]>
Commit | Line | Data |
---|---|---|
3595e2eb VK |
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. | |
5c93c473 VK |
16 | * - implement all request handlers. Still not implemented: |
17 | * vubr_get_queue_num_exec() | |
18 | * vubr_send_rarp_exec() | |
3595e2eb VK |
19 | * - test for broken requests and virtqueue. |
20 | * - implement features defined by Virtio 1.0 spec. | |
21 | * - support mergeable buffers and indirect descriptors. | |
3595e2eb VK |
22 | * - implement clean shutdown. |
23 | * - implement non-blocking writes to UDP backend. | |
24 | * - implement polling strategy. | |
5c93c473 VK |
25 | * - implement clean starting/stopping of vq processing |
26 | * - implement clean starting/stopping of used and buffers | |
27 | * dirty page logging. | |
3595e2eb VK |
28 | */ |
29 | ||
5c93c473 VK |
30 | #define _FILE_OFFSET_BITS 64 |
31 | ||
681c28a3 | 32 | #include "qemu/osdep.h" |
d18dc3af | 33 | #include "qemu-common.h" |
b7d89466 | 34 | #include "qemu/atomic.h" |
e10e798c | 35 | #include "qemu/iov.h" |
3595e2eb | 36 | #include "standard-headers/linux/virtio_net.h" |
e10e798c | 37 | #include "contrib/libvhost-user/libvhost-user.h" |
3595e2eb VK |
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 | typedef void (*CallbackFunc)(int sock, void *ctx); | |
49 | ||
50 | typedef struct Event { | |
51 | void *ctx; | |
52 | CallbackFunc callback; | |
53 | } Event; | |
54 | ||
55 | typedef struct Dispatcher { | |
56 | int max_sock; | |
57 | fd_set fdset; | |
58 | Event events[FD_SETSIZE]; | |
59 | } Dispatcher; | |
60 | ||
e10e798c MAL |
61 | typedef struct VubrDev { |
62 | VuDev vudev; | |
63 | Dispatcher dispatcher; | |
64 | int backend_udp_sock; | |
65 | struct sockaddr_in backend_udp_dest; | |
66 | int hdrlen; | |
67 | int sock; | |
68 | int ready; | |
69 | int quit; | |
e3af2928 TB |
70 | struct { |
71 | int fd; | |
72 | void *addr; | |
73 | pthread_t thread; | |
74 | } notifier; | |
e10e798c MAL |
75 | } VubrDev; |
76 | ||
3595e2eb VK |
77 | static void |
78 | vubr_die(const char *s) | |
79 | { | |
80 | perror(s); | |
81 | exit(1); | |
82 | } | |
83 | ||
84 | static int | |
85 | dispatcher_init(Dispatcher *dispr) | |
86 | { | |
87 | FD_ZERO(&dispr->fdset); | |
88 | dispr->max_sock = -1; | |
89 | return 0; | |
90 | } | |
91 | ||
92 | static int | |
93 | dispatcher_add(Dispatcher *dispr, int sock, void *ctx, CallbackFunc cb) | |
94 | { | |
95 | if (sock >= FD_SETSIZE) { | |
96 | fprintf(stderr, | |
97 | "Error: Failed to add new event. sock %d should be less than %d\n", | |
98 | sock, FD_SETSIZE); | |
99 | return -1; | |
100 | } | |
101 | ||
102 | dispr->events[sock].ctx = ctx; | |
103 | dispr->events[sock].callback = cb; | |
104 | ||
105 | FD_SET(sock, &dispr->fdset); | |
106 | if (sock > dispr->max_sock) { | |
107 | dispr->max_sock = sock; | |
108 | } | |
109 | DPRINT("Added sock %d for watching. max_sock: %d\n", | |
110 | sock, dispr->max_sock); | |
111 | return 0; | |
112 | } | |
113 | ||
3595e2eb VK |
114 | static int |
115 | dispatcher_remove(Dispatcher *dispr, int sock) | |
116 | { | |
117 | if (sock >= FD_SETSIZE) { | |
118 | fprintf(stderr, | |
119 | "Error: Failed to remove event. sock %d should be less than %d\n", | |
120 | sock, FD_SETSIZE); | |
121 | return -1; | |
122 | } | |
123 | ||
124 | FD_CLR(sock, &dispr->fdset); | |
6d0b908a | 125 | DPRINT("Sock %d removed from dispatcher watch.\n", sock); |
3595e2eb VK |
126 | return 0; |
127 | } | |
3595e2eb VK |
128 | |
129 | /* timeout in us */ | |
130 | static int | |
131 | dispatcher_wait(Dispatcher *dispr, uint32_t timeout) | |
132 | { | |
133 | struct timeval tv; | |
134 | tv.tv_sec = timeout / 1000000; | |
135 | tv.tv_usec = timeout % 1000000; | |
136 | ||
137 | fd_set fdset = dispr->fdset; | |
138 | ||
139 | /* wait until some of sockets become readable. */ | |
140 | int rc = select(dispr->max_sock + 1, &fdset, 0, 0, &tv); | |
141 | ||
142 | if (rc == -1) { | |
143 | vubr_die("select"); | |
144 | } | |
145 | ||
146 | /* Timeout */ | |
147 | if (rc == 0) { | |
148 | return 0; | |
149 | } | |
150 | ||
151 | /* Now call callback for every ready socket. */ | |
152 | ||
153 | int sock; | |
6d0b908a VK |
154 | for (sock = 0; sock < dispr->max_sock + 1; sock++) { |
155 | /* The callback on a socket can remove other sockets from the | |
156 | * dispatcher, thus we have to check that the socket is | |
157 | * still not removed from dispatcher's list | |
158 | */ | |
159 | if (FD_ISSET(sock, &fdset) && FD_ISSET(sock, &dispr->fdset)) { | |
3595e2eb VK |
160 | Event *e = &dispr->events[sock]; |
161 | e->callback(sock, e->ctx); | |
162 | } | |
6d0b908a | 163 | } |
3595e2eb VK |
164 | |
165 | return 0; | |
166 | } | |
167 | ||
3595e2eb | 168 | static void |
e10e798c | 169 | vubr_handle_tx(VuDev *dev, int qidx) |
3595e2eb | 170 | { |
e10e798c MAL |
171 | VuVirtq *vq = vu_get_queue(dev, qidx); |
172 | VubrDev *vubr = container_of(dev, VubrDev, vudev); | |
173 | int hdrlen = vubr->hdrlen; | |
174 | VuVirtqElement *elem = NULL; | |
3595e2eb | 175 | |
e10e798c | 176 | assert(qidx % 2); |
3595e2eb | 177 | |
e10e798c MAL |
178 | for (;;) { |
179 | ssize_t ret; | |
180 | unsigned int out_num; | |
181 | struct iovec sg[VIRTQUEUE_MAX_SIZE], *out_sg; | |
3595e2eb | 182 | |
e10e798c MAL |
183 | elem = vu_queue_pop(dev, vq, sizeof(VuVirtqElement)); |
184 | if (!elem) { | |
185 | break; | |
3595e2eb | 186 | } |
3595e2eb | 187 | |
e10e798c MAL |
188 | out_num = elem->out_num; |
189 | out_sg = elem->out_sg; | |
190 | if (out_num < 1) { | |
191 | fprintf(stderr, "virtio-net header not in first element\n"); | |
3595e2eb VK |
192 | break; |
193 | } | |
e10e798c MAL |
194 | if (VHOST_USER_BRIDGE_DEBUG) { |
195 | iov_hexdump(out_sg, out_num, stderr, "TX:", 1024); | |
5c93c473 | 196 | } |
e10e798c MAL |
197 | |
198 | if (hdrlen) { | |
199 | unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), | |
200 | out_sg, out_num, | |
201 | hdrlen, -1); | |
202 | out_num = sg_num; | |
203 | out_sg = sg; | |
3595e2eb VK |
204 | } |
205 | ||
e10e798c MAL |
206 | struct msghdr msg = { |
207 | .msg_name = (struct sockaddr *) &vubr->backend_udp_dest, | |
208 | .msg_namelen = sizeof(struct sockaddr_in), | |
209 | .msg_iov = out_sg, | |
210 | .msg_iovlen = out_num, | |
211 | }; | |
212 | do { | |
213 | ret = sendmsg(vubr->backend_udp_sock, &msg, 0); | |
214 | } while (ret == -1 && (errno == EAGAIN || errno == EINTR)); | |
3595e2eb | 215 | |
e10e798c MAL |
216 | if (ret == -1) { |
217 | vubr_die("sendmsg()"); | |
218 | } | |
3595e2eb | 219 | |
e10e798c MAL |
220 | vu_queue_push(dev, vq, elem, 0); |
221 | vu_queue_notify(dev, vq); | |
3595e2eb | 222 | |
e10e798c MAL |
223 | free(elem); |
224 | elem = NULL; | |
3595e2eb | 225 | } |
3595e2eb | 226 | |
e10e798c | 227 | free(elem); |
3595e2eb VK |
228 | } |
229 | ||
277238f9 MAL |
230 | |
231 | /* this function reverse the effect of iov_discard_front() it must be | |
232 | * called with 'front' being the original struct iovec and 'bytes' | |
233 | * being the number of bytes you shaved off | |
234 | */ | |
e10e798c MAL |
235 | static void |
236 | iov_restore_front(struct iovec *front, struct iovec *iov, size_t bytes) | |
3595e2eb | 237 | { |
e10e798c | 238 | struct iovec *cur; |
3595e2eb | 239 | |
277238f9 MAL |
240 | for (cur = front; cur != iov; cur++) { |
241 | assert(bytes >= cur->iov_len); | |
e10e798c | 242 | bytes -= cur->iov_len; |
3595e2eb VK |
243 | } |
244 | ||
e10e798c MAL |
245 | cur->iov_base -= bytes; |
246 | cur->iov_len += bytes; | |
3595e2eb VK |
247 | } |
248 | ||
249 | static void | |
e10e798c | 250 | iov_truncate(struct iovec *iov, unsigned iovc, size_t bytes) |
3595e2eb | 251 | { |
e10e798c | 252 | unsigned i; |
3595e2eb | 253 | |
e10e798c MAL |
254 | for (i = 0; i < iovc; i++, iov++) { |
255 | if (bytes < iov->iov_len) { | |
256 | iov->iov_len = bytes; | |
257 | return; | |
258 | } | |
3595e2eb | 259 | |
e10e798c | 260 | bytes -= iov->iov_len; |
5c93c473 | 261 | } |
5c93c473 | 262 | |
e10e798c | 263 | assert(!"couldn't truncate iov"); |
3595e2eb VK |
264 | } |
265 | ||
5c93c473 | 266 | static void |
e10e798c | 267 | vubr_backend_recv_cb(int sock, void *ctx) |
5c93c473 | 268 | { |
e10e798c MAL |
269 | VubrDev *vubr = (VubrDev *) ctx; |
270 | VuDev *dev = &vubr->vudev; | |
271 | VuVirtq *vq = vu_get_queue(dev, 0); | |
272 | VuVirtqElement *elem = NULL; | |
273 | struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; | |
274 | struct virtio_net_hdr_mrg_rxbuf mhdr; | |
275 | unsigned mhdr_cnt = 0; | |
276 | int hdrlen = vubr->hdrlen; | |
277 | int i = 0; | |
278 | struct virtio_net_hdr hdr = { | |
279 | .flags = 0, | |
280 | .gso_type = VIRTIO_NET_HDR_GSO_NONE | |
281 | }; | |
5c93c473 | 282 | |
e10e798c MAL |
283 | DPRINT("\n\n *** IN UDP RECEIVE CALLBACK ***\n\n"); |
284 | DPRINT(" hdrlen = %d\n", hdrlen); | |
5c93c473 | 285 | |
e10e798c | 286 | if (!vu_queue_enabled(dev, vq) || |
12176528 | 287 | !vu_queue_started(dev, vq) || |
e10e798c MAL |
288 | !vu_queue_avail_bytes(dev, vq, hdrlen, 0)) { |
289 | DPRINT("Got UDP packet, but no available descriptors on RX virtq.\n"); | |
5c93c473 VK |
290 | return; |
291 | } | |
292 | ||
241187c1 | 293 | while (1) { |
e10e798c MAL |
294 | struct iovec *sg; |
295 | ssize_t ret, total = 0; | |
296 | unsigned int num; | |
3595e2eb | 297 | |
e10e798c MAL |
298 | elem = vu_queue_pop(dev, vq, sizeof(VuVirtqElement)); |
299 | if (!elem) { | |
a28c393c VK |
300 | break; |
301 | } | |
302 | ||
e10e798c MAL |
303 | if (elem->in_num < 1) { |
304 | fprintf(stderr, "virtio-net contains no in buffers\n"); | |
3595e2eb VK |
305 | break; |
306 | } | |
307 | ||
e10e798c MAL |
308 | sg = elem->in_sg; |
309 | num = elem->in_num; | |
310 | if (i == 0) { | |
311 | if (hdrlen == 12) { | |
312 | mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), | |
313 | sg, elem->in_num, | |
314 | offsetof(typeof(mhdr), num_buffers), | |
315 | sizeof(mhdr.num_buffers)); | |
316 | } | |
317 | iov_from_buf(sg, elem->in_num, 0, &hdr, sizeof hdr); | |
318 | total += hdrlen; | |
277238f9 MAL |
319 | ret = iov_discard_front(&sg, &num, hdrlen); |
320 | assert(ret == hdrlen); | |
3595e2eb VK |
321 | } |
322 | ||
e10e798c MAL |
323 | struct msghdr msg = { |
324 | .msg_name = (struct sockaddr *) &vubr->backend_udp_dest, | |
325 | .msg_namelen = sizeof(struct sockaddr_in), | |
326 | .msg_iov = sg, | |
8f1d22d9 | 327 | .msg_iovlen = num, |
e10e798c MAL |
328 | .msg_flags = MSG_DONTWAIT, |
329 | }; | |
330 | do { | |
331 | ret = recvmsg(vubr->backend_udp_sock, &msg, 0); | |
332 | } while (ret == -1 && (errno == EINTR)); | |
3595e2eb | 333 | |
e10e798c MAL |
334 | if (i == 0) { |
335 | iov_restore_front(elem->in_sg, sg, hdrlen); | |
336 | } | |
3595e2eb | 337 | |
e10e798c MAL |
338 | if (ret == -1) { |
339 | if (errno == EWOULDBLOCK) { | |
340 | vu_queue_rewind(dev, vq, 1); | |
341 | break; | |
342 | } | |
3595e2eb | 343 | |
e10e798c MAL |
344 | vubr_die("recvmsg()"); |
345 | } | |
5c93c473 | 346 | |
e10e798c MAL |
347 | total += ret; |
348 | iov_truncate(elem->in_sg, elem->in_num, total); | |
349 | vu_queue_fill(dev, vq, elem, total, i++); | |
3595e2eb | 350 | |
e10e798c MAL |
351 | free(elem); |
352 | elem = NULL; | |
241187c1 EB |
353 | |
354 | break; /* could loop if DONTWAIT worked? */ | |
355 | } | |
3595e2eb | 356 | |
e10e798c MAL |
357 | if (mhdr_cnt) { |
358 | mhdr.num_buffers = i; | |
359 | iov_from_buf(mhdr_sg, mhdr_cnt, | |
360 | 0, | |
361 | &mhdr.num_buffers, sizeof mhdr.num_buffers); | |
3595e2eb VK |
362 | } |
363 | ||
e10e798c MAL |
364 | vu_queue_flush(dev, vq, i); |
365 | vu_queue_notify(dev, vq); | |
3595e2eb | 366 | |
e10e798c | 367 | free(elem); |
3595e2eb VK |
368 | } |
369 | ||
370 | static void | |
e10e798c | 371 | vubr_receive_cb(int sock, void *ctx) |
3595e2eb | 372 | { |
e10e798c | 373 | VubrDev *vubr = (VubrDev *)ctx; |
3595e2eb | 374 | |
e10e798c MAL |
375 | if (!vu_dispatch(&vubr->vudev)) { |
376 | fprintf(stderr, "Error while dispatching\n"); | |
3595e2eb VK |
377 | } |
378 | } | |
379 | ||
e10e798c MAL |
380 | typedef struct WatchData { |
381 | VuDev *dev; | |
382 | vu_watch_cb cb; | |
383 | void *data; | |
384 | } WatchData; | |
3595e2eb | 385 | |
e10e798c MAL |
386 | static void |
387 | watch_cb(int sock, void *ctx) | |
3595e2eb | 388 | { |
e10e798c | 389 | struct WatchData *wd = ctx; |
3595e2eb | 390 | |
e10e798c | 391 | wd->cb(wd->dev, VU_WATCH_IN, wd->data); |
3595e2eb VK |
392 | } |
393 | ||
e10e798c MAL |
394 | static void |
395 | vubr_set_watch(VuDev *dev, int fd, int condition, | |
396 | vu_watch_cb cb, void *data) | |
3595e2eb | 397 | { |
e10e798c MAL |
398 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
399 | static WatchData watches[FD_SETSIZE]; | |
400 | struct WatchData *wd = &watches[fd]; | |
a28c393c | 401 | |
e10e798c MAL |
402 | wd->cb = cb; |
403 | wd->data = data; | |
404 | wd->dev = dev; | |
405 | dispatcher_add(&vubr->dispatcher, fd, wd, watch_cb); | |
3595e2eb VK |
406 | } |
407 | ||
5c93c473 | 408 | static void |
e10e798c | 409 | vubr_remove_watch(VuDev *dev, int fd) |
5c93c473 | 410 | { |
e10e798c | 411 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
5c93c473 | 412 | |
e10e798c | 413 | dispatcher_remove(&vubr->dispatcher, fd); |
3595e2eb VK |
414 | } |
415 | ||
416 | static int | |
e10e798c | 417 | vubr_send_rarp_exec(VuDev *dev, VhostUserMsg *vmsg) |
3595e2eb | 418 | { |
e10e798c | 419 | DPRINT("Function %s() not implemented yet.\n", __func__); |
3595e2eb VK |
420 | return 0; |
421 | } | |
422 | ||
423 | static int | |
e10e798c | 424 | vubr_process_msg(VuDev *dev, VhostUserMsg *vmsg, int *do_reply) |
3595e2eb | 425 | { |
e10e798c MAL |
426 | switch (vmsg->request) { |
427 | case VHOST_USER_SEND_RARP: | |
428 | *do_reply = vubr_send_rarp_exec(dev, vmsg); | |
429 | return 1; | |
430 | default: | |
431 | /* let the library handle the rest */ | |
432 | return 0; | |
5c93c473 | 433 | } |
5c93c473 | 434 | |
3595e2eb VK |
435 | return 0; |
436 | } | |
437 | ||
e10e798c MAL |
438 | static void |
439 | vubr_set_features(VuDev *dev, uint64_t features) | |
3595e2eb | 440 | { |
e10e798c | 441 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
3595e2eb | 442 | |
e10e798c MAL |
443 | if ((features & (1ULL << VIRTIO_F_VERSION_1)) || |
444 | (features & (1ULL << VIRTIO_NET_F_MRG_RXBUF))) { | |
445 | vubr->hdrlen = 12; | |
446 | } else { | |
447 | vubr->hdrlen = 10; | |
523b018d | 448 | } |
3595e2eb VK |
449 | } |
450 | ||
e10e798c MAL |
451 | static uint64_t |
452 | vubr_get_features(VuDev *dev) | |
3595e2eb | 453 | { |
e10e798c | 454 | return 1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE | |
e3af2928 TB |
455 | 1ULL << VIRTIO_NET_F_MRG_RXBUF | |
456 | 1ULL << VIRTIO_F_VERSION_1; | |
3595e2eb VK |
457 | } |
458 | ||
e10e798c MAL |
459 | static void |
460 | vubr_queue_set_started(VuDev *dev, int qidx, bool started) | |
3595e2eb | 461 | { |
e3af2928 | 462 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
e10e798c | 463 | VuVirtq *vq = vu_get_queue(dev, qidx); |
3595e2eb | 464 | |
e3af2928 TB |
465 | if (started && vubr->notifier.fd >= 0) { |
466 | vu_set_queue_host_notifier(dev, vq, vubr->notifier.fd, | |
467 | getpagesize(), | |
468 | qidx * getpagesize()); | |
469 | } | |
470 | ||
e10e798c MAL |
471 | if (qidx % 2 == 1) { |
472 | vu_set_queue_handler(dev, vq, started ? vubr_handle_tx : NULL); | |
6d0b908a | 473 | } |
3595e2eb VK |
474 | } |
475 | ||
e10e798c MAL |
476 | static void |
477 | vubr_panic(VuDev *dev, const char *msg) | |
3595e2eb | 478 | { |
e10e798c | 479 | VubrDev *vubr = container_of(dev, VubrDev, vudev); |
5c93c473 | 480 | |
e10e798c | 481 | fprintf(stderr, "PANIC: %s\n", msg); |
3595e2eb | 482 | |
e10e798c MAL |
483 | dispatcher_remove(&vubr->dispatcher, dev->sock); |
484 | vubr->quit = 1; | |
3595e2eb VK |
485 | } |
486 | ||
672339f7 MAL |
487 | static bool |
488 | vubr_queue_is_processed_in_order(VuDev *dev, int qidx) | |
489 | { | |
490 | return true; | |
491 | } | |
492 | ||
e10e798c MAL |
493 | static const VuDevIface vuiface = { |
494 | .get_features = vubr_get_features, | |
495 | .set_features = vubr_set_features, | |
496 | .process_msg = vubr_process_msg, | |
497 | .queue_set_started = vubr_queue_set_started, | |
672339f7 | 498 | .queue_is_processed_in_order = vubr_queue_is_processed_in_order, |
e10e798c | 499 | }; |
3595e2eb VK |
500 | |
501 | static void | |
502 | vubr_accept_cb(int sock, void *ctx) | |
503 | { | |
504 | VubrDev *dev = (VubrDev *)ctx; | |
505 | int conn_fd; | |
506 | struct sockaddr_un un; | |
507 | socklen_t len = sizeof(un); | |
508 | ||
509 | conn_fd = accept(sock, (struct sockaddr *) &un, &len); | |
5c93c473 | 510 | if (conn_fd == -1) { |
3595e2eb VK |
511 | vubr_die("accept()"); |
512 | } | |
513 | DPRINT("Got connection from remote peer on sock %d\n", conn_fd); | |
e10e798c MAL |
514 | |
515 | vu_init(&dev->vudev, | |
516 | conn_fd, | |
517 | vubr_panic, | |
518 | vubr_set_watch, | |
519 | vubr_remove_watch, | |
520 | &vuiface); | |
521 | ||
3595e2eb | 522 | dispatcher_add(&dev->dispatcher, conn_fd, ctx, vubr_receive_cb); |
98206d4e | 523 | dispatcher_remove(&dev->dispatcher, sock); |
3595e2eb VK |
524 | } |
525 | ||
526 | static VubrDev * | |
aef8486e | 527 | vubr_new(const char *path, bool client) |
3595e2eb VK |
528 | { |
529 | VubrDev *dev = (VubrDev *) calloc(1, sizeof(VubrDev)); | |
3595e2eb | 530 | struct sockaddr_un un; |
aef8486e | 531 | CallbackFunc cb; |
3595e2eb VK |
532 | size_t len; |
533 | ||
3595e2eb VK |
534 | /* Get a UNIX socket. */ |
535 | dev->sock = socket(AF_UNIX, SOCK_STREAM, 0); | |
536 | if (dev->sock == -1) { | |
537 | vubr_die("socket"); | |
538 | } | |
539 | ||
e3af2928 TB |
540 | dev->notifier.fd = -1; |
541 | ||
3595e2eb VK |
542 | un.sun_family = AF_UNIX; |
543 | strcpy(un.sun_path, path); | |
544 | len = sizeof(un.sun_family) + strlen(path); | |
3595e2eb | 545 | |
aef8486e MAL |
546 | if (!client) { |
547 | unlink(path); | |
3595e2eb | 548 | |
aef8486e MAL |
549 | if (bind(dev->sock, (struct sockaddr *) &un, len) == -1) { |
550 | vubr_die("bind"); | |
551 | } | |
552 | ||
553 | if (listen(dev->sock, 1) == -1) { | |
554 | vubr_die("listen"); | |
555 | } | |
556 | cb = vubr_accept_cb; | |
3595e2eb | 557 | |
aef8486e MAL |
558 | DPRINT("Waiting for connections on UNIX socket %s ...\n", path); |
559 | } else { | |
560 | if (connect(dev->sock, (struct sockaddr *)&un, len) == -1) { | |
561 | vubr_die("connect"); | |
562 | } | |
e10e798c MAL |
563 | vu_init(&dev->vudev, |
564 | dev->sock, | |
565 | vubr_panic, | |
566 | vubr_set_watch, | |
567 | vubr_remove_watch, | |
568 | &vuiface); | |
aef8486e | 569 | cb = vubr_receive_cb; |
3595e2eb VK |
570 | } |
571 | ||
572 | dispatcher_init(&dev->dispatcher); | |
e10e798c | 573 | |
aef8486e | 574 | dispatcher_add(&dev->dispatcher, dev->sock, (void *)dev, cb); |
3595e2eb | 575 | |
3595e2eb VK |
576 | return dev; |
577 | } | |
578 | ||
e3af2928 TB |
579 | static void *notifier_thread(void *arg) |
580 | { | |
581 | VuDev *dev = (VuDev *)arg; | |
582 | VubrDev *vubr = container_of(dev, VubrDev, vudev); | |
583 | int pagesize = getpagesize(); | |
584 | int qidx; | |
585 | ||
586 | while (true) { | |
587 | for (qidx = 0; qidx < VHOST_MAX_NR_VIRTQUEUE; qidx++) { | |
588 | uint16_t *n = vubr->notifier.addr + pagesize * qidx; | |
589 | ||
590 | if (*n == qidx) { | |
591 | *n = 0xffff; | |
592 | /* We won't miss notifications if we reset | |
593 | * the memory first. */ | |
594 | smp_mb(); | |
595 | ||
596 | DPRINT("Got a notification for queue%d via host notifier.\n", | |
597 | qidx); | |
598 | ||
599 | if (qidx % 2 == 1) { | |
600 | vubr_handle_tx(dev, qidx); | |
601 | } | |
602 | } | |
603 | usleep(1000); | |
604 | } | |
605 | } | |
606 | ||
607 | return NULL; | |
608 | } | |
609 | ||
610 | static void | |
611 | vubr_host_notifier_setup(VubrDev *dev) | |
612 | { | |
613 | char template[] = "/tmp/vubr-XXXXXX"; | |
614 | pthread_t thread; | |
615 | size_t length; | |
616 | void *addr; | |
617 | int fd; | |
618 | ||
619 | length = getpagesize() * VHOST_MAX_NR_VIRTQUEUE; | |
620 | ||
621 | fd = mkstemp(template); | |
622 | if (fd < 0) { | |
623 | vubr_die("mkstemp()"); | |
624 | } | |
625 | ||
626 | if (posix_fallocate(fd, 0, length) != 0) { | |
627 | vubr_die("posix_fallocate()"); | |
628 | } | |
629 | ||
630 | addr = mmap(NULL, length, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); | |
631 | if (addr == MAP_FAILED) { | |
632 | vubr_die("mmap()"); | |
633 | } | |
634 | ||
635 | memset(addr, 0xff, length); | |
636 | ||
637 | if (pthread_create(&thread, NULL, notifier_thread, &dev->vudev) != 0) { | |
638 | vubr_die("pthread_create()"); | |
639 | } | |
640 | ||
641 | dev->notifier.fd = fd; | |
642 | dev->notifier.addr = addr; | |
643 | dev->notifier.thread = thread; | |
644 | } | |
645 | ||
7cf32491 VK |
646 | static void |
647 | vubr_set_host(struct sockaddr_in *saddr, const char *host) | |
648 | { | |
d18dc3af | 649 | if (qemu_isdigit(host[0])) { |
7cf32491 VK |
650 | if (!inet_aton(host, &saddr->sin_addr)) { |
651 | fprintf(stderr, "inet_aton() failed.\n"); | |
652 | exit(1); | |
653 | } | |
654 | } else { | |
655 | struct hostent *he = gethostbyname(host); | |
656 | ||
657 | if (!he) { | |
658 | fprintf(stderr, "gethostbyname() failed.\n"); | |
659 | exit(1); | |
660 | } | |
661 | saddr->sin_addr = *(struct in_addr *)he->h_addr; | |
662 | } | |
663 | } | |
664 | ||
3595e2eb VK |
665 | static void |
666 | vubr_backend_udp_setup(VubrDev *dev, | |
667 | const char *local_host, | |
7cf32491 VK |
668 | const char *local_port, |
669 | const char *remote_host, | |
670 | const char *remote_port) | |
3595e2eb VK |
671 | { |
672 | int sock; | |
7cf32491 VK |
673 | const char *r; |
674 | ||
675 | int lport, rport; | |
676 | ||
677 | lport = strtol(local_port, (char **)&r, 0); | |
678 | if (r == local_port) { | |
679 | fprintf(stderr, "lport parsing failed.\n"); | |
680 | exit(1); | |
681 | } | |
682 | ||
683 | rport = strtol(remote_port, (char **)&r, 0); | |
684 | if (r == remote_port) { | |
685 | fprintf(stderr, "rport parsing failed.\n"); | |
686 | exit(1); | |
687 | } | |
688 | ||
3595e2eb VK |
689 | struct sockaddr_in si_local = { |
690 | .sin_family = AF_INET, | |
7cf32491 | 691 | .sin_port = htons(lport), |
3595e2eb VK |
692 | }; |
693 | ||
7cf32491 | 694 | vubr_set_host(&si_local, local_host); |
3595e2eb VK |
695 | |
696 | /* setup destination for sends */ | |
697 | dev->backend_udp_dest = (struct sockaddr_in) { | |
698 | .sin_family = AF_INET, | |
7cf32491 | 699 | .sin_port = htons(rport), |
3595e2eb | 700 | }; |
7cf32491 | 701 | vubr_set_host(&dev->backend_udp_dest, remote_host); |
3595e2eb VK |
702 | |
703 | sock = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP); | |
704 | if (sock == -1) { | |
705 | vubr_die("socket"); | |
706 | } | |
707 | ||
708 | if (bind(sock, (struct sockaddr *)&si_local, sizeof(si_local)) == -1) { | |
709 | vubr_die("bind"); | |
710 | } | |
711 | ||
712 | dev->backend_udp_sock = sock; | |
713 | dispatcher_add(&dev->dispatcher, sock, dev, vubr_backend_recv_cb); | |
714 | DPRINT("Waiting for data from udp backend on %s:%d...\n", | |
7cf32491 | 715 | local_host, lport); |
3595e2eb VK |
716 | } |
717 | ||
718 | static void | |
719 | vubr_run(VubrDev *dev) | |
720 | { | |
e10e798c | 721 | while (!dev->quit) { |
3595e2eb VK |
722 | /* timeout 200ms */ |
723 | dispatcher_wait(&dev->dispatcher, 200000); | |
724 | /* Here one can try polling strategy. */ | |
725 | } | |
726 | } | |
727 | ||
7cf32491 VK |
728 | static int |
729 | vubr_parse_host_port(const char **host, const char **port, const char *buf) | |
730 | { | |
731 | char *p = strchr(buf, ':'); | |
732 | ||
733 | if (!p) { | |
734 | return -1; | |
735 | } | |
736 | *p = '\0'; | |
737 | *host = strdup(buf); | |
738 | *port = strdup(p + 1); | |
739 | return 0; | |
740 | } | |
741 | ||
742 | #define DEFAULT_UD_SOCKET "/tmp/vubr.sock" | |
743 | #define DEFAULT_LHOST "127.0.0.1" | |
744 | #define DEFAULT_LPORT "4444" | |
745 | #define DEFAULT_RHOST "127.0.0.1" | |
746 | #define DEFAULT_RPORT "5555" | |
747 | ||
748 | static const char *ud_socket_path = DEFAULT_UD_SOCKET; | |
749 | static const char *lhost = DEFAULT_LHOST; | |
750 | static const char *lport = DEFAULT_LPORT; | |
751 | static const char *rhost = DEFAULT_RHOST; | |
752 | static const char *rport = DEFAULT_RPORT; | |
753 | ||
3595e2eb VK |
754 | int |
755 | main(int argc, char *argv[]) | |
756 | { | |
757 | VubrDev *dev; | |
7cf32491 | 758 | int opt; |
aef8486e | 759 | bool client = false; |
e3af2928 | 760 | bool host_notifier = false; |
3595e2eb | 761 | |
e3af2928 | 762 | while ((opt = getopt(argc, argv, "l:r:u:cH")) != -1) { |
7cf32491 VK |
763 | |
764 | switch (opt) { | |
765 | case 'l': | |
766 | if (vubr_parse_host_port(&lhost, &lport, optarg) < 0) { | |
767 | goto out; | |
768 | } | |
769 | break; | |
770 | case 'r': | |
771 | if (vubr_parse_host_port(&rhost, &rport, optarg) < 0) { | |
772 | goto out; | |
773 | } | |
774 | break; | |
775 | case 'u': | |
776 | ud_socket_path = strdup(optarg); | |
777 | break; | |
aef8486e MAL |
778 | case 'c': |
779 | client = true; | |
780 | break; | |
e3af2928 TB |
781 | case 'H': |
782 | host_notifier = true; | |
783 | break; | |
7cf32491 VK |
784 | default: |
785 | goto out; | |
786 | } | |
787 | } | |
788 | ||
aef8486e MAL |
789 | DPRINT("ud socket: %s (%s)\n", ud_socket_path, |
790 | client ? "client" : "server"); | |
7cf32491 VK |
791 | DPRINT("local: %s:%s\n", lhost, lport); |
792 | DPRINT("remote: %s:%s\n", rhost, rport); | |
793 | ||
aef8486e | 794 | dev = vubr_new(ud_socket_path, client); |
3595e2eb VK |
795 | if (!dev) { |
796 | return 1; | |
797 | } | |
798 | ||
e3af2928 TB |
799 | if (host_notifier) { |
800 | vubr_host_notifier_setup(dev); | |
801 | } | |
802 | ||
7cf32491 | 803 | vubr_backend_udp_setup(dev, lhost, lport, rhost, rport); |
3595e2eb | 804 | vubr_run(dev); |
e10e798c MAL |
805 | |
806 | vu_deinit(&dev->vudev); | |
807 | ||
3595e2eb | 808 | return 0; |
7cf32491 VK |
809 | |
810 | out: | |
811 | fprintf(stderr, "Usage: %s ", argv[0]); | |
e3af2928 | 812 | fprintf(stderr, "[-c] [-H] [-u ud_socket_path] [-l lhost:lport] [-r rhost:rport]\n"); |
7cf32491 VK |
813 | fprintf(stderr, "\t-u path to unix doman socket. default: %s\n", |
814 | DEFAULT_UD_SOCKET); | |
815 | fprintf(stderr, "\t-l local host and port. default: %s:%s\n", | |
816 | DEFAULT_LHOST, DEFAULT_LPORT); | |
817 | fprintf(stderr, "\t-r remote host and port. default: %s:%s\n", | |
818 | DEFAULT_RHOST, DEFAULT_RPORT); | |
aef8486e | 819 | fprintf(stderr, "\t-c client mode\n"); |
e3af2928 | 820 | fprintf(stderr, "\t-H use host notifier\n"); |
7cf32491 VK |
821 | |
822 | return 1; | |
3595e2eb | 823 | } |