socket.c@ 26837

Last change on this file since 26837 was 26598, checked in by vboxsync, 15 years ago
NAT: read no more than available space in the mbuf.
Property svn:eol-style set to `native`
File size: 37.2 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
35	struct socket *
36	solookup(struct socket *head, struct in_addr laddr,
37	u_int lport, struct in_addr faddr, u_int fport)
38	{
39	struct socket *so;
40
41	for (so = head->so_next; so != head; so = so->so_next)
42	{
43	if ( so->so_lport == lport
44	&& so->so_laddr.s_addr == laddr.s_addr
45	&& so->so_faddr.s_addr == faddr.s_addr
46	&& so->so_fport == fport)
47	return so;
48	}
49
50	return (struct socket *)NULL;
51	}
52
53	/*
54	* Create a new socket, initialise the fields
55	* It is the responsibility of the caller to
56	* insque() it into the correct linked-list
57	*/
58	struct socket *
59	socreate()
60	{
61	struct socket *so;
62
63	so = (struct socket *)RTMemAllocZ(sizeof(struct socket));
64	if (so)
65	{
66	so->so_state = SS_NOFDREF;
67	so->s = -1;
68	#if !defined(RT_OS_WINDOWS)
69	so->so_poll_index = -1;
70	#endif
71	}
72	return so;
73	}
74
75	/*
76	* remque and free a socket, clobber cache
77	* VBOX_WITH_SLIRP_MT: before sofree queue should be locked, because
78	* in sofree we don't know from which queue item beeing removed.
79	*/
80	void
81	sofree(PNATState pData, struct socket *so)
82	{
83	struct socket *so_prev = NULL;
84	if (so == tcp_last_so)
85	tcp_last_so = &tcb;
86	else if (so == udp_last_so)
87	udp_last_so = &udb;
88
89	/* check if mbuf haven't been already freed */
90	if (so->so_m != NULL)
91	m_free(pData, so->so_m);
92	#ifndef VBOX_WITH_SLIRP_MT
93	if (so->so_next && so->so_prev)
94	{
95	remque(pData, so); /* crashes if so is not in a queue */
96	NSOCK_DEC();
97	}
98
99	RTMemFree(so);
100	#else
101	so->so_deleted = 1;
102	#endif
103	}
104
105	#ifdef VBOX_WITH_SLIRP_MT
106	void
107	soread_queue(PNATState pData, struct socket so, int ret)
108	{
109	*ret = soread(pData, so);
110	}
111	#endif
112
113	/*
114	* Read from so's socket into sb_snd, updating all relevant sbuf fields
115	* NOTE: This will only be called if it is select()ed for reading, so
116	* a read() of 0 (or less) means it's disconnected
117	*/
118	int
119	soread(PNATState pData, struct socket *so)
120	{
121	int n, nn, lss, total;
122	struct sbuf *sb = &so->so_snd;
123	size_t len = sb->sb_datalen - sb->sb_cc;
124	struct iovec iov[2];
125	int mss = so->so_tcpcb->t_maxseg;
126
127	STAM_PROFILE_START(&pData->StatIOread, a);
128	STAM_COUNTER_RESET(&pData->StatIORead_in_1);
129	STAM_COUNTER_RESET(&pData->StatIORead_in_2);
130
131	QSOCKET_LOCK(tcb);
132	SOCKET_LOCK(so);
133	QSOCKET_UNLOCK(tcb);
134
135	DEBUG_CALL("soread");
136	DEBUG_ARG("so = %lx", (long )so);
137
138	/*
139	* No need to check if there's enough room to read.
140	* soread wouldn't have been called if there weren't
141	*/
142
143	len = sb->sb_datalen - sb->sb_cc;
144
145	iov[0].iov_base = sb->sb_wptr;
146	iov[1].iov_base = 0;
147	iov[1].iov_len = 0;
148	if (sb->sb_wptr < sb->sb_rptr)
149	{
150	iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
151	/* Should never succeed, but... */
152	if (iov[0].iov_len > len)
153	iov[0].iov_len = len;
154	if (iov[0].iov_len > mss)
155	iov[0].iov_len -= iov[0].iov_len%mss;
156	n = 1;
157	}
158	else
159	{
160	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
161	/* Should never succeed, but... */
162	if (iov[0].iov_len > len)
163	iov[0].iov_len = len;
164	len -= iov[0].iov_len;
165	if (len)
166	{
167	iov[1].iov_base = sb->sb_data;
168	iov[1].iov_len = sb->sb_rptr - sb->sb_data;
169	if (iov[1].iov_len > len)
170	iov[1].iov_len = len;
171	total = iov[0].iov_len + iov[1].iov_len;
172	if (total > mss)
173	{
174	lss = total % mss;
175	if (iov[1].iov_len > lss)
176	{
177	iov[1].iov_len -= lss;
178	n = 2;
179	}
180	else
181	{
182	lss -= iov[1].iov_len;
183	iov[0].iov_len -= lss;
184	n = 1;
185	}
186	}
187	else
188	n = 2;
189	}
190	else
191	{
192	if (iov[0].iov_len > mss)
193	iov[0].iov_len -= iov[0].iov_len%mss;
194	n = 1;
195	}
196	}
197
198	#ifdef HAVE_READV
199	nn = readv(so->s, (struct iovec *)iov, n);
200	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
201	#else
202	nn = recv(so->s, iov[0].iov_base, iov[0].iov_len, 0);
203	#endif
204	if (nn <= 0)
205	{
206	#if defined(RT_OS_WINDOWS)
207	/*
208	* Special case for WSAEnumNetworkEvents: If we receive 0 bytes that
209	* _could_ mean that the connection is closed. But we will receive an
210	* FD_CLOSE event later if the connection was _really_ closed. With
211	* www.youtube.com I see this very often. Closing the socket too early
212	* would be dangerous.
213	*/
214	int status, ignored;
215	unsigned long pending = 0;
216	status = WSAIoctl(so->s, FIONREAD, NULL, 0, &pending, sizeof(unsigned long), &ignored, NULL, NULL);
217	if (status < 0)
218	LogRel(("NAT:error in WSAIoctl: %d\n", WSAGetLastError()));
219	if (nn == 0 && (pending != 0))
220	{
221	SOCKET_UNLOCK(so);
222	STAM_PROFILE_STOP(&pData->StatIOread, a);
223	return 0;
224	}
225	#endif
226	if ( nn < 0
227	&& ( errno == EINTR
228	\|\| errno == EAGAIN
229	\|\| errno == EWOULDBLOCK))
230	{
231	SOCKET_UNLOCK(so);
232	STAM_PROFILE_STOP(&pData->StatIOread, a);
233	return 0;
234	}
235	else
236	{
237	/* nn == 0 means peer has performed an orderly shutdown */
238	DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n",
239	nn, errno, strerror(errno)));
240	sofcantrcvmore(so);
241	tcp_sockclosed(pData, sototcpcb(so));
242	SOCKET_UNLOCK(so);
243	STAM_PROFILE_STOP(&pData->StatIOread, a);
244	return -1;
245	}
246	}
247	STAM_STATS(
248	if (n == 1)
249	{
250	STAM_COUNTER_INC(&pData->StatIORead_in_1);
251	STAM_COUNTER_ADD(&pData->StatIORead_in_1_bytes, nn);
252	}
253	else
254	{
255	STAM_COUNTER_INC(&pData->StatIORead_in_2);
256	STAM_COUNTER_ADD(&pData->StatIORead_in_2_1st_bytes, nn);
257	}
258	);
259
260	#ifndef HAVE_READV
261	/*
262	* If there was no error, try and read the second time round
263	* We read again if n = 2 (ie, there's another part of the buffer)
264	* and we read as much as we could in the first read
265	* We don't test for <= 0 this time, because there legitimately
266	* might not be any more data (since the socket is non-blocking),
267	* a close will be detected on next iteration.
268	* A return of -1 wont (shouldn't) happen, since it didn't happen above
269	*/
270	if (n == 2 && nn == iov[0].iov_len)
271	{
272	int ret;
273	ret = recv(so->s, iov[1].iov_base, iov[1].iov_len, 0);
274	if (ret > 0)
275	nn += ret;
276	STAM_STATS(
277	if (ret > 0)
278	{
279	STAM_COUNTER_INC(&pData->StatIORead_in_2);
280	STAM_COUNTER_ADD(&pData->StatIORead_in_2_2nd_bytes, ret);
281	}
282	);
283	}
284
285	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
286	#endif
287
288	/* Update fields */
289	sb->sb_cc += nn;
290	sb->sb_wptr += nn;
291	if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
292	sb->sb_wptr -= sb->sb_datalen;
293	STAM_PROFILE_STOP(&pData->StatIOread, a);
294	SOCKET_UNLOCK(so);
295	return nn;
296	}
297
298	/*
299	* Get urgent data
300	*
301	* When the socket is created, we set it SO_OOBINLINE,
302	* so when OOB data arrives, we soread() it and everything
303	* in the send buffer is sent as urgent data
304	*/
305	void
306	sorecvoob(PNATState pData, struct socket *so)
307	{
308	struct tcpcb *tp = sototcpcb(so);
309
310	DEBUG_CALL("sorecvoob");
311	DEBUG_ARG("so = %lx", (long)so);
312
313	/*
314	* We take a guess at how much urgent data has arrived.
315	* In most situations, when urgent data arrives, the next
316	* read() should get all the urgent data. This guess will
317	* be wrong however if more data arrives just after the
318	* urgent data, or the read() doesn't return all the
319	* urgent data.
320	*/
321	soread(pData, so);
322	tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
323	tp->t_force = 1;
324	tcp_output(pData, tp);
325	tp->t_force = 0;
326	}
327
328	/*
329	* Send urgent data
330	* There's a lot duplicated code here, but...
331	*/
332	int
333	sosendoob(struct socket *so)
334	{
335	struct sbuf *sb = &so->so_rcv;
336	char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
337
338	int n, len;
339
340	DEBUG_CALL("sosendoob");
341	DEBUG_ARG("so = %lx", (long)so);
342	DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
343
344	if (so->so_urgc > sizeof(buff))
345	so->so_urgc = sizeof(buff); /* XXX */
346
347	if (sb->sb_rptr < sb->sb_wptr)
348	{
349	/* We can send it directly */
350	n = send(so->s, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
351	so->so_urgc -= n;
352
353	DEBUG_MISC((dfd, " --- sent %d bytes urgent data, %d urgent bytes left\n",
354	n, so->so_urgc));
355	}
356	else
357	{
358	/*
359	* Since there's no sendv or sendtov like writev,
360	* we must copy all data to a linear buffer then
361	* send it all
362	*/
363	len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
364	if (len > so->so_urgc)
365	len = so->so_urgc;
366	memcpy(buff, sb->sb_rptr, len);
367	so->so_urgc -= len;
368	if (so->so_urgc)
369	{
370	n = sb->sb_wptr - sb->sb_data;
371	if (n > so->so_urgc)
372	n = so->so_urgc;
373	memcpy(buff + len, sb->sb_data, n);
374	so->so_urgc -= n;
375	len += n;
376	}
377	n = send(so->s, buff, len, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
378	#ifdef DEBUG
379	if (n != len)
380	DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
381	#endif
382	DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n",
383	n, so->so_urgc));
384	}
385
386	sb->sb_cc -= n;
387	sb->sb_rptr += n;
388	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
389	sb->sb_rptr -= sb->sb_datalen;
390
391	return n;
392	}
393
394	/*
395	* Write data from so_rcv to so's socket,
396	* updating all sbuf field as necessary
397	*/
398	int
399	sowrite(PNATState pData, struct socket *so)
400	{
401	int n, nn;
402	struct sbuf *sb = &so->so_rcv;
403	size_t len = sb->sb_cc;
404	struct iovec iov[2];
405
406	STAM_PROFILE_START(&pData->StatIOwrite, a);
407	STAM_COUNTER_RESET(&pData->StatIOWrite_in_1);
408	STAM_COUNTER_RESET(&pData->StatIOWrite_in_1_bytes);
409	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2);
410	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2_1st_bytes);
411	STAM_COUNTER_RESET(&pData->StatIOWrite_in_2_2nd_bytes);
412	STAM_COUNTER_RESET(&pData->StatIOWrite_no_w);
413	STAM_COUNTER_RESET(&pData->StatIOWrite_rest);
414	STAM_COUNTER_RESET(&pData->StatIOWrite_rest_bytes);
415	DEBUG_CALL("sowrite");
416	DEBUG_ARG("so = %lx", (long)so);
417	QSOCKET_LOCK(tcb);
418	SOCKET_LOCK(so);
419	QSOCKET_UNLOCK(tcb);
420	if (so->so_urgc)
421	{
422	sosendoob(so);
423	if (sb->sb_cc == 0)
424	{
425	SOCKET_UNLOCK(so);
426	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
427	return 0;
428	}
429	}
430
431	/*
432	* No need to check if there's something to write,
433	* sowrite wouldn't have been called otherwise
434	*/
435
436	len = sb->sb_cc;
437
438	iov[0].iov_base = sb->sb_rptr;
439	iov[1].iov_base = 0;
440	iov[1].iov_len = 0;
441	if (sb->sb_rptr < sb->sb_wptr)
442	{
443	iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
444	/* Should never succeed, but... */
445	if (iov[0].iov_len > len)
446	iov[0].iov_len = len;
447	n = 1;
448	}
449	else
450	{
451	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
452	if (iov[0].iov_len > len)
453	iov[0].iov_len = len;
454	len -= iov[0].iov_len;
455	if (len)
456	{
457	iov[1].iov_base = sb->sb_data;
458	iov[1].iov_len = sb->sb_wptr - sb->sb_data;
459	if (iov[1].iov_len > len)
460	iov[1].iov_len = len;
461	n = 2;
462	}
463	else
464	n = 1;
465	}
466	STAM_STATS({
467	if (n == 1)
468	{
469	STAM_COUNTER_INC(&pData->StatIOWrite_in_1);
470	STAM_COUNTER_ADD(&pData->StatIOWrite_in_1_bytes, iov[0].iov_len);
471	}
472	else
473	{
474	STAM_COUNTER_INC(&pData->StatIOWrite_in_2);
475	STAM_COUNTER_ADD(&pData->StatIOWrite_in_2_1st_bytes, iov[0].iov_len);
476	STAM_COUNTER_ADD(&pData->StatIOWrite_in_2_2nd_bytes, iov[1].iov_len);
477	}
478	});
479	/* Check if there's urgent data to send, and if so, send it */
480	#ifdef HAVE_READV
481	nn = writev(so->s, (const struct iovec *)iov, n);
482	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
483	#else
484	nn = send(so->s, iov[0].iov_base, iov[0].iov_len, 0);
485	#endif
486	/* This should never happen, but people tell me it does shrug */
487	if ( nn < 0
488	&& ( errno == EAGAIN
489	\|\| errno == EINTR
490	\|\| errno == EWOULDBLOCK))
491	{
492	SOCKET_UNLOCK(so);
493	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
494	return 0;
495	}
496
497	if (nn < 0 \|\| (nn == 0 && iov[0].iov_len > 0))
498	{
499	DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
500	so->so_state, errno));
501	sofcantsendmore(so);
502	tcp_sockclosed(pData, sototcpcb(so));
503	SOCKET_UNLOCK(so);
504	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
505	return -1;
506	}
507
508	#ifndef HAVE_READV
509	if (n == 2 && nn == iov[0].iov_len)
510	{
511	int ret;
512	ret = send(so->s, iov[1].iov_base, iov[1].iov_len, 0);
513	if (ret > 0)
514	nn += ret;
515	STAM_STATS({
516	if (ret > 0 && ret != iov[1].iov_len)
517	{
518	STAM_COUNTER_INC(&pData->StatIOWrite_rest);
519	STAM_COUNTER_ADD(&pData->StatIOWrite_rest_bytes, (ret - iov[1].iov_len));
520	}
521	});
522	}
523	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
524	#endif
525
526	/* Update sbuf */
527	sb->sb_cc -= nn;
528	sb->sb_rptr += nn;
529	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
530	sb->sb_rptr -= sb->sb_datalen;
531
532	/*
533	* If in DRAIN mode, and there's no more data, set
534	* it CANTSENDMORE
535	*/
536	if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
537	sofcantsendmore(so);
538
539	SOCKET_UNLOCK(so);
540	STAM_PROFILE_STOP(&pData->StatIOwrite, a);
541	return nn;
542	}
543
544	/*
545	* recvfrom() a UDP socket
546	*/
547	void
548	sorecvfrom(PNATState pData, struct socket *so)
549	{
550	ssize_t ret = 0;
551	struct sockaddr_in addr;
552	socklen_t addrlen = sizeof(struct sockaddr_in);
553
554	DEBUG_CALL("sorecvfrom");
555	DEBUG_ARG("so = %lx", (long)so);
556
557	if (so->so_type == IPPROTO_ICMP)
558	{
559	/* This is a "ping" reply */
560	#ifdef RT_OS_WINDOWS
561	sorecvfrom_icmp_win(pData, so);
562	#else /* RT_OS_WINDOWS */
563	sorecvfrom_icmp_unix(pData, so);
564	#endif /* !RT_OS_WINDOWS */
565	udp_detach(pData, so);
566	}
567	else
568	{
569	/* A "normal" UDP packet */
570	struct mbuf *m;
571	struct ethhdr *eh;
572	ssize_t len;
573	u_long n = 0;
574	#ifdef VBOX_WITH_SLIRP_BSD_MBUF
575	int size;
576	#endif
577	int rc = 0;
578	static int signalled = 0;
579
580	QSOCKET_LOCK(udb);
581	SOCKET_LOCK(so);
582	QSOCKET_UNLOCK(udb);
583
584	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
585	if (!(m = m_get(pData)))
586	{
587	SOCKET_UNLOCK(so);
588	return;
589	}
590	/* adjust both parameters to maks M_FREEROOM calculate correct */
591	m->m_data += if_maxlinkhdr + sizeof(struct udphdr) + sizeof(struct ip);
592
593	/*
594	* XXX Shouldn't FIONREAD packets destined for port 53,
595	* but I don't know the max packet size for DNS lookups
596	*/
597	len = M_FREEROOM(m);
598	/* if (so->so_fport != RT_H2N_U16_C(53)) */
599	rc = ioctlsocket(so->s, FIONREAD, &n);
600	if ( rc == -1
601	&& ( errno == EAGAIN
602	\|\| errno == EWOULDBLOCK
603	\|\| errno == EINPROGRESS
604	\|\| errno == ENOTCONN))
605	{
606	m_free(pData, m);
607	return;
608	}
609
610	Log2(("NAT: %R[natsock] ioctlsocket before read "
611	"(rc:%d errno:%d, n:%d)\n", so, rc, errno, n));
612
613	if (rc == -1 && signalled == 0)
614	{
615	LogRel(("NAT: can't fetch amount of bytes on socket %R[natsock], so message will be truncated.\n", so));
616	signalled = 1;
617	m_free(pData, m);
618	return;
619	}
620
621	if (rc != -1 && n > len)
622	{
623	n = (m->m_data - m->m_dat) + m->m_len + n + 1;
624	m_inc(m, n);
625	len = M_FREEROOM(m);
626	}
627	ret = recvfrom(so->s, m->m_data, len, 0,
628	(struct sockaddr *)&addr, &addrlen);
629	Log2(("NAT: %R[natsock] ioctlsocket after read "
630	"(rc:%d errno:%d, n:%d) ret:%d, len:%d\n", so,
631	rc, errno, n, ret, len));
632	#else
633	/How many data has been received ?/
634	/*
635	* 1. calculate how much we can read
636	* 2. read as much as possible
637	* 3. attach buffer to allocated header mbuf
638	*/
639	rc = ioctlsocket(so->s, FIONREAD, &n);
640	if (rc == -1 && signalled == 0)
641	{
642	LogRel(("NAT: can't fetch amount of bytes on socket %R[natsock], so message will be truncated.\n", so));
643	signalled = 1;
644	}
645
646	len = sizeof(struct udpiphdr) + ETH_HLEN;
647	if (n > (if_mtu - len))
648	{
649	n = if_mtu - len; /* can't read than we can put in the mbuf*/
650	}
651	len += n;
652
653	size = MCLBYTES;
654	if (len < MSIZE)
655	size = MCLBYTES;
656	else if (len < MCLBYTES)
657	size = MCLBYTES;
658	else if (len < MJUM9BYTES)
659	size = MJUM9BYTES;
660	else if (len < MJUM16BYTES)
661	size = MJUM16BYTES;
662	else
663	AssertMsgFailed(("Unsupported size"));
664
665	m = m_getjcl(pData, M_NOWAIT, MT_HEADER, M_PKTHDR, size);
666	m->m_data += ETH_HLEN;
667	m->m_pkthdr.header = mtod(m, void *);
668	m->m_data += sizeof(struct udpiphdr);
669	ret = recvfrom(so->s, mtod(m, char *), n, 0,
670	(struct sockaddr *)&addr, &addrlen);
671	/* @todo (vvl) check which flags and type should be passed */
672	#endif
673	m->m_len = ret;
674	if (ret < 0)
675	{
676	u_char code = ICMP_UNREACH_PORT;
677
678	if (errno == EHOSTUNREACH)
679	code = ICMP_UNREACH_HOST;
680	else if (errno == ENETUNREACH)
681	code = ICMP_UNREACH_NET;
682
683	m_free(pData, m);
684	if ( errno == EAGAIN
685	\|\| errno == EWOULDBLOCK
686	\|\| errno == EINPROGRESS
687	\|\| errno == ENOTCONN)
688	{
689	return;
690	}
691
692	Log2((" rx error, tx icmp ICMP_UNREACH:%i\n", code));
693	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
694	so->so_m = NULL;
695	}
696	else
697	{
698	/*
699	* Hack: domain name lookup will be used the most for UDP,
700	* and since they'll only be used once there's no need
701	* for the 4 minute (or whatever) timeout... So we time them
702	* out much quicker (10 seconds for now...)
703	*/
704	if (so->so_expire)
705	{
706	if (so->so_fport != RT_H2N_U16_C(53))
707	so->so_expire = curtime + SO_EXPIRE;
708	}
709	/*
710	* last argument should be changed if Slirp will inject IP attributes
711	* Note: Here we can't check if dnsproxy's sent initial request
712	*/
713	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
714	if (so->so_fport == RT_H2N_U16_C(53))
715	dnsproxy_answer(pData, so, m);
716	#endif
717
718	#if 0
719	if (m->m_len == len)
720	{
721	m_inc(m, MINCSIZE);
722	m->m_len = 0;
723	}
724	#endif
725
726	/*
727	* If this packet was destined for CTL_ADDR,
728	* make it look like that's where it came from, done by udp_output
729	*/
730	udp_output(pData, so, m, &addr);
731	SOCKET_UNLOCK(so);
732	} /* rx error */
733	} /* if ping packet */
734	}
735
736	/*
737	* sendto() a socket
738	*/
739	int
740	sosendto(PNATState pData, struct socket so, struct mbuf m)
741	{
742	int ret;
743	struct sockaddr_in *paddr;
744	struct sockaddr addr;
745	#if 0
746	struct sockaddr_in host_addr;
747	#endif
748	#ifdef VBOX_WITH_SLIRP_BSD_MBUF
749	caddr_t buf;
750	int mlen;
751	#endif
752
753	DEBUG_CALL("sosendto");
754	DEBUG_ARG("so = %lx", (long)so);
755	DEBUG_ARG("m = %lx", (long)m);
756
757	memset(&addr, 0, sizeof(struct sockaddr));
758	#ifdef RT_OS_DARWIN
759	addr.sa_len = sizeof(struct sockaddr_in);
760	#endif
761	paddr = (struct sockaddr_in *)&addr;
762	paddr->sin_family = AF_INET;
763	if ((so->so_faddr.s_addr & RT_H2N_U32(pData->netmask)) == pData->special_addr.s_addr)
764	{
765	/* It's an alias */
766	uint32_t last_byte = RT_N2H_U32(so->so_faddr.s_addr) & ~pData->netmask;
767	switch(last_byte)
768	{
769	#if 0
770	/* handle this case at 'default:' */
771	case CTL_BROADCAST:
772	addr.sin_addr.s_addr = INADDR_BROADCAST;
773	/* Send the packet to host to fully emulate broadcast */
774	/** @todo r=klaus: on Linux host this causes the host to receive
775	* the packet twice for some reason. And I cannot find any place
776	* in the man pages which states that sending a broadcast does not
777	* reach the host itself. */
778	host_addr.sin_family = AF_INET;
779	host_addr.sin_port = so->so_fport;
780	host_addr.sin_addr = our_addr;
781	sendto(so->s, m->m_data, m->m_len, 0,
782	(struct sockaddr *)&host_addr, sizeof (struct sockaddr));
783	break;
784	#endif
785	case CTL_DNS:
786	case CTL_ALIAS:
787	default:
788	if (last_byte == ~pData->netmask)
789	paddr->sin_addr.s_addr = INADDR_BROADCAST;
790	else
791	paddr->sin_addr = loopback_addr;
792	break;
793	}
794	}
795	else
796	paddr->sin_addr = so->so_faddr;
797	paddr->sin_port = so->so_fport;
798
799	DEBUG_MISC((dfd, " sendto()ing, addr.sin_port=%d, addr.sin_addr.s_addr=%.16s\n",
800	RT_N2H_U16(paddr->sin_port), inet_ntoa(paddr->sin_addr)));
801
802	/* Don't care what port we get */
803	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
804	ret = sendto(so->s, m->m_data, m->m_len, 0, &addr, sizeof (struct sockaddr_in));
805	#else
806	mlen = m_length(m, NULL);
807	buf = RTMemAlloc(mlen);
808	if (buf == NULL)
809	{
810	return -1;
811	}
812	m_copydata(m, 0, mlen, buf);
813	ret = sendto(so->s, buf, mlen, 0,
814	(struct sockaddr *)&addr, sizeof (struct sockaddr));
815	#endif
816	if (ret < 0)
817	{
818	Log2(("UDP: sendto fails (%s)\n", strerror(errno)));
819	return -1;
820	}
821
822	/*
823	* Kill the socket if there's no reply in 4 minutes,
824	* but only if it's an expirable socket
825	*/
826	if (so->so_expire)
827	so->so_expire = curtime + SO_EXPIRE;
828	so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
829	return 0;
830	}
831
832	/*
833	* XXX This should really be tcp_listen
834	*/
835	struct socket *
836	solisten(PNATState pData, u_int32_t bind_addr, u_int port, u_int32_t laddr, u_int lport, int flags)
837	{
838	struct sockaddr_in addr;
839	struct socket *so;
840	socklen_t addrlen = sizeof(addr);
841	int s, opt = 1;
842	int status;
843
844	DEBUG_CALL("solisten");
845	DEBUG_ARG("port = %d", port);
846	DEBUG_ARG("laddr = %x", laddr);
847	DEBUG_ARG("lport = %d", lport);
848	DEBUG_ARG("flags = %x", flags);
849
850	if ((so = socreate()) == NULL)
851	{
852	/* RTMemFree(so); Not sofree() ??? free(NULL) == NOP */
853	return NULL;
854	}
855
856	/* Don't tcp_attach... we don't need so_snd nor so_rcv */
857	if ((so->so_tcpcb = tcp_newtcpcb(pData, so)) == NULL)
858	{
859	RTMemFree(so);
860	return NULL;
861	}
862
863	SOCKET_LOCK_CREATE(so);
864	SOCKET_LOCK(so);
865	QSOCKET_LOCK(tcb);
866	insque(pData, so,&tcb);
867	NSOCK_INC();
868	QSOCKET_UNLOCK(tcb);
869
870	/*
871	* SS_FACCEPTONCE sockets must time out.
872	*/
873	if (flags & SS_FACCEPTONCE)
874	so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
875
876	so->so_state = (SS_FACCEPTCONN\|flags);
877	so->so_lport = lport; /* Kept in network format */
878	so->so_laddr.s_addr = laddr; /* Ditto */
879
880	memset(&addr, 0, sizeof(addr));
881	#ifdef RT_OS_DARWIN
882	addr.sin_len = sizeof(addr);
883	#endif
884	addr.sin_family = AF_INET;
885	addr.sin_addr.s_addr = bind_addr;
886	addr.sin_port = port;
887
888	if ( ((s = socket(AF_INET, SOCK_STREAM, 0)) < 0)
889	\|\| (setsockopt(s, SOL_SOCKET, SO_REUSEADDR,(char *)&opt, sizeof(int)) < 0)
890	\|\| (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0)
891	\|\| (listen(s, 1) < 0))
892	{
893	#ifdef RT_OS_WINDOWS
894	int tmperrno = WSAGetLastError(); /* Don't clobber the real reason we failed */
895	closesocket(s);
896	QSOCKET_LOCK(tcb);
897	sofree(pData, so);
898	QSOCKET_UNLOCK(tcb);
899	/* Restore the real errno */
900	WSASetLastError(tmperrno);
901	#else
902	int tmperrno = errno; /* Don't clobber the real reason we failed */
903	close(s);
904	QSOCKET_LOCK(tcb);
905	sofree(pData, so);
906	QSOCKET_UNLOCK(tcb);
907	/* Restore the real errno */
908	errno = tmperrno;
909	#endif
910	return NULL;
911	}
912	fd_nonblock(s);
913	setsockopt(s, SOL_SOCKET, SO_OOBINLINE,(char *)&opt, sizeof(int));
914
915	getsockname(s,(struct sockaddr *)&addr,&addrlen);
916	so->so_fport = addr.sin_port;
917	/* set socket buffers */
918	opt = pData->socket_rcv;
919	status = setsockopt(s, SOL_SOCKET, SO_RCVBUF, (char *)&opt, sizeof(int));
920	if (status < 0)
921	{
922	LogRel(("NAT: Error(%d) while setting RCV capacity to (%d)\n", errno, opt));
923	goto no_sockopt;
924	}
925	opt = pData->socket_snd;
926	status = setsockopt(s, SOL_SOCKET, SO_SNDBUF, (char *)&opt, sizeof(int));
927	if (status < 0)
928	{
929	LogRel(("NAT: Error(%d) while setting SND capacity to (%d)\n", errno, opt));
930	goto no_sockopt;
931	}
932	no_sockopt:
933	if (addr.sin_addr.s_addr == 0 \|\| addr.sin_addr.s_addr == loopback_addr.s_addr)
934	so->so_faddr = alias_addr;
935	else
936	so->so_faddr = addr.sin_addr;
937
938	so->s = s;
939	SOCKET_UNLOCK(so);
940	return so;
941	}
942
943	/*
944	* Data is available in so_rcv
945	* Just write() the data to the socket
946	* XXX not yet...
947	*/
948	void
949	sorwakeup(struct socket *so)
950	{
951	#if 0
952	sowrite(so);
953	FD_CLR(so->s,&writefds);
954	#endif
955	}
956
957	/*
958	* Data has been freed in so_snd
959	* We have room for a read() if we want to
960	* For now, don't read, it'll be done in the main loop
961	*/
962	void
963	sowwakeup(struct socket *so)
964	{
965	}
966
967	/*
968	* Various session state calls
969	* XXX Should be #define's
970	* The socket state stuff needs work, these often get call 2 or 3
971	* times each when only 1 was needed
972	*/
973	void
974	soisfconnecting(struct socket *so)
975	{
976	so->so_state &= ~(SS_NOFDREF\|SS_ISFCONNECTED\|SS_FCANTRCVMORE\|
977	SS_FCANTSENDMORE\|SS_FWDRAIN);
978	so->so_state \|= SS_ISFCONNECTING; /* Clobber other states */
979	}
980
981	void
982	soisfconnected(struct socket *so)
983	{
984	so->so_state &= ~(SS_ISFCONNECTING\|SS_FWDRAIN\|SS_NOFDREF);
985	so->so_state \|= SS_ISFCONNECTED; /* Clobber other states */
986	}
987
988	void
989	sofcantrcvmore(struct socket *so)
990	{
991	if ((so->so_state & SS_NOFDREF) == 0)
992	{
993	shutdown(so->s, 0);
994	}
995	so->so_state &= ~(SS_ISFCONNECTING);
996	if (so->so_state & SS_FCANTSENDMORE)
997	so->so_state = SS_NOFDREF; /* Don't select it */
998	/* XXX close() here as well? */
999	else
1000	so->so_state \|= SS_FCANTRCVMORE;
1001	}
1002
1003	void
1004	sofcantsendmore(struct socket *so)
1005	{
1006	if ((so->so_state & SS_NOFDREF) == 0)
1007	shutdown(so->s, 1); /* send FIN to fhost */
1008
1009	so->so_state &= ~(SS_ISFCONNECTING);
1010	if (so->so_state & SS_FCANTRCVMORE)
1011	so->so_state = SS_NOFDREF; /* as above */
1012	else
1013	so->so_state \|= SS_FCANTSENDMORE;
1014	}
1015
1016	void
1017	soisfdisconnected(struct socket *so)
1018	{
1019	#if 0
1020	so->so_state &= ~(SS_ISFCONNECTING\|SS_ISFCONNECTED);
1021	close(so->s);
1022	so->so_state = SS_ISFDISCONNECTED;
1023	/*
1024	* XXX Do nothing ... ?
1025	*/
1026	#endif
1027	}
1028
1029	/*
1030	* Set write drain mode
1031	* Set CANTSENDMORE once all data has been write()n
1032	*/
1033	void
1034	sofwdrain(struct socket *so)
1035	{
1036	if (so->so_rcv.sb_cc)
1037	so->so_state \|= SS_FWDRAIN;
1038	else
1039	sofcantsendmore(so);
1040	}
1041
1042	static void
1043	send_icmp_to_guest(PNATState pData, char buff, size_t len, struct socket so, const struct sockaddr_in *addr)
1044	{
1045	struct ip *ip;
1046	uint32_t dst, src;
1047	char ip_copy[256];
1048	struct icmp *icp;
1049	int old_ip_len = 0;
1050	int hlen, original_hlen = 0;
1051	struct mbuf *m;
1052	struct icmp_msg *icm;
1053	uint8_t proto;
1054	int type = 0;
1055
1056	ip = (struct ip *)buff;
1057	hlen = (ip->ip_hl << 2);
1058	icp = (struct icmp )((char )ip + hlen);
1059
1060	Log(("ICMP:received msg(t:%d, c:%d)\n", icp->icmp_type, icp->icmp_code));
1061	if ( icp->icmp_type != ICMP_ECHOREPLY
1062	&& icp->icmp_type != ICMP_TIMXCEED
1063	&& icp->icmp_type != ICMP_UNREACH)
1064	{
1065	return;
1066	}
1067
1068	type = icp->icmp_type;
1069	if ( type == ICMP_TIMXCEED
1070	\|\| type == ICMP_UNREACH)
1071	{
1072	ip = &icp->icmp_ip;
1073	}
1074
1075	icm = icmp_find_original_mbuf(pData, ip);
1076
1077	if (icm == NULL)
1078	{
1079	Log(("NAT: Can't find the corresponding packet for the received ICMP\n"));
1080	return;
1081	}
1082
1083	m = icm->im_m;
1084	Assert(m != NULL);
1085
1086	src = addr->sin_addr.s_addr;
1087
1088	ip = mtod(m, struct ip *);
1089	proto = ip->ip_p;
1090	/* Now ip is pointing on header we've sent from guest */
1091	if ( icp->icmp_type == ICMP_TIMXCEED
1092	\|\| icp->icmp_type == ICMP_UNREACH)
1093	{
1094	old_ip_len = (ip->ip_hl << 2) + 64;
1095	if (old_ip_len > sizeof(ip_copy))
1096	old_ip_len = sizeof(ip_copy);
1097	memcpy(ip_copy, ip, old_ip_len);
1098	}
1099
1100	/* source address from original IP packet*/
1101	dst = ip->ip_src.s_addr;
1102
1103	/* overide ther tail of old packet */
1104	ip = mtod(m, struct ip ); / ip is from mbuf we've overrided */
1105	original_hlen = ip->ip_hl << 2;
1106	/* saves original ip header and options */
1107	memcpy(m->m_data + original_hlen, buff + hlen, len - hlen);
1108	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
1109	m->m_len = len - hlen + original_hlen;
1110	ip->ip_len = m->m_len;
1111	#else
1112	ip->ip_len = m_length(m, NULL);
1113	#endif
1114	ip->ip_p = IPPROTO_ICMP; /* the original package could be whatever, but we're response via ICMP*/
1115
1116	icp = (struct icmp )((char )ip + (ip->ip_hl << 2));
1117	type = icp->icmp_type;
1118	if ( type == ICMP_TIMXCEED
1119	\|\| type == ICMP_UNREACH)
1120	{
1121	/* according RFC 793 error messages required copy of initial IP header + 64 bit */
1122	memcpy(&icp->icmp_ip, ip_copy, old_ip_len);
1123	ip->ip_tos = ((ip->ip_tos & 0x1E) \| 0xC0); /* high priority for errors */
1124	}
1125
1126	ip->ip_src.s_addr = src;
1127	ip->ip_dst.s_addr = dst;
1128	icmp_reflect(pData, m);
1129	LIST_REMOVE(icm, im_list);
1130	/* Don't call m_free here*/
1131
1132	if ( type == ICMP_TIMXCEED
1133	\|\| type == ICMP_UNREACH)
1134	{
1135	icm->im_so->so_m = NULL;
1136	switch (proto)
1137	{
1138	case IPPROTO_UDP:
1139	/XXX: so->so_m already freed so we shouldn't call sofree /
1140	udp_detach(pData, icm->im_so);
1141	break;
1142	case IPPROTO_TCP:
1143	/close tcp should be here /
1144	break;
1145	default:
1146	/* do nothing */
1147	break;
1148	}
1149	}
1150	RTMemFree(icm);
1151	}
1152
1153	#ifdef RT_OS_WINDOWS
1154	static void
1155	sorecvfrom_icmp_win(PNATState pData, struct socket *so)
1156	{
1157	int len;
1158	int i;
1159	struct ip *ip;
1160	struct mbuf *m;
1161	struct icmp *icp;
1162	struct icmp_msg *icm;
1163	struct ip ip_broken; / ICMP returns header + 64 bit of packet */
1164	uint32_t src;
1165	ICMP_ECHO_REPLY *icr;
1166	int hlen = 0;
1167	int data_len = 0;
1168	int nbytes = 0;
1169	u_char code = ~0;
1170
1171	len = pData->pfIcmpParseReplies(pData->pvIcmpBuffer, pData->szIcmpBuffer);
1172	if (len < 0)
1173	{
1174	LogRel(("NAT: Error (%d) occurred on ICMP receiving\n", GetLastError()));
1175	return;
1176	}
1177	if (len == 0)
1178	return; /* no error */
1179
1180	icr = (ICMP_ECHO_REPLY *)pData->pvIcmpBuffer;
1181	for (i = 0; i < len; ++i)
1182	{
1183	switch(icr[i].Status)
1184	{
1185	case IP_DEST_HOST_UNREACHABLE:
1186	code = (code != ~0 ? code : ICMP_UNREACH_HOST);
1187	case IP_DEST_NET_UNREACHABLE:
1188	code = (code != ~0 ? code : ICMP_UNREACH_NET);
1189	case IP_DEST_PROT_UNREACHABLE:
1190	code = (code != ~0 ? code : ICMP_UNREACH_PROTOCOL);
1191	/* UNREACH error inject here */
1192	case IP_DEST_PORT_UNREACHABLE:
1193	code = (code != ~0 ? code : ICMP_UNREACH_PORT);
1194	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, "Error occurred!!!");
1195	so->so_m = NULL;
1196	break;
1197	case IP_SUCCESS: /* echo replied */
1198	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
1199	m = m_get(pData);
1200	#else
1201	m = m_gethdr(pData, M_NOWAIT, MT_HEADER);
1202	#endif
1203	m->m_data += if_maxlinkhdr;
1204	ip = mtod(m, struct ip *);
1205	ip->ip_src.s_addr = icr[i].Address;
1206	ip->ip_p = IPPROTO_ICMP;
1207	ip->ip_dst.s_addr = so->so_laddr.s_addr; /XXX: still the hack/
1208	data_len = sizeof(struct ip);
1209	ip->ip_hl = data_len >> 2; /* requiered for icmp_reflect, no IP options */
1210	ip->ip_ttl = icr[i].Options.Ttl;
1211
1212	icp = (struct icmp )&ip[1]; / no options */
1213	icp->icmp_type = ICMP_ECHOREPLY;
1214	icp->icmp_code = 0;
1215	icp->icmp_id = so->so_icmp_id;
1216	icp->icmp_seq = so->so_icmp_seq;
1217
1218	data_len += ICMP_MINLEN;
1219
1220	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
1221	nbytes = (data_len + icr[i].DataSize > m->m_size? m->m_size - data_len: icr[i].DataSize);
1222	memcpy(icp->icmp_data, icr[i].Data, nbytes);
1223	#else
1224	AssertMsgFailed(("ICMP"));
1225	#endif
1226
1227	data_len += icr[i].DataSize;
1228
1229	ip->ip_len = data_len;
1230	m->m_len = ip->ip_len;
1231
1232	icmp_reflect(pData, m);
1233	break;
1234	case IP_TTL_EXPIRED_TRANSIT: /* TTL expired */
1235
1236	ip_broken = icr[i].Data;
1237	icm = icmp_find_original_mbuf(pData, ip_broken);
1238	if (icm == NULL) {
1239	Log(("ICMP: can't find original package (first double word %x)\n", (uint32_t )ip_broken));
1240	return;
1241	}
1242	m = icm->im_m;
1243	ip = mtod(m, struct ip *);
1244	ip->ip_ttl = icr[i].Options.Ttl;
1245	src = ip->ip_src.s_addr;
1246	ip->ip_dst.s_addr = src;
1247	ip->ip_dst.s_addr = icr[i].Address;
1248
1249	hlen = (ip->ip_hl << 2);
1250	icp = (struct icmp )((char )ip + hlen);
1251	ip_broken->ip_src.s_addr = src; /it packet sent from host not from guest/
1252	data_len = (ip_broken->ip_hl << 2) + 64;
1253
1254	#ifndef VBOX_WITH_SLIRP_BSD_MBUF
1255	nbytes =(hlen + ICMP_MINLEN + data_len > m->m_size? m->m_size - (hlen + ICMP_MINLEN): data_len);
1256	memcpy(icp->icmp_data, ip_broken, nbytes);
1257	#else
1258	AssertMsgFailed(("ICMP"));
1259	#endif
1260	icmp_reflect(pData, m);
1261	break;
1262	default:
1263	Log(("ICMP(default): message with Status: %x was received from %x\n", icr[i].Status, icr[i].Address));
1264	break;
1265	}
1266	}
1267	}
1268	#else /* RT_OS_WINDOWS */
1269	static void sorecvfrom_icmp_unix(PNATState pData, struct socket *so)
1270	{
1271	struct sockaddr_in addr;
1272	socklen_t addrlen = sizeof(struct sockaddr_in);
1273	char *buff;
1274	int len = 0;
1275	int rc = 0;
1276	static int signalled = 0;
1277
1278	rc = ioctlsocket(so->s, FIONREAD, &len);
1279	if ( rc == -1
1280	&& ( errno == EAGAIN
1281	\|\| errno == EWOULDBLOCK
1282	\|\| errno == EINPROGRESS
1283	\|\| errno == ENOTCONN))
1284	{
1285	return;
1286	}
1287	if (rc == -1 && signalled == 0)
1288	{
1289	signalled = 1;
1290	LogRel(("NAT: fetching number of bits has been failed for ICMP socket (%d: %s)\n",
1291	errno, strerror(errno)));
1292	return;
1293	}
1294	len = (len != 0 && rc != -1 ? len : 1500);
1295	buff = RTMemAlloc(len);
1296	len = recvfrom(so->s, buff, len, 0,
1297	(struct sockaddr *)&addr, &addrlen);
1298	/* XXX Check if reply is "correct"? */
1299
1300	if (len == -1 \|\| len == 0)
1301	{
1302	u_char code;
1303	if ( len == -1
1304	&& (errno == EAGAIN
1305	\|\| errno == EWOULDBLOCK
1306	\|\| errno == EINPROGRESS
1307	\|\| errno == ENOTCONN))
1308	{
1309	return;
1310	}
1311	code = ICMP_UNREACH_PORT;
1312
1313	if (errno == EHOSTUNREACH)
1314	code = ICMP_UNREACH_HOST;
1315	else if (errno == ENETUNREACH)
1316	code = ICMP_UNREACH_NET;
1317
1318	LogRel((" udp icmp rx errno = %d-%s\n",
1319	errno, strerror(errno)));
1320	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
1321	so->so_m = NULL;
1322	}
1323	else
1324	{
1325	send_icmp_to_guest(pData, buff, len, so, &addr);
1326	}
1327	RTMemFree(buff);
1328	}
1329	#endif /* !RT_OS_WINDOWS */

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/socket.c@ 26837

Download in other formats: