socket.c@ 64137

Last change on this file since 64137 was 63676, checked in by vboxsync, 8 years ago
NAT: soread - errno is valid only when syscall indicates failure, so set saved sockerr to 0 on success (because POLA).
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 41.6 KB

Line
1	/* $Id: socket.c 63676 2016-08-31 16:17:19Z vboxsync $ */
2	/** @file
3	* NAT - socket handling.
4	*/
5
6	/*
7	* Copyright (C) 2006-2016 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/*
19	* This code is based on:
20	*
21	* Copyright (c) 1995 Danny Gasparovski.
22	*
23	* Please read the file COPYRIGHT for the
24	* terms and conditions of the copyright.
25	*/
26
27	#include <slirp.h>
28	#include "ip_icmp.h"
29	#include "main.h"
30	#ifdef __sun__
31	#include <sys/filio.h>
32	#endif
33	#include <VBox/vmm/pdmdrv.h>
34	#if defined (RT_OS_WINDOWS)
35	#include <iprt/win/iphlpapi.h>
36	#include <icmpapi.h>
37	#endif
38
39	#if defined(DECLARE_IOVEC) && defined(RT_OS_WINDOWS)
40	AssertCompileMembersSameSizeAndOffset(struct iovec, iov_base, WSABUF, buf);
41	AssertCompileMembersSameSizeAndOffset(struct iovec, iov_len, WSABUF, len);
42	#endif
43
44	#ifdef VBOX_WITH_NAT_UDP_SOCKET_CLONE
45	/**
46	*
47	*/
48	struct socket * soCloneUDPSocketWithForegnAddr(PNATState pData, bool fBindSocket, struct socket *pSo, uint32_t u32ForeignAddr)
49	{
50	struct socket *pNewSocket = NULL;
51	LogFlowFunc(("Enter: fBindSocket:%RTbool, so:%R[natsock], u32ForeignAddr:%RTnaipv4\n", fBindSocket, pSo, u32ForeignAddr));
52	pNewSocket = socreate();
53	if (!pNewSocket)
54	{
55	LogFunc(("Can't create socket\n"));
56	LogFlowFunc(("Leave: NULL\n"));
57	return NULL;
58	}
59	if (fBindSocket)
60	{
61	if (udp_attach(pData, pNewSocket, 0) <= 0)
62	{
63	sofree(pData, pNewSocket);
64	LogFunc(("Can't attach fresh created socket\n"));
65	return NULL;
66	}
67	}
68	else
69	{
70	pNewSocket->so_cloneOf = (struct socket *)pSo;
71	pNewSocket->s = pSo->s;
72	insque(pData, pNewSocket, &udb);
73	}
74	pNewSocket->so_laddr = pSo->so_laddr;
75	pNewSocket->so_lport = pSo->so_lport;
76	pNewSocket->so_faddr.s_addr = u32ForeignAddr;
77	pNewSocket->so_fport = pSo->so_fport;
78	pSo->so_cCloneCounter++;
79	LogFlowFunc(("Leave: %R[natsock]\n", pNewSocket));
80	return pNewSocket;
81	}
82
83	struct socket soLookUpClonedUDPSocket(PNATState pData, const struct socket pcSo, uint32_t u32ForeignAddress)
84	{
85	struct socket *pSoClone = NULL;
86	LogFlowFunc(("Enter: pcSo:%R[natsock], u32ForeignAddress:%RTnaipv4\n", pcSo, u32ForeignAddress));
87	for (pSoClone = udb.so_next; pSoClone != &udb; pSoClone = pSoClone->so_next)
88	{
89	if ( pSoClone->so_cloneOf
90	&& pSoClone->so_cloneOf == pcSo
91	&& pSoClone->so_lport == pcSo->so_lport
92	&& pSoClone->so_fport == pcSo->so_fport
93	&& pSoClone->so_laddr.s_addr == pcSo->so_laddr.s_addr
94	&& pSoClone->so_faddr.s_addr == u32ForeignAddress)
95	goto done;
96	}
97	pSoClone = NULL;
98	done:
99	LogFlowFunc(("Leave: pSoClone: %R[natsock]\n", pSoClone));
100	return pSoClone;
101	}
102	#endif
103
104	#ifdef VBOX_WITH_NAT_SEND2HOME
105	DECLINLINE(bool) slirpSend2Home(PNATState pData, struct socket pSo, const void pvBuf, uint32_t cbBuf, int iFlags)
106	{
107	int idxAddr;
108	int ret = 0;
109	bool fSendDone = false;
110	LogFlowFunc(("Enter pSo:%R[natsock] pvBuf: %p, cbBuf: %d, iFlags: %d\n", pSo, pvBuf, cbBuf, iFlags));
111	for (idxAddr = 0; idxAddr < pData->cInHomeAddressSize; ++idxAddr)
112	{
113
114	struct socket *pNewSocket = soCloneUDPSocketWithForegnAddr(pData, pSo, pData->pInSockAddrHomeAddress[idxAddr].sin_addr);
115	AssertReturn((pNewSocket, false));
116	pData->pInSockAddrHomeAddress[idxAddr].sin_port = pSo->so_fport;
117	/** @todo more verbose on errors,
118	* @note: we shouldn't care if this send fail or not (we're in broadcast).
119	*/
120	LogFunc(("send %d bytes to %RTnaipv4 from %R[natsock]\n", cbBuf, pData->pInSockAddrHomeAddress[idxAddr].sin_addr.s_addr, pNewSocket));
121	ret = sendto(pNewSocket->s, pvBuf, cbBuf, iFlags, (struct sockaddr *)&pData->pInSockAddrHomeAddress[idxAddr], sizeof(struct sockaddr_in));
122	if (ret < 0)
123	LogFunc(("Failed to send %d bytes to %RTnaipv4\n", cbBuf, pData->pInSockAddrHomeAddress[idxAddr].sin_addr.s_addr));
124	fSendDone \|= ret > 0;
125	}
126	LogFlowFunc(("Leave %RTbool\n", fSendDone));
127	return fSendDone;
128	}
129	#endif /* !VBOX_WITH_NAT_SEND2HOME */
130
131	#if !defined(RT_OS_WINDOWS)
132	static void send_icmp_to_guest(PNATState, char , size_t, const struct sockaddr_in );
133	static void sorecvfrom_icmp_unix(PNATState, struct socket *);
134	#endif /* !RT_OS_WINDOWS */
135
136	void
137	so_init(void)
138	{
139	}
140
141	struct socket *
142	solookup(struct socket *head, struct in_addr laddr,
143	u_int lport, struct in_addr faddr, u_int fport)
144	{
145	struct socket *so;
146
147	for (so = head->so_next; so != head; so = so->so_next)
148	{
149	if ( so->so_lport == lport
150	&& so->so_laddr.s_addr == laddr.s_addr
151	&& so->so_faddr.s_addr == faddr.s_addr
152	&& so->so_fport == fport)
153	return so;
154	}
155
156	return (struct socket *)NULL;
157	}
158
159	/*
160	* Create a new socket, initialise the fields
161	* It is the responsibility of the caller to
162	* insque() it into the correct linked-list
163	*/
164	struct socket *
165	socreate(void)
166	{
167	struct socket *so;
168
169	so = (struct socket *)RTMemAllocZ(sizeof(struct socket));
170	if (so)
171	{
172	so->so_state = SS_NOFDREF;
173	so->s = -1;
174	#if !defined(RT_OS_WINDOWS)
175	so->so_poll_index = -1;
176	#endif
177	}
178	return so;
179	}
180
181	/*
182	* remque and free a socket, clobber cache
183	*/
184	void
185	sofree(PNATState pData, struct socket *so)
186	{
187	LogFlowFunc(("ENTER:%R[natsock]\n", so));
188	/*
189	* We should not remove socket when polling routine do the polling
190	* instead we mark it for deletion.
191	*/
192	if (so->fUnderPolling)
193	{
194	so->fShouldBeRemoved = 1;
195	LogFlowFunc(("LEAVE:%R[natsock] postponed deletion\n", so));
196	return;
197	}
198	/**
199	* Check that we don't freeng socket with tcbcb
200	*/
201	Assert(!sototcpcb(so));
202	/* udp checks */
203	Assert(!so->so_timeout);
204	Assert(!so->so_timeout_arg);
205	if (so == tcp_last_so)
206	tcp_last_so = &tcb;
207	else if (so == udp_last_so)
208	udp_last_so = &udb;
209
210	/* check if mbuf haven't been already freed */
211	if (so->so_m != NULL)
212	{
213	m_freem(pData, so->so_m);
214	so->so_m = NULL;
215	}
216
217	if (so->so_ohdr != NULL)
218	{
219	RTMemFree(so->so_ohdr);
220	so->so_ohdr = NULL;
221	}
222
223	if (so->so_next && so->so_prev)
224	{
225	remque(pData, so); /* crashes if so is not in a queue */
226	NSOCK_DEC();
227	}
228
229	RTMemFree(so);
230	LogFlowFuncLeave();
231	}
232
233	/*
234	* Read from so's socket into sb_snd, updating all relevant sbuf fields
235	* NOTE: This will only be called if it is select()ed for reading, so
236	* a read() of 0 (or less) means it's disconnected
237	*/
238	int
239	soread(PNATState pData, struct socket *so)
240	{
241	int n, nn, lss, total;
242	struct sbuf *sb = &so->so_snd;
243	u_int len = sb->sb_datalen - sb->sb_cc;
244	struct iovec iov[2];
245	int mss = so->so_tcpcb->t_maxseg;
246	int sockerr;
247
248	STAM_PROFILE_START(&pData->StatIOread, a);
249	STAM_COUNTER_RESET(&pData->StatIORead_in_1);
250	STAM_COUNTER_RESET(&pData->StatIORead_in_2);
251
252	QSOCKET_LOCK(tcb);
253	SOCKET_LOCK(so);
254	QSOCKET_UNLOCK(tcb);
255
256	LogFlow(("soread: so = %R[natsock]\n", so));
257	Log2(("%s: so = %R[natsock] so->so_snd = %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, so, sb));
258
259	/*
260	* No need to check if there's enough room to read.
261	* soread wouldn't have been called if there weren't
262	*/
263
264	len = sb->sb_datalen - sb->sb_cc;
265
266	iov[0].iov_base = sb->sb_wptr;
267	iov[1].iov_base = 0;
268	iov[1].iov_len = 0;
269	if (sb->sb_wptr < sb->sb_rptr)
270	{
271	iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
272	/* Should never succeed, but... */
273	if (iov[0].iov_len > len)
274	iov[0].iov_len = len;
275	if (iov[0].iov_len > mss)
276	iov[0].iov_len -= iov[0].iov_len%mss;
277	n = 1;
278	}
279	else
280	{
281	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
282	/* Should never succeed, but... */
283	if (iov[0].iov_len > len)
284	iov[0].iov_len = len;
285	len -= iov[0].iov_len;
286	if (len)
287	{
288	iov[1].iov_base = sb->sb_data;
289	iov[1].iov_len = sb->sb_rptr - sb->sb_data;
290	if (iov[1].iov_len > len)
291	iov[1].iov_len = len;
292	total = iov[0].iov_len + iov[1].iov_len;
293	if (total > mss)
294	{
295	lss = total % mss;
296	if (iov[1].iov_len > lss)
297	{
298	iov[1].iov_len -= lss;
299	n = 2;
300	}
301	else
302	{
303	lss -= iov[1].iov_len;
304	iov[0].iov_len -= lss;
305	n = 1;
306	}
307	}
308	else
309	n = 2;
310	}
311	else
312	{
313	if (iov[0].iov_len > mss)
314	iov[0].iov_len -= iov[0].iov_len%mss;
315	n = 1;
316	}
317	}
318
319	#ifdef HAVE_READV
320	nn = readv(so->s, (struct iovec *)iov, n);
321	#else
322	nn = recv(so->s, iov[0].iov_base, iov[0].iov_len, (so->so_tcpcb->t_force? MSG_OOB:0));
323	#endif
324	if (nn < 0)
325	sockerr = errno; /* save it, as it may be clobbered by logging */
326	else
327	sockerr = 0;
328
329	Log2(("%s: read(1) nn = %d bytes\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn));
330	Log2(("%s: so = %R[natsock] so->so_snd = %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, so, sb));
331	if (nn <= 0)
332	{
333	if (nn == 0) /* XXX: should this be inside #if defined(RT_OS_WINDOWS)? */
334	{
335	/*
336	* Special case for WSAEnumNetworkEvents: If we receive 0 bytes that
337	* _could_ mean that the connection is closed. But we will receive an
338	* FD_CLOSE event later if the connection was _really_ closed. With
339	* www.youtube.com I see this very often. Closing the socket too early
340	* would be dangerous.
341	*/
342	int status;
343	unsigned long pending = 0;
344	status = ioctlsocket(so->s, FIONREAD, &pending);
345	if (status < 0)
346	Log(("NAT:%s: error in WSAIoctl: %d\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, errno));
347	if (pending != 0)
348	{
349	SOCKET_UNLOCK(so);
350	STAM_PROFILE_STOP(&pData->StatIOread, a);
351	return 0;
352	}
353	}
354
355	if ( nn < 0
356	&& soIgnorableErrorCode(sockerr))
357	{
358	SOCKET_UNLOCK(so);
359	STAM_PROFILE_STOP(&pData->StatIOread, a);
360	return 0;
361	}
362	else
363	{
364	int fUninitializedTemplate = 0;
365	fUninitializedTemplate = RT_BOOL(( sototcpcb(so)
366	&& ( sototcpcb(so)->t_template.ti_src.s_addr == INADDR_ANY
367	\|\| sototcpcb(so)->t_template.ti_dst.s_addr == INADDR_ANY)));
368	/* nn == 0 means peer has performed an orderly shutdown */
369	Log2(("%s: disconnected, nn = %d, errno = %d (%s)\n",
370	RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn, sockerr, strerror(sockerr)));
371	sofcantrcvmore(so);
372	if (!fUninitializedTemplate)
373	tcp_sockclosed(pData, sototcpcb(so));
374	else
375	tcp_drop(pData, sototcpcb(so), sockerr);
376	SOCKET_UNLOCK(so);
377	STAM_PROFILE_STOP(&pData->StatIOread, a);
378	return -1;
379	}
380	}
381	STAM_STATS(
382	if (n == 1)
383	{
384	STAM_COUNTER_INC(&pData->StatIORead_in_1);
385	STAM_COUNTER_ADD(&pData->StatIORead_in_1_bytes, nn);
386	}
387	else
388	{
389	STAM_COUNTER_INC(&pData->StatIORead_in_2);
390	STAM_COUNTER_ADD(&pData->StatIORead_in_2_1st_bytes, nn);
391	}
392	);
393
394	#ifndef HAVE_READV
395	/*
396	* If there was no error, try and read the second time round
397	* We read again if n = 2 (ie, there's another part of the buffer)
398	* and we read as much as we could in the first read
399	* We don't test for <= 0 this time, because there legitimately
400	* might not be any more data (since the socket is non-blocking),
401	* a close will be detected on next iteration.
402	* A return of -1 wont (shouldn't) happen, since it didn't happen above
403	*/
404	if (n == 2 && (unsigned)nn == iov[0].iov_len)
405	{
406	int ret;
407	ret = recv(so->s, iov[1].iov_base, iov[1].iov_len, 0);
408	if (ret > 0)
409	nn += ret;
410	STAM_STATS(
411	if (ret > 0)
412	{
413	STAM_COUNTER_INC(&pData->StatIORead_in_2);
414	STAM_COUNTER_ADD(&pData->StatIORead_in_2_2nd_bytes, ret);
415	}
416	);
417	}
418
419	Log2(("%s: read(2) nn = %d bytes\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn));
420	#endif
421
422	/* Update fields */
423	sb->sb_cc += nn;
424	sb->sb_wptr += nn;
425	Log2(("%s: update so_snd (readed nn = %d) %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn, sb));
426	if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
427	{
428	sb->sb_wptr -= sb->sb_datalen;
429	Log2(("%s: alter sb_wptr so_snd = %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, sb));
430	}
431	STAM_PROFILE_STOP(&pData->StatIOread, a);
432	SOCKET_UNLOCK(so);
433	return nn;
434	}
435
436	/*
437	* Get urgent data
438	*
439	* When the socket is created, we set it SO_OOBINLINE,
440	* so when OOB data arrives, we soread() it and everything
441	* in the send buffer is sent as urgent data
442	*/
443	void
444	sorecvoob(PNATState pData, struct socket *so)
445	{
446	struct tcpcb *tp = sototcpcb(so);
447	ssize_t ret;
448
449	LogFlowFunc(("sorecvoob: so = %R[natsock]\n", so));
450
451	/*
452	* We take a guess at how much urgent data has arrived.
453	* In most situations, when urgent data arrives, the next
454	* read() should get all the urgent data. This guess will
455	* be wrong however if more data arrives just after the
456	* urgent data, or the read() doesn't return all the
457	* urgent data.
458	*/
459	ret = soread(pData, so);
460	if (RT_LIKELY(ret > 0))
461	{
462	tp->snd_up = tp->snd_una + SBUF_LEN(&so->so_snd);
463	tp->t_force = 1;
464	tcp_output(pData, tp);
465	tp->t_force = 0;
466	}
467	}
468
469	/*
470	* Send urgent data
471	* There's a lot duplicated code here, but...
472	*/
473	int
474	sosendoob(struct socket *so)
475	{
476	struct sbuf *sb = &so->so_rcv;
477	char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
478
479	int n, len;
480
481	LogFlowFunc(("sosendoob so = %R[natsock]\n", so));
482
483	if (so->so_urgc > sizeof(buff))
484	so->so_urgc = sizeof(buff); /* XXX */
485
486	if (sb->sb_rptr < sb->sb_wptr)
487	{
488	/* We can send it directly */
489	n = send(so->s, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
490	so->so_urgc -= n;
491
492	Log2((" --- sent %d bytes urgent data, %d urgent bytes left\n",
493	n, so->so_urgc));
494	}
495	else
496	{
497	/*
498	* Since there's no sendv or sendtov like writev,
499	* we must copy all data to a linear buffer then
500	* send it all
501	*/
502	len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
503	if (len > so->so_urgc)
504	len = so->so_urgc;
505	memcpy(buff, sb->sb_rptr, len);
506	so->so_urgc -= len;
507	if (so->so_urgc)
508	{
509	n = sb->sb_wptr - sb->sb_data;
510	if (n > so->so_urgc)
511	n = so->so_urgc;
512	memcpy(buff + len, sb->sb_data, n);
513	so->so_urgc -= n;
514	len += n;
515	}
516	n = send(so->s, buff, len, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
517	#ifdef DEBUG
518	if (n != len)
519	Log(("Didn't send all data urgently XXXXX\n"));
520	#endif
521	Log2((" ---2 sent %d bytes urgent data, %d urgent bytes left\n",
522	n, so->so_urgc));
523	}
524
525	sb->sb_cc -= n;
526	sb->sb_rptr += n;
527	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
528	sb->sb_rptr -= sb->sb_datalen;
529
530	return n;
531	}
532
533	/*
534	* Write data from so_rcv to so's socket,
535	* updating all sbuf field as necessary
536	*/
537	int
538	sowrite(PNATState pData, struct socket *so)
539	{
540	int n, nn;
541	struct sbuf *sb = &so->so_rcv;
542	u_int len = sb->sb_cc;
543	struct iovec iov[2];
544
545	STAM_PROFILE_START(&pData->StatIOwrite, a);
546	STAM_COUNTER_RESET(&pData->StatIOWrite_in_1);
547	STAM_COUNTER_RESET(&pData->StatIOWrite_in_1_bytes);
548	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2);
549	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2_1st_bytes);
550	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2_2nd_bytes);
551	STAM_COUNTER_RESET(&pData->StatIOWrite_no_w);
552	STAM_COUNTER_RESET(&pData->StatIOWrite_rest);
553	STAM_COUNTER_RESET(&pData->StatIOWrite_rest_bytes);
554	LogFlowFunc(("so = %R[natsock]\n", so));
555	Log2(("%s: so = %R[natsock] so->so_rcv = %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, so, sb));
556	QSOCKET_LOCK(tcb);
557	SOCKET_LOCK(so);
558	QSOCKET_UNLOCK(tcb);
559	if (so->so_urgc)
560	{
561	sosendoob(so);
562	if (sb->sb_cc == 0)
563	{
564	SOCKET_UNLOCK(so);
565	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
566	return 0;
567	}
568	}
569
570	/*
571	* No need to check if there's something to write,
572	* sowrite wouldn't have been called otherwise
573	*/
574
575	len = sb->sb_cc;
576
577	iov[0].iov_base = sb->sb_rptr;
578	iov[1].iov_base = 0;
579	iov[1].iov_len = 0;
580	if (sb->sb_rptr < sb->sb_wptr)
581	{
582	iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
583	/* Should never succeed, but... */
584	if (iov[0].iov_len > len)
585	iov[0].iov_len = len;
586	n = 1;
587	}
588	else
589	{
590	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
591	if (iov[0].iov_len > len)
592	iov[0].iov_len = len;
593	len -= iov[0].iov_len;
594	if (len)
595	{
596	iov[1].iov_base = sb->sb_data;
597	iov[1].iov_len = sb->sb_wptr - sb->sb_data;
598	if (iov[1].iov_len > len)
599	iov[1].iov_len = len;
600	n = 2;
601	}
602	else
603	n = 1;
604	}
605	STAM_STATS({
606	if (n == 1)
607	{
608	STAM_COUNTER_INC(&pData->StatIOWrite_in_1);
609	STAM_COUNTER_ADD(&pData->StatIOWrite_in_1_bytes, iov[0].iov_len);
610	}
611	else
612	{
613	STAM_COUNTER_INC(&pData->StatIOWrite_in_2);
614	STAM_COUNTER_ADD(&pData->StatIOWrite_in_2_1st_bytes, iov[0].iov_len);
615	STAM_COUNTER_ADD(&pData->StatIOWrite_in_2_2nd_bytes, iov[1].iov_len);
616	}
617	});
618	/* Check if there's urgent data to send, and if so, send it */
619	#ifdef HAVE_READV
620	nn = writev(so->s, (const struct iovec *)iov, n);
621	#else
622	nn = send(so->s, iov[0].iov_base, iov[0].iov_len, 0);
623	#endif
624	Log2(("%s: wrote(1) nn = %d bytes\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn));
625	/* This should never happen, but people tell me it does shrug */
626	if ( nn < 0
627	&& soIgnorableErrorCode(errno))
628	{
629	SOCKET_UNLOCK(so);
630	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
631	return 0;
632	}
633
634	if (nn < 0 \|\| (nn == 0 && iov[0].iov_len > 0))
635	{
636	Log2(("%s: disconnected, so->so_state = %x, errno = %d\n",
637	RT_GCC_EXTENSION __PRETTY_FUNCTION__, so->so_state, errno));
638	sofcantsendmore(so);
639	tcp_sockclosed(pData, sototcpcb(so));
640	SOCKET_UNLOCK(so);
641	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
642	return -1;
643	}
644
645	#ifndef HAVE_READV
646	if (n == 2 && (unsigned)nn == iov[0].iov_len)
647	{
648	int ret;
649	ret = send(so->s, iov[1].iov_base, iov[1].iov_len, 0);
650	if (ret > 0)
651	nn += ret;
652	# ifdef VBOX_WITH_STATISTICS
653	if (ret > 0 && ret != (ssize_t)iov[1].iov_len)
654	{
655	STAM_COUNTER_INC(&pData->StatIOWrite_rest);
656	STAM_COUNTER_ADD(&pData->StatIOWrite_rest_bytes, (iov[1].iov_len - ret));
657	}
658	#endif
659	}
660	Log2(("%s: wrote(2) nn = %d bytes\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn));
661	#endif
662
663	/* Update sbuf */
664	sb->sb_cc -= nn;
665	sb->sb_rptr += nn;
666	Log2(("%s: update so_rcv (written nn = %d) %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn, sb));
667	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
668	{
669	sb->sb_rptr -= sb->sb_datalen;
670	Log2(("%s: alter sb_rptr of so_rcv %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, sb));
671	}
672
673	/*
674	* If in DRAIN mode, and there's no more data, set
675	* it CANTSENDMORE
676	*/
677	if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
678	sofcantsendmore(so);
679
680	SOCKET_UNLOCK(so);
681	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
682	return nn;
683	}
684
685	/*
686	* recvfrom() a UDP socket
687	*/
688	void
689	sorecvfrom(PNATState pData, struct socket *so)
690	{
691	LogFlowFunc(("sorecvfrom: so = %p\n", so));
692
693	#ifdef RT_OS_WINDOWS
694	/* ping is handled with ICMP API in ip_icmpwin.c */
695	Assert(so->so_type == IPPROTO_UDP);
696	#else
697	if (so->so_type == IPPROTO_ICMP)
698	{
699	/* This is a "ping" reply */
700	sorecvfrom_icmp_unix(pData, so);
701	udp_detach(pData, so);
702	}
703	else
704	#endif /* !RT_OS_WINDOWS */
705	{
706	static char achBuf[64 * 1024];
707
708	/* A "normal" UDP packet */
709	struct sockaddr_in addr;
710	socklen_t addrlen = sizeof(struct sockaddr_in);
711	struct iovec iov[2];
712	ssize_t nread;
713	struct mbuf *m;
714
715	QSOCKET_LOCK(udb);
716	SOCKET_LOCK(so);
717	QSOCKET_UNLOCK(udb);
718
719	m = m_getjcl(pData, M_NOWAIT, MT_HEADER, M_PKTHDR, slirp_size(pData));
720	if (m == NULL)
721	{
722	SOCKET_UNLOCK(so);
723	return;
724	}
725
726	m->m_data += ETH_HLEN;
727	m->m_pkthdr.header = mtod(m, void *);
728
729	m->m_data += sizeof(struct udpiphdr);
730
731	/* small packets will fit without copying */
732	iov[0].iov_base = mtod(m, char *);
733	iov[0].iov_len = M_TRAILINGSPACE(m);
734
735	/* large packets will spill into a temp buffer */
736	iov[1].iov_base = achBuf;
737	iov[1].iov_len = sizeof(achBuf);
738
739	#if !defined(RT_OS_WINDOWS)
740	{
741	struct msghdr mh;
742	memset(&mh, 0, sizeof(mh));
743
744	mh.msg_iov = iov;
745	mh.msg_iovlen = 2;
746	mh.msg_name = &addr;
747	mh.msg_namelen = addrlen;
748
749	nread = recvmsg(so->s, &mh, 0);
750	}
751	#else /* RT_OS_WINDOWS */
752	{
753	DWORD nbytes; /* NB: can't use nread b/c of different size */
754	DWORD flags = 0;
755	int status;
756	AssertCompile(sizeof(WSABUF) == sizeof(struct iovec));
757	AssertCompileMembersSameSizeAndOffset(WSABUF, len, struct iovec, iov_len);
758	AssertCompileMembersSameSizeAndOffset(WSABUF, buf, struct iovec, iov_base);
759	status = WSARecvFrom(so->s, (WSABUF *)&iov[0], 2, &nbytes, &flags,
760	(struct sockaddr *)&addr, &addrlen,
761	NULL, NULL);
762	if (status != SOCKET_ERROR)
763	nread = nbytes;
764	else
765	nread = -1;
766	}
767	#endif
768	if (nread >= 0)
769	{
770	if (nread <= iov[0].iov_len)
771	m->m_len = nread;
772	else
773	{
774	m->m_len = iov[0].iov_len;
775	m_append(pData, m, nread - iov[0].iov_len, iov[1].iov_base);
776	}
777	Assert(m_length(m, NULL) == (size_t)nread);
778
779	/*
780	* Hack: domain name lookup will be used the most for UDP,
781	* and since they'll only be used once there's no need
782	* for the 4 minute (or whatever) timeout... So we time them
783	* out much quicker (10 seconds for now...)
784	*/
785	if (so->so_expire)
786	{
787	if (so->so_fport != RT_H2N_U16_C(53))
788	so->so_expire = curtime + SO_EXPIRE;
789	}
790
791	/*
792	* DNS proxy requests are forwarded to the real resolver,
793	* but its socket's so_faddr is that of the DNS proxy
794	* itself.
795	*
796	* last argument should be changed if Slirp will inject IP attributes
797	*/
798	if ( pData->fUseDnsProxy
799	&& so->so_fport == RT_H2N_U16_C(53)
800	&& CTL_CHECK(so->so_faddr.s_addr, CTL_DNS))
801	dnsproxy_answer(pData, so, m);
802
803	/* packets definetly will be fragmented, could confuse receiver peer. */
804	if (nread > if_mtu)
805	m->m_flags \|= M_SKIP_FIREWALL;
806
807	/*
808	* If this packet was destined for CTL_ADDR,
809	* make it look like that's where it came from, done by udp_output
810	*/
811	udp_output(pData, so, m, &addr);
812	}
813	else
814	{
815	m_freem(pData, m);
816
817	if (!soIgnorableErrorCode(errno))
818	{
819	u_char code;
820	if (errno == EHOSTUNREACH)
821	code = ICMP_UNREACH_HOST;
822	else if (errno == ENETUNREACH)
823	code = ICMP_UNREACH_NET;
824	else
825	code = ICMP_UNREACH_PORT;
826
827	Log2((" rx error, tx icmp ICMP_UNREACH:%i\n", code));
828	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
829	so->so_m = NULL;
830	}
831	}
832
833	SOCKET_UNLOCK(so);
834	}
835	}
836
837	/*
838	* sendto() a socket
839	*/
840	int
841	sosendto(PNATState pData, struct socket so, struct mbuf m)
842	{
843	int ret;
844	struct sockaddr_in *paddr;
845	struct sockaddr addr;
846	#if 0
847	struct sockaddr_in host_addr;
848	#endif
849	caddr_t buf = 0;
850	int mlen;
851
852	LogFlowFunc(("sosendto: so = %R[natsock], m = %p\n", so, m));
853
854	memset(&addr, 0, sizeof(struct sockaddr));
855	#ifdef RT_OS_DARWIN
856	addr.sa_len = sizeof(struct sockaddr_in);
857	#endif
858	paddr = (struct sockaddr_in *)&addr;
859	paddr->sin_family = AF_INET;
860	if ((so->so_faddr.s_addr & RT_H2N_U32(pData->netmask)) == pData->special_addr.s_addr)
861	{
862	/* It's an alias */
863	uint32_t last_byte = RT_N2H_U32(so->so_faddr.s_addr) & ~pData->netmask;
864	switch(last_byte)
865	{
866	#if 0
867	/* handle this case at 'default:' */
868	case CTL_BROADCAST:
869	addr.sin_addr.s_addr = INADDR_BROADCAST;
870	/* Send the packet to host to fully emulate broadcast */
871	/** @todo r=klaus: on Linux host this causes the host to receive
872	* the packet twice for some reason. And I cannot find any place
873	* in the man pages which states that sending a broadcast does not
874	* reach the host itself. */
875	host_addr.sin_family = AF_INET;
876	host_addr.sin_port = so->so_fport;
877	host_addr.sin_addr = our_addr;
878	sendto(so->s, m->m_data, m->m_len, 0,
879	(struct sockaddr *)&host_addr, sizeof (struct sockaddr));
880	break;
881	#endif
882	case CTL_DNS:
883	case CTL_ALIAS:
884	default:
885	if (last_byte == ~pData->netmask)
886	paddr->sin_addr.s_addr = INADDR_BROADCAST;
887	else
888	paddr->sin_addr = loopback_addr;
889	break;
890	}
891	}
892	else
893	paddr->sin_addr = so->so_faddr;
894	paddr->sin_port = so->so_fport;
895
896	Log2((" sendto()ing, addr.sin_port=%d, addr.sin_addr.s_addr=%.16s\n",
897	RT_N2H_U16(paddr->sin_port), inet_ntoa(paddr->sin_addr)));
898
899	/* Don't care what port we get */
900	/*
901	* > nmap -sV -T4 -O -A -v -PU3483 255.255.255.255
902	* generates bodyless messages, annoying memmory management system.
903	*/
904	mlen = m_length(m, NULL);
905	if (mlen > 0)
906	{
907	buf = RTMemAlloc(mlen);
908	if (buf == NULL)
909	{
910	return -1;
911	}
912	m_copydata(m, 0, mlen, buf);
913	}
914	ret = sendto(so->s, buf, mlen, 0,
915	(struct sockaddr *)&addr, sizeof (struct sockaddr));
916	#ifdef VBOX_WITH_NAT_SEND2HOME
917	if (slirpIsWideCasting(pData, so->so_faddr.s_addr))
918	{
919	slirpSend2Home(pData, so, buf, mlen, 0);
920	}
921	#endif
922	if (buf)
923	RTMemFree(buf);
924	if (ret < 0)
925	{
926	Log2(("UDP: sendto fails (%s)\n", strerror(errno)));
927	return -1;
928	}
929
930	/*
931	* Kill the socket if there's no reply in 4 minutes,
932	* but only if it's an expirable socket
933	*/
934	if (so->so_expire)
935	so->so_expire = curtime + SO_EXPIRE;
936	so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
937	return 0;
938	}
939
940	/*
941	* XXX This should really be tcp_listen
942	*/
943	struct socket *
944	solisten(PNATState pData, u_int32_t bind_addr, u_int port, u_int32_t laddr, u_int lport, int flags)
945	{
946	struct sockaddr_in addr;
947	struct socket *so;
948	socklen_t addrlen = sizeof(addr);
949	int s, opt = 1;
950	int status;
951
952	LogFlowFunc(("solisten: port = %d, laddr = %x, lport = %d, flags = %x\n", port, laddr, lport, flags));
953
954	if ((so = socreate()) == NULL)
955	{
956	/* RTMemFree(so); Not sofree() ??? free(NULL) == NOP */
957	return NULL;
958	}
959
960	/* Don't tcp_attach... we don't need so_snd nor so_rcv */
961	if ((so->so_tcpcb = tcp_newtcpcb(pData, so)) == NULL)
962	{
963	RTMemFree(so);
964	return NULL;
965	}
966
967	SOCKET_LOCK_CREATE(so);
968	SOCKET_LOCK(so);
969	QSOCKET_LOCK(tcb);
970	insque(pData, so,&tcb);
971	NSOCK_INC();
972	QSOCKET_UNLOCK(tcb);
973
974	/*
975	* SS_FACCEPTONCE sockets must time out.
976	*/
977	if (flags & SS_FACCEPTONCE)
978	so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
979
980	so->so_state = (SS_FACCEPTCONN\|flags);
981	so->so_lport = lport; /* Kept in network format */
982	so->so_laddr.s_addr = laddr; /* Ditto */
983
984	memset(&addr, 0, sizeof(addr));
985	#ifdef RT_OS_DARWIN
986	addr.sin_len = sizeof(addr);
987	#endif
988	addr.sin_family = AF_INET;
989	addr.sin_addr.s_addr = bind_addr;
990	addr.sin_port = port;
991
992	/**
993	* changing listen(,1->SOMAXCONN) shouldn't be harmful for NAT's TCP/IP stack,
994	* kernel will choose the optimal value for requests queue length.
995	* @note: MSDN recommends low (2-4) values for bluetooth networking devices.
996	*/
997	if ( ((s = socket(AF_INET, SOCK_STREAM, 0)) < 0)
998	\|\| (setsockopt(s, SOL_SOCKET, SO_REUSEADDR,(char *)&opt, sizeof(int)) < 0)
999	\|\| (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0)
1000	\|\| (listen(s, pData->soMaxConn) < 0))
1001	{
1002	#ifdef RT_OS_WINDOWS
1003	int tmperrno = WSAGetLastError(); /* Don't clobber the real reason we failed */
1004	closesocket(s);
1005	QSOCKET_LOCK(tcb);
1006	sofree(pData, so);
1007	QSOCKET_UNLOCK(tcb);
1008	/* Restore the real errno */
1009	WSASetLastError(tmperrno);
1010	#else
1011	int tmperrno = errno; /* Don't clobber the real reason we failed */
1012	close(s);
1013	if (sototcpcb(so))
1014	tcp_close(pData, sototcpcb(so));
1015	else
1016	sofree(pData, so);
1017	/* Restore the real errno */
1018	errno = tmperrno;
1019	#endif
1020	return NULL;
1021	}
1022	fd_nonblock(s);
1023	setsockopt(s, SOL_SOCKET, SO_OOBINLINE,(char *)&opt, sizeof(int));
1024
1025	getsockname(s,(struct sockaddr *)&addr,&addrlen);
1026	so->so_fport = addr.sin_port;
1027	/* set socket buffers */
1028	opt = pData->socket_rcv;
1029	status = setsockopt(s, SOL_SOCKET, SO_RCVBUF, (char *)&opt, sizeof(int));
1030	if (status < 0)
1031	{
1032	LogRel(("NAT: Error(%d) while setting RCV capacity to (%d)\n", errno, opt));
1033	goto no_sockopt;
1034	}
1035	opt = pData->socket_snd;
1036	status = setsockopt(s, SOL_SOCKET, SO_SNDBUF, (char *)&opt, sizeof(int));
1037	if (status < 0)
1038	{
1039	LogRel(("NAT: Error(%d) while setting SND capacity to (%d)\n", errno, opt));
1040	goto no_sockopt;
1041	}
1042	no_sockopt:
1043	if (addr.sin_addr.s_addr == 0 \|\| addr.sin_addr.s_addr == loopback_addr.s_addr)
1044	so->so_faddr = alias_addr;
1045	else
1046	so->so_faddr = addr.sin_addr;
1047
1048	so->s = s;
1049	SOCKET_UNLOCK(so);
1050	return so;
1051	}
1052
1053	/*
1054	* Data is available in so_rcv
1055	* Just write() the data to the socket
1056	* XXX not yet...
1057	* @todo do we really need this function, what it's intended to do?
1058	*/
1059	void
1060	sorwakeup(struct socket *so)
1061	{
1062	NOREF(so);
1063	#if 0
1064	sowrite(so);
1065	FD_CLR(so->s,&writefds);
1066	#endif
1067	}
1068
1069	/*
1070	* Data has been freed in so_snd
1071	* We have room for a read() if we want to
1072	* For now, don't read, it'll be done in the main loop
1073	*/
1074	void
1075	sowwakeup(struct socket *so)
1076	{
1077	NOREF(so);
1078	}
1079
1080	/*
1081	* Various session state calls
1082	* XXX Should be #define's
1083	* The socket state stuff needs work, these often get call 2 or 3
1084	* times each when only 1 was needed
1085	*/
1086	void
1087	soisfconnecting(struct socket *so)
1088	{
1089	so->so_state &= ~(SS_NOFDREF\|SS_ISFCONNECTED\|SS_FCANTRCVMORE\|
1090	SS_FCANTSENDMORE\|SS_FWDRAIN);
1091	so->so_state \|= SS_ISFCONNECTING; /* Clobber other states */
1092	}
1093
1094	void
1095	soisfconnected(struct socket *so)
1096	{
1097	LogFlowFunc(("ENTER: so:%R[natsock]\n", so));
1098	so->so_state &= ~(SS_ISFCONNECTING\|SS_FWDRAIN\|SS_NOFDREF);
1099	so->so_state \|= SS_ISFCONNECTED; /* Clobber other states */
1100	LogFlowFunc(("LEAVE: so:%R[natsock]\n", so));
1101	}
1102
1103	void
1104	sofcantrcvmore(struct socket *so)
1105	{
1106	LogFlowFunc(("ENTER: so:%R[natsock]\n", so));
1107	if ((so->so_state & SS_NOFDREF) == 0)
1108	{
1109	shutdown(so->s, 0);
1110	}
1111	so->so_state &= ~(SS_ISFCONNECTING);
1112	if (so->so_state & SS_FCANTSENDMORE)
1113	so->so_state = SS_NOFDREF; /* Don't select it */
1114	/* XXX close() here as well? */
1115	else
1116	so->so_state \|= SS_FCANTRCVMORE;
1117	LogFlowFuncLeave();
1118	}
1119
1120	void
1121	sofcantsendmore(struct socket *so)
1122	{
1123	LogFlowFunc(("ENTER: so:%R[natsock]\n", so));
1124	if ((so->so_state & SS_NOFDREF) == 0)
1125	shutdown(so->s, 1); /* send FIN to fhost */
1126
1127	so->so_state &= ~(SS_ISFCONNECTING);
1128	if (so->so_state & SS_FCANTRCVMORE)
1129	so->so_state = SS_NOFDREF; /* as above */
1130	else
1131	so->so_state \|= SS_FCANTSENDMORE;
1132	LogFlowFuncLeave();
1133	}
1134
1135	void
1136	soisfdisconnected(struct socket *so)
1137	{
1138	NOREF(so);
1139	#if 0
1140	so->so_state &= ~(SS_ISFCONNECTING\|SS_ISFCONNECTED);
1141	close(so->s);
1142	so->so_state = SS_ISFDISCONNECTED;
1143	/*
1144	* XXX Do nothing ... ?
1145	*/
1146	#endif
1147	}
1148
1149	/*
1150	* Set write drain mode
1151	* Set CANTSENDMORE once all data has been write()n
1152	*/
1153	void
1154	sofwdrain(struct socket *so)
1155	{
1156	if (SBUF_LEN(&so->so_rcv))
1157	so->so_state \|= SS_FWDRAIN;
1158	else
1159	sofcantsendmore(so);
1160	}
1161
1162	#if !defined(RT_OS_WINDOWS)
1163	static void
1164	send_icmp_to_guest(PNATState pData, char buff, size_t len, const struct sockaddr_in addr)
1165	{
1166	struct ip *ip;
1167	uint32_t dst, src;
1168	char ip_copy[256];
1169	struct icmp *icp;
1170	int old_ip_len = 0;
1171	int hlen, original_hlen = 0;
1172	struct mbuf *m;
1173	struct icmp_msg *icm;
1174	uint8_t proto;
1175	int type = 0;
1176
1177	ip = (struct ip *)buff;
1178	/* Fix ip->ip_len to contain the total packet length including the header
1179	* in _host_ byte order for all OSes. On Darwin, that value already is in
1180	* host byte order. Solaris and Darwin report only the payload. */
1181	#ifndef RT_OS_DARWIN
1182	ip->ip_len = RT_N2H_U16(ip->ip_len);
1183	#endif
1184	hlen = (ip->ip_hl << 2);
1185	#if defined(RT_OS_SOLARIS) \|\| defined(RT_OS_DARWIN)
1186	ip->ip_len += hlen;
1187	#endif
1188	if (ip->ip_len < hlen + ICMP_MINLEN)
1189	{
1190	Log(("send_icmp_to_guest: ICMP header is too small to understand which type/subtype of the datagram\n"));
1191	return;
1192	}
1193	icp = (struct icmp )((char )ip + hlen);
1194
1195	Log(("ICMP:received msg(t:%d, c:%d)\n", icp->icmp_type, icp->icmp_code));
1196	if ( icp->icmp_type != ICMP_ECHOREPLY
1197	&& icp->icmp_type != ICMP_TIMXCEED
1198	&& icp->icmp_type != ICMP_UNREACH)
1199	{
1200	return;
1201	}
1202
1203	/*
1204	* ICMP_ECHOREPLY, ICMP_TIMXCEED, ICMP_UNREACH minimal header size is
1205	* ICMP_ECHOREPLY assuming data 0
1206	* icmp_{type(8), code(8), cksum(16),identifier(16),seqnum(16)}
1207	*/
1208	if (ip->ip_len < hlen + 8)
1209	{
1210	Log(("send_icmp_to_guest: NAT accept ICMP_{ECHOREPLY, TIMXCEED, UNREACH} the minimum size is 64 (see rfc792)\n"));
1211	return;
1212	}
1213
1214	type = icp->icmp_type;
1215	if ( type == ICMP_TIMXCEED
1216	\|\| type == ICMP_UNREACH)
1217	{
1218	/*
1219	* ICMP_TIMXCEED, ICMP_UNREACH minimal header size is
1220	* icmp_{type(8), code(8), cksum(16),unused(32)} + IP header + 64 bit of original datagram
1221	*/
1222	if (ip->ip_len < hlen + 2*8 + sizeof(struct ip))
1223	{
1224	Log(("send_icmp_to_guest: NAT accept ICMP_{TIMXCEED, UNREACH} the minimum size of ipheader + 64 bit of data (see rfc792)\n"));
1225	return;
1226	}
1227	ip = &icp->icmp_ip;
1228	}
1229
1230	icm = icmp_find_original_mbuf(pData, ip);
1231	if (icm == NULL)
1232	{
1233	Log(("NAT: Can't find the corresponding packet for the received ICMP\n"));
1234	return;
1235	}
1236
1237	m = icm->im_m;
1238	if (!m)
1239	{
1240	LogFunc(("%R[natsock] hasn't stored it's mbuf on sent\n", icm->im_so));
1241	goto done;
1242	}
1243
1244	src = addr->sin_addr.s_addr;
1245	if (type == ICMP_ECHOREPLY)
1246	{
1247	struct ip ip0 = mtod(m, struct ip );
1248	struct icmp icp0 = (struct icmp )((char *)ip0 + (ip0->ip_hl << 2));
1249	if (icp0->icmp_type != ICMP_ECHO)
1250	{
1251	Log(("NAT: we haven't found echo for this reply\n"));
1252	goto done;
1253	}
1254	/*
1255	* while combining buffer to send (see ip_icmp.c) we control ICMP header only,
1256	* IP header combined by OS network stack, our local copy of IP header contians values
1257	* in host byte order so no byte order conversion is required. IP headers fields are converting
1258	* in ip_output0 routine only.
1259	*/
1260	if ( (ip->ip_len - hlen)
1261	!= (ip0->ip_len - (ip0->ip_hl << 2)))
1262	{
1263	Log(("NAT: ECHO(%d) lenght doesn't match ECHOREPLY(%d)\n",
1264	(ip->ip_len - hlen), (ip0->ip_len - (ip0->ip_hl << 2))));
1265	goto done;
1266	}
1267	}
1268
1269	/* ip points on origianal ip header */
1270	ip = mtod(m, struct ip *);
1271	proto = ip->ip_p;
1272	/* Now ip is pointing on header we've sent from guest */
1273	if ( icp->icmp_type == ICMP_TIMXCEED
1274	\|\| icp->icmp_type == ICMP_UNREACH)
1275	{
1276	old_ip_len = (ip->ip_hl << 2) + 64;
1277	if (old_ip_len > sizeof(ip_copy))
1278	old_ip_len = sizeof(ip_copy);
1279	memcpy(ip_copy, ip, old_ip_len);
1280	}
1281
1282	/* source address from original IP packet*/
1283	dst = ip->ip_src.s_addr;
1284
1285	/* overide ther tail of old packet */
1286	ip = mtod(m, struct ip ); / ip is from mbuf we've overrided */
1287	original_hlen = ip->ip_hl << 2;
1288	/* saves original ip header and options */
1289	m_copyback(pData, m, original_hlen, len - hlen, buff + hlen);
1290	ip->ip_len = m_length(m, NULL);
1291	ip->ip_p = IPPROTO_ICMP; /* the original package could be whatever, but we're response via ICMP*/
1292
1293	icp = (struct icmp )((char )ip + (ip->ip_hl << 2));
1294	type = icp->icmp_type;
1295	if ( type == ICMP_TIMXCEED
1296	\|\| type == ICMP_UNREACH)
1297	{
1298	/* according RFC 793 error messages required copy of initial IP header + 64 bit */
1299	memcpy(&icp->icmp_ip, ip_copy, old_ip_len);
1300
1301	/* undo byte order conversions done in ip_input() */
1302	HTONS(icp->icmp_ip.ip_len);
1303	HTONS(icp->icmp_ip.ip_id);
1304	HTONS(icp->icmp_ip.ip_off);
1305
1306	ip->ip_tos = ((ip->ip_tos & 0x1E) \| 0xC0); /* high priority for errors */
1307	}
1308
1309	ip->ip_src.s_addr = src;
1310	ip->ip_dst.s_addr = dst;
1311	icmp_reflect(pData, m);
1312	/* m was freed */
1313	icm->im_m = NULL;
1314
1315	done:
1316	icmp_msg_delete(pData, icm);
1317	}
1318
1319	static void sorecvfrom_icmp_unix(PNATState pData, struct socket *so)
1320	{
1321	struct sockaddr_in addr;
1322	socklen_t addrlen = sizeof(struct sockaddr_in);
1323	struct ip ip;
1324	char *buff;
1325	int len = 0;
1326
1327	/* 1- step: read the ip header */
1328	len = recvfrom(so->s, &ip, sizeof(struct ip), MSG_PEEK,
1329	(struct sockaddr *)&addr, &addrlen);
1330	if ( len < 0
1331	&& ( soIgnorableErrorCode(errno)
1332	\|\| errno == ENOTCONN))
1333	{
1334	Log(("sorecvfrom_icmp_unix: 1 - step can't read IP datagramm (would block)\n"));
1335	return;
1336	}
1337
1338	if ( len < sizeof(struct ip)
1339	\|\| len < 0
1340	\|\| len == 0)
1341	{
1342	u_char code;
1343	code = ICMP_UNREACH_PORT;
1344
1345	if (errno == EHOSTUNREACH)
1346	code = ICMP_UNREACH_HOST;
1347	else if (errno == ENETUNREACH)
1348	code = ICMP_UNREACH_NET;
1349
1350	LogRel(("NAT: UDP ICMP rx errno=%d (%s)\n", errno, strerror(errno)));
1351	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
1352	so->so_m = NULL;
1353	Log(("sorecvfrom_icmp_unix: 1 - step can't read IP datagramm\n"));
1354	return;
1355	}
1356	/* basic check of IP header */
1357	if ( ip.ip_v != IPVERSION
1358	# ifndef RT_OS_DARWIN
1359	\|\| ip.ip_p != IPPROTO_ICMP
1360	# endif
1361	)
1362	{
1363	Log(("sorecvfrom_icmp_unix: 1 - step IP isn't IPv4\n"));
1364	return;
1365	}
1366	# ifndef RT_OS_DARWIN
1367	/* Darwin reports the IP length already in host byte order. */
1368	ip.ip_len = RT_N2H_U16(ip.ip_len);
1369	# endif
1370	# if defined(RT_OS_SOLARIS) \|\| defined(RT_OS_DARWIN)
1371	/* Solaris and Darwin report the payload only */
1372	ip.ip_len += (ip.ip_hl << 2);
1373	# endif
1374	/* Note: ip->ip_len in host byte order (all OS) */
1375	len = ip.ip_len;
1376	buff = RTMemAlloc(len);
1377	if (buff == NULL)
1378	{
1379	Log(("sorecvfrom_icmp_unix: 1 - step can't allocate enought room for datagram\n"));
1380	return;
1381	}
1382	/* 2 - step: we're reading rest of the datagramm to the buffer */
1383	addrlen = sizeof(struct sockaddr_in);
1384	memset(&addr, 0, addrlen);
1385	len = recvfrom(so->s, buff, len, 0,
1386	(struct sockaddr *)&addr, &addrlen);
1387	if ( len < 0
1388	&& ( soIgnorableErrorCode(errno)
1389	\|\| errno == ENOTCONN))
1390	{
1391	Log(("sorecvfrom_icmp_unix: 2 - step can't read IP body (would block expected:%d)\n",
1392	ip.ip_len));
1393	RTMemFree(buff);
1394	return;
1395	}
1396	if ( len < 0
1397	\|\| len == 0)
1398	{
1399	Log(("sorecvfrom_icmp_unix: 2 - step read of the rest of datagramm is fallen (errno:%d, len:%d expected: %d)\n",
1400	errno, len, (ip.ip_len - sizeof(struct ip))));
1401	RTMemFree(buff);
1402	return;
1403	}
1404	/* len is modified in 2nd read, when the rest of the datagramm was read */
1405	send_icmp_to_guest(pData, buff, len, &addr);
1406	RTMemFree(buff);
1407	}
1408	#endif /* !RT_OS_WINDOWS */

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/socket.c@ 64137

Download in other formats: