socket.c@ 19383

Last change on this file since 19383 was 19383, checked in by vboxsync, 16 years ago
NAT: UDP gaps removed (detected with VPN)
Property svn:eol-style set to `native`
File size: 31.1 KB

Line
1	/*
2	* Copyright (c) 1995 Danny Gasparovski.
3	*
4	* Please read the file COPYRIGHT for the
5	* terms and conditions of the copyright.
6	*/
7
8	#define WANT_SYS_IOCTL_H
9	#include <slirp.h>
10	#include "ip_icmp.h"
11	#include "main.h"
12	#ifdef __sun__
13	#include <sys/filio.h>
14	#endif
15	#if defined (RT_OS_WINDOWS)
16	#include <iphlpapi.h>
17	#include <icmpapi.h>
18	#endif
19
20
21	static void send_icmp_to_guest(PNATState, char , size_t, struct socket , const struct sockaddr_in *);
22	#ifdef RT_OS_WINDOWS
23	static void sorecvfrom_icmp_win(PNATState, struct socket *);
24	#else /* RT_OS_WINDOWS */
25	static void sorecvfrom_icmp_unix(PNATState, struct socket *);
26	#endif /* !RT_OS_WINDOWS */
27
28	void
29	so_init()
30	{
31	}
32
33
34	struct socket *
35	solookup(struct socket *head, struct in_addr laddr,
36	u_int lport, struct in_addr faddr, u_int fport)
37	{
38	struct socket *so;
39
40	for (so = head->so_next; so != head; so = so->so_next)
41	{
42	if ( so->so_lport == lport
43	&& so->so_laddr.s_addr == laddr.s_addr
44	&& so->so_faddr.s_addr == faddr.s_addr
45	&& so->so_fport == fport)
46	return so;
47	}
48
49	return (struct socket *)NULL;
50	}
51
52	/*
53	* Create a new socket, initialise the fields
54	* It is the responsibility of the caller to
55	* insque() it into the correct linked-list
56	*/
57	struct socket *
58	socreate()
59	{
60	struct socket *so;
61
62	so = (struct socket *)RTMemAllocZ(sizeof(struct socket));
63	if(so)
64	{
65	so->so_state = SS_NOFDREF;
66	so->s = -1;
67	#if !defined(RT_OS_WINDOWS)
68	so->so_poll_index = -1;
69	#endif
70	}
71	return so;
72	}
73
74	/*
75	* remque and free a socket, clobber cache
76	* VBOX_WITH_SLIRP_MT: before sofree queue should be locked, because
77	* in sofree we don't know from which queue item beeing removed.
78	*/
79	void
80	sofree(PNATState pData, struct socket *so)
81	{
82	struct socket *so_prev = NULL;
83	if (so == tcp_last_so)
84	tcp_last_so = &tcb;
85	else if (so == udp_last_so)
86	udp_last_so = &udb;
87
88	/* check if mbuf haven't been already freed */
89	if (so->so_m != NULL)
90	m_free(pData, so->so_m);
91	#ifndef VBOX_WITH_SLIRP_MT
92	if(so->so_next && so->so_prev)
93	{
94	remque(pData, so); /* crashes if so is not in a queue */
95	NSOCK_DEC();
96	}
97
98	RTMemFree(so);
99	#else
100	so->so_deleted = 1;
101	#endif
102	}
103
104	#ifdef VBOX_WITH_SLIRP_MT
105	void
106	soread_queue(PNATState pData, struct socket so, int ret)
107	{
108	*ret = soread(pData, so);
109	}
110	#endif
111
112	/*
113	* Read from so's socket into sb_snd, updating all relevant sbuf fields
114	* NOTE: This will only be called if it is select()ed for reading, so
115	* a read() of 0 (or less) means it's disconnected
116	*/
117	int
118	soread(PNATState pData, struct socket *so)
119	{
120	int n, nn, lss, total;
121	struct sbuf *sb = &so->so_snd;
122	size_t len = sb->sb_datalen - sb->sb_cc;
123	struct iovec iov[2];
124	int mss = so->so_tcpcb->t_maxseg;
125	QSOCKET_LOCK(tcb);
126	SOCKET_LOCK(so);
127	QSOCKET_UNLOCK(tcb);
128
129	DEBUG_CALL("soread");
130	DEBUG_ARG("so = %lx", (long )so);
131
132	/*
133	* No need to check if there's enough room to read.
134	* soread wouldn't have been called if there weren't
135	*/
136
137	len = sb->sb_datalen - sb->sb_cc;
138
139	iov[0].iov_base = sb->sb_wptr;
140	iov[1].iov_base = 0;
141	iov[1].iov_len = 0;
142	if (sb->sb_wptr < sb->sb_rptr)
143	{
144	iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
145	/* Should never succeed, but... */
146	if (iov[0].iov_len > len)
147	iov[0].iov_len = len;
148	if (iov[0].iov_len > mss)
149	iov[0].iov_len -= iov[0].iov_len%mss;
150	n = 1;
151	}
152	else
153	{
154	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
155	/* Should never succeed, but... */
156	if (iov[0].iov_len > len)
157	iov[0].iov_len = len;
158	len -= iov[0].iov_len;
159	if (len)
160	{
161	iov[1].iov_base = sb->sb_data;
162	iov[1].iov_len = sb->sb_rptr - sb->sb_data;
163	if(iov[1].iov_len > len)
164	iov[1].iov_len = len;
165	total = iov[0].iov_len + iov[1].iov_len;
166	if (total > mss)
167	{
168	lss = total % mss;
169	if (iov[1].iov_len > lss)
170	{
171	iov[1].iov_len -= lss;
172	n = 2;
173	}
174	else
175	{
176	lss -= iov[1].iov_len;
177	iov[0].iov_len -= lss;
178	n = 1;
179	}
180	}
181	else
182	n = 2;
183	}
184	else
185	{
186	if (iov[0].iov_len > mss)
187	iov[0].iov_len -= iov[0].iov_len%mss;
188	n = 1;
189	}
190	}
191
192	#ifdef HAVE_READV
193	nn = readv(so->s, (struct iovec *)iov, n);
194	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
195	#else
196	nn = recv(so->s, iov[0].iov_base, iov[0].iov_len, 0);
197	#endif
198	if (nn <= 0)
199	{
200	#if defined(RT_OS_WINDOWS)
201	/*
202	* Special case for WSAEnumNetworkEvents: If we receive 0 bytes that
203	* _could_ mean that the connection is closed. But we will receive an
204	* FD_CLOSE event later if the connection was _really_ closed. With
205	* www.youtube.com I see this very often. Closing the socket too early
206	* would be dangerous.
207	*/
208	int status, ignored;
209	unsigned long pending = 0;
210	status = WSAIoctl(so->s, FIONREAD, NULL, 0, &pending, sizeof(unsigned long), &ignored, NULL, NULL);
211	if (status < 0)
212	LogRel(("NAT:error in WSAIoctl: %d\n", WSAGetLastError()));
213	if (nn == 0 && (pending != 0))
214	{
215	SOCKET_UNLOCK(so);
216	return 0;
217	}
218	#endif
219	if (nn < 0 && (errno == EINTR \|\| errno == EAGAIN \|\| errno == EWOULDBLOCK))
220	{
221	SOCKET_UNLOCK(so);
222	return 0;
223	}
224	else
225	{
226	/* nn == 0 means peer has performed an orderly shutdown */
227	DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n",
228	nn, errno, strerror(errno)));
229	sofcantrcvmore(so);
230	tcp_sockclosed(pData, sototcpcb(so));
231	SOCKET_UNLOCK(so);
232	return -1;
233	}
234	}
235
236	#ifndef HAVE_READV
237	/*
238	* If there was no error, try and read the second time round
239	* We read again if n = 2 (ie, there's another part of the buffer)
240	* and we read as much as we could in the first read
241	* We don't test for <= 0 this time, because there legitimately
242	* might not be any more data (since the socket is non-blocking),
243	* a close will be detected on next iteration.
244	* A return of -1 wont (shouldn't) happen, since it didn't happen above
245	*/
246	if (n == 2 && nn == iov[0].iov_len)
247	{
248	int ret;
249	ret = recv(so->s, iov[1].iov_base, iov[1].iov_len, 0);
250	if (ret > 0)
251	nn += ret;
252	}
253
254	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
255	#endif
256
257	/* Update fields */
258	sb->sb_cc += nn;
259	sb->sb_wptr += nn;
260	if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
261	sb->sb_wptr -= sb->sb_datalen;
262	SOCKET_UNLOCK(so);
263	return nn;
264	}
265
266	/*
267	* Get urgent data
268	*
269	* When the socket is created, we set it SO_OOBINLINE,
270	* so when OOB data arrives, we soread() it and everything
271	* in the send buffer is sent as urgent data
272	*/
273	void
274	sorecvoob(PNATState pData, struct socket *so)
275	{
276	struct tcpcb *tp = sototcpcb(so);
277
278	DEBUG_CALL("sorecvoob");
279	DEBUG_ARG("so = %lx", (long)so);
280
281	/*
282	* We take a guess at how much urgent data has arrived.
283	* In most situations, when urgent data arrives, the next
284	* read() should get all the urgent data. This guess will
285	* be wrong however if more data arrives just after the
286	* urgent data, or the read() doesn't return all the
287	* urgent data.
288	*/
289	soread(pData, so);
290	tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
291	tp->t_force = 1;
292	tcp_output(pData, tp);
293	tp->t_force = 0;
294	}
295
296	/*
297	* Send urgent data
298	* There's a lot duplicated code here, but...
299	*/
300	int
301	sosendoob(struct socket *so)
302	{
303	struct sbuf *sb = &so->so_rcv;
304	char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
305
306	int n, len;
307
308	DEBUG_CALL("sosendoob");
309	DEBUG_ARG("so = %lx", (long)so);
310	DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
311
312	if (so->so_urgc > sizeof(buff))
313	so->so_urgc = sizeof(buff); /* XXX */
314
315	if (sb->sb_rptr < sb->sb_wptr)
316	{
317	/* We can send it directly */
318	n = send(so->s, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
319	so->so_urgc -= n;
320
321	DEBUG_MISC((dfd, " --- sent %d bytes urgent data, %d urgent bytes left\n",
322	n, so->so_urgc));
323	}
324	else
325	{
326	/*
327	* Since there's no sendv or sendtov like writev,
328	* we must copy all data to a linear buffer then
329	* send it all
330	*/
331	len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
332	if (len > so->so_urgc)
333	len = so->so_urgc;
334	memcpy(buff, sb->sb_rptr, len);
335	so->so_urgc -= len;
336	if (so->so_urgc)
337	{
338	n = sb->sb_wptr - sb->sb_data;
339	if (n > so->so_urgc)
340	n = so->so_urgc;
341	memcpy(buff + len, sb->sb_data, n);
342	so->so_urgc -= n;
343	len += n;
344	}
345	n = send(so->s, buff, len, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
346	#ifdef DEBUG
347	if (n != len)
348	DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
349	#endif
350	DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n",
351	n, so->so_urgc));
352	}
353
354	sb->sb_cc -= n;
355	sb->sb_rptr += n;
356	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
357	sb->sb_rptr -= sb->sb_datalen;
358
359	return n;
360	}
361
362	/*
363	* Write data from so_rcv to so's socket,
364	* updating all sbuf field as necessary
365	*/
366	int
367	sowrite(PNATState pData, struct socket *so)
368	{
369	int n, nn;
370	struct sbuf *sb = &so->so_rcv;
371	size_t len = sb->sb_cc;
372	struct iovec iov[2];
373
374	DEBUG_CALL("sowrite");
375	DEBUG_ARG("so = %lx", (long)so);
376	QSOCKET_LOCK(tcb);
377	SOCKET_LOCK(so);
378	QSOCKET_UNLOCK(tcb);
379	if (so->so_urgc)
380	{
381	sosendoob(so);
382	if (sb->sb_cc == 0)
383	{
384	SOCKET_UNLOCK(so);
385	return 0;
386	}
387	}
388
389	/*
390	* No need to check if there's something to write,
391	* sowrite wouldn't have been called otherwise
392	*/
393
394	len = sb->sb_cc;
395
396	iov[0].iov_base = sb->sb_rptr;
397	iov[1].iov_base = 0;
398	iov[1].iov_len = 0;
399	if (sb->sb_rptr < sb->sb_wptr)
400	{
401	iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
402	/* Should never succeed, but... */
403	if (iov[0].iov_len > len)
404	iov[0].iov_len = len;
405	n = 1;
406	}
407	else
408	{
409	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
410	if (iov[0].iov_len > len)
411	iov[0].iov_len = len;
412	len -= iov[0].iov_len;
413	if (len)
414	{
415	iov[1].iov_base = sb->sb_data;
416	iov[1].iov_len = sb->sb_wptr - sb->sb_data;
417	if (iov[1].iov_len > len)
418	iov[1].iov_len = len;
419	n = 2;
420	}
421	else
422	n = 1;
423	}
424	/* Check if there's urgent data to send, and if so, send it */
425	#ifdef HAVE_READV
426	nn = writev(so->s, (const struct iovec *)iov, n);
427	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
428	#else
429	nn = send(so->s, iov[0].iov_base, iov[0].iov_len, 0);
430	#endif
431	/* This should never happen, but people tell me it does shrug */
432	if (nn < 0 && (errno == EAGAIN \|\| errno == EINTR \|\| errno == EWOULDBLOCK))
433	{
434	SOCKET_UNLOCK(so);
435	return 0;
436	}
437
438	if (nn < 0 \|\| (nn == 0 && iov[0].iov_len > 0))
439	{
440	DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
441	so->so_state, errno));
442	sofcantsendmore(so);
443	tcp_sockclosed(pData, sototcpcb(so));
444	SOCKET_UNLOCK(so);
445	return -1;
446	}
447
448	#ifndef HAVE_READV
449	if (n == 2 && nn == iov[0].iov_len)
450	{
451	int ret;
452	ret = send(so->s, iov[1].iov_base, iov[1].iov_len, 0);
453	if (ret > 0)
454	nn += ret;
455	}
456	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
457	#endif
458
459	/* Update sbuf */
460	sb->sb_cc -= nn;
461	sb->sb_rptr += nn;
462	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
463	sb->sb_rptr -= sb->sb_datalen;
464
465	/*
466	* If in DRAIN mode, and there's no more data, set
467	* it CANTSENDMORE
468	*/
469	if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
470	sofcantsendmore(so);
471
472	SOCKET_UNLOCK(so);
473	return nn;
474	}
475
476	/*
477	* recvfrom() a UDP socket
478	*/
479	void
480	sorecvfrom(PNATState pData, struct socket *so)
481	{
482	struct sockaddr_in addr;
483	socklen_t addrlen = sizeof(struct sockaddr_in);
484
485	DEBUG_CALL("sorecvfrom");
486	DEBUG_ARG("so = %lx", (long)so);
487
488	if (so->so_type == IPPROTO_ICMP)
489	{
490	/* This is a "ping" reply */
491	#ifdef RT_OS_WINDOWS
492	sorecvfrom_icmp_win(pData, so);
493	#else /* RT_OS_WINDOWS */
494	sorecvfrom_icmp_unix(pData, so);
495	#endif /* !RT_OS_WINDOWS */
496	udp_detach(pData, so);
497	}
498	else
499	{
500	/* A "normal" UDP packet */
501	struct mbuf *m;
502	struct ethhdr *eh;
503	size_t len;
504	u_long n;
505
506	QSOCKET_LOCK(udb);
507	SOCKET_LOCK(so);
508	QSOCKET_UNLOCK(udb);
509
510	if (!(m = m_get(pData)))
511	{
512	SOCKET_UNLOCK(so);
513	return;
514	}
515	m->m_data += if_maxlinkhdr;
516	m->m_data += sizeof(struct udphdr)
517	+ sizeof(struct ip); /XXX: no options atm/
518
519	/*
520	* XXX Shouldn't FIONREAD packets destined for port 53,
521	* but I don't know the max packet size for DNS lookups
522	*/
523	#if 0
524	len = M_FREEROOM(m);
525	/* if (so->so_fport != htons(53)) */
526	{
527	ioctlsocket(so->s, FIONREAD, &n);
528
529	if (n > len)
530	{
531	n = (m->m_data - m->m_dat) + m->m_len + n + 1;
532	m_inc(m, n);
533	len = M_FREEROOM(m);
534	}
535	}
536	#else
537	len = m->m_size - (if_maxlinkhdr + sizeof(struct udpiphdr)); /* get max free room here*/
538	#endif
539	m->m_len = recvfrom(so->s, m->m_data, len, 0,
540	(struct sockaddr *)&addr, &addrlen);
541	Log2((" did recvfrom %d, errno = %d-%s\n",
542	m->m_len, errno, strerror(errno)));
543	if(m->m_len < 0)
544	{
545	u_char code = ICMP_UNREACH_PORT;
546
547	if (errno == EHOSTUNREACH)
548	code = ICMP_UNREACH_HOST;
549	else if(errno == ENETUNREACH)
550	code = ICMP_UNREACH_NET;
551
552	Log2((dfd," rx error, tx icmp ICMP_UNREACH:%i\n", code));
553	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
554	so->so_m = NULL;
555	m_free(pData, m);
556	}
557	else
558	{
559	/*
560	* Hack: domain name lookup will be used the most for UDP,
561	* and since they'll only be used once there's no need
562	* for the 4 minute (or whatever) timeout... So we time them
563	* out much quicker (10 seconds for now...)
564	*/
565	#ifndef VBOX_WITH_SLIRP_DNS_PROXY
566	if (so->so_expire)
567	{
568	if (so->so_fport == htons(53))
569	so->so_expire = curtime + SO_EXPIREFAST;
570	else
571	so->so_expire = curtime + SO_EXPIRE;
572	}
573	#else
574	if (so->so_expire)
575	{
576	if (so->so_fport != htons(53))
577	so->so_expire = curtime + SO_EXPIRE;
578	}
579	/*
580	* last argument should be changed if Slirp will inject IP attributes
581	* Note: Here we can't check if dnsproxy's sent initial request
582	*/
583	if (so->so_fport == htons(53))
584	dnsproxy_answer(pData, so, m);
585	#endif
586
587	#if 0
588	if (m->m_len == len)
589	{
590	m_inc(m, MINCSIZE);
591	m->m_len = 0;
592	}
593	#endif
594
595	/*
596	* If this packet was destined for CTL_ADDR,
597	* make it look like that's where it came from, done by udp_output
598	*/
599	#ifdef VBOX_WITH_NAT_SERVICE
600	{
601	struct ethhdr *eh0;
602	struct ethhdr *eh;
603	Assert(so->so_m);
604	eh0 = (struct ethhdr *)so->so_m->m_dat;
605	eh = (struct ethhdr *)m->m_dat;
606
607	memcpy(eh->h_source, eh0->h_source, ETH_ALEN);
608	}
609	#endif
610	udp_output(pData, so, m, &addr);
611	SOCKET_UNLOCK(so);
612	} /* rx error */
613	} /* if ping packet */
614	}
615
616	/*
617	* sendto() a socket
618	*/
619	int
620	sosendto(PNATState pData, struct socket so, struct mbuf m)
621	{
622	int ret;
623	struct sockaddr_in addr;
624	#if 0
625	struct sockaddr_in host_addr;
626	#endif
627
628	DEBUG_CALL("sosendto");
629	DEBUG_ARG("so = %lx", (long)so);
630	DEBUG_ARG("m = %lx", (long)m);
631
632	addr.sin_family = AF_INET;
633	if ((so->so_faddr.s_addr & htonl(pData->netmask)) == special_addr.s_addr)
634	{
635	/* It's an alias */
636	uint32_t last_byte = ntohl(so->so_faddr.s_addr) & ~pData->netmask;
637	switch(last_byte)
638	{
639	#if 0
640	/* handle this case at 'default:' */
641	case CTL_BROADCAST:
642	addr.sin_addr.s_addr = INADDR_BROADCAST;
643	/* Send the packet to host to fully emulate broadcast */
644	/** @todo r=klaus: on Linux host this causes the host to receive
645	* the packet twice for some reason. And I cannot find any place
646	* in the man pages which states that sending a broadcast does not
647	* reach the host itself. */
648	host_addr.sin_family = AF_INET;
649	host_addr.sin_port = so->so_fport;
650	host_addr.sin_addr = our_addr;
651	sendto(so->s, m->m_data, m->m_len, 0,
652	(struct sockaddr *)&host_addr, sizeof (struct sockaddr));
653	break;
654	#endif
655	case CTL_DNS:
656	#ifndef VBOX_WITH_MULTI_DNS
657	if (!get_dns_addr(pData, &dns_addr))
658	addr.sin_addr = dns_addr;
659	else
660	addr.sin_addr = loopback_addr;
661	break;
662	#endif
663	case CTL_ALIAS:
664	default:
665	if (last_byte == ~pData->netmask)
666	addr.sin_addr.s_addr = INADDR_BROADCAST;
667	else
668	addr.sin_addr = loopback_addr;
669	break;
670	}
671	}
672	else
673	addr.sin_addr = so->so_faddr;
674	addr.sin_port = so->so_fport;
675
676	DEBUG_MISC((dfd, " sendto()ing, addr.sin_port=%d, addr.sin_addr.s_addr=%.16s\n",
677	ntohs(addr.sin_port), inet_ntoa(addr.sin_addr)));
678
679	/* Don't care what port we get */
680	ret = sendto(so->s, m->m_data, m->m_len, 0,
681	(struct sockaddr *)&addr, sizeof (struct sockaddr));
682	if (ret < 0)
683	{
684	LogRel(("UDP: sendto fails (%s)\n", strerror(errno)));
685	return -1;
686	}
687
688	/*
689	* Kill the socket if there's no reply in 4 minutes,
690	* but only if it's an expirable socket
691	*/
692	if (so->so_expire)
693	so->so_expire = curtime + SO_EXPIRE;
694	so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
695	return 0;
696	}
697
698	/*
699	* XXX This should really be tcp_listen
700	*/
701	struct socket *
702	solisten(PNATState pData, u_int port, u_int32_t laddr, u_int lport, int flags)
703	{
704	struct sockaddr_in addr;
705	struct socket *so;
706	socklen_t addrlen = sizeof(addr);
707	int s, opt = 1;
708
709	DEBUG_CALL("solisten");
710	DEBUG_ARG("port = %d", port);
711	DEBUG_ARG("laddr = %x", laddr);
712	DEBUG_ARG("lport = %d", lport);
713	DEBUG_ARG("flags = %x", flags);
714
715	if ((so = socreate()) == NULL)
716	{
717	/* RTMemFree(so); Not sofree() ??? free(NULL) == NOP */
718	return NULL;
719	}
720
721	/* Don't tcp_attach... we don't need so_snd nor so_rcv */
722	if ((so->so_tcpcb = tcp_newtcpcb(pData, so)) == NULL)
723	{
724	RTMemFree(so);
725	return NULL;
726	}
727
728	SOCKET_LOCK_CREATE(so);
729	SOCKET_LOCK(so);
730	QSOCKET_LOCK(tcb);
731	insque(pData, so,&tcb);
732	NSOCK_INC();
733	QSOCKET_UNLOCK(tcb);
734
735	/*
736	* SS_FACCEPTONCE sockets must time out.
737	*/
738	if (flags & SS_FACCEPTONCE)
739	so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
740
741	so->so_state = (SS_FACCEPTCONN\|flags);
742	so->so_lport = lport; /* Kept in network format */
743	so->so_laddr.s_addr = laddr; /* Ditto */
744
745	addr.sin_family = AF_INET;
746	addr.sin_addr.s_addr = INADDR_ANY;
747	addr.sin_port = port;
748
749	if ( ((s = socket(AF_INET, SOCK_STREAM, 0)) < 0)
750	\|\| (setsockopt(s, SOL_SOCKET, SO_REUSEADDR,(char *)&opt, sizeof(int)) < 0)
751	\|\| (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0)
752	\|\| (listen(s, 1) < 0))
753	{
754	#ifdef RT_OS_WINDOWS
755	int tmperrno = WSAGetLastError(); /* Don't clobber the real reason we failed */
756	closesocket(s);
757	QSOCKET_LOCK(tcb);
758	sofree(pData, so);
759	QSOCKET_UNLOCK(tcb);
760	/* Restore the real errno */
761	WSASetLastError(tmperrno);
762	#else
763	int tmperrno = errno; /* Don't clobber the real reason we failed */
764	close(s);
765	QSOCKET_LOCK(tcb);
766	sofree(pData, so);
767	QSOCKET_UNLOCK(tcb);
768	/* Restore the real errno */
769	errno = tmperrno;
770	#endif
771	return NULL;
772	}
773	setsockopt(s, SOL_SOCKET, SO_OOBINLINE,(char *)&opt, sizeof(int));
774
775	getsockname(s,(struct sockaddr *)&addr,&addrlen);
776	so->so_fport = addr.sin_port;
777	if (addr.sin_addr.s_addr == 0 \|\| addr.sin_addr.s_addr == loopback_addr.s_addr)
778	so->so_faddr = alias_addr;
779	else
780	so->so_faddr = addr.sin_addr;
781
782	so->s = s;
783	SOCKET_UNLOCK(so);
784	return so;
785	}
786
787	/*
788	* Data is available in so_rcv
789	* Just write() the data to the socket
790	* XXX not yet...
791	*/
792	void
793	sorwakeup(struct socket *so)
794	{
795	#if 0
796	sowrite(so);
797	FD_CLR(so->s,&writefds);
798	#endif
799	}
800
801	/*
802	* Data has been freed in so_snd
803	* We have room for a read() if we want to
804	* For now, don't read, it'll be done in the main loop
805	*/
806	void
807	sowwakeup(struct socket *so)
808	{
809	}
810
811	/*
812	* Various session state calls
813	* XXX Should be #define's
814	* The socket state stuff needs work, these often get call 2 or 3
815	* times each when only 1 was needed
816	*/
817	void
818	soisfconnecting(struct socket *so)
819	{
820	so->so_state &= ~(SS_NOFDREF\|SS_ISFCONNECTED\|SS_FCANTRCVMORE\|
821	SS_FCANTSENDMORE\|SS_FWDRAIN);
822	so->so_state \|= SS_ISFCONNECTING; /* Clobber other states */
823	}
824
825	void
826	soisfconnected(struct socket *so)
827	{
828	so->so_state &= ~(SS_ISFCONNECTING\|SS_FWDRAIN\|SS_NOFDREF);
829	so->so_state \|= SS_ISFCONNECTED; /* Clobber other states */
830	}
831
832	void
833	sofcantrcvmore(struct socket *so)
834	{
835	if ((so->so_state & SS_NOFDREF) == 0)
836	{
837	shutdown(so->s, 0);
838	}
839	so->so_state &= ~(SS_ISFCONNECTING);
840	if (so->so_state & SS_FCANTSENDMORE)
841	so->so_state = SS_NOFDREF; /* Don't select it */
842	/* XXX close() here as well? */
843	else
844	so->so_state \|= SS_FCANTRCVMORE;
845	}
846
847	void
848	sofcantsendmore(struct socket *so)
849	{
850	if ((so->so_state & SS_NOFDREF) == 0)
851	shutdown(so->s, 1); /* send FIN to fhost */
852
853	so->so_state &= ~(SS_ISFCONNECTING);
854	if (so->so_state & SS_FCANTRCVMORE)
855	so->so_state = SS_NOFDREF; /* as above */
856	else
857	so->so_state \|= SS_FCANTSENDMORE;
858	}
859
860	void
861	soisfdisconnected(struct socket *so)
862	{
863	#if 0
864	so->so_state &= ~(SS_ISFCONNECTING\|SS_ISFCONNECTED);
865	close(so->s);
866	so->so_state = SS_ISFDISCONNECTED;
867	/*
868	* XXX Do nothing ... ?
869	*/
870	#endif
871	}
872
873	/*
874	* Set write drain mode
875	* Set CANTSENDMORE once all data has been write()n
876	*/
877	void
878	sofwdrain(struct socket *so)
879	{
880	if (so->so_rcv.sb_cc)
881	so->so_state \|= SS_FWDRAIN;
882	else
883	sofcantsendmore(so);
884	}
885
886	static void
887	send_icmp_to_guest(PNATState pData, char buff, size_t len, struct socket so, const struct sockaddr_in *addr)
888	{
889	struct ip *ip;
890	uint32_t dst, src;
891	char ip_copy[256];
892	struct icmp *icp;
893	int old_ip_len = 0;
894	int hlen, original_hlen = 0;
895	struct mbuf *m;
896	struct icmp_msg *icm;
897	uint8_t proto;
898	int type = 0;
899
900	ip = (struct ip *)buff;
901	hlen = (ip->ip_hl << 2);
902	icp = (struct icmp )((char )ip + hlen);
903
904	Log(("ICMP:received msg(t:%d, c:%d)\n", icp->icmp_type, icp->icmp_code));
905	if ( icp->icmp_type != ICMP_ECHOREPLY
906	&& icp->icmp_type != ICMP_TIMXCEED
907	&& icp->icmp_type != ICMP_UNREACH)
908	{
909	return;
910	}
911
912	type = icp->icmp_type;
913	if ( type == ICMP_TIMXCEED
914	\|\| type == ICMP_UNREACH)
915	{
916	ip = &icp->icmp_ip;
917	DO_ALIAS(&ip->ip_dst);
918	}
919	else
920	{
921	DO_ALIAS(&ip->ip_src);
922	}
923
924	icm = icmp_find_original_mbuf(pData, ip);
925
926	if (icm == NULL)
927	{
928	Log(("NAT: Can't find the corresponding packet for the received ICMP\n"));
929	return;
930	}
931
932	m = icm->im_m;
933	Assert(m != NULL);
934
935	src = addr->sin_addr.s_addr;
936
937	ip = mtod(m, struct ip *);
938	proto = ip->ip_p;
939	/* Now ip is pointing on header we've sent from guest */
940	if ( icp->icmp_type == ICMP_TIMXCEED
941	\|\| icp->icmp_type == ICMP_UNREACH)
942	{
943	old_ip_len = (ip->ip_hl << 2) + 64;
944	if (old_ip_len > sizeof(ip_copy))
945	old_ip_len = sizeof(ip_copy);
946	memcpy(ip_copy, ip, old_ip_len);
947	}
948
949	/* source address from original IP packet*/
950	dst = ip->ip_src.s_addr;
951
952	/* overide ther tail of old packet */
953	ip = mtod(m, struct ip ); / ip is from mbuf we've overrided */
954	original_hlen = ip->ip_hl << 2;
955	/* saves original ip header and options */
956	memcpy(m->m_data + original_hlen, buff + hlen, len - hlen);
957	m->m_len = len - hlen + original_hlen;
958	ip->ip_len = m->m_len;
959	ip->ip_p = IPPROTO_ICMP; /* the original package could be whatever, but we're response via ICMP*/
960
961	icp = (struct icmp )((char )ip + (ip->ip_hl << 2));
962	type = icp->icmp_type;
963	if ( type == ICMP_TIMXCEED
964	\|\| type == ICMP_UNREACH)
965	{
966	/* according RFC 793 error messages required copy of initial IP header + 64 bit */
967	memcpy(&icp->icmp_ip, ip_copy, old_ip_len);
968	ip->ip_tos = ((ip->ip_tos & 0x1E) \| 0xC0); /* high priority for errors */
969	}
970
971	ip->ip_src.s_addr = src;
972	ip->ip_dst.s_addr = dst;
973	icmp_reflect(pData, m);
974	LIST_REMOVE(icm, im_list);
975	/* Don't call m_free here*/
976
977	if ( type == ICMP_TIMXCEED
978	\|\| type == ICMP_UNREACH)
979	{
980	icm->im_so->so_m = NULL;
981	switch (proto)
982	{
983	case IPPROTO_UDP:
984	/XXX: so->so_m already freed so we shouldn't call sofree /
985	udp_detach(pData, icm->im_so);
986	break;
987	case IPPROTO_TCP:
988	/close tcp should be here /
989	break;
990	default:
991	/* do nothing */
992	break;
993	}
994	}
995	RTMemFree(icm);
996	}
997
998	#ifdef RT_OS_WINDOWS
999	static void
1000	sorecvfrom_icmp_win(PNATState pData, struct socket *so)
1001	{
1002	int len;
1003	int i;
1004	struct ip *ip;
1005	struct mbuf *m;
1006	struct icmp *icp;
1007	struct icmp_msg *icm;
1008	struct ip ip_broken; / ICMP returns header + 64 bit of packet */
1009	uint32_t src;
1010	ICMP_ECHO_REPLY *icr;
1011	int hlen = 0;
1012	int data_len = 0;
1013	int nbytes = 0;
1014	u_char code = ~0;
1015
1016	len = pData->pfIcmpParseReplies(pData->pvIcmpBuffer, pData->szIcmpBuffer);
1017	if (len < 0)
1018	{
1019	LogRel(("NAT: Error (%d) occurred on ICMP receiving\n", GetLastError()));
1020	return;
1021	}
1022	if (len == 0)
1023	return; /* no error */
1024
1025	icr = (ICMP_ECHO_REPLY *)pData->pvIcmpBuffer;
1026	for (i = 0; i < len; ++i)
1027	{
1028	switch(icr[i].Status)
1029	{
1030	case IP_DEST_HOST_UNREACHABLE:
1031	code = (code != ~0 ? code : ICMP_UNREACH_HOST);
1032	case IP_DEST_NET_UNREACHABLE:
1033	code = (code != ~0 ? code : ICMP_UNREACH_NET);
1034	case IP_DEST_PROT_UNREACHABLE:
1035	code = (code != ~0 ? code : ICMP_UNREACH_PROTOCOL);
1036	/* UNREACH error inject here */
1037	case IP_DEST_PORT_UNREACHABLE:
1038	code = (code != ~0 ? code : ICMP_UNREACH_PORT);
1039	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, "Error occurred!!!");
1040	so->so_m = NULL;
1041	break;
1042	case IP_SUCCESS: /* echo replied */
1043	m = m_get(pData);
1044	m->m_data += if_maxlinkhdr;
1045	ip = mtod(m, struct ip *);
1046	ip->ip_src.s_addr = icr[i].Address;
1047	DO_ALIAS(&ip->ip_src);
1048	ip->ip_p = IPPROTO_ICMP;
1049	ip->ip_dst.s_addr = so->so_laddr.s_addr; /XXX: still the hack/
1050	data_len = sizeof(struct ip);
1051	ip->ip_hl = data_len >> 2; /* requiered for icmp_reflect, no IP options */
1052	ip->ip_ttl = icr[i].Options.Ttl;
1053
1054	icp = (struct icmp )&ip[1]; / no options */
1055	icp->icmp_type = ICMP_ECHOREPLY;
1056	icp->icmp_code = 0;
1057	icp->icmp_id = so->so_icmp_id;
1058	icp->icmp_seq = so->so_icmp_seq;
1059
1060	data_len += ICMP_MINLEN;
1061
1062	nbytes = (data_len + icr[i].DataSize > m->m_size? m->m_size - data_len: icr[i].DataSize);
1063	memcpy(icp->icmp_data, icr[i].Data, nbytes);
1064
1065	data_len += icr[i].DataSize;
1066
1067	ip->ip_len = data_len;
1068	m->m_len = ip->ip_len;
1069
1070	icmp_reflect(pData, m);
1071	break;
1072	case IP_TTL_EXPIRED_TRANSIT: /* TTL expired */
1073
1074	ip_broken = icr[i].Data;
1075	icm = icmp_find_original_mbuf(pData, ip_broken);
1076	if (icm == NULL) {
1077	Log(("ICMP: can't find original package (first double word %x)\n", (uint32_t )ip_broken));
1078	return;
1079	}
1080	m = icm->im_m;
1081	ip = mtod(m, struct ip *);
1082	ip->ip_ttl = icr[i].Options.Ttl;
1083	src = ip->ip_src.s_addr;
1084	ip->ip_dst.s_addr = src;
1085	ip->ip_dst.s_addr = icr[i].Address;
1086
1087	hlen = (ip->ip_hl << 2);
1088	icp = (struct icmp )((char )ip + hlen);
1089	ip_broken->ip_src.s_addr = src; /it packet sent from host not from guest/
1090	data_len = (ip_broken->ip_hl << 2) + 64;
1091
1092	nbytes =(hlen + ICMP_MINLEN + data_len > m->m_size? m->m_size - (hlen + ICMP_MINLEN): data_len);
1093	memcpy(icp->icmp_data, ip_broken, nbytes);
1094	icmp_reflect(pData, m);
1095	break;
1096	default:
1097	Log(("ICMP(default): message with Status: %x was received from %x\n", icr[i].Status, icr[i].Address));
1098	break;
1099	}
1100	}
1101	}
1102	#else /* RT_OS_WINDOWS */
1103	static void sorecvfrom_icmp_unix(PNATState pData, struct socket *so)
1104	{
1105	struct sockaddr_in addr;
1106	socklen_t addrlen = sizeof(struct sockaddr_in);
1107	char buff[1500];
1108	int len;
1109	len = recvfrom(so->s, buff, 1500, 0,
1110	(struct sockaddr *)&addr, &addrlen);
1111	/* XXX Check if reply is "correct"? */
1112
1113	if (len == -1 \|\| len == 0)
1114	{
1115	u_char code = ICMP_UNREACH_PORT;
1116
1117	if (errno == EHOSTUNREACH)
1118	code = ICMP_UNREACH_HOST;
1119	else if(errno == ENETUNREACH)
1120	code = ICMP_UNREACH_NET;
1121
1122	DEBUG_MISC((dfd," udp icmp rx errno = %d-%s\n",
1123	errno, strerror(errno)));
1124	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
1125	so->so_m = NULL;
1126	}
1127	else
1128	{
1129	send_icmp_to_guest(pData, buff, len, so, &addr);
1130	}
1131	}
1132	#endif /* !RT_OS_WINDOWS */

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/Devices/Network/slirp/socket.c@ 19383

Download in other formats: