socket.cpp@ 49971

Last change on this file since 49971 was 48935, checked in by vboxsync, 11 years ago
Runtime: Whitespace and svn:keyword cleanups by scm.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 59.8 KB

Line
1	/* $Id: socket.cpp 48935 2013-10-07 21:19:37Z vboxsync $ */
2	/** @file
3	* IPRT - Network Sockets.
4	*/
5
6	/*
7	* Copyright (C) 2006-2013 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* The contents of this file may alternatively be used under the terms
18	* of the Common Development and Distribution License Version 1.0
19	* (CDDL) only, as it comes in the "COPYING.CDDL" file of the
20	* VirtualBox OSE distribution, in which case the provisions of the
21	* CDDL are applicable instead of those of the GPL.
22	*
23	* You may elect to license modified versions of this file under the
24	* terms and conditions of either the GPL or the CDDL or both.
25	*/
26
27
28	/*******************************************************************************
29	* Header Files *
30	*******************************************************************************/
31	#ifdef RT_OS_WINDOWS
32	# include <winsock2.h>
33	# include <ws2tcpip.h>
34	#else /* !RT_OS_WINDOWS */
35	# include <errno.h>
36	# include <sys/select.h>
37	# include <sys/stat.h>
38	# include <sys/socket.h>
39	# include <netinet/in.h>
40	# include <netinet/tcp.h>
41	# include <arpa/inet.h>
42	# ifdef IPRT_WITH_TCPIP_V6
43	# include <netinet6/in6.h>
44	# endif
45	# include <sys/un.h>
46	# include <netdb.h>
47	# include <unistd.h>
48	# include <fcntl.h>
49	# include <sys/uio.h>
50	#endif /* !RT_OS_WINDOWS */
51	#include <limits.h>
52
53	#include "internal/iprt.h"
54	#include <iprt/socket.h>
55
56	#include <iprt/alloca.h>
57	#include <iprt/asm.h>
58	#include <iprt/assert.h>
59	#include <iprt/ctype.h>
60	#include <iprt/err.h>
61	#include <iprt/mempool.h>
62	#include <iprt/poll.h>
63	#include <iprt/string.h>
64	#include <iprt/thread.h>
65	#include <iprt/time.h>
66	#include <iprt/mem.h>
67	#include <iprt/sg.h>
68	#include <iprt/log.h>
69
70	#include "internal/magics.h"
71	#include "internal/socket.h"
72	#include "internal/string.h"
73
74
75	/*******************************************************************************
76	* Defined Constants And Macros *
77	*******************************************************************************/
78	/* non-standard linux stuff (it seems). */
79	#ifndef MSG_NOSIGNAL
80	# define MSG_NOSIGNAL 0
81	#endif
82
83	/* Windows has different names for SHUT_XXX. */
84	#ifndef SHUT_RDWR
85	# ifdef SD_BOTH
86	# define SHUT_RDWR SD_BOTH
87	# else
88	# define SHUT_RDWR 2
89	# endif
90	#endif
91	#ifndef SHUT_WR
92	# ifdef SD_SEND
93	# define SHUT_WR SD_SEND
94	# else
95	# define SHUT_WR 1
96	# endif
97	#endif
98	#ifndef SHUT_RD
99	# ifdef SD_RECEIVE
100	# define SHUT_RD SD_RECEIVE
101	# else
102	# define SHUT_RD 0
103	# endif
104	#endif
105
106	/* fixup backlevel OSes. */
107	#if defined(RT_OS_OS2) \|\| defined(RT_OS_WINDOWS)
108	# define socklen_t int
109	#endif
110
111	/** How many pending connection. */
112	#define RTTCP_SERVER_BACKLOG 10
113
114	/* Limit read and write sizes on Windows and OS/2. */
115	#ifdef RT_OS_WINDOWS
116	# define RTSOCKET_MAX_WRITE (INT_MAX / 2)
117	# define RTSOCKET_MAX_READ (INT_MAX / 2)
118	#elif defined(RT_OS_OS2)
119	# define RTSOCKET_MAX_WRITE 0x10000
120	# define RTSOCKET_MAX_READ 0x10000
121	#endif
122
123
124	/*******************************************************************************
125	* Structures and Typedefs *
126	*******************************************************************************/
127	/**
128	* Socket handle data.
129	*
130	* This is mainly required for implementing RTPollSet on Windows.
131	*/
132	typedef struct RTSOCKETINT
133	{
134	/** Magic number (RTSOCKET_MAGIC). */
135	uint32_t u32Magic;
136	/** Exclusive user count.
137	* This is used to prevent two threads from accessing the handle concurrently.
138	* It can be higher than 1 if this handle is reference multiple times in a
139	* polling set (Windows). */
140	uint32_t volatile cUsers;
141	/** The native socket handle. */
142	RTSOCKETNATIVE hNative;
143	/** Indicates whether the handle has been closed or not. */
144	bool volatile fClosed;
145	/** Indicates whether the socket is operating in blocking or non-blocking mode
146	* currently. */
147	bool fBlocking;
148	#if defined(RT_OS_WINDOWS) \|\| defined(RT_OS_OS2)
149	/** The pollset currently polling this socket. This is NIL if no one is
150	* polling. */
151	RTPOLLSET hPollSet;
152	#endif
153	#ifdef RT_OS_WINDOWS
154	/** The event semaphore we've associated with the socket handle.
155	* This is WSA_INVALID_EVENT if not done. */
156	WSAEVENT hEvent;
157	/** The events we're polling for. */
158	uint32_t fPollEvts;
159	/** The events we're currently subscribing to with WSAEventSelect.
160	* This is ZERO if we're currently not subscribing to anything. */
161	uint32_t fSubscribedEvts;
162	/** Saved events which are only posted once. */
163	uint32_t fEventsSaved;
164	#endif /* RT_OS_WINDOWS */
165	} RTSOCKETINT;
166
167
168	/**
169	* Address union used internally for things like getpeername and getsockname.
170	*/
171	typedef union RTSOCKADDRUNION
172	{
173	struct sockaddr Addr;
174	struct sockaddr_in IPv4;
175	#ifdef IPRT_WITH_TCPIP_V6
176	struct sockaddr_in6 IPv6;
177	#endif
178	} RTSOCKADDRUNION;
179
180
181	/**
182	* Get the last error as an iprt status code.
183	*
184	* @returns IPRT status code.
185	*/
186	DECLINLINE(int) rtSocketError(void)
187	{
188	#ifdef RT_OS_WINDOWS
189	return RTErrConvertFromWin32(WSAGetLastError());
190	#else
191	return RTErrConvertFromErrno(errno);
192	#endif
193	}
194
195
196	/**
197	* Resets the last error.
198	*/
199	DECLINLINE(void) rtSocketErrorReset(void)
200	{
201	#ifdef RT_OS_WINDOWS
202	WSASetLastError(0);
203	#else
204	errno = 0;
205	#endif
206	}
207
208
209	/**
210	* Get the last resolver error as an iprt status code.
211	*
212	* @returns iprt status code.
213	*/
214	int rtSocketResolverError(void)
215	{
216	#ifdef RT_OS_WINDOWS
217	return RTErrConvertFromWin32(WSAGetLastError());
218	#else
219	switch (h_errno)
220	{
221	case HOST_NOT_FOUND:
222	return VERR_NET_HOST_NOT_FOUND;
223	case NO_DATA:
224	return VERR_NET_ADDRESS_NOT_AVAILABLE;
225	case NO_RECOVERY:
226	return VERR_IO_GEN_FAILURE;
227	case TRY_AGAIN:
228	return VERR_TRY_AGAIN;
229
230	default:
231	return VERR_UNRESOLVED_ERROR;
232	}
233	#endif
234	}
235
236
237	/**
238	* Converts from a native socket address to a generic IPRT network address.
239	*
240	* @returns IPRT status code.
241	* @param pSrc The source address.
242	* @param cbSrc The size of the source address.
243	* @param pAddr Where to return the generic IPRT network
244	* address.
245	*/
246	static int rtSocketNetAddrFromAddr(RTSOCKADDRUNION const *pSrc, size_t cbSrc, PRTNETADDR pAddr)
247	{
248	/*
249	* Convert the address.
250	*/
251	if ( cbSrc == sizeof(struct sockaddr_in)
252	&& pSrc->Addr.sa_family == AF_INET)
253	{
254	RT_ZERO(*pAddr);
255	pAddr->enmType = RTNETADDRTYPE_IPV4;
256	pAddr->uPort = RT_N2H_U16(pSrc->IPv4.sin_port);
257	pAddr->uAddr.IPv4.u = pSrc->IPv4.sin_addr.s_addr;
258	}
259	#ifdef IPRT_WITH_TCPIP_V6
260	else if ( cbSrc == sizeof(struct sockaddr_in6)
261	&& pSrc->Addr.sa_family == AF_INET6)
262	{
263	RT_ZERO(*pAddr);
264	pAddr->enmType = RTNETADDRTYPE_IPV6;
265	pAddr->uPort = RT_N2H_U16(pSrc->IPv6.sin6_port);
266	pAddr->uAddr.IPv6.au32[0] = pSrc->IPv6.sin6_addr.s6_addr32[0];
267	pAddr->uAddr.IPv6.au32[1] = pSrc->IPv6.sin6_addr.s6_addr32[1];
268	pAddr->uAddr.IPv6.au32[2] = pSrc->IPv6.sin6_addr.s6_addr32[2];
269	pAddr->uAddr.IPv6.au32[3] = pSrc->IPv6.sin6_addr.s6_addr32[3];
270	}
271	#endif
272	else
273	return VERR_NET_ADDRESS_FAMILY_NOT_SUPPORTED;
274	return VINF_SUCCESS;
275	}
276
277
278	/**
279	* Converts from a generic IPRT network address to a native socket address.
280	*
281	* @returns IPRT status code.
282	* @param pAddr Pointer to the generic IPRT network address.
283	* @param pDst The source address.
284	* @param cbSrc The size of the source address.
285	* @param pcbAddr Where to store the size of the returned address.
286	* Optional
287	*/
288	static int rtSocketAddrFromNetAddr(PCRTNETADDR pAddr, RTSOCKADDRUNION pDst, size_t cbDst, int pcbAddr)
289	{
290	RT_BZERO(pDst, cbDst);
291	if ( pAddr->enmType == RTNETADDRTYPE_IPV4
292	&& cbDst >= sizeof(struct sockaddr_in))
293	{
294	pDst->Addr.sa_family = AF_INET;
295	pDst->IPv4.sin_port = RT_H2N_U16(pAddr->uPort);
296	pDst->IPv4.sin_addr.s_addr = pAddr->uAddr.IPv4.u;
297	if (pcbAddr)
298	*pcbAddr = sizeof(pDst->IPv4);
299	}
300	#ifdef IPRT_WITH_TCPIP_V6
301	else if ( pAddr->enmType == RTNETADDRTYPE_IPV6
302	&& cbDst >= sizeof(struct sockaddr_in6))
303	{
304	pDst->Addr.sa_family = AF_INET6;
305	pDst->IPv6.sin6_port = RT_H2N_U16(pAddr->uPort);
306	pSrc->IPv6.sin6_addr.s6_addr32[0] = pAddr->uAddr.IPv6.au32[0];
307	pSrc->IPv6.sin6_addr.s6_addr32[1] = pAddr->uAddr.IPv6.au32[1];
308	pSrc->IPv6.sin6_addr.s6_addr32[2] = pAddr->uAddr.IPv6.au32[2];
309	pSrc->IPv6.sin6_addr.s6_addr32[3] = pAddr->uAddr.IPv6.au32[3];
310	if (pcbAddr)
311	*pcbAddr = sizeof(pDst->IPv6);
312	}
313	#endif
314	else
315	return VERR_NET_ADDRESS_FAMILY_NOT_SUPPORTED;
316	return VINF_SUCCESS;
317	}
318
319
320	/**
321	* Tries to lock the socket for exclusive usage by the calling thread.
322	*
323	* Call rtSocketUnlock() to unlock.
324	*
325	* @returns @c true if locked, @c false if not.
326	* @param pThis The socket structure.
327	*/
328	DECLINLINE(bool) rtSocketTryLock(RTSOCKETINT *pThis)
329	{
330	return ASMAtomicCmpXchgU32(&pThis->cUsers, 1, 0);
331	}
332
333
334	/**
335	* Unlocks the socket.
336	*
337	* @param pThis The socket structure.
338	*/
339	DECLINLINE(void) rtSocketUnlock(RTSOCKETINT *pThis)
340	{
341	ASMAtomicCmpXchgU32(&pThis->cUsers, 0, 1);
342	}
343
344
345	/**
346	* The slow path of rtSocketSwitchBlockingMode that does the actual switching.
347	*
348	* @returns IPRT status code.
349	* @param pThis The socket structure.
350	* @param fBlocking The desired mode of operation.
351	* @remarks Do not call directly.
352	*/
353	static int rtSocketSwitchBlockingModeSlow(RTSOCKETINT *pThis, bool fBlocking)
354	{
355	#ifdef RT_OS_WINDOWS
356	u_long uBlocking = fBlocking ? 0 : 1;
357	if (ioctlsocket(pThis->hNative, FIONBIO, &uBlocking))
358	return rtSocketError();
359
360	#else
361	int fFlags = fcntl(pThis->hNative, F_GETFL, 0);
362	if (fFlags == -1)
363	return rtSocketError();
364
365	if (fBlocking)
366	fFlags &= ~O_NONBLOCK;
367	else
368	fFlags \|= O_NONBLOCK;
369	if (fcntl(pThis->hNative, F_SETFL, fFlags) == -1)
370	return rtSocketError();
371	#endif
372
373	pThis->fBlocking = fBlocking;
374	return VINF_SUCCESS;
375	}
376
377
378	/**
379	* Switches the socket to the desired blocking mode if necessary.
380	*
381	* The socket must be locked.
382	*
383	* @returns IPRT status code.
384	* @param pThis The socket structure.
385	* @param fBlocking The desired mode of operation.
386	*/
387	DECLINLINE(int) rtSocketSwitchBlockingMode(RTSOCKETINT *pThis, bool fBlocking)
388	{
389	if (pThis->fBlocking != fBlocking)
390	return rtSocketSwitchBlockingModeSlow(pThis, fBlocking);
391	return VINF_SUCCESS;
392	}
393
394
395	/**
396	* Creates an IPRT socket handle for a native one.
397	*
398	* @returns IPRT status code.
399	* @param ppSocket Where to return the IPRT socket handle.
400	* @param hNative The native handle.
401	*/
402	int rtSocketCreateForNative(RTSOCKETINT **ppSocket, RTSOCKETNATIVE hNative)
403	{
404	RTSOCKETINT pThis = (RTSOCKETINT )RTMemPoolAlloc(RTMEMPOOL_DEFAULT, sizeof(*pThis));
405	if (!pThis)
406	return VERR_NO_MEMORY;
407	pThis->u32Magic = RTSOCKET_MAGIC;
408	pThis->cUsers = 0;
409	pThis->hNative = hNative;
410	pThis->fClosed = false;
411	pThis->fBlocking = true;
412	#if defined(RT_OS_WINDOWS) \|\| defined(RT_OS_OS2)
413	pThis->hPollSet = NIL_RTPOLLSET;
414	#endif
415	#ifdef RT_OS_WINDOWS
416	pThis->hEvent = WSA_INVALID_EVENT;
417	pThis->fPollEvts = 0;
418	pThis->fSubscribedEvts = 0;
419	#endif
420	*ppSocket = pThis;
421	return VINF_SUCCESS;
422	}
423
424
425	RTDECL(int) RTSocketFromNative(PRTSOCKET phSocket, RTHCINTPTR uNative)
426	{
427	AssertReturn(uNative != NIL_RTSOCKETNATIVE, VERR_INVALID_PARAMETER);
428	#ifndef RT_OS_WINDOWS
429	AssertReturn(uNative >= 0, VERR_INVALID_PARAMETER);
430	#endif
431	AssertPtrReturn(phSocket, VERR_INVALID_POINTER);
432	return rtSocketCreateForNative(phSocket, uNative);
433	}
434
435
436	/**
437	* Wrapper around socket().
438	*
439	* @returns IPRT status code.
440	* @param phSocket Where to store the handle to the socket on
441	* success.
442	* @param iDomain The protocol family (PF_XXX).
443	* @param iType The socket type (SOCK_XXX).
444	* @param iProtocol Socket parameter, usually 0.
445	*/
446	int rtSocketCreate(PRTSOCKET phSocket, int iDomain, int iType, int iProtocol)
447	{
448	/*
449	* Create the socket.
450	*/
451	RTSOCKETNATIVE hNative = socket(iDomain, iType, iProtocol);
452	if (hNative == NIL_RTSOCKETNATIVE)
453	return rtSocketError();
454
455	/*
456	* Wrap it.
457	*/
458	int rc = rtSocketCreateForNative(phSocket, hNative);
459	if (RT_FAILURE(rc))
460	{
461	#ifdef RT_OS_WINDOWS
462	closesocket(hNative);
463	#else
464	close(hNative);
465	#endif
466	}
467	return rc;
468	}
469
470
471	RTDECL(uint32_t) RTSocketRetain(RTSOCKET hSocket)
472	{
473	RTSOCKETINT *pThis = hSocket;
474	AssertPtrReturn(pThis, UINT32_MAX);
475	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, UINT32_MAX);
476	return RTMemPoolRetain(pThis);
477	}
478
479
480	/**
481	* Worker for RTSocketRelease and RTSocketClose.
482	*
483	* @returns IPRT status code.
484	* @param pThis The socket handle instance data.
485	* @param fDestroy Whether we're reaching ref count zero.
486	*/
487	static int rtSocketCloseIt(RTSOCKETINT *pThis, bool fDestroy)
488	{
489	/*
490	* Invalidate the handle structure on destroy.
491	*/
492	if (fDestroy)
493	{
494	Assert(ASMAtomicReadU32(&pThis->u32Magic) == RTSOCKET_MAGIC);
495	ASMAtomicWriteU32(&pThis->u32Magic, RTSOCKET_MAGIC_DEAD);
496	}
497
498	int rc = VINF_SUCCESS;
499	if (ASMAtomicCmpXchgBool(&pThis->fClosed, true, false))
500	{
501	/*
502	* Close the native handle.
503	*/
504	RTSOCKETNATIVE hNative = pThis->hNative;
505	if (hNative != NIL_RTSOCKETNATIVE)
506	{
507	pThis->hNative = NIL_RTSOCKETNATIVE;
508
509	#ifdef RT_OS_WINDOWS
510	if (closesocket(hNative))
511	#else
512	if (close(hNative))
513	#endif
514	{
515	rc = rtSocketError();
516	#ifdef RT_OS_WINDOWS
517	AssertMsgFailed(("\"%s\": closesocket(%p) -> %Rrc\n", (uintptr_t)hNative, rc));
518	#else
519	AssertMsgFailed(("\"%s\": close(%d) -> %Rrc\n", hNative, rc));
520	#endif
521	}
522	}
523
524	#ifdef RT_OS_WINDOWS
525	/*
526	* Close the event.
527	*/
528	WSAEVENT hEvent = pThis->hEvent;
529	if (hEvent == WSA_INVALID_EVENT)
530	{
531	pThis->hEvent = WSA_INVALID_EVENT;
532	WSACloseEvent(hEvent);
533	}
534	#endif
535	}
536
537	return rc;
538	}
539
540
541	RTDECL(uint32_t) RTSocketRelease(RTSOCKET hSocket)
542	{
543	RTSOCKETINT *pThis = hSocket;
544	if (pThis == NIL_RTSOCKET)
545	return 0;
546	AssertPtrReturn(pThis, UINT32_MAX);
547	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, UINT32_MAX);
548
549	/* get the refcount without killing it... */
550	uint32_t cRefs = RTMemPoolRefCount(pThis);
551	AssertReturn(cRefs != UINT32_MAX, UINT32_MAX);
552	if (cRefs == 1)
553	rtSocketCloseIt(pThis, true);
554
555	return RTMemPoolRelease(RTMEMPOOL_DEFAULT, pThis);
556	}
557
558
559	RTDECL(int) RTSocketClose(RTSOCKET hSocket)
560	{
561	RTSOCKETINT *pThis = hSocket;
562	if (pThis == NIL_RTSOCKET)
563	return VINF_SUCCESS;
564	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
565	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
566
567	uint32_t cRefs = RTMemPoolRefCount(pThis);
568	AssertReturn(cRefs != UINT32_MAX, UINT32_MAX);
569
570	int rc = rtSocketCloseIt(pThis, cRefs == 1);
571
572	RTMemPoolRelease(RTMEMPOOL_DEFAULT, pThis);
573	return rc;
574	}
575
576
577	RTDECL(RTHCUINTPTR) RTSocketToNative(RTSOCKET hSocket)
578	{
579	RTSOCKETINT *pThis = hSocket;
580	AssertPtrReturn(pThis, RTHCUINTPTR_MAX);
581	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, RTHCUINTPTR_MAX);
582	return (RTHCUINTPTR)pThis->hNative;
583	}
584
585
586	RTDECL(int) RTSocketSetInheritance(RTSOCKET hSocket, bool fInheritable)
587	{
588	RTSOCKETINT *pThis = hSocket;
589	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
590	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
591	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
592
593	int rc = VINF_SUCCESS;
594	#ifdef RT_OS_WINDOWS
595	if (!SetHandleInformation((HANDLE)pThis->hNative, HANDLE_FLAG_INHERIT, fInheritable ? HANDLE_FLAG_INHERIT : 0))
596	rc = RTErrConvertFromWin32(GetLastError());
597	#else
598	if (fcntl(pThis->hNative, F_SETFD, fInheritable ? 0 : FD_CLOEXEC) < 0)
599	rc = RTErrConvertFromErrno(errno);
600	#endif
601
602	return rc;
603	}
604
605
606	static bool rtSocketIsIPv4Numerical(const char *pszAddress, PRTNETADDRIPV4 pAddr)
607	{
608
609	/* Empty address resolves to the INADDR_ANY address (good for bind). */
610	if (!pszAddress \|\| !*pszAddress)
611	{
612	pAddr->u = INADDR_ANY;
613	return true;
614	}
615
616	/* Four quads? */
617	char psz = (char )pszAddress;
618	for (int i = 0; i < 4; i++)
619	{
620	uint8_t u8;
621	int rc = RTStrToUInt8Ex(psz, &psz, 0, &u8);
622	if (rc != VINF_SUCCESS && rc != VWRN_TRAILING_CHARS)
623	return false;
624	if (*psz != (i < 3 ? '.' : '\0'))
625	return false;
626	psz++;
627
628	pAddr->au8[i] = u8; /* big endian */
629	}
630
631	return true;
632	}
633
634	RTDECL(int) RTSocketParseInetAddress(const char *pszAddress, unsigned uPort, PRTNETADDR pAddr)
635	{
636	int rc;
637
638	/*
639	* Validate input.
640	*/
641	AssertReturn(uPort > 0, VERR_INVALID_PARAMETER);
642	AssertPtrNullReturn(pszAddress, VERR_INVALID_POINTER);
643
644	#ifdef RT_OS_WINDOWS
645	/*
646	* Initialize WinSock and check version.
647	*/
648	WORD wVersionRequested = MAKEWORD(1, 1);
649	WSADATA wsaData;
650	rc = WSAStartup(wVersionRequested, &wsaData);
651	if (wsaData.wVersion != wVersionRequested)
652	{
653	AssertMsgFailed(("Wrong winsock version\n"));
654	return VERR_NOT_SUPPORTED;
655	}
656	#endif
657
658	/*
659	* Resolve the address. Pretty crude at the moment, but we have to make
660	* sure to not ask the NT 4 gethostbyname about an IPv4 address as it may
661	* give a wrong answer.
662	*/
663	/** @todo this only supports IPv4, and IPv6 support needs to be added.
664	* It probably needs to be converted to getaddrinfo(). */
665	RTNETADDRIPV4 IPv4Quad;
666	if (rtSocketIsIPv4Numerical(pszAddress, &IPv4Quad))
667	{
668	Log3(("rtSocketIsIPv4Numerical: %#x (%RTnaipv4)\n", pszAddress, IPv4Quad.u, IPv4Quad));
669	RT_ZERO(*pAddr);
670	pAddr->enmType = RTNETADDRTYPE_IPV4;
671	pAddr->uPort = uPort;
672	pAddr->uAddr.IPv4 = IPv4Quad;
673	return VINF_SUCCESS;
674	}
675
676	struct hostent *pHostEnt;
677	pHostEnt = gethostbyname(pszAddress);
678	if (!pHostEnt)
679	{
680	rc = rtSocketResolverError();
681	AssertMsgFailed(("Could not resolve '%s', rc=%Rrc\n", pszAddress, rc));
682	return rc;
683	}
684
685	if (pHostEnt->h_addrtype == AF_INET)
686	{
687	RT_ZERO(*pAddr);
688	pAddr->enmType = RTNETADDRTYPE_IPV4;
689	pAddr->uPort = uPort;
690	pAddr->uAddr.IPv4.u = ((struct in_addr *)pHostEnt->h_addr)->s_addr;
691	Log3(("gethostbyname: %s -> %#x (%RTnaipv4)\n", pszAddress, pAddr->uAddr.IPv4.u, pAddr->uAddr.IPv4));
692	}
693	else
694	return VERR_NET_ADDRESS_FAMILY_NOT_SUPPORTED;
695
696	return VINF_SUCCESS;
697	}
698
699
700	/*
701	* New function to allow both ipv4 and ipv6 addresses to be resolved.
702	* Breaks compatibility with windows before 2000.
703	*/
704	RTDECL(int) RTSocketQueryAddressStr(const char pszHost, char pszResult, size_t *pcbResult, PRTNETADDRTYPE penmAddrType)
705	{
706	AssertPtrReturn(pszHost, VERR_INVALID_POINTER);
707	AssertPtrReturn(pcbResult, VERR_INVALID_POINTER);
708	AssertPtrNullReturn(penmAddrType, VERR_INVALID_POINTER);
709	AssertPtrNullReturn(pszResult, VERR_INVALID_POINTER);
710
711	#if defined(RT_OS_OS2) \|\| defined(RT_OS_WINDOWS) /** @todo dynamically resolve the APIs not present in NT4! */
712	return VERR_NOT_SUPPORTED;
713
714	#else
715	int rc;
716	if (*pcbResult < 16)
717	return VERR_NET_ADDRESS_NOT_AVAILABLE;
718
719	/* Setup the hint. */
720	struct addrinfo grHints;
721	RT_ZERO(grHints);
722	grHints.ai_socktype = 0;
723	grHints.ai_flags = 0;
724	grHints.ai_protocol = 0;
725	grHints.ai_family = AF_UNSPEC;
726	if (penmAddrType)
727	{
728	switch (*penmAddrType)
729	{
730	case RTNETADDRTYPE_INVALID:
731	/grHints.ai_family = AF_UNSPEC;/
732	break;
733	case RTNETADDRTYPE_IPV4:
734	grHints.ai_family = AF_INET;
735	break;
736	case RTNETADDRTYPE_IPV6:
737	grHints.ai_family = AF_INET6;
738	break;
739	default:
740	AssertFailedReturn(VERR_INVALID_PARAMETER);
741	}
742	}
743
744	# ifdef RT_OS_WINDOWS
745	/*
746	* Winsock2 init
747	*/
748	/** @todo someone should check if we really need 2, 2 here */
749	WORD wVersionRequested = MAKEWORD(2, 2);
750	WSADATA wsaData;
751	rc = WSAStartup(wVersionRequested, &wsaData);
752	if (wsaData.wVersion != wVersionRequested)
753	{
754	AssertMsgFailed(("Wrong winsock version\n"));
755	return VERR_NOT_SUPPORTED;
756	}
757	# endif
758
759	/** @todo r=bird: getaddrinfo and freeaddrinfo breaks the additions on NT4. */
760	struct addrinfo *pgrResults = NULL;
761	rc = getaddrinfo(pszHost, "", &grHints, &pgrResults);
762	if (rc != 0)
763	return VERR_NET_ADDRESS_NOT_AVAILABLE;
764
765	// return data
766	// on multiple matches return only the first one
767
768	if (!pgrResults)
769	return VERR_NET_ADDRESS_NOT_AVAILABLE;
770
771	struct addrinfo const *pgrResult = pgrResults->ai_next;
772	if (!pgrResult)
773	{
774	freeaddrinfo(pgrResults);
775	return VERR_NET_ADDRESS_NOT_AVAILABLE;
776	}
777
778	uint8_t const *pbDummy;
779	RTNETADDRTYPE enmAddrType = RTNETADDRTYPE_INVALID;
780	size_t cchIpAddress;
781	char szIpAddress[48];
782	if (pgrResult->ai_family == AF_INET)
783	{
784	struct sockaddr_in const pgrSa = (struct sockaddr_in const )pgrResult->ai_addr;
785	pbDummy = (uint8_t const *)&pgrSa->sin_addr;
786	cchIpAddress = RTStrPrintf(szIpAddress, sizeof(szIpAddress), "%u.%u.%u.%u",
787	pbDummy[0], pbDummy[1], pbDummy[2], pbDummy[3]);
788	Assert(cchIpAddress >= 7 && cchIpAddress < sizeof(szIpAddress) - 1);
789	enmAddrType = RTNETADDRTYPE_IPV4;
790	rc = VINF_SUCCESS;
791	}
792	else if (pgrResult->ai_family == AF_INET6)
793	{
794	struct sockaddr_in6 const pgrSa6 = (struct sockaddr_in6 const )pgrResult->ai_addr;
795	pbDummy = (uint8_t const *) &pgrSa6->sin6_addr;
796	char szTmp[32+1];
797	size_t cchTmp = RTStrPrintf(szTmp, sizeof(szTmp),
798	"%02x%02x%02x%02x"
799	"%02x%02x%02x%02x"
800	"%02x%02x%02x%02x"
801	"%02x%02x%02x%02x",
802	pbDummy[0], pbDummy[1], pbDummy[2], pbDummy[3],
803	pbDummy[4], pbDummy[5], pbDummy[6], pbDummy[7],
804	pbDummy[8], pbDummy[9], pbDummy[10], pbDummy[11],
805	pbDummy[12], pbDummy[13], pbDummy[14], pbDummy[15]);
806	Assert(cchTmp == 32);
807	rc = rtStrToIpAddr6Str(szTmp, szIpAddress, sizeof(szIpAddress), NULL, 0, true);
808	if (RT_SUCCESS(rc))
809	cchIpAddress = strlen(szIpAddress);
810	else
811	{
812	szIpAddress[0] = '\0';
813	cchIpAddress = 0;
814	}
815	enmAddrType = RTNETADDRTYPE_IPV6;
816	}
817	else
818	{
819	rc = VERR_NET_ADDRESS_NOT_AVAILABLE;
820	szIpAddress[0] = '\0';
821	cchIpAddress = 0;
822	}
823	freeaddrinfo(pgrResults);
824
825	/*
826	* Copy out the result.
827	*/
828	size_t const cbResult = *pcbResult;
829	*pcbResult = cchIpAddress + 1;
830	if (cchIpAddress < cbResult)
831	memcpy(pszResult, szIpAddress, cchIpAddress + 1);
832	else
833	{
834	RT_BZERO(pszResult, cbResult);
835	if (RT_SUCCESS(rc))
836	rc = VERR_BUFFER_OVERFLOW;
837	}
838	if (penmAddrType && RT_SUCCESS(rc))
839	*penmAddrType = enmAddrType;
840	return rc;
841	#endif /* !RT_OS_OS2 */
842	}
843
844
845	RTDECL(int) RTSocketRead(RTSOCKET hSocket, void pvBuffer, size_t cbBuffer, size_t pcbRead)
846	{
847	/*
848	* Validate input.
849	*/
850	RTSOCKETINT *pThis = hSocket;
851	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
852	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
853	AssertReturn(cbBuffer > 0, VERR_INVALID_PARAMETER);
854	AssertPtr(pvBuffer);
855	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
856
857	int rc = rtSocketSwitchBlockingMode(pThis, true /* fBlocking */);
858	if (RT_FAILURE(rc))
859	return rc;
860
861	/*
862	* Read loop.
863	* If pcbRead is NULL we have to fill the entire buffer!
864	*/
865	size_t cbRead = 0;
866	size_t cbToRead = cbBuffer;
867	for (;;)
868	{
869	rtSocketErrorReset();
870	#ifdef RTSOCKET_MAX_READ
871	int cbNow = cbToRead >= RTSOCKET_MAX_READ ? RTSOCKET_MAX_READ : (int)cbToRead;
872	#else
873	size_t cbNow = cbToRead;
874	#endif
875	ssize_t cbBytesRead = recv(pThis->hNative, (char *)pvBuffer + cbRead, cbNow, MSG_NOSIGNAL);
876	if (cbBytesRead <= 0)
877	{
878	rc = rtSocketError();
879	Assert(RT_FAILURE_NP(rc) \|\| cbBytesRead == 0);
880	if (RT_SUCCESS_NP(rc))
881	{
882	if (!pcbRead)
883	rc = VERR_NET_SHUTDOWN;
884	else
885	{
886	*pcbRead = 0;
887	rc = VINF_SUCCESS;
888	}
889	}
890	break;
891	}
892	if (pcbRead)
893	{
894	/* return partial data */
895	*pcbRead = cbBytesRead;
896	break;
897	}
898
899	/* read more? */
900	cbRead += cbBytesRead;
901	if (cbRead == cbBuffer)
902	break;
903
904	/* next */
905	cbToRead = cbBuffer - cbRead;
906	}
907
908	rtSocketUnlock(pThis);
909	return rc;
910	}
911
912
913	RTDECL(int) RTSocketReadFrom(RTSOCKET hSocket, void pvBuffer, size_t cbBuffer, size_t pcbRead, PRTNETADDR pSrcAddr)
914	{
915	/*
916	* Validate input.
917	*/
918	RTSOCKETINT *pThis = hSocket;
919	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
920	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
921	AssertReturn(cbBuffer > 0, VERR_INVALID_PARAMETER);
922	AssertPtr(pvBuffer);
923	AssertPtr(pcbRead);
924	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
925
926	int rc = rtSocketSwitchBlockingMode(pThis, true /* fBlocking */);
927	if (RT_FAILURE(rc))
928	return rc;
929
930	/*
931	* Read data.
932	*/
933	size_t cbRead = 0;
934	size_t cbToRead = cbBuffer;
935	rtSocketErrorReset();
936	RTSOCKADDRUNION u;
937	#ifdef RTSOCKET_MAX_READ
938	int cbNow = cbToRead >= RTSOCKET_MAX_READ ? RTSOCKET_MAX_READ : (int)cbToRead;
939	int cbAddr = sizeof(u);
940	#else
941	size_t cbNow = cbToRead;
942	socklen_t cbAddr = sizeof(u);
943	#endif
944	ssize_t cbBytesRead = recvfrom(pThis->hNative, (char *)pvBuffer + cbRead, cbNow, MSG_NOSIGNAL, &u.Addr, &cbAddr);
945	if (cbBytesRead <= 0)
946	{
947	rc = rtSocketError();
948	Assert(RT_FAILURE_NP(rc) \|\| cbBytesRead == 0);
949	if (RT_SUCCESS_NP(rc))
950	{
951	*pcbRead = 0;
952	rc = VINF_SUCCESS;
953	}
954	}
955	else
956	{
957	if (pSrcAddr)
958	rc = rtSocketNetAddrFromAddr(&u, cbAddr, pSrcAddr);
959	*pcbRead = cbBytesRead;
960	}
961
962	rtSocketUnlock(pThis);
963	return rc;
964	}
965
966
967	RTDECL(int) RTSocketWrite(RTSOCKET hSocket, const void *pvBuffer, size_t cbBuffer)
968	{
969	/*
970	* Validate input.
971	*/
972	RTSOCKETINT *pThis = hSocket;
973	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
974	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
975	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
976
977	int rc = rtSocketSwitchBlockingMode(pThis, true /* fBlocking */);
978	if (RT_FAILURE(rc))
979	return rc;
980
981	/*
982	* Try write all at once.
983	*/
984	#ifdef RTSOCKET_MAX_WRITE
985	int cbNow = cbBuffer >= RTSOCKET_MAX_WRITE ? RTSOCKET_MAX_WRITE : (int)cbBuffer;
986	#else
987	size_t cbNow = cbBuffer >= SSIZE_MAX ? SSIZE_MAX : cbBuffer;
988	#endif
989	ssize_t cbWritten = send(pThis->hNative, (const char *)pvBuffer, cbNow, MSG_NOSIGNAL);
990	if (RT_LIKELY((size_t)cbWritten == cbBuffer && cbWritten >= 0))
991	rc = VINF_SUCCESS;
992	else if (cbWritten < 0)
993	rc = rtSocketError();
994	else
995	{
996	/*
997	* Unfinished business, write the remainder of the request. Must ignore
998	* VERR_INTERRUPTED here if we've managed to send something.
