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         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
 100         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
 101         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
 102         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
 103         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
 104 };
 105
 106 /**
 107  * Create a unix domain socket, bind to path and listen for connection.
 108  * @return
 109  *  socket fd or -1 on failure
 110  */
 111 static int
 112 uds_socket(const char *path)
 113 {
 114         struct sockaddr_un un;
 115         int sockfd;
 116         int ret;
 117
 118         if (path == NULL)
 119                 return -1;
 120
 121         sockfd = socket(AF_UNIX, SOCK_STREAM, 0);
 122         if (sockfd < 0)
 123                 return -1;
 124         RTE_LOG(INFO, VHOST_CONFIG, "socket created, fd:%d\n", sockfd);
 125
 126         memset(&un, 0, sizeof(un));
 127         un.sun_family = AF_UNIX;
 128         snprintf(un.sun_path, sizeof(un.sun_path), "%s", path);
 129         ret = bind(sockfd, (struct sockaddr *)&un, sizeof(un));
 130         if (ret == -1) {
 131                 RTE_LOG(ERR, VHOST_CONFIG, "fail to bind fd:%d, remove file:%s and try again.\n",
 132                         sockfd, path);
 133                 goto err;
 134         }
 135         RTE_LOG(INFO, VHOST_CONFIG, "bind to %s\n", path);
 136
 137         ret = listen(sockfd, MAX_VIRTIO_BACKLOG);
 138         if (ret == -1)
 139                 goto err;
 140
 141         return sockfd;
 142
 143 err:
 144         close(sockfd);
 145         return -1;
 146 }
 147
 148 /* return bytes# of read on success or negative val on failure. */
 149 static int
 150 read_fd_message(int sockfd, char *buf, int buflen, int *fds, int fd_num)
 151 {
 152         struct iovec iov;
 153         struct msghdr msgh;
 154         size_t fdsize = fd_num * sizeof(int);
 155         char control[CMSG_SPACE(fdsize)];
 156         struct cmsghdr *cmsg;
 157         int ret;
 158
 159         memset(&msgh, 0, sizeof(msgh));
 160         iov.iov_base = buf;
 161         iov.iov_len  = buflen;
 162
 163         msgh.msg_iov = &iov;
 164         msgh.msg_iovlen = 1;
 165         msgh.msg_control = control;
 166         msgh.msg_controllen = sizeof(control);
 167
 168         ret = recvmsg(sockfd, &msgh, 0);
 169         if (ret <= 0) {
 170                 RTE_LOG(ERR, VHOST_CONFIG, "recvmsg failed\n");
 171                 return ret;
 172         }
 173
 174         if (msgh.msg_flags & (MSG_TRUNC | MSG_CTRUNC)) {
 175                 RTE_LOG(ERR, VHOST_CONFIG, "truncted msg\n");
 176                 return -1;
 177         }
 178
 179         for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg != NULL;
 180                 cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
 181                 if ((cmsg->cmsg_level == SOL_SOCKET) &&
 182                         (cmsg->cmsg_type == SCM_RIGHTS)) {
 183                         memcpy(fds, CMSG_DATA(cmsg), fdsize);
 184                         break;
 185                 }
 186         }
 187
 188         return ret;
 189 }
 190
 191 /* return bytes# of read on success or negative val on failure. */
 192 static int
 193 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
 194 {
 195         int ret;
 196
 197         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
 198                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
 199         if (ret <= 0)
 200                 return ret;
 201
 202         if (msg && msg->size) {
 203                 if (msg->size > sizeof(msg->payload)) {
 204                         RTE_LOG(ERR, VHOST_CONFIG,
 205                                 "invalid msg size: %d\n", msg->size);
 206                         return -1;
 207                 }
 208                 ret = read(sockfd, &msg->payload, msg->size);
 209                 if (ret <= 0)
 210                         return ret;
 211                 if (ret != (int)msg->size) {
 212                         RTE_LOG(ERR, VHOST_CONFIG,
 213                                 "read control message failed\n");
 214                         return -1;
 215                 }
 216         }
 217
 218         return ret;
 219 }
 220
 221 static int
 222 send_fd_message(int sockfd, char *buf, int buflen, int *fds, int fd_num)
 223 {
 224
 225         struct iovec iov;
 226         struct msghdr msgh;
 227         size_t fdsize = fd_num * sizeof(int);
 228         char control[CMSG_SPACE(fdsize)];
 229         struct cmsghdr *cmsg;
 230         int ret;
 231
 232         memset(&msgh, 0, sizeof(msgh));
 233         iov.iov_base = buf;
 234         iov.iov_len = buflen;
 235
 236         msgh.msg_iov = &iov;
 237         msgh.msg_iovlen = 1;
 238
 239         if (fds && fd_num > 0) {
 240                 msgh.msg_control = control;
 241                 msgh.msg_controllen = sizeof(control);
 242                 cmsg = CMSG_FIRSTHDR(&msgh);
 243                 cmsg->cmsg_len = CMSG_LEN(fdsize);
 244                 cmsg->cmsg_level = SOL_SOCKET;
 245                 cmsg->cmsg_type = SCM_RIGHTS;
 246                 memcpy(CMSG_DATA(cmsg), fds, fdsize);
 247         } else {
 248                 msgh.msg_control = NULL;
 249                 msgh.msg_controllen = 0;
 250         }
 251
 252         do {
 253                 ret = sendmsg(sockfd, &msgh, 0);
 254         } while (ret < 0 && errno == EINTR);
 255
 256         if (ret < 0) {
 257                 RTE_LOG(ERR, VHOST_CONFIG,  "sendmsg error\n");
 258                 return ret;
 259         }
 260
 261         return ret;
 262 }
 263
 264 static int
 265 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
 266 {
 267         int ret;
 268
 269         if (!msg)
 270                 return 0;
 271
 272         msg->flags &= ~VHOST_USER_VERSION_MASK;
 273         msg->flags |= VHOST_USER_VERSION;
 274         msg->flags |= VHOST_USER_REPLY_MASK;
 275
 276         ret = send_fd_message(sockfd, (char *)msg,
 277                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
 278
 279         return ret;
 280 }
 281
 282 /* call back when there is new virtio connection.  */
 283 static void
 284 vserver_new_vq_conn(int fd, void *dat, __rte_unused int *remove)
 285 {
 286         struct vhost_server *vserver = (struct vhost_server *)dat;
 287         int conn_fd;
 288         struct connfd_ctx *ctx;
 289         int fh;
 290         struct vhost_device_ctx vdev_ctx = { (pid_t)0, 0 };
 291         unsigned int size;
 292
 293         conn_fd = accept(fd, NULL, NULL);
 294         RTE_LOG(INFO, VHOST_CONFIG,
 295                 "new virtio connection is %d\n", conn_fd);
 296         if (conn_fd < 0)
 297                 return;
 298
 299         ctx = calloc(1, sizeof(*ctx));
 300         if (ctx == NULL) {
 301                 close(conn_fd);
 302                 return;
 303         }
 304
 305         fh = ops->new_device(vdev_ctx);
 306         if (fh == -1) {
 307                 free(ctx);
 308                 close(conn_fd);
 309                 return;
 310         }
 311
 312         vdev_ctx.fh = fh;
 313         size = strnlen(vserver->path, PATH_MAX);
 314         ops->set_ifname(vdev_ctx, vserver->path,
 315                 size);
 316
 317         RTE_LOG(INFO, VHOST_CONFIG, "new device, handle is %d\n", fh);
 318
 319         ctx->vserver = vserver;
 320         ctx->fh = fh;
 321         fdset_add(&g_vhost_server.fdset,
 322                 conn_fd, vserver_message_handler, NULL, ctx);
 323 }
 324
 325 /* callback when there is message on the connfd */
 326 static void
 327 vserver_message_handler(int connfd, void *dat, int *remove)
 328 {
 329         struct vhost_device_ctx ctx;
 330         struct connfd_ctx *cfd_ctx = (struct connfd_ctx *)dat;
 331         struct VhostUserMsg msg;
 332         uint64_t features;
 333         int ret;
 334
 335         ctx.fh = cfd_ctx->fh;
 336         ret = read_vhost_message(connfd, &msg);
 337         if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
 338                 if (ret < 0)
 339                         RTE_LOG(ERR, VHOST_CONFIG,
 340                                 "vhost read message failed\n");
 341                 else if (ret == 0)
 342                         RTE_LOG(INFO, VHOST_CONFIG,
 343                                 "vhost peer closed\n");
 344                 else
 345                         RTE_LOG(ERR, VHOST_CONFIG,
 346                                 "vhost read incorrect message\n");
 347
 348                 close(connfd);
 349                 *remove = 1;
 350                 free(cfd_ctx);
 351                 ops->destroy_device(ctx);
 352
 353                 return;
 354         }
 355
 356         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
 357                 vhost_message_str[msg.request]);
 358         switch (msg.request) {
 359         case VHOST_USER_GET_FEATURES:
 360                 ret = ops->get_features(ctx, &features);
 361                 msg.payload.u64 = features;
 362                 msg.size = sizeof(msg.payload.u64);
 363                 send_vhost_message(connfd, &msg);
 364                 break;
 365         case VHOST_USER_SET_FEATURES:
 366                 features = msg.payload.u64;
 367                 ops->set_features(ctx, &features);
 368                 break;
 369
 370         case VHOST_USER_GET_PROTOCOL_FEATURES:
 371                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
 372                 msg.size = sizeof(msg.payload.u64);
 373                 send_vhost_message(connfd, &msg);
 374                 break;
 375         case VHOST_USER_SET_PROTOCOL_FEATURES:
 376                 user_set_protocol_features(ctx, msg.payload.u64);
 377                 break;
 378
 379         case VHOST_USER_SET_OWNER:
 380                 ops->set_owner(ctx);
 381                 break;
 382         case VHOST_USER_RESET_OWNER:
 383                 ops->reset_owner(ctx);
 384                 break;
 385
 386         case VHOST_USER_SET_MEM_TABLE:
 387                 user_set_mem_table(ctx, &msg);
 388                 break;
 389
 390         case VHOST_USER_SET_LOG_BASE:
 391                 user_set_log_base(ctx, &msg);
 392
 393                 /* it needs a reply */
 394                 msg.size = sizeof(msg.payload.u64);
 395                 send_vhost_message(connfd, &msg);
 396                 break;
 397         case VHOST_USER_SET_LOG_FD:
 398                 close(msg.fds[0]);
 399                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
 400                 break;
 401
 402         case VHOST_USER_SET_VRING_NUM:
 403                 ops->set_vring_num(ctx, &msg.payload.state);
 404                 break;
 405         case VHOST_USER_SET_VRING_ADDR:
 406                 ops->set_vring_addr(ctx, &msg.payload.addr);
 407                 break;
 408         case VHOST_USER_SET_VRING_BASE:
 409                 ops->set_vring_base(ctx, &msg.payload.state);
 410                 break;
 411
 412         case VHOST_USER_GET_VRING_BASE:
 413                 ret = user_get_vring_base(ctx, &msg.payload.state);
 414                 msg.size = sizeof(msg.payload.state);
 415                 send_vhost_message(connfd, &msg);
 416                 break;
 417
 418         case VHOST_USER_SET_VRING_KICK:
 419                 user_set_vring_kick(ctx, &msg);
 420                 break;
 421         case VHOST_USER_SET_VRING_CALL:
 422                 user_set_vring_call(ctx, &msg);
 423                 break;
 424
 425         case VHOST_USER_SET_VRING_ERR:
 426                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
 427                         close(msg.fds[0]);
 428                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
 429                 break;
 430
 431         case VHOST_USER_GET_QUEUE_NUM:
 432                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
 433                 msg.size = sizeof(msg.payload.u64);
 434                 send_vhost_message(connfd, &msg);
 435                 break;
 436
 437         case VHOST_USER_SET_VRING_ENABLE:
 438                 user_set_vring_enable(ctx, &msg.payload.state);
 439                 break;
 440         case VHOST_USER_SEND_RARP:
 441                 user_send_rarp(&msg);
 442                 break;
 443
 444         default:
 445                 break;
 446
 447         }
 448 }
 449
 450 /**
 451  * Creates and initialise the vhost server.
 452  */
 453 int
 454 rte_vhost_driver_register(const char *path)
 455 {
 456         struct vhost_server *vserver;
 457
 458         pthread_mutex_lock(&g_vhost_server.server_mutex);
 459         if (ops == NULL)
 460                 ops = get_virtio_net_callbacks();
 461
 462         if (g_vhost_server.vserver_cnt == MAX_VHOST_SERVER) {
 463                 RTE_LOG(ERR, VHOST_CONFIG,
 464                         "error: the number of servers reaches maximum\n");
 465                 pthread_mutex_unlock(&g_vhost_server.server_mutex);
 466                 return -1;
 467         }
 468
 469         vserver = calloc(sizeof(struct vhost_server), 1);
 470         if (vserver == NULL) {
 471                 pthread_mutex_unlock(&g_vhost_server.server_mutex);
 472                 return -1;
 473         }
 474
 475         vserver->listenfd = uds_socket(path);
 476         if (vserver->listenfd < 0) {
 477                 free(vserver);
 478                 pthread_mutex_unlock(&g_vhost_server.server_mutex);
 479                 return -1;
 480         }
 481
 482         vserver->path = strdup(path);
 483
 484         fdset_add(&g_vhost_server.fdset, vserver->listenfd,
 485                 vserver_new_vq_conn, NULL, vserver);
 486
 487         g_vhost_server.server[g_vhost_server.vserver_cnt++] = vserver;
 488         pthread_mutex_unlock(&g_vhost_server.server_mutex);
 489
 490         return 0;
 491 }
 492
 493
 494 /**
 495  * Unregister the specified vhost server
 496  */
 497 int
 498 rte_vhost_driver_unregister(const char *path)
 499 {
 500         int i;
 501         int count;
 502
 503         pthread_mutex_lock(&g_vhost_server.server_mutex);
 504
 505         for (i = 0; i < g_vhost_server.vserver_cnt; i++) {
 506                 if (!strcmp(g_vhost_server.server[i]->path, path)) {
 507                         fdset_del(&g_vhost_server.fdset,
 508                                 g_vhost_server.server[i]->listenfd);
 509
 510                         close(g_vhost_server.server[i]->listenfd);
 511                         free(g_vhost_server.server[i]->path);
 512                         free(g_vhost_server.server[i]);
 513
 514                         unlink(path);
 515
 516                         count = --g_vhost_server.vserver_cnt;
 517                         g_vhost_server.server[i] = g_vhost_server.server[count];
 518                         g_vhost_server.server[count] = NULL;
 519                         pthread_mutex_unlock(&g_vhost_server.server_mutex);
 520
 521                         return 0;
 522                 }
 523         }
 524         pthread_mutex_unlock(&g_vhost_server.server_mutex);
 525
 526         return -1;
 527 }
 528
 529 int
 530 rte_vhost_driver_session_start(void)
 531 {
 532         fdset_event_dispatch(&g_vhost_server.fdset);
 533         return 0;
 534 }