| 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/atomic.h" |
| 34 | #include "qemu/ctype.h" |
| 35 | #include "qemu/iov.h" |
| 36 | #include "standard-headers/linux/virtio_net.h" |
| 37 | #include "libvhost-user.h" |
| 38 | |
| 39 | #define VHOST_USER_BRIDGE_DEBUG 1 |
| 40 | |
| 41 | #define DPRINT(...) \ |
| 42 | do { \ |
| 43 | if (VHOST_USER_BRIDGE_DEBUG) { \ |
| 44 | printf(__VA_ARGS__); \ |
| 45 | } \ |
| 46 | } while (0) |
| 47 | |
| 48 | enum { |
| 49 | VHOST_USER_BRIDGE_MAX_QUEUES = 8, |
| 50 | }; |
| 51 | |
| 52 | typedef void (*CallbackFunc)(int sock, void *ctx); |
| 53 | |
| 54 | typedef struct Event { |
| 55 | void *ctx; |
| 56 | CallbackFunc callback; |
| 57 | } Event; |
| 58 | |
| 59 | typedef struct Dispatcher { |
| 60 | int max_sock; |
| 61 | fd_set fdset; |
| 62 | Event events[FD_SETSIZE]; |
| 63 | } Dispatcher; |
| 64 | |
| 65 | typedef struct VubrDev { |
| 66 | VuDev vudev; |
| 67 | Dispatcher dispatcher; |
| 68 | int backend_udp_sock; |
| 69 | struct sockaddr_in backend_udp_dest; |
| 70 | int hdrlen; |
| 71 | int sock; |
| 72 | int ready; |
| 73 | int quit; |
| 74 | struct { |
| 75 | int fd; |
| 76 | void *addr; |
| 77 | pthread_t thread; |
| 78 | } notifier; |
| 79 | } VubrDev; |
| 80 | |
| 81 | static void |
| 82 | vubr_die(const char *s) |
| 83 | { |
| 84 | perror(s); |
| 85 | exit(1); |
| 86 | } |
| 87 | |
| 88 | static int |
| 89 | dispatcher_init(Dispatcher *dispr) |
| 90 | { |
| 91 | FD_ZERO(&dispr->fdset); |
| 92 | dispr->max_sock = -1; |
| 93 | return 0; |
| 94 | } |
| 95 | |
| 96 | static int |
| 97 | dispatcher_add(Dispatcher *dispr, int sock, void *ctx, CallbackFunc cb) |
| 98 | { |
| 99 | if (sock >= FD_SETSIZE) { |
| 100 | fprintf(stderr, |
| 101 | "Error: Failed to add new event. sock %d should be less than %d\n", |
| 102 | sock, FD_SETSIZE); |
| 103 | return -1; |
| 104 | } |
| 105 | |
| 106 | dispr->events[sock].ctx = ctx; |
| 107 | dispr->events[sock].callback = cb; |
| 108 | |
| 109 | FD_SET(sock, &dispr->fdset); |
| 110 | if (sock > dispr->max_sock) { |
| 111 | dispr->max_sock = sock; |
| 112 | } |
| 113 | DPRINT("Added sock %d for watching. max_sock: %d\n", |
| 114 | sock, dispr->max_sock); |
| 115 | return 0; |
| 116 | } |
| 117 | |
| 118 | static int |
| 119 | dispatcher_remove(Dispatcher *dispr, int sock) |
| 120 | { |
| 121 | if (sock >= FD_SETSIZE) { |
| 122 | fprintf(stderr, |
| 123 | "Error: Failed to remove event. sock %d should be less than %d\n", |
| 124 | sock, FD_SETSIZE); |
| 125 | return -1; |
| 126 | } |
| 127 | |
| 128 | FD_CLR(sock, &dispr->fdset); |
| 129 | DPRINT("Sock %d removed from dispatcher watch.\n", sock); |
| 130 | return 0; |
| 131 | } |
| 132 | |
| 133 | /* timeout in us */ |
| 134 | static int |
| 135 | dispatcher_wait(Dispatcher *dispr, uint32_t timeout) |
| 136 | { |
| 137 | struct timeval tv; |
| 138 | tv.tv_sec = timeout / 1000000; |
| 139 | tv.tv_usec = timeout % 1000000; |
| 140 | |
| 141 | fd_set fdset = dispr->fdset; |
| 142 | |
| 143 | /* wait until some of sockets become readable. */ |
| 144 | int rc = select(dispr->max_sock + 1, &fdset, 0, 0, &tv); |
| 145 | |
| 146 | if (rc == -1) { |
| 147 | vubr_die("select"); |
| 148 | } |
| 149 | |
| 150 | /* Timeout */ |
| 151 | if (rc == 0) { |
| 152 | return 0; |
| 153 | } |
| 154 | |
| 155 | /* Now call callback for every ready socket. */ |
| 156 | |
| 157 | int sock; |
| 158 | for (sock = 0; sock < dispr->max_sock + 1; sock++) { |
| 159 | /* The callback on a socket can remove other sockets from the |
| 160 | * dispatcher, thus we have to check that the socket is |
| 161 | * still not removed from dispatcher's list |
| 162 | */ |
| 163 | if (FD_ISSET(sock, &fdset) && FD_ISSET(sock, &dispr->fdset)) { |
| 164 | Event *e = &dispr->events[sock]; |
| 165 | e->callback(sock, e->ctx); |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | return 0; |
| 170 | } |
| 171 | |
| 172 | static void |
| 173 | vubr_handle_tx(VuDev *dev, int qidx) |
| 174 | { |
| 175 | VuVirtq *vq = vu_get_queue(dev, qidx); |
| 176 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
| 177 | int hdrlen = vubr->hdrlen; |
| 178 | VuVirtqElement *elem = NULL; |
| 179 | |
| 180 | assert(qidx % 2); |
| 181 | |
| 182 | DPRINT("\n\n *** IN UDP TRANSMIT HANDLER ***\n\n"); |
| 183 | DPRINT(" hdrlen = %d\n", hdrlen); |
| 184 | |
| 185 | for (;;) { |
| 186 | ssize_t ret; |
| 187 | unsigned int out_num; |
| 188 | struct iovec sg[VIRTQUEUE_MAX_SIZE], *out_sg; |
| 189 | |
| 190 | elem = vu_queue_pop(dev, vq, sizeof(VuVirtqElement)); |
| 191 | if (!elem) { |
| 192 | break; |
| 193 | } |
| 194 | |
| 195 | out_num = elem->out_num; |
| 196 | out_sg = elem->out_sg; |
| 197 | if (out_num < 1) { |
| 198 | fprintf(stderr, "virtio-net header not in first element\n"); |
| 199 | break; |
| 200 | } |
| 201 | if (VHOST_USER_BRIDGE_DEBUG) { |
| 202 | iov_hexdump(out_sg, out_num, stderr, "TX:", 1024); |
| 203 | } |
| 204 | |
| 205 | if (hdrlen) { |
| 206 | unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), |
| 207 | out_sg, out_num, |
| 208 | hdrlen, -1); |
| 209 | out_num = sg_num; |
| 210 | out_sg = sg; |
| 211 | } |
| 212 | |
| 213 | struct msghdr msg = { |
| 214 | .msg_name = (struct sockaddr *) &vubr->backend_udp_dest, |
| 215 | .msg_namelen = sizeof(struct sockaddr_in), |
| 216 | .msg_iov = out_sg, |
| 217 | .msg_iovlen = out_num, |
| 218 | }; |
| 219 | do { |
| 220 | ret = sendmsg(vubr->backend_udp_sock, &msg, 0); |
| 221 | } while (ret == -1 && (errno == EAGAIN || errno == EINTR)); |
| 222 | |
| 223 | if (ret == -1) { |
| 224 | vubr_die("sendmsg()"); |
| 225 | } |
| 226 | |
| 227 | vu_queue_push(dev, vq, elem, 0); |
| 228 | vu_queue_notify(dev, vq); |
| 229 | |
| 230 | free(elem); |
| 231 | elem = NULL; |
| 232 | } |
| 233 | |
| 234 | free(elem); |
| 235 | } |
| 236 | |
| 237 | |
| 238 | /* this function reverse the effect of iov_discard_front() it must be |
| 239 | * called with 'front' being the original struct iovec and 'bytes' |
| 240 | * being the number of bytes you shaved off |
| 241 | */ |
| 242 | static void |
| 243 | iov_restore_front(struct iovec *front, struct iovec *iov, size_t bytes) |
| 244 | { |
| 245 | struct iovec *cur; |
| 246 | |
| 247 | for (cur = front; cur != iov; cur++) { |
| 248 | assert(bytes >= cur->iov_len); |
| 249 | bytes -= cur->iov_len; |
| 250 | } |
| 251 | |
| 252 | cur->iov_base -= bytes; |
| 253 | cur->iov_len += bytes; |
| 254 | } |
| 255 | |
| 256 | static void |
| 257 | iov_truncate(struct iovec *iov, unsigned iovc, size_t bytes) |
| 258 | { |
| 259 | unsigned i; |
| 260 | |
| 261 | for (i = 0; i < iovc; i++, iov++) { |
| 262 | if (bytes < iov->iov_len) { |
| 263 | iov->iov_len = bytes; |
| 264 | return; |
| 265 | } |
| 266 | |
| 267 | bytes -= iov->iov_len; |
| 268 | } |
| 269 | |
| 270 | assert(!"couldn't truncate iov"); |
| 271 | } |
| 272 | |
| 273 | static void |
| 274 | vubr_backend_recv_cb(int sock, void *ctx) |
| 275 | { |
| 276 | VubrDev *vubr = (VubrDev *) ctx; |
| 277 | VuDev *dev = &vubr->vudev; |
| 278 | VuVirtq *vq = vu_get_queue(dev, 0); |
| 279 | VuVirtqElement *elem = NULL; |
| 280 | struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; |
| 281 | struct virtio_net_hdr_mrg_rxbuf mhdr; |
| 282 | unsigned mhdr_cnt = 0; |
| 283 | int hdrlen = vubr->hdrlen; |
| 284 | int i = 0; |
| 285 | struct virtio_net_hdr hdr = { |
| 286 | .flags = 0, |
| 287 | .gso_type = VIRTIO_NET_HDR_GSO_NONE |
| 288 | }; |
| 289 | |
| 290 | DPRINT("\n\n *** IN UDP RECEIVE CALLBACK ***\n\n"); |
| 291 | DPRINT(" hdrlen = %d\n", hdrlen); |
| 292 | |
| 293 | if (!vu_queue_enabled(dev, vq) || |
| 294 | !vu_queue_started(dev, vq) || |
| 295 | !vu_queue_avail_bytes(dev, vq, hdrlen, 0)) { |
| 296 | DPRINT("Got UDP packet, but no available descriptors on RX virtq.\n"); |
| 297 | return; |
| 298 | } |
| 299 | |
| 300 | while (1) { |
| 301 | struct iovec *sg; |
| 302 | ssize_t ret, total = 0; |
| 303 | unsigned int num; |
| 304 | |
| 305 | elem = vu_queue_pop(dev, vq, sizeof(VuVirtqElement)); |
| 306 | if (!elem) { |
| 307 | break; |
| 308 | } |
| 309 | |
| 310 | if (elem->in_num < 1) { |
| 311 | fprintf(stderr, "virtio-net contains no in buffers\n"); |
| 312 | break; |
| 313 | } |
| 314 | |
| 315 | sg = elem->in_sg; |
| 316 | num = elem->in_num; |
| 317 | if (i == 0) { |
| 318 | if (hdrlen == 12) { |
| 319 | mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), |
| 320 | sg, elem->in_num, |
| 321 | offsetof(typeof(mhdr), num_buffers), |
| 322 | sizeof(mhdr.num_buffers)); |
| 323 | } |
| 324 | iov_from_buf(sg, elem->in_num, 0, &hdr, sizeof hdr); |
| 325 | total += hdrlen; |
| 326 | ret = iov_discard_front(&sg, &num, hdrlen); |
| 327 | assert(ret == hdrlen); |
| 328 | } |
| 329 | |
| 330 | struct msghdr msg = { |
| 331 | .msg_name = (struct sockaddr *) &vubr->backend_udp_dest, |
| 332 | .msg_namelen = sizeof(struct sockaddr_in), |
| 333 | .msg_iov = sg, |
| 334 | .msg_iovlen = num, |
| 335 | .msg_flags = MSG_DONTWAIT, |
| 336 | }; |
| 337 | ret = RETRY_ON_EINTR(recvmsg(vubr->backend_udp_sock, &msg, 0)); |
| 338 | |
| 339 | if (ret > 0 && VHOST_USER_BRIDGE_DEBUG) { |
| 340 | iov_hexdump(sg, num, stderr, "RX:", ret); |
| 341 | } |
| 342 | |
| 343 | if (i == 0) { |
| 344 | iov_restore_front(elem->in_sg, sg, hdrlen); |
| 345 | } |
| 346 | |
| 347 | if (ret == -1) { |
| 348 | if (errno == EWOULDBLOCK) { |
| 349 | vu_queue_rewind(dev, vq, 1); |
| 350 | break; |
| 351 | } |
| 352 | |
| 353 | vubr_die("recvmsg()"); |
| 354 | } |
| 355 | |
| 356 | total += ret; |
| 357 | iov_truncate(elem->in_sg, elem->in_num, total); |
| 358 | vu_queue_fill(dev, vq, elem, total, i++); |
| 359 | |
| 360 | free(elem); |
| 361 | elem = NULL; |
| 362 | |
| 363 | break; /* could loop if DONTWAIT worked? */ |
| 364 | } |
| 365 | |
| 366 | if (mhdr_cnt) { |
| 367 | mhdr.num_buffers = i; |
| 368 | iov_from_buf(mhdr_sg, mhdr_cnt, |
| 369 | 0, |
| 370 | &mhdr.num_buffers, sizeof mhdr.num_buffers); |
| 371 | } |
| 372 | |
| 373 | vu_queue_flush(dev, vq, i); |
| 374 | vu_queue_notify(dev, vq); |
| 375 | |
| 376 | free(elem); |
| 377 | } |
| 378 | |
| 379 | static void |
| 380 | vubr_receive_cb(int sock, void *ctx) |
| 381 | { |
| 382 | VubrDev *vubr = (VubrDev *)ctx; |
| 383 | |
| 384 | if (!vu_dispatch(&vubr->vudev)) { |
| 385 | fprintf(stderr, "Error while dispatching\n"); |
| 386 | } |
| 387 | } |
| 388 | |
| 389 | typedef struct WatchData { |
| 390 | VuDev *dev; |
| 391 | vu_watch_cb cb; |
| 392 | void *data; |
| 393 | } WatchData; |
| 394 | |
| 395 | static void |
| 396 | watch_cb(int sock, void *ctx) |
| 397 | { |
| 398 | struct WatchData *wd = ctx; |
| 399 | |
| 400 | wd->cb(wd->dev, VU_WATCH_IN, wd->data); |
| 401 | } |
| 402 | |
| 403 | static void |
| 404 | vubr_set_watch(VuDev *dev, int fd, int condition, |
| 405 | vu_watch_cb cb, void *data) |
| 406 | { |
| 407 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
| 408 | static WatchData watches[FD_SETSIZE]; |
| 409 | struct WatchData *wd = &watches[fd]; |
| 410 | |
| 411 | wd->cb = cb; |
| 412 | wd->data = data; |
| 413 | wd->dev = dev; |
| 414 | dispatcher_add(&vubr->dispatcher, fd, wd, watch_cb); |
| 415 | } |
| 416 | |
| 417 | static void |
| 418 | vubr_remove_watch(VuDev *dev, int fd) |
| 419 | { |
| 420 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
| 421 | |
| 422 | dispatcher_remove(&vubr->dispatcher, fd); |
| 423 | } |
| 424 | |
| 425 | static int |
| 426 | vubr_send_rarp_exec(VuDev *dev, VhostUserMsg *vmsg) |
| 427 | { |
| 428 | DPRINT("Function %s() not implemented yet.\n", __func__); |
| 429 | return 0; |
| 430 | } |
| 431 | |
| 432 | static int |
| 433 | vubr_process_msg(VuDev *dev, VhostUserMsg *vmsg, int *do_reply) |
| 434 | { |
| 435 | switch (vmsg->request) { |
| 436 | case VHOST_USER_SEND_RARP: |
| 437 | *do_reply = vubr_send_rarp_exec(dev, vmsg); |
| 438 | return 1; |
| 439 | default: |
| 440 | /* let the library handle the rest */ |
| 441 | return 0; |
| 442 | } |
| 443 | |
| 444 | return 0; |
| 445 | } |
| 446 | |
| 447 | static void |
| 448 | vubr_set_features(VuDev *dev, uint64_t features) |
| 449 | { |
| 450 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
| 451 | |
| 452 | if ((features & (1ULL << VIRTIO_F_VERSION_1)) || |
| 453 | (features & (1ULL << VIRTIO_NET_F_MRG_RXBUF))) { |
| 454 | vubr->hdrlen = 12; |
| 455 | } else { |
| 456 | vubr->hdrlen = 10; |
| 457 | } |
| 458 | } |
| 459 | |
| 460 | static uint64_t |
| 461 | vubr_get_features(VuDev *dev) |
| 462 | { |
| 463 | return 1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE | |
| 464 | 1ULL << VIRTIO_NET_F_MRG_RXBUF | |
| 465 | 1ULL << VIRTIO_F_VERSION_1; |
| 466 | } |
| 467 | |
| 468 | static void |
| 469 | vubr_queue_set_started(VuDev *dev, int qidx, bool started) |
| 470 | { |
| 471 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
| 472 | VuVirtq *vq = vu_get_queue(dev, qidx); |
| 473 | |
| 474 | if (started && vubr->notifier.fd >= 0) { |
| 475 | vu_set_queue_host_notifier(dev, vq, vubr->notifier.fd, |
| 476 | qemu_real_host_page_size(), |
| 477 | qidx * qemu_real_host_page_size()); |
| 478 | } |
| 479 | |
| 480 | if (qidx % 2 == 1) { |
| 481 | vu_set_queue_handler(dev, vq, started ? vubr_handle_tx : NULL); |
| 482 | } |
| 483 | } |
| 484 | |
| 485 | static void |
| 486 | vubr_panic(VuDev *dev, const char *msg) |
| 487 | { |
| 488 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
| 489 | |
| 490 | fprintf(stderr, "PANIC: %s\n", msg); |
| 491 | |
| 492 | dispatcher_remove(&vubr->dispatcher, dev->sock); |
| 493 | vubr->quit = 1; |
| 494 | } |
| 495 | |
| 496 | static bool |
| 497 | vubr_queue_is_processed_in_order(VuDev *dev, int qidx) |
| 498 | { |
| 499 | return true; |
| 500 | } |
| 501 | |
| 502 | static const VuDevIface vuiface = { |
| 503 | .get_features = vubr_get_features, |
| 504 | .set_features = vubr_set_features, |
| 505 | .process_msg = vubr_process_msg, |
| 506 | .queue_set_started = vubr_queue_set_started, |
| 507 | .queue_is_processed_in_order = vubr_queue_is_processed_in_order, |
| 508 | }; |
| 509 | |
| 510 | static void |
| 511 | vubr_accept_cb(int sock, void *ctx) |
| 512 | { |
| 513 | VubrDev *dev = (VubrDev *)ctx; |
| 514 | int conn_fd; |
| 515 | struct sockaddr_un un; |
| 516 | socklen_t len = sizeof(un); |
| 517 | |
| 518 | conn_fd = accept(sock, (struct sockaddr *) &un, &len); |
| 519 | if (conn_fd == -1) { |
| 520 | vubr_die("accept()"); |
| 521 | } |
| 522 | DPRINT("Got connection from remote peer on sock %d\n", conn_fd); |
| 523 | |
| 524 | if (!vu_init(&dev->vudev, |
| 525 | VHOST_USER_BRIDGE_MAX_QUEUES, |
| 526 | conn_fd, |
| 527 | vubr_panic, |
| 528 | NULL, |
| 529 | vubr_set_watch, |
| 530 | vubr_remove_watch, |
| 531 | &vuiface)) { |
| 532 | fprintf(stderr, "Failed to initialize libvhost-user\n"); |
| 533 | exit(1); |
| 534 | } |
| 535 | |
| 536 | dispatcher_add(&dev->dispatcher, conn_fd, ctx, vubr_receive_cb); |
| 537 | dispatcher_remove(&dev->dispatcher, sock); |
| 538 | } |
| 539 | |
| 540 | static VubrDev * |
| 541 | vubr_new(const char *path, bool client) |
| 542 | { |
| 543 | VubrDev *dev = (VubrDev *) calloc(1, sizeof(VubrDev)); |
| 544 | struct sockaddr_un un; |
| 545 | CallbackFunc cb; |
| 546 | size_t len; |
| 547 | |
| 548 | if (strlen(path) >= sizeof(un.sun_path)) { |
| 549 | fprintf(stderr, "unix domain socket path '%s' is too long\n", path); |
| 550 | exit(1); |
| 551 | } |
| 552 | |
| 553 | /* Get a UNIX socket. */ |
| 554 | dev->sock = socket(AF_UNIX, SOCK_STREAM, 0); |
| 555 | if (dev->sock == -1) { |
| 556 | vubr_die("socket"); |
| 557 | } |
| 558 | |
| 559 | dev->notifier.fd = -1; |
| 560 | |
| 561 | un.sun_family = AF_UNIX; |
| 562 | strcpy(un.sun_path, path); |
| 563 | len = sizeof(un.sun_family) + strlen(path); |
| 564 | |
| 565 | if (!client) { |
| 566 | unlink(path); |
| 567 | |
| 568 | if (bind(dev->sock, (struct sockaddr *) &un, len) == -1) { |
| 569 | vubr_die("bind"); |
| 570 | } |
| 571 | |
| 572 | if (listen(dev->sock, 1) == -1) { |
| 573 | vubr_die("listen"); |
| 574 | } |
| 575 | cb = vubr_accept_cb; |
| 576 | |
| 577 | DPRINT("Waiting for connections on UNIX socket %s ...\n", path); |
| 578 | } else { |
| 579 | if (connect(dev->sock, (struct sockaddr *)&un, len) == -1) { |
| 580 | vubr_die("connect"); |
| 581 | } |
| 582 | |
| 583 | if (!vu_init(&dev->vudev, |
| 584 | VHOST_USER_BRIDGE_MAX_QUEUES, |
| 585 | dev->sock, |
| 586 | vubr_panic, |
| 587 | NULL, |
| 588 | vubr_set_watch, |
| 589 | vubr_remove_watch, |
| 590 | &vuiface)) { |
| 591 | fprintf(stderr, "Failed to initialize libvhost-user\n"); |
| 592 | exit(1); |
| 593 | } |
| 594 | |
| 595 | cb = vubr_receive_cb; |
| 596 | } |
| 597 | |
| 598 | dispatcher_init(&dev->dispatcher); |
| 599 | |
| 600 | dispatcher_add(&dev->dispatcher, dev->sock, (void *)dev, cb); |
| 601 | |
| 602 | return dev; |
| 603 | } |
| 604 | |
| 605 | static void *notifier_thread(void *arg) |
| 606 | { |
| 607 | VuDev *dev = (VuDev *)arg; |
| 608 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
| 609 | int pagesize = qemu_real_host_page_size(); |
| 610 | int qidx; |
| 611 | |
| 612 | while (true) { |
| 613 | for (qidx = 0; qidx < VHOST_USER_BRIDGE_MAX_QUEUES; qidx++) { |
| 614 | uint16_t *n = vubr->notifier.addr + pagesize * qidx; |
| 615 | |
| 616 | if (*n == qidx) { |
| 617 | *n = 0xffff; |
| 618 | /* We won't miss notifications if we reset |
| 619 | * the memory first. */ |
| 620 | smp_mb(); |
| 621 | |
| 622 | DPRINT("Got a notification for queue%d via host notifier.\n", |
| 623 | qidx); |
| 624 | |
| 625 | if (qidx % 2 == 1) { |
| 626 | vubr_handle_tx(dev, qidx); |
| 627 | } |
| 628 | } |
| 629 | usleep(1000); |
| 630 | } |
| 631 | } |
| 632 | |
| 633 | return NULL; |
| 634 | } |
| 635 | |
| 636 | static void |
| 637 | vubr_host_notifier_setup(VubrDev *dev) |
| 638 | { |
| 639 | pthread_t thread; |
| 640 | size_t length; |
| 641 | void *addr; |
| 642 | int fd; |
| 643 | |
| 644 | length = qemu_real_host_page_size() * VHOST_USER_BRIDGE_MAX_QUEUES; |
| 645 | |
| 646 | fd = g_file_open_tmp("vubr-XXXXXX", NULL, NULL); |
| 647 | if (fd < 0) { |
| 648 | vubr_die("mkstemp()"); |
| 649 | } |
| 650 | |
| 651 | if (posix_fallocate(fd, 0, length) != 0) { |
| 652 | vubr_die("posix_fallocate()"); |
| 653 | } |
| 654 | |
| 655 | addr = mmap(NULL, length, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); |
| 656 | if (addr == MAP_FAILED) { |
| 657 | vubr_die("mmap()"); |
| 658 | } |
| 659 | |
| 660 | memset(addr, 0xff, length); |
| 661 | |
| 662 | if (pthread_create(&thread, NULL, notifier_thread, &dev->vudev) != 0) { |
| 663 | vubr_die("pthread_create()"); |
| 664 | } |
| 665 | |
| 666 | dev->notifier.fd = fd; |
| 667 | dev->notifier.addr = addr; |
| 668 | dev->notifier.thread = thread; |
| 669 | } |
| 670 | |
| 671 | static void |
| 672 | vubr_set_host(struct sockaddr_in *saddr, const char *host) |
| 673 | { |
| 674 | if (qemu_isdigit(host[0])) { |
| 675 | if (!inet_aton(host, &saddr->sin_addr)) { |
| 676 | fprintf(stderr, "inet_aton() failed.\n"); |
| 677 | exit(1); |
| 678 | } |
| 679 | } else { |
| 680 | struct hostent *he = gethostbyname(host); |
| 681 | |
| 682 | if (!he) { |
| 683 | fprintf(stderr, "gethostbyname() failed.\n"); |
| 684 | exit(1); |
| 685 | } |
| 686 | saddr->sin_addr = *(struct in_addr *)he->h_addr; |
| 687 | } |
| 688 | } |
| 689 | |
| 690 | static void |
| 691 | vubr_backend_udp_setup(VubrDev *dev, |
| 692 | const char *local_host, |
| 693 | const char *local_port, |
| 694 | const char *remote_host, |
| 695 | const char *remote_port) |
| 696 | { |
| 697 | int sock; |
| 698 | const char *r; |
| 699 | |
| 700 | int lport, rport; |
| 701 | |
| 702 | lport = strtol(local_port, (char **)&r, 0); |
| 703 | if (r == local_port) { |
| 704 | fprintf(stderr, "lport parsing failed.\n"); |
| 705 | exit(1); |
| 706 | } |
| 707 | |
| 708 | rport = strtol(remote_port, (char **)&r, 0); |
| 709 | if (r == remote_port) { |
| 710 | fprintf(stderr, "rport parsing failed.\n"); |
| 711 | exit(1); |
| 712 | } |
| 713 | |
| 714 | struct sockaddr_in si_local = { |
| 715 | .sin_family = AF_INET, |
| 716 | .sin_port = htons(lport), |
| 717 | }; |
| 718 | |
| 719 | vubr_set_host(&si_local, local_host); |
| 720 | |
| 721 | /* setup destination for sends */ |
| 722 | dev->backend_udp_dest = (struct sockaddr_in) { |
| 723 | .sin_family = AF_INET, |
| 724 | .sin_port = htons(rport), |
| 725 | }; |
| 726 | vubr_set_host(&dev->backend_udp_dest, remote_host); |
| 727 | |
| 728 | sock = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP); |
| 729 | if (sock == -1) { |
| 730 | vubr_die("socket"); |
| 731 | } |
| 732 | |
| 733 | if (bind(sock, (struct sockaddr *)&si_local, sizeof(si_local)) == -1) { |
| 734 | vubr_die("bind"); |
| 735 | } |
| 736 | |
| 737 | dev->backend_udp_sock = sock; |
| 738 | dispatcher_add(&dev->dispatcher, sock, dev, vubr_backend_recv_cb); |
| 739 | DPRINT("Waiting for data from udp backend on %s:%d...\n", |
| 740 | local_host, lport); |
| 741 | } |
| 742 | |
| 743 | static void |
| 744 | vubr_run(VubrDev *dev) |
| 745 | { |
| 746 | while (!dev->quit) { |
| 747 | /* timeout 200ms */ |
| 748 | dispatcher_wait(&dev->dispatcher, 200000); |
| 749 | /* Here one can try polling strategy. */ |
| 750 | } |
| 751 | } |
| 752 | |
| 753 | static int |
| 754 | vubr_parse_host_port(const char **host, const char **port, const char *buf) |
| 755 | { |
| 756 | g_auto(GStrv) tokens = g_strsplit(buf, ":", 2); |
| 757 | if (!tokens[0] || !tokens[1]) { |
| 758 | return -1; |
| 759 | } |
| 760 | *host = g_steal_pointer(&tokens[0]); |
| 761 | *port = g_steal_pointer(&tokens[1]); |
| 762 | return 0; |
| 763 | } |
| 764 | |
| 765 | #define DEFAULT_UD_SOCKET "/tmp/vubr.sock" |
| 766 | #define DEFAULT_LHOST "127.0.0.1" |
| 767 | #define DEFAULT_LPORT "4444" |
| 768 | #define DEFAULT_RHOST "127.0.0.1" |
| 769 | #define DEFAULT_RPORT "5555" |
| 770 | |
| 771 | static const char *ud_socket_path = DEFAULT_UD_SOCKET; |
| 772 | static const char *lhost = DEFAULT_LHOST; |
| 773 | static const char *lport = DEFAULT_LPORT; |
| 774 | static const char *rhost = DEFAULT_RHOST; |
| 775 | static const char *rport = DEFAULT_RPORT; |
| 776 | |
| 777 | int |
| 778 | main(int argc, char *argv[]) |
| 779 | { |
| 780 | VubrDev *dev; |
| 781 | int opt; |
| 782 | bool client = false; |
| 783 | bool host_notifier = false; |
| 784 | |
| 785 | while ((opt = getopt(argc, argv, "l:r:u:cH")) != -1) { |
| 786 | |
| 787 | switch (opt) { |
| 788 | case 'l': |
| 789 | if (vubr_parse_host_port(&lhost, &lport, optarg) < 0) { |
| 790 | goto out; |
| 791 | } |
| 792 | break; |
| 793 | case 'r': |
| 794 | if (vubr_parse_host_port(&rhost, &rport, optarg) < 0) { |
| 795 | goto out; |
| 796 | } |
| 797 | break; |
| 798 | case 'u': |
| 799 | ud_socket_path = strdup(optarg); |
| 800 | break; |
| 801 | case 'c': |
| 802 | client = true; |
| 803 | break; |
| 804 | case 'H': |
| 805 | host_notifier = true; |
| 806 | break; |
| 807 | default: |
| 808 | goto out; |
| 809 | } |
| 810 | } |
| 811 | |
| 812 | DPRINT("ud socket: %s (%s)\n", ud_socket_path, |
| 813 | client ? "client" : "server"); |
| 814 | DPRINT("local: %s:%s\n", lhost, lport); |
| 815 | DPRINT("remote: %s:%s\n", rhost, rport); |
| 816 | |
| 817 | dev = vubr_new(ud_socket_path, client); |
| 818 | if (!dev) { |
| 819 | return 1; |
| 820 | } |
| 821 | |
| 822 | if (host_notifier) { |
| 823 | vubr_host_notifier_setup(dev); |
| 824 | } |
| 825 | |
| 826 | vubr_backend_udp_setup(dev, lhost, lport, rhost, rport); |
| 827 | vubr_run(dev); |
| 828 | |
| 829 | vu_deinit(&dev->vudev); |
| 830 | |
| 831 | return 0; |
| 832 | |
| 833 | out: |
| 834 | fprintf(stderr, "Usage: %s ", argv[0]); |
| 835 | fprintf(stderr, "[-c] [-H] [-u ud_socket_path] [-l lhost:lport] [-r rhost:rport]\n"); |
| 836 | fprintf(stderr, "\t-u path to unix domain socket. default: %s\n", |
| 837 | DEFAULT_UD_SOCKET); |
| 838 | fprintf(stderr, "\t-l local host and port. default: %s:%s\n", |
| 839 | DEFAULT_LHOST, DEFAULT_LPORT); |
| 840 | fprintf(stderr, "\t-r remote host and port. default: %s:%s\n", |
| 841 | DEFAULT_RHOST, DEFAULT_RPORT); |
| 842 | fprintf(stderr, "\t-c client mode\n"); |
| 843 | fprintf(stderr, "\t-H use host notifier\n"); |
| 844 | |
| 845 | return 1; |
| 846 | } |