lib/librte_vhost/vhost_user/vhost-net-user.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 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 <limits.h>
  37 #include <stdlib.h>
  38 #include <unistd.h>
  39 #include <string.h>
  40 #include <sys/types.h>
  41 #include <sys/socket.h>
  42 #include <sys/un.h>
  43 #include <errno.h>
  44 #include <pthread.h>
  45
  46 #include <rte_log.h>
  47 #include <rte_virtio_net.h>
  48
  49 #include "fd_man.h"
  50 #include "vhost-net-user.h"
  51 #include "vhost-net.h"
  52 #include "virtio-net-user.h"
  53
  54 #define MAX_VIRTIO_BACKLOG 128
  55
  56 static void vserver_new_vq_conn(int fd, void *data, int *remove);
  57 static void vserver_message_handler(int fd, void *dat, int *remove);
  58 struct vhost_net_device_ops const *ops;
  59
  60 struct connfd_ctx {
  61         struct vhost_server *vserver;
  62         uint32_t fh;
  63 };
  64
  65 #define MAX_VHOST_SERVER 1024
  66 struct _vhost_server {
  67         struct vhost_server *server[MAX_VHOST_SERVER];
  68         struct fdset fdset;
  69         int vserver_cnt;
  70         pthread_mutex_t server_mutex;
  71 };
  72
  73 static struct _vhost_server g_vhost_server = {
  74         .fdset = {
  75                 .fd = { [0 ... MAX_FDS - 1] = {-1, NULL, NULL, NULL, 0} },
  76                 .fd_mutex = PTHREAD_MUTEX_INITIALIZER,
  77                 .num = 0
  78         },
  79         .vserver_cnt = 0,
  80         .server_mutex = PTHREAD_MUTEX_INITIALIZER,
  81 };
  82
  83 static const char *vhost_message_str[VHOST_USER_MAX] = {
  84         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  85         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  86         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  87         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  88         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  89         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  90         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  91         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  92         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  93         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  94         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  95         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  96         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  97         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  98         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR"
  99 };
 100
 101 /**
 102  * Create a unix domain socket, bind to path and listen for connection.
 103  * @return
 104  *  socket fd or -1 on failure
 105  */
 106 static int
 107 uds_socket(const char *path)
 108 {
 109         struct sockaddr_un un;
 110         int sockfd;
 111         int ret;
 112
 113         if (path == NULL)
 114                 return -1;
 115
 116         sockfd = socket(AF_UNIX, SOCK_STREAM, 0);
 117         if (sockfd < 0)
 118                 return -1;
 119         RTE_LOG(INFO, VHOST_CONFIG, "socket created, fd:%d\n", sockfd);
 120
 121         memset(&un, 0, sizeof(un));
 122         un.sun_family = AF_UNIX;
 123         snprintf(un.sun_path, sizeof(un.sun_path), "%s", path);
 124         ret = bind(sockfd, (struct sockaddr *)&un, sizeof(un));
 125         if (ret == -1) {
 126                 RTE_LOG(ERR, VHOST_CONFIG, "fail to bind fd:%d, remove file:%s and try again.\n",
 127                         sockfd, path);
 128                 goto err;
 129         }
 130         RTE_LOG(INFO, VHOST_CONFIG, "bind to %s\n", path);
 131
 132         ret = listen(sockfd, MAX_VIRTIO_BACKLOG);
 133         if (ret == -1)
 134                 goto err;
 135
 136         return sockfd;
 137
 138 err:
 139         close(sockfd);
 140         return -1;
 141 }
 142
 143 /* return bytes# of read on success or negative val on failure. */
 144 static int
 145 read_fd_message(int sockfd, char *buf, int buflen, int *fds, int fd_num)
 146 {
 147         struct iovec iov;
 148         struct msghdr msgh;
 149         size_t fdsize = fd_num * sizeof(int);
 150         char control[CMSG_SPACE(fdsize)];
 151         struct cmsghdr *cmsg;
 152         int ret;
 153
 154         memset(&msgh, 0, sizeof(msgh));
 155         iov.iov_base = buf;
 156         iov.iov_len  = buflen;
 157
 158         msgh.msg_iov = &iov;
 159         msgh.msg_iovlen = 1;
 160         msgh.msg_control = control;
 161         msgh.msg_controllen = sizeof(control);
 162
 163         ret = recvmsg(sockfd, &msgh, 0);
 164         if (ret <= 0) {
 165                 RTE_LOG(ERR, VHOST_CONFIG, "recvmsg failed\n");
 166                 return ret;
 167         }
 168
 169         if (msgh.msg_flags & (MSG_TRUNC | MSG_CTRUNC)) {
 170                 RTE_LOG(ERR, VHOST_CONFIG, "truncted msg\n");
 171                 return -1;
 172         }
 173
 174         for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg != NULL;
 175                 cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
 176                 if ((cmsg->cmsg_level == SOL_SOCKET) &&
 177                         (cmsg->cmsg_type == SCM_RIGHTS)) {
 178                         memcpy(fds, CMSG_DATA(cmsg), fdsize);
 179                         break;
 180                 }
 181         }
 182
 183         return ret;
 184 }
 185
 186 /* return bytes# of read on success or negative val on failure. */
 187 static int
 188 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
 189 {
 190         int ret;
 191
 192         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
 193                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
 194         if (ret <= 0)
 195                 return ret;
 196
 197         if (msg && msg->size) {
 198                 if (msg->size > sizeof(msg->payload)) {
 199                         RTE_LOG(ERR, VHOST_CONFIG,
 200                                 "invalid msg size: %d\n", msg->size);
 201                         return -1;
 202                 }
 203                 ret = read(sockfd, &msg->payload, msg->size);
 204                 if (ret <= 0)
 205                         return ret;
 206                 if (ret != (int)msg->size) {
 207                         RTE_LOG(ERR, VHOST_CONFIG,
 208                                 "read control message failed\n");
 209                         return -1;
 210                 }
 211         }
 212
 213         return ret;
 214 }
 215
 216 static int
 217 send_fd_message(int sockfd, char *buf, int buflen, int *fds, int fd_num)
 218 {
 219
 220         struct iovec iov;
 221         struct msghdr msgh;
 222         size_t fdsize = fd_num * sizeof(int);
 223         char control[CMSG_SPACE(fdsize)];
 224         struct cmsghdr *cmsg;
 225         int ret;
 226
 227         memset(&msgh, 0, sizeof(msgh));
 228         iov.iov_base = buf;
 229         iov.iov_len = buflen;
 230
 231         msgh.msg_iov = &iov;
 232         msgh.msg_iovlen = 1;
 233
 234         if (fds && fd_num > 0) {
 235                 msgh.msg_control = control;
 236                 msgh.msg_controllen = sizeof(control);
 237                 cmsg = CMSG_FIRSTHDR(&msgh);
 238                 cmsg->cmsg_len = CMSG_LEN(fdsize);
 239                 cmsg->cmsg_level = SOL_SOCKET;
 240                 cmsg->cmsg_type = SCM_RIGHTS;
 241                 memcpy(CMSG_DATA(cmsg), fds, fdsize);
 242         } else {
 243                 msgh.msg_control = NULL;
 244                 msgh.msg_controllen = 0;
 245         }
 246
 247         do {
 248                 ret = sendmsg(sockfd, &msgh, 0);
 249         } while (ret < 0 && errno == EINTR);
 250
 251         if (ret < 0) {
 252                 RTE_LOG(ERR, VHOST_CONFIG,  "sendmsg error\n");
 253                 return ret;
 254         }
 255
 256         return ret;
 257 }
 258
 259 static int
 260 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
 261 {
 262         int ret;
 263
 264         if (!msg)
 265                 return 0;
 266
 267         msg->flags &= ~VHOST_USER_VERSION_MASK;
 268         msg->flags |= VHOST_USER_VERSION;
 269         msg->flags |= VHOST_USER_REPLY_MASK;
 270
 271         ret = send_fd_message(sockfd, (char *)msg,
 272                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
 273
 274         return ret;
 275 }
 276
 277 /* call back when there is new virtio connection.  */
 278 static void
 279 vserver_new_vq_conn(int fd, void *dat, __rte_unused int *remove)
 280 {
 281         struct vhost_server *vserver = (struct vhost_server *)dat;
 282         int conn_fd;
 283         struct connfd_ctx *ctx;
 284         int fh;
 285         struct vhost_device_ctx vdev_ctx = { (pid_t)0, 0 };
 286         unsigned int size;
 287
 288         conn_fd = accept(fd, NULL, NULL);
 289         RTE_LOG(INFO, VHOST_CONFIG,
 290                 "new virtio connection is %d\n", conn_fd);
 291         if (conn_fd < 0)
 292                 return;
 293
 294         ctx = calloc(1, sizeof(*ctx));
 295         if (ctx == NULL) {
 296                 close(conn_fd);
 297                 return;
 298         }
 299
 300         fh = ops->new_device(vdev_ctx);
 301         if (fh == -1) {
 302                 free(ctx);
 303                 close(conn_fd);
 304                 return;
 305         }
 306
 307         vdev_ctx.fh = fh;
 308         size = strnlen(vserver->path, PATH_MAX);
 309         ops->set_ifname(vdev_ctx, vserver->path,
 310                 size);
 311
 312         RTE_LOG(INFO, VHOST_CONFIG, "new device, handle is %d\n", fh);
 313
 314         ctx->vserver = vserver;
 315         ctx->fh = fh;
 316         fdset_add(&g_vhost_server.fdset,
 317                 conn_fd, vserver_message_handler, NULL, ctx);
 318 }
 319
 320 /* callback when there is message on the connfd */
 321 static void
 322 vserver_message_handler(int connfd, void *dat, int *remove)
 323 {
 324         struct vhost_device_ctx ctx;
 325         struct connfd_ctx *cfd_ctx = (struct connfd_ctx *)dat;
 326         struct VhostUserMsg msg;
 327         uint64_t features;
 328         int ret;
 329
 330         ctx.fh = cfd_ctx->fh;
 331         ret = read_vhost_message(connfd, &msg);
 332         if (ret <= 0 || msg.request > VHOST_USER_MAX) {
 333                 if (ret < 0)
 334                         RTE_LOG(ERR, VHOST_CONFIG,
 335                                 "vhost read message failed\n");
 336                 else if (ret == 0)
 337                         RTE_LOG(INFO, VHOST_CONFIG,
 338                                 "vhost peer closed\n");
 339                 else
 340                         RTE_LOG(ERR, VHOST_CONFIG,
 341                                 "vhost read incorrect message\n");
 342
 343                 close(connfd);
 344                 *remove = 1;
 345                 free(cfd_ctx);
 346                 user_destroy_device(ctx);
 347                 ops->destroy_device(ctx);
 348
 349                 return;
 350         }
 351
 352         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
 353                 vhost_message_str[msg.request]);
 354         switch (msg.request) {
 355         case VHOST_USER_GET_FEATURES:
 356                 ret = ops->get_features(ctx, &features);
 357                 msg.payload.u64 = features;
 358                 msg.size = sizeof(msg.payload.u64);
 359                 send_vhost_message(connfd, &msg);
 360                 break;
 361         case VHOST_USER_SET_FEATURES:
 362                 features = msg.payload.u64;
 363                 ops->set_features(ctx, &features);
 364                 break;
 365
 366         case VHOST_USER_SET_OWNER:
 367                 ops->set_owner(ctx);
 368                 break;
 369         case VHOST_USER_RESET_OWNER:
 370                 ops->reset_owner(ctx);
 371                 break;
 372
 373         case VHOST_USER_SET_MEM_TABLE:
 374                 user_set_mem_table(ctx, &msg);
 375                 break;
 376
 377         case VHOST_USER_SET_LOG_BASE:
 378                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
 379         case VHOST_USER_SET_LOG_FD:
 380                 close(msg.fds[0]);
 381                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
 382                 break;
 383
 384         case VHOST_USER_SET_VRING_NUM:
 385                 ops->set_vring_num(ctx, &msg.payload.state);
 386                 break;
 387         case VHOST_USER_SET_VRING_ADDR:
 388                 ops->set_vring_addr(ctx, &msg.payload.addr);
 389                 break;
 390         case VHOST_USER_SET_VRING_BASE:
 391                 ops->set_vring_base(ctx, &msg.payload.state);
 392                 break;
 393
 394         case VHOST_USER_GET_VRING_BASE:
 395                 ret = user_get_vring_base(ctx, &msg.payload.state);
 396                 msg.size = sizeof(msg.payload.state);
 397                 send_vhost_message(connfd, &msg);
 398                 break;
 399
 400         case VHOST_USER_SET_VRING_KICK:
 401                 user_set_vring_kick(ctx, &msg);
 402                 break;
 403         case VHOST_USER_SET_VRING_CALL:
 404                 user_set_vring_call(ctx, &msg);
 405                 break;
 406
 407         case VHOST_USER_SET_VRING_ERR:
 408                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
 409                         close(msg.fds[0]);
 410                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
 411                 break;
 412
 413         default:
 414                 break;
 415
 416         }
 417 }
 418
 419 /**
 420  * Creates and initialise the vhost server.
