MAINTAINERS: Merge sections CPU, NUMA into Machine core
[qemu/kevin.git] / net / socket.c
blobc92354049bcaee7581045174a503365eb27a4c40
1 /*
2 * QEMU System Emulator
4 * Copyright (c) 2003-2008 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
24 #include "qemu/osdep.h"
26 #include "net/net.h"
27 #include "clients.h"
28 #include "monitor/monitor.h"
29 #include "qapi/error.h"
30 #include "qemu-common.h"
31 #include "qemu/error-report.h"
32 #include "qemu/option.h"
33 #include "qemu/sockets.h"
34 #include "qemu/iov.h"
35 #include "qemu/main-loop.h"
37 typedef struct NetSocketState {
38 NetClientState nc;
39 int listen_fd;
40 int fd;
41 SocketReadState rs;
42 unsigned int send_index; /* number of bytes sent (only SOCK_STREAM) */
43 struct sockaddr_in dgram_dst; /* contains inet host and port destination iff connectionless (SOCK_DGRAM) */
44 IOHandler *send_fn; /* differs between SOCK_STREAM/SOCK_DGRAM */
45 bool read_poll; /* waiting to receive data? */
46 bool write_poll; /* waiting to transmit data? */
47 } NetSocketState;
49 static void net_socket_accept(void *opaque);
50 static void net_socket_writable(void *opaque);
52 static void net_socket_update_fd_handler(NetSocketState *s)
54 qemu_set_fd_handler(s->fd,
55 s->read_poll ? s->send_fn : NULL,
56 s->write_poll ? net_socket_writable : NULL,
57 s);
60 static void net_socket_read_poll(NetSocketState *s, bool enable)
62 s->read_poll = enable;
63 net_socket_update_fd_handler(s);
66 static void net_socket_write_poll(NetSocketState *s, bool enable)
68 s->write_poll = enable;
69 net_socket_update_fd_handler(s);
72 static void net_socket_writable(void *opaque)
74 NetSocketState *s = opaque;
76 net_socket_write_poll(s, false);
78 qemu_flush_queued_packets(&s->nc);
81 static ssize_t net_socket_receive(NetClientState *nc, const uint8_t *buf, size_t size)
83 NetSocketState *s = DO_UPCAST(NetSocketState, nc, nc);
84 uint32_t len = htonl(size);
85 struct iovec iov[] = {
87 .iov_base = &len,
88 .iov_len = sizeof(len),
89 }, {
90 .iov_base = (void *)buf,
91 .iov_len = size,
94 size_t remaining;
95 ssize_t ret;
97 remaining = iov_size(iov, 2) - s->send_index;
98 ret = iov_send(s->fd, iov, 2, s->send_index, remaining);
100 if (ret == -1 && errno == EAGAIN) {
101 ret = 0; /* handled further down */
103 if (ret == -1) {
104 s->send_index = 0;
105 return -errno;
107 if (ret < (ssize_t)remaining) {
108 s->send_index += ret;
109 net_socket_write_poll(s, true);
110 return 0;
112 s->send_index = 0;
113 return size;
116 static ssize_t net_socket_receive_dgram(NetClientState *nc, const uint8_t *buf, size_t size)
118 NetSocketState *s = DO_UPCAST(NetSocketState, nc, nc);
119 ssize_t ret;
121 do {
122 if (s->dgram_dst.sin_family != AF_UNIX) {
123 ret = qemu_sendto(s->fd, buf, size, 0,
124 (struct sockaddr *)&s->dgram_dst,
125 sizeof(s->dgram_dst));
126 } else {
127 ret = send(s->fd, buf, size, 0);
129 } while (ret == -1 && errno == EINTR);
131 if (ret == -1 && errno == EAGAIN) {
132 net_socket_write_poll(s, true);
133 return 0;
135 return ret;
138 static void net_socket_send_completed(NetClientState *nc, ssize_t len)
140 NetSocketState *s = DO_UPCAST(NetSocketState, nc, nc);
142 if (!s->read_poll) {
143 net_socket_read_poll(s, true);
147 static void net_socket_rs_finalize(SocketReadState *rs)
149 NetSocketState *s = container_of(rs, NetSocketState, rs);
151 if (qemu_send_packet_async(&s->nc, rs->buf,
152 rs->packet_len,
153 net_socket_send_completed) == 0) {
154 net_socket_read_poll(s, false);
158 static void net_socket_send(void *opaque)
160 NetSocketState *s = opaque;
161 int size;
162 int ret;
163 uint8_t buf1[NET_BUFSIZE];
164 const uint8_t *buf;
166 size = qemu_recv(s->fd, buf1, sizeof(buf1), 0);
167 if (size < 0) {
168 if (errno != EWOULDBLOCK)
169 goto eoc;
170 } else if (size == 0) {
171 /* end of connection */
172 eoc:
173 net_socket_read_poll(s, false);
174 net_socket_write_poll(s, false);
175 if (s->listen_fd != -1) {
176 qemu_set_fd_handler(s->listen_fd, net_socket_accept, NULL, s);
178 closesocket(s->fd);
180 s->fd = -1;
181 net_socket_rs_init(&s->rs, net_socket_rs_finalize, false);
182 s->nc.link_down = true;
183 memset(s->nc.info_str, 0, sizeof(s->nc.info_str));
185 return;
187 buf = buf1;
189 ret = net_fill_rstate(&s->rs, buf, size);
191 if (ret == -1) {
192 goto eoc;
196 static void net_socket_send_dgram(void *opaque)
198 NetSocketState *s = opaque;
199 int size;
201 size = qemu_recv(s->fd, s->rs.buf, sizeof(s->rs.buf), 0);
202 if (size < 0)
203 return;
204 if (size == 0) {
205 /* end of connection */
206 net_socket_read_poll(s, false);
207 net_socket_write_poll(s, false);
208 return;
210 if (qemu_send_packet_async(&s->nc, s->rs.buf, size,
211 net_socket_send_completed) == 0) {
212 net_socket_read_poll(s, false);
216 static int net_socket_mcast_create(struct sockaddr_in *mcastaddr,
217 struct in_addr *localaddr,
218 Error **errp)
220 struct ip_mreq imr;
221 int fd;
222 int val, ret;
223 #ifdef __OpenBSD__
224 unsigned char loop;
225 #else
226 int loop;
227 #endif
229 if (!IN_MULTICAST(ntohl(mcastaddr->sin_addr.s_addr))) {
230 error_setg(errp, "specified mcastaddr %s (0x%08x) "
231 "does not contain a multicast address",
232 inet_ntoa(mcastaddr->sin_addr),
233 (int)ntohl(mcastaddr->sin_addr.s_addr));
234 return -1;
237 fd = qemu_socket(PF_INET, SOCK_DGRAM, 0);
238 if (fd < 0) {
239 error_setg_errno(errp, errno, "can't create datagram socket");
240 return -1;
243 /* Allow multiple sockets to bind the same multicast ip and port by setting
244 * SO_REUSEADDR. This is the only situation where SO_REUSEADDR should be set
245 * on windows. Use socket_set_fast_reuse otherwise as it sets SO_REUSEADDR
246 * only on posix systems.
248 val = 1;
249 ret = qemu_setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof(val));
250 if (ret < 0) {
251 error_setg_errno(errp, errno,
252 "can't set socket option SO_REUSEADDR");
253 goto fail;
256 ret = bind(fd, (struct sockaddr *)mcastaddr, sizeof(*mcastaddr));
257 if (ret < 0) {
258 error_setg_errno(errp, errno, "can't bind ip=%s to socket",
259 inet_ntoa(mcastaddr->sin_addr));
260 goto fail;
263 /* Add host to multicast group */
264 imr.imr_multiaddr = mcastaddr->sin_addr;
265 if (localaddr) {
266 imr.imr_interface = *localaddr;
267 } else {
268 imr.imr_interface.s_addr = htonl(INADDR_ANY);
271 ret = qemu_setsockopt(fd, IPPROTO_IP, IP_ADD_MEMBERSHIP,
272 &imr, sizeof(struct ip_mreq));
273 if (ret < 0) {
274 error_setg_errno(errp, errno,
275 "can't add socket to multicast group %s",
276 inet_ntoa(imr.imr_multiaddr));
277 goto fail;
280 /* Force mcast msgs to loopback (eg. several QEMUs in same host */
281 loop = 1;
282 ret = qemu_setsockopt(fd, IPPROTO_IP, IP_MULTICAST_LOOP,
283 &loop, sizeof(loop));
284 if (ret < 0) {
285 error_setg_errno(errp, errno,
286 "can't force multicast message to loopback");
287 goto fail;
290 /* If a bind address is given, only send packets from that address */
291 if (localaddr != NULL) {
292 ret = qemu_setsockopt(fd, IPPROTO_IP, IP_MULTICAST_IF,
293 localaddr, sizeof(*localaddr));
294 if (ret < 0) {
295 error_setg_errno(errp, errno,
296 "can't set the default network send interface");
297 goto fail;
301 qemu_set_nonblock(fd);
302 return fd;
303 fail:
304 if (fd >= 0)
305 closesocket(fd);
306 return -1;
309 static void net_socket_cleanup(NetClientState *nc)
311 NetSocketState *s = DO_UPCAST(NetSocketState, nc, nc);
312 if (s->fd != -1) {
313 net_socket_read_poll(s, false);
314 net_socket_write_poll(s, false);
315 close(s->fd);
316 s->fd = -1;
318 if (s->listen_fd != -1) {
319 qemu_set_fd_handler(s->listen_fd, NULL, NULL, NULL);
320 closesocket(s->listen_fd);
321 s->listen_fd = -1;
325 static NetClientInfo net_dgram_socket_info = {
326 .type = NET_CLIENT_DRIVER_SOCKET,
327 .size = sizeof(NetSocketState),
328 .receive = net_socket_receive_dgram,
329 .cleanup = net_socket_cleanup,
332 static NetSocketState *net_socket_fd_init_dgram(NetClientState *peer,
333 const char *model,
334 const char *name,
335 int fd, int is_connected,
336 const char *mcast,
337 Error **errp)
339 struct sockaddr_in saddr;
340 int newfd;
341 NetClientState *nc;
342 NetSocketState *s;
343 SocketAddress *sa;
344 SocketAddressType sa_type;
346 sa = socket_local_address(fd, errp);
347 if (!sa) {
348 return NULL;
350 sa_type = sa->type;
351 qapi_free_SocketAddress(sa);
353 /* fd passed: multicast: "learn" dgram_dst address from bound address and save it
354 * Because this may be "shared" socket from a "master" process, datagrams would be recv()
355 * by ONLY ONE process: we must "clone" this dgram socket --jjo
358 if (is_connected && mcast != NULL) {
359 if (parse_host_port(&saddr, mcast, errp) < 0) {
360 goto err;
362 /* must be bound */
363 if (saddr.sin_addr.s_addr == 0) {
364 error_setg(errp, "can't setup multicast destination address");
365 goto err;
367 /* clone dgram socket */
368 newfd = net_socket_mcast_create(&saddr, NULL, errp);
369 if (newfd < 0) {
370 goto err;
372 /* clone newfd to fd, close newfd */
373 dup2(newfd, fd);
374 close(newfd);
378 nc = qemu_new_net_client(&net_dgram_socket_info, peer, model, name);
380 s = DO_UPCAST(NetSocketState, nc, nc);
382 s->fd = fd;
383 s->listen_fd = -1;
384 s->send_fn = net_socket_send_dgram;
385 net_socket_rs_init(&s->rs, net_socket_rs_finalize, false);
386 net_socket_read_poll(s, true);
388 /* mcast: save bound address as dst */
389 if (is_connected && mcast != NULL) {
390 s->dgram_dst = saddr;
391 snprintf(nc->info_str, sizeof(nc->info_str),
392 "socket: fd=%d (cloned mcast=%s:%d)",
393 fd, inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
394 } else {
395 if (sa_type == SOCKET_ADDRESS_TYPE_UNIX) {
396 s->dgram_dst.sin_family = AF_UNIX;
399 snprintf(nc->info_str, sizeof(nc->info_str),
400 "socket: fd=%d %s", fd, SocketAddressType_str(sa_type));
403 return s;
405 err:
406 closesocket(fd);
407 return NULL;
410 static void net_socket_connect(void *opaque)
412 NetSocketState *s = opaque;
413 s->send_fn = net_socket_send;
414 net_socket_read_poll(s, true);
417 static NetClientInfo net_socket_info = {
418 .type = NET_CLIENT_DRIVER_SOCKET,
419 .size = sizeof(NetSocketState),
420 .receive = net_socket_receive,
421 .cleanup = net_socket_cleanup,
424 static NetSocketState *net_socket_fd_init_stream(NetClientState *peer,
425 const char *model,
426 const char *name,
427 int fd, int is_connected)
429 NetClientState *nc;
430 NetSocketState *s;
432 nc = qemu_new_net_client(&net_socket_info, peer, model, name);
434 snprintf(nc->info_str, sizeof(nc->info_str), "socket: fd=%d", fd);
436 s = DO_UPCAST(NetSocketState, nc, nc);
438 s->fd = fd;
439 s->listen_fd = -1;
440 net_socket_rs_init(&s->rs, net_socket_rs_finalize, false);
442 /* Disable Nagle algorithm on TCP sockets to reduce latency */
443 socket_set_nodelay(fd);
445 if (is_connected) {
446 net_socket_connect(s);
447 } else {
448 qemu_set_fd_handler(s->fd, NULL, net_socket_connect, s);
450 return s;
453 static NetSocketState *net_socket_fd_init(NetClientState *peer,
454 const char *model, const char *name,
455 int fd, int is_connected,
456 const char *mc, Error **errp)
458 int so_type = -1, optlen=sizeof(so_type);
460 if(getsockopt(fd, SOL_SOCKET, SO_TYPE, (char *)&so_type,
461 (socklen_t *)&optlen)< 0) {
462 error_setg(errp, "can't get socket option SO_TYPE");
463 closesocket(fd);
464 return NULL;
466 switch(so_type) {
467 case SOCK_DGRAM:
468 return net_socket_fd_init_dgram(peer, model, name, fd, is_connected,
469 mc, errp);
470 case SOCK_STREAM:
471 return net_socket_fd_init_stream(peer, model, name, fd, is_connected);
472 default:
473 error_setg(errp, "socket type=%d for fd=%d must be either"
474 " SOCK_DGRAM or SOCK_STREAM", so_type, fd);
475 closesocket(fd);
477 return NULL;
480 static void net_socket_accept(void *opaque)
482 NetSocketState *s = opaque;
483 struct sockaddr_in saddr;
484 socklen_t len;
485 int fd;
487 for(;;) {
488 len = sizeof(saddr);
489 fd = qemu_accept(s->listen_fd, (struct sockaddr *)&saddr, &len);
490 if (fd < 0 && errno != EINTR) {
491 return;
492 } else if (fd >= 0) {
493 qemu_set_fd_handler(s->listen_fd, NULL, NULL, NULL);
494 break;
498 s->fd = fd;
499 s->nc.link_down = false;
500 net_socket_connect(s);
501 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
502 "socket: connection from %s:%d",
503 inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
506 static int net_socket_listen_init(NetClientState *peer,
507 const char *model,
508 const char *name,
509 const char *host_str,
510 Error **errp)
512 NetClientState *nc;
513 NetSocketState *s;
514 struct sockaddr_in saddr;
515 int fd, ret;
517 if (parse_host_port(&saddr, host_str, errp) < 0) {
518 return -1;
521 fd = qemu_socket(PF_INET, SOCK_STREAM, 0);
522 if (fd < 0) {
523 error_setg_errno(errp, errno, "can't create stream socket");
524 return -1;
526 qemu_set_nonblock(fd);
528 socket_set_fast_reuse(fd);
530 ret = bind(fd, (struct sockaddr *)&saddr, sizeof(saddr));
531 if (ret < 0) {
532 error_setg_errno(errp, errno, "can't bind ip=%s to socket",
533 inet_ntoa(saddr.sin_addr));
534 closesocket(fd);
535 return -1;
537 ret = listen(fd, 0);
538 if (ret < 0) {
539 error_setg_errno(errp, errno, "can't listen on socket");
540 closesocket(fd);
541 return -1;
544 nc = qemu_new_net_client(&net_socket_info, peer, model, name);
545 s = DO_UPCAST(NetSocketState, nc, nc);
546 s->fd = -1;
547 s->listen_fd = fd;
548 s->nc.link_down = true;
549 net_socket_rs_init(&s->rs, net_socket_rs_finalize, false);
551 qemu_set_fd_handler(s->listen_fd, net_socket_accept, NULL, s);
552 return 0;
555 static int net_socket_connect_init(NetClientState *peer,
556 const char *model,
557 const char *name,
558 const char *host_str,
559 Error **errp)
561 NetSocketState *s;
562 int fd, connected, ret;
563 struct sockaddr_in saddr;
565 if (parse_host_port(&saddr, host_str, errp) < 0) {
566 return -1;
569 fd = qemu_socket(PF_INET, SOCK_STREAM, 0);
570 if (fd < 0) {
571 error_setg_errno(errp, errno, "can't create stream socket");
572 return -1;
574 qemu_set_nonblock(fd);
576 connected = 0;
577 for(;;) {
578 ret = connect(fd, (struct sockaddr *)&saddr, sizeof(saddr));
579 if (ret < 0) {
580 if (errno == EINTR || errno == EWOULDBLOCK) {
581 /* continue */
582 } else if (errno == EINPROGRESS ||
583 errno == EALREADY ||
584 errno == EINVAL) {
585 break;
586 } else {
587 error_setg_errno(errp, errno, "can't connect socket");
588 closesocket(fd);
589 return -1;
591 } else {
592 connected = 1;
593 break;
596 s = net_socket_fd_init(peer, model, name, fd, connected, NULL, errp);
597 if (!s) {
598 return -1;
601 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
602 "socket: connect to %s:%d",
603 inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
604 return 0;
607 static int net_socket_mcast_init(NetClientState *peer,
608 const char *model,
609 const char *name,
610 const char *host_str,
611 const char *localaddr_str,
612 Error **errp)
614 NetSocketState *s;
615 int fd;
616 struct sockaddr_in saddr;
617 struct in_addr localaddr, *param_localaddr;
619 if (parse_host_port(&saddr, host_str, errp) < 0) {
620 return -1;
623 if (localaddr_str != NULL) {
624 if (inet_aton(localaddr_str, &localaddr) == 0) {
625 error_setg(errp, "localaddr '%s' is not a valid IPv4 address",
626 localaddr_str);
627 return -1;
629 param_localaddr = &localaddr;
630 } else {
631 param_localaddr = NULL;
634 fd = net_socket_mcast_create(&saddr, param_localaddr, errp);
635 if (fd < 0) {
636 return -1;
639 s = net_socket_fd_init(peer, model, name, fd, 0, NULL, errp);
640 if (!s) {
641 return -1;
644 s->dgram_dst = saddr;
646 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
647 "socket: mcast=%s:%d",
648 inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
649 return 0;
653 static int net_socket_udp_init(NetClientState *peer,
654 const char *model,
655 const char *name,
656 const char *rhost,
657 const char *lhost,
658 Error **errp)
660 NetSocketState *s;
661 int fd, ret;
662 struct sockaddr_in laddr, raddr;
664 if (parse_host_port(&laddr, lhost, errp) < 0) {
665 return -1;
668 if (parse_host_port(&raddr, rhost, errp) < 0) {
669 return -1;
672 fd = qemu_socket(PF_INET, SOCK_DGRAM, 0);
673 if (fd < 0) {
674 error_setg_errno(errp, errno, "can't create datagram socket");
675 return -1;
678 ret = socket_set_fast_reuse(fd);
679 if (ret < 0) {
680 error_setg_errno(errp, errno,
681 "can't set socket option SO_REUSEADDR");
682 closesocket(fd);
683 return -1;
685 ret = bind(fd, (struct sockaddr *)&laddr, sizeof(laddr));
686 if (ret < 0) {
687 error_setg_errno(errp, errno, "can't bind ip=%s to socket",
688 inet_ntoa(laddr.sin_addr));
689 closesocket(fd);
690 return -1;
692 qemu_set_nonblock(fd);
694 s = net_socket_fd_init(peer, model, name, fd, 0, NULL, errp);
695 if (!s) {
696 return -1;
699 s->dgram_dst = raddr;
701 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
702 "socket: udp=%s:%d",
703 inet_ntoa(raddr.sin_addr), ntohs(raddr.sin_port));
704 return 0;
707 int net_init_socket(const Netdev *netdev, const char *name,
708 NetClientState *peer, Error **errp)
710 const NetdevSocketOptions *sock;
712 assert(netdev->type == NET_CLIENT_DRIVER_SOCKET);
713 sock = &netdev->u.socket;
715 if (sock->has_fd + sock->has_listen + sock->has_connect + sock->has_mcast +
716 sock->has_udp != 1) {
717 error_setg(errp, "exactly one of listen=, connect=, mcast= or udp="
718 " is required");
719 return -1;
722 if (sock->has_localaddr && !sock->has_mcast && !sock->has_udp) {
723 error_setg(errp, "localaddr= is only valid with mcast= or udp=");
724 return -1;
727 if (sock->has_fd) {
728 int fd;
730 fd = monitor_fd_param(cur_mon, sock->fd, errp);
731 if (fd == -1) {
732 return -1;
734 qemu_set_nonblock(fd);
735 if (!net_socket_fd_init(peer, "socket", name, fd, 1, sock->mcast,
736 errp)) {
737 return -1;
739 return 0;
742 if (sock->has_listen) {
743 if (net_socket_listen_init(peer, "socket", name, sock->listen, errp)
744 < 0) {
745 return -1;
747 return 0;
750 if (sock->has_connect) {
751 if (net_socket_connect_init(peer, "socket", name, sock->connect, errp)
752 < 0) {
753 return -1;
755 return 0;
758 if (sock->has_mcast) {
759 /* if sock->localaddr is missing, it has been initialized to "all bits
760 * zero" */
761 if (net_socket_mcast_init(peer, "socket", name, sock->mcast,
762 sock->localaddr, errp) < 0) {
763 return -1;
765 return 0;
768 assert(sock->has_udp);
769 if (!sock->has_localaddr) {
770 error_setg(errp, "localaddr= is mandatory with udp=");
771 return -1;
773 if (net_socket_udp_init(peer, "socket", name, sock->udp, sock->localaddr,
774 errp) < 0) {
775 return -1;
777 return 0;