1 /* 2 * Vhost User Bridge 3 * 4 * Copyright (c) 2015 Red Hat, Inc. 5 * 6 * Authors: 7 * Victor Kaplansky <victork@redhat.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2 or 10 * later. See the COPYING file in the top-level directory. 11 */ 12 13 /* 14 * TODO: 15 * - main should get parameters from the command line. 16 * - implement all request handlers. Still not implemented: 17 * vubr_get_queue_num_exec() 18 * vubr_send_rarp_exec() 19 * - test for broken requests and virtqueue. 20 * - implement features defined by Virtio 1.0 spec. 21 * - support mergeable buffers and indirect descriptors. 22 * - implement clean shutdown. 23 * - implement non-blocking writes to UDP backend. 24 * - implement polling strategy. 25 * - implement clean starting/stopping of vq processing 26 * - implement clean starting/stopping of used and buffers 27 * dirty page logging. 28 */ 29 30 #define _FILE_OFFSET_BITS 64 31 32 #include "qemu/osdep.h" 33 #include "qemu/iov.h" 34 #include "standard-headers/linux/virtio_net.h" 35 #include "contrib/libvhost-user/libvhost-user.h" 36 37 #define VHOST_USER_BRIDGE_DEBUG 1 38 39 #define DPRINT(...) \ 40 do { \ 41 if (VHOST_USER_BRIDGE_DEBUG) { \ 42 printf(__VA_ARGS__); \ 43 } \ 44 } while (0) 45 46 typedef void (*CallbackFunc)(int sock, void *ctx); 47 48 typedef struct Event { 49 void *ctx; 50 CallbackFunc callback; 51 } Event; 52 53 typedef struct Dispatcher { 54 int max_sock; 55 fd_set fdset; 56 Event events[FD_SETSIZE]; 57 } Dispatcher; 58 59 typedef struct VubrDev { 60 VuDev vudev; 61 Dispatcher dispatcher; 62 int backend_udp_sock; 63 struct sockaddr_in backend_udp_dest; 64 int hdrlen; 65 int sock; 66 int ready; 67 int quit; 68 } VubrDev; 69 70 static void 71 vubr_die(const char *s) 72 { 73 perror(s); 74 exit(1); 75 } 76 77 static int 78 dispatcher_init(Dispatcher *dispr) 79 { 80 FD_ZERO(&dispr->fdset); 81 dispr->max_sock = -1; 82 return 0; 83 } 84 85 static int 86 dispatcher_add(Dispatcher *dispr, int sock, void *ctx, CallbackFunc cb) 87 { 88 if (sock >= FD_SETSIZE) { 89 fprintf(stderr, 90 "Error: Failed to add new event. sock %d should be less than %d\n", 91 sock, FD_SETSIZE); 92 return -1; 93 } 94 95 dispr->events[sock].ctx = ctx; 96 dispr->events[sock].callback = cb; 97 98 FD_SET(sock, &dispr->fdset); 99 if (sock > dispr->max_sock) { 100 dispr->max_sock = sock; 101 } 102 DPRINT("Added sock %d for watching. max_sock: %d\n", 103 sock, dispr->max_sock); 104 return 0; 105 } 106 107 static int 108 dispatcher_remove(Dispatcher *dispr, int sock) 109 { 110 if (sock >= FD_SETSIZE) { 111 fprintf(stderr, 112 "Error: Failed to remove event. sock %d should be less than %d\n", 113 sock, FD_SETSIZE); 114 return -1; 115 } 116 117 FD_CLR(sock, &dispr->fdset); 118 DPRINT("Sock %d removed from dispatcher watch.\n", sock); 119 return 0; 120 } 121 122 /* timeout in us */ 123 static int 124 dispatcher_wait(Dispatcher *dispr, uint32_t timeout) 125 { 126 struct timeval tv; 127 tv.tv_sec = timeout / 1000000; 128 tv.tv_usec = timeout % 1000000; 129 130 fd_set fdset = dispr->fdset; 131 132 /* wait until some of sockets become readable. */ 133 int rc = select(dispr->max_sock + 1, &fdset, 0, 0, &tv); 134 135 if (rc == -1) { 136 vubr_die("select"); 137 } 138 139 /* Timeout */ 140 if (rc == 0) { 141 return 0; 142 } 143 144 /* Now call callback for every ready socket. */ 145 146 int sock; 147 for (sock = 0; sock < dispr->max_sock + 1; sock++) { 148 /* The callback on a socket can remove other sockets from the 149 * dispatcher, thus we have to check that the socket is 150 * still not removed from dispatcher's list 151 */ 152 if (FD_ISSET(sock, &fdset) && FD_ISSET(sock, &dispr->fdset)) { 153 Event *e = &dispr->events[sock]; 154 e->callback(sock, e->ctx); 155 } 156 } 157 158 return 0; 159 } 160 161 static void 162 vubr_handle_tx(VuDev *dev, int qidx) 163 { 164 VuVirtq *vq = vu_get_queue(dev, qidx); 165 VubrDev *vubr = container_of(dev, VubrDev, vudev); 166 int hdrlen = vubr->hdrlen; 167 VuVirtqElement *elem = NULL; 168 169 assert(qidx % 2); 170 171 for (;;) { 172 ssize_t ret; 173 unsigned int out_num; 174 struct iovec sg[VIRTQUEUE_MAX_SIZE], *out_sg; 175 176 elem = vu_queue_pop(dev, vq, sizeof(VuVirtqElement)); 177 if (!elem) { 178 break; 179 } 180 181 out_num = elem->out_num; 182 out_sg = elem->out_sg; 183 if (out_num < 1) { 184 fprintf(stderr, "virtio-net header not in first element\n"); 185 break; 186 } 187 if (VHOST_USER_BRIDGE_DEBUG) { 188 iov_hexdump(out_sg, out_num, stderr, "TX:", 1024); 189 } 190 191 if (hdrlen) { 192 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), 193 out_sg, out_num, 194 hdrlen, -1); 195 out_num = sg_num; 196 out_sg = sg; 197 } 198 199 struct msghdr msg = { 200 .msg_name = (struct sockaddr *) &vubr->backend_udp_dest, 201 .msg_namelen = sizeof(struct sockaddr_in), 202 .msg_iov = out_sg, 203 .msg_iovlen = out_num, 204 }; 205 do { 206 ret = sendmsg(vubr->backend_udp_sock, &msg, 0); 207 } while (ret == -1 && (errno == EAGAIN || errno == EINTR)); 208 209 if (ret == -1) { 210 vubr_die("sendmsg()"); 211 } 212 213 vu_queue_push(dev, vq, elem, 0); 214 vu_queue_notify(dev, vq); 215 216 free(elem); 217 elem = NULL; 218 } 219 220 free(elem); 221 } 222 223 224 /* this function reverse the effect of iov_discard_front() it must be 225 * called with 'front' being the original struct iovec and 'bytes' 226 * being the number of bytes you shaved off 227 */ 228 static void 229 iov_restore_front(struct iovec *front, struct iovec *iov, size_t bytes) 230 { 231 struct iovec *cur; 232 233 for (cur = front; cur != iov; cur++) { 234 assert(bytes >= cur->iov_len); 235 bytes -= cur->iov_len; 236 } 237 238 cur->iov_base -= bytes; 239 cur->iov_len += bytes; 240 } 241 242 static void 243 iov_truncate(struct iovec *iov, unsigned iovc, size_t bytes) 244 { 245 unsigned i; 246 247 for (i = 0; i < iovc; i++, iov++) { 248 if (bytes < iov->iov_len) { 249 iov->iov_len = bytes; 250 return; 251 } 252 253 bytes -= iov->iov_len; 254 } 255 256 assert(!"couldn't truncate iov"); 257 } 258 259 static void 260 vubr_backend_recv_cb(int sock, void *ctx) 261 { 262 VubrDev *vubr = (VubrDev *) ctx; 263 VuDev *dev = &vubr->vudev; 264 VuVirtq *vq = vu_get_queue(dev, 0); 265 VuVirtqElement *elem = NULL; 266 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; 267 struct virtio_net_hdr_mrg_rxbuf mhdr; 268 unsigned mhdr_cnt = 0; 269 int hdrlen = vubr->hdrlen; 270 int i = 0; 271 struct virtio_net_hdr hdr = { 272 .flags = 0, 273 .gso_type = VIRTIO_NET_HDR_GSO_NONE 274 }; 275 276 DPRINT("\n\n *** IN UDP RECEIVE CALLBACK ***\n\n"); 277 DPRINT(" hdrlen = %d\n", hdrlen); 278 279 if (!vu_queue_enabled(dev, vq) || 280 !vu_queue_avail_bytes(dev, vq, hdrlen, 0)) { 281 DPRINT("Got UDP packet, but no available descriptors on RX virtq.