socket.c@ 28390

Last change on this file since 28390 was 28390, checked in by vboxsync, 15 years ago
NAT: ICMP with BSD buffers.
Property svn:eol-style set to `native`
File size: 42.1 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	#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, (so->so_tcpcb->t_force? MSG_OOB:0));
202	#endif
203	if (nn <= 0)
204	{
205	/*
206	* Special case for WSAEnumNetworkEvents: If we receive 0 bytes that
207	* _could_ mean that the connection is closed. But we will receive an
208	* FD_CLOSE event later if the connection was _really_ closed. With
209	* www.youtube.com I see this very often. Closing the socket too early
210	* would be dangerous.
211	*/
212	int status, ignored;
213	unsigned long pending = 0;
214	status = ioctlsocket(so->s, FIONREAD, &pending);
215	if (status < 0)
216	LogRel(("NAT:error in WSAIoctl: %d\n", errno));
217	if (nn == 0 && (pending != 0))
218	{
219	SOCKET_UNLOCK(so);
220	STAM_PROFILE_STOP(&pData->StatIOread, a);
221	return 0;
222	}
223	if ( nn < 0
224	&& ( errno == EINTR
225	\|\| errno == EAGAIN
226	\|\| errno == EWOULDBLOCK))
227	{
228	SOCKET_UNLOCK(so);
229	STAM_PROFILE_STOP(&pData->StatIOread, a);
230	return 0;
231	}
232	else
233	{
234	/* nn == 0 means peer has performed an orderly shutdown */
235	DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n",
236	nn, errno, strerror(errno)));
237	sofcantrcvmore(so);
238	tcp_sockclosed(pData, sototcpcb(so));
239	SOCKET_UNLOCK(so);
240	STAM_PROFILE_STOP(&pData->StatIOread, a);
241	return -1;
242	}
243	}
244	STAM_STATS(
245	if (n == 1)
246	{
247	STAM_COUNTER_INC(&pData->StatIORead_in_1);
248	STAM_COUNTER_ADD(&pData->StatIORead_in_1_bytes, nn);
249	}
250	else
251	{
252	STAM_COUNTER_INC(&pData->StatIORead_in_2);
253	STAM_COUNTER_ADD(&pData->StatIORead_in_2_1st_bytes, nn);
254	}
255	);
256
257	#ifndef HAVE_READV
258	/*
259	* If there was no error, try and read the second time round
260	* We read again if n = 2 (ie, there's another part of the buffer)
261	* and we read as much as we could in the first read
262	* We don't test for <= 0 this time, because there legitimately
263	* might not be any more data (since the socket is non-blocking),
264	* a close will be detected on next iteration.
265	* A return of -1 wont (shouldn't) happen, since it didn't happen above
266	*/
267	if (n == 2 && nn == iov[0].iov_len)
268	{
269	int ret;
270	ret = recv(so->s, iov[1].iov_base, iov[1].iov_len, 0);
271	if (ret > 0)
272	nn += ret;
273	STAM_STATS(
274	if (ret > 0)
275	{
276	STAM_COUNTER_INC(&pData->StatIORead_in_2);
277	STAM_COUNTER_ADD(&pData->StatIORead_in_2_2nd_bytes, ret);
278	}
279	);
280	}
281
282	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
283	#endif
284
285	/* Update fields */
286	sb->sb_cc += nn;
287	sb->sb_wptr += nn;
288	if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
289	sb->sb_wptr -= sb->sb_datalen;
290	STAM_PROFILE_STOP(&pData->StatIOread, a);
291	SOCKET_UNLOCK(so);
292	return nn;
293	}
294
295	/*
296	* Get urgent data
297	*
298	* When the socket is created, we set it SO_OOBINLINE,
299	* so when OOB data arrives, we soread() it and everything
300	* in the send buffer is sent as urgent data
301	*/
302	void
303	sorecvoob(PNATState pData, struct socket *so)
304	{
305	struct tcpcb *tp = sototcpcb(so);
306	ssize_t ret;
307
308	DEBUG_CALL("sorecvoob");
309	DEBUG_ARG("so = %lx", (long)so);
310
311	/*
312	* We take a guess at how much urgent data has arrived.
313	* In most situations, when urgent data arrives, the next
314	* read() should get all the urgent data. This guess will
315	* be wrong however if more data arrives just after the
316	* urgent data, or the read() doesn't return all the
317	* urgent data.
318	*/
319	ret = soread(pData, so);
320	tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
321	tp->t_force = 1;
322	tcp_output(pData, tp);
323	tp->t_force = 0;
324	}
325
326	/*
327	* Send urgent data
328	* There's a lot duplicated code here, but...