 421  */
 422 int
 423 rte_vhost_driver_register(const char *path)
 424 {
 425         struct vhost_server *vserver;
 426
 427         pthread_mutex_lock(&g_vhost_server.server_mutex);
 428         if (ops == NULL)
 429                 ops = get_virtio_net_callbacks();
 430
 431         if (g_vhost_server.vserver_cnt == MAX_VHOST_SERVER) {
 432                 RTE_LOG(ERR, VHOST_CONFIG,
 433                         "error: the number of servers reaches maximum\n");
 434                 pthread_mutex_unlock(&g_vhost_server.server_mutex);
 435                 return -1;
 436         }
 437
 438         vserver = calloc(sizeof(struct vhost_server), 1);
 439         if (vserver == NULL) {
 440                 pthread_mutex_unlock(&g_vhost_server.server_mutex);
 441                 return -1;
 442         }
 443
 444         vserver->listenfd = uds_socket(path);
 445         if (vserver->listenfd < 0) {
 446                 free(vserver);
 447                 pthread_mutex_unlock(&g_vhost_server.server_mutex);
 448                 return -1;
 449         }
 450
 451         vserver->path = strdup(path);
 452
 453         fdset_add(&g_vhost_server.fdset, vserver->listenfd,
 454                 vserver_new_vq_conn, NULL, vserver);
 455
 456         g_vhost_server.server[g_vhost_server.vserver_cnt++] = vserver;
 457         pthread_mutex_unlock(&g_vhost_server.server_mutex);
 458
 459         return 0;
 460 }
 461
 462
 463 /**
 464  * Unregister the specified vhost server
 465  */
 466 int
 467 rte_vhost_driver_unregister(const char *path)
 468 {
 469         int i;
 470         int count;
 471
 472         pthread_mutex_lock(&g_vhost_server.server_mutex);
 473
 474         for (i = 0; i < g_vhost_server.vserver_cnt; i++) {
 475                 if (!strcmp(g_vhost_server.server[i]->path, path)) {
 476                         fdset_del(&g_vhost_server.fdset,
 477                                 g_vhost_server.server[i]->listenfd);
 478
 479                         close(g_vhost_server.server[i]->listenfd);
 480                         free(g_vhost_server.server[i]->path);
 481                         free(g_vhost_server.server[i]);
 482
 483                         unlink(path);
 484
 485                         count = --g_vhost_server.vserver_cnt;
 486                         g_vhost_server.server[i] = g_vhost_server.server[count];
 487                         g_vhost_server.server[count] = NULL;
 488                         pthread_mutex_unlock(&g_vhost_server.server_mutex);
 489
 490                         return 0;
 491                 }
 492         }
 493         pthread_mutex_unlock(&g_vhost_server.server_mutex);
 494
 495         return -1;
 496 }
 497
 498 int
 499 rte_vhost_driver_session_start(void)
 500 {
 501         fdset_event_dispatch(&g_vhost_server.fdset);
 502         return 0;
 503 }