\n"); 282 return; 283 } 284 285 do { 286 struct iovec *sg; 287 ssize_t ret, total = 0; 288 unsigned int num; 289 290 elem = vu_queue_pop(dev, vq, sizeof(VuVirtqElement)); 291 if (!elem) { 292 break; 293 } 294 295 if (elem->in_num < 1) { 296 fprintf(stderr, "virtio-net contains no in buffers\n"); 297 break; 298 } 299 300 sg = elem->in_sg; 301 num = elem->in_num; 302 if (i == 0) { 303 if (hdrlen == 12) { 304 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), 305 sg, elem->in_num, 306 offsetof(typeof(mhdr), num_buffers), 307 sizeof(mhdr.num_buffers)); 308 } 309 iov_from_buf(sg, elem->in_num, 0, &hdr, sizeof hdr); 310 total += hdrlen; 311 ret = iov_discard_front(&sg, &num, hdrlen); 312 assert(ret == hdrlen); 313 } 314 315 struct msghdr msg = { 316 .msg_name = (struct sockaddr *) &vubr->backend_udp_dest, 317 .msg_namelen = sizeof(struct sockaddr_in), 318 .msg_iov = sg, 319 .msg_iovlen = elem->in_num, 320 .msg_flags = MSG_DONTWAIT, 321 }; 322 do { 323 ret = recvmsg(vubr->backend_udp_sock, &msg, 0); 324 } while (ret == -1 && (errno == EINTR)); 325 326 if (i == 0) { 327 iov_restore_front(elem->in_sg, sg, hdrlen); 328 } 329 330 if (ret == -1) { 331 if (errno == EWOULDBLOCK) { 332 vu_queue_rewind(dev, vq, 1); 333 break; 334 } 335 336 vubr_die("recvmsg()"); 337 } 338 339 total += ret; 340 iov_truncate(elem->in_sg, elem->in_num, total); 341 vu_queue_fill(dev, vq, elem, total, i++); 342 343 free(elem); 344 elem = NULL; 345 } while (false); /* could loop if DONTWAIT worked? */ 346 347 if (mhdr_cnt) { 348 mhdr.num_buffers = i; 349 iov_from_buf(mhdr_sg, mhdr_cnt, 350 0, 351 &mhdr.num_buffers, sizeof mhdr.num_buffers); 352 } 353 354 vu_queue_flush(dev, vq, i); 355 vu_queue_notify(dev, vq); 356 357 free(elem); 358 } 359 360 static void 361 vubr_receive_cb(int sock, void *ctx) 362 { 363 VubrDev *vubr = (VubrDev *)ctx; 364 365 if (!vu_dispatch(&vubr->vudev)) { 366 fprintf(stderr, "Error while dispatching\n"); 367 } 368 } 369 370 typedef struct WatchData { 371 VuDev *dev; 372 vu_watch_cb cb; 373 void *data; 374 } WatchData; 375 376 static void 377 watch_cb(int sock, void *ctx) 378 { 379 struct WatchData *wd = ctx; 380 381 wd->cb(wd->dev, VU_WATCH_IN, wd->data); 382 } 383 384 static void 385 vubr_set_watch(VuDev *dev, int fd, int condition, 386 vu_watch_cb cb, void *data) 387 { 388 VubrDev *vubr = container_of(dev, VubrDev, vudev); 389 static WatchData watches[FD_SETSIZE]; 390 struct WatchData *wd = &watches[fd]; 391 392 wd->cb = cb; 393 wd->data = data; 394 wd->dev = dev; 395 dispatcher_add(&vubr->dispatcher, fd, wd, watch_cb); 396 } 397 398 static void 399 vubr_remove_watch(VuDev *dev, int fd) 400 { 401 VubrDev *vubr = container_of(dev, VubrDev, vudev); 402 403 dispatcher_remove(&vubr->dispatcher, fd); 404 } 405 406 static int 407 vubr_send_rarp_exec(VuDev *dev, VhostUserMsg *vmsg) 408 { 409 DPRINT("Function %s() not implemented yet.\n", __func__); 410 return 0; 411 } 412 413 static int 414 vubr_process_msg(VuDev *dev, VhostUserMsg *vmsg, int *do_reply) 415 { 416 switch (vmsg->request) { 417 case VHOST_USER_SEND_RARP: 418 *do_reply = vubr_send_rarp_exec(dev, vmsg); 419 return 1; 420 default: 421 /* let the library handle the rest */ 422 return 0; 423 } 424 425 return 0; 426 } 427 428 static void 429 vubr_set_features(VuDev *dev, uint64_t features) 430 { 431 VubrDev *vubr = container_of(dev, VubrDev, vudev); 432 433 if ((features & (1ULL << VIRTIO_F_VERSION_1)) || 434 (features & (1ULL << VIRTIO_NET_F_MRG_RXBUF))) { 435 vubr->hdrlen = 12; 436 } else { 437 vubr->hdrlen = 10; 438 } 439 } 440 441 static uint64_t 442 vubr_get_features(VuDev *dev) 443 { 444 return 1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE | 445 1ULL << VIRTIO_NET_F_MRG_RXBUF; 446 } 447 448 static void 449 vubr_queue_set_started(VuDev *dev, int qidx, bool started) 450 { 451 VuVirtq *vq = vu_get_queue(dev, qidx); 452 453 if (qidx % 2 == 1) { 454 vu_set_queue_handler(dev, vq, started ? vubr_handle_tx : NULL); 455 } 456 } 457 458 static void 459 vubr_panic(VuDev *dev, const char *msg) 460 { 461 VubrDev *vubr = container_of(dev, VubrDev, vudev); 462 463 fprintf(stderr, "PANIC: %s\n", msg); 464 465 dispatcher_remove(&vubr->dispatcher, dev->sock); 466 vubr->quit = 1; 467 } 468 469 static bool 470 vubr_queue_is_processed_in_order(VuDev *dev, int qidx) 471 { 472 return true; 473 } 474 475 static const VuDevIface vuiface = { 476 .get_features = vubr_get_features, 477 .set_features = vubr_set_features, 478 .process_msg = vubr_process_msg, 479 .queue_set_started = vubr_queue_set_started, 480 .queue_is_processed_in_order = vubr_queue_is_processed_in_order, 481 }; 482 483 static void 484 vubr_accept_cb(int sock, void *ctx) 485 { 486 VubrDev *dev = (VubrDev *)ctx; 487 int conn_fd; 488 struct sockaddr_un un; 489 socklen_t len = sizeof(un); 490 491 conn_fd = accept(sock, (struct sockaddr *) &un, &len); 492 if (conn_fd == -1) { 493 vubr_die("accept()"); 494 } 495 DPRINT("Got connection from remote peer on sock %d\n", conn_fd); 496 497 vu_init(&dev->vudev, 498 conn_fd, 499 vubr_panic, 500 vubr_set_watch, 501 vubr_remove_watch, 502 &vuiface); 503 504 dispatcher_add(&dev->dispatcher, conn_fd, ctx, vubr_receive_cb); 505 dispatcher_remove(&dev->dispatcher, sock); 506 } 507 508 static VubrDev * 509 vubr_new(const char *path, bool client) 510 { 511 VubrDev *dev = (VubrDev *) calloc(1, sizeof(VubrDev)); 512 struct sockaddr_un un; 513 CallbackFunc cb; 514 size_t len; 515 516 /* Get a UNIX socket. */ 517 dev->sock = socket(AF_UNIX, SOCK_STREAM, 0); 518 if (dev->sock == -1) { 519 vubr_die("socket"); 520 } 521 522 un.sun_family = AF_UNIX; 523 strcpy(un.sun_path, path); 524 len = sizeof(un.sun_family) + strlen(path); 525 526 if (!client) { 527 unlink(path); 528 529 if (bind(dev->sock, (struct sockaddr *) &un, len) == -1) { 530 vubr_die("bind"); 531 } 532 533 if (listen(dev->sock, 1) == -1) { 534 vubr_die("listen"); 535 } 536 cb = vubr_accept_cb; 537 538 DPRINT("Waiting for connections on UNIX socket %s ...\n", path); 539 } else { 540 if (connect(dev->sock, (struct sockaddr *)&un, len) == -1) { 541 vubr_die("connect"); 542 } 543 vu_init(&dev->vudev, 544 dev->sock, 545 vubr_panic, 546 vubr_set_watch, 547 vubr_remove_watch, 548 &vuiface); 549 cb = vubr_receive_cb; 550 } 551 552 dispatcher_init(&dev->dispatcher); 553 554 dispatcher_add(&dev->dispatcher, dev->sock, (void *)dev, cb); 555 556 return dev; 557 } 558 559 static void 560 vubr_set_host(struct sockaddr_in *saddr, const char *host) 561 { 562 if (isdigit(host[0])) { 563 if (!inet_aton(host, &saddr->sin_addr)) { 564 fprintf(stderr, "inet_aton() failed.\n"); 565 exit(1); 566 } 567 } else { 568 struct hostent *he = gethostbyname(host); 569 570 if (!he) { 571 fprintf(stderr, "gethostbyname() failed.\n"); 572 exit(1); 573 } 574 saddr->sin_addr = *(struct in_addr *)he->h_addr; 575 } 576 } 577 578 static void 579 vubr_backend_udp_setup(VubrDev *dev, 580 const char *local_host, 581 const char *local_port, 582 const char *remote_host, 583 const char *remote_port) 584 { 585 int sock; 586 const char *r; 587 588 int lport, rport; 589 590 lport = strtol(local_port, (char **)&r, 0); 591 if (r == local_port) { 592 fprintf(stderr, "lport parsing failed.\n"); 593 exit(1); 594 } 595 596 rport = strtol(remote_port, (char **)&r, 0); 597 if (r == remote_port) { 598 fprintf(stderr, "rport parsing failed.\n"); 599 exit(1); 600 } 601 602 struct sockaddr_in si_local = { 603 .sin_family = AF_INET, 604 .sin_port = htons(lport), 605 }; 606 607 vubr_set_host(&si_local, local_host); 608 609 /* setup destination for sends */ 610 dev->backend_udp_dest = (struct sockaddr_in) { 611 .sin_family = AF_INET, 612 .sin_port = htons(rport), 613 }; 614 vubr_set_host(&dev->backend_udp_dest, remote_host); 615 616 sock = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP); 617 if (sock == -1) { 618 vubr_die("socket"); 619 } 620 621 if (bind(sock, (struct sockaddr *)&si_local, sizeof(si_local)) == -1) { 622 vubr_die("bind"); 623 } 624 625 dev->backend_udp_sock = sock; 626 dispatcher_add(&dev->dispatcher, sock, dev, vubr_backend_recv_cb); 627 DPRINT("Waiting for data from udp backend on %s:%d...\n", 628 local_host, lport); 629 } 630 631 static void 632 vubr_run(VubrDev *dev) 633 { 634 while (!dev->quit) { 635 /* timeout 200ms */ 636 dispatcher_wait(&dev->dispatcher, 200000); 637 /* Here one can try polling strategy. */ 638 } 639 } 640 641 static int 642 vubr_parse_host_port(const char **host, const char **port, const char *buf) 643 { 644 char *p = strchr(buf, ':'); 645 646 if (!p) { 647 return -1; 648 } 649 *p = '\0'; 650 *host = strdup(buf); 651 *port = strdup(p + 1); 652 return 0; 653 } 654 655 #define DEFAULT_UD_SOCKET "/tmp/vubr.sock" 656 #define DEFAULT_LHOST "127.0.0.1" 657 #define DEFAULT_LPORT "4444" 658 #define DEFAULT_RHOST "127.0.0.1" 659 #define DEFAULT_RPORT "5555" 660 661 static const char *ud_socket_path = DEFAULT_UD_SOCKET; 662 static const char *lhost = DEFAULT_LHOST; 663 static const char *lport = DEFAULT_LPORT; 664 static const char *rhost = DEFAULT_RHOST; 665 static const char *rport = DEFAULT_RPORT; 666 667 int 668 main(int argc, char *argv[]) 669 { 670 VubrDev *dev; 671 int opt; 672 bool client = false; 673 674 while ((opt = getopt(argc, argv, "l:r:u:c")) != -1) { 675 676 switch (opt) { 677 case 'l': 678 if (vubr_parse_host_port(&lhost, &lport, optarg) < 0) { 679 goto out; 680 } 681 break; 682 case 'r': 683 if (vubr_parse_host_port(&rhost, &rport, optarg) < 0) { 684 goto out; 685 } 686 break; 687 case 'u': 688 ud_socket_path = strdup(optarg); 689 break; 690 case 'c': 691 client = true; 692 break; 693 default: 694 goto out; 695 } 696 } 697 698 DPRINT("ud socket: %s (%s)\n", ud_socket_path, 699 client ? "client" : "server"); 700 DPRINT("local: %s:%s\n", lhost, lport); 701 DPRINT("remote: %s:%s\n", rhost, rport); 702 703 dev = vubr_new(ud_socket_path, client); 704 if (!dev) { 705 return 1; 706 } 707 708 vubr_backend_udp_setup(dev, lhost, lport, rhost, rport); 709 vubr_run(dev); 710 711 vu_deinit(&dev->vudev); 712 713 return 0; 714 715 out: 716 fprintf(stderr, "Usage: %s ", argv[0]); 717 fprintf(stderr, "[-c] [-u ud_socket_path] [-l lhost:lport] [-r rhost:rport]\n"); 718 fprintf(stderr, "\t-u path to unix doman socket. default: %s\n", 719 DEFAULT_UD_SOCKET); 720 fprintf(stderr, "\t-l local host and port. default: %s:%s\n", 721 DEFAULT_LHOST, DEFAULT_LPORT); 722 fprintf(stderr, "\t-r remote host and port. default: %s:%s\n", 723 DEFAULT_RHOST, DEFAULT_RPORT); 724 fprintf(stderr, "\t-c client mode\n"); 725 726 return 1; 727 } 728