329	*/
330	int
331	sosendoob(struct socket *so)
332	{
333	struct sbuf *sb = &so->so_rcv;
334	char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
335
336	int n, len;
337
338	DEBUG_CALL("sosendoob");
339	DEBUG_ARG("so = %lx", (long)so);
340	DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
341
342	if (so->so_urgc > sizeof(buff))
343	so->so_urgc = sizeof(buff); /* XXX */
344
345	if (sb->sb_rptr < sb->sb_wptr)
346	{
347	/* We can send it directly */
348	n = send(so->s, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
349	so->so_urgc -= n;
350
351	DEBUG_MISC((dfd, " --- sent %d bytes urgent data, %d urgent bytes left\n",
352	n, so->so_urgc));
353	}
354	else
355	{
356	/*
357	* Since there's no sendv or sendtov like writev,
358	* we must copy all data to a linear buffer then
359	* send it all
360	*/
361	len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
362	if (len > so->so_urgc)
363	len = so->so_urgc;
364	memcpy(buff, sb->sb_rptr, len);
365	so->so_urgc -= len;
366	if (so->so_urgc)
367	{
368	n = sb->sb_wptr - sb->sb_data;
369	if (n > so->so_urgc)
370	n = so->so_urgc;
371	memcpy(buff + len, sb->sb_data, n);
372	so->so_urgc -= n;
373	len += n;
374	}
375	n = send(so->s, buff, len, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
376	#ifdef DEBUG
377	if (n != len)
378	DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
379	#endif
380	DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n",
381	n, so->so_urgc));
382	}
383
384	sb->sb_cc -= n;
385	sb->sb_rptr += n;
386	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
387	sb->sb_rptr -= sb->sb_datalen;
388
389	return n;
390	}
391
392	/*
393	* Write data from so_rcv to so's socket,
394	* updating all sbuf field as necessary
395	*/
396	int
397	sowrite(PNATState pData, struct socket *so)
398	{
399	int n, nn;
400	struct sbuf *sb = &so->so_rcv;
401	size_t len = sb->sb_cc;
402	struct iovec iov[2];
403
404	STAM_PROFILE_START(&pData->StatIOwrite, a);
405	STAM_COUNTER_RESET(&pData->StatIOWrite_in_1);
406	STAM_COUNTER_RESET(&pData->StatIOWrite_in_1_bytes);
407	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2);
408	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2_1st_bytes);
409	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2_2nd_bytes);
410	STAM_COUNTER_RESET(&pData->StatIOWrite_no_w);
411	STAM_COUNTER_RESET(&pData->StatIOWrite_rest);
412	STAM_COUNTER_RESET(&pData->StatIOWrite_rest_bytes);
413	DEBUG_CALL("sowrite");
414	DEBUG_ARG("so = %lx", (long)so);
415	QSOCKET_LOCK(tcb);
416	SOCKET_LOCK(so);
417	QSOCKET_UNLOCK(tcb);
418	if (so->so_urgc)
419	{
420	sosendoob(so);
421	if (sb->sb_cc == 0)
422	{
423	SOCKET_UNLOCK(so);
424	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
425	return 0;
426	}
427	}
428
429	/*
430	* No need to check if there's something to write,
431	* sowrite wouldn't have been called otherwise
432	*/
433
434	len = sb->sb_cc;
435
436	iov[0].iov_base = sb->sb_rptr;
437	iov[1].iov_base = 0;
438	iov[1].iov_len = 0;
439	if (sb->sb_rptr < sb->sb_wptr)
440	{
441	iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
442	/* Should never succeed, but... */
443	if (iov[0].iov_len > len)
444	iov[0].iov_len = len;
445	n = 1;
446	}
447	else
448	{
449	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
450	if (iov[0].iov_len > len)
451	iov[0].iov_len = len;
452	len -= iov[0].iov_len;
453	if (len)
454	{
455	iov[1].iov_base = sb->sb_data;
456	iov[1].iov_len = sb->sb_wptr - sb->sb_data;
457	if (iov[1].iov_len > len)
458	iov[1].iov_len = len;
459	n = 2;
460	}
461	else
462	n = 1;
463	}
464	STAM_STATS({
465	if (n == 1)
466	{
467	STAM_COUNTER_INC(&pData->StatIOWrite_in_1);
468	STAM_COUNTER_ADD(&pData->StatIOWrite_in_1_bytes, iov[0].iov_len);
469	}
470	else
471	{
472	STAM_COUNTER_INC(&pData->StatIOWrite_in_2);
473	STAM_COUNTER_ADD(&pData->StatIOWrite_in_2_1st_bytes, iov[0].iov_len);
474	STAM_COUNTER_ADD(&pData->StatIOWrite_in_2_2nd_bytes, iov[1].iov_len);
475	}
476	});
477	/* Check if there's urgent data to send, and if so, send it */
478	#ifdef HAVE_READV
479	nn = writev(so->s, (const struct iovec *)iov, n);
480	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
481	#else
482	nn = send(so->s, iov[0].iov_base, iov[0].iov_len, 0);
483	#endif
484	/* This should never happen, but people tell me it does shrug */
485	if ( nn < 0
486	&& ( errno == EAGAIN
487	\|\| errno == EINTR
488	\|\| errno == EWOULDBLOCK))
489	{
490	SOCKET_UNLOCK(so);
491	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
492	return 0;
493	}
494
495	if (nn < 0 \|\| (nn == 0 && iov[0].iov_len > 0))
496	{
497	DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
498	so->so_state, errno));
499	sofcantsendmore(so);
500	tcp_sockclosed(pData, sototcpcb(so));
501	SOCKET_UNLOCK(so);
502	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
503	return -1;
504	}
505
506	#ifndef HAVE_READV
507	if (n == 2 && nn == iov[0].iov_len)
508	{
509	int ret;
510	ret = send(so->s, iov[1].iov_base, iov[1].iov_len, 0);
511	if (ret > 0)
512	nn += ret;
513	STAM_STATS({
514	if (ret > 0 && ret != iov[1].iov_len)
515	{
516	STAM_COUNTER_INC(&pData->StatIOWrite_rest);
517	STAM_COUNTER_ADD(&pData->StatIOWrite_rest_bytes, (ret - iov[1].iov_len));
518	}
519	});
520	}
521	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
522	#endif
523
524	/* Update sbuf */
525	sb->sb_cc -= nn;
526	sb->sb_rptr += nn;
527	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
528	sb->sb_rptr -= sb->sb_datalen;
529
530	/*
531	* If in DRAIN mode, and there's no more data, set
532	* it CANTSENDMORE
533	*/
534	if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
535	sofcantsendmore(so);
536
537	SOCKET_UNLOCK(so);
538	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
539	return nn;
540	}
541
542	/*
543	* recvfrom() a UDP socket
544	*/
545	void
546	sorecvfrom(PNATState pData, struct socket *so)
547	{
548	ssize_t ret = 0;
549	struct sockaddr_in addr;
550	socklen_t addrlen = sizeof(struct sockaddr_in);
551
552	DEBUG_CALL("sorecvfrom");
553	DEBUG_ARG("so = %lx", (long)so);
554
555	if (so->so_type == IPPROTO_ICMP)
556	{
557	/* This is a "ping" reply */
558	#ifdef RT_OS_WINDOWS
559	sorecvfrom_icmp_win(pData, so);
560	#else /* RT_OS_WINDOWS */
561	sorecvfrom_icmp_unix(pData, so);
562	#endif /* !RT_OS_WINDOWS */
563	udp_detach(pData, so);
564	}
565	else
566	{
567	/* A "normal" UDP packet */
568	struct mbuf *m;
569	ssize_t len;
570	u_long n = 0;
571	#ifdef VBOX_WITH_SLIRP_BSD_MBUF
572	int size;
573	#endif
574	int rc = 0;
575	static int signalled = 0;
576
577	QSOCKET_LOCK(udb);
578	SOCKET_LOCK(so);
579	QSOCKET_UNLOCK(udb);
580
581	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
582	if (!(m = m_get(pData)))
583	{
584	SOCKET_UNLOCK(so);
585	return;
586	}
587	/* adjust both parameters to maks M_FREEROOM calculate correct */
588	m->m_data += if_maxlinkhdr + sizeof(struct udphdr) + sizeof(struct ip);
589
590	/*
591	* XXX Shouldn't FIONREAD packets destined for port 53,
592	* but I don't know the max packet size for DNS lookups
593	*/
594	len = M_FREEROOM(m);
595	/* if (so->so_fport != RT_H2N_U16_C(53)) */
596	rc = ioctlsocket(so->s, FIONREAD, &n);
597	if ( rc == -1
598	&& ( errno == EAGAIN
599	\|\| errno == EWOULDBLOCK
600	\|\| errno == EINPROGRESS
601	\|\| errno == ENOTCONN))
602	{
603	m_free(pData, m);
604	return;
605	}
606
607	Log2(("NAT: %R[natsock] ioctlsocket before read "
608	"(rc:%d errno:%d, n:%d)\n", so, rc, errno, n));
609
610	if (rc == -1 && signalled == 0)
611	{
612	LogRel(("NAT: can't fetch amount of bytes on socket %R[natsock], so message will be truncated.\n", so));
613	signalled = 1;
614	m_free(pData, m);
615	return;
616	}
617
618	if (rc != -1 && n > len)
619	{
620	n = (m->m_data - m->m_dat) + m->m_len + n + 1;
621	m_inc(m, n);
622	len = M_FREEROOM(m);
623	}
624	ret = recvfrom(so->s, m->m_data, len, 0,
625	(struct sockaddr *)&addr, &addrlen);
626	Log2(("NAT: %R[natsock] ioctlsocket after read "
627	"(rc:%d errno:%d, n:%d) ret:%d, len:%d\n", so,
628	rc, errno, n, ret, len));
629	#else
630	/How many data has been received ?/
631	/*
632	* 1. calculate how much we can read
633	* 2. read as much as possible
634	* 3. attach buffer to allocated header mbuf
635	*/
636	rc = ioctlsocket(so->s, FIONREAD, &n);
637	if (rc == -1 && signalled == 0)
638	{
639	LogRel(("NAT: can't fetch amount of bytes on socket %R[natsock], so message will be truncated.\n", so));
640	signalled = 1;
641	}
642
643	len = sizeof(struct udpiphdr) + ETH_HLEN;
644	if (n > (if_mtu - len))
645	{
646	n = if_mtu - len; /* can't read than we can put in the mbuf*/
647	}
648	len += n;
649
650	size = MCLBYTES;
651	if (len < MSIZE)
652	size = MCLBYTES;
653	else if (len < MCLBYTES)
654	size = MCLBYTES;
655	else if (len < MJUM9BYTES)
656	size = MJUM9BYTES;
657	else if (len < MJUM16BYTES)
658	size = MJUM16BYTES;
659	else
660	AssertMsgFailed(("Unsupported size"));
661
662	m = m_getjcl(pData, M_NOWAIT, MT_HEADER, M_PKTHDR, size);
663	m->m_data += ETH_HLEN;
664	m->m_pkthdr.header = mtod(m, void *);
665	m->m_data += sizeof(struct udpiphdr);
666	ret = recvfrom(so->s, mtod(m, char *), n, 0,
667	(struct sockaddr *)&addr, &addrlen);
668	/* @todo (vvl) check which flags and type should be passed */
669	#endif
670	m->m_len = ret;
671	if (ret < 0)
672	{
673	u_char code = ICMP_UNREACH_PORT;
674
675	if (errno == EHOSTUNREACH)
676	code = ICMP_UNREACH_HOST;
677	else if (errno == ENETUNREACH)
678	code = ICMP_UNREACH_NET;
679
680	m_free(pData, m);
681	if ( errno == EAGAIN
682	\|\| errno == EWOULDBLOCK
683	\|\| errno == EINPROGRESS
684	\|\| errno == ENOTCONN)
685	{
686	return;
687	}
688
689	Log2((" rx error, tx icmp ICMP_UNREACH:%i\n", code));
690	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
691	so->so_m = NULL;
692	}
693	else
694	{
695	/*
696	* Hack: domain name lookup will be used the most for UDP,
697	* and since they'll only be used once there's no need
698	* for the 4 minute (or whatever) timeout... So we time them
699	* out much quicker (10 seconds for now...)
700	*/
701	if (so->so_expire)
702	{
703	if (so->so_fport != RT_H2N_U16_C(53))
704	so->so_expire = curtime + SO_EXPIRE;
705	}
706	/*
707	* last argument should be changed if Slirp will inject IP attributes
708	* Note: Here we can't check if dnsproxy's sent initial request
709	*/
710	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
711	if (so->so_fport == RT_H2N_U16_C(53))
712	dnsproxy_answer(pData, so, m);
713	#endif
714
715	#if 0
716	if (m->m_len == len)
717	{
718	m_inc(m, MINCSIZE);
719	m->m_len = 0;
720	}
721	#endif
722
723	/*
724	* If this packet was destined for CTL_ADDR,
725	* make it look like that's where it came from, done by udp_output
726	*/
727	udp_output(pData, so, m, &addr);
728	SOCKET_UNLOCK(so);
729	} /* rx error */
730	} /* if ping packet */
731	}
732
733	/*
734	* sendto() a socket
735	*/
736	int
737	sosendto(PNATState pData, struct socket so, struct mbuf m)
738	{
739	int ret;
740	struct sockaddr_in *paddr;
741	struct sockaddr addr;
742	#if 0
743	struct sockaddr_in host_addr;
744	#endif
745	#ifdef VBOX_WITH_SLIRP_BSD_MBUF
746	caddr_t buf;
747	int mlen;
748	#endif
749
750	DEBUG_CALL("sosendto");
751	DEBUG_ARG("so = %lx", (long)so);
752	DEBUG_ARG("m = %lx", (long)m);
753
754	memset(&addr, 0, sizeof(struct sockaddr));
755	#ifdef RT_OS_DARWIN
756	addr.sa_len = sizeof(struct sockaddr_in);
757	#endif
758	paddr = (struct sockaddr_in *)&addr;
759	paddr->sin_family = AF_INET;
760	if ((so->so_faddr.s_addr & RT_H2N_U32(pData->netmask)) == pData->special_addr.s_addr)
761	{
762	/* It's an alias */
763	uint32_t last_byte = RT_N2H_U32(so->so_faddr.s_addr) & ~pData->netmask;
764	switch(last_byte)
765	{
766	#if 0
767	/* handle this case at 'default:' */
768	case CTL_BROADCAST:
769	addr.sin_addr.s_addr = INADDR_BROADCAST;
770	/* Send the packet to host to fully emulate broadcast */
771	/** @todo r=klaus: on Linux host this causes the host to receive
772	* the packet twice for some reason. And I cannot find any place
773	* in the man pages which states that sending a broadcast does not
774	* reach the host itself. */
775	host_addr.sin_family = AF_INET;
776	host_addr.sin_port = so->so_fport;
777	host_addr.sin_addr = our_addr;
778	sendto(so->s, m->m_data, m->m_len, 0,
779	(struct sockaddr *)&host_addr, sizeof (struct sockaddr));
780	break;
781	#endif
782	case CTL_DNS:
783	case CTL_ALIAS:
784	default:
785	if (last_byte == ~pData->netmask)
786	paddr->sin_addr.s_addr = INADDR_BROADCAST;
787	else
788	paddr->sin_addr = loopback_addr;
789	break;
790	}
791	}
792	else
793	paddr->sin_addr = so->so_faddr;
794	paddr->sin_port = so->so_fport;
795
796	DEBUG_MISC((dfd, " sendto()ing, addr.sin_port=%d, addr.sin_addr.s_addr=%.16s\n",
797	RT_N2H_U16(paddr->sin_port), inet_ntoa(paddr->sin_addr)));
798
799	/* Don't care what port we get */
800	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
801	ret = sendto(so->s, m->m_data, m->m_len, 0, &addr, sizeof (struct sockaddr_in));
802	#else
803	mlen = m_length(m, NULL);
804	buf = RTMemAlloc(mlen);
805	if (buf == NULL)
806	{
807	return -1;
808	}
809	m_copydata(m, 0, mlen, buf);
810	ret = sendto(so->s, buf, mlen, 0,
811	(struct sockaddr *)&addr, sizeof (struct sockaddr));
812	#endif
813	if (ret < 0)
814	{
815	Log2(("UDP: sendto fails (%s)\n", strerror(errno)));
816	return -1;
817	}
818
819	/*
820	* Kill the socket if there's no reply in 4 minutes,
821	* but only if it's an expirable socket
822	*/
823	if (so->so_expire)
824	so->so_expire = curtime + SO_EXPIRE;
825	so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
826	return 0;
827	}
828
829	/*
830	* XXX This should really be tcp_listen
831	*/
832	struct socket *
833	solisten(PNATState pData, u_int32_t bind_addr, u_int port, u_int32_t laddr, u_int lport, int flags)
834	{
835	struct sockaddr_in addr;
836	struct socket *so;
837	socklen_t addrlen = sizeof(addr);
838	int s, opt = 1;
839	int status;
840
841	DEBUG_CALL("solisten");
842	DEBUG_ARG("port = %d", port);
843	DEBUG_ARG("laddr = %x", laddr);
844	DEBUG_ARG("lport = %d", lport);
845	DEBUG_ARG("flags = %x", flags);
846
847	if ((so = socreate()) == NULL)
848	{
849	/* RTMemFree(so); Not sofree() ??? free(NULL) == NOP */
850	return NULL;
851	}
852
853	/* Don't tcp_attach... we don't need so_snd nor so_rcv */
854	if ((so->so_tcpcb = tcp_newtcpcb(pData, so)) == NULL)
855	{
856	RTMemFree(so);
857	return NULL;
858	}
859
860	SOCKET_LOCK_CREATE(so);
861	SOCKET_LOCK(so);
862	QSOCKET_LOCK(tcb);
863	insque(pData, so,&tcb);
864	NSOCK_INC();
865	QSOCKET_UNLOCK(tcb);
866
867	/*
868	* SS_FACCEPTONCE sockets must time out.
869	*/
870	if (flags & SS_FACCEPTONCE)
871	so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
872
873	so->so_state = (SS_FACCEPTCONN\|flags);
874	so->so_lport = lport; /* Kept in network format */
875	so->so_laddr.s_addr = laddr; /* Ditto */
876
877	memset(&addr, 0, sizeof(addr));
878	#ifdef RT_OS_DARWIN
879	addr.sin_len = sizeof(addr);
880	#endif
881	addr.sin_family = AF_INET;
882	addr.sin_addr.s_addr = bind_addr;
883	addr.sin_port = port;
884
885	if ( ((s = socket(AF_INET, SOCK_STREAM, 0)) < 0)
886	\|\| (setsockopt(s, SOL_SOCKET, SO_REUSEADDR,(char *)&opt, sizeof(int)) < 0)
887	\|\| (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0)
888	\|\| (listen(s, 1) < 0))
889	{
890	#ifdef RT_OS_WINDOWS
891	int tmperrno = WSAGetLastError(); /* Don't clobber the real reason we failed */
892	closesocket(s);
893	QSOCKET_LOCK(tcb);
894	sofree(pData, so);
895	QSOCKET_UNLOCK(tcb);
896	/* Restore the real errno */
897	WSASetLastError(tmperrno);
898	#else
899	int tmperrno = errno; /* Don't clobber the real reason we failed */
900	close(s);
901	QSOCKET_LOCK(tcb);
902	sofree(pData, so);
903	QSOCKET_UNLOCK(tcb);
904	/* Restore the real errno */
905	errno = tmperrno;
906	#endif
907	return NULL;
908	}
909	fd_nonblock(s);
910	setsockopt(s, SOL_SOCKET, SO_OOBINLINE,(char *)&opt, sizeof(int));
911
912	getsockname(s,(struct sockaddr *)&addr,&addrlen);
913	so->so_fport = addr.sin_port;
914	/* set socket buffers */
915	opt = pData->socket_rcv;
916	status = setsockopt(s, SOL_SOCKET, SO_RCVBUF, (char *)&opt, sizeof(int));
917	if (status < 0)
918	{
919	LogRel(("NAT: Error(%d) while setting RCV capacity to (%d)\n", errno, opt));
920	goto no_sockopt;
921	}
922	opt = pData->socket_snd;
923	status = setsockopt(s, SOL_SOCKET, SO_SNDBUF, (char *)&opt, sizeof(int));
924	if (status < 0)
925	{
926	LogRel(("NAT: Error(%d) while setting SND capacity to (%d)\n", errno, opt));
927	goto no_sockopt;
928	}
929	no_sockopt:
930	if (addr.sin_addr.s_addr == 0 \|\| addr.sin_addr.s_addr == loopback_addr.s_addr)
931	so->so_faddr = alias_addr;
932	else
933	so->so_faddr = addr.sin_addr;
934
935	so->s = s;
936	SOCKET_UNLOCK(so);
937	return so;
938	}
939
940	/*
941	* Data is available in so_rcv
942	* Just write() the data to the socket
943	* XXX not yet...
944	*/
945	void
946	sorwakeup(struct socket *so)
947	{
948	#if 0
949	sowrite(so);
950	FD_CLR(so->s,&writefds);
951	#endif
952	}
953
954	/*
955	* Data has been freed in so_snd
956	* We have room for a read() if we want to
957	* For now, don't read, it'll be done in the main loop
958	*/
959	void
960	sowwakeup(struct socket *so)
961	{
962	}
963
964	/*
965	* Various session state calls
966	* XXX Should be #define's
967	* The socket state stuff needs work, these often get call 2 or 3
968	* times each when only 1 was needed
969	*/
970	void
971	soisfconnecting(struct socket *so)
972	{
973	so->so_state &= ~(SS_NOFDREF\|SS_ISFCONNECTED\|SS_FCANTRCVMORE\|
974	SS_FCANTSENDMORE\|SS_FWDRAIN);
975	so->so_state \|= SS_ISFCONNECTING; /* Clobber other states */
976	}
977
978	void
979	soisfconnected(struct socket *so)
980	{
981	so->so_state &= ~(SS_ISFCONNECTING\|SS_FWDRAIN\|SS_NOFDREF);
982	so->so_state \|= SS_ISFCONNECTED; /* Clobber other states */
983	}
984
985	void
986	sofcantrcvmore(struct socket *so)
987	{
988	if ((so->so_state & SS_NOFDREF) == 0)
989	{
990	shutdown(so->s, 0);
991	}
992	so->so_state &= ~(SS_ISFCONNECTING);
993	if (so->so_state & SS_FCANTSENDMORE)
994	so->so_state = SS_NOFDREF; /* Don't select it */
995	/* XXX close() here as well? */
996	else
997	so->so_state \|= SS_FCANTRCVMORE;
998	}
999
1000	void
1001	sofcantsendmore(struct socket *so)
1002	{
1003	if ((so->so_state & SS_NOFDREF) == 0)
1004	shutdown(so->s, 1); /* send FIN to fhost */
1005
1006	so->so_state &= ~(SS_ISFCONNECTING);
1007	if (so->so_state & SS_FCANTRCVMORE)
1008	so->so_state = SS_NOFDREF; /* as above */
1009	else
1010	so->so_state \|= SS_FCANTSENDMORE;
1011	}
1012
1013	void
1014	soisfdisconnected(struct socket *so)
1015	{
1016	#if 0
1017	so->so_state &= ~(SS_ISFCONNECTING\|SS_ISFCONNECTED);
1018	close(so->s);
1019	so->so_state = SS_ISFDISCONNECTED;
1020	/*
1021	* XXX Do nothing ... ?
1022	*/
1023	#endif
1024	}
1025
1026	/*
1027	* Set write drain mode
1028	* Set CANTSENDMORE once all data has been write()n
1029	*/
1030	void
1031	sofwdrain(struct socket *so)
1032	{
1033	if (so->so_rcv.sb_cc)
1034	so->so_state \|= SS_FWDRAIN;
1035	else
1036	sofcantsendmore(so);
1037	}
1038
1039	static void
1040	send_icmp_to_guest(PNATState pData, char buff, size_t len, struct socket so, const struct sockaddr_in *addr)
1041	{
1042	struct ip *ip;
1043	uint32_t dst, src;
1044	char ip_copy[256];
1045	struct icmp *icp;
1046	int old_ip_len = 0;
1047	int hlen, original_hlen = 0;
1048	struct mbuf *m;
1049	struct icmp_msg *icm;
1050	uint8_t proto;
1051	int type = 0;
1052	int m_room;
1053
1054	ip = (struct ip *)buff;
1055	/* Fix ip->ip_len to contain the total packet length including the header
1056	* in _host_ byte order for all OSes. On Darwin, that value already is in
1057	* host byte order. Solaris and Darwin report only the payload. */
1058	#ifndef RT_OS_DARWIN
1059	ip->ip_len = RT_N2H_U16(ip->ip_len);
1060	#endif
1061	hlen = (ip->ip_hl << 2);
1062	#if defined(RT_OS_SOLARIS) \|\| defined(RT_OS_DARWIN)
1063	ip->ip_len += hlen;
1064	#endif
1065	if (ip->ip_len < hlen + ICMP_MINLEN)
1066	{
1067	Log(("send_icmp_to_guest: ICMP header is too small to understand which type/subtype of the datagram\n"));
1068	return;
1069	}
1070	icp = (struct icmp )((char )ip + hlen);
1071
1072	Log(("ICMP:received msg(t:%d, c:%d)\n", icp->icmp_type, icp->icmp_code));
1073	if ( icp->icmp_type != ICMP_ECHOREPLY
1074	&& icp->icmp_type != ICMP_TIMXCEED
1075	&& icp->icmp_type != ICMP_UNREACH)
1076	{
1077	return;
1078	}
1079
1080	/*
1081	* ICMP_ECHOREPLY, ICMP_TIMXCEED, ICMP_UNREACH minimal header size is
1082	* ICMP_ECHOREPLY assuming data 0
1083	* icmp_{type(8), code(8), cksum(16),identifier(16),seqnum(16)}
1084	*/
1085	if (ip->ip_len < hlen + 8)
1086	{
1087	Log(("send_icmp_to_guest: NAT accept ICMP_{ECHOREPLY, TIMXCEED, UNREACH} the minimum size is 64 (see rfc792)\n"));
1088	return;
1089	}
1090
1091	type = icp->icmp_type;
1092	if ( type == ICMP_TIMXCEED
1093	\|\| type == ICMP_UNREACH)
1094	{
1095	/*
1096	* ICMP_TIMXCEED, ICMP_UNREACH minimal header size is
1097	* icmp_{type(8), code(8), cksum(16),unused(32)} + IP header + 64 bit of original datagram
1098	*/
1099	if (ip->ip_len < hlen + 2*8 + sizeof(struct ip))
1100	{
1101	Log(("send_icmp_to_guest: NAT accept ICMP_{TIMXCEED, UNREACH} the minimum size of ipheader + 64 bit of data (see rfc792)\n"));
1102	return;
1103	}
1104	ip = &icp->icmp_ip;
1105	}
1106
1107	icm = icmp_find_original_mbuf(pData, ip);
1108	if (icm == NULL)
1109	{
1110	Log(("NAT: Can't find the corresponding packet for the received ICMP\n"));
1111	return;
1112	}
1113
1114	m = icm->im_m;
1115	Assert(m != NULL);
1116
1117	src = addr->sin_addr.s_addr;
1118	if (type == ICMP_ECHOREPLY)
1119	{
1120	struct ip ip0 = mtod(m, struct ip );
1121	struct icmp icp0 = (struct icmp )((char *)ip0 + (ip0->ip_hl << 2));
1122	if (icp0->icmp_type != ICMP_ECHO)
1123	{
1124	Log(("NAT: we haven't found echo for this reply\n"));
1125	return;
1126	}
1127	/*
1128	* while combining buffer to send (see ip_icmp.c) we control ICMP header only,
1129	* IP header combined by OS network stack, our local copy of IP header contians values
1130	* in host byte order so no byte order conversion is required. IP headers fields are converting
1131	* in ip_output0 routine only.
1132	*/
1133	if ( (ip->ip_len - hlen)
1134	!= (ip0->ip_len - (ip0->ip_hl << 2)))
1135	{
1136	Log(("NAT: ECHO(%d) lenght doesn't match ECHOREPLY(%d)\n",
1137	(ip->ip_len - hlen), (ip0->ip_len - (ip0->ip_hl << 2))));
1138	return;
1139	}
1140	}
1141
1142	/* ip points on origianal ip header */
1143	ip = mtod(m, struct ip *);
1144	proto = ip->ip_p;
1145	/* Now ip is pointing on header we've sent from guest */
1146	if ( icp->icmp_type == ICMP_TIMXCEED
1147	\|\| icp->icmp_type == ICMP_UNREACH)
1148	{
1149	old_ip_len = (ip->ip_hl << 2) + 64;
1150	if (old_ip_len > sizeof(ip_copy))
1151	old_ip_len = sizeof(ip_copy);
1152	memcpy(ip_copy, ip, old_ip_len);
1153	}
1154
1155	/* source address from original IP packet*/
1156	dst = ip->ip_src.s_addr;
1157
1158	/* overide ther tail of old packet */
1159	ip = mtod(m, struct ip ); / ip is from mbuf we've overrided */
1160	original_hlen = ip->ip_hl << 2;
1161	/* saves original ip header and options */
1162	#ifdef VBOX_WITH_SLIRP_BSD_MBUF
1163	m_copyback(pData, m, original_hlen, len - hlen, buff + hlen);
1164	ip->ip_len = m_length(m, NULL);
1165	#else
1166	/* m_room space in the saved m buffer */
1167	m_room = M_ROOM(m);
1168	if (m_room < len - hlen + original_hlen)
1169	{
1170	/* we need involve ether header length into new buffer buffer calculation */
1171	m_inc(m, if_maxlinkhdr + len - hlen + original_hlen);
1172	if (m->m_size < if_maxlinkhdr + len - hlen + original_hlen)
1173	{
1174	Log(("send_icmp_to_guest: extending buffer was failed (packet is dropped)\n"));
1175	return;
1176	}
1177	}
1178	memcpy(m->m_data + original_hlen, buff + hlen, len - hlen);
1179	m->m_len = len - hlen + original_hlen;
1180	ip->ip_len = m->m_len;
1181	#endif
1182	ip->ip_p = IPPROTO_ICMP; /* the original package could be whatever, but we're response via ICMP*/
1183
1184	icp = (struct icmp )((char )ip + (ip->ip_hl << 2));
1185	type = icp->icmp_type;
1186	if ( type == ICMP_TIMXCEED
1187	\|\| type == ICMP_UNREACH)
1188	{
1189	/* according RFC 793 error messages required copy of initial IP header + 64 bit */
1190	memcpy(&icp->icmp_ip, ip_copy, old_ip_len);
1191	ip->ip_tos = ((ip->ip_tos & 0x1E) \| 0xC0); /* high priority for errors */
1192	}
1193
1194	ip->ip_src.s_addr = src;
1195	ip->ip_dst.s_addr = dst;
1196	icmp_reflect(pData, m);
1197	LIST_REMOVE(icm, im_list);
1198	/* Don't call m_free here*/
1199
1200	if ( type == ICMP_TIMXCEED
1201	\|\| type == ICMP_UNREACH)
1202	{
1203	icm->im_so->so_m = NULL;
1204	switch (proto)
1205	{
1206	case IPPROTO_UDP:
1207	/XXX: so->so_m already freed so we shouldn't call sofree /
1208	udp_detach(pData, icm->im_so);
1209	break;
1210	case IPPROTO_TCP:
1211	/close tcp should be here /
1212	break;
1213	default:
1214	/* do nothing */
1215	break;
1216	}
1217	}
1218	RTMemFree(icm);
1219	}
1220
1221	#ifdef RT_OS_WINDOWS
1222	static void
1223	sorecvfrom_icmp_win(PNATState pData, struct socket *so)
1224	{
1225	int len;
1226	int i;
1227	struct ip *ip;
1228	struct mbuf *m;
1229	struct icmp *icp;
1230	struct icmp_msg *icm;
1231	struct ip ip_broken; / ICMP returns header + 64 bit of packet */
1232	uint32_t src;
1233	ICMP_ECHO_REPLY *icr;
1234	int hlen = 0;
1235	int data_len = 0;
1236	int nbytes = 0;
1237	u_char code = ~0;
1238	int out_len;
1239	int size;
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	out_len = ETH_HLEN + sizeof(struct ip) + 8;
1272	size;
1273	size = MCLBYTES;
1274	if (out_len < MSIZE)
1275	size = MCLBYTES;
1276	else if (out_len < MCLBYTES)
1277	size = MCLBYTES;
1278	else if (out_len < MJUM9BYTES)
1279	size = MJUM9BYTES;
1280	else if (out_len < MJUM16BYTES)
1281	size = MJUM16BYTES;
1282	else
1283	AssertMsgFailed(("Unsupported size"));
1284
1285	m = m_getjcl(pData, M_NOWAIT, MT_HEADER, M_PKTHDR, size);
1286	# endif
1287	m->m_len = 0;
1288	m->m_data += if_maxlinkhdr;
1289	ip = mtod(m, struct ip *);
1290	ip->ip_src.s_addr = icr[i].Address;
1291	ip->ip_p = IPPROTO_ICMP;
1292	ip->ip_dst.s_addr = so->so_laddr.s_addr; /XXX: still the hack/
1293	data_len = sizeof(struct ip);
1294	ip->ip_hl = data_len >> 2; /* requiered for icmp_reflect, no IP options */
1295	ip->ip_ttl = icr[i].Options.Ttl;
1296
1297	icp = (struct icmp )&ip[1]; / no options */
1298	icp->icmp_type = ICMP_ECHOREPLY;
1299	icp->icmp_code = 0;
1300	icp->icmp_id = so->so_icmp_id;
1301	icp->icmp_seq = so->so_icmp_seq;
1302
1303	data_len += ICMP_MINLEN;
1304
1305	# ifndef VBOX_WITH_SLIRP_BSD_MBUF
1306	nbytes = (data_len + icr[i].DataSize > m->m_size? m->m_size - data_len: icr[i].DataSize);
1307	memcpy(icp->icmp_data, icr[i].Data, nbytes);
1308	# else
1309	hlen = (ip->ip_hl << 2);
1310	m->m_pkthdr.header = mtod(m, void *);
1311	m->m_len = data_len;
1312
1313	m_copyback(pData, m, hlen + 8, icr[i].DataSize, icr[i].Data);
1314	# endif
1315
1316	data_len += icr[i].DataSize;
1317
1318	ip->ip_len = data_len;
1319	m->m_len = ip->ip_len;
1320
1321	icmp_reflect(pData, m);
1322	break;
1323	case IP_TTL_EXPIRED_TRANSIT: /* TTL expired */
1324
1325	ip_broken = icr[i].Data;
1326	icm = icmp_find_original_mbuf(pData, ip_broken);
1327	if (icm == NULL) {
1328	Log(("ICMP: can't find original package (first double word %x)\n", (uint32_t )ip_broken));
1329	return;
1330	}
1331	m = icm->im_m;
1332	ip = mtod(m, struct ip *);
1333	ip->ip_ttl = icr[i].Options.Ttl;
1334	src = ip->ip_src.s_addr;
1335	ip->ip_dst.s_addr = src;
1336	ip->ip_dst.s_addr = icr[i].Address;
1337
1338	hlen = (ip->ip_hl << 2);
1339	icp = (struct icmp )((char )ip + hlen);
1340	ip_broken->ip_src.s_addr = src; /it packet sent from host not from guest/
1341	data_len = (ip_broken->ip_hl << 2) + 64;
1342
1343	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
1344	nbytes =(hlen + ICMP_MINLEN + data_len > m->m_size? m->m_size - (hlen + ICMP_MINLEN): data_len);
1345	memcpy(icp->icmp_data, ip_broken, nbytes);
1346	#else
1347	m->m_len = data_len;
1348	m->m_pkthdr.header = mtod(m, void *);
1349	m_copyback(pData, m, ip->ip_hl >> 2, icr[i].DataSize, icr[i].Data);
1350	#endif
1351	icmp_reflect(pData, m);
1352	break;
1353	default:
1354	Log(("ICMP(default): message with Status: %x was received from %x\n", icr[i].Status, icr[i].Address));
1355	break;
1356	}
1357	}
1358	}
1359	#else /* !RT_OS_WINDOWS */
1360	static void sorecvfrom_icmp_unix(PNATState pData, struct socket *so)
1361	{
1362	struct sockaddr_in addr;
1363	socklen_t addrlen = sizeof(struct sockaddr_in);
1364	struct ip ip;
1365	char *buff;
1366	int len = 0;
1367
1368	/* 1- step: read the ip header */
1369	len = recvfrom(so->s, &ip, sizeof(struct ip), MSG_PEEK,
1370	(struct sockaddr *)&addr, &addrlen);
1371	if ( len < 0
1372	&& ( errno == EAGAIN
1373	\|\| errno == EWOULDBLOCK
1374	\|\| errno == EINPROGRESS
1375	\|\| errno == ENOTCONN))
1376	{
1377	Log(("sorecvfrom_icmp_unix: 1 - step can't read IP datagramm (would block)\n"));
1378	return;
1379	}
1380
1381	if ( len < sizeof(struct ip)
1382	\|\| len < 0
1383	\|\| len == 0)
1384	{
1385	u_char code;
1386	code = ICMP_UNREACH_PORT;
1387
1388	if (errno == EHOSTUNREACH)
1389	code = ICMP_UNREACH_HOST;
1390	else if (errno == ENETUNREACH)
1391	code = ICMP_UNREACH_NET;
1392
1393	LogRel((" udp icmp rx errno = %d-%s\n",
1394	errno, strerror(errno)));
1395	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
1396	so->so_m = NULL;
1397	Log(("sorecvfrom_icmp_unix: 1 - step can't read IP datagramm \n"));
1398	return;
1399	}
1400	/* basic check of IP header */
1401	if ( ip.ip_v != IPVERSION
1402	# ifndef RT_OS_DARWIN
1403	\|\| ip.ip_p != IPPROTO_ICMP
1404	# endif
1405	)
1406	{
1407	Log(("sorecvfrom_icmp_unix: 1 - step IP isn't IPv4 \n"));
1408	return;
1409	}
1410	# ifndef RT_OS_DARWIN
1411	/* Darwin reports the IP length already in host byte order. */
1412	ip.ip_len = RT_N2H_U16(ip.ip_len);
1413	# endif
1414	# if defined(RT_OS_SOLARIS) \|\| defined(RT_OS_DARWIN)
1415	/* Solaris and Darwin report the payload only */
1416	ip.ip_len += (ip.ip_hl << 2);
1417	# endif
1418	/* Note: ip->ip_len in host byte order (all OS) */
1419	len = ip.ip_len;
1420	buff = RTMemAlloc(len);
1421	if (buff == NULL)
1422	{
1423	Log(("sorecvfrom_icmp_unix: 1 - step can't allocate enought room for datagram\n"));
1424	return;
1425	}
1426	/* 2 - step: we're reading rest of the datagramm to the buffer */
1427	addrlen = sizeof(struct sockaddr_in);
1428	memset(&addr, 0, addrlen);
1429	len = recvfrom(so->s, buff, len, 0,
1430	(struct sockaddr *)&addr, &addrlen);
1431	if ( len < 0
1432	&& ( errno == EAGAIN
1433	\|\| errno == EWOULDBLOCK
1434	\|\| errno == EINPROGRESS
1435	\|\| errno == ENOTCONN))
1436	{
1437	Log(("sorecvfrom_icmp_unix: 2 - step can't read IP body (would block expected:%d)\n",
1438	ip.ip_len));
1439	RTMemFree(buff);
1440	return;
1441	}
1442	if ( len < 0
1443	\|\| len == 0)
1444	{
1445	Log(("sorecvfrom_icmp_unix: 2 - step read of the rest of datagramm is fallen (errno:%d, len:%d expected: %d)\n",
1446	errno, len, (ip.ip_len - sizeof(struct ip))));
1447	RTMemFree(buff);
1448	return;
1449	}
1450	/* len is modified in 2nd read, when the rest of the datagramm was read */
1451	send_icmp_to_guest(pData, buff, len, so, &addr);
1452	RTMemFree(buff);
1453	}
1454	#endif /* !RT_OS_WINDOWS */

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/socket.c@ 28390

Download in other formats: