qemu-doc: include version number
[qemu/ar7.git] / tests / vhost-user-bridge.c
blob1e5b5ca3da62741d1d5475352e43a305aab6107f
1 /*
2 * Vhost User Bridge
4 * Copyright (c) 2015 Red Hat, Inc.
6 * Authors:
7 * Victor Kaplansky <victork@redhat.com>
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.
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.
30 #define _FILE_OFFSET_BITS 64
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"
37 #define VHOST_USER_BRIDGE_DEBUG 1
39 #define DPRINT(...) \
40 do { \
41 if (VHOST_USER_BRIDGE_DEBUG) { \
42 printf(__VA_ARGS__); \
43 } \
44 } while (0)
46 typedef void (*CallbackFunc)(int sock, void *ctx);
48 typedef struct Event {
49 void *ctx;
50 CallbackFunc callback;
51 } Event;
53 typedef struct Dispatcher {
54 int max_sock;
55 fd_set fdset;
56 Event events[FD_SETSIZE];
57 } Dispatcher;
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;
70 static void
71 vubr_die(const char *s)
73 perror(s);
74 exit(1);
77 static int
78 dispatcher_init(Dispatcher *dispr)
80 FD_ZERO(&dispr->fdset);
81 dispr->max_sock = -1;
82 return 0;
85 static int
86 dispatcher_add(Dispatcher *dispr, int sock, void *ctx, CallbackFunc cb)
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;
95 dispr->events[sock].ctx = ctx;
96 dispr->events[sock].callback = cb;
98 FD_SET(sock, &dispr->fdset);
99 if (sock > dispr->max_sock) {
100 dispr->max_sock = sock;
102 DPRINT("Added sock %d for watching. max_sock: %d\n",
103 sock, dispr->max_sock);
104 return 0;
107 static int
108 dispatcher_remove(Dispatcher *dispr, int sock)
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;
117 FD_CLR(sock, &dispr->fdset);
118 DPRINT("Sock %d removed from dispatcher watch.\n", sock);
119 return 0;
122 /* timeout in us */
123 static int
124 dispatcher_wait(Dispatcher *dispr, uint32_t timeout)
126 struct timeval tv;
127 tv.tv_sec = timeout / 1000000;
128 tv.tv_usec = timeout % 1000000;
130 fd_set fdset = dispr->fdset;
132 /* wait until some of sockets become readable. */
133 int rc = select(dispr->max_sock + 1, &fdset, 0, 0, &tv);
135 if (rc == -1) {
136 vubr_die("select");
139 /* Timeout */
140 if (rc == 0) {
141 return 0;
144 /* Now call callback for every ready socket. */
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
152 if (FD_ISSET(sock, &fdset) && FD_ISSET(sock, &dispr->fdset)) {
153 Event *e = &dispr->events[sock];
154 e->callback(sock, e->ctx);
158 return 0;
161 static void
162 vubr_handle_tx(VuDev *dev, int qidx)
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;
169 assert(qidx % 2);
171 for (;;) {
172 ssize_t ret;
173 unsigned int out_num;
174 struct iovec sg[VIRTQUEUE_MAX_SIZE], *out_sg;
176 elem = vu_queue_pop(dev, vq, sizeof(VuVirtqElement));
177 if (!elem) {
178 break;
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;
187 if (VHOST_USER_BRIDGE_DEBUG) {
188 iov_hexdump(out_sg, out_num, stderr, "TX:", 1024);
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;
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,
205 do {
206 ret = sendmsg(vubr->backend_udp_sock, &msg, 0);
207 } while (ret == -1 && (errno == EAGAIN || errno == EINTR));
209 if (ret == -1) {
210 vubr_die("sendmsg()");
213 vu_queue_push(dev, vq, elem, 0);
214 vu_queue_notify(dev, vq);
216 free(elem);
217 elem = NULL;
220 free(elem);
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
228 static void
229 iov_restore_front(struct iovec *front, struct iovec *iov, size_t bytes)
231 struct iovec *cur;
233 for (cur = front; cur != iov; cur++) {
234 assert(bytes >= cur->iov_len);
235 bytes -= cur->iov_len;
238 cur->iov_base -= bytes;
239 cur->iov_len += bytes;
242 static void
243 iov_truncate(struct iovec *iov, unsigned iovc, size_t bytes)
245 unsigned i;
247 for (i = 0; i < iovc; i++, iov++) {
248 if (bytes < iov->iov_len) {
249 iov->iov_len = bytes;
250 return;
253 bytes -= iov->iov_len;
256 assert(!"couldn't truncate iov");
259 static void
260 vubr_backend_recv_cb(int sock, void *ctx)
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
276 DPRINT("\n\n *** IN UDP RECEIVE CALLBACK ***\n\n");
277 DPRINT(" hdrlen = %d\n", hdrlen);
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;
285 do {
286 struct iovec *sg;
287 ssize_t ret, total = 0;
288 unsigned int num;
290 elem = vu_queue_pop(dev, vq, sizeof(VuVirtqElement));
291 if (!elem) {
292 break;
295 if (elem->in_num < 1) {
296 fprintf(stderr, "virtio-net contains no in buffers\n");
297 break;
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));
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);
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,
322 do {
323 ret = recvmsg(vubr->backend_udp_sock, &msg, 0);
324 } while (ret == -1 && (errno == EINTR));
326 if (i == 0) {
327 iov_restore_front(elem->in_sg, sg, hdrlen);
330 if (ret == -1) {
331 if (errno == EWOULDBLOCK) {
332 vu_queue_rewind(dev, vq, 1);
333 break;
336 vubr_die("recvmsg()");
339 total += ret;
340 iov_truncate(elem->in_sg, elem->in_num, total);
341 vu_queue_fill(dev, vq, elem, total, i++);
343 free(elem);
344 elem = NULL;
345 } while (false); /* could loop if DONTWAIT worked? */
347 if (mhdr_cnt) {
348 mhdr.num_buffers = i;
349 iov_from_buf(mhdr_sg, mhdr_cnt,
351 &mhdr.num_buffers, sizeof mhdr.num_buffers);
354 vu_queue_flush(dev, vq, i);
355 vu_queue_notify(dev, vq);
357 free(elem);
360 static void
361 vubr_receive_cb(int sock, void *ctx)
363 VubrDev *vubr = (VubrDev *)ctx;
365 if (!vu_dispatch(&vubr->vudev)) {
366 fprintf(stderr, "Error while dispatching\n");
370 typedef struct WatchData {
371 VuDev *dev;
372 vu_watch_cb cb;
373 void *data;
374 } WatchData;
376 static void
377 watch_cb(int sock, void *ctx)
379 struct WatchData *wd = ctx;
381 wd->cb(wd->dev, VU_WATCH_IN, wd->data);
384 static void
385 vubr_set_watch(VuDev *dev, int fd, int condition,
386 vu_watch_cb cb, void *data)
388 VubrDev *vubr = container_of(dev, VubrDev, vudev);
389 static WatchData watches[FD_SETSIZE];
390 struct WatchData *wd = &watches[fd];
392 wd->cb = cb;
393 wd->data = data;
394 wd->dev = dev;
395 dispatcher_add(&vubr->dispatcher, fd, wd, watch_cb);
398 static void
399 vubr_remove_watch(VuDev *dev, int fd)
401 VubrDev *vubr = container_of(dev, VubrDev, vudev);
403 dispatcher_remove(&vubr->dispatcher, fd);
406 static int
407 vubr_send_rarp_exec(VuDev *dev, VhostUserMsg *vmsg)
409 DPRINT("Function %s() not implemented yet.\n", __func__);
410 return 0;
413 static int
414 vubr_process_msg(VuDev *dev, VhostUserMsg *vmsg, int *do_reply)
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;
425 return 0;
428 static void
429 vubr_set_features(VuDev *dev, uint64_t features)
431 VubrDev *vubr = container_of(dev, VubrDev, vudev);
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;
441 static uint64_t
442 vubr_get_features(VuDev *dev)
444 return 1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE |
445 1ULL << VIRTIO_NET_F_MRG_RXBUF;
448 static void
449 vubr_queue_set_started(VuDev *dev, int qidx, bool started)
451 VuVirtq *vq = vu_get_queue(dev, qidx);
453 if (qidx % 2 == 1) {
454 vu_set_queue_handler(dev, vq, started ? vubr_handle_tx : NULL);
458 static void
459 vubr_panic(VuDev *dev, const char *msg)
461 VubrDev *vubr = container_of(dev, VubrDev, vudev);
463 fprintf(stderr, "PANIC: %s\n", msg);
465 dispatcher_remove(&vubr->dispatcher, dev->sock);
466 vubr->quit = 1;
469 static const VuDevIface vuiface = {
470 .get_features = vubr_get_features,
471 .set_features = vubr_set_features,
472 .process_msg = vubr_process_msg,
473 .queue_set_started = vubr_queue_set_started,
476 static void
477 vubr_accept_cb(int sock, void *ctx)
479 VubrDev *dev = (VubrDev *)ctx;
480 int conn_fd;
481 struct sockaddr_un un;
482 socklen_t len = sizeof(un);
484 conn_fd = accept(sock, (struct sockaddr *) &un, &len);
485 if (conn_fd == -1) {
486 vubr_die("accept()");
488 DPRINT("Got connection from remote peer on sock %d\n", conn_fd);
490 vu_init(&dev->vudev,
491 conn_fd,
492 vubr_panic,
493 vubr_set_watch,
494 vubr_remove_watch,
495 &vuiface);
497 dispatcher_add(&dev->dispatcher, conn_fd, ctx, vubr_receive_cb);
498 dispatcher_remove(&dev->dispatcher, sock);
501 static VubrDev *
502 vubr_new(const char *path, bool client)
504 VubrDev *dev = (VubrDev *) calloc(1, sizeof(VubrDev));
505 struct sockaddr_un un;
506 CallbackFunc cb;
507 size_t len;
509 /* Get a UNIX socket. */
510 dev->sock = socket(AF_UNIX, SOCK_STREAM, 0);
511 if (dev->sock == -1) {
512 vubr_die("socket");
515 un.sun_family = AF_UNIX;
516 strcpy(un.sun_path, path);
517 len = sizeof(un.sun_family) + strlen(path);
519 if (!client) {
520 unlink(path);
522 if (bind(dev->sock, (struct sockaddr *) &un, len) == -1) {
523 vubr_die("bind");
526 if (listen(dev->sock, 1) == -1) {
527 vubr_die("listen");
529 cb = vubr_accept_cb;
531 DPRINT("Waiting for connections on UNIX socket %s ...\n", path);
532 } else {
533 if (connect(dev->sock, (struct sockaddr *)&un, len) == -1) {
534 vubr_die("connect");
536 vu_init(&dev->vudev,
537 dev->sock,
538 vubr_panic,
539 vubr_set_watch,
540 vubr_remove_watch,
541 &vuiface);
542 cb = vubr_receive_cb;
545 dispatcher_init(&dev->dispatcher);
547 dispatcher_add(&dev->dispatcher, dev->sock, (void *)dev, cb);
549 return dev;
552 static void
553 vubr_set_host(struct sockaddr_in *saddr, const char *host)
555 if (isdigit(host[0])) {
556 if (!inet_aton(host, &saddr->sin_addr)) {
557 fprintf(stderr, "inet_aton() failed.\n");
558 exit(1);
560 } else {
561 struct hostent *he = gethostbyname(host);
563 if (!he) {
564 fprintf(stderr, "gethostbyname() failed.\n");
565 exit(1);
567 saddr->sin_addr = *(struct in_addr *)he->h_addr;
571 static void
572 vubr_backend_udp_setup(VubrDev *dev,
573 const char *local_host,
574 const char *local_port,
575 const char *remote_host,
576 const char *remote_port)
578 int sock;
579 const char *r;
581 int lport, rport;
583 lport = strtol(local_port, (char **)&r, 0);
584 if (r == local_port) {
585 fprintf(stderr, "lport parsing failed.\n");
586 exit(1);
589 rport = strtol(remote_port, (char **)&r, 0);
590 if (r == remote_port) {
591 fprintf(stderr, "rport parsing failed.\n");
592 exit(1);
595 struct sockaddr_in si_local = {
596 .sin_family = AF_INET,
597 .sin_port = htons(lport),
600 vubr_set_host(&si_local, local_host);
602 /* setup destination for sends */
603 dev->backend_udp_dest = (struct sockaddr_in) {
604 .sin_family = AF_INET,
605 .sin_port = htons(rport),
607 vubr_set_host(&dev->backend_udp_dest, remote_host);
609 sock = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
610 if (sock == -1) {
611 vubr_die("socket");
614 if (bind(sock, (struct sockaddr *)&si_local, sizeof(si_local)) == -1) {
615 vubr_die("bind");
618 dev->backend_udp_sock = sock;
619 dispatcher_add(&dev->dispatcher, sock, dev, vubr_backend_recv_cb);
620 DPRINT("Waiting for data from udp backend on %s:%d...\n",
621 local_host, lport);
624 static void
625 vubr_run(VubrDev *dev)
627 while (!dev->quit) {
628 /* timeout 200ms */
629 dispatcher_wait(&dev->dispatcher, 200000);
630 /* Here one can try polling strategy. */
634 static int
635 vubr_parse_host_port(const char **host, const char **port, const char *buf)
637 char *p = strchr(buf, ':');
639 if (!p) {
640 return -1;
642 *p = '\0';
643 *host = strdup(buf);
644 *port = strdup(p + 1);
645 return 0;
648 #define DEFAULT_UD_SOCKET "/tmp/vubr.sock"
649 #define DEFAULT_LHOST "127.0.0.1"
650 #define DEFAULT_LPORT "4444"
651 #define DEFAULT_RHOST "127.0.0.1"
652 #define DEFAULT_RPORT "5555"
654 static const char *ud_socket_path = DEFAULT_UD_SOCKET;
655 static const char *lhost = DEFAULT_LHOST;
656 static const char *lport = DEFAULT_LPORT;
657 static const char *rhost = DEFAULT_RHOST;
658 static const char *rport = DEFAULT_RPORT;
661 main(int argc, char *argv[])
663 VubrDev *dev;
664 int opt;
665 bool client = false;
667 while ((opt = getopt(argc, argv, "l:r:u:c")) != -1) {
669 switch (opt) {
670 case 'l':
671 if (vubr_parse_host_port(&lhost, &lport, optarg) < 0) {
672 goto out;
674 break;
675 case 'r':
676 if (vubr_parse_host_port(&rhost, &rport, optarg) < 0) {
677 goto out;
679 break;
680 case 'u':
681 ud_socket_path = strdup(optarg);
682 break;
683 case 'c':
684 client = true;
685 break;
686 default:
687 goto out;
691 DPRINT("ud socket: %s (%s)\n", ud_socket_path,
692 client ? "client" : "server");
693 DPRINT("local: %s:%s\n", lhost, lport);
694 DPRINT("remote: %s:%s\n", rhost, rport);
696 dev = vubr_new(ud_socket_path, client);
697 if (!dev) {
698 return 1;
701 vubr_backend_udp_setup(dev, lhost, lport, rhost, rport);
702 vubr_run(dev);
704 vu_deinit(&dev->vudev);
706 return 0;
708 out:
709 fprintf(stderr, "Usage: %s ", argv[0]);
710 fprintf(stderr, "[-c] [-u ud_socket_path] [-l lhost:lport] [-r rhost:rport]\n");
711 fprintf(stderr, "\t-u path to unix doman socket. default: %s\n",
712 DEFAULT_UD_SOCKET);
713 fprintf(stderr, "\t-l local host and port. default: %s:%s\n",
714 DEFAULT_LHOST, DEFAULT_LPORT);
715 fprintf(stderr, "\t-r remote host and port. default: %s:%s\n",
716 DEFAULT_RHOST, DEFAULT_RPORT);
717 fprintf(stderr, "\t-c client mode\n");
719 return 1;