socket.c@ 16688

Last change on this file since 16688 was 16653, checked in by vboxsync, 16 years ago
NAT: fixed memmory corruption in new Unix multiplexing mechanism
Property svn:eol-style set to `native`
File size: 30.3 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	memset(so, 0, sizeof(struct socket));
66	so->so_state = SS_NOFDREF;
67	so->s = -1;
68	#if defined(VBOX_WITH_SIMPLIFIED_SLIRP_SYNC) && !defined(RT_OS_WINDOWS)
69	so->so_poll_index = -1;
70	#endif
71	}
72	return so;
73	}
74
75	/*
76	* remque and free a socket, clobber cache
77	* VBOX_WITH_SLIRP_MT: before sofree queue should be locked, because
78	* in sofree we don't know from which queue item beeing removed.
79	*/
80	void
81	sofree(PNATState pData, struct socket *so)
82	{
83	struct socket *so_prev = NULL;
84	if (so == tcp_last_so)
85	tcp_last_so = &tcb;
86	else if (so == udp_last_so)
87	udp_last_so = &udb;
88
89	/* check if mbuf haven't been already freed */
90	if (so->so_m != NULL)
91	m_free(pData, so->so_m);
92	#ifndef VBOX_WITH_SLIRP_MT
93	if(so->so_next && so->so_prev)
94	{
95	remque(pData, so); /* crashes if so is not in a queue */
96	NSOCK_DEC();
97	}
98
99	RTMemFree(so);
100	#else
101	so->so_deleted = 1;
102	#endif
103	}
104
105	#ifdef VBOX_WITH_SLIRP_MT
106	void
107	soread_queue(PNATState pData, struct socket so, int ret)
108	{
109	*ret = soread(pData, so);
110	}
111	#endif
112
113	/*
114	* Read from so's socket into sb_snd, updating all relevant sbuf fields
115	* NOTE: This will only be called if it is select()ed for reading, so
116	* a read() of 0 (or less) means it's disconnected
117	*/
118	int
119	soread(PNATState pData, struct socket *so)
120	{
121	int n, nn, lss, total;
122	struct sbuf *sb = &so->so_snd;
123	size_t len = sb->sb_datalen - sb->sb_cc;
124	struct iovec iov[2];
125	int mss = so->so_tcpcb->t_maxseg;
126	QSOCKET_LOCK(tcb);
127	SOCKET_LOCK(so);
128	QSOCKET_UNLOCK(tcb);
129
130	DEBUG_CALL("soread");
131	DEBUG_ARG("so = %lx", (long )so);
132
133	/*
134	* No need to check if there's enough room to read.
135	* soread wouldn't have been called if there weren't
136	*/
137
138	len = sb->sb_datalen - sb->sb_cc;
139
140	iov[0].iov_base = sb->sb_wptr;
141	iov[1].iov_base = 0;
142	iov[1].iov_len = 0;
143	if (sb->sb_wptr < sb->sb_rptr)
144	{
145	iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
146	/* Should never succeed, but... */
147	if (iov[0].iov_len > len)
148	iov[0].iov_len = len;
149	if (iov[0].iov_len > mss)
150	iov[0].iov_len -= iov[0].iov_len%mss;
151	n = 1;
152	}
153	else
154	{
155	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
156	/* Should never succeed, but... */
157	if (iov[0].iov_len > len)
158	iov[0].iov_len = len;
159	len -= iov[0].iov_len;
160	if (len)
161	{
162	iov[1].iov_base = sb->sb_data;
163	iov[1].iov_len = sb->sb_rptr - sb->sb_data;
164	if(iov[1].iov_len > len)
165	iov[1].iov_len = len;
166	total = iov[0].iov_len + iov[1].iov_len;
167	if (total > mss)
168	{
169	lss = total % mss;
170	if (iov[1].iov_len > lss)
171	{
172	iov[1].iov_len -= lss;
173	n = 2;
174	}
175	else
176	{
177	lss -= iov[1].iov_len;
178	iov[0].iov_len -= lss;
179	n = 1;
180	}
181	}
182	else
183	n = 2;
184	}
185	else
186	{
187	if (iov[0].iov_len > mss)
188	iov[0].iov_len -= iov[0].iov_len%mss;
189	n = 1;
190	}
191	}
192
193	#ifdef HAVE_READV
194	nn = readv(so->s, (struct iovec *)iov, n);
195	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
196	#else
197	nn = recv(so->s, iov[0].iov_base, iov[0].iov_len,0);
198	#endif
199	if (nn <= 0)
200	{
201	#if defined(VBOX_WITH_SIMPLIFIED_SLIRP_SYNC) && defined(RT_OS_WINDOWS)
202	/*
203	* Special case for WSAEnumNetworkEvents: If we receive 0 bytes that
204	* _could_ mean that the connection is closed. But we will receive an
205	* FD_CLOSE event later if the connection was _really_ closed. With
206	* www.youtube.com I see this very often. Closing the socket too early
207	* would be dangerous.
208	*/
209	int status, ignored;
210	unsigned long pending = 0;
211	status = WSAIoctl(so->s, FIONREAD, NULL, 0, &pending, sizeof(unsigned long), &ignored, NULL, NULL);
212	if (status < 0)
213	Log2(("error in WSAIoctl: %d\n", WSAGetLastError()));
214	if (nn == 0 && (pending != 0))
215	{
216	SOCKET_UNLOCK(so);
217	return 0;
218	}
219	#endif
220	if (nn < 0 && (errno == EINTR \|\| errno == EAGAIN \|\| errno == EWOULDBLOCK))
221	{
222	SOCKET_UNLOCK(so);
223	return 0;
224	}
225	else
226	{
227	/* nn == 0 means peer has performed an orderly shutdown */
228	DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n",
229	nn, errno,strerror(errno)));
230	sofcantrcvmore(so);
231	tcp_sockclosed(pData, sototcpcb(so));
232	SOCKET_UNLOCK(so);
233	return -1;
234	}
235	}
236
237	#ifndef HAVE_READV
238	/*
239	* If there was no error, try and read the second time round
240	* We read again if n = 2 (ie, there's another part of the buffer)
241	* and we read as much as we could in the first read
242	* We don't test for <= 0 this time, because there legitimately
243	* might not be any more data (since the socket is non-blocking),
244	* a close will be detected on next iteration.
245	* A return of -1 wont (shouldn't) happen, since it didn't happen above
246	*/
247	if (n == 2 && nn == iov[0].iov_len)
248	{
249	int ret;
250	ret = recv(so->s, iov[1].iov_base, iov[1].iov_len,0);
251	if (ret > 0)
252	nn += ret;
253	}
254
255	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
256	#endif
257
258	/* Update fields */
259	sb->sb_cc += nn;
260	sb->sb_wptr += nn;
261	if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
262	sb->sb_wptr -= sb->sb_datalen;
263	SOCKET_UNLOCK(so);
264	return nn;
265	}
266
267	/*
268	* Get urgent data
269	*
270	* When the socket is created, we set it SO_OOBINLINE,
271	* so when OOB data arrives, we soread() it and everything
272	* in the send buffer is sent as urgent data
273	*/
274	void
275	sorecvoob(PNATState pData, struct socket *so)
276	{
277	struct tcpcb *tp = sototcpcb(so);
278
279	DEBUG_CALL("sorecvoob");
280	DEBUG_ARG("so = %lx", (long)so);
281
282	/*
283	* We take a guess at how much urgent data has arrived.
284	* In most situations, when urgent data arrives, the next
285	* read() should get all the urgent data. This guess will
286	* be wrong however if more data arrives just after the
287	* urgent data, or the read() doesn't return all the
288	* urgent data.
289	*/
290	soread(pData, so);
291	tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
292	tp->t_force = 1;
293	tcp_output(pData, tp);
294	tp->t_force = 0;
295	}
296
297	/*
298	* Send urgent data
299	* There's a lot duplicated code here, but...
300	*/
301	int
302	sosendoob(struct socket *so)
303	{
304	struct sbuf *sb = &so->so_rcv;
305	char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
306
307	int n, len;
308
309	DEBUG_CALL("sosendoob");
310	DEBUG_ARG("so = %lx", (long)so);
311	DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
312
313	if (so->so_urgc > sizeof(buff))
314	so->so_urgc = sizeof(buff); /* XXX */
315
316	if (sb->sb_rptr < sb->sb_wptr)
317	{
318	/* We can send it directly */
319	n = send(so->s, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
320	so->so_urgc -= n;
321
322	DEBUG_MISC((dfd, " --- sent %d bytes urgent data, %d urgent bytes left\n",
323	n, so->so_urgc));
324	}
325	else
326	{
327	/*
328	* Since there's no sendv or sendtov like writev,
329	* we must copy all data to a linear buffer then
330	* send it all
331	*/
332	len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
333	if (len > so->so_urgc)
334	len = so->so_urgc;
335	memcpy(buff, sb->sb_rptr, len);
336	so->so_urgc -= len;
337	if (so->so_urgc)
338	{
339	n = sb->sb_wptr - sb->sb_data;
340	if (n > so->so_urgc)
341	n = so->so_urgc;
342	memcpy(buff + len, sb->sb_data, n);
343	so->so_urgc -= n;
344	len += n;
345	}
346	n = send(so->s, buff, len, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
347	#ifdef DEBUG
348	if (n != len)
349	DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
350	#endif
351	DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n",
352	n, so->so_urgc));
353	}
354
355	sb->sb_cc -= n;
356	sb->sb_rptr += n;
357	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
358	sb->sb_rptr -= sb->sb_datalen;
359
360	return n;
361	}
362
363	/*
364	* Write data from so_rcv to so's socket,
365	* updating all sbuf field as necessary
366	*/
367	int
368	sowrite(PNATState pData, struct socket *so)
369	{
370	int n,nn;
371	struct sbuf *sb = &so->so_rcv;
372	size_t len = sb->sb_cc;
373	struct iovec iov[2];
374
375	DEBUG_CALL("sowrite");
376	DEBUG_ARG("so = %lx", (long)so);
377	QSOCKET_LOCK(tcb);
378	SOCKET_LOCK(so);
379	QSOCKET_UNLOCK(tcb);
380	if (so->so_urgc)
381	{
382	sosendoob(so);
383	if (sb->sb_cc == 0)
384	{
385	SOCKET_UNLOCK(so);
386	return 0;
387	}
388	}
389
390	/*
391	* No need to check if there's something to write,
392	* sowrite wouldn't have been called otherwise
393	*/
394
395	len = sb->sb_cc;
396
397	iov[0].iov_base = sb->sb_rptr;
398	iov[1].iov_base = 0;
399	iov[1].iov_len = 0;
400	if (sb->sb_rptr < sb->sb_wptr)
401	{
402	iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
403	/* Should never succeed, but... */
404	if (iov[0].iov_len > len)
405	iov[0].iov_len = len;
406	n = 1;
407	}
408	else
409	{
410	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
411	if (iov[0].iov_len > len)
412	iov[0].iov_len = len;
413	len -= iov[0].iov_len;
414	if (len)
415	{
416	iov[1].iov_base = sb->sb_data;
417	iov[1].iov_len = sb->sb_wptr - sb->sb_data;
418	if (iov[1].iov_len > len)
419	iov[1].iov_len = len;
420	n = 2;
421	}
422	else
423	n = 1;
424	}
425	/* Check if there's urgent data to send, and if so, send it */
426	#ifdef HAVE_READV
427	nn = writev(so->s, (const struct iovec *)iov, n);
428	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
429	#else
430	nn = send(so->s, iov[0].iov_base, iov[0].iov_len, 0);
431	#endif
432	/* This should never happen, but people tell me it does shrug */
433	if (nn < 0 && (errno == EAGAIN \|\| errno == EINTR \|\| errno == EWOULDBLOCK))
434	{
435	SOCKET_UNLOCK(so);
436	return 0;
437	}
438
439	if (nn < 0 \|\| (nn == 0 && iov[0].iov_len > 0))
440	{
441	DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
442	so->so_state, errno));
443	sofcantsendmore(so);
444	tcp_sockclosed(pData, sototcpcb(so));
445	SOCKET_UNLOCK(so);
446	return -1;
447	}
448
449	#ifndef HAVE_READV
450	if (n == 2 && nn == iov[0].iov_len)
451	{
452	int ret;
453	ret = send(so->s, iov[1].iov_base, iov[1].iov_len,0);
454	if (ret > 0)
455	nn += ret;
456	}
457	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
458	#endif
459
460	/* Update sbuf */
461	sb->sb_cc -= nn;
462	sb->sb_rptr += nn;
463	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
464	sb->sb_rptr -= sb->sb_datalen;
465
466	/*
467	* If in DRAIN mode, and there's no more data, set
468	* it CANTSENDMORE
469	*/
470	if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
471	sofcantsendmore(so);
472
473	SOCKET_UNLOCK(so);
474	return nn;
475	}
476
477	/*
478	* recvfrom() a UDP socket
479	*/
480	void
481	sorecvfrom(PNATState pData, struct socket *so)
482	{
483	struct sockaddr_in addr;
484	socklen_t addrlen = sizeof(struct sockaddr_in);
485
486	DEBUG_CALL("sorecvfrom");
487	DEBUG_ARG("so = %lx", (long)so);
488
489	if (so->so_type == IPPROTO_ICMP)
490	{
491	/* This is a "ping" reply */
492	#ifdef RT_OS_WINDOWS
493	sorecvfrom_icmp_win(pData, so);
494	#else /* RT_OS_WINDOWS */
495	sorecvfrom_icmp_unix(pData, so);
496	#endif /* !RT_OS_WINDOWS */
497	udp_detach(pData, so);
498	}
499	else
500	{
501	/* A "normal" UDP packet */
502	struct mbuf *m;
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	#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
517	m->m_data += sizeof(struct udphdr)
518	+ sizeof(struct ip); /XXX: no options atm/
519	#endif
520
521	/*
522	* XXX Shouldn't FIONREAD packets destined for port 53,
523	* but I don't know the max packet size for DNS lookups
524	*/
525	len = M_FREEROOM(m);
526	/* if (so->so_fport != htons(53)) */
527	{
528	ioctlsocket(so->s, FIONREAD, &n);
529
530	if (n > len)
531	{
532	n = (m->m_data - m->m_dat) + m->m_len + n + 1;
533	m_inc(m, n);
534	len = M_FREEROOM(m);
535	}
536	}
537
538	m->m_len = recvfrom(so->s, m->m_data, len, 0,
539	(struct sockaddr *)&addr, &addrlen);
540	Log2((" did recvfrom %d, errno = %d-%s\n",
541	m->m_len, errno,strerror(errno)));
542	if(m->m_len < 0)
543	{
544	u_char code = ICMP_UNREACH_PORT;
545
546	if (errno == EHOSTUNREACH)
547	code = ICMP_UNREACH_HOST;
548	else if(errno == ENETUNREACH)
549	code = ICMP_UNREACH_NET;
550
551	Log2((dfd," rx error, tx icmp ICMP_UNREACH:%i\n", code));
552	icmp_error(pData, so->so_m, ICMP_UNREACH,code, 0,strerror(errno));
553	m_free(pData, m);
554	}
555	else
556	{
557	/*
558	* Hack: domain name lookup will be used the most for UDP,
559	* and since they'll only be used once there's no need
560	* for the 4 minute (or whatever) timeout... So we time them
561	* out much quicker (10 seconds for now...)
562	*/
563	if (so->so_expire)
564	{
565	if (so->so_fport == htons(53))
566	so->so_expire = curtime + SO_EXPIREFAST;
567	else
568	so->so_expire = curtime + SO_EXPIRE;
569	}
570
571	#if 0
572	if (m->m_len == len)
573	{
574	m_inc(m, MINCSIZE);
575	m->m_len = 0;
576	}
577	#endif
578
579	/*
580	* If this packet was destined for CTL_ADDR,
581	* make it look like that's where it came from, done by udp_output
582	*/
583	udp_output(pData, so, m, &addr);
584	SOCKET_UNLOCK(so);
585	} /* rx error */
586	} /* if ping packet */
587	}
588
589	/*
590	* sendto() a socket
591	*/
592	int
593	sosendto(PNATState pData, struct socket so, struct mbuf m)
594	{
595	int ret;
596	struct sockaddr_in addr;
597	#if 0
598	struct sockaddr_in host_addr;
599	#endif
600
601	DEBUG_CALL("sosendto");
602	DEBUG_ARG("so = %lx", (long)so);
603	DEBUG_ARG("m = %lx", (long)m);
604
605	addr.sin_family = AF_INET;
606	if ((so->so_faddr.s_addr & htonl(pData->netmask)) == special_addr.s_addr)
607	{
608	/* It's an alias */
609	uint32_t last_byte = ntohl(so->so_faddr.s_addr) & ~pData->netmask;
610	switch(last_byte)
611	{
612	#if 0
613	/* handle this case at 'default:' */
614	case CTL_BROADCAST:
615	addr.sin_addr.s_addr = INADDR_BROADCAST;
616	/* Send the packet to host to fully emulate broadcast */
617	/** @todo r=klaus: on Linux host this causes the host to receive
618	* the packet twice for some reason. And I cannot find any place
619	* in the man pages which states that sending a broadcast does not
620	* reach the host itself. */
621	host_addr.sin_family = AF_INET;
622	host_addr.sin_port = so->so_fport;
623	host_addr.sin_addr = our_addr;
624	sendto(so->s, m->m_data, m->m_len, 0,
625	(struct sockaddr *)&host_addr, sizeof (struct sockaddr));
626	break;
627	#endif
628	case CTL_DNS:
629	#ifndef VBOX_WITH_MULTI_DNS
630	if (!get_dns_addr(pData, &dns_addr))
631	addr.sin_addr = dns_addr;
632	else
633	addr.sin_addr = loopback_addr;
634	break;
635	#endif
636	case CTL_ALIAS:
637	default:
638	if (last_byte == ~pData->netmask)
639	addr.sin_addr.s_addr = INADDR_BROADCAST;
640	else
641	addr.sin_addr = loopback_addr;
642	break;
643	}
644	}
645	else
646	addr.sin_addr = so->so_faddr;
647	addr.sin_port = so->so_fport;
648
649	DEBUG_MISC((dfd, " sendto()ing, addr.sin_port=%d, addr.sin_addr.s_addr=%.16s\n",
650	ntohs(addr.sin_port), inet_ntoa(addr.sin_addr)));
651
652	/* Don't care what port we get */
653	ret = sendto(so->s, m->m_data, m->m_len, 0,
654	(struct sockaddr *)&addr, sizeof (struct sockaddr));
655	if (ret < 0)
656	{
657	LogRel(("UDP: sendto fails (%s)\n", strerror(errno)));
658	return -1;
659	}
660
661	/*
662	* Kill the socket if there's no reply in 4 minutes,
663	* but only if it's an expirable socket
664	*/
665	if (so->so_expire)
666	so->so_expire = curtime + SO_EXPIRE;
667	so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
668	return 0;
669	}
670
671	/*
672	* XXX This should really be tcp_listen
673	*/
674	struct socket *
675	solisten(PNATState pData, u_int port, u_int32_t laddr, u_int lport, int flags)
676	{
677	struct sockaddr_in addr;
678	struct socket *so;
679	socklen_t addrlen = sizeof(addr);
680	int s, opt = 1;
681
682	DEBUG_CALL("solisten");
683	DEBUG_ARG("port = %d", port);
684	DEBUG_ARG("laddr = %x", laddr);
685	DEBUG_ARG("lport = %d", lport);
686	DEBUG_ARG("flags = %x", flags);
687
688	if ((so = socreate()) == NULL)
689	{
690	/* RTMemFree(so); Not sofree() ??? free(NULL) == NOP */
691	return NULL;
692	}
693
694	/* Don't tcp_attach... we don't need so_snd nor so_rcv */
695	if ((so->so_tcpcb = tcp_newtcpcb(pData, so)) == NULL)
696	{
697	RTMemFree(so);
698	return NULL;
699	}
700
701	SOCKET_LOCK_CREATE(so);
702	SOCKET_LOCK(so);
703	QSOCKET_LOCK(tcb);
704	insque(pData, so,&tcb);
705	NSOCK_INC();
706	QSOCKET_UNLOCK(tcb);
707
708	/*
709	* SS_FACCEPTONCE sockets must time out.
710	*/
711	if (flags & SS_FACCEPTONCE)
712	so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
713
714	so->so_state = (SS_FACCEPTCONN\|flags);
715	so->so_lport = lport; /* Kept in network format */
716	so->so_laddr.s_addr = laddr; /* Ditto */
717
718	addr.sin_family = AF_INET;
719	addr.sin_addr.s_addr = INADDR_ANY;
720	addr.sin_port = port;
721
722	if ( ((s = socket(AF_INET,SOCK_STREAM,0)) < 0)
723	\|\| (setsockopt(s,SOL_SOCKET,SO_REUSEADDR,(char *)&opt,sizeof(int)) < 0)
724	\|\| (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0)
725	\|\| (listen(s,1) < 0))
726	{
727	#ifdef RT_OS_WINDOWS
728	int tmperrno = WSAGetLastError(); /* Don't clobber the real reason we failed */
729	closesocket(s);
730	QSOCKET_LOCK(tcb);
731	sofree(pData, so);
732	QSOCKET_UNLOCK(tcb);
733	/* Restore the real errno */
734	WSASetLastError(tmperrno);
735	#else
736	int tmperrno = errno; /* Don't clobber the real reason we failed */
737	close(s);
738	QSOCKET_LOCK(tcb);
739	sofree(pData, so);
740	QSOCKET_UNLOCK(tcb);
741	/* Restore the real errno */
742	errno = tmperrno;
743	#endif
744	return NULL;
745	}
746	setsockopt(s,SOL_SOCKET,SO_OOBINLINE,(char *)&opt,sizeof(int));
747
748	getsockname(s,(struct sockaddr *)&addr,&addrlen);
749	so->so_fport = addr.sin_port;
750	if (addr.sin_addr.s_addr == 0 \|\| addr.sin_addr.s_addr == loopback_addr.s_addr)
751	so->so_faddr = alias_addr;
752	else
753	so->so_faddr = addr.sin_addr;
754
755	so->s = s;
756	SOCKET_UNLOCK(so);
757	return so;
758	}
759
760	/*
761	* Data is available in so_rcv
762	* Just write() the data to the socket
763	* XXX not yet...
764	*/
765	void
766	sorwakeup(struct socket *so)
767	{
768	#if 0
769	sowrite(so);
770	FD_CLR(so->s,&writefds);
771	#endif
772	}
773
774	/*
775	* Data has been freed in so_snd
776	* We have room for a read() if we want to
777	* For now, don't read, it'll be done in the main loop
778	*/
779	void
780	sowwakeup(struct socket *so)
781	{
782	}
783
784	/*
785	* Various session state calls
786	* XXX Should be #define's
787	* The socket state stuff needs work, these often get call 2 or 3
788	* times each when only 1 was needed
789	*/
790	void
791	soisfconnecting(struct socket *so)
792	{
793	so->so_state &= ~(SS_NOFDREF\|SS_ISFCONNECTED\|SS_FCANTRCVMORE\|
794	SS_FCANTSENDMORE\|SS_FWDRAIN);
795	so->so_state \|= SS_ISFCONNECTING; /* Clobber other states */
796	}
797
798	void
799	soisfconnected(struct socket *so)
800	{
801	so->so_state &= ~(SS_ISFCONNECTING\|SS_FWDRAIN\|SS_NOFDREF);
802	so->so_state \|= SS_ISFCONNECTED; /* Clobber other states */
803	}
804
805	void
806	sofcantrcvmore(struct socket *so)
807	{
808	if ((so->so_state & SS_NOFDREF) == 0)
809	{
810	shutdown(so->s,0);
811	}
812	so->so_state &= ~(SS_ISFCONNECTING);
813	if (so->so_state & SS_FCANTSENDMORE)
814	so->so_state = SS_NOFDREF; /* Don't select it */
815	/* XXX close() here as well? */
816	else
817	so->so_state \|= SS_FCANTRCVMORE;
818	}
819
820	void
821	sofcantsendmore(struct socket *so)
822	{
823	if ((so->so_state & SS_NOFDREF) == 0)
824	shutdown(so->s, 1); /* send FIN to fhost */
825
826	so->so_state &= ~(SS_ISFCONNECTING);
827	if (so->so_state & SS_FCANTRCVMORE)
828	so->so_state = SS_NOFDREF; /* as above */
829	else
830	so->so_state \|= SS_FCANTSENDMORE;
831	}
832
833	void
834	soisfdisconnected(struct socket *so)
835	{
836	#if 0
837	so->so_state &= ~(SS_ISFCONNECTING\|SS_ISFCONNECTED);
838	close(so->s);
839	so->so_state = SS_ISFDISCONNECTED;
840	/*
841	* XXX Do nothing ... ?
842	*/
843	#endif
844	}
845
846	/*
847	* Set write drain mode
848	* Set CANTSENDMORE once all data has been write()n
849	*/
850	void
851	sofwdrain(struct socket *so)
852	{
853	if (so->so_rcv.sb_cc)
854	so->so_state \|= SS_FWDRAIN;
855	else
856	sofcantsendmore(so);
857	}
858
859	static void
860	send_icmp_to_guest(PNATState pData, char buff, size_t len, struct socket so, const struct sockaddr_in *addr)
861	{
862	struct ip *ip;
863	uint32_t dst,src;
864	char ip_copy[256];
865	struct icmp *icp;
866	int old_ip_len = 0;
867	int hlen, original_hlen = 0;
868	struct mbuf *m;
869	struct icmp_msg *icm;
870	uint8_t proto;
871	int type = 0;
872
873	ip = (struct ip *)buff;
874	hlen = (ip->ip_hl << 2);
875	icp = (struct icmp )((char )ip + hlen);
876
877	Log(("ICMP:received msg(t:%d, c:%d)\n", icp->icmp_type, icp->icmp_code));
878	if ( icp->icmp_type != ICMP_ECHOREPLY
879	&& icp->icmp_type != ICMP_TIMXCEED
880	&& icp->icmp_type != ICMP_UNREACH)
881	{
882	return;
883	}
884
885	type = icp->icmp_type;
886	if ( type == ICMP_TIMXCEED
887	\|\| type == ICMP_UNREACH)
888	{
889	ip = &icp->icmp_ip;
890	DO_ALIAS(&ip->ip_dst);
891	}
892	else
893	{
894	DO_ALIAS(&ip->ip_src);
895	}
896
897	icm = icmp_find_original_mbuf(pData, ip);
898
899	if (icm == NULL)
900	{
901	Log(("NAT: Can't find the corresponding packet for the received ICMP\n"));
902	return;
903	}
904
905	m = icm->im_m;
906	Assert(m != NULL);
907
908	src = addr->sin_addr.s_addr;
909
910	ip = mtod(m, struct ip *);
911	proto = ip->ip_p;
912	/* Now ip is pointing on header we've sent from guest */
913	if ( icp->icmp_type == ICMP_TIMXCEED
914	\|\| icp->icmp_type == ICMP_UNREACH)
915	{
916	old_ip_len = (ip->ip_hl << 2) + 64;
917	if (old_ip_len > sizeof(ip_copy))
918	old_ip_len = sizeof(ip_copy);
919	memcpy(ip_copy, ip, old_ip_len);
920	}
921
922	/* source address from original IP packet*/
923	dst = ip->ip_src.s_addr;
924
925	/* overide ther tail of old packet */
926	ip = mtod(m, struct ip ); / ip is from mbuf we've overrided */
927	original_hlen = ip->ip_hl << 2;
928	/* saves original ip header and options */
929	memcpy(m->m_data + original_hlen, buff + hlen, len - hlen);
930	m->m_len = len - hlen + original_hlen;
931	ip->ip_len = m->m_len;
932	ip->ip_p = IPPROTO_ICMP; /* the original package could be whatever, but we're response via ICMP*/
933
934	icp = (struct icmp )((char )ip + (ip->ip_hl << 2));
935	type = icp->icmp_type;
936	if ( type == ICMP_TIMXCEED
937	\|\| type == ICMP_UNREACH)
938	{
939	/* according RFC 793 error messages required copy of initial IP header + 64 bit */
940	memcpy(&icp->icmp_ip, ip_copy, old_ip_len);
941	ip->ip_tos = ((ip->ip_tos & 0x1E) \| 0xC0); /* high priority for errors */
942	}
943
944	ip->ip_src.s_addr = src;
945	ip->ip_dst.s_addr = dst;
946	icmp_reflect(pData, m);
947	LIST_REMOVE(icm, im_list);
948	/* Don't call m_free here*/
949
950	if ( type == ICMP_TIMXCEED
951	\|\| type == ICMP_UNREACH)
952	{
953	icm->im_so->so_m = NULL;
954	switch (proto)
955	{
956	case IPPROTO_UDP:
957	/XXX: so->so_m already freed so we shouldn't call sofree /
958	udp_detach(pData, icm->im_so);
959	break;
960	case IPPROTO_TCP:
961	/close tcp should be here /
962	break;
963	default:
964	/* do nothing */
965	break;
966	}
967	}
968	RTMemFree(icm);
969	}
970
971	#ifdef RT_OS_WINDOWS
972	static void
973	sorecvfrom_icmp_win(PNATState pData, struct socket *so)
974	{
975	int len;
976	int i;
977	struct ip *ip;
978	struct mbuf *m;
979	struct icmp *icp;
980	struct icmp_msg *icm;
981	struct ip ip_broken; / ICMP returns header + 64 bit of packet */
982	uint32_t src;
983	ICMP_ECHO_REPLY *icr;
984	int hlen = 0;
985	int data_len = 0;
986	int nbytes = 0;
987	u_char code = ~0;
988
989	len = pData->pfIcmpParseReplies(pData->pvIcmpBuffer, pData->szIcmpBuffer);
990	#ifndef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
991	fIcmp = 0; /* reply processed */
992	#endif
993	if (len < 0)
994	{
995	LogRel(("NAT: Error (%d) occurred on ICMP receiving\n", GetLastError()));
996	return;
997	}
998	if (len == 0)
999	return; /* no error */
1000
1001	icr = (ICMP_ECHO_REPLY *)pData->pvIcmpBuffer;
1002	for (i = 0; i < len; ++i)
1003	{
1004	switch(icr[i].Status)
1005	{
1006	case IP_DEST_HOST_UNREACHABLE:
1007	code = (code != ~0 ? code : ICMP_UNREACH_HOST);
1008	case IP_DEST_NET_UNREACHABLE:
1009	code = (code != ~0 ? code : ICMP_UNREACH_NET);
1010	case IP_DEST_PROT_UNREACHABLE:
1011	code = (code != ~0 ? code : ICMP_UNREACH_PROTOCOL);
1012	/* UNREACH error inject here */
1013	case IP_DEST_PORT_UNREACHABLE:
1014	code = (code != ~0 ? code : ICMP_UNREACH_PORT);
1015	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, "Error occurred!!!");
1016	break;
1017	case IP_SUCCESS: /* echo replied */
1018	m = m_get(pData);
1019	m->m_data += if_maxlinkhdr;
1020	ip = mtod(m, struct ip *);
1021	ip->ip_src.s_addr = icr[i].Address;
1022	DO_ALIAS(&ip->ip_src);
1023	ip->ip_p = IPPROTO_ICMP;
1024	ip->ip_dst.s_addr = so->so_laddr.s_addr; /XXX: still the hack/
1025	data_len = sizeof(struct ip);
1026	ip->ip_hl = data_len >> 2; /* requiered for icmp_reflect, no IP options */
1027	ip->ip_ttl = icr[i].Options.Ttl;
1028
1029	icp = (struct icmp )&ip[1]; / no options */
1030	icp->icmp_type = ICMP_ECHOREPLY;
1031	icp->icmp_code = 0;
1032	icp->icmp_id = so->so_icmp_id;
1033	icp->icmp_seq = so->so_icmp_seq;
1034
1035	data_len += ICMP_MINLEN;
1036
1037	nbytes = (data_len + icr[i].DataSize > m->m_size? m->m_size - data_len: icr[i].DataSize);
1038	memcpy(icp->icmp_data, icr[i].Data, nbytes);
1039
1040	data_len += icr[i].DataSize;
1041
1042	ip->ip_len = data_len;
1043	m->m_len = ip->ip_len;
1044
1045	icmp_reflect(pData, m);
1046	break;
1047	case IP_TTL_EXPIRED_TRANSIT: /* TTL expired */
1048
1049	ip_broken = icr[i].Data;
1050	icm = icmp_find_original_mbuf(pData, ip_broken);
1051	if (icm == NULL) {
1052	Log(("ICMP: can't find original package (first double word %x)\n", (uint32_t )ip_broken));
1053	return;
1054	}
1055	m = icm->im_m;
1056	ip = mtod(m, struct ip *);
1057	ip->ip_ttl = icr[i].Options.Ttl;
1058	src = ip->ip_src.s_addr;
1059	ip->ip_dst.s_addr = src;
1060	ip->ip_dst.s_addr = icr[i].Address;
1061
1062	hlen = (ip->ip_hl << 2);
1063	icp = (struct icmp )((char )ip + hlen);
1064	ip_broken->ip_src.s_addr = src; /it packet sent from host not from guest/
1065	data_len = (ip_broken->ip_hl << 2) + 64;
1066
1067	nbytes =(hlen + ICMP_MINLEN + data_len > m->m_size? m->m_size - (hlen + ICMP_MINLEN): data_len);
1068	memcpy(icp->icmp_data, ip_broken, nbytes);
1069	icmp_reflect(pData, m);
1070	break;
1071	default:
1072	Log(("ICMP(default): message with Status: %x was received from %x\n", icr[i].Status, icr[i].Address));
1073	break;
1074	}
1075	}
1076	}
1077	#else /* RT_OS_WINDOWS */
1078	static void sorecvfrom_icmp_unix(PNATState pData, struct socket *so)
1079	{
1080	struct sockaddr_in addr;
1081	socklen_t addrlen = sizeof(struct sockaddr_in);
1082	char buff[1500];
1083	int len;
1084	len = recvfrom(so->s, buff, 1500, 0,
1085	(struct sockaddr *)&addr, &addrlen);
1086	/* XXX Check if reply is "correct"? */
1087
1088	if (len == -1 \|\| len == 0)
1089	{
1090	u_char code = ICMP_UNREACH_PORT;
1091
1092	if (errno == EHOSTUNREACH)
1093	code = ICMP_UNREACH_HOST;
1094	else if(errno == ENETUNREACH)
1095	code = ICMP_UNREACH_NET;
1096
1097	DEBUG_MISC((dfd," udp icmp rx errno = %d-%s\n",
1098	errno,strerror(errno)));
1099	icmp_error(pData, so->so_m, ICMP_UNREACH,code, 0,strerror(errno));
1100	}
1101	else
1102	{
1103	send_icmp_to_guest(pData, buff, len, so, &addr);
1104	}
1105	}
1106	#endif /* !RT_OS_WINDOWS */

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/socket.c@ 16688

Download in other formats: