socket.c@ 20958

Last change on this file since 20958 was 20712, checked in by vboxsync, 15 years ago
DrvNat,slirp: simplify statistics and deregister them.
Property svn:eol-style set to `native`
File size: 33.5 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 && (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	struct sockaddr_in addr;
545	socklen_t addrlen = sizeof(struct sockaddr_in);
546
547	DEBUG_CALL("sorecvfrom");
548	DEBUG_ARG("so = %lx", (long)so);
549
550	if (so->so_type == IPPROTO_ICMP)
551	{
552	/* This is a "ping" reply */
553	#ifdef RT_OS_WINDOWS
554	sorecvfrom_icmp_win(pData, so);
555	#else /* RT_OS_WINDOWS */
556	sorecvfrom_icmp_unix(pData, so);
557	#endif /* !RT_OS_WINDOWS */
558	udp_detach(pData, so);
559	}
560	else
561	{
562	/* A "normal" UDP packet */
563	struct mbuf *m;
564	struct ethhdr *eh;
565	size_t len;
566	u_long n;
567
568	QSOCKET_LOCK(udb);
569	SOCKET_LOCK(so);
570	QSOCKET_UNLOCK(udb);
571
572	if (!(m = m_get(pData)))
573	{
574	SOCKET_UNLOCK(so);
575	return;
576	}
577	/* adjust both parameters to maks M_FREEROOM calculate correct */
578	m_adj(m, if_maxlinkhdr + sizeof(struct udphdr) + sizeof(struct ip));
579
580	/*
581	* XXX Shouldn't FIONREAD packets destined for port 53,
582	* but I don't know the max packet size for DNS lookups
583	*/
584	len = M_FREEROOM(m);
585	/* if (so->so_fport != htons(53)) */
586	{
587	ioctlsocket(so->s, FIONREAD, &n);
588
589	if (n > len)
590	{
591	n = (m->m_data - m->m_dat) + m->m_len + n + 1;
592	m_inc(m, n);
593	len = M_FREEROOM(m);
594	}
595	}
596
597	m->m_len = recvfrom(so->s, m->m_data, len, 0,
598	(struct sockaddr *)&addr, &addrlen);
599	Log2((" did recvfrom %d, errno = %d-%s\n",
600	m->m_len, errno, strerror(errno)));
601	if(m->m_len < 0)
602	{
603	u_char code = ICMP_UNREACH_PORT;
604
605	if (errno == EHOSTUNREACH)
606	code = ICMP_UNREACH_HOST;
607	else if(errno == ENETUNREACH)
608	code = ICMP_UNREACH_NET;
609
610	Log2((dfd," rx error, tx icmp ICMP_UNREACH:%i\n", code));
611	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
612	so->so_m = NULL;
613	m_free(pData, m);
614	}
615	else
616	{
617	/*
618	* Hack: domain name lookup will be used the most for UDP,
619	* and since they'll only be used once there's no need
620	* for the 4 minute (or whatever) timeout... So we time them
621	* out much quicker (10 seconds for now...)
622	*/
623	#ifndef VBOX_WITH_SLIRP_DNS_PROXY
624	if (so->so_expire)
625	{
626	if (so->so_fport == htons(53))
627	so->so_expire = curtime + SO_EXPIREFAST;
628	else
629	so->so_expire = curtime + SO_EXPIRE;
630	}
631	#else
632	if (so->so_expire)
633	{
634	if (so->so_fport != htons(53))
635	so->so_expire = curtime + SO_EXPIRE;
636	}
637	/*
638	* last argument should be changed if Slirp will inject IP attributes
639	* Note: Here we can't check if dnsproxy's sent initial request
640	*/
641	if (so->so_fport == htons(53))
642	dnsproxy_answer(pData, so, m);
643	#endif
644
645	#if 0
646	if (m->m_len == len)
647	{
648	m_inc(m, MINCSIZE);
649	m->m_len = 0;
650	}
651	#endif
652
653	/*
654	* If this packet was destined for CTL_ADDR,
655	* make it look like that's where it came from, done by udp_output
656	*/
657	udp_output(pData, so, m, &addr);
658	SOCKET_UNLOCK(so);
659	} /* rx error */
660	} /* if ping packet */
661	}
662
663	/*
664	* sendto() a socket
665	*/
666	int
667	sosendto(PNATState pData, struct socket so, struct mbuf m)
668	{
669	int ret;
670	struct sockaddr_in addr;
671	#if 0
672	struct sockaddr_in host_addr;
673	#endif
674
675	DEBUG_CALL("sosendto");
676	DEBUG_ARG("so = %lx", (long)so);
677	DEBUG_ARG("m = %lx", (long)m);
678
679	addr.sin_family = AF_INET;
680	if ((so->so_faddr.s_addr & htonl(pData->netmask)) == special_addr.s_addr)
681	{
682	/* It's an alias */
683	uint32_t last_byte = ntohl(so->so_faddr.s_addr) & ~pData->netmask;
684	switch(last_byte)
685	{
686	#if 0
687	/* handle this case at 'default:' */
688	case CTL_BROADCAST:
689	addr.sin_addr.s_addr = INADDR_BROADCAST;
690	/* Send the packet to host to fully emulate broadcast */
691	/** @todo r=klaus: on Linux host this causes the host to receive
692	* the packet twice for some reason. And I cannot find any place
693	* in the man pages which states that sending a broadcast does not
694	* reach the host itself. */
695	host_addr.sin_family = AF_INET;
696	host_addr.sin_port = so->so_fport;
697	host_addr.sin_addr = our_addr;
698	sendto(so->s, m->m_data, m->m_len, 0,
699	(struct sockaddr *)&host_addr, sizeof (struct sockaddr));
700	break;
701	#endif
702	case CTL_DNS:
703	#ifndef VBOX_WITH_MULTI_DNS
704	if (!get_dns_addr(pData, &dns_addr))
705	addr.sin_addr = dns_addr;
706	else
707	addr.sin_addr = loopback_addr;
708	break;
709	#endif
710	case CTL_ALIAS:
711	default:
712	if (last_byte == ~pData->netmask)
713	addr.sin_addr.s_addr = INADDR_BROADCAST;
714	else
715	addr.sin_addr = loopback_addr;
716	break;
717	}
718	}
719	else
720	addr.sin_addr = so->so_faddr;
721	addr.sin_port = so->so_fport;
722
723	DEBUG_MISC((dfd, " sendto()ing, addr.sin_port=%d, addr.sin_addr.s_addr=%.16s\n",
724	ntohs(addr.sin_port), inet_ntoa(addr.sin_addr)));
725
726	/* Don't care what port we get */
727	ret = sendto(so->s, m->m_data, m->m_len, 0,
728	(struct sockaddr *)&addr, sizeof (struct sockaddr));
729	if (ret < 0)
730	{
731	LogRel(("UDP: sendto fails (%s)\n", strerror(errno)));
732	return -1;
733	}
734
735	/*
736	* Kill the socket if there's no reply in 4 minutes,
737	* but only if it's an expirable socket
738	*/
739	if (so->so_expire)
740	so->so_expire = curtime + SO_EXPIRE;
741	so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
742	return 0;
743	}
744
745	/*
746	* XXX This should really be tcp_listen
747	*/
748	struct socket *
749	solisten(PNATState pData, u_int port, u_int32_t laddr, u_int lport, int flags)
750	{
751	struct sockaddr_in addr;
752	struct socket *so;
753	socklen_t addrlen = sizeof(addr);
754	int s, opt = 1;
755	int status;
756
757	DEBUG_CALL("solisten");
758	DEBUG_ARG("port = %d", port);
759	DEBUG_ARG("laddr = %x", laddr);
760	DEBUG_ARG("lport = %d", lport);
761	DEBUG_ARG("flags = %x", flags);
762
763	if ((so = socreate()) == NULL)
764	{
765	/* RTMemFree(so); Not sofree() ??? free(NULL) == NOP */
766	return NULL;
767	}
768
769	/* Don't tcp_attach... we don't need so_snd nor so_rcv */
770	if ((so->so_tcpcb = tcp_newtcpcb(pData, so)) == NULL)
771	{
772	RTMemFree(so);
773	return NULL;
774	}
775
776	SOCKET_LOCK_CREATE(so);
777	SOCKET_LOCK(so);
778	QSOCKET_LOCK(tcb);
779	insque(pData, so,&tcb);
780	NSOCK_INC();
781	QSOCKET_UNLOCK(tcb);
782
783	/*
784	* SS_FACCEPTONCE sockets must time out.
785	*/
786	if (flags & SS_FACCEPTONCE)
787	so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
788
789	so->so_state = (SS_FACCEPTCONN\|flags);
790	so->so_lport = lport; /* Kept in network format */
791	so->so_laddr.s_addr = laddr; /* Ditto */
792
793	addr.sin_family = AF_INET;
794	addr.sin_addr.s_addr = INADDR_ANY;
795	addr.sin_port = port;
796
797	if ( ((s = socket(AF_INET, SOCK_STREAM, 0)) < 0)
798	\|\| (setsockopt(s, SOL_SOCKET, SO_REUSEADDR,(char *)&opt, sizeof(int)) < 0)
799	\|\| (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0)
800	\|\| (listen(s, 1) < 0))
801	{
802	#ifdef RT_OS_WINDOWS
803	int tmperrno = WSAGetLastError(); /* Don't clobber the real reason we failed */
804	closesocket(s);
805	QSOCKET_LOCK(tcb);
806	sofree(pData, so);
807	QSOCKET_UNLOCK(tcb);
808	/* Restore the real errno */
809	WSASetLastError(tmperrno);
810	#else
811	int tmperrno = errno; /* Don't clobber the real reason we failed */
812	close(s);
813	QSOCKET_LOCK(tcb);
814	sofree(pData, so);
815	QSOCKET_UNLOCK(tcb);
816	/* Restore the real errno */
817	errno = tmperrno;
818	#endif
819	return NULL;
820	}
821	fd_nonblock(s);
822	setsockopt(s, SOL_SOCKET, SO_OOBINLINE,(char *)&opt, sizeof(int));
823
824	getsockname(s,(struct sockaddr *)&addr,&addrlen);
825	so->so_fport = addr.sin_port;
826	/* set socket buffers */
827	opt = pData->socket_rcv;
828	status = setsockopt(s, SOL_SOCKET, SO_RCVBUF, (char *)&opt, sizeof(int));
829	if (status < 0)
830	{
831	LogRel(("NAT: Error(%d) while setting RCV capacity to (%d)\n", errno, opt));
832	goto no_sockopt;
833	}
834	opt = pData->socket_snd;
835	status = setsockopt(s, SOL_SOCKET, SO_SNDBUF, (char *)&opt, sizeof(int));
836	if (status < 0)
837	{
838	LogRel(("NAT: Error(%d) while setting SND capacity to (%d)\n", errno, opt));
839	goto no_sockopt;
840	}
841	no_sockopt:
842	if (addr.sin_addr.s_addr == 0 \|\| addr.sin_addr.s_addr == loopback_addr.s_addr)
843	so->so_faddr = alias_addr;
844	else
845	so->so_faddr = addr.sin_addr;
846
847	so->s = s;
848	SOCKET_UNLOCK(so);
849	return so;
850	}
851
852	/*
853	* Data is available in so_rcv
854	* Just write() the data to the socket
855	* XXX not yet...
856	*/
857	void
858	sorwakeup(struct socket *so)
859	{
860	#if 0
861	sowrite(so);
862	FD_CLR(so->s,&writefds);
863	#endif
864	}
865
866	/*
867	* Data has been freed in so_snd
868	* We have room for a read() if we want to
869	* For now, don't read, it'll be done in the main loop
870	*/
871	void
872	sowwakeup(struct socket *so)
873	{
874	}
875
876	/*
877	* Various session state calls
878	* XXX Should be #define's
879	* The socket state stuff needs work, these often get call 2 or 3
880	* times each when only 1 was needed
881	*/
882	void
883	soisfconnecting(struct socket *so)
884	{
885	so->so_state &= ~(SS_NOFDREF\|SS_ISFCONNECTED\|SS_FCANTRCVMORE\|
886	SS_FCANTSENDMORE\|SS_FWDRAIN);
887	so->so_state \|= SS_ISFCONNECTING; /* Clobber other states */
888	}
889
890	void
891	soisfconnected(struct socket *so)
892	{
893	so->so_state &= ~(SS_ISFCONNECTING\|SS_FWDRAIN\|SS_NOFDREF);
894	so->so_state \|= SS_ISFCONNECTED; /* Clobber other states */
895	}
896
897	void
898	sofcantrcvmore(struct socket *so)
899	{
900	if ((so->so_state & SS_NOFDREF) == 0)
901	{
902	shutdown(so->s, 0);
903	}
904	so->so_state &= ~(SS_ISFCONNECTING);
905	if (so->so_state & SS_FCANTSENDMORE)
906	so->so_state = SS_NOFDREF; /* Don't select it */
907	/* XXX close() here as well? */
908	else
909	so->so_state \|= SS_FCANTRCVMORE;
910	}
911
912	void
913	sofcantsendmore(struct socket *so)
914	{
915	if ((so->so_state & SS_NOFDREF) == 0)
916	shutdown(so->s, 1); /* send FIN to fhost */
917
918	so->so_state &= ~(SS_ISFCONNECTING);
919	if (so->so_state & SS_FCANTRCVMORE)
920	so->so_state = SS_NOFDREF; /* as above */
921	else
922	so->so_state \|= SS_FCANTSENDMORE;
923	}
924
925	void
926	soisfdisconnected(struct socket *so)
927	{
928	#if 0
929	so->so_state &= ~(SS_ISFCONNECTING\|SS_ISFCONNECTED);
930	close(so->s);
931	so->so_state = SS_ISFDISCONNECTED;
932	/*
933	* XXX Do nothing ... ?
934	*/
935	#endif
936	}
937
938	/*
939	* Set write drain mode
940	* Set CANTSENDMORE once all data has been write()n
941	*/
942	void
943	sofwdrain(struct socket *so)
944	{
945	if (so->so_rcv.sb_cc)
946	so->so_state \|= SS_FWDRAIN;
947	else
948	sofcantsendmore(so);
949	}
950
951	static void
952	send_icmp_to_guest(PNATState pData, char buff, size_t len, struct socket so, const struct sockaddr_in *addr)
953	{
954	struct ip *ip;
955	uint32_t dst, src;
956	char ip_copy[256];
957	struct icmp *icp;
958	int old_ip_len = 0;
959	int hlen, original_hlen = 0;
960	struct mbuf *m;
961	struct icmp_msg *icm;
962	uint8_t proto;
963	int type = 0;
964
965	ip = (struct ip *)buff;
966	hlen = (ip->ip_hl << 2);
967	icp = (struct icmp )((char )ip + hlen);
968
969	Log(("ICMP:received msg(t:%d, c:%d)\n", icp->icmp_type, icp->icmp_code));
970	if ( icp->icmp_type != ICMP_ECHOREPLY
971	&& icp->icmp_type != ICMP_TIMXCEED
972	&& icp->icmp_type != ICMP_UNREACH)
973	{
974	return;
975	}
976
977	type = icp->icmp_type;
978	if ( type == ICMP_TIMXCEED
979	\|\| type == ICMP_UNREACH)
980	{
981	ip = &icp->icmp_ip;
982	DO_ALIAS(&ip->ip_dst);
983	}
984	else
985	{
986	DO_ALIAS(&ip->ip_src);
987	}
988
989	icm = icmp_find_original_mbuf(pData, ip);
990
991	if (icm == NULL)
992	{
993	Log(("NAT: Can't find the corresponding packet for the received ICMP\n"));
994	return;
995	}
996
997	m = icm->im_m;
998	Assert(m != NULL);
999
1000	src = addr->sin_addr.s_addr;
1001
1002	ip = mtod(m, struct ip *);
1003	proto = ip->ip_p;
1004	/* Now ip is pointing on header we've sent from guest */
1005	if ( icp->icmp_type == ICMP_TIMXCEED
1006	\|\| icp->icmp_type == ICMP_UNREACH)
1007	{
1008	old_ip_len = (ip->ip_hl << 2) + 64;
1009	if (old_ip_len > sizeof(ip_copy))
1010	old_ip_len = sizeof(ip_copy);
1011	memcpy(ip_copy, ip, old_ip_len);
1012	}
1013
1014	/* source address from original IP packet*/
1015	dst = ip->ip_src.s_addr;
1016
1017	/* overide ther tail of old packet */
1018	ip = mtod(m, struct ip ); / ip is from mbuf we've overrided */
1019	original_hlen = ip->ip_hl << 2;
1020	/* saves original ip header and options */
1021	memcpy(m->m_data + original_hlen, buff + hlen, len - hlen);
1022	m->m_len = len - hlen + original_hlen;
1023	ip->ip_len = m->m_len;
1024	ip->ip_p = IPPROTO_ICMP; /* the original package could be whatever, but we're response via ICMP*/
1025
1026	icp = (struct icmp )((char )ip + (ip->ip_hl << 2));
1027	type = icp->icmp_type;
1028	if ( type == ICMP_TIMXCEED
1029	\|\| type == ICMP_UNREACH)
1030	{
1031	/* according RFC 793 error messages required copy of initial IP header + 64 bit */
1032	memcpy(&icp->icmp_ip, ip_copy, old_ip_len);
1033	ip->ip_tos = ((ip->ip_tos & 0x1E) \| 0xC0); /* high priority for errors */
1034	}
1035
1036	ip->ip_src.s_addr = src;
1037	ip->ip_dst.s_addr = dst;
1038	icmp_reflect(pData, m);
1039	LIST_REMOVE(icm, im_list);
1040	/* Don't call m_free here*/
1041
1042	if ( type == ICMP_TIMXCEED
1043	\|\| type == ICMP_UNREACH)
1044	{
1045	icm->im_so->so_m = NULL;
1046	switch (proto)
1047	{
1048	case IPPROTO_UDP:
1049	/XXX: so->so_m already freed so we shouldn't call sofree /
1050	udp_detach(pData, icm->im_so);
1051	break;
1052	case IPPROTO_TCP:
1053	/close tcp should be here /
1054	break;
1055	default:
1056	/* do nothing */
1057	break;
1058	}
1059	}
1060	RTMemFree(icm);
1061	}
1062
1063	#ifdef RT_OS_WINDOWS
1064	static void
1065	sorecvfrom_icmp_win(PNATState pData, struct socket *so)
1066	{
1067	int len;
1068	int i;
1069	struct ip *ip;
1070	struct mbuf *m;
1071	struct icmp *icp;
1072	struct icmp_msg *icm;
1073	struct ip ip_broken; / ICMP returns header + 64 bit of packet */
1074	uint32_t src;
1075	ICMP_ECHO_REPLY *icr;
1076	int hlen = 0;
1077	int data_len = 0;
1078	int nbytes = 0;
1079	u_char code = ~0;
1080
1081	len = pData->pfIcmpParseReplies(pData->pvIcmpBuffer, pData->szIcmpBuffer);
1082	if (len < 0)
1083	{
1084	LogRel(("NAT: Error (%d) occurred on ICMP receiving\n", GetLastError()));
1085	return;
1086	}
1087	if (len == 0)
1088	return; /* no error */
1089
1090	icr = (ICMP_ECHO_REPLY *)pData->pvIcmpBuffer;
1091	for (i = 0; i < len; ++i)
1092	{
1093	switch(icr[i].Status)
1094	{
1095	case IP_DEST_HOST_UNREACHABLE:
1096	code = (code != ~0 ? code : ICMP_UNREACH_HOST);
1097	case IP_DEST_NET_UNREACHABLE:
1098	code = (code != ~0 ? code : ICMP_UNREACH_NET);
1099	case IP_DEST_PROT_UNREACHABLE:
1100	code = (code != ~0 ? code : ICMP_UNREACH_PROTOCOL);
1101	/* UNREACH error inject here */
1102	case IP_DEST_PORT_UNREACHABLE:
1103	code = (code != ~0 ? code : ICMP_UNREACH_PORT);
1104	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, "Error occurred!!!");
1105	so->so_m = NULL;
1106	break;
1107	case IP_SUCCESS: /* echo replied */
1108	m = m_get(pData);
1109	m->m_data += if_maxlinkhdr;
1110	ip = mtod(m, struct ip *);
1111	ip->ip_src.s_addr = icr[i].Address;
1112	DO_ALIAS(&ip->ip_src);
1113	ip->ip_p = IPPROTO_ICMP;
1114	ip->ip_dst.s_addr = so->so_laddr.s_addr; /XXX: still the hack/
1115	data_len = sizeof(struct ip);
1116	ip->ip_hl = data_len >> 2; /* requiered for icmp_reflect, no IP options */
1117	ip->ip_ttl = icr[i].Options.Ttl;
1118
1119	icp = (struct icmp )&ip[1]; / no options */
1120	icp->icmp_type = ICMP_ECHOREPLY;
1121	icp->icmp_code = 0;
1122	icp->icmp_id = so->so_icmp_id;
1123	icp->icmp_seq = so->so_icmp_seq;
1124
1125	data_len += ICMP_MINLEN;
1126
1127	nbytes = (data_len + icr[i].DataSize > m->m_size? m->m_size - data_len: icr[i].DataSize);
1128	memcpy(icp->icmp_data, icr[i].Data, nbytes);
1129
1130	data_len += icr[i].DataSize;
1131
1132	ip->ip_len = data_len;
1133	m->m_len = ip->ip_len;
1134
1135	icmp_reflect(pData, m);
1136	break;
1137	case IP_TTL_EXPIRED_TRANSIT: /* TTL expired */
1138
1139	ip_broken = icr[i].Data;
1140	icm = icmp_find_original_mbuf(pData, ip_broken);
1141	if (icm == NULL) {
1142	Log(("ICMP: can't find original package (first double word %x)\n", (uint32_t )ip_broken));
1143	return;
1144	}
1145	m = icm->im_m;
1146	ip = mtod(m, struct ip *);
1147	ip->ip_ttl = icr[i].Options.Ttl;
1148	src = ip->ip_src.s_addr;
1149	ip->ip_dst.s_addr = src;
1150	ip->ip_dst.s_addr = icr[i].Address;
1151
1152	hlen = (ip->ip_hl << 2);
1153	icp = (struct icmp )((char )ip + hlen);
1154	ip_broken->ip_src.s_addr = src; /it packet sent from host not from guest/
1155	data_len = (ip_broken->ip_hl << 2) + 64;
1156
1157	nbytes =(hlen + ICMP_MINLEN + data_len > m->m_size? m->m_size - (hlen + ICMP_MINLEN): data_len);
1158	memcpy(icp->icmp_data, ip_broken, nbytes);
1159	icmp_reflect(pData, m);
1160	break;
1161	default:
1162	Log(("ICMP(default): message with Status: %x was received from %x\n", icr[i].Status, icr[i].Address));
1163	break;
1164	}
1165	}
1166	}
1167	#else /* RT_OS_WINDOWS */
1168	static void sorecvfrom_icmp_unix(PNATState pData, struct socket *so)
1169	{
1170	struct sockaddr_in addr;
1171	socklen_t addrlen = sizeof(struct sockaddr_in);
1172	char buff[1500];
1173	int len;
1174	len = recvfrom(so->s, buff, 1500, 0,
1175	(struct sockaddr *)&addr, &addrlen);
1176	/* XXX Check if reply is "correct"? */
1177
1178	if (len == -1 \|\| len == 0)
1179	{
1180	u_char code = ICMP_UNREACH_PORT;
1181
1182	if (errno == EHOSTUNREACH)
1183	code = ICMP_UNREACH_HOST;
1184	else if(errno == ENETUNREACH)
1185	code = ICMP_UNREACH_NET;
1186
1187	DEBUG_MISC((dfd," udp icmp rx errno = %d-%s\n",
1188	errno, strerror(errno)));
1189	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
1190	so->so_m = NULL;
1191	}
1192	else
1193	{
1194	send_icmp_to_guest(pData, buff, len, so, &addr);
1195	}
1196	}
1197	#endif /* !RT_OS_WINDOWS */

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

source: vbox/trunk/src/VBox/Devices/Network/slirp/socket.c@ 20958

Download in other formats: