]>
Commit | Line | Data |
---|---|---|
559607ea DB |
1 | /* |
2 | * QEMU I/O channels sockets driver | |
3 | * | |
4 | * Copyright (c) 2015 Red Hat, Inc. | |
5 | * | |
6 | * This library is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU Lesser General Public | |
8 | * License as published by the Free Software Foundation; either | |
c8198bd5 | 9 | * version 2.1 of the License, or (at your option) any later version. |
559607ea DB |
10 | * |
11 | * This library is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 | * Lesser General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU Lesser General Public | |
17 | * License along with this library; if not, see <http://www.gnu.org/licenses/>. | |
559607ea DB |
18 | */ |
19 | ||
cae9fc56 | 20 | #include "qemu/osdep.h" |
da34e65c | 21 | #include "qapi/error.h" |
9af23989 | 22 | #include "qapi/qapi-visit-sockets.h" |
0b8fa32f | 23 | #include "qemu/module.h" |
559607ea | 24 | #include "io/channel-socket.h" |
06e0f098 | 25 | #include "io/channel-util.h" |
559607ea DB |
26 | #include "io/channel-watch.h" |
27 | #include "trace.h" | |
37f9e0a2 | 28 | #include "qapi/clone-visitor.h" |
2bc58ffc LB |
29 | #ifdef CONFIG_LINUX |
30 | #include <linux/errqueue.h> | |
31 | #include <sys/socket.h> | |
32 | ||
33 | #if (defined(MSG_ZEROCOPY) && defined(SO_ZEROCOPY)) | |
34 | #define QEMU_MSG_ZEROCOPY | |
35 | #endif | |
36 | #endif | |
559607ea DB |
37 | |
38 | #define SOCKET_MAX_FDS 16 | |
39 | ||
40 | SocketAddress * | |
41 | qio_channel_socket_get_local_address(QIOChannelSocket *ioc, | |
42 | Error **errp) | |
43 | { | |
44 | return socket_sockaddr_to_address(&ioc->localAddr, | |
45 | ioc->localAddrLen, | |
46 | errp); | |
47 | } | |
48 | ||
49 | SocketAddress * | |
50 | qio_channel_socket_get_remote_address(QIOChannelSocket *ioc, | |
51 | Error **errp) | |
52 | { | |
53 | return socket_sockaddr_to_address(&ioc->remoteAddr, | |
54 | ioc->remoteAddrLen, | |
55 | errp); | |
56 | } | |
57 | ||
58 | QIOChannelSocket * | |
59 | qio_channel_socket_new(void) | |
60 | { | |
61 | QIOChannelSocket *sioc; | |
62 | QIOChannel *ioc; | |
63 | ||
64 | sioc = QIO_CHANNEL_SOCKET(object_new(TYPE_QIO_CHANNEL_SOCKET)); | |
65 | sioc->fd = -1; | |
2bc58ffc LB |
66 | sioc->zero_copy_queued = 0; |
67 | sioc->zero_copy_sent = 0; | |
559607ea DB |
68 | |
69 | ioc = QIO_CHANNEL(sioc); | |
d8d3c7cc | 70 | qio_channel_set_feature(ioc, QIO_CHANNEL_FEATURE_SHUTDOWN); |
559607ea | 71 | |
a5897205 PB |
72 | #ifdef WIN32 |
73 | ioc->event = CreateEvent(NULL, FALSE, FALSE, NULL); | |
74 | #endif | |
75 | ||
559607ea DB |
76 | trace_qio_channel_socket_new(sioc); |
77 | ||
78 | return sioc; | |
79 | } | |
80 | ||
81 | ||
82 | static int | |
83 | qio_channel_socket_set_fd(QIOChannelSocket *sioc, | |
84 | int fd, | |
85 | Error **errp) | |
86 | { | |
87 | if (sioc->fd != -1) { | |
88 | error_setg(errp, "Socket is already open"); | |
89 | return -1; | |
90 | } | |
91 | ||
92 | sioc->fd = fd; | |
93 | sioc->remoteAddrLen = sizeof(sioc->remoteAddr); | |
94 | sioc->localAddrLen = sizeof(sioc->localAddr); | |
95 | ||
96 | ||
97 | if (getpeername(fd, (struct sockaddr *)&sioc->remoteAddr, | |
98 | &sioc->remoteAddrLen) < 0) { | |
b16a44e1 | 99 | if (errno == ENOTCONN) { |
559607ea DB |
100 | memset(&sioc->remoteAddr, 0, sizeof(sioc->remoteAddr)); |
101 | sioc->remoteAddrLen = sizeof(sioc->remoteAddr); | |
102 | } else { | |
b16a44e1 | 103 | error_setg_errno(errp, errno, |
559607ea DB |
104 | "Unable to query remote socket address"); |
105 | goto error; | |
106 | } | |
107 | } | |
108 | ||
109 | if (getsockname(fd, (struct sockaddr *)&sioc->localAddr, | |
110 | &sioc->localAddrLen) < 0) { | |
b16a44e1 | 111 | error_setg_errno(errp, errno, |
559607ea DB |
112 | "Unable to query local socket address"); |
113 | goto error; | |
114 | } | |
115 | ||
116 | #ifndef WIN32 | |
117 | if (sioc->localAddr.ss_family == AF_UNIX) { | |
118 | QIOChannel *ioc = QIO_CHANNEL(sioc); | |
d8d3c7cc | 119 | qio_channel_set_feature(ioc, QIO_CHANNEL_FEATURE_FD_PASS); |
559607ea DB |
120 | } |
121 | #endif /* WIN32 */ | |
122 | ||
123 | return 0; | |
124 | ||
125 | error: | |
126 | sioc->fd = -1; /* Let the caller close FD on failure */ | |
127 | return -1; | |
128 | } | |
129 | ||
130 | QIOChannelSocket * | |
131 | qio_channel_socket_new_fd(int fd, | |
132 | Error **errp) | |
133 | { | |
134 | QIOChannelSocket *ioc; | |
135 | ||
136 | ioc = qio_channel_socket_new(); | |
137 | if (qio_channel_socket_set_fd(ioc, fd, errp) < 0) { | |
138 | object_unref(OBJECT(ioc)); | |
139 | return NULL; | |
140 | } | |
141 | ||
142 | trace_qio_channel_socket_new_fd(ioc, fd); | |
143 | ||
144 | return ioc; | |
145 | } | |
146 | ||
147 | ||
148 | int qio_channel_socket_connect_sync(QIOChannelSocket *ioc, | |
149 | SocketAddress *addr, | |
150 | Error **errp) | |
151 | { | |
152 | int fd; | |
153 | ||
154 | trace_qio_channel_socket_connect_sync(ioc, addr); | |
b2587932 | 155 | fd = socket_connect(addr, errp); |
559607ea DB |
156 | if (fd < 0) { |
157 | trace_qio_channel_socket_connect_fail(ioc); | |
158 | return -1; | |
159 | } | |
160 | ||
161 | trace_qio_channel_socket_connect_complete(ioc, fd); | |
162 | if (qio_channel_socket_set_fd(ioc, fd, errp) < 0) { | |
25657fc6 | 163 | close(fd); |
559607ea DB |
164 | return -1; |
165 | } | |
166 | ||
2bc58ffc LB |
167 | #ifdef QEMU_MSG_ZEROCOPY |
168 | int ret, v = 1; | |
169 | ret = setsockopt(fd, SOL_SOCKET, SO_ZEROCOPY, &v, sizeof(v)); | |
170 | if (ret == 0) { | |
171 | /* Zero copy available on host */ | |
172 | qio_channel_set_feature(QIO_CHANNEL(ioc), | |
173 | QIO_CHANNEL_FEATURE_WRITE_ZERO_COPY); | |
174 | } | |
175 | #endif | |
176 | ||
84615a19 | 177 | qio_channel_set_feature(QIO_CHANNEL(ioc), |
178 | QIO_CHANNEL_FEATURE_READ_MSG_PEEK); | |
179 | ||
559607ea DB |
180 | return 0; |
181 | } | |
182 | ||
183 | ||
59de517d DB |
184 | static void qio_channel_socket_connect_worker(QIOTask *task, |
185 | gpointer opaque) | |
559607ea DB |
186 | { |
187 | QIOChannelSocket *ioc = QIO_CHANNEL_SOCKET(qio_task_get_source(task)); | |
188 | SocketAddress *addr = opaque; | |
59de517d | 189 | Error *err = NULL; |
559607ea | 190 | |
59de517d | 191 | qio_channel_socket_connect_sync(ioc, addr, &err); |
559607ea | 192 | |
59de517d | 193 | qio_task_set_error(task, err); |
559607ea DB |
194 | } |
195 | ||
196 | ||
197 | void qio_channel_socket_connect_async(QIOChannelSocket *ioc, | |
198 | SocketAddress *addr, | |
199 | QIOTaskFunc callback, | |
200 | gpointer opaque, | |
8005fdd8 PX |
201 | GDestroyNotify destroy, |
202 | GMainContext *context) | |
559607ea DB |
203 | { |
204 | QIOTask *task = qio_task_new( | |
205 | OBJECT(ioc), callback, opaque, destroy); | |
206 | SocketAddress *addrCopy; | |
207 | ||
37f9e0a2 | 208 | addrCopy = QAPI_CLONE(SocketAddress, addr); |
559607ea DB |
209 | |
210 | /* socket_connect() does a non-blocking connect(), but it | |
211 | * still blocks in DNS lookups, so we must use a thread */ | |
212 | trace_qio_channel_socket_connect_async(ioc, addr); | |
213 | qio_task_run_in_thread(task, | |
214 | qio_channel_socket_connect_worker, | |
215 | addrCopy, | |
a17536c5 | 216 | (GDestroyNotify)qapi_free_SocketAddress, |
8005fdd8 | 217 | context); |
559607ea DB |
218 | } |
219 | ||
220 | ||
221 | int qio_channel_socket_listen_sync(QIOChannelSocket *ioc, | |
222 | SocketAddress *addr, | |
4e2d8bf6 | 223 | int num, |
559607ea DB |
224 | Error **errp) |
225 | { | |
226 | int fd; | |
227 | ||
4e2d8bf6 JQ |
228 | trace_qio_channel_socket_listen_sync(ioc, addr, num); |
229 | fd = socket_listen(addr, num, errp); | |
559607ea DB |
230 | if (fd < 0) { |
231 | trace_qio_channel_socket_listen_fail(ioc); | |
232 | return -1; | |
233 | } | |
234 | ||
235 | trace_qio_channel_socket_listen_complete(ioc, fd); | |
236 | if (qio_channel_socket_set_fd(ioc, fd, errp) < 0) { | |
25657fc6 | 237 | close(fd); |
559607ea DB |
238 | return -1; |
239 | } | |
bf535208 | 240 | qio_channel_set_feature(QIO_CHANNEL(ioc), QIO_CHANNEL_FEATURE_LISTEN); |
559607ea DB |
241 | |
242 | return 0; | |
243 | } | |
244 | ||
245 | ||
7959e29e JQ |
246 | struct QIOChannelListenWorkerData { |
247 | SocketAddress *addr; | |
248 | int num; /* amount of expected connections */ | |
249 | }; | |
250 | ||
251 | static void qio_channel_listen_worker_free(gpointer opaque) | |
252 | { | |
253 | struct QIOChannelListenWorkerData *data = opaque; | |
254 | ||
255 | qapi_free_SocketAddress(data->addr); | |
256 | g_free(data); | |
257 | } | |
258 | ||
59de517d DB |
259 | static void qio_channel_socket_listen_worker(QIOTask *task, |
260 | gpointer opaque) | |
559607ea DB |
261 | { |
262 | QIOChannelSocket *ioc = QIO_CHANNEL_SOCKET(qio_task_get_source(task)); | |
7959e29e | 263 | struct QIOChannelListenWorkerData *data = opaque; |
59de517d | 264 | Error *err = NULL; |
559607ea | 265 | |
7959e29e | 266 | qio_channel_socket_listen_sync(ioc, data->addr, data->num, &err); |
559607ea | 267 | |
59de517d | 268 | qio_task_set_error(task, err); |
559607ea DB |
269 | } |
270 | ||
271 | ||
272 | void qio_channel_socket_listen_async(QIOChannelSocket *ioc, | |
273 | SocketAddress *addr, | |
7959e29e | 274 | int num, |
559607ea DB |
275 | QIOTaskFunc callback, |
276 | gpointer opaque, | |
8005fdd8 PX |
277 | GDestroyNotify destroy, |
278 | GMainContext *context) | |
559607ea DB |
279 | { |
280 | QIOTask *task = qio_task_new( | |
281 | OBJECT(ioc), callback, opaque, destroy); | |
7959e29e | 282 | struct QIOChannelListenWorkerData *data; |
559607ea | 283 | |
7959e29e JQ |
284 | data = g_new0(struct QIOChannelListenWorkerData, 1); |
285 | data->addr = QAPI_CLONE(SocketAddress, addr); | |
286 | data->num = num; | |
559607ea DB |
287 | |
288 | /* socket_listen() blocks in DNS lookups, so we must use a thread */ | |
7959e29e | 289 | trace_qio_channel_socket_listen_async(ioc, addr, num); |
559607ea DB |
290 | qio_task_run_in_thread(task, |
291 | qio_channel_socket_listen_worker, | |
7959e29e JQ |
292 | data, |
293 | qio_channel_listen_worker_free, | |
8005fdd8 | 294 | context); |
559607ea DB |
295 | } |
296 | ||
297 | ||
298 | int qio_channel_socket_dgram_sync(QIOChannelSocket *ioc, | |
299 | SocketAddress *localAddr, | |
300 | SocketAddress *remoteAddr, | |
301 | Error **errp) | |
302 | { | |
303 | int fd; | |
304 | ||
305 | trace_qio_channel_socket_dgram_sync(ioc, localAddr, remoteAddr); | |
150dcd1a | 306 | fd = socket_dgram(remoteAddr, localAddr, errp); |
559607ea DB |
307 | if (fd < 0) { |
308 | trace_qio_channel_socket_dgram_fail(ioc); | |
309 | return -1; | |
310 | } | |
311 | ||
312 | trace_qio_channel_socket_dgram_complete(ioc, fd); | |
313 | if (qio_channel_socket_set_fd(ioc, fd, errp) < 0) { | |
25657fc6 | 314 | close(fd); |
559607ea DB |
315 | return -1; |
316 | } | |
317 | ||
318 | return 0; | |
319 | } | |
320 | ||
321 | ||
322 | struct QIOChannelSocketDGramWorkerData { | |
323 | SocketAddress *localAddr; | |
324 | SocketAddress *remoteAddr; | |
325 | }; | |
326 | ||
327 | ||
328 | static void qio_channel_socket_dgram_worker_free(gpointer opaque) | |
329 | { | |
330 | struct QIOChannelSocketDGramWorkerData *data = opaque; | |
331 | qapi_free_SocketAddress(data->localAddr); | |
332 | qapi_free_SocketAddress(data->remoteAddr); | |
333 | g_free(data); | |
334 | } | |
335 | ||
59de517d DB |
336 | static void qio_channel_socket_dgram_worker(QIOTask *task, |
337 | gpointer opaque) | |
559607ea DB |
338 | { |
339 | QIOChannelSocket *ioc = QIO_CHANNEL_SOCKET(qio_task_get_source(task)); | |
340 | struct QIOChannelSocketDGramWorkerData *data = opaque; | |
59de517d | 341 | Error *err = NULL; |
559607ea DB |
342 | |
343 | /* socket_dgram() blocks in DNS lookups, so we must use a thread */ | |
59de517d DB |
344 | qio_channel_socket_dgram_sync(ioc, data->localAddr, |
345 | data->remoteAddr, &err); | |
559607ea | 346 | |
59de517d | 347 | qio_task_set_error(task, err); |
559607ea DB |
348 | } |
349 | ||
350 | ||
351 | void qio_channel_socket_dgram_async(QIOChannelSocket *ioc, | |
352 | SocketAddress *localAddr, | |
353 | SocketAddress *remoteAddr, | |
354 | QIOTaskFunc callback, | |
355 | gpointer opaque, | |
8005fdd8 PX |
356 | GDestroyNotify destroy, |
357 | GMainContext *context) | |
559607ea DB |
358 | { |
359 | QIOTask *task = qio_task_new( | |
360 | OBJECT(ioc), callback, opaque, destroy); | |
361 | struct QIOChannelSocketDGramWorkerData *data = g_new0( | |
362 | struct QIOChannelSocketDGramWorkerData, 1); | |
363 | ||
37f9e0a2 EB |
364 | data->localAddr = QAPI_CLONE(SocketAddress, localAddr); |
365 | data->remoteAddr = QAPI_CLONE(SocketAddress, remoteAddr); | |
559607ea DB |
366 | |
367 | trace_qio_channel_socket_dgram_async(ioc, localAddr, remoteAddr); | |
368 | qio_task_run_in_thread(task, | |
369 | qio_channel_socket_dgram_worker, | |
370 | data, | |
a17536c5 | 371 | qio_channel_socket_dgram_worker_free, |
8005fdd8 | 372 | context); |
559607ea DB |
373 | } |
374 | ||
375 | ||
376 | QIOChannelSocket * | |
377 | qio_channel_socket_accept(QIOChannelSocket *ioc, | |
378 | Error **errp) | |
379 | { | |
380 | QIOChannelSocket *cioc; | |
381 | ||
0e5d6327 | 382 | cioc = qio_channel_socket_new(); |
559607ea DB |
383 | cioc->remoteAddrLen = sizeof(ioc->remoteAddr); |
384 | cioc->localAddrLen = sizeof(ioc->localAddr); | |
385 | ||
386 | retry: | |
387 | trace_qio_channel_socket_accept(ioc); | |
de7971ff DB |
388 | cioc->fd = qemu_accept(ioc->fd, (struct sockaddr *)&cioc->remoteAddr, |
389 | &cioc->remoteAddrLen); | |
559607ea | 390 | if (cioc->fd < 0) { |
b16a44e1 | 391 | if (errno == EINTR) { |
559607ea DB |
392 | goto retry; |
393 | } | |
8bd9c4e6 PX |
394 | error_setg_errno(errp, errno, "Unable to accept connection"); |
395 | trace_qio_channel_socket_accept_fail(ioc); | |
559607ea DB |
396 | goto error; |
397 | } | |
398 | ||
bead5994 DB |
399 | if (getsockname(cioc->fd, (struct sockaddr *)&cioc->localAddr, |
400 | &cioc->localAddrLen) < 0) { | |
b16a44e1 | 401 | error_setg_errno(errp, errno, |
559607ea DB |
402 | "Unable to query local socket address"); |
403 | goto error; | |
404 | } | |
405 | ||
bead5994 DB |
406 | #ifndef WIN32 |
407 | if (cioc->localAddr.ss_family == AF_UNIX) { | |
d8d3c7cc FF |
408 | QIOChannel *ioc_local = QIO_CHANNEL(cioc); |
409 | qio_channel_set_feature(ioc_local, QIO_CHANNEL_FEATURE_FD_PASS); | |
bead5994 DB |
410 | } |
411 | #endif /* WIN32 */ | |
412 | ||
84615a19 | 413 | qio_channel_set_feature(QIO_CHANNEL(cioc), |
414 | QIO_CHANNEL_FEATURE_READ_MSG_PEEK); | |
415 | ||
559607ea DB |
416 | trace_qio_channel_socket_accept_complete(ioc, cioc, cioc->fd); |
417 | return cioc; | |
418 | ||
419 | error: | |
420 | object_unref(OBJECT(cioc)); | |
421 | return NULL; | |
422 | } | |
423 | ||
424 | static void qio_channel_socket_init(Object *obj) | |
425 | { | |
426 | QIOChannelSocket *ioc = QIO_CHANNEL_SOCKET(obj); | |
427 | ioc->fd = -1; | |
428 | } | |
429 | ||
430 | static void qio_channel_socket_finalize(Object *obj) | |
431 | { | |
432 | QIOChannelSocket *ioc = QIO_CHANNEL_SOCKET(obj); | |
74b6ce43 | 433 | |
559607ea | 434 | if (ioc->fd != -1) { |
e413ae0c FF |
435 | QIOChannel *ioc_local = QIO_CHANNEL(ioc); |
436 | if (qio_channel_has_feature(ioc_local, QIO_CHANNEL_FEATURE_LISTEN)) { | |
74b6ce43 MAL |
437 | Error *err = NULL; |
438 | ||
439 | socket_listen_cleanup(ioc->fd, &err); | |
440 | if (err) { | |
441 | error_report_err(err); | |
442 | err = NULL; | |
443 | } | |
444 | } | |
a5897205 | 445 | #ifdef WIN32 |
a4aafea2 | 446 | qemu_socket_unselect(ioc->fd, NULL); |
a5897205 | 447 | #endif |
25657fc6 | 448 | close(ioc->fd); |
559607ea DB |
449 | ioc->fd = -1; |
450 | } | |
451 | } | |
452 | ||
453 | ||
454 | #ifndef WIN32 | |
455 | static void qio_channel_socket_copy_fds(struct msghdr *msg, | |
456 | int **fds, size_t *nfds) | |
457 | { | |
458 | struct cmsghdr *cmsg; | |
459 | ||
460 | *nfds = 0; | |
461 | *fds = NULL; | |
462 | ||
463 | for (cmsg = CMSG_FIRSTHDR(msg); cmsg; cmsg = CMSG_NXTHDR(msg, cmsg)) { | |
464 | int fd_size, i; | |
465 | int gotfds; | |
466 | ||
467 | if (cmsg->cmsg_len < CMSG_LEN(sizeof(int)) || | |
468 | cmsg->cmsg_level != SOL_SOCKET || | |
469 | cmsg->cmsg_type != SCM_RIGHTS) { | |
470 | continue; | |
471 | } | |
472 | ||
473 | fd_size = cmsg->cmsg_len - CMSG_LEN(0); | |
474 | ||
475 | if (!fd_size) { | |
476 | continue; | |
477 | } | |
478 | ||
479 | gotfds = fd_size / sizeof(int); | |
480 | *fds = g_renew(int, *fds, *nfds + gotfds); | |
481 | memcpy(*fds + *nfds, CMSG_DATA(cmsg), fd_size); | |
482 | ||
483 | for (i = 0; i < gotfds; i++) { | |
484 | int fd = (*fds)[*nfds + i]; | |
485 | if (fd < 0) { | |
486 | continue; | |
487 | } | |
488 | ||
489 | /* O_NONBLOCK is preserved across SCM_RIGHTS so reset it */ | |
ff5927ba | 490 | qemu_socket_set_block(fd); |
559607ea DB |
491 | |
492 | #ifndef MSG_CMSG_CLOEXEC | |
493 | qemu_set_cloexec(fd); | |
494 | #endif | |
495 | } | |
496 | *nfds += gotfds; | |
497 | } | |
498 | } | |
499 | ||
500 | ||
501 | static ssize_t qio_channel_socket_readv(QIOChannel *ioc, | |
502 | const struct iovec *iov, | |
503 | size_t niov, | |
504 | int **fds, | |
505 | size_t *nfds, | |
84615a19 | 506 | int flags, |
559607ea DB |
507 | Error **errp) |
508 | { | |
509 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
510 | ssize_t ret; | |
511 | struct msghdr msg = { NULL, }; | |
512 | char control[CMSG_SPACE(sizeof(int) * SOCKET_MAX_FDS)]; | |
513 | int sflags = 0; | |
514 | ||
ccf1e2dc DB |
515 | memset(control, 0, CMSG_SPACE(sizeof(int) * SOCKET_MAX_FDS)); |
516 | ||
559607ea DB |
517 | msg.msg_iov = (struct iovec *)iov; |
518 | msg.msg_iovlen = niov; | |
519 | if (fds && nfds) { | |
520 | msg.msg_control = control; | |
521 | msg.msg_controllen = sizeof(control); | |
d80f54ce DDAG |
522 | #ifdef MSG_CMSG_CLOEXEC |
523 | sflags |= MSG_CMSG_CLOEXEC; | |
524 | #endif | |
525 | ||
559607ea DB |
526 | } |
527 | ||
84615a19 | 528 | if (flags & QIO_CHANNEL_READ_FLAG_MSG_PEEK) { |
529 | sflags |= MSG_PEEK; | |
530 | } | |
531 | ||
559607ea DB |
532 | retry: |
533 | ret = recvmsg(sioc->fd, &msg, sflags); | |
534 | if (ret < 0) { | |
b16a44e1 | 535 | if (errno == EAGAIN) { |
559607ea DB |
536 | return QIO_CHANNEL_ERR_BLOCK; |
537 | } | |
b16a44e1 | 538 | if (errno == EINTR) { |
559607ea DB |
539 | goto retry; |
540 | } | |
541 | ||
b16a44e1 | 542 | error_setg_errno(errp, errno, |
559607ea DB |
543 | "Unable to read from socket"); |
544 | return -1; | |
545 | } | |
546 | ||
547 | if (fds && nfds) { | |
548 | qio_channel_socket_copy_fds(&msg, fds, nfds); | |
549 | } | |
550 | ||
551 | return ret; | |
552 | } | |
553 | ||
554 | static ssize_t qio_channel_socket_writev(QIOChannel *ioc, | |
555 | const struct iovec *iov, | |
556 | size_t niov, | |
557 | int *fds, | |
558 | size_t nfds, | |
b88651cb | 559 | int flags, |
559607ea DB |
560 | Error **errp) |
561 | { | |
562 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
563 | ssize_t ret; | |
564 | struct msghdr msg = { NULL, }; | |
ccf1e2dc | 565 | char control[CMSG_SPACE(sizeof(int) * SOCKET_MAX_FDS)]; |
7b3c618a DB |
566 | size_t fdsize = sizeof(int) * nfds; |
567 | struct cmsghdr *cmsg; | |
2bc58ffc | 568 | int sflags = 0; |
559607ea | 569 | |
ccf1e2dc DB |
570 | memset(control, 0, CMSG_SPACE(sizeof(int) * SOCKET_MAX_FDS)); |
571 | ||
559607ea DB |
572 | msg.msg_iov = (struct iovec *)iov; |
573 | msg.msg_iovlen = niov; | |
574 | ||
575 | if (nfds) { | |
559607ea | 576 | if (nfds > SOCKET_MAX_FDS) { |
cc75a50c | 577 | error_setg_errno(errp, EINVAL, |
559607ea DB |
578 | "Only %d FDs can be sent, got %zu", |
579 | SOCKET_MAX_FDS, nfds); | |
580 | return -1; | |
581 | } | |
582 | ||
583 | msg.msg_control = control; | |
584 | msg.msg_controllen = CMSG_SPACE(sizeof(int) * nfds); | |
585 | ||
586 | cmsg = CMSG_FIRSTHDR(&msg); | |
587 | cmsg->cmsg_len = CMSG_LEN(fdsize); | |
588 | cmsg->cmsg_level = SOL_SOCKET; | |
589 | cmsg->cmsg_type = SCM_RIGHTS; | |
590 | memcpy(CMSG_DATA(cmsg), fds, fdsize); | |
591 | } | |
592 | ||
2bc58ffc | 593 | if (flags & QIO_CHANNEL_WRITE_FLAG_ZERO_COPY) { |
803ca43e | 594 | #ifdef QEMU_MSG_ZEROCOPY |
2bc58ffc | 595 | sflags = MSG_ZEROCOPY; |
803ca43e LB |
596 | #else |
597 | /* | |
598 | * We expect QIOChannel class entry point to have | |
599 | * blocked this code path already | |
600 | */ | |
601 | g_assert_not_reached(); | |
2bc58ffc | 602 | #endif |
803ca43e | 603 | } |
2bc58ffc | 604 | |
559607ea | 605 | retry: |
2bc58ffc | 606 | ret = sendmsg(sioc->fd, &msg, sflags); |
559607ea | 607 | if (ret <= 0) { |
2bc58ffc LB |
608 | switch (errno) { |
609 | case EAGAIN: | |
559607ea | 610 | return QIO_CHANNEL_ERR_BLOCK; |
2bc58ffc | 611 | case EINTR: |
559607ea | 612 | goto retry; |
2bc58ffc | 613 | case ENOBUFS: |
803ca43e | 614 | if (flags & QIO_CHANNEL_WRITE_FLAG_ZERO_COPY) { |
2bc58ffc LB |
615 | error_setg_errno(errp, errno, |
616 | "Process can't lock enough memory for using MSG_ZEROCOPY"); | |
617 | return -1; | |
618 | } | |
619 | break; | |
559607ea | 620 | } |
2bc58ffc | 621 | |
b16a44e1 | 622 | error_setg_errno(errp, errno, |
559607ea DB |
623 | "Unable to write to socket"); |
624 | return -1; | |
625 | } | |
4f5a0971 LB |
626 | |
627 | if (flags & QIO_CHANNEL_WRITE_FLAG_ZERO_COPY) { | |
628 | sioc->zero_copy_queued++; | |
629 | } | |
630 | ||
559607ea DB |
631 | return ret; |
632 | } | |
633 | #else /* WIN32 */ | |
634 | static ssize_t qio_channel_socket_readv(QIOChannel *ioc, | |
635 | const struct iovec *iov, | |
636 | size_t niov, | |
637 | int **fds, | |
638 | size_t *nfds, | |
84615a19 | 639 | int flags, |
559607ea DB |
640 | Error **errp) |
641 | { | |
642 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
643 | ssize_t done = 0; | |
644 | ssize_t i; | |
84615a19 | 645 | int sflags = 0; |
646 | ||
647 | if (flags & QIO_CHANNEL_READ_FLAG_MSG_PEEK) { | |
648 | sflags |= MSG_PEEK; | |
649 | } | |
559607ea DB |
650 | |
651 | for (i = 0; i < niov; i++) { | |
652 | ssize_t ret; | |
653 | retry: | |
654 | ret = recv(sioc->fd, | |
655 | iov[i].iov_base, | |
656 | iov[i].iov_len, | |
84615a19 | 657 | sflags); |
559607ea | 658 | if (ret < 0) { |
b16a44e1 | 659 | if (errno == EAGAIN) { |
559607ea DB |
660 | if (done) { |
661 | return done; | |
662 | } else { | |
663 | return QIO_CHANNEL_ERR_BLOCK; | |
664 | } | |
b16a44e1 | 665 | } else if (errno == EINTR) { |
559607ea DB |
666 | goto retry; |
667 | } else { | |
b16a44e1 | 668 | error_setg_errno(errp, errno, |
5151d23e | 669 | "Unable to read from socket"); |
559607ea DB |
670 | return -1; |
671 | } | |
672 | } | |
673 | done += ret; | |
674 | if (ret < iov[i].iov_len) { | |
675 | return done; | |
676 | } | |
677 | } | |
678 | ||
679 | return done; | |
680 | } | |
681 | ||
682 | static ssize_t qio_channel_socket_writev(QIOChannel *ioc, | |
683 | const struct iovec *iov, | |
684 | size_t niov, | |
685 | int *fds, | |
686 | size_t nfds, | |
b88651cb | 687 | int flags, |
559607ea DB |
688 | Error **errp) |
689 | { | |
690 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
691 | ssize_t done = 0; | |
692 | ssize_t i; | |
693 | ||
694 | for (i = 0; i < niov; i++) { | |
695 | ssize_t ret; | |
696 | retry: | |
697 | ret = send(sioc->fd, | |
698 | iov[i].iov_base, | |
699 | iov[i].iov_len, | |
700 | 0); | |
701 | if (ret < 0) { | |
b16a44e1 | 702 | if (errno == EAGAIN) { |
559607ea DB |
703 | if (done) { |
704 | return done; | |
705 | } else { | |
706 | return QIO_CHANNEL_ERR_BLOCK; | |
707 | } | |
b16a44e1 | 708 | } else if (errno == EINTR) { |
559607ea DB |
709 | goto retry; |
710 | } else { | |
b16a44e1 | 711 | error_setg_errno(errp, errno, |
559607ea DB |
712 | "Unable to write to socket"); |
713 | return -1; | |
714 | } | |
715 | } | |
716 | done += ret; | |
717 | if (ret < iov[i].iov_len) { | |
718 | return done; | |
719 | } | |
720 | } | |
721 | ||
722 | return done; | |
723 | } | |
724 | #endif /* WIN32 */ | |
725 | ||
2bc58ffc LB |
726 | |
727 | #ifdef QEMU_MSG_ZEROCOPY | |
728 | static int qio_channel_socket_flush(QIOChannel *ioc, | |
729 | Error **errp) | |
730 | { | |
731 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
732 | struct msghdr msg = {}; | |
733 | struct sock_extended_err *serr; | |
734 | struct cmsghdr *cm; | |
735 | char control[CMSG_SPACE(sizeof(*serr))]; | |
736 | int received; | |
927f93e0 LB |
737 | int ret; |
738 | ||
739 | if (sioc->zero_copy_queued == sioc->zero_copy_sent) { | |
740 | return 0; | |
741 | } | |
2bc58ffc LB |
742 | |
743 | msg.msg_control = control; | |
744 | msg.msg_controllen = sizeof(control); | |
745 | memset(control, 0, sizeof(control)); | |
746 | ||
927f93e0 LB |
747 | ret = 1; |
748 | ||
2bc58ffc LB |
749 | while (sioc->zero_copy_sent < sioc->zero_copy_queued) { |
750 | received = recvmsg(sioc->fd, &msg, MSG_ERRQUEUE); | |
751 | if (received < 0) { | |
752 | switch (errno) { | |
753 | case EAGAIN: | |
754 | /* Nothing on errqueue, wait until something is available */ | |
755 | qio_channel_wait(ioc, G_IO_ERR); | |
756 | continue; | |
757 | case EINTR: | |
758 | continue; | |
759 | default: | |
760 | error_setg_errno(errp, errno, | |
761 | "Unable to read errqueue"); | |
762 | return -1; | |
763 | } | |
764 | } | |
765 | ||
766 | cm = CMSG_FIRSTHDR(&msg); | |
5258a7e2 LB |
767 | if (cm->cmsg_level != SOL_IP && cm->cmsg_type != IP_RECVERR && |
768 | cm->cmsg_level != SOL_IPV6 && cm->cmsg_type != IPV6_RECVERR) { | |
2bc58ffc LB |
769 | error_setg_errno(errp, EPROTOTYPE, |
770 | "Wrong cmsg in errqueue"); | |
771 | return -1; | |
772 | } | |
773 | ||
774 | serr = (void *) CMSG_DATA(cm); | |
775 | if (serr->ee_errno != SO_EE_ORIGIN_NONE) { | |
776 | error_setg_errno(errp, serr->ee_errno, | |
777 | "Error on socket"); | |
778 | return -1; | |
779 | } | |
780 | if (serr->ee_origin != SO_EE_ORIGIN_ZEROCOPY) { | |
781 | error_setg_errno(errp, serr->ee_origin, | |
782 | "Error not from zero copy"); | |
783 | return -1; | |
784 | } | |
785 | ||
786 | /* No errors, count successfully finished sendmsg()*/ | |
787 | sioc->zero_copy_sent += serr->ee_data - serr->ee_info + 1; | |
788 | ||
789 | /* If any sendmsg() succeeded using zero copy, return 0 at the end */ | |
790 | if (serr->ee_code != SO_EE_CODE_ZEROCOPY_COPIED) { | |
791 | ret = 0; | |
792 | } | |
793 | } | |
794 | ||
795 | return ret; | |
796 | } | |
797 | ||
798 | #endif /* QEMU_MSG_ZEROCOPY */ | |
799 | ||
559607ea DB |
800 | static int |
801 | qio_channel_socket_set_blocking(QIOChannel *ioc, | |
802 | bool enabled, | |
803 | Error **errp) | |
804 | { | |
805 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
806 | ||
807 | if (enabled) { | |
ff5927ba | 808 | qemu_socket_set_block(sioc->fd); |
559607ea | 809 | } else { |
ff5927ba | 810 | qemu_socket_set_nonblock(sioc->fd); |
559607ea DB |
811 | } |
812 | return 0; | |
813 | } | |
814 | ||
815 | ||
816 | static void | |
817 | qio_channel_socket_set_delay(QIOChannel *ioc, | |
818 | bool enabled) | |
819 | { | |
820 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
821 | int v = enabled ? 0 : 1; | |
822 | ||
e7b79428 MAL |
823 | setsockopt(sioc->fd, |
824 | IPPROTO_TCP, TCP_NODELAY, | |
825 | &v, sizeof(v)); | |
559607ea DB |
826 | } |
827 | ||
828 | ||
829 | static void | |
830 | qio_channel_socket_set_cork(QIOChannel *ioc, | |
831 | bool enabled) | |
832 | { | |
833 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
834 | int v = enabled ? 1 : 0; | |
835 | ||
836 | socket_set_cork(sioc->fd, v); | |
837 | } | |
838 | ||
839 | ||
840 | static int | |
841 | qio_channel_socket_close(QIOChannel *ioc, | |
842 | Error **errp) | |
843 | { | |
844 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
d66f78e1 | 845 | int rc = 0; |
fdceb4ab | 846 | Error *err = NULL; |
559607ea | 847 | |
a5897205 PB |
848 | if (sioc->fd != -1) { |
849 | #ifdef WIN32 | |
a4aafea2 | 850 | qemu_socket_unselect(sioc->fd, NULL); |
a5897205 | 851 | #endif |
73564c40 DB |
852 | if (qio_channel_has_feature(ioc, QIO_CHANNEL_FEATURE_LISTEN)) { |
853 | socket_listen_cleanup(sioc->fd, errp); | |
854 | } | |
855 | ||
25657fc6 | 856 | if (close(sioc->fd) < 0) { |
a5897205 | 857 | sioc->fd = -1; |
fdceb4ab MA |
858 | error_setg_errno(&err, errno, "Unable to close socket"); |
859 | error_propagate(errp, err); | |
a5897205 PB |
860 | return -1; |
861 | } | |
559607ea | 862 | sioc->fd = -1; |
559607ea | 863 | } |
d66f78e1 | 864 | return rc; |
559607ea DB |
865 | } |
866 | ||
867 | static int | |
868 | qio_channel_socket_shutdown(QIOChannel *ioc, | |
869 | QIOChannelShutdown how, | |
870 | Error **errp) | |
871 | { | |
872 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
873 | int sockhow; | |
874 | ||
875 | switch (how) { | |
876 | case QIO_CHANNEL_SHUTDOWN_READ: | |
877 | sockhow = SHUT_RD; | |
878 | break; | |
879 | case QIO_CHANNEL_SHUTDOWN_WRITE: | |
880 | sockhow = SHUT_WR; | |
881 | break; | |
882 | case QIO_CHANNEL_SHUTDOWN_BOTH: | |
883 | default: | |
884 | sockhow = SHUT_RDWR; | |
885 | break; | |
886 | } | |
887 | ||
888 | if (shutdown(sioc->fd, sockhow) < 0) { | |
b16a44e1 | 889 | error_setg_errno(errp, errno, |
559607ea DB |
890 | "Unable to shutdown socket"); |
891 | return -1; | |
892 | } | |
893 | return 0; | |
894 | } | |
895 | ||
bf88c124 | 896 | static void qio_channel_socket_set_aio_fd_handler(QIOChannel *ioc, |
06e0f098 | 897 | AioContext *read_ctx, |
bf88c124 | 898 | IOHandler *io_read, |
06e0f098 | 899 | AioContext *write_ctx, |
bf88c124 PB |
900 | IOHandler *io_write, |
901 | void *opaque) | |
902 | { | |
903 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
06e0f098 SH |
904 | |
905 | qio_channel_util_set_aio_fd_handler(sioc->fd, read_ctx, io_read, | |
906 | sioc->fd, write_ctx, io_write, | |
907 | opaque); | |
bf88c124 PB |
908 | } |
909 | ||
559607ea DB |
910 | static GSource *qio_channel_socket_create_watch(QIOChannel *ioc, |
911 | GIOCondition condition) | |
912 | { | |
913 | QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc); | |
b83b68a0 PB |
914 | return qio_channel_create_socket_watch(ioc, |
915 | sioc->fd, | |
916 | condition); | |
559607ea DB |
917 | } |
918 | ||
919 | static void qio_channel_socket_class_init(ObjectClass *klass, | |
920 | void *class_data G_GNUC_UNUSED) | |
921 | { | |
922 | QIOChannelClass *ioc_klass = QIO_CHANNEL_CLASS(klass); | |
923 | ||
924 | ioc_klass->io_writev = qio_channel_socket_writev; | |
925 | ioc_klass->io_readv = qio_channel_socket_readv; | |
926 | ioc_klass->io_set_blocking = qio_channel_socket_set_blocking; | |
927 | ioc_klass->io_close = qio_channel_socket_close; | |
928 | ioc_klass->io_shutdown = qio_channel_socket_shutdown; | |
929 | ioc_klass->io_set_cork = qio_channel_socket_set_cork; | |
930 | ioc_klass->io_set_delay = qio_channel_socket_set_delay; | |
931 | ioc_klass->io_create_watch = qio_channel_socket_create_watch; | |
bf88c124 | 932 | ioc_klass->io_set_aio_fd_handler = qio_channel_socket_set_aio_fd_handler; |
2bc58ffc LB |
933 | #ifdef QEMU_MSG_ZEROCOPY |
934 | ioc_klass->io_flush = qio_channel_socket_flush; | |
935 | #endif | |
559607ea DB |
936 | } |
937 | ||
938 | static const TypeInfo qio_channel_socket_info = { | |
939 | .parent = TYPE_QIO_CHANNEL, | |
940 | .name = TYPE_QIO_CHANNEL_SOCKET, | |
941 | .instance_size = sizeof(QIOChannelSocket), | |
942 | .instance_init = qio_channel_socket_init, | |
943 | .instance_finalize = qio_channel_socket_finalize, | |
944 | .class_init = qio_channel_socket_class_init, | |
945 | }; | |
946 | ||
947 | static void qio_channel_socket_register_types(void) | |
948 | { | |
949 | type_register_static(&qio_channel_socket_info); | |
950 | } | |
951 | ||
952 | type_init(qio_channel_socket_register_types); |