socket.c@ 27764

Last change on this file since 27764 was 27573, checked in by vboxsync, 15 years ago
NAT: cosmetics; refined comments
Property svn:eol-style set to `native`
File size: 41.3 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	#include <VBox/pdmdrv.h>
16	#if defined (RT_OS_WINDOWS)
17	#include <iphlpapi.h>
18	#include <icmpapi.h>
19	#endif
20
21
22	static void send_icmp_to_guest(PNATState, char , size_t, struct socket , const struct sockaddr_in *);
23	#ifdef RT_OS_WINDOWS
24	static void sorecvfrom_icmp_win(PNATState, struct socket *);
25	#else /* RT_OS_WINDOWS */
26	static void sorecvfrom_icmp_unix(PNATState, struct socket *);
27	#endif /* !RT_OS_WINDOWS */
28
29	void
30	so_init()
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
126	STAM_PROFILE_START(&pData->StatIOread, a);
127	STAM_COUNTER_RESET(&pData->StatIORead_in_1);
128	STAM_COUNTER_RESET(&pData->StatIORead_in_2);
129
130	QSOCKET_LOCK(tcb);
131	SOCKET_LOCK(so);
132	QSOCKET_UNLOCK(tcb);
133
134	DEBUG_CALL("soread");
135	DEBUG_ARG("so = %lx", (long)so);
136
137	/*
138	* No need to check if there's enough room to read.
139	* soread wouldn't have been called if there weren't
140	*/
141
142	len = sb->sb_datalen - sb->sb_cc;
143
144	iov[0].iov_base = sb->sb_wptr;
145	iov[1].iov_base = 0;
146	iov[1].iov_len = 0;
147	if (sb->sb_wptr < sb->sb_rptr)
148	{
149	iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
150	/* Should never succeed, but... */
151	if (iov[0].iov_len > len)
152	iov[0].iov_len = len;
153	if (iov[0].iov_len > mss)
154	iov[0].iov_len -= iov[0].iov_len%mss;
155	n = 1;
156	}
157	else
158	{
159	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
160	/* Should never succeed, but... */
161	if (iov[0].iov_len > len)
162	iov[0].iov_len = len;
163	len -= iov[0].iov_len;
164	if (len)
165	{
166	iov[1].iov_base = sb->sb_data;
167	iov[1].iov_len = sb->sb_rptr - sb->sb_data;
168	if (iov[1].iov_len > len)
169	iov[1].iov_len = len;
170	total = iov[0].iov_len + iov[1].iov_len;
171	if (total > mss)
172	{
173	lss = total % mss;
174	if (iov[1].iov_len > lss)
175	{
176	iov[1].iov_len -= lss;
177	n = 2;
178	}
179	else
180	{
181	lss -= iov[1].iov_len;
182	iov[0].iov_len -= lss;
183	n = 1;
184	}
185	}
186	else
187	n = 2;
188	}
189	else
190	{
191	if (iov[0].iov_len > mss)
192	iov[0].iov_len -= iov[0].iov_len%mss;
193	n = 1;
194	}
195	}
196
197	#ifdef HAVE_READV
198	nn = readv(so->s, (struct iovec *)iov, n);
199	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
200	#else
201	nn = recv(so->s, iov[0].iov_base, iov[0].iov_len, 0);
202	#endif
203	if (nn <= 0)
204	{
205	#ifdef RT_OS_WINDOWS
206	/*
207	* Special case for WSAEnumNetworkEvents: If we receive 0 bytes that
208	* _could_ mean that the connection is closed. But we will receive an
209	* FD_CLOSE event later if the connection was _really_ closed. With
210	* www.youtube.com I see this very often. Closing the socket too early
211	* would be dangerous.
212	*/
213	int status, ignored;
214	unsigned long pending = 0;
215	status = WSAIoctl(so->s, FIONREAD, NULL, 0, &pending, sizeof(unsigned long), &ignored, NULL, NULL);
216	if (status < 0)
217	LogRel(("NAT:error in WSAIoctl: %d\n", WSAGetLastError()));
218	if (nn == 0 && (pending != 0))
219	{
220	SOCKET_UNLOCK(so);
221	STAM_PROFILE_STOP(&pData->StatIOread, a);
222	return 0;
223	}
224	#endif
225	if ( nn < 0
226	&& ( errno == EINTR
227	\|\| errno == EAGAIN
228	\|\| errno == EWOULDBLOCK))
229	{
230	SOCKET_UNLOCK(so);
231	STAM_PROFILE_STOP(&pData->StatIOread, a);
232	return 0;
233	}
234	else
235	{
236	/* nn == 0 means peer has performed an orderly shutdown */
237	DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n",
238	nn, errno, strerror(errno)));
239	sofcantrcvmore(so);
240	tcp_sockclosed(pData, sototcpcb(so));
241	SOCKET_UNLOCK(so);
242	STAM_PROFILE_STOP(&pData->StatIOread, a);
243	return -1;
244	}
245	}
246	STAM_STATS(
247	if (n == 1)
248	{
249	STAM_COUNTER_INC(&pData->StatIORead_in_1);
250	STAM_COUNTER_ADD(&pData->StatIORead_in_1_bytes, nn);
251	}
252	else
253	{
254	STAM_COUNTER_INC(&pData->StatIORead_in_2);
255	STAM_COUNTER_ADD(&pData->StatIORead_in_2_1st_bytes, nn);
256	}
257	);
258
259	#ifndef HAVE_READV
260	/*
261	* If there was no error, try and read the second time round
262	* We read again if n = 2 (ie, there's another part of the buffer)
263	* and we read as much as we could in the first read
264	* We don't test for <= 0 this time, because there legitimately
265	* might not be any more data (since the socket is non-blocking),
266	* a close will be detected on next iteration.
267	* A return of -1 wont (shouldn't) happen, since it didn't happen above
268	*/
269	if (n == 2 && nn == iov[0].iov_len)
270	{
271	int ret;
272	ret = recv(so->s, iov[1].iov_base, iov[1].iov_len, 0);
273	if (ret > 0)
274	nn += ret;
275	STAM_STATS(
276	if (ret > 0)
277	{
278	STAM_COUNTER_INC(&pData->StatIORead_in_2);
279	STAM_COUNTER_ADD(&pData->StatIORead_in_2_2nd_bytes, ret);
280	}
281	);
282	}
283
284	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
285	#endif
286
287	/* Update fields */
288	sb->sb_cc += nn;
289	sb->sb_wptr += nn;
290	if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
291	sb->sb_wptr -= sb->sb_datalen;
292	STAM_PROFILE_STOP(&pData->StatIOread, a);
293	SOCKET_UNLOCK(so);
294	return nn;
295	}
296
297	/*
298	* Get urgent data
299	*
300	* When the socket is created, we set it SO_OOBINLINE,
301	* so when OOB data arrives, we soread() it and everything
302	* in the send buffer is sent as urgent data
303	*/
304	void
305	sorecvoob(PNATState pData, struct socket *so)
306	{
307	struct tcpcb *tp = sototcpcb(so);
308
309	DEBUG_CALL("sorecvoob");
310	DEBUG_ARG("so = %lx", (long)so);
311
312	/*
313	* We take a guess at how much urgent data has arrived.
314	* In most situations, when urgent data arrives, the next
315	* read() should get all the urgent data. This guess will
316	* be wrong however if more data arrives just after the
317	* urgent data, or the read() doesn't return all the
318	* urgent data.
319	*/
320	soread(pData, so);
321	tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
322	tp->t_force = 1;
323	tcp_output(pData, tp);
324	tp->t_force = 0;
325	}
326
327	/*
328	* Send urgent data
329	* There's a lot duplicated code here, but...
330	*/
331	int
332	sosendoob(struct socket *so)
333	{
334	struct sbuf *sb = &so->so_rcv;
335	char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
336
337	int n, len;
338
339	DEBUG_CALL("sosendoob");
340	DEBUG_ARG("so = %lx", (long)so);
341	DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
342
343	if (so->so_urgc > sizeof(buff))
344	so->so_urgc = sizeof(buff); /* XXX */
345
346	if (sb->sb_rptr < sb->sb_wptr)
347	{
348	/* We can send it directly */
349	n = send(so->s, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
350	so->so_urgc -= n;
351
352	DEBUG_MISC((dfd, " --- sent %d bytes urgent data, %d urgent bytes left\n",
353	n, so->so_urgc));
354	}
355	else
356	{
357	/*
358	* Since there's no sendv or sendtov like writev,
359	* we must copy all data to a linear buffer then
360	* send it all
361	*/
362	len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
363	if (len > so->so_urgc)
364	len = so->so_urgc;
365	memcpy(buff, sb->sb_rptr, len);
366	so->so_urgc -= len;
367	if (so->so_urgc)
368	{
369	n = sb->sb_wptr - sb->sb_data;
370	if (n > so->so_urgc)
371	n = so->so_urgc;
372	memcpy(buff + len, sb->sb_data, n);
373	so->so_urgc -= n;
374	len += n;
375	}
376	n = send(so->s, buff, len, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
377	#ifdef DEBUG
378	if (n != len)
379	DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
380	#endif
381	DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n",
382	n, so->so_urgc));
383	}
384
385	sb->sb_cc -= n;
386	sb->sb_rptr += n;
387	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
388	sb->sb_rptr -= sb->sb_datalen;
389
390	return n;
391	}
392
393	/*
394	* Write data from so_rcv to so's socket,
395	* updating all sbuf field as necessary
396	*/
397	int
398	sowrite(PNATState pData, struct socket *so)
399	{
400	int n, nn;
401	struct sbuf *sb = &so->so_rcv;
402	size_t len = sb->sb_cc;
403	struct iovec iov[2];
404
405	STAM_PROFILE_START(&pData->StatIOwrite, a);
406	STAM_COUNTER_RESET(&pData->StatIOWrite_in_1);
407	STAM_COUNTER_RESET(&pData->StatIOWrite_in_1_bytes);
408	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2);
409	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2_1st_bytes);
410	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2_2nd_bytes);
411	STAM_COUNTER_RESET(&pData->StatIOWrite_no_w);
412	STAM_COUNTER_RESET(&pData->StatIOWrite_rest);
413	STAM_COUNTER_RESET(&pData->StatIOWrite_rest_bytes);
414	DEBUG_CALL("sowrite");
415	DEBUG_ARG("so = %lx", (long)so);
416	QSOCKET_LOCK(tcb);
417	SOCKET_LOCK(so);
418	QSOCKET_UNLOCK(tcb);
419	if (so->so_urgc)
420	{
421	sosendoob(so);
422	if (sb->sb_cc == 0)
423	{
424	SOCKET_UNLOCK(so);
425	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
426	return 0;
427	}
428	}
429
430	/*
431	* No need to check if there's something to write,
432	* sowrite wouldn't have been called otherwise
433	*/
434
435	len = sb->sb_cc;
436
437	iov[0].iov_base = sb->sb_rptr;
438	iov[1].iov_base = 0;
439	iov[1].iov_len = 0;
440	if (sb->sb_rptr < sb->sb_wptr)
441	{
442	iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
443	/* Should never succeed, but... */
444	if (iov[0].iov_len > len)
445	iov[0].iov_len = len;
446	n = 1;
447	}
448	else
449	{
450	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
451	if (iov[0].iov_len > len)
452	iov[0].iov_len = len;
453	len -= iov[0].iov_len;
454	if (len)
455	{
456	iov[1].iov_base = sb->sb_data;
457	iov[1].iov_len = sb->sb_wptr - sb->sb_data;
458	if (iov[1].iov_len > len)
459	iov[1].iov_len = len;
460	n = 2;
461	}
462	else
463	n = 1;
464	}
465	STAM_STATS({
466	if (n == 1)
467	{
468	STAM_COUNTER_INC(&pData->StatIOWrite_in_1);
469	STAM_COUNTER_ADD(&pData->StatIOWrite_in_1_bytes, iov[0].iov_len);
470	}
471	else
472	{
473	STAM_COUNTER_INC(&pData->StatIOWrite_in_2);
474	STAM_COUNTER_ADD(&pData->StatIOWrite_in_2_1st_bytes, iov[0].iov_len);
475	STAM_COUNTER_ADD(&pData->StatIOWrite_in_2_2nd_bytes, iov[1].iov_len);
476	}
477	});
478	/* Check if there's urgent data to send, and if so, send it */
479	#ifdef HAVE_READV
480	nn = writev(so->s, (const struct iovec *)iov, n);
481	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
482	#else
483	nn = send(so->s, iov[0].iov_base, iov[0].iov_len, 0);
484	#endif
485	/* This should never happen, but people tell me it does shrug */
486	if ( nn < 0
487	&& ( errno == EAGAIN
488	\|\| errno == EINTR
489	\|\| errno == EWOULDBLOCK))
490	{
491	SOCKET_UNLOCK(so);
492	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
493	return 0;
494	}
495
496	if (nn < 0 \|\| (nn == 0 && iov[0].iov_len > 0))
497	{
498	DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
499	so->so_state, errno));
500	sofcantsendmore(so);
501	tcp_sockclosed(pData, sototcpcb(so));
502	SOCKET_UNLOCK(so);
503	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
504	return -1;
505	}
506
507	#ifndef HAVE_READV
508	if (n == 2 && nn == iov[0].iov_len)
509	{
510	int ret;
511	ret = send(so->s, iov[1].iov_base, iov[1].iov_len, 0);
512	if (ret > 0)
513	nn += ret;
514	STAM_STATS({
515	if (ret > 0 && ret != iov[1].iov_len)
516	{
517	STAM_COUNTER_INC(&pData->StatIOWrite_rest);
518	STAM_COUNTER_ADD(&pData->StatIOWrite_rest_bytes, (ret - iov[1].iov_len));
519	}
520	});
521	}
522	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
523	#endif
524
525	/* Update sbuf */
526	sb->sb_cc -= nn;
527	sb->sb_rptr += nn;
528	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
529	sb->sb_rptr -= sb->sb_datalen;
530
531	/*
532	* If in DRAIN mode, and there's no more data, set
533	* it CANTSENDMORE
534	*/
535	if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
536	sofcantsendmore(so);
537
538	SOCKET_UNLOCK(so);
539	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
540	return nn;
541	}
542
543	/*
544	* recvfrom() a UDP socket
545	*/
546	void
547	sorecvfrom(PNATState pData, struct socket *so)
548	{
549	ssize_t ret = 0;
550	struct sockaddr_in addr;
551	socklen_t addrlen = sizeof(struct sockaddr_in);
552
553	DEBUG_CALL("sorecvfrom");
554	DEBUG_ARG("so = %lx", (long)so);
555
556	if (so->so_type == IPPROTO_ICMP)
557	{
558	/* This is a "ping" reply */
559	#ifdef RT_OS_WINDOWS
560	sorecvfrom_icmp_win(pData, so);
561	#else /* RT_OS_WINDOWS */
562	sorecvfrom_icmp_unix(pData, so);
563	#endif /* !RT_OS_WINDOWS */
564	udp_detach(pData, so);
565	}
566	else
567	{
568	/* A "normal" UDP packet */
569	struct mbuf *m;
570	struct ethhdr *eh;
571	ssize_t len;
572	u_long n = 0;
573	#ifdef VBOX_WITH_SLIRP_BSD_MBUF
574	int size;
575	#endif
576	int rc = 0;
577	static int signalled = 0;
578
579	QSOCKET_LOCK(udb);
580	SOCKET_LOCK(so);
581	QSOCKET_UNLOCK(udb);
582
583	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
584	if (!(m = m_get(pData)))
585	{
586	SOCKET_UNLOCK(so);
587	return;
588	}
589	/* adjust both parameters to maks M_FREEROOM calculate correct */
590	m->m_data += if_maxlinkhdr + sizeof(struct udphdr) + sizeof(struct ip);
591
592	/*
593	* XXX Shouldn't FIONREAD packets destined for port 53,
594	* but I don't know the max packet size for DNS lookups
595	*/
596	len = M_FREEROOM(m);
597	/* if (so->so_fport != RT_H2N_U16_C(53)) */
598	rc = ioctlsocket(so->s, FIONREAD, &n);
599	if ( rc == -1
600	&& ( errno == EAGAIN
601	\|\| errno == EWOULDBLOCK
602	\|\| errno == EINPROGRESS
603	\|\| errno == ENOTCONN))
604	{
605	m_free(pData, m);
606	return;
607	}
608
609	Log2(("NAT: %R[natsock] ioctlsocket before read "
610	"(rc:%d errno:%d, n:%d)\n", so, rc, errno, n));
611
612	if (rc == -1 && signalled == 0)
613	{
614	LogRel(("NAT: can't fetch amount of bytes on socket %R[natsock], so message will be truncated.\n", so));
615	signalled = 1;
616	m_free(pData, m);
617	return;
618	}
619
620	if (rc != -1 && n > len)
621	{
622	n = (m->m_data - m->m_dat) + m->m_len + n + 1;
623	m_inc(m, n);
624	len = M_FREEROOM(m);
625	}
626	ret = recvfrom(so->s, m->m_data, len, 0,
627	(struct sockaddr *)&addr, &addrlen);
628	Log2(("NAT: %R[natsock] ioctlsocket after read "
629	"(rc:%d errno:%d, n:%d) ret:%d, len:%d\n", so,
630	rc, errno, n, ret, len));
631	#else
632	/How many data has been received ?/
633	/*
634	* 1. calculate how much we can read
635	* 2. read as much as possible
636	* 3. attach buffer to allocated header mbuf
637	*/
638	rc = ioctlsocket(so->s, FIONREAD, &n);
639	if (rc == -1 && signalled == 0)
640	{
641	LogRel(("NAT: can't fetch amount of bytes on socket %R[natsock], so message will be truncated.\n", so));
642	signalled = 1;
643	}
644
645	len = sizeof(struct udpiphdr) + ETH_HLEN;
646	if (n > (if_mtu - len))
647	{
648	n = if_mtu - len; /* can't read than we can put in the mbuf*/
649	}
650	len += n;
651
652	size = MCLBYTES;
653	if (len < MSIZE)
654	size = MCLBYTES;
655	else if (len < MCLBYTES)
656	size = MCLBYTES;
657	else if (len < MJUM9BYTES)
658	size = MJUM9BYTES;
659	else if (len < MJUM16BYTES)
660	size = MJUM16BYTES;
661	else
662	AssertMsgFailed(("Unsupported size"));
663
664	m = m_getjcl(pData, M_NOWAIT, MT_HEADER, M_PKTHDR, size);
665	m->m_data += ETH_HLEN;
666	m->m_pkthdr.header = mtod(m, void *);
667	m->m_data += sizeof(struct udpiphdr);
668	ret = recvfrom(so->s, mtod(m, char *), n, 0,
669	(struct sockaddr *)&addr, &addrlen);
670	/* @todo (vvl) check which flags and type should be passed */
671	#endif
672	m->m_len = ret;
673	if (ret < 0)
674	{
675	u_char code = ICMP_UNREACH_PORT;
676
677	if (errno == EHOSTUNREACH)
678	code = ICMP_UNREACH_HOST;
679	else if (errno == ENETUNREACH)
680	code = ICMP_UNREACH_NET;
681
682	m_free(pData, m);
683	if ( errno == EAGAIN
684	\|\| errno == EWOULDBLOCK
685	\|\| errno == EINPROGRESS
686	\|\| errno == ENOTCONN)
687	{
688	return;
689	}
690
691	Log2((" rx error, tx icmp ICMP_UNREACH:%i\n", code));
692	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
693	so->so_m = NULL;
694	}
695	else
696	{
697	/*
698	* Hack: domain name lookup will be used the most for UDP,
699	* and since they'll only be used once there's no need
700	* for the 4 minute (or whatever) timeout... So we time them
701	* out much quicker (10 seconds for now...)
702	*/
703	if (so->so_expire)
704	{
705	if (so->so_fport != RT_H2N_U16_C(53))
706	so->so_expire = curtime + SO_EXPIRE;
707	}
708	/*
709	* last argument should be changed if Slirp will inject IP attributes
710	* Note: Here we can't check if dnsproxy's sent initial request
711	*/
712	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
713	if (so->so_fport == RT_H2N_U16_C(53))
714	dnsproxy_answer(pData, so, m);
715	#endif
716
717	#if 0
718	if (m->m_len == len)
719	{
720	m_inc(m, MINCSIZE);
721	m->m_len = 0;
722	}
723	#endif
724
725	/*
726	* If this packet was destined for CTL_ADDR,
727	* make it look like that's where it came from, done by udp_output
728	*/
729	udp_output(pData, so, m, &addr);
730	SOCKET_UNLOCK(so);
731	} /* rx error */
732	} /* if ping packet */
733	}
734
735	/*
736	* sendto() a socket
737	*/
738	int
739	sosendto(PNATState pData, struct socket so, struct mbuf m)
740	{
741	int ret;
742	struct sockaddr_in *paddr;
743	struct sockaddr addr;
744	#if 0
745	struct sockaddr_in host_addr;
746	#endif
747	#ifdef VBOX_WITH_SLIRP_BSD_MBUF
748	caddr_t buf;
749	int mlen;
750	#endif
751
752	DEBUG_CALL("sosendto");
753	DEBUG_ARG("so = %lx", (long)so);
754	DEBUG_ARG("m = %lx", (long)m);
755
756	memset(&addr, 0, sizeof(struct sockaddr));
757	#ifdef RT_OS_DARWIN
758	addr.sa_len = sizeof(struct sockaddr_in);
759	#endif
760	paddr = (struct sockaddr_in *)&addr;
761	paddr->sin_family = AF_INET;
762	if ((so->so_faddr.s_addr & RT_H2N_U32(pData->netmask)) == pData->special_addr.s_addr)
763	{
764	/* It's an alias */
765	uint32_t last_byte = RT_N2H_U32(so->so_faddr.s_addr) & ~pData->netmask;
766	switch(last_byte)
767	{
768	#if 0
769	/* handle this case at 'default:' */
770	case CTL_BROADCAST:
771	addr.sin_addr.s_addr = INADDR_BROADCAST;
772	/* Send the packet to host to fully emulate broadcast */
773	/** @todo r=klaus: on Linux host this causes the host to receive
774	* the packet twice for some reason. And I cannot find any place
775	* in the man pages which states that sending a broadcast does not
776	* reach the host itself. */
777	host_addr.sin_family = AF_INET;
778	host_addr.sin_port = so->so_fport;
779	host_addr.sin_addr = our_addr;
780	sendto(so->s, m->m_data, m->m_len, 0,
781	(struct sockaddr *)&host_addr, sizeof (struct sockaddr));
782	break;
783	#endif
784	case CTL_DNS:
785	case CTL_ALIAS:
786	default:
787	if (last_byte == ~pData->netmask)
788	paddr->sin_addr.s_addr = INADDR_BROADCAST;
789	else
790	paddr->sin_addr = loopback_addr;
791	break;
792	}
793	}
794	else
795	paddr->sin_addr = so->so_faddr;
796	paddr->sin_port = so->so_fport;
797
798	DEBUG_MISC((dfd, " sendto()ing, addr.sin_port=%d, addr.sin_addr.s_addr=%.16s\n",
799	RT_N2H_U16(paddr->sin_port), inet_ntoa(paddr->sin_addr)));
800
801	/* Don't care what port we get */
802	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
803	ret = sendto(so->s, m->m_data, m->m_len, 0, &addr, sizeof (struct sockaddr_in));
804	#else
805	mlen = m_length(m, NULL);
806	buf = RTMemAlloc(mlen);
807	if (buf == NULL)
808	{
809	return -1;
810	}
811	m_copydata(m, 0, mlen, buf);
812	ret = sendto(so->s, buf, mlen, 0,
813	(struct sockaddr *)&addr, sizeof (struct sockaddr));
814	#endif
815	if (ret < 0)
816	{
817	Log2(("UDP: sendto fails (%s)\n", strerror(errno)));
818	return -1;
819	}
820
821	/*
822	* Kill the socket if there's no reply in 4 minutes,
823	* but only if it's an expirable socket
824	*/
825	if (so->so_expire)
826	so->so_expire = curtime + SO_EXPIRE;
827	so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
828	return 0;
829	}
830
831	/*
832	* XXX This should really be tcp_listen
833	*/
834	struct socket *
835	solisten(PNATState pData, u_int32_t bind_addr, u_int port, u_int32_t laddr, u_int lport, int flags)
836	{
837	struct sockaddr_in addr;
838	struct socket *so;
839	socklen_t addrlen = sizeof(addr);
840	int s, opt = 1;
841	int status;
842
843	DEBUG_CALL("solisten");
844	DEBUG_ARG("port = %d", port);
845	DEBUG_ARG("laddr = %x", laddr);
846	DEBUG_ARG("lport = %d", lport);
847	DEBUG_ARG("flags = %x", flags);
848
849	if ((so = socreate()) == NULL)
850	{
851	/* RTMemFree(so); Not sofree() ??? free(NULL) == NOP */
852	return NULL;
853	}
854
855	/* Don't tcp_attach... we don't need so_snd nor so_rcv */
856	if ((so->so_tcpcb = tcp_newtcpcb(pData, so)) == NULL)
857	{
858	RTMemFree(so);
859	return NULL;
860	}
861
862	SOCKET_LOCK_CREATE(so);
863	SOCKET_LOCK(so);
864	QSOCKET_LOCK(tcb);
865	insque(pData, so,&tcb);
866	NSOCK_INC();
867	QSOCKET_UNLOCK(tcb);
868
869	/*
870	* SS_FACCEPTONCE sockets must time out.
871	*/
872	if (flags & SS_FACCEPTONCE)
873	so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
874
875	so->so_state = (SS_FACCEPTCONN\|flags);
876	so->so_lport = lport; /* Kept in network format */
877	so->so_laddr.s_addr = laddr; /* Ditto */
878
879	memset(&addr, 0, sizeof(addr));
880	#ifdef RT_OS_DARWIN
881	addr.sin_len = sizeof(addr);
882	#endif
883	addr.sin_family = AF_INET;
884	addr.sin_addr.s_addr = bind_addr;
885	addr.sin_port = port;
886
887	if ( ((s = socket(AF_INET, SOCK_STREAM, 0)) < 0)
888	\|\| (setsockopt(s, SOL_SOCKET, SO_REUSEADDR,(char *)&opt, sizeof(int)) < 0)
889	\|\| (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0)
890	\|\| (listen(s, 1) < 0))
891	{
892	#ifdef RT_OS_WINDOWS
893	int tmperrno = WSAGetLastError(); /* Don't clobber the real reason we failed */
894	closesocket(s);
895	QSOCKET_LOCK(tcb);
896	sofree(pData, so);
897	QSOCKET_UNLOCK(tcb);
898	/* Restore the real errno */
899	WSASetLastError(tmperrno);
900	#else
901	int tmperrno = errno; /* Don't clobber the real reason we failed */
902	close(s);
903	QSOCKET_LOCK(tcb);
904	sofree(pData, so);
905	QSOCKET_UNLOCK(tcb);
906	/* Restore the real errno */
907	errno = tmperrno;
908	#endif
909	return NULL;
910	}
911	fd_nonblock(s);
912	setsockopt(s, SOL_SOCKET, SO_OOBINLINE,(char *)&opt, sizeof(int));
913
914	getsockname(s,(struct sockaddr *)&addr,&addrlen);
915	so->so_fport = addr.sin_port;
916	/* set socket buffers */
917	opt = pData->socket_rcv;
918	status = setsockopt(s, SOL_SOCKET, SO_RCVBUF, (char *)&opt, sizeof(int));
919	if (status < 0)
920	{
921	LogRel(("NAT: Error(%d) while setting RCV capacity to (%d)\n", errno, opt));
922	goto no_sockopt;
923	}
924	opt = pData->socket_snd;
925	status = setsockopt(s, SOL_SOCKET, SO_SNDBUF, (char *)&opt, sizeof(int));
926	if (status < 0)
927	{
928	LogRel(("NAT: Error(%d) while setting SND capacity to (%d)\n", errno, opt));
929	goto no_sockopt;
930	}
931	no_sockopt:
932	if (addr.sin_addr.s_addr == 0 \|\| addr.sin_addr.s_addr == loopback_addr.s_addr)
933	so->so_faddr = alias_addr;
934	else
935	so->so_faddr = addr.sin_addr;
936
937	so->s = s;
938	SOCKET_UNLOCK(so);
939	return so;
940	}
941
942	/*
943	* Data is available in so_rcv
944	* Just write() the data to the socket
945	* XXX not yet...
946	*/
947	void
948	sorwakeup(struct socket *so)
949	{
950	#if 0
951	sowrite(so);
952	FD_CLR(so->s,&writefds);
953	#endif
954	}
955
956	/*
957	* Data has been freed in so_snd
958	* We have room for a read() if we want to
959	* For now, don't read, it'll be done in the main loop
960	*/
961	void
962	sowwakeup(struct socket *so)
963	{
964	}
965
966	/*
967	* Various session state calls
968	* XXX Should be #define's
969	* The socket state stuff needs work, these often get call 2 or 3
970	* times each when only 1 was needed
971	*/
972	void
973	soisfconnecting(struct socket *so)
974	{
975	so->so_state &= ~(SS_NOFDREF\|SS_ISFCONNECTED\|SS_FCANTRCVMORE\|
976	SS_FCANTSENDMORE\|SS_FWDRAIN);
977	so->so_state \|= SS_ISFCONNECTING; /* Clobber other states */
978	}
979
980	void
981	soisfconnected(struct socket *so)
982	{
983	so->so_state &= ~(SS_ISFCONNECTING\|SS_FWDRAIN\|SS_NOFDREF);
984	so->so_state \|= SS_ISFCONNECTED; /* Clobber other states */
985	}
986
987	void
988	sofcantrcvmore(struct socket *so)
989	{
990	if ((so->so_state & SS_NOFDREF) == 0)
991	{
992	shutdown(so->s, 0);
993	}
994	so->so_state &= ~(SS_ISFCONNECTING);
995	if (so->so_state & SS_FCANTSENDMORE)
996	so->so_state = SS_NOFDREF; /* Don't select it */
997	/* XXX close() here as well? */
998	else
999	so->so_state \|= SS_FCANTRCVMORE;
1000	}
1001
1002	void
1003	sofcantsendmore(struct socket *so)
1004	{
1005	if ((so->so_state & SS_NOFDREF) == 0)
1006	shutdown(so->s, 1); /* send FIN to fhost */
1007
1008	so->so_state &= ~(SS_ISFCONNECTING);
1009	if (so->so_state & SS_FCANTRCVMORE)
1010	so->so_state = SS_NOFDREF; /* as above */
1011	else
1012	so->so_state \|= SS_FCANTSENDMORE;
1013	}
1014
1015	void
1016	soisfdisconnected(struct socket *so)
1017	{
1018	#if 0
1019	so->so_state &= ~(SS_ISFCONNECTING\|SS_ISFCONNECTED);
1020	close(so->s);
1021	so->so_state = SS_ISFDISCONNECTED;
1022	/*
1023	* XXX Do nothing ... ?
1024	*/
1025	#endif
1026	}
1027
1028	/*
1029	* Set write drain mode
1030	* Set CANTSENDMORE once all data has been write()n
1031	*/
1032	void
1033	sofwdrain(struct socket *so)
1034	{
1035	if (so->so_rcv.sb_cc)
1036	so->so_state \|= SS_FWDRAIN;
1037	else
1038	sofcantsendmore(so);
1039	}
1040
1041	static void
1042	send_icmp_to_guest(PNATState pData, char buff, size_t len, struct socket so, const struct sockaddr_in *addr)
1043	{
1044	struct ip *ip;
1045	uint32_t dst, src;
1046	char ip_copy[256];
1047	struct icmp *icp;
1048	int old_ip_len = 0;
1049	int hlen, original_hlen = 0;
1050	struct mbuf *m;
1051	struct icmp_msg *icm;
1052	uint8_t proto;
1053	int type = 0;
1054	int m_room;
1055
1056	ip = (struct ip *)buff;
1057	/* Fix ip->ip_len to contain the total packet length including the header
1058	* in _host_ byte order for all OSes. On Darwin, that value already is in
1059	* host byte order. Solaris and Darwin report only the payload. */
1060	#ifndef RT_OS_DARWIN
1061	ip->ip_len = RT_N2H_U16(ip->ip_len);
1062	#endif
1063	hlen = (ip->ip_hl << 2);
1064	#if defined(RT_OS_SOLARIS) \|\| defined(RT_OS_DARWIN)
1065	ip->ip_len += hlen;
1066	#endif
1067	if (ip->ip_len < hlen + ICMP_MINLEN)
1068	{
1069	Log(("send_icmp_to_guest: ICMP header is too small to understand which type/subtype of the datagram\n"));
1070	return;
1071	}
1072	icp = (struct icmp )((char )ip + hlen);
1073
1074	Log(("ICMP:received msg(t:%d, c:%d)\n", icp->icmp_type, icp->icmp_code));
1075	if ( icp->icmp_type != ICMP_ECHOREPLY
1076	&& icp->icmp_type != ICMP_TIMXCEED
1077	&& icp->icmp_type != ICMP_UNREACH)
1078	{
1079	return;
1080	}
1081
1082	/*
1083	* ICMP_ECHOREPLY, ICMP_TIMXCEED, ICMP_UNREACH minimal header size is
1084	* ICMP_ECHOREPLY assuming data 0
1085	* icmp_{type(8), code(8), cksum(16),identifier(16),seqnum(16)}
1086	*/
1087	if (ip->ip_len < hlen + 8)
1088	{
1089	Log(("send_icmp_to_guest: NAT accept ICMP_{ECHOREPLY, TIMXCEED, UNREACH} the minimum size is 64 (see rfc792)\n"));
1090	return;
1091	}
1092
1093	type = icp->icmp_type;
1094	if ( type == ICMP_TIMXCEED
1095	\|\| type == ICMP_UNREACH)
1096	{
1097	/*
1098	* ICMP_TIMXCEED, ICMP_UNREACH minimal header size is
1099	* icmp_{type(8), code(8), cksum(16),unused(32)} + IP header + 64 bit of original datagram
1100	*/
1101	if (ip->ip_len < hlen + 2*8 + sizeof(struct ip))
1102	{
1103	Log(("send_icmp_to_guest: NAT accept ICMP_{TIMXCEED, UNREACH} the minimum size of ipheader + 64 bit of data (see rfc792)\n"));
1104	return;
1105	}
1106	ip = &icp->icmp_ip;
1107	}
1108
1109	icm = icmp_find_original_mbuf(pData, ip);
1110	if (icm == NULL)
1111	{
1112	Log(("NAT: Can't find the corresponding packet for the received ICMP\n"));
1113	return;
1114	}
1115
1116	m = icm->im_m;
1117	Assert(m != NULL);
1118
1119	src = addr->sin_addr.s_addr;
1120	if (type == ICMP_ECHOREPLY)
1121	{
1122	struct ip ip0 = mtod(m, struct ip );
1123	struct icmp icp0 = (struct icmp )((char *)ip0 + (ip0->ip_hl << 2));
1124	if (icp0->icmp_type != ICMP_ECHO)
1125	{
1126	Log(("NAT: we haven't found echo for this reply\n"));
1127	return;
1128	}
1129	/*
1130	* while combining buffer to send (see ip_icmp.c) we control ICMP header only,
1131	* IP header combined by OS network stack, our local copy of IP header contians values
1132	* in host byte order so no byte order conversion is required. IP headers fields are converting
1133	* in ip_output0 routine only.
1134	*/
1135	if ( (ip->ip_len - hlen)
1136	!= (ip0->ip_len - (ip0->ip_hl << 2)))
1137	{
1138	Log(("NAT: ECHO(%d) lenght doesn't match ECHOREPLY(%d)\n",
1139	(ip->ip_len - hlen), (ip0->ip_len - (ip0->ip_hl << 2))));
1140	return;
1141	}
1142	}
1143
1144	/* ip points on origianal ip header */
1145	ip = mtod(m, struct ip *);
1146	proto = ip->ip_p;
1147	/* Now ip is pointing on header we've sent from guest */
1148	if ( icp->icmp_type == ICMP_TIMXCEED
1149	\|\| icp->icmp_type == ICMP_UNREACH)
1150	{
1151	old_ip_len = (ip->ip_hl << 2) + 64;
1152	if (old_ip_len > sizeof(ip_copy))
1153	old_ip_len = sizeof(ip_copy);
1154	memcpy(ip_copy, ip, old_ip_len);
1155	}
1156
1157	/* source address from original IP packet*/
1158	dst = ip->ip_src.s_addr;
1159
1160	/* overide ther tail of old packet */
1161	ip = mtod(m, struct ip ); / ip is from mbuf we've overrided */
1162	original_hlen = ip->ip_hl << 2;
1163	/* saves original ip header and options */
1164	#ifdef VBOX_WITH_SLIRP_BSD_MBUF
1165	m_copyback(pData, m, original_hlen, len - hlen, buff + hlen);
1166	ip->ip_len = m_length(m, NULL);
1167	#else
1168	/* m_room space in the saved m buffer */
1169	m_room = M_ROOM(m);
1170	if (m_room < len - hlen + original_hlen)
1171	{
1172	/* we need involve ether header length into new buffer buffer calculation */
1173	m_inc(m, if_maxlinkhdr + len - hlen + original_hlen);
1174	if (m->m_size < if_maxlinkhdr + len - hlen + original_hlen)
1175	{
1176	Log(("send_icmp_to_guest: extending buffer was failed (packet is dropped)\n"));
1177	return;
1178	}
1179	}
1180	memcpy(m->m_data + original_hlen, buff + hlen, len - hlen);
1181	m->m_len = len - hlen + original_hlen;
1182	ip->ip_len = m->m_len;
1183	#endif
1184	ip->ip_p = IPPROTO_ICMP; /* the original package could be whatever, but we're response via ICMP*/
1185
1186	icp = (struct icmp )((char )ip + (ip->ip_hl << 2));
1187	type = icp->icmp_type;
1188	if ( type == ICMP_TIMXCEED
1189	\|\| type == ICMP_UNREACH)
1190	{
1191	/* according RFC 793 error messages required copy of initial IP header + 64 bit */
1192	memcpy(&icp->icmp_ip, ip_copy, old_ip_len);
1193	ip->ip_tos = ((ip->ip_tos & 0x1E) \| 0xC0); /* high priority for errors */
1194	}
1195
1196	ip->ip_src.s_addr = src;
1197	ip->ip_dst.s_addr = dst;
1198	icmp_reflect(pData, m);
1199	LIST_REMOVE(icm, im_list);
1200	/* Don't call m_free here*/
1201
1202	if ( type == ICMP_TIMXCEED
1203	\|\| type == ICMP_UNREACH)
1204	{
1205	icm->im_so->so_m = NULL;
1206	switch (proto)
1207	{
1208	case IPPROTO_UDP:
1209	/XXX: so->so_m already freed so we shouldn't call sofree /
1210	udp_detach(pData, icm->im_so);
1211	break;
1212	case IPPROTO_TCP:
1213	/close tcp should be here /
1214	break;
1215	default:
1216	/* do nothing */
1217	break;
1218	}
1219	}
1220	RTMemFree(icm);
1221	}
1222
1223	#ifdef RT_OS_WINDOWS
1224	static void
1225	sorecvfrom_icmp_win(PNATState pData, struct socket *so)
1226	{
1227	int len;
1228	int i;
1229	struct ip *ip;
1230	struct mbuf *m;
1231	struct icmp *icp;
1232	struct icmp_msg *icm;
1233	struct ip ip_broken; / ICMP returns header + 64 bit of packet */
1234	uint32_t src;
1235	ICMP_ECHO_REPLY *icr;
1236	int hlen = 0;
1237	int data_len = 0;
1238	int nbytes = 0;
1239	u_char code = ~0;
1240
1241	len = pData->pfIcmpParseReplies(pData->pvIcmpBuffer, pData->szIcmpBuffer);
1242	if (len < 0)
1243	{
1244	LogRel(("NAT: Error (%d) occurred on ICMP receiving\n", GetLastError()));
1245	return;
1246	}
1247	if (len == 0)
1248	return; /* no error */
1249
1250	icr = (ICMP_ECHO_REPLY *)pData->pvIcmpBuffer;
1251	for (i = 0; i < len; ++i)
1252	{
1253	switch(icr[i].Status)
1254	{
1255	case IP_DEST_HOST_UNREACHABLE:
1256	code = (code != ~0 ? code : ICMP_UNREACH_HOST);
1257	case IP_DEST_NET_UNREACHABLE:
1258	code = (code != ~0 ? code : ICMP_UNREACH_NET);
1259	case IP_DEST_PROT_UNREACHABLE:
1260	code = (code != ~0 ? code : ICMP_UNREACH_PROTOCOL);
1261	/* UNREACH error inject here */
1262	case IP_DEST_PORT_UNREACHABLE:
1263	code = (code != ~0 ? code : ICMP_UNREACH_PORT);
1264	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, "Error occurred!!!");
1265	so->so_m = NULL;
1266	break;
1267	case IP_SUCCESS: /* echo replied */
1268	# ifndef VBOX_WITH_SLIRP_BSD_MBUF
1269	m = m_get(pData);
1270	# else
1271	m = m_gethdr(pData, M_NOWAIT, MT_HEADER);
1272	# endif
1273	m->m_data += if_maxlinkhdr;
1274	ip = mtod(m, struct ip *);
1275	ip->ip_src.s_addr = icr[i].Address;
1276	ip->ip_p = IPPROTO_ICMP;
1277	ip->ip_dst.s_addr = so->so_laddr.s_addr; /XXX: still the hack/
1278	data_len = sizeof(struct ip);
1279	ip->ip_hl = data_len >> 2; /* requiered for icmp_reflect, no IP options */
1280	ip->ip_ttl = icr[i].Options.Ttl;
1281
1282	icp = (struct icmp )&ip[1]; / no options */
1283	icp->icmp_type = ICMP_ECHOREPLY;
1284	icp->icmp_code = 0;
1285	icp->icmp_id = so->so_icmp_id;
1286	icp->icmp_seq = so->so_icmp_seq;
1287
1288	data_len += ICMP_MINLEN;
1289
1290	# ifndef VBOX_WITH_SLIRP_BSD_MBUF
1291	nbytes = (data_len + icr[i].DataSize > m->m_size? m->m_size - data_len: icr[i].DataSize);
1292	memcpy(icp->icmp_data, icr[i].Data, nbytes);
1293	# else
1294	AssertMsgFailed(("ICMP"));
1295	# endif
1296
1297	data_len += icr[i].DataSize;
1298
1299	ip->ip_len = data_len;
1300	m->m_len = ip->ip_len;
1301
1302	icmp_reflect(pData, m);
1303	break;
1304	case IP_TTL_EXPIRED_TRANSIT: /* TTL expired */
1305
1306	ip_broken = icr[i].Data;
1307	icm = icmp_find_original_mbuf(pData, ip_broken);
1308	if (icm == NULL) {
1309	Log(("ICMP: can't find original package (first double word %x)\n", (uint32_t )ip_broken));
1310	return;
1311	}
1312	m = icm->im_m;
1313	ip = mtod(m, struct ip *);
1314	ip->ip_ttl = icr[i].Options.Ttl;
1315	src = ip->ip_src.s_addr;
1316	ip->ip_dst.s_addr = src;
1317	ip->ip_dst.s_addr = icr[i].Address;
1318
1319	hlen = (ip->ip_hl << 2);
1320	icp = (struct icmp )((char )ip + hlen);
1321	ip_broken->ip_src.s_addr = src; /it packet sent from host not from guest/
1322	data_len = (ip_broken->ip_hl << 2) + 64;
1323
1324	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
1325	nbytes =(hlen + ICMP_MINLEN + data_len > m->m_size? m->m_size - (hlen + ICMP_MINLEN): data_len);
1326	memcpy(icp->icmp_data, ip_broken, nbytes);
1327	#else
1328	AssertMsgFailed(("ICMP"));
1329	#endif
1330	icmp_reflect(pData, m);
1331	break;
1332	default:
1333	Log(("ICMP(default): message with Status: %x was received from %x\n", icr[i].Status, icr[i].Address));
1334	break;
1335	}
1336	}
1337	}
1338	#else /* !RT_OS_WINDOWS */
1339	static void sorecvfrom_icmp_unix(PNATState pData, struct socket *so)
1340	{
1341	struct sockaddr_in addr;
1342	socklen_t addrlen = sizeof(struct sockaddr_in);
1343	struct ip ip;
1344	char *buff;
1345	int len = 0;
1346
1347	/* 1- step: read the ip header */
1348	len = recvfrom(so->s, &ip, sizeof(struct ip), MSG_PEEK,
1349	(struct sockaddr *)&addr, &addrlen);
1350	if ( len < 0
1351	&& ( errno == EAGAIN
1352	\|\| errno == EWOULDBLOCK
1353	\|\| errno == EINPROGRESS
1354	\|\| errno == ENOTCONN))
1355	{
1356	Log(("sorecvfrom_icmp_unix: 1 - step can't read IP datagramm (would block)\n"));
1357	return;
1358	}
1359
1360	if ( len < sizeof(struct ip)
1361	\|\| len < 0
1362	\|\| len == 0)
1363	{
1364	u_char code;
1365	code = ICMP_UNREACH_PORT;
1366
1367	if (errno == EHOSTUNREACH)
1368	code = ICMP_UNREACH_HOST;
1369	else if (errno == ENETUNREACH)
1370	code = ICMP_UNREACH_NET;
1371
1372	LogRel((" udp icmp rx errno = %d-%s\n",
1373	errno, strerror(errno)));
1374	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
1375	so->so_m = NULL;
1376	Log(("sorecvfrom_icmp_unix: 1 - step can't read IP datagramm \n"));
1377	return;
1378	}
1379	/* basic check of IP header */
1380	if ( ip.ip_v != IPVERSION
1381	# ifndef RT_OS_DARWIN
1382	\|\| ip.ip_p != IPPROTO_ICMP
1383	# endif
1384	)
1385	{
1386	Log(("sorecvfrom_icmp_unix: 1 - step IP isn't IPv4 \n"));
1387	return;
1388	}
1389	# ifndef RT_OS_DARWIN
1390	/* Darwin reports the IP length already in host byte order. */
1391	ip.ip_len = RT_N2H_U16(ip.ip_len);
1392	# endif
1393	# if defined(RT_OS_SOLARIS) \|\| defined(RT_OS_DARWIN)
1394	/* Solaris and Darwin report the payload only */
1395	ip.ip_len += (ip.ip_hl << 2);
1396	# endif
1397	/* Note: ip->ip_len in host byte order (all OS) */
1398	len = ip.ip_len;
1399	buff = RTMemAlloc(len);
1400	if (buff == NULL)
1401	{
1402	Log(("sorecvfrom_icmp_unix: 1 - step can't allocate enought room for datagram\n"));
1403	return;
1404	}
1405	/* 2 - step: we're reading rest of the datagramm to the buffer */
1406	addrlen = sizeof(struct sockaddr_in);
1407	memset(&addr, 0, addrlen);
1408	len = recvfrom(so->s, buff, len, 0,
1409	(struct sockaddr *)&addr, &addrlen);
1410	if ( len < 0
1411	&& ( errno == EAGAIN
1412	\|\| errno == EWOULDBLOCK
1413	\|\| errno == EINPROGRESS
1414	\|\| errno == ENOTCONN))
1415	{
1416	Log(("sorecvfrom_icmp_unix: 2 - step can't read IP body (would block expected:%d)\n",
1417	ip.ip_len));
1418	RTMemFree(buff);
1419	return;
1420	}
1421	if ( len < 0
1422	\|\| len == 0)
1423	{
1424	Log(("sorecvfrom_icmp_unix: 2 - step read of the rest of datagramm is fallen (errno:%d, len:%d expected: %d)\n",
1425	errno, len, (ip.ip_len - sizeof(struct ip))));
1426	RTMemFree(buff);
1427	return;
1428	}
1429	/* len is modified in 2nd read, when the rest of the datagramm was read */
1430	send_icmp_to_guest(pData, buff, len, so, &addr);
1431	RTMemFree(buff);
1432	}
1433	#endif /* !RT_OS_WINDOWS */

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/socket.c@ 27764

Download in other formats: