contrib/ivshmem-server/ivshmem-server.c

   1 /*
   2  * Copyright 6WIND S.A., 2014
   3  *
   4  * This work is licensed under the terms of the GNU GPL, version 2 or
   5  * (at your option) any later version.  See the COPYING file in the
   6  * top-level directory.
   7  */
   8 #include "qemu-common.h"
   9 #include "qemu/sockets.h"
  10
  11 #include <sys/mman.h>
  12 #include <sys/types.h>
  13 #include <sys/socket.h>
  14 #include <sys/un.h>
  15 #ifdef CONFIG_LINUX
  16 #include <sys/vfs.h>
  17 #endif
  18
  19 #include "ivshmem-server.h"
  20
  21 /* log a message on stdout if verbose=1 */
  22 #define IVSHMEM_SERVER_DEBUG(server, fmt, ...) do { \
  23         if ((server)->verbose) {         \
  24             printf(fmt, ## __VA_ARGS__); \
  25         }                                \
  26     } while (0)
  27
  28 /** maximum size of a huge page, used by ivshmem_server_ftruncate() */
  29 #define IVSHMEM_SERVER_MAX_HUGEPAGE_SIZE (1024 * 1024 * 1024)
  30
  31 /** default listen backlog (number of sockets not accepted) */
  32 #define IVSHMEM_SERVER_LISTEN_BACKLOG 10
  33
  34 /* send message to a client unix socket */
  35 static int
  36 ivshmem_server_send_one_msg(int sock_fd, long peer_id, int fd)
  37 {
  38     int ret;
  39     struct msghdr msg;
  40     struct iovec iov[1];
  41     union {
  42         struct cmsghdr cmsg;
  43         char control[CMSG_SPACE(sizeof(int))];
  44     } msg_control;
  45     struct cmsghdr *cmsg;
  46
  47     iov[0].iov_base = &peer_id;
  48     iov[0].iov_len = sizeof(peer_id);
  49
  50     memset(&msg, 0, sizeof(msg));
  51     msg.msg_iov = iov;
  52     msg.msg_iovlen = 1;
  53
  54     /* if fd is specified, add it in a cmsg */
  55     if (fd >= 0) {
  56         memset(&msg_control, 0, sizeof(msg_control));
  57         msg.msg_control = &msg_control;
  58         msg.msg_controllen = sizeof(msg_control);
  59         cmsg = CMSG_FIRSTHDR(&msg);
  60         cmsg->cmsg_level = SOL_SOCKET;
  61         cmsg->cmsg_type = SCM_RIGHTS;
  62         cmsg->cmsg_len = CMSG_LEN(sizeof(int));
  63         memcpy(CMSG_DATA(cmsg), &fd, sizeof(fd));
  64     }
  65
  66     ret = sendmsg(sock_fd, &msg, 0);
  67     if (ret <= 0) {
  68         return -1;
  69     }
  70
  71     return 0;
  72 }
  73
  74 /* free a peer when the server advertises a disconnection or when the
  75  * server is freed */
  76 static void
  77 ivshmem_server_free_peer(IvshmemServer *server, IvshmemServerPeer *peer)
  78 {
  79     unsigned vector;
  80     IvshmemServerPeer *other_peer;
  81
  82     IVSHMEM_SERVER_DEBUG(server, "free peer %ld\n", peer->id);
  83     close(peer->sock_fd);
  84     QTAILQ_REMOVE(&server->peer_list, peer, next);
  85
  86     /* advertise the deletion to other peers */
  87     QTAILQ_FOREACH(other_peer, &server->peer_list, next) {
  88         ivshmem_server_send_one_msg(other_peer->sock_fd, peer->id, -1);
  89     }
  90
  91     for (vector = 0; vector < peer->vectors_count; vector++) {
  92         event_notifier_cleanup(&peer->vectors[vector]);
  93     }
  94
  95     g_free(peer);
  96 }
  97
  98 /* send the peer id and the shm_fd just after a new client connection */
  99 static int
 100 ivshmem_server_send_initial_info(IvshmemServer *server, IvshmemServerPeer *peer)
 101 {
 102     int ret;
 103
 104     /* send the peer id to the client */
 105     ret = ivshmem_server_send_one_msg(peer->sock_fd, peer->id, -1);
 106     if (ret < 0) {
 107         IVSHMEM_SERVER_DEBUG(server, "cannot send peer id: %s\n",
 108                              strerror(errno));
 109         return -1;
 110     }
 111
 112     /* send the shm_fd */
 113     ret = ivshmem_server_send_one_msg(peer->sock_fd, -1, server->shm_fd);
 114     if (ret < 0) {
 115         IVSHMEM_SERVER_DEBUG(server, "cannot send shm fd: %s\n",
 116                              strerror(errno));
 117         return -1;
 118     }
 119
 120     return 0;
 121 }
 122
 123 /* handle message on listening unix socket (new client connection) */
 124 static int
 125 ivshmem_server_handle_new_conn(IvshmemServer *server)
 126 {
 127     IvshmemServerPeer *peer, *other_peer;
 128     struct sockaddr_un unaddr;
 129     socklen_t unaddr_len;
 130     int newfd;
 131     unsigned i;
 132
 133     /* accept the incoming connection */
 134     unaddr_len = sizeof(unaddr);
 135     newfd = qemu_accept(server->sock_fd,
 136                         (struct sockaddr *)&unaddr, &unaddr_len);
 137
 138     if (newfd < 0) {
 139         IVSHMEM_SERVER_DEBUG(server, "cannot accept() %s\n", strerror(errno));
 140         return -1;
 141     }
 142
 143     qemu_set_nonblock(newfd);
 144     IVSHMEM_SERVER_DEBUG(server, "accept()=%d\n", newfd);
 145
 146     /* allocate new structure for this peer */
 147     peer = g_malloc0(sizeof(*peer));
 148     peer->sock_fd = newfd;
 149
 150     /* get an unused peer id */
 151     /* XXX: this could use id allocation such as Linux IDA, or simply
 152      * a free-list */
 153     for (i = 0; i < G_MAXUINT16; i++) {
 154         if (ivshmem_server_search_peer(server, server->cur_id) == NULL) {
 155             break;
 156         }
 157         server->cur_id++;
 158     }
 159     if (i == G_MAXUINT16) {
 160         IVSHMEM_SERVER_DEBUG(server, "cannot allocate new client id\n");
 161         goto fail;
 162     }
 163     peer->id = server->cur_id++;
 164
 165     /* create eventfd, one per vector */
 166     peer->vectors_count = server->n_vectors;
 167     for (i = 0; i < peer->vectors_count; i++) {
 168         if (event_notifier_init(&peer->vectors[i], FALSE) < 0) {
 169             IVSHMEM_SERVER_DEBUG(server, "cannot create eventfd\n");
 170             goto fail;
 171         }
 172     }
 173
 174     /* send peer id and shm fd */
 175     if (ivshmem_server_send_initial_info(server, peer) < 0) {
 176         IVSHMEM_SERVER_DEBUG(server, "cannot send initial info\n");
 177         goto fail;
 178     }
 179
 180     /* advertise the new peer to others */
 181     QTAILQ_FOREACH(other_peer, &server->peer_list, next) {
 182         for (i = 0; i < peer->vectors_count; i++) {
 183             ivshmem_server_send_one_msg(other_peer->sock_fd, peer->id,
 184                                         peer->vectors[i].wfd);
 185         }
 186     }
 187
 188     /* advertise the other peers to the new one */
 189     QTAILQ_FOREACH(other_peer, &server->peer_list, next) {
 190         for (i = 0; i < peer->vectors_count; i++) {
 191             ivshmem_server_send_one_msg(peer->sock_fd, other_peer->id,
 192                                         other_peer->vectors[i].wfd);
 193         }
 194     }
 195
 196     /* advertise the new peer to itself */
 197     for (i = 0; i < peer->vectors_count; i++) {
 198         ivshmem_server_send_one_msg(peer->sock_fd, peer->id,
 199                                     event_notifier_get_fd(&peer->vectors[i]));
 200     }
 201
 202     QTAILQ_INSERT_TAIL(&server->peer_list, peer, next);
 203     IVSHMEM_SERVER_DEBUG(server, "new peer id = %ld\n",
 204                          peer->id);
 205     return 0;
 206
 207 fail:
 208     while (i--) {
 209         event_notifier_cleanup(&peer->vectors[i]);
 210     }
 211     close(newfd);
 212     g_free(peer);
 213     return -1;
 214 }
 215
 216 /* Try to ftruncate a file to next power of 2 of shmsize.
 217  * If it fails; all power of 2 above shmsize are tested until
 218  * we reach the maximum huge page size. This is useful
 219  * if the shm file is in a hugetlbfs that cannot be truncated to the
 220  * shm_size value. */
 221 static int
 222 ivshmem_server_ftruncate(int fd, unsigned shmsize)
 223 {
 224     int ret;
 225     struct stat mapstat;
 226
 227     /* align shmsize to next power of 2 */
 228     shmsize = pow2ceil(shmsize);
 229
 230     if (fstat(fd, &mapstat) != -1 && mapstat.st_size == shmsize) {
 231         return 0;
 232     }
 233
 234     while (shmsize <= IVSHMEM_SERVER_MAX_HUGEPAGE_SIZE) {
 235         ret = ftruncate(fd, shmsize);
 236         if (ret == 0) {
 237             return ret;
 238         }
 239         shmsize *= 2;
 240     }
 241
 242     return -1;
 243 }
 244
 245 /* Init a new ivshmem server */
 246 int
 247 ivshmem_server_init(IvshmemServer *server, const char *unix_sock_path,
 248                     const char *shm_path, size_t shm_size, unsigned n_vectors,
 249                     bool verbose)
 250 {
 251     int ret;
 252
 253     memset(server, 0, sizeof(*server));
 254     server->verbose = verbose;
 255
 256     ret = snprintf(server->unix_sock_path, sizeof(server->unix_sock_path),
 257                    "%s", unix_sock_path);
 258     if (ret < 0 || ret >= sizeof(server->unix_sock_path)) {
 259         IVSHMEM_SERVER_DEBUG(server, "could not copy unix socket path\n");
 260         return -1;
 261     }
 262     ret = snprintf(server->shm_path, sizeof(server->shm_path),
 263                    "%s", shm_path);
 264     if (ret < 0 || ret >= sizeof(server->shm_path)) {
 265         IVSHMEM_SERVER_DEBUG(server, "could not copy shm path\n");
 266         return -1;
 267     }
 268
 269     server->shm_size = shm_size;
 270     server->n_vectors = n_vectors;
 271
 272     QTAILQ_INIT(&server->peer_list);
 273
 274     return 0;
 275 }
 276
 277 #ifdef CONFIG_LINUX
 278
 279 #define HUGETLBFS_MAGIC       0x958458f6
 280
 281 static long gethugepagesize(const char *path)
 282 {
 283     struct statfs fs;
 284     int ret;
 285
 286     do {
 287         ret = statfs(path, &fs);
 288     } while (ret != 0 && errno == EINTR);
 289
 290     if (ret != 0) {
 291         return -1;
 292     }
 293
 294     if (fs.f_type != HUGETLBFS_MAGIC) {
 295         return -1;
 296     }
 297
 298     return fs.f_bsize;
 299 }
 300 #endif
 301
 302 /* open shm, create and bind to the unix socket */
 303 int
 304 ivshmem_server_start(IvshmemServer *server)
 305 {
 306     struct sockaddr_un sun;
 307     int shm_fd, sock_fd, ret;
 308
 309     /* open shm file */
 310 #ifdef CONFIG_LINUX
 311     long hpagesize;
 312
 313     hpagesize = gethugepagesize(server->shm_path);
 314     if (hpagesize < 0 && errno != ENOENT) {
 315         IVSHMEM_SERVER_DEBUG(server, "cannot stat shm file %s: %s\n",
 316                              server->shm_path, strerror(errno));
 317     }
 318
 319     if (hpagesize > 0) {
 320         gchar *filename = g_strdup_printf("%s/ivshmem.XXXXXX", server->shm_path);
 321         IVSHMEM_SERVER_DEBUG(server, "Using hugepages: %s\n", server->shm_path);
 322         shm_fd = mkstemp(filename);
 323         unlink(filename);
 324         g_free(filename);
 325     } else
 326 #endif
 327     {
 328         IVSHMEM_SERVER_DEBUG(server, "Using POSIX shared memory: %s\n",
 329                              server->shm_path);
 330         shm_fd = shm_open(server->shm_path, O_CREAT|O_RDWR, S_IRWXU);
 331     }
 332
 333     if (shm_fd < 0) {
 334         fprintf(stderr, "cannot open shm file %s: %s\n", server->shm_path,
 335                 strerror(errno));
 336         return -1;
 337     }
 338     if (ivshmem_server_ftruncate(shm_fd, server->shm_size) < 0) {
 339         fprintf(stderr, "ftruncate(%s) failed: %s\n", server->shm_path,
 340                 strerror(errno));
 341         goto err_close_shm;
 342     }
 343
 344     IVSHMEM_SERVER_DEBUG(server, "create & bind socket %s\n",
 345                          server->unix_sock_path);
 346
 347     /* create the unix listening socket */
 348     sock_fd = socket(AF_UNIX, SOCK_STREAM, 0);
 349     if (sock_fd < 0) {
 350         IVSHMEM_SERVER_DEBUG(server, "cannot create socket: %s\n",
 351                              strerror(errno));
 352         goto err_close_shm;
 353     }
 354
 355     sun.sun_family = AF_UNIX;
 356     ret = snprintf(sun.sun_path, sizeof(sun.sun_path), "%s",
 357                    server->unix_sock_path);
 358     if (ret < 0 || ret >= sizeof(sun.sun_path)) {
 359         IVSHMEM_SERVER_DEBUG(server, "could not copy unix socket path\n");
 360         goto err_close_sock;
 361     }
 362     if (bind(sock_fd, (struct sockaddr *)&sun, sizeof(sun)) < 0) {
 363         IVSHMEM_SERVER_DEBUG(server, "cannot connect to %s: %s\n", sun.sun_path,
 364                              strerror(errno));
 365         goto err_close_sock;
 366     }
 367
 368     if (listen(sock_fd, IVSHMEM_SERVER_LISTEN_BACKLOG) < 0) {
 369         IVSHMEM_SERVER_DEBUG(server, "listen() failed: %s\n", strerror(errno));
 370         goto err_close_sock;
 371     }
 372
 373     server->sock_fd = sock_fd;
 374     server->shm_fd = shm_fd;
 375
 376     return 0;
 377
 378 err_close_sock:
 379     close(sock_fd);
 380 err_close_shm:
 381     close(shm_fd);
 382     return -1;
 383 }
 384
 385 /* close connections to clients, the unix socket and the shm fd */
 386 void
 387 ivshmem_server_close(IvshmemServer *server)
 388 {
 389     IvshmemServerPeer *peer, *npeer;
 390
 391     IVSHMEM_SERVER_DEBUG(server, "close server\n");
 392
 393     QTAILQ_FOREACH_SAFE(peer, &server->peer_list, next, npeer) {
 394         ivshmem_server_free_peer(server, peer);
 395     }
 396
 397     unlink(server->unix_sock_path);
 398     close(server->sock_fd);
 399     close(server->shm_fd);
 400     server->sock_fd = -1;
 401     server->shm_fd = -1;
 402 }
 403
 404 /* get the fd_set according to the unix socket and the peer list */
 405 void
 406 ivshmem_server_get_fds(const IvshmemServer *server, fd_set *fds, int *maxfd)
 407 {
 408     IvshmemServerPeer *peer;
 409
 410     if (server->sock_fd == -1) {
 411         return;
 412     }
 413
 414     FD_SET(server->sock_fd, fds);
 415     if (server->sock_fd >= *maxfd) {
 416         *maxfd = server->sock_fd + 1;
 417     }
 418
 419     QTAILQ_FOREACH(peer, &server->peer_list, next) {
 420         FD_SET(peer->sock_fd, fds);
 421         if (peer->sock_fd >= *maxfd) {
 422             *maxfd = peer->sock_fd + 1;
 423         }
 424     }
 425 }
 426
 427 /* process incoming messages on the sockets in fd_set */
 428 int
 429 ivshmem_server_handle_fds(IvshmemServer *server, fd_set *fds, int maxfd)
 430 {
 431     IvshmemServerPeer *peer, *peer_next;
 432
 433     if (server->sock_fd < maxfd && FD_ISSET(server->sock_fd, fds) &&
 434         ivshmem_server_handle_new_conn(server) < 0 && errno != EINTR) {
 435         IVSHMEM_SERVER_DEBUG(server, "ivshmem_server_handle_new_conn() "
 436                              "failed\n");
 437         return -1;
 438     }
 439
 440     QTAILQ_FOREACH_SAFE(peer, &server->peer_list, next, peer_next) {
 441         /* any message from a peer socket result in a close() */
 442         IVSHMEM_SERVER_DEBUG(server, "peer->sock_fd=%d\n", peer->sock_fd);
 443         if (peer->sock_fd < maxfd && FD_ISSET(peer->sock_fd, fds)) {
 444             ivshmem_server_free_peer(server, peer);
 445         }
 446     }
 447
 448     return 0;
 449 }
 450
 451 /* lookup peer from its id */
 452 IvshmemServerPeer *
 453 ivshmem_server_search_peer(IvshmemServer *server, long peer_id)
 454 {
 455     IvshmemServerPeer *peer;
 456
 457     QTAILQ_FOREACH(peer, &server->peer_list, next) {
 458         if (peer->id == peer_id) {
 459             return peer;
 460         }
 461     }
 462     return NULL;
 463 }
 464
 465 /* dump our info, the list of peers their vectors on stdout */
 466 void
 467 ivshmem_server_dump(const IvshmemServer *server)
 468 {
 469     const IvshmemServerPeer *peer;
 470     unsigned vector;
 471
 472     /* dump peers */
 473     QTAILQ_FOREACH(peer, &server->peer_list, next) {
 474         printf("peer_id = %ld\n", peer->id);
 475
 476         for (vector = 0; vector < peer->vectors_count; vector++) {
 477             printf("  vector %d is enabled (fd=%d)\n", vector,
 478                    event_notifier_get_fd(&peer->vectors[vector]));
 479         }
 480     }
 481 }