->features           = VIRTIO_NET_SUPPORTED_FEATURES;
 665
 666         if ((flags & RTE_VHOST_USER_CLIENT) != 0) {
 667                 vsocket->reconnect = !(flags & RTE_VHOST_USER_NO_RECONNECT);
 668                 if (vsocket->reconnect && reconn_tid == 0) {
 669                         if (vhost_user_reconnect_init() < 0) {
 670                                 free(vsocket->path);
 671                                 free(vsocket);
 672                                 goto out;
 673                         }
 674                 }
 675         } else {
 676                 vsocket->is_server = true;
 677         }
 678         ret = create_unix_socket(vsocket);
 679         if (ret < 0) {
 680                 free(vsocket->path);
 681                 free(vsocket);
 682                 goto out;
 683         }
 684
 685         pthread_mutex_init(&vsocket->conn_mutex, NULL);
 686         vhost_user.vsockets[vhost_user.vsocket_cnt++] = vsocket;
 687
 688 out:
 689         pthread_mutex_unlock(&vhost_user.mutex);
 690
 691         return ret;
 692 }
 693
 694 static bool
 695 vhost_user_remove_reconnect(struct vhost_user_socket *vsocket)
 696 {
 697         int found = false;
 698         struct vhost_user_reconnect *reconn, *next;
 699
 700         pthread_mutex_lock(&reconn_list.mutex);
 701
 702         for (reconn = TAILQ_FIRST(&reconn_list.head);
 703              reconn != NULL; reconn = next) {
 704                 next = TAILQ_NEXT(reconn, next);
 705
 706                 if (reconn->vsocket == vsocket) {
 707                         TAILQ_REMOVE(&reconn_list.head, reconn, next);
 708                         close(reconn->fd);
 709                         free(reconn);
 710                         found = true;
 711                         break;
 712                 }
 713         }
 714         pthread_mutex_unlock(&reconn_list.mutex);
 715         return found;
 716 }
 717
 718 /**
 719  * Unregister the specified vhost socket
 720  */
 721 int
 722 rte_vhost_driver_unregister(const char *path)
 723 {
 724         int i;
 725         int count;
 726         struct vhost_user_connection *conn;
 727
 728         pthread_mutex_lock(&vhost_user.mutex);
 729
 730         for (i = 0; i < vhost_user.vsocket_cnt; i++) {
 731                 struct vhost_user_socket *vsocket = vhost_user.vsockets[i];
 732
 733                 if (!strcmp(vsocket->path, path)) {
 734                         if (vsocket->is_server) {
 735                                 fdset_del(&vhost_user.fdset, vsocket->socket_fd);
 736                                 close(vsocket->socket_fd);
 737                                 unlink(path);
 738                         } else if (vsocket->reconnect) {
 739                                 vhost_user_remove_reconnect(vsocket);
 740                         }
 741
 742                         pthread_mutex_lock(&vsocket->conn_mutex);
 743                         TAILQ_FOREACH(conn, &vsocket->conn_list, next) {
 744                                 close(conn->connfd);
 745                                 conn->connfd = -1;
 746                         }
 747                         pthread_mutex_unlock(&vsocket->conn_mutex);
 748
 749                         do {
 750                                 pthread_mutex_lock(&vsocket->conn_mutex);
 751                                 conn = TAILQ_FIRST(&vsocket->conn_list);
 752                                 pthread_mutex_unlock(&vsocket->conn_mutex);
 753                         } while (conn != NULL);
 754
 755                         free(vsocket->path);
 756                         free(vsocket);
 757
 758                         count = --vhost_user.vsocket_cnt;
 759                         vhost_user.vsockets[i] = vhost_user.vsockets[count];
 760                         vhost_user.vsockets[count] = NULL;
 761                         pthread_mutex_unlock(&vhost_user.mutex);
 762
 763                         return 0;
 764                 }
 765         }
 766         pthread_mutex_unlock(&vhost_user.mutex);
 767
 768         return -1;
 769 }
 770
 771 /*
 772  * Register ops so that we can add/remove device to data core.
 773  */
 774 int
 775 rte_vhost_driver_callback_register(const char *path,
 776         struct vhost_device_ops const * const ops)
 777 {
 778         struct vhost_user_socket *vsocket;
 779
 780         pthread_mutex_lock(&vhost_user.mutex);
 781         vsocket = find_vhost_user_socket(path);
 782         if (vsocket)
 783                 vsocket->notify_ops = ops;
 784         pthread_mutex_unlock(&vhost_user.mutex);
 785
 786         return vsocket ? 0 : -1;
 787 }
 788
 789 struct vhost_device_ops const *
 790 vhost_driver_callback_get(const char *path)
 791 {
 792         struct vhost_user_socket *vsocket;
 793
 794         pthread_mutex_lock(&vhost_user.mutex);
 795         vsocket = find_vhost_user_socket(path);
 796         pthread_mutex_unlock(&vhost_user.mutex);
 797
 798         return vsocket ? vsocket->notify_ops : NULL;
 799 }
 800
 801 int
 802 rte_vhost_driver_start(const char *path)
 803 {
 804         struct vhost_user_socket *vsocket;
 805         static pthread_t fdset_tid;
 806
 807         pthread_mutex_lock(&vhost_user.mutex);
 808         vsocket = find_vhost_user_socket(path);
 809         pthread_mutex_unlock(&vhost_user.mutex);
 810
 811         if (!vsocket)
 812                 return -1;
 813
 814         if (fdset_tid == 0) {
 815                 rte_cpuset_t orig_cpuset;
 816                 rte_cpuset_t tmp_cpuset;
 817                 long num_cores, i;
 818                 int ret;
 819
 820                 CPU_ZERO(&tmp_cpuset);
 821                 num_cores = sysconf(_SC_NPROCESSORS_CONF);
 822                 /* Create a mask containing all CPUs */
 823                 for (i = 0; i < num_cores; i++) {
 824                         CPU_SET(i, &tmp_cpuset);
 825                 }
 826
 827                 rte_thread_get_affinity(&orig_cpuset);
 828                 rte_thread_set_affinity(&tmp_cpuset);
 829                 ret = pthread_create(&fdset_tid, NULL, fdset_event_dispatch,
 830                                      &vhost_user.fdset);
 831                 rte_thread_set_affinity(&orig_cpuset);
 832                 if (ret < 0)
 833                         RTE_LOG(ERR, VHOST_CONFIG,
 834                                 "failed to create fdset handling thread");
 835         }
 836
 837         if (vsocket->is_server)
 838                 return vhost_user_start_server(vsocket);
 839         else
 840                 return vhost_user_start_client(vsocket);
 841 }