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