qemu-options: Rewrite -numa documentation
[qemu/kevin.git] / net / socket.c
blobfe3547b0184d04b1a958b435352bc18f2a6bffb0
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 ret = qemu_sendto(s->fd, buf, size, 0,
123 (struct sockaddr *)&s->dgram_dst,
124 sizeof(s->dgram_dst));
125 } while (ret == -1 && errno == EINTR);
127 if (ret == -1 && errno == EAGAIN) {
128 net_socket_write_poll(s, true);
129 return 0;
131 return ret;
134 static void net_socket_send_completed(NetClientState *nc, ssize_t len)
136 NetSocketState *s = DO_UPCAST(NetSocketState, nc, nc);
138 if (!s->read_poll) {
139 net_socket_read_poll(s, true);
143 static void net_socket_rs_finalize(SocketReadState *rs)
145 NetSocketState *s = container_of(rs, NetSocketState, rs);
147 if (qemu_send_packet_async(&s->nc, rs->buf,
148 rs->packet_len,
149 net_socket_send_completed) == 0) {
150 net_socket_read_poll(s, false);
154 static void net_socket_send(void *opaque)
156 NetSocketState *s = opaque;
157 int size;
158 int ret;
159 uint8_t buf1[NET_BUFSIZE];
160 const uint8_t *buf;
162 size = qemu_recv(s->fd, buf1, sizeof(buf1), 0);
163 if (size < 0) {
164 if (errno != EWOULDBLOCK)
165 goto eoc;
166 } else if (size == 0) {
167 /* end of connection */
168 eoc:
169 net_socket_read_poll(s, false);
170 net_socket_write_poll(s, false);
171 if (s->listen_fd != -1) {
172 qemu_set_fd_handler(s->listen_fd, net_socket_accept, NULL, s);
174 closesocket(s->fd);
176 s->fd = -1;
177 net_socket_rs_init(&s->rs, net_socket_rs_finalize);
178 s->nc.link_down = true;
179 memset(s->nc.info_str, 0, sizeof(s->nc.info_str));
181 return;
183 buf = buf1;
185 ret = net_fill_rstate(&s->rs, buf, size);
187 if (ret == -1) {
188 goto eoc;
192 static void net_socket_send_dgram(void *opaque)
194 NetSocketState *s = opaque;
195 int size;
197 size = qemu_recv(s->fd, s->rs.buf, sizeof(s->rs.buf), 0);
198 if (size < 0)
199 return;
200 if (size == 0) {
201 /* end of connection */
202 net_socket_read_poll(s, false);
203 net_socket_write_poll(s, false);
204 return;
206 if (qemu_send_packet_async(&s->nc, s->rs.buf, size,
207 net_socket_send_completed) == 0) {
208 net_socket_read_poll(s, false);
212 static int net_socket_mcast_create(struct sockaddr_in *mcastaddr, struct in_addr *localaddr)
214 struct ip_mreq imr;
215 int fd;
216 int val, ret;
217 #ifdef __OpenBSD__
218 unsigned char loop;
219 #else
220 int loop;
221 #endif
223 if (!IN_MULTICAST(ntohl(mcastaddr->sin_addr.s_addr))) {
224 fprintf(stderr, "qemu: error: specified mcastaddr \"%s\" (0x%08x) "
225 "does not contain a multicast address\n",
226 inet_ntoa(mcastaddr->sin_addr),
227 (int)ntohl(mcastaddr->sin_addr.s_addr));
228 return -1;
231 fd = qemu_socket(PF_INET, SOCK_DGRAM, 0);
232 if (fd < 0) {
233 perror("socket(PF_INET, SOCK_DGRAM)");
234 return -1;
237 /* Allow multiple sockets to bind the same multicast ip and port by setting
238 * SO_REUSEADDR. This is the only situation where SO_REUSEADDR should be set
239 * on windows. Use socket_set_fast_reuse otherwise as it sets SO_REUSEADDR
240 * only on posix systems.
242 val = 1;
243 ret = qemu_setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof(val));
244 if (ret < 0) {
245 perror("setsockopt(SOL_SOCKET, SO_REUSEADDR)");
246 goto fail;
249 ret = bind(fd, (struct sockaddr *)mcastaddr, sizeof(*mcastaddr));
250 if (ret < 0) {
251 perror("bind");
252 goto fail;
255 /* Add host to multicast group */
256 imr.imr_multiaddr = mcastaddr->sin_addr;
257 if (localaddr) {
258 imr.imr_interface = *localaddr;
259 } else {
260 imr.imr_interface.s_addr = htonl(INADDR_ANY);
263 ret = qemu_setsockopt(fd, IPPROTO_IP, IP_ADD_MEMBERSHIP,
264 &imr, sizeof(struct ip_mreq));
265 if (ret < 0) {
266 perror("setsockopt(IP_ADD_MEMBERSHIP)");
267 goto fail;
270 /* Force mcast msgs to loopback (eg. several QEMUs in same host */
271 loop = 1;
272 ret = qemu_setsockopt(fd, IPPROTO_IP, IP_MULTICAST_LOOP,
273 &loop, sizeof(loop));
274 if (ret < 0) {
275 perror("setsockopt(SOL_IP, IP_MULTICAST_LOOP)");
276 goto fail;
279 /* If a bind address is given, only send packets from that address */
280 if (localaddr != NULL) {
281 ret = qemu_setsockopt(fd, IPPROTO_IP, IP_MULTICAST_IF,
282 localaddr, sizeof(*localaddr));
283 if (ret < 0) {
284 perror("setsockopt(IP_MULTICAST_IF)");
285 goto fail;
289 qemu_set_nonblock(fd);
290 return fd;
291 fail:
292 if (fd >= 0)
293 closesocket(fd);
294 return -1;
297 static void net_socket_cleanup(NetClientState *nc)
299 NetSocketState *s = DO_UPCAST(NetSocketState, nc, nc);
300 if (s->fd != -1) {
301 net_socket_read_poll(s, false);
302 net_socket_write_poll(s, false);
303 close(s->fd);
304 s->fd = -1;
306 if (s->listen_fd != -1) {
307 qemu_set_fd_handler(s->listen_fd, NULL, NULL, NULL);
308 closesocket(s->listen_fd);
309 s->listen_fd = -1;
313 static NetClientInfo net_dgram_socket_info = {
314 .type = NET_CLIENT_DRIVER_SOCKET,
315 .size = sizeof(NetSocketState),
316 .receive = net_socket_receive_dgram,
317 .cleanup = net_socket_cleanup,
320 static NetSocketState *net_socket_fd_init_dgram(NetClientState *peer,
321 const char *model,
322 const char *name,
323 int fd, int is_connected)
325 struct sockaddr_in saddr;
326 int newfd;
327 socklen_t saddr_len = sizeof(saddr);
328 NetClientState *nc;
329 NetSocketState *s;
331 /* fd passed: multicast: "learn" dgram_dst address from bound address and save it
332 * Because this may be "shared" socket from a "master" process, datagrams would be recv()
333 * by ONLY ONE process: we must "clone" this dgram socket --jjo
336 if (is_connected) {
337 if (getsockname(fd, (struct sockaddr *) &saddr, &saddr_len) == 0) {
338 /* must be bound */
339 if (saddr.sin_addr.s_addr == 0) {
340 fprintf(stderr, "qemu: error: init_dgram: fd=%d unbound, "
341 "cannot setup multicast dst addr\n", fd);
342 goto err;
344 /* clone dgram socket */
345 newfd = net_socket_mcast_create(&saddr, NULL);
346 if (newfd < 0) {
347 /* error already reported by net_socket_mcast_create() */
348 goto err;
350 /* clone newfd to fd, close newfd */
351 dup2(newfd, fd);
352 close(newfd);
354 } else {
355 fprintf(stderr,
356 "qemu: error: init_dgram: fd=%d failed getsockname(): %s\n",
357 fd, strerror(errno));
358 goto err;
362 nc = qemu_new_net_client(&net_dgram_socket_info, peer, model, name);
364 s = DO_UPCAST(NetSocketState, nc, nc);
366 s->fd = fd;
367 s->listen_fd = -1;
368 s->send_fn = net_socket_send_dgram;
369 net_socket_rs_init(&s->rs, net_socket_rs_finalize);
370 net_socket_read_poll(s, true);
372 /* mcast: save bound address as dst */
373 if (is_connected) {
374 s->dgram_dst = saddr;
375 snprintf(nc->info_str, sizeof(nc->info_str),
376 "socket: fd=%d (cloned mcast=%s:%d)",
377 fd, inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
378 } else {
379 snprintf(nc->info_str, sizeof(nc->info_str),
380 "socket: fd=%d", fd);
383 return s;
385 err:
386 closesocket(fd);
387 return NULL;
390 static void net_socket_connect(void *opaque)
392 NetSocketState *s = opaque;
393 s->send_fn = net_socket_send;
394 net_socket_read_poll(s, true);
397 static NetClientInfo net_socket_info = {
398 .type = NET_CLIENT_DRIVER_SOCKET,
399 .size = sizeof(NetSocketState),
400 .receive = net_socket_receive,
401 .cleanup = net_socket_cleanup,
404 static NetSocketState *net_socket_fd_init_stream(NetClientState *peer,
405 const char *model,
406 const char *name,
407 int fd, int is_connected)
409 NetClientState *nc;
410 NetSocketState *s;
412 nc = qemu_new_net_client(&net_socket_info, peer, model, name);
414 snprintf(nc->info_str, sizeof(nc->info_str), "socket: fd=%d", fd);
416 s = DO_UPCAST(NetSocketState, nc, nc);
418 s->fd = fd;
419 s->listen_fd = -1;
420 net_socket_rs_init(&s->rs, net_socket_rs_finalize);
422 /* Disable Nagle algorithm on TCP sockets to reduce latency */
423 socket_set_nodelay(fd);
425 if (is_connected) {
426 net_socket_connect(s);
427 } else {
428 qemu_set_fd_handler(s->fd, NULL, net_socket_connect, s);
430 return s;
433 static NetSocketState *net_socket_fd_init(NetClientState *peer,
434 const char *model, const char *name,
435 int fd, int is_connected)
437 int so_type = -1, optlen=sizeof(so_type);
439 if(getsockopt(fd, SOL_SOCKET, SO_TYPE, (char *)&so_type,
440 (socklen_t *)&optlen)< 0) {
441 fprintf(stderr, "qemu: error: getsockopt(SO_TYPE) for fd=%d failed\n",
442 fd);
443 closesocket(fd);
444 return NULL;
446 switch(so_type) {
447 case SOCK_DGRAM:
448 return net_socket_fd_init_dgram(peer, model, name, fd, is_connected);
449 case SOCK_STREAM:
450 return net_socket_fd_init_stream(peer, model, name, fd, is_connected);
451 default:
452 /* who knows ... this could be a eg. a pty, do warn and continue as stream */
453 fprintf(stderr, "qemu: warning: socket type=%d for fd=%d is not SOCK_DGRAM or SOCK_STREAM\n", so_type, fd);
454 return net_socket_fd_init_stream(peer, model, name, fd, is_connected);
456 return NULL;
459 static void net_socket_accept(void *opaque)
461 NetSocketState *s = opaque;
462 struct sockaddr_in saddr;
463 socklen_t len;
464 int fd;
466 for(;;) {
467 len = sizeof(saddr);
468 fd = qemu_accept(s->listen_fd, (struct sockaddr *)&saddr, &len);
469 if (fd < 0 && errno != EINTR) {
470 return;
471 } else if (fd >= 0) {
472 qemu_set_fd_handler(s->listen_fd, NULL, NULL, NULL);
473 break;
477 s->fd = fd;
478 s->nc.link_down = false;
479 net_socket_connect(s);
480 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
481 "socket: connection from %s:%d",
482 inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
485 static int net_socket_listen_init(NetClientState *peer,
486 const char *model,
487 const char *name,
488 const char *host_str)
490 NetClientState *nc;
491 NetSocketState *s;
492 SocketAddress *saddr;
493 int ret;
494 Error *local_error = NULL;
496 saddr = socket_parse(host_str, &local_error);
497 if (saddr == NULL) {
498 error_report_err(local_error);
499 return -1;
502 ret = socket_listen(saddr, &local_error);
503 if (ret < 0) {
504 qapi_free_SocketAddress(saddr);
505 error_report_err(local_error);
506 return -1;
509 nc = qemu_new_net_client(&net_socket_info, peer, model, name);
510 s = DO_UPCAST(NetSocketState, nc, nc);
511 s->fd = -1;
512 s->listen_fd = ret;
513 s->nc.link_down = true;
514 net_socket_rs_init(&s->rs, net_socket_rs_finalize);
516 qemu_set_fd_handler(s->listen_fd, net_socket_accept, NULL, s);
517 qapi_free_SocketAddress(saddr);
518 return 0;
521 typedef struct {
522 NetClientState *peer;
523 SocketAddress *saddr;
524 char *model;
525 char *name;
526 } socket_connect_data;
528 static void socket_connect_data_free(socket_connect_data *c)
530 qapi_free_SocketAddress(c->saddr);
531 g_free(c->model);
532 g_free(c->name);
533 g_free(c);
536 static void net_socket_connected(int fd, Error *err, void *opaque)
538 socket_connect_data *c = opaque;
539 NetSocketState *s;
540 char *addr_str = NULL;
541 Error *local_error = NULL;
543 addr_str = socket_address_to_string(c->saddr, &local_error);
544 if (addr_str == NULL) {
545 error_report_err(local_error);
546 closesocket(fd);
547 goto end;
550 s = net_socket_fd_init(c->peer, c->model, c->name, fd, true);
551 if (!s) {
552 closesocket(fd);
553 goto end;
556 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
557 "socket: connect to %s", addr_str);
559 end:
560 g_free(addr_str);
561 socket_connect_data_free(c);
564 static int net_socket_connect_init(NetClientState *peer,
565 const char *model,
566 const char *name,
567 const char *host_str)
569 socket_connect_data *c = g_new0(socket_connect_data, 1);
570 int fd = -1;
571 Error *local_error = NULL;
573 c->peer = peer;
574 c->model = g_strdup(model);
575 c->name = g_strdup(name);
576 c->saddr = socket_parse(host_str, &local_error);
577 if (c->saddr == NULL) {
578 goto err;
581 fd = socket_connect(c->saddr, &local_error, net_socket_connected, c);
582 if (fd < 0) {
583 goto err;
586 return 0;
588 err:
589 error_report_err(local_error);
590 socket_connect_data_free(c);
591 return -1;
594 static int net_socket_mcast_init(NetClientState *peer,
595 const char *model,
596 const char *name,
597 const char *host_str,
598 const char *localaddr_str)
600 NetSocketState *s;
601 int fd;
602 struct sockaddr_in saddr;
603 struct in_addr localaddr, *param_localaddr;
605 if (parse_host_port(&saddr, host_str) < 0)
606 return -1;
608 if (localaddr_str != NULL) {
609 if (inet_aton(localaddr_str, &localaddr) == 0)
610 return -1;
611 param_localaddr = &localaddr;
612 } else {
613 param_localaddr = NULL;
616 fd = net_socket_mcast_create(&saddr, param_localaddr);
617 if (fd < 0)
618 return -1;
620 s = net_socket_fd_init(peer, model, name, fd, 0);
621 if (!s)
622 return -1;
624 s->dgram_dst = saddr;
626 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
627 "socket: mcast=%s:%d",
628 inet_ntoa(saddr.sin_addr), ntohs(saddr.sin_port));
629 return 0;
633 static int net_socket_udp_init(NetClientState *peer,
634 const char *model,
635 const char *name,
636 const char *rhost,
637 const char *lhost)
639 NetSocketState *s;
640 int fd, ret;
641 struct sockaddr_in laddr, raddr;
643 if (parse_host_port(&laddr, lhost) < 0) {
644 return -1;
647 if (parse_host_port(&raddr, rhost) < 0) {
648 return -1;
651 fd = qemu_socket(PF_INET, SOCK_DGRAM, 0);
652 if (fd < 0) {
653 perror("socket(PF_INET, SOCK_DGRAM)");
654 return -1;
657 ret = socket_set_fast_reuse(fd);
658 if (ret < 0) {
659 closesocket(fd);
660 return -1;
662 ret = bind(fd, (struct sockaddr *)&laddr, sizeof(laddr));
663 if (ret < 0) {
664 perror("bind");
665 closesocket(fd);
666 return -1;
668 qemu_set_nonblock(fd);
670 s = net_socket_fd_init(peer, model, name, fd, 0);
671 if (!s) {
672 return -1;
675 s->dgram_dst = raddr;
677 snprintf(s->nc.info_str, sizeof(s->nc.info_str),
678 "socket: udp=%s:%d",
679 inet_ntoa(raddr.sin_addr), ntohs(raddr.sin_port));
680 return 0;
683 int net_init_socket(const Netdev *netdev, const char *name,
684 NetClientState *peer, Error **errp)
686 /* FIXME error_setg(errp, ...) on failure */
687 Error *err = NULL;
688 const NetdevSocketOptions *sock;
690 assert(netdev->type == NET_CLIENT_DRIVER_SOCKET);
691 sock = &netdev->u.socket;
693 if (sock->has_fd + sock->has_listen + sock->has_connect + sock->has_mcast +
694 sock->has_udp != 1) {
695 error_report("exactly one of fd=, listen=, connect=, mcast= or udp="
696 " is required");
697 return -1;
700 if (sock->has_localaddr && !sock->has_mcast && !sock->has_udp) {
701 error_report("localaddr= is only valid with mcast= or udp=");
702 return -1;
705 if (sock->has_fd) {
706 int fd;
708 fd = monitor_fd_param(cur_mon, sock->fd, &err);
709 if (fd == -1) {
710 error_report_err(err);
711 return -1;
713 qemu_set_nonblock(fd);
714 if (!net_socket_fd_init(peer, "socket", name, fd, 1)) {
715 return -1;
717 return 0;
720 if (sock->has_listen) {
721 if (net_socket_listen_init(peer, "socket", name, sock->listen) == -1) {
722 return -1;
724 return 0;
727 if (sock->has_connect) {
728 if (net_socket_connect_init(peer, "socket", name, sock->connect) ==
729 -1) {
730 return -1;
732 return 0;
735 if (sock->has_mcast) {
736 /* if sock->localaddr is missing, it has been initialized to "all bits
737 * zero" */
738 if (net_socket_mcast_init(peer, "socket", name, sock->mcast,
739 sock->localaddr) == -1) {
740 return -1;
742 return 0;
745 assert(sock->has_udp);
746 if (!sock->has_localaddr) {
747 error_report("localaddr= is mandatory with udp=");
748 return -1;
750 if (net_socket_udp_init(peer, "socket", name, sock->udp, sock->localaddr) ==
751 -1) {
752 return -1;
754 return 0;