socket.c@ 74950

Last change on this file since 74950 was 72292, checked in by vboxsync, 7 years ago
NAT: sobind - untabify.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 43.1 KB

Line
1	/* $Id: socket.c 72292 2018-05-22 23:42:27Z vboxsync $ */
2	/** @file
3	* NAT - socket handling.
4	*/
5
6	/*
7	* Copyright (C) 2006-2017 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	#include <alias.h>
39
40	#if defined(DECLARE_IOVEC) && defined(RT_OS_WINDOWS)
41	AssertCompileMembersSameSizeAndOffset(struct iovec, iov_base, WSABUF, buf);
42	AssertCompileMembersSameSizeAndOffset(struct iovec, iov_len, WSABUF, len);
43	#endif
44
45	#ifdef VBOX_WITH_NAT_SEND2HOME
46	DECLINLINE(bool) slirpSend2Home(PNATState pData, struct socket pSo, const void pvBuf, uint32_t cbBuf, int iFlags)
47	{
48	int idxAddr;
49	int ret = 0;
50	bool fSendDone = false;
51	LogFlowFunc(("Enter pSo:%R[natsock] pvBuf: %p, cbBuf: %d, iFlags: %d\n", pSo, pvBuf, cbBuf, iFlags));
52	for (idxAddr = 0; idxAddr < pData->cInHomeAddressSize; ++idxAddr)
53	{
54
55	struct socket *pNewSocket = soCloneUDPSocketWithForegnAddr(pData, pSo, pData->pInSockAddrHomeAddress[idxAddr].sin_addr);
56	AssertReturn((pNewSocket, false));
57	pData->pInSockAddrHomeAddress[idxAddr].sin_port = pSo->so_fport;
58	/** @todo more verbose on errors,
59	* @note: we shouldn't care if this send fail or not (we're in broadcast).
60	*/
61	LogFunc(("send %d bytes to %RTnaipv4 from %R[natsock]\n", cbBuf, pData->pInSockAddrHomeAddress[idxAddr].sin_addr.s_addr, pNewSocket));
62	ret = sendto(pNewSocket->s, pvBuf, cbBuf, iFlags, (struct sockaddr *)&pData->pInSockAddrHomeAddress[idxAddr], sizeof(struct sockaddr_in));
63	if (ret < 0)
64	LogFunc(("Failed to send %d bytes to %RTnaipv4\n", cbBuf, pData->pInSockAddrHomeAddress[idxAddr].sin_addr.s_addr));
65	fSendDone \|= ret > 0;
66	}
67	LogFlowFunc(("Leave %RTbool\n", fSendDone));
68	return fSendDone;
69	}
70	#endif /* !VBOX_WITH_NAT_SEND2HOME */
71
72	#if !defined(RT_OS_WINDOWS)
73	static void send_icmp_to_guest(PNATState, char , size_t, const struct sockaddr_in );
74	static void sorecvfrom_icmp_unix(PNATState, struct socket *);
75	#endif /* !RT_OS_WINDOWS */
76
77	void
78	so_init(void)
79	{
80	}
81
82	struct socket *
83	solookup(struct socket *head, struct in_addr laddr,
84	u_int lport, struct in_addr faddr, u_int fport)
85	{
86	struct socket *so;
87
88	for (so = head->so_next; so != head; so = so->so_next)
89	{
90	if ( so->so_lport == lport
91	&& so->so_laddr.s_addr == laddr.s_addr
92	&& so->so_faddr.s_addr == faddr.s_addr
93	&& so->so_fport == fport)
94	return so;
95	}
96
97	return (struct socket *)NULL;
98	}
99
100	/*
101	* Create a new socket, initialise the fields
102	* It is the responsibility of the caller to
103	* insque() it into the correct linked-list
104	*/
105	struct socket *
106	socreate(void)
107	{
108	struct socket *so;
109
110	so = (struct socket *)RTMemAllocZ(sizeof(struct socket));
111	if (so)
112	{
113	so->so_state = SS_NOFDREF;
114	so->s = -1;
115	#if !defined(RT_OS_WINDOWS)
116	so->so_poll_index = -1;
117	#endif
118	}
119	return so;
120	}
121
122	/*
123	* remque and free a socket, clobber cache
124	*/
125	void
126	sofree(PNATState pData, struct socket *so)
127	{
128	LogFlowFunc(("ENTER:%R[natsock]\n", so));
129	/*
130	* We should not remove socket when polling routine do the polling
131	* instead we mark it for deletion.
132	*/
133	if (so->fUnderPolling)
134	{
135	so->fShouldBeRemoved = 1;
136	LogFlowFunc(("LEAVE:%R[natsock] postponed deletion\n", so));
137	return;
138	}
139	/**
140	* Check that we don't freeng socket with tcbcb
141	*/
142	Assert(!sototcpcb(so));
143	/* udp checks */
144	Assert(!so->so_timeout);
145	Assert(!so->so_timeout_arg);
146	if (so == tcp_last_so)
147	tcp_last_so = &tcb;
148	else if (so == udp_last_so)
149	udp_last_so = &udb;
150
151	/* check if mbuf haven't been already freed */
152	if (so->so_m != NULL)
153	{
154	m_freem(pData, so->so_m);
155	so->so_m = NULL;
156	}
157
158	if (so->so_ohdr != NULL)
159	{
160	RTMemFree(so->so_ohdr);
161	so->so_ohdr = NULL;
162	}
163
164	if (so->so_next && so->so_prev)
165	{
166	remque(pData, so); /* crashes if so is not in a queue */
167	NSOCK_DEC();
168	}
169
170	RTMemFree(so);
171	LogFlowFuncLeave();
172	}
173
174
175	/*
176	* Worker for sobind() below.
177	*/
178	static int
179	sobindto(struct socket *so, uint32_t addr, uint16_t port)
180	{
181	struct sockaddr_in self;
182	int status;
183
184	if (addr == INADDR_ANY && port == 0 && so->so_type != IPPROTO_UDP)
185	{
186	/* TCP sockets without constraints don't need to be bound */
187	Log2(("NAT: sobind: %s guest %RTnaipv4:%d - nothing to do\n",
188	so->so_type == IPPROTO_UDP ? "udp" : "tcp",
189	so->so_laddr.s_addr, ntohs(so->so_lport)));
190	return 0;
191	}
192
193	RT_ZERO(self);
194	#ifdef RT_OS_DARWIN
195	self.sin_len = sizeof(self);
196	#endif
197	self.sin_family = AF_INET;
198	self.sin_addr.s_addr = addr;
199	self.sin_port = port;
200
201	status = bind(so->s, (struct sockaddr *)&self, sizeof(self));
202	if (status == 0)
203	{
204	Log2(("NAT: sobind: %s guest %RTnaipv4:%d to host %RTnaipv4:%d\n",
205	so->so_type == IPPROTO_UDP ? "udp" : "tcp",
206	so->so_laddr.s_addr, ntohs(so->so_lport), addr, ntohs(port)));
207	return 0;
208	}
209
210	Log2(("NAT: sobind: %s guest %RTnaipv4:%d to host %RTnaipv4:%d error %d%s\n",
211	so->so_type == IPPROTO_UDP ? "udp" : "tcp",
212	so->so_laddr.s_addr, ntohs(so->so_lport),
213	addr, ntohs(port),
214	errno, port ? " (will retry with random port)" : ""));
215
216	if (port) /* retry without */
217	status = sobindto(so, addr, 0);
218
219	if (addr)
220	return status;
221	else
222	return 0;
223	}
224
225
226	/*
227	* Bind the socket to specific host address and/or port if necessary.
228	* We also always bind udp sockets to force the local port to be
229	* allocated and known in advance.
230	*/
231	int
232	sobind(PNATState pData, struct socket *so)
233	{
234	uint32_t addr = pData->bindIP.s_addr; /* may be INADDR_ANY */
235	bool fSamePorts = !!(pData->i32AliasMode & PKT_ALIAS_SAME_PORTS);
236	uint16_t port;
237	int status;
238
239	if (fSamePorts)
240	{
241	int opt = 1;
242	setsockopt(so->s, SOL_SOCKET, SO_REUSEADDR, (char *)&opt, sizeof(opt));
243	port = so->so_lport;
244	}
245	else
246	{
247	port = 0;
248	}
249
250	status = sobindto(so, addr, port);
251	return status;
252	}
253
254
255	/*
256	* Read from so's socket into sb_snd, updating all relevant sbuf fields
257	* NOTE: This will only be called if it is select()ed for reading, so
258	* a read() of 0 (or less) means it's disconnected
259	*/
260	int
261	soread(PNATState pData, struct socket *so)
262	{
263	int n, nn, lss, total;
264	struct sbuf *sb = &so->so_snd;
265	u_int len = sb->sb_datalen - sb->sb_cc;
266	struct iovec iov[2];
267	int mss = so->so_tcpcb->t_maxseg;
268	int sockerr;
269
270	STAM_PROFILE_START(&pData->StatIOread, a);
271	STAM_COUNTER_RESET(&pData->StatIORead_in_1);
272	STAM_COUNTER_RESET(&pData->StatIORead_in_2);
273
274	QSOCKET_LOCK(tcb);
275	SOCKET_LOCK(so);
276	QSOCKET_UNLOCK(tcb);
277
278	LogFlow(("soread: so = %R[natsock]\n", so));
279	Log2(("%s: so = %R[natsock] so->so_snd = %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, so, sb));
280
281	/*
282	* No need to check if there's enough room to read.
283	* soread wouldn't have been called if there weren't
284	*/
285
286	len = sb->sb_datalen - sb->sb_cc;
287
288	iov[0].iov_base = sb->sb_wptr;
289	iov[1].iov_base = 0;
290	iov[1].iov_len = 0;
291	if (sb->sb_wptr < sb->sb_rptr)
292	{
293	iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
294	/* Should never succeed, but... */
295	if (iov[0].iov_len > len)
296	iov[0].iov_len = len;
297	if (iov[0].iov_len > mss)
298	iov[0].iov_len -= iov[0].iov_len%mss;
299	n = 1;
300	}
301	else
302	{
303	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
304	/* Should never succeed, but... */
305	if (iov[0].iov_len > len)
306	iov[0].iov_len = len;
307	len -= iov[0].iov_len;
308	if (len)
309	{
310	iov[1].iov_base = sb->sb_data;
311	iov[1].iov_len = sb->sb_rptr - sb->sb_data;
312	if (iov[1].iov_len > len)
313	iov[1].iov_len = len;
314	total = iov[0].iov_len + iov[1].iov_len;
315	if (total > mss)
316	{
317	lss = total % mss;
318	if (iov[1].iov_len > lss)
319	{
320	iov[1].iov_len -= lss;
321	n = 2;
322	}
323	else
324	{
325	lss -= iov[1].iov_len;
326	iov[0].iov_len -= lss;
327	n = 1;
328	}
329	}
330	else
331	n = 2;
332	}
333	else
334	{
335	if (iov[0].iov_len > mss)
336	iov[0].iov_len -= iov[0].iov_len%mss;
337	n = 1;
338	}
339	}
340
341	#ifdef HAVE_READV
342	nn = readv(so->s, (struct iovec *)iov, n);
343	#else
344	nn = recv(so->s, iov[0].iov_base, iov[0].iov_len, (so->so_tcpcb->t_force? MSG_OOB:0));
345	#endif
346	if (nn < 0)
347	sockerr = errno; /* save it, as it may be clobbered by logging */
348	else
349	sockerr = 0;
350
351	Log2(("%s: read(1) nn = %d bytes\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn));
352	Log2(("%s: so = %R[natsock] so->so_snd = %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, so, sb));
353	if (nn <= 0)
354	{
355	#ifdef RT_OS_WINDOWS
356	/*
357	* Windows reports ESHUTDOWN after SHUT_RD (SD_RECEIVE)
358	* instead of just returning EOF indication.
359	*/
360	if (nn < 0 && sockerr == ESHUTDOWN)
361	{
362	nn = 0;
363	sockerr = 0;
364	}
365	#endif
366
367	if (nn == 0) /* XXX: should this be inside #if defined(RT_OS_WINDOWS)? */
368	{
369	/*
370	* Special case for WSAEnumNetworkEvents: If we receive 0 bytes that
371	* _could_ mean that the connection is closed. But we will receive an
372	* FD_CLOSE event later if the connection was _really_ closed. With
373	* www.youtube.com I see this very often. Closing the socket too early
374	* would be dangerous.
375	*/
376	int status;
377	unsigned long pending = 0;
378	status = ioctlsocket(so->s, FIONREAD, &pending);
379	if (status < 0)
380	Log(("NAT:%s: error in WSAIoctl: %d\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, errno));
381	if (pending != 0)
382	{
383	SOCKET_UNLOCK(so);
384	STAM_PROFILE_STOP(&pData->StatIOread, a);
385	return 0;
386	}
387	}
388
389	if ( nn < 0
390	&& soIgnorableErrorCode(sockerr))
391	{
392	SOCKET_UNLOCK(so);
393	STAM_PROFILE_STOP(&pData->StatIOread, a);
394	return 0;
395	}
396	else
397	{
398	int fUninitializedTemplate = 0;
399	int shuterr;
400
401	fUninitializedTemplate = RT_BOOL(( sototcpcb(so)
402	&& ( sototcpcb(so)->t_template.ti_src.s_addr == INADDR_ANY
403	\|\| sototcpcb(so)->t_template.ti_dst.s_addr == INADDR_ANY)));
404	/* nn == 0 means peer has performed an orderly shutdown */
405	Log2(("%s: disconnected, nn = %d, errno = %d (%s)\n",
406	RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn, sockerr, strerror(sockerr)));
407
408	shuterr = sofcantrcvmore(so);
409	if (!sockerr && !shuterr && !fUninitializedTemplate)
410	tcp_sockclosed(pData, sototcpcb(so));
411	else
412	{
413	LogRel2(("NAT: sockerr %d, shuterr %d - %R[natsock]\n", sockerr, shuterr, so));
414	tcp_drop(pData, sototcpcb(so), sockerr);
415	}
416	SOCKET_UNLOCK(so);
417	STAM_PROFILE_STOP(&pData->StatIOread, a);
418	return -1;
419	}
420	}
421	STAM_STATS(
422	if (n == 1)
423	{
424	STAM_COUNTER_INC(&pData->StatIORead_in_1);
425	STAM_COUNTER_ADD(&pData->StatIORead_in_1_bytes, nn);
426	}
427	else
428	{
429	STAM_COUNTER_INC(&pData->StatIORead_in_2);
430	STAM_COUNTER_ADD(&pData->StatIORead_in_2_1st_bytes, nn);
431	}
432	);
433
434	#ifndef HAVE_READV
435	/*
436	* If there was no error, try and read the second time round
437	* We read again if n = 2 (ie, there's another part of the buffer)
438	* and we read as much as we could in the first read
439	* We don't test for <= 0 this time, because there legitimately
440	* might not be any more data (since the socket is non-blocking),
441	* a close will be detected on next iteration.
442	* A return of -1 wont (shouldn't) happen, since it didn't happen above
443	*/
444	if (n == 2 && (unsigned)nn == iov[0].iov_len)
445	{
446	int ret;
447	ret = recv(so->s, iov[1].iov_base, iov[1].iov_len, 0);
448	if (ret > 0)
449	nn += ret;
450	STAM_STATS(
451	if (ret > 0)
452	{
453	STAM_COUNTER_INC(&pData->StatIORead_in_2);
454	STAM_COUNTER_ADD(&pData->StatIORead_in_2_2nd_bytes, ret);
455	}
456	);
457	}
458
459	Log2(("%s: read(2) nn = %d bytes\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn));
460	#endif
461
462	/* Update fields */
463	sb->sb_cc += nn;
464	sb->sb_wptr += nn;
465	Log2(("%s: update so_snd (readed nn = %d) %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn, sb));
466	if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
467	{
468	sb->sb_wptr -= sb->sb_datalen;
469	Log2(("%s: alter sb_wptr so_snd = %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, sb));
470	}
471	STAM_PROFILE_STOP(&pData->StatIOread, a);
472	SOCKET_UNLOCK(so);
473	return nn;
474	}
475
476	/*
477	* Get urgent data
478	*
479	* When the socket is created, we set it SO_OOBINLINE,
480	* so when OOB data arrives, we soread() it and everything
481	* in the send buffer is sent as urgent data
482	*/
483	void
484	sorecvoob(PNATState pData, struct socket *so)
485	{
486	struct tcpcb *tp = sototcpcb(so);
487	ssize_t ret;
488
489	LogFlowFunc(("sorecvoob: so = %R[natsock]\n", so));
490
491	/*
492	* We take a guess at how much urgent data has arrived.
493	* In most situations, when urgent data arrives, the next
494	* read() should get all the urgent data. This guess will
495	* be wrong however if more data arrives just after the
496	* urgent data, or the read() doesn't return all the
497	* urgent data.
498	*/
499	ret = soread(pData, so);
500	if (RT_LIKELY(ret > 0))
501	{
502	tp->snd_up = tp->snd_una + SBUF_LEN(&so->so_snd);
503	tp->t_force = 1;
504	tcp_output(pData, tp);
505	tp->t_force = 0;
506	}
507	}
508
509	/*
510	* Send urgent data
511	* There's a lot duplicated code here, but...
512	*/
513	int
514	sosendoob(struct socket *so)
515	{
516	struct sbuf *sb = &so->so_rcv;
517	char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
518
519	int n, len;
520
521	LogFlowFunc(("sosendoob so = %R[natsock]\n", so));
522
523	if (so->so_urgc > sizeof(buff))
524	so->so_urgc = sizeof(buff); /* XXX */
525
526	if (sb->sb_rptr < sb->sb_wptr)
527	{
528	/* We can send it directly */
529	n = send(so->s, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
530	so->so_urgc -= n;
531
532	Log2((" --- sent %d bytes urgent data, %d urgent bytes left\n",
533	n, so->so_urgc));
534	}
535	else
536	{
537	/*
538	* Since there's no sendv or sendtov like writev,
539	* we must copy all data to a linear buffer then
540	* send it all
541	*/
542	len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
543	if (len > so->so_urgc)
544	len = so->so_urgc;
545	memcpy(buff, sb->sb_rptr, len);
546	so->so_urgc -= len;
547	if (so->so_urgc)
548	{
549	n = sb->sb_wptr - sb->sb_data;
550	if (n > so->so_urgc)
551	n = so->so_urgc;
552	memcpy(buff + len, sb->sb_data, n);
553	so->so_urgc -= n;
554	len += n;
555	}
556	n = send(so->s, buff, len, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
557	#ifdef DEBUG
558	if (n != len)
559	Log(("Didn't send all data urgently XXXXX\n"));
560	#endif
561	Log2((" ---2 sent %d bytes urgent data, %d urgent bytes left\n",
562	n, so->so_urgc));
563	}
564
565	sb->sb_cc -= n;
566	sb->sb_rptr += n;
567	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
568	sb->sb_rptr -= sb->sb_datalen;
569
570	return n;
571	}
572
573	/*
574	* Write data from so_rcv to so's socket,
575	* updating all sbuf field as necessary
576	*/
577	int
578	sowrite(PNATState pData, struct socket *so)
579	{
580	int n, nn;
581	struct sbuf *sb = &so->so_rcv;
582	u_int len = sb->sb_cc;
583	struct iovec iov[2];
584
585	STAM_PROFILE_START(&pData->StatIOwrite, a);
586	STAM_COUNTER_RESET(&pData->StatIOWrite_in_1);
587	STAM_COUNTER_RESET(&pData->StatIOWrite_in_1_bytes);
588	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2);
589	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2_1st_bytes);
590	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2_2nd_bytes);
591	STAM_COUNTER_RESET(&pData->StatIOWrite_no_w);
592	STAM_COUNTER_RESET(&pData->StatIOWrite_rest);
593	STAM_COUNTER_RESET(&pData->StatIOWrite_rest_bytes);
594	LogFlowFunc(("so = %R[natsock]\n", so));
595	Log2(("%s: so = %R[natsock] so->so_rcv = %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, so, sb));
596	QSOCKET_LOCK(tcb);
597	SOCKET_LOCK(so);
598	QSOCKET_UNLOCK(tcb);
599	if (so->so_urgc)
600	{
601	sosendoob(so);
602	if (sb->sb_cc == 0)
603	{
604	SOCKET_UNLOCK(so);
605	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
606	return 0;
607	}
608	}
609
610	/*
611	* No need to check if there's something to write,
612	* sowrite wouldn't have been called otherwise
613	*/
614
615	len = sb->sb_cc;
616
617	iov[0].iov_base = sb->sb_rptr;
618	iov[1].iov_base = 0;
619	iov[1].iov_len = 0;
620	if (sb->sb_rptr < sb->sb_wptr)
621	{
622	iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
623	/* Should never succeed, but... */
624	if (iov[0].iov_len > len)
625	iov[0].iov_len = len;
626	n = 1;
627	}
628	else
629	{
630	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
631	if (iov[0].iov_len > len)
632	iov[0].iov_len = len;
633	len -= iov[0].iov_len;
634	if (len)
635	{
636	iov[1].iov_base = sb->sb_data;
637	iov[1].iov_len = sb->sb_wptr - sb->sb_data;
638	if (iov[1].iov_len > len)
639	iov[1].iov_len = len;
640	n = 2;
641	}
642	else
643	n = 1;
644	}
645	STAM_STATS({
646	if (n == 1)
647	{
648	STAM_COUNTER_INC(&pData->StatIOWrite_in_1);
649	STAM_COUNTER_ADD(&pData->StatIOWrite_in_1_bytes, iov[0].iov_len);
650	}
651	else
652	{
653	STAM_COUNTER_INC(&pData->StatIOWrite_in_2);
654	STAM_COUNTER_ADD(&pData->StatIOWrite_in_2_1st_bytes, iov[0].iov_len);
655	STAM_COUNTER_ADD(&pData->StatIOWrite_in_2_2nd_bytes, iov[1].iov_len);
656	}
657	});
658	/* Check if there's urgent data to send, and if so, send it */
659	#ifdef HAVE_READV
660	nn = writev(so->s, (const struct iovec *)iov, n);
661	#else
662	nn = send(so->s, iov[0].iov_base, iov[0].iov_len, 0);
663	#endif
664	Log2(("%s: wrote(1) nn = %d bytes\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn));
665	/* This should never happen, but people tell me it does shrug */
666	if ( nn < 0
667	&& soIgnorableErrorCode(errno))
668	{
669	SOCKET_UNLOCK(so);
670	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
671	return 0;
672	}
673
674	if (nn < 0 \|\| (nn == 0 && iov[0].iov_len > 0))
675	{
676	Log2(("%s: disconnected, so->so_state = %x, errno = %d\n",
677	RT_GCC_EXTENSION __PRETTY_FUNCTION__, so->so_state, errno));
678	sofcantsendmore(so);
679	tcp_sockclosed(pData, sototcpcb(so));
680	SOCKET_UNLOCK(so);
681	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
682	return -1;
683	}
684
685	#ifndef HAVE_READV
686	if (n == 2 && (unsigned)nn == iov[0].iov_len)
687	{
688	int ret;
689	ret = send(so->s, iov[1].iov_base, iov[1].iov_len, 0);
690	if (ret > 0)
691	nn += ret;
692	# ifdef VBOX_WITH_STATISTICS
693	if (ret > 0 && ret != (ssize_t)iov[1].iov_len)
694	{
695	STAM_COUNTER_INC(&pData->StatIOWrite_rest);
696	STAM_COUNTER_ADD(&pData->StatIOWrite_rest_bytes, (iov[1].iov_len - ret));
697	}
698	#endif
699	}
700	Log2(("%s: wrote(2) nn = %d bytes\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn));
701	#endif
702
703	/* Update sbuf */
704	sb->sb_cc -= nn;
705	sb->sb_rptr += nn;
706	Log2(("%s: update so_rcv (written nn = %d) %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, nn, sb));
707	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
708	{
709	sb->sb_rptr -= sb->sb_datalen;
710	Log2(("%s: alter sb_rptr of so_rcv %R[sbuf]\n", RT_GCC_EXTENSION __PRETTY_FUNCTION__, sb));
711	}
712
713	/*
714	* If in DRAIN mode, and there's no more data, set
715	* it CANTSENDMORE
716	*/
717	if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
718	sofcantsendmore(so);
719
720	SOCKET_UNLOCK(so);
721	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
722	return nn;
723	}
724
725	/*
726	* recvfrom() a UDP socket
727	*/
728	void
729	sorecvfrom(PNATState pData, struct socket *so)
730	{
731	LogFlowFunc(("sorecvfrom: so = %p\n", so));
732
733	#ifdef RT_OS_WINDOWS
734	/* ping is handled with ICMP API in ip_icmpwin.c */
735	Assert(so->so_type == IPPROTO_UDP);
736	#else
737	if (so->so_type == IPPROTO_ICMP)
738	{
739	/* This is a "ping" reply */
740	sorecvfrom_icmp_unix(pData, so);
741	udp_detach(pData, so);
742	}
743	else
744	#endif /* !RT_OS_WINDOWS */
745	{
746	static char achBuf[64 * 1024];
747
748	/* A "normal" UDP packet */
749	struct sockaddr_in addr;
750	socklen_t addrlen = sizeof(struct sockaddr_in);
751	struct iovec iov[2];
752	ssize_t nread;
753	struct mbuf *m;
754
755	QSOCKET_LOCK(udb);
756	SOCKET_LOCK(so);
757	QSOCKET_UNLOCK(udb);
758
759	m = m_getjcl(pData, M_NOWAIT, MT_HEADER, M_PKTHDR, slirp_size(pData));
760	if (m == NULL)
761	{
762	SOCKET_UNLOCK(so);
763	return;
764	}
765
766	m->m_data += ETH_HLEN;
767	m->m_pkthdr.header = mtod(m, void *);
768
769	m->m_data += sizeof(struct udpiphdr);
770
771	/* small packets will fit without copying */
772	iov[0].iov_base = mtod(m, char *);
773	iov[0].iov_len = M_TRAILINGSPACE(m);
774
775	/* large packets will spill into a temp buffer */
776	iov[1].iov_base = achBuf;
777	iov[1].iov_len = sizeof(achBuf);
778
779	#if !defined(RT_OS_WINDOWS)
780	{
781	struct msghdr mh;
782	memset(&mh, 0, sizeof(mh));
783
784	mh.msg_iov = iov;
785	mh.msg_iovlen = 2;
786	mh.msg_name = &addr;
787	mh.msg_namelen = addrlen;
788
789	nread = recvmsg(so->s, &mh, 0);
790	}
791	#else /* RT_OS_WINDOWS */
792	{
793	DWORD nbytes; /* NB: can't use nread b/c of different size */
794	DWORD flags = 0;
795	int status;
796	AssertCompile(sizeof(WSABUF) == sizeof(struct iovec));
797	AssertCompileMembersSameSizeAndOffset(WSABUF, len, struct iovec, iov_len);
798	AssertCompileMembersSameSizeAndOffset(WSABUF, buf, struct iovec, iov_base);
799	status = WSARecvFrom(so->s, (WSABUF *)&iov[0], 2, &nbytes, &flags,
800	(struct sockaddr *)&addr, &addrlen,
801	NULL, NULL);
802	if (status != SOCKET_ERROR)
803	nread = nbytes;
804	else
805	nread = -1;
806	}
807	#endif
808	if (nread >= 0)
809	{
810	if (nread <= iov[0].iov_len)
811	m->m_len = nread;
812	else
813	{
814	m->m_len = iov[0].iov_len;
815	m_append(pData, m, nread - iov[0].iov_len, iov[1].iov_base);
816	}
817	Assert(m_length(m, NULL) == (size_t)nread);
818
819	/*
820	* Hack: domain name lookup will be used the most for UDP,
821	* and since they'll only be used once there's no need
822	* for the 4 minute (or whatever) timeout... So we time them
823	* out much quicker (10 seconds for now...)
824	*/
825	if (so->so_expire)
826	{
827	if (so->so_fport != RT_H2N_U16_C(53))
828	so->so_expire = curtime + SO_EXPIRE;
829	}
830
831	/*
832	* DNS proxy requests are forwarded to the real resolver,
833	* but its socket's so_faddr is that of the DNS proxy
834	* itself.
835	*
836	* last argument should be changed if Slirp will inject IP attributes
837	*/
838	if ( pData->fUseDnsProxy
839	&& so->so_fport == RT_H2N_U16_C(53)
840	&& CTL_CHECK(so->so_faddr.s_addr, CTL_DNS))
841	dnsproxy_answer(pData, so, m);
842
843	/* packets definetly will be fragmented, could confuse receiver peer. */
844	if (nread > if_mtu)
845	m->m_flags \|= M_SKIP_FIREWALL;
846
847	/*
848	* If this packet was destined for CTL_ADDR,
849	* make it look like that's where it came from, done by udp_output
850	*/
851	udp_output(pData, so, m, &addr);
852	}
853	else
854	{
855	m_freem(pData, m);
856
857	if (!soIgnorableErrorCode(errno))
858	{
859	u_char code;
860	if (errno == EHOSTUNREACH)
861	code = ICMP_UNREACH_HOST;
862	else if (errno == ENETUNREACH)
863	code = ICMP_UNREACH_NET;
864	else
865	code = ICMP_UNREACH_PORT;
866
867	Log2((" rx error, tx icmp ICMP_UNREACH:%i\n", code));
868	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
869	so->so_m = NULL;
870	}
871	}
872
873	SOCKET_UNLOCK(so);
874	}
875	}
876
877	/*
878	* sendto() a socket
879	*/
880	int
881	sosendto(PNATState pData, struct socket so, struct mbuf m)
882	{
883	int ret;
884	struct sockaddr_in *paddr;
885	struct sockaddr addr;
886	#if 0
887	struct sockaddr_in host_addr;
888	#endif
889	caddr_t buf = 0;
890	int mlen;
891
892	LogFlowFunc(("sosendto: so = %R[natsock], m = %p\n", so, m));
893
894	memset(&addr, 0, sizeof(struct sockaddr));
895	#ifdef RT_OS_DARWIN
896	addr.sa_len = sizeof(struct sockaddr_in);
897	#endif
898	paddr = (struct sockaddr_in *)&addr;
899	paddr->sin_family = AF_INET;
900	if ((so->so_faddr.s_addr & RT_H2N_U32(pData->netmask)) == pData->special_addr.s_addr)
901	{
902	/* It's an alias */
903	uint32_t last_byte = RT_N2H_U32(so->so_faddr.s_addr) & ~pData->netmask;
904	switch(last_byte)
905	{
906	#if 0
907	/* handle this case at 'default:' */
908	case CTL_BROADCAST:
909	addr.sin_addr.s_addr = INADDR_BROADCAST;
910	/* Send the packet to host to fully emulate broadcast */
911	/** @todo r=klaus: on Linux host this causes the host to receive
912	* the packet twice for some reason. And I cannot find any place
913	* in the man pages which states that sending a broadcast does not
914	* reach the host itself. */
915	host_addr.sin_family = AF_INET;
916	host_addr.sin_port = so->so_fport;
917	host_addr.sin_addr = our_addr;
918	sendto(so->s, m->m_data, m->m_len, 0,
919	(struct sockaddr *)&host_addr, sizeof (struct sockaddr));
920	break;
921	#endif
922	case CTL_DNS:
923	case CTL_ALIAS:
924	default:
925	if (last_byte == ~pData->netmask)
926	paddr->sin_addr.s_addr = INADDR_BROADCAST;
927	else
928	paddr->sin_addr = loopback_addr;
929	break;
930	}
931	}
932	else
933	paddr->sin_addr = so->so_faddr;
934	paddr->sin_port = so->so_fport;
935
936	Log2((" sendto()ing, addr.sin_port=%d, addr.sin_addr.s_addr=%.16s\n",
937	RT_N2H_U16(paddr->sin_port), inet_ntoa(paddr->sin_addr)));
938
939	/* Don't care what port we get */
940	/*
941	* > nmap -sV -T4 -O -A -v -PU3483 255.255.255.255
942	* generates bodyless messages, annoying memmory management system.
943	*/
944	mlen = m_length(m, NULL);
945	if (mlen > 0)
946	{
947	buf = RTMemAlloc(mlen);
948	if (buf == NULL)
949	{
950	return -1;
951	}
952	m_copydata(m, 0, mlen, buf);
953	}
954	ret = sendto(so->s, buf, mlen, 0,
955	(struct sockaddr *)&addr, sizeof (struct sockaddr));
956	#ifdef VBOX_WITH_NAT_SEND2HOME
957	if (slirpIsWideCasting(pData, so->so_faddr.s_addr))
958	{
959	slirpSend2Home(pData, so, buf, mlen, 0);
960	}
961	#endif
962	if (buf)
963	RTMemFree(buf);
964	if (ret < 0)
965	{
966	Log2(("UDP: sendto fails (%s)\n", strerror(errno)));
967	return -1;
968	}
969
970	/*
971	* Kill the socket if there's no reply in 4 minutes,
972	* but only if it's an expirable socket
973	*/
974	if (so->so_expire)
975	so->so_expire = curtime + SO_EXPIRE;
976	so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
977	return 0;
978	}
979
980	/*
981	* XXX This should really be tcp_listen
982	*/
983	struct socket *
984	solisten(PNATState pData, u_int32_t bind_addr, u_int port, u_int32_t laddr, u_int lport, int flags)
985	{
986	struct sockaddr_in addr;
987	struct socket *so;
988	socklen_t addrlen = sizeof(addr);
989	int s, opt = 1;
990	int status;
991
992	LogFlowFunc(("solisten: port = %d, laddr = %x, lport = %d, flags = %x\n", port, laddr, lport, flags));
993
994	if ((so = socreate()) == NULL)
995	{
996	/* RTMemFree(so); Not sofree() ??? free(NULL) == NOP */
997	return NULL;
998	}
999
1000	/* Don't tcp_attach... we don't need so_snd nor so_rcv */
1001	if ((so->so_tcpcb = tcp_newtcpcb(pData, so)) == NULL)
1002	{
1003	RTMemFree(so);
1004	return NULL;
1005	}
1006
1007	SOCKET_LOCK_CREATE(so);
1008	SOCKET_LOCK(so);
1009	QSOCKET_LOCK(tcb);
1010	insque(pData, so,&tcb);
1011	NSOCK_INC();
1012	QSOCKET_UNLOCK(tcb);
1013
1014	/*
1015	* SS_FACCEPTONCE sockets must time out.
1016	*/
1017	if (flags & SS_FACCEPTONCE)
1018	so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
1019
1020	so->so_state = (SS_FACCEPTCONN\|flags);
1021	so->so_lport = lport; /* Kept in network format */
1022	so->so_laddr.s_addr = laddr; /* Ditto */
1023
1024	memset(&addr, 0, sizeof(addr));
1025	#ifdef RT_OS_DARWIN
1026	addr.sin_len = sizeof(addr);
1027	#endif
1028	addr.sin_family = AF_INET;
1029	addr.sin_addr.s_addr = bind_addr;
1030	addr.sin_port = port;
1031
1032	/**
1033	* changing listen(,1->SOMAXCONN) shouldn't be harmful for NAT's TCP/IP stack,
1034	* kernel will choose the optimal value for requests queue length.
1035	* @note: MSDN recommends low (2-4) values for bluetooth networking devices.
1036	*/
1037	if ( ((s = socket(AF_INET, SOCK_STREAM, 0)) < 0)
1038	\|\| (setsockopt(s, SOL_SOCKET, SO_REUSEADDR,(char *)&opt, sizeof(int)) < 0)
1039	\|\| (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0)
1040	\|\| (listen(s, pData->soMaxConn) < 0))
1041	{
1042	#ifdef RT_OS_WINDOWS
1043	int tmperrno = WSAGetLastError(); /* Don't clobber the real reason we failed */
1044	closesocket(s);
1045	QSOCKET_LOCK(tcb);
1046	sofree(pData, so);
1047	QSOCKET_UNLOCK(tcb);
1048	/* Restore the real errno */
1049	WSASetLastError(tmperrno);
1050	#else
1051	int tmperrno = errno; /* Don't clobber the real reason we failed */
1052	close(s);
1053	if (sototcpcb(so))
1054	tcp_close(pData, sototcpcb(so));
1055	else
1056	sofree(pData, so);
1057	/* Restore the real errno */
1058	errno = tmperrno;
1059	#endif
1060	return NULL;
1061	}
1062	fd_nonblock(s);
1063	setsockopt(s, SOL_SOCKET, SO_OOBINLINE,(char *)&opt, sizeof(int));
1064
1065	getsockname(s,(struct sockaddr *)&addr,&addrlen);
1066	so->so_fport = addr.sin_port;
1067	/* set socket buffers */
1068	opt = pData->socket_rcv;
1069	status = setsockopt(s, SOL_SOCKET, SO_RCVBUF, (char *)&opt, sizeof(int));
1070	if (status < 0)
1071	{
1072	LogRel(("NAT: Error(%d) while setting RCV capacity to (%d)\n", errno, opt));
1073	goto no_sockopt;
1074	}
1075	opt = pData->socket_snd;
1076	status = setsockopt(s, SOL_SOCKET, SO_SNDBUF, (char *)&opt, sizeof(int));
1077	if (status < 0)
1078	{
1079	LogRel(("NAT: Error(%d) while setting SND capacity to (%d)\n", errno, opt));
1080	goto no_sockopt;
1081	}
1082	no_sockopt:
1083	if (addr.sin_addr.s_addr == 0 \|\| addr.sin_addr.s_addr == loopback_addr.s_addr)
1084	so->so_faddr = alias_addr;
1085	else
1086	so->so_faddr = addr.sin_addr;
1087
1088	so->s = s;
1089	SOCKET_UNLOCK(so);
1090	return so;
1091	}
1092
1093	/*
1094	* Data is available in so_rcv
1095	* Just write() the data to the socket
1096	* XXX not yet...
1097	* @todo do we really need this function, what it's intended to do?
1098	*/
1099	void
1100	sorwakeup(struct socket *so)
1101	{
1102	NOREF(so);
1103	#if 0
1104	sowrite(so);
1105	FD_CLR(so->s,&writefds);
1106	#endif
1107	}
1108
1109	/*
1110	* Data has been freed in so_snd
1111	* We have room for a read() if we want to
1112	* For now, don't read, it'll be done in the main loop
1113	*/
1114	void
1115	sowwakeup(struct socket *so)
1116	{
1117	NOREF(so);
1118	}
1119
1120	/*
1121	* Various session state calls
1122	* XXX Should be #define's
1123	* The socket state stuff needs work, these often get call 2 or 3
1124	* times each when only 1 was needed
1125	*/
1126	void
1127	soisfconnecting(struct socket *so)
1128	{
1129	so->so_state &= ~(SS_NOFDREF\|SS_ISFCONNECTED\|SS_FCANTRCVMORE\|
1130	SS_FCANTSENDMORE\|SS_FWDRAIN);
1131	so->so_state \|= SS_ISFCONNECTING; /* Clobber other states */
1132	}
1133
1134	void
1135	soisfconnected(struct socket *so)
1136	{
1137	LogFlowFunc(("ENTER: so:%R[natsock]\n", so));
1138	so->so_state &= ~(SS_ISFCONNECTING\|SS_FWDRAIN\|SS_NOFDREF);
1139	so->so_state \|= SS_ISFCONNECTED; /* Clobber other states */
1140	LogFlowFunc(("LEAVE: so:%R[natsock]\n", so));
1141	}
1142
1143	int
1144	sofcantrcvmore(struct socket *so)
1145	{
1146	int err = 0;
1147
1148	LogFlowFunc(("ENTER: so:%R[natsock]\n", so));
1149	if ((so->so_state & SS_NOFDREF) == 0)
1150	{
1151	/*
1152	* If remote closes first and then sends an RST, the recv() in
1153	* soread() will keep reporting EOF without any error
1154	* indication. As far as I can tell the only way to detect
1155	* this on Linux is to check if shutdown() succeeds here (but
1156	* see below).
1157	*
1158	* OTOH on OS X shutdown() "helpfully" checks if remote has
1159	* already closed and then always returns ENOTCONN
1160	* immediately.
1161	*/
1162	int status = shutdown(so->s, SHUT_RD);
1163	#if defined(RT_OS_LINUX)
1164	if (status < 0)
1165	err = errno;
1166	#else
1167	RT_NOREF(status);
1168	#endif
1169	}
1170	so->so_state &= ~(SS_ISFCONNECTING);
1171	if (so->so_state & SS_FCANTSENDMORE)
1172	{
1173	#if defined(RT_OS_LINUX)
1174	/*
1175	* If we have closed first, and remote closes, shutdown will
1176	* return ENOTCONN, but this is expected. Don't tell the
1177	* caller there was an error.
1178	*/
1179	if (err == ENOTCONN)
1180	err = 0;
1181	#endif
1182	so->so_state = SS_NOFDREF; /* Don't select it */
1183	/* XXX close() here as well? */
1184	}
1185	else
1186	so->so_state \|= SS_FCANTRCVMORE;
1187
1188	LogFlowFunc(("LEAVE: %d\n", err));
1189	return err;
1190	}
1191
1192	void
1193	sofcantsendmore(struct socket *so)
1194	{
1195	LogFlowFunc(("ENTER: so:%R[natsock]\n", so));
1196	if ((so->so_state & SS_NOFDREF) == 0)
1197	shutdown(so->s, 1); /* send FIN to fhost */
1198
1199	so->so_state &= ~(SS_ISFCONNECTING);
1200	if (so->so_state & SS_FCANTRCVMORE)
1201	so->so_state = SS_NOFDREF; /* as above */
1202	else
1203	so->so_state \|= SS_FCANTSENDMORE;
1204	LogFlowFuncLeave();
1205	}
1206
1207	void
1208	soisfdisconnected(struct socket *so)
1209	{
1210	NOREF(so);
1211	#if 0
1212	so->so_state &= ~(SS_ISFCONNECTING\|SS_ISFCONNECTED);
1213	close(so->s);
1214	so->so_state = SS_ISFDISCONNECTED;
1215	/*
1216	* XXX Do nothing ... ?
1217	*/
1218	#endif
1219	}
1220
1221	/*
1222	* Set write drain mode
1223	* Set CANTSENDMORE once all data has been write()n
1224	*/
1225	void
1226	sofwdrain(struct socket *so)
1227	{
1228	if (SBUF_LEN(&so->so_rcv))
1229	so->so_state \|= SS_FWDRAIN;
1230	else
1231	sofcantsendmore(so);
1232	}
1233
1234	#if !defined(RT_OS_WINDOWS)
1235	static void
1236	send_icmp_to_guest(PNATState pData, char buff, size_t len, const struct sockaddr_in addr)
1237	{
1238	struct ip *ip;
1239	uint32_t dst, src;
1240	char ip_copy[256];
1241	struct icmp *icp;
1242	int old_ip_len = 0;
1243	int hlen, original_hlen = 0;
1244	struct mbuf *m;
1245	struct icmp_msg *icm;
1246	uint8_t proto;
1247	int type = 0;
1248
1249	ip = (struct ip *)buff;
1250	/* Fix ip->ip_len to contain the total packet length including the header
1251	* in _host_ byte order for all OSes. On Darwin, that value already is in
1252	* host byte order. Solaris and Darwin report only the payload. */
1253	#ifndef RT_OS_DARWIN
1254	ip->ip_len = RT_N2H_U16(ip->ip_len);
1255	#endif
1256	hlen = (ip->ip_hl << 2);
1257	#if defined(RT_OS_SOLARIS) \|\| defined(RT_OS_DARWIN)
1258	ip->ip_len += hlen;
1259	#endif
1260	if (ip->ip_len < hlen + ICMP_MINLEN)
1261	{
1262	Log(("send_icmp_to_guest: ICMP header is too small to understand which type/subtype of the datagram\n"));
1263	return;
1264	}
1265	icp = (struct icmp )((char )ip + hlen);
1266
1267	Log(("ICMP:received msg(t:%d, c:%d)\n", icp->icmp_type, icp->icmp_code));
1268	if ( icp->icmp_type != ICMP_ECHOREPLY
1269	&& icp->icmp_type != ICMP_TIMXCEED
1270	&& icp->icmp_type != ICMP_UNREACH)
1271	{
1272	return;
1273	}
1274
1275	/*
1276	* ICMP_ECHOREPLY, ICMP_TIMXCEED, ICMP_UNREACH minimal header size is
1277	* ICMP_ECHOREPLY assuming data 0
1278	* icmp_{type(8), code(8), cksum(16),identifier(16),seqnum(16)}
1279	*/
1280	if (ip->ip_len < hlen + 8)
1281	{
1282	Log(("send_icmp_to_guest: NAT accept ICMP_{ECHOREPLY, TIMXCEED, UNREACH} the minimum size is 64 (see rfc792)\n"));
1283	return;
1284	}
1285
1286	type = icp->icmp_type;
1287	if ( type == ICMP_TIMXCEED
1288	\|\| type == ICMP_UNREACH)
1289	{
1290	/*
1291	* ICMP_TIMXCEED, ICMP_UNREACH minimal header size is
1292	* icmp_{type(8), code(8), cksum(16),unused(32)} + IP header + 64 bit of original datagram
1293	*/
1294	if (ip->ip_len < hlen + 2*8 + sizeof(struct ip))
1295	{
1296	Log(("send_icmp_to_guest: NAT accept ICMP_{TIMXCEED, UNREACH} the minimum size of ipheader + 64 bit of data (see rfc792)\n"));
1297	return;
1298	}
1299	ip = &icp->icmp_ip;
1300	}
1301
1302	icm = icmp_find_original_mbuf(pData, ip);
1303	if (icm == NULL)
1304	{
1305	Log(("NAT: Can't find the corresponding packet for the received ICMP\n"));
1306	return;
1307	}
1308
1309	m = icm->im_m;
1310	if (!m)
1311	{
1312	LogFunc(("%R[natsock] hasn't stored it's mbuf on sent\n", icm->im_so));
1313	goto done;
1314	}
1315
1316	src = addr->sin_addr.s_addr;
1317	if (type == ICMP_ECHOREPLY)
1318	{
1319	struct ip ip0 = mtod(m, struct ip );
1320	struct icmp icp0 = (struct icmp )((char *)ip0 + (ip0->ip_hl << 2));
1321	if (icp0->icmp_type != ICMP_ECHO)
1322	{
1323	Log(("NAT: we haven't found echo for this reply\n"));
1324	goto done;
1325	}
1326	/*
1327	* while combining buffer to send (see ip_icmp.c) we control ICMP header only,
1328	* IP header combined by OS network stack, our local copy of IP header contians values
1329	* in host byte order so no byte order conversion is required. IP headers fields are converting
1330	* in ip_output0 routine only.
1331	*/
1332	if ( (ip->ip_len - hlen)
1333	!= (ip0->ip_len - (ip0->ip_hl << 2)))
1334	{
1335	Log(("NAT: ECHO(%d) lenght doesn't match ECHOREPLY(%d)\n",
1336	(ip->ip_len - hlen), (ip0->ip_len - (ip0->ip_hl << 2))));
1337	goto done;
1338	}
1339	}
1340
1341	/* ip points on origianal ip header */
1342	ip = mtod(m, struct ip *);
1343	proto = ip->ip_p;
1344	/* Now ip is pointing on header we've sent from guest */
1345	if ( icp->icmp_type == ICMP_TIMXCEED
1346	\|\| icp->icmp_type == ICMP_UNREACH)
1347	{
1348	old_ip_len = (ip->ip_hl << 2) + 64;
1349	if (old_ip_len > sizeof(ip_copy))
1350	old_ip_len = sizeof(ip_copy);
1351	memcpy(ip_copy, ip, old_ip_len);
1352	}
1353
1354	/* source address from original IP packet*/
1355	dst = ip->ip_src.s_addr;
1356
1357	/* overide ther tail of old packet */
1358	ip = mtod(m, struct ip ); / ip is from mbuf we've overrided */
1359	original_hlen = ip->ip_hl << 2;
1360	/* saves original ip header and options */
1361	m_copyback(pData, m, original_hlen, len - hlen, buff + hlen);
1362	ip->ip_len = m_length(m, NULL);
1363	ip->ip_p = IPPROTO_ICMP; /* the original package could be whatever, but we're response via ICMP*/
1364
1365	icp = (struct icmp )((char )ip + (ip->ip_hl << 2));
1366	type = icp->icmp_type;
1367	if ( type == ICMP_TIMXCEED
1368	\|\| type == ICMP_UNREACH)
1369	{
1370	/* according RFC 793 error messages required copy of initial IP header + 64 bit */
1371	memcpy(&icp->icmp_ip, ip_copy, old_ip_len);
1372
1373	/* undo byte order conversions done in ip_input() */
1374	HTONS(icp->icmp_ip.ip_len);
1375	HTONS(icp->icmp_ip.ip_id);
1376	HTONS(icp->icmp_ip.ip_off);
1377
1378	ip->ip_tos = ((ip->ip_tos & 0x1E) \| 0xC0); /* high priority for errors */
1379	}
1380
1381	ip->ip_src.s_addr = src;
1382	ip->ip_dst.s_addr = dst;
1383	icmp_reflect(pData, m);
1384	/* m was freed */
1385	icm->im_m = NULL;
1386
1387	done:
1388	icmp_msg_delete(pData, icm);
1389	}
1390
1391	static void sorecvfrom_icmp_unix(PNATState pData, struct socket *so)
1392	{
1393	struct sockaddr_in addr;
1394	socklen_t addrlen = sizeof(struct sockaddr_in);
1395	struct ip ip;
1396	char *buff;
1397	int len = 0;
1398
1399	/* 1- step: read the ip header */
1400	len = recvfrom(so->s, &ip, sizeof(struct ip), MSG_PEEK,
1401	(struct sockaddr *)&addr, &addrlen);
1402	if ( len < 0
1403	&& ( soIgnorableErrorCode(errno)
1404	\|\| errno == ENOTCONN))
1405	{
1406	Log(("sorecvfrom_icmp_unix: 1 - step can't read IP datagramm (would block)\n"));
1407	return;
1408	}
1409
1410	if ( len < sizeof(struct ip)
1411	\|\| len < 0
1412	\|\| len == 0)
1413	{
1414	u_char code;
1415	code = ICMP_UNREACH_PORT;
1416
1417	if (errno == EHOSTUNREACH)
1418	code = ICMP_UNREACH_HOST;
1419	else if (errno == ENETUNREACH)
1420	code = ICMP_UNREACH_NET;
1421
1422	LogRel(("NAT: UDP ICMP rx errno=%d (%s)\n", errno, strerror(errno)));
1423	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
1424	so->so_m = NULL;
1425	Log(("sorecvfrom_icmp_unix: 1 - step can't read IP datagramm\n"));
1426	return;
1427	}
1428	/* basic check of IP header */
1429	if ( ip.ip_v != IPVERSION
1430	# ifndef RT_OS_DARWIN
1431	\|\| ip.ip_p != IPPROTO_ICMP
1432	# endif
1433	)
1434	{
1435	Log(("sorecvfrom_icmp_unix: 1 - step IP isn't IPv4\n"));
1436	return;
1437	}
1438	# ifndef RT_OS_DARWIN
1439	/* Darwin reports the IP length already in host byte order. */
1440	ip.ip_len = RT_N2H_U16(ip.ip_len);
1441	# endif
1442	# if defined(RT_OS_SOLARIS) \|\| defined(RT_OS_DARWIN)
1443	/* Solaris and Darwin report the payload only */
1444	ip.ip_len += (ip.ip_hl << 2);
1445	# endif
1446	/* Note: ip->ip_len in host byte order (all OS) */
1447	len = ip.ip_len;
1448	buff = RTMemAlloc(len);
1449	if (buff == NULL)
1450	{
1451	Log(("sorecvfrom_icmp_unix: 1 - step can't allocate enought room for datagram\n"));
1452	return;
1453	}
1454	/* 2 - step: we're reading rest of the datagramm to the buffer */
1455	addrlen = sizeof(struct sockaddr_in);
1456	memset(&addr, 0, addrlen);
1457	len = recvfrom(so->s, buff, len, 0,
1458	(struct sockaddr *)&addr, &addrlen);
1459	if ( len < 0
1460	&& ( soIgnorableErrorCode(errno)
1461	\|\| errno == ENOTCONN))
1462	{
1463	Log(("sorecvfrom_icmp_unix: 2 - step can't read IP body (would block expected:%d)\n",
1464	ip.ip_len));
1465	RTMemFree(buff);
1466	return;
1467	}
1468	if ( len < 0
1469	\|\| len == 0)
1470	{
1471	Log(("sorecvfrom_icmp_unix: 2 - step read of the rest of datagramm is fallen (errno:%d, len:%d expected: %d)\n",
1472	errno, len, (ip.ip_len - sizeof(struct ip))));
1473	RTMemFree(buff);
1474	return;
1475	}
1476	/* len is modified in 2nd read, when the rest of the datagramm was read */
1477	send_icmp_to_guest(pData, buff, len, &addr);
1478	RTMemFree(buff);
1479	}
1480	#endif /* !RT_OS_WINDOWS */

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/socket.c@ 74950

Download in other formats: