socket.c@ 23253

最後變更在這個檔案從23253是 23241,由 vboxsync 提交於 16 年前
NAT: do UDP/ICMP errno analisys.
屬性 svn:eol-style 設為 `native`
檔案大小: 37.0 KB

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

注意: 瀏覽 TracBrowser 來幫助您使用儲存庫瀏覽器

source: vbox/trunk/src/VBox/Devices/Network/slirp/socket.c@ 23253

以其他格式下載: