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
  59 struct connfd_ctx {
  60         struct vhost_server *vserver;
  61         int vid;
  62 };
  63
  64 #define MAX_VHOST_SERVER 1024
  65 struct _vhost_server {
  66         struct vhost_server *server[MAX_VHOST_SERVER];
  67         struct fdset fdset;
  68         int vserver_cnt;
  69         pthread_mutex_t server_mutex;
  70 };
  71
  72 static struct _vhost_server g_vhost_server = {
  73         .fdset = {
  74                 .fd = { [0 ... MAX_FDS - 1] = {-1, NULL, NULL, NULL, 0} },
  75                 .fd_mutex = PTHREAD_MUTEX_INITIALIZER,
  76                 .num = 0
  77         },
  78         .vserver_cnt = 0,
  79         .server_mutex = PTHREAD_MUTEX_INITIALIZER,
  80 };
  81
  82 static const char *vhost_message_str[VHOST_USER_MAX] = {
  83         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  84         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  85         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  86         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  87         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  88         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  89         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  90         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  91         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  92         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  93         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  94         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  95         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  96         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  97         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  98         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  99         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
 100         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
 101         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
 102         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
 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 vid;
 289         unsigned int size;
 290
 291         conn_fd = accept(fd, NULL, NULL);
 292         RTE_LOG(INFO, VHOST_CONFIG,
 293                 "new virtio connection is %d\n", conn_fd);
 294         if (conn_fd < 0)
 295                 return;
 296
 297         ctx = calloc(1, sizeof(*ctx));
 298         if (ctx == NULL) {
 299                 close(conn_fd);
 300                 return;
 301         }
 302
 303         vid = vhost_new_device();
 304         if (vid == -1) {
 305                 free(ctx);
 306                 close(conn_fd);
 307                 return;
 308         }
 309
 310         size = strnlen(vserver->path, PATH_MAX);
 311         vhost_set_ifname(vid, vserver->path, size);
 312
 313         RTE_LOG(INFO, VHOST_CONFIG, "new device, handle is %d\n", vid);
 314
 315         ctx->vserver = vserver;
 316         ctx->vid = vid;
 317         fdset_add(&g_vhost_server.fdset,
 318                 conn_fd, vserver_message_handler, NULL, ctx);
 319 }
 320
 321 /* callback when there is message on the connfd */
 322 static void
 323 vserver_message_handler(int connfd, void *dat, int *remove)
 324 {
 325         int vid;
 326         struct connfd_ctx *cfd_ctx = (struct connfd_ctx *)dat;
 327         struct VhostUserMsg msg;
 328         uint64_t features;
 329         int ret;
 330
 331         vid = cfd_ctx->vid;
 332         ret = read_vhost_message(connfd, &msg);
 333         if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
 334                 if (ret < 0)
 335                         RTE_LOG(ERR, VHOST_CONFIG,
 336                                 "vhost read message failed\n");
 337                 else if (ret == 0)
 338                         RTE_LOG(INFO, VHOST_CONFIG,
 339                                 "vhost peer closed\n");
 340                 else
 341                         RTE_LOG(ERR, VHOST_CONFIG,
 342                                 "vhost read incorrect message\n");
 343
 344                 close(connfd);
 345                 *remove = 1;
 346                 free(cfd_ctx);
 347                 vhost_destroy_device(vid);
 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 = vhost_get_features(vid, &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                 vhost_set_features(vid, &features);
 364                 break;
 365
 366         case VHOST_USER_GET_PROTOCOL_FEATURES:
 367                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
 368                 msg.size = sizeof(msg.payload.u64);
 369                 send_vhost_message(connfd, &msg);
 370                 break;
 371         case VHOST_USER_SET_PROTOCOL_FEATURES:
 372                 user_set_protocol_features(vid, msg.payload.u64);
 373                 break;
 374
 375         case VHOST_USER_SET_OWNER:
 376                 vhost_set_owner(vid);
 377                 break;
 378         case VHOST_USER_RESET_OWNER:
 379                 vhost_reset_owner(vid);
 380                 break;
 381
 382         case VHOST_USER_SET_MEM_TABLE:
 383                 user_set_mem_table(vid, &msg);
 384                 break;
 385
 386         case VHOST_USER_SET_LOG_BASE:
 387                 user_set_log_base(vid, &msg);
 388
 389                 /* it needs a reply */
 390                 msg.size = sizeof(msg.payload.u64);
 391                 send_vhost_message(connfd, &msg);
 392                 break;
 393         case VHOST_USER_SET_LOG_FD:
 394                 close(msg.fds[0]);
 395                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
 396                 break;
 397
 398         case VHOST_USER_SET_VRING_NUM:
 399                 vhost_set_vring_num(vid, &msg.payload.state);
 400                 break;
 401         case VHOST_USER_SET_VRING_ADDR:
 402                 vhost_set_vring_addr(vid, &msg.payload.addr);
 403                 break;
 404         case VHOST_USER_SET_VRING_BASE:
 405                 vhost_set_vring_base(vid, &msg.payload.state);
 406                 break;
 407
 408         case VHOST_USER_GET_VRING_BASE:
 409                 ret = user_get_vring_base(vid, &msg.payload.state);
 410                 msg.size = sizeof(msg.payload.state);
 411                 send_vhost_message(connfd, &msg);
 412                 break;
 413
 414         case VHOST_USER_SET_VRING_KICK:
 415                 user_set_vring_kick(vid, &msg);
 416                 break;
 417         case VHOST_USER_SET_VRING_CALL:
 418                 user_set_vring_call(vid, &msg);
 419                 break;
 420
 421         case VHOST_USER_SET_VRING_ERR:
 422                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
 423                         close(msg.fds[0]);
 424                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
 425                 break;
 426
 427         case VHOST_USER_GET_QUEUE_NUM:
 428                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
 429                 msg.size = sizeof(msg.payload.u64);
 430                 send_vhost_message(connfd, &msg);
 431                 break;
 432
 433         case VHOST_USER_SET_VRING_ENABLE:
 434                 user_set_vring_enable(vid, &msg.payload.state);
 435                 break;
 436         case VHOST_USER_SEND_RARP:
 437                 user_send_rarp(vid, &msg);
 438                 break;
 439
 440         default:
 441                 break;
 442
 443         }
 444 }
 445
 446 /**
 447  * Creates and initialise the vhost server.
 448  */
 449 int
 450 rte_vhost_driver_register(const char *path)
 451 {
 452         struct vhost_server *vserver;
 453
 454         pthread_mutex_lock(&g_vhost_server.server_mutex);
 455
 456         if (g_vhost_server.vserver_cnt == MAX_VHOST_SERVER) {
 457                 RTE_LOG(ERR, VHOST_CONFIG,
 458                         "error: the number of servers reaches maximum\n");
 459                 pthread_mutex_unlock(&g_vhost_server.server_mutex);
 460                 return -1;
 461         }
 462
 463         vserver = calloc(sizeof(struct vhost_server), 1);
 464         if (vserver == NULL) {
 465                 pthread_mutex_unlock(&g_vhost_server.server_mutex);
 466                 return -1;
 467         }
 468
 469         vserver->listenfd = uds_socket(path);
 470         if (vserver->listenfd < 0) {
 471                 free(vserver);
 472                 pthread_mutex_unlock(&g_vhost_server.server_mutex);
 473                 return -1;
 474         }
 475
 476         vserver->path = strdup(path);
 477
 478         fdset_add(&g_vhost_server.fdset, vserver->listenfd,
 479                 vserver_new_vq_conn, NULL, vserver);
 480
 481         g_vhost_server.server[g_vhost_server.vserver_cnt++] = vserver;
 482         pthread_mutex_unlock(&g_vhost_server.server_mutex);
 483
 484         return 0;
 485 }
 486
 487
 488 /**
 489  * Unregister the specified vhost server
 490  */
 491 int
 492 rte_vhost_driver_unregister(const char *path)
 493 {
 494         int i;
 495         int count;
 496
 497         pthread_mutex_lock(&g_vhost_server.server_mutex);
 498
 499         for (i = 0; i < g_vhost_server.vserver_cnt; i++) {
 500                 if (!strcmp(g_vhost_server.server[i]->path, path)) {
 501                         fdset_del(&g_vhost_server.fdset,
 502                                 g_vhost_server.server[i]->listenfd);
 503
 504                         close(g_vhost_server.server[i]->listenfd);
 505                         free(g_vhost_server.server[i]->path);
 506                         free(g_vhost_server.server[i]);
 507
 508                         unlink(path);
 509
 510                         count = --g_vhost_server.vserver_cnt;
 511                         g_vhost_server.server[i] = g_vhost_server.server[count];
 512                         g_vhost_server.server[count] = NULL;
 513                         pthread_mutex_unlock(&g_vhost_server.server_mutex);
 514
 515                         return 0;
 516                 }
 517         }
 518         pthread_mutex_unlock(&g_vhost_server.server_mutex);
 519
 520         return -1;
 521 }
 522
 523 int
 524 rte_vhost_driver_session_start(void)
 525 {
 526         fdset_event_dispatch(&g_vhost_server.fdset);
 527         return 0;
 528 }