999	*/
1000	size_t cbSentSoFar = 0;
1001	for (;;)
1002	{
1003	/* advance */
1004	cbBuffer -= (size_t)cbWritten;
1005	if (!cbBuffer)
1006	break;
1007	cbSentSoFar += (size_t)cbWritten;
1008	pvBuffer = (char const *)pvBuffer + cbWritten;
1009
1010	/* send */
1011	#ifdef RTSOCKET_MAX_WRITE
1012	cbNow = cbBuffer >= RTSOCKET_MAX_WRITE ? RTSOCKET_MAX_WRITE : (int)cbBuffer;
1013	#else
1014	cbNow = cbBuffer >= SSIZE_MAX ? SSIZE_MAX : cbBuffer;
1015	#endif
1016	cbWritten = send(pThis->hNative, (const char *)pvBuffer, cbNow, MSG_NOSIGNAL);
1017	if (cbWritten >= 0)
1018	AssertMsg(cbBuffer >= (size_t)cbWritten, ("Wrote more than we requested!!! cbWritten=%zu cbBuffer=%zu rtSocketError()=%d\n",
1019	cbWritten, cbBuffer, rtSocketError()));
1020	else
1021	{
1022	rc = rtSocketError();
1023	if (rc != VERR_INTERNAL_ERROR \|\| cbSentSoFar == 0)
1024	break;
1025	cbWritten = 0;
1026	rc = VINF_SUCCESS;
1027	}
1028	}
1029	}
1030
1031	rtSocketUnlock(pThis);
1032	return rc;
1033	}
1034
1035
1036	RTDECL(int) RTSocketWriteTo(RTSOCKET hSocket, const void *pvBuffer, size_t cbBuffer, PCRTNETADDR pAddr)
1037	{
1038	/*
1039	* Validate input.
1040	*/
1041	RTSOCKETINT *pThis = hSocket;
1042	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1043	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1044
1045	/* no locking since UDP reads may be done concurrently to writes, and
1046	* this is the normal use case of this code. */
1047
1048	int rc = rtSocketSwitchBlockingMode(pThis, true /* fBlocking */);
1049	if (RT_FAILURE(rc))
1050	return rc;
1051
1052	/* Figure out destination address. */
1053	struct sockaddr *pSA = NULL;
1054	#ifdef RT_OS_WINDOWS
1055	int cbSA = 0;
1056	#else
1057	socklen_t cbSA = 0;
1058	#endif
1059	RTSOCKADDRUNION u;
1060	if (pAddr)
1061	{
1062	rc = rtSocketAddrFromNetAddr(pAddr, &u, sizeof(u), NULL);
1063	if (RT_FAILURE(rc))
1064	return rc;
1065	pSA = &u.Addr;
1066	cbSA = sizeof(u);
1067	}
1068
1069	/*
1070	* Must write all at once, otherwise it is a failure.
1071	*/
1072	#ifdef RT_OS_WINDOWS
1073	int cbNow = cbBuffer >= RTSOCKET_MAX_WRITE ? RTSOCKET_MAX_WRITE : (int)cbBuffer;
1074	#else
1075	size_t cbNow = cbBuffer >= SSIZE_MAX ? SSIZE_MAX : cbBuffer;
1076	#endif
1077	ssize_t cbWritten = sendto(pThis->hNative, (const char *)pvBuffer, cbNow, MSG_NOSIGNAL, pSA, cbSA);
1078	if (RT_LIKELY((size_t)cbWritten == cbBuffer && cbWritten >= 0))
1079	rc = VINF_SUCCESS;
1080	else if (cbWritten < 0)
1081	rc = rtSocketError();
1082	else
1083	rc = VERR_TOO_MUCH_DATA;
1084
1085	rtSocketUnlock(pThis);
1086	return rc;
1087	}
1088
1089
1090	RTDECL(int) RTSocketSgWrite(RTSOCKET hSocket, PCRTSGBUF pSgBuf)
1091	{
1092	/*
1093	* Validate input.
1094	*/
1095	RTSOCKETINT *pThis = hSocket;
1096	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1097	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1098	AssertPtrReturn(pSgBuf, VERR_INVALID_PARAMETER);
1099	AssertReturn(pSgBuf->cSegs > 0, VERR_INVALID_PARAMETER);
1100	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1101
1102	int rc = rtSocketSwitchBlockingMode(pThis, true /* fBlocking */);
1103	if (RT_FAILURE(rc))
1104	return rc;
1105
1106	/*
1107	* Construct message descriptor (translate pSgBuf) and send it.
1108	*/
1109	rc = VERR_NO_TMP_MEMORY;
1110	#ifdef RT_OS_WINDOWS
1111	AssertCompileSize(WSABUF, sizeof(RTSGSEG));
1112	AssertCompileMemberSize(WSABUF, buf, RT_SIZEOFMEMB(RTSGSEG, pvSeg));
1113
1114	LPWSABUF paMsg = (LPWSABUF)RTMemTmpAllocZ(pSgBuf->cSegs * sizeof(WSABUF));
1115	if (paMsg)
1116	{
1117	for (unsigned i = 0; i < pSgBuf->cSegs; i++)
1118	{
1119	paMsg[i].buf = (char *)pSgBuf->paSegs[i].pvSeg;
1120	paMsg[i].len = (u_long)pSgBuf->paSegs[i].cbSeg;
1121	}
1122
1123	DWORD dwSent;
1124	int hrc = WSASend(pThis->hNative, paMsg, pSgBuf->cSegs, &dwSent,
1125	MSG_NOSIGNAL, NULL, NULL);
1126	if (!hrc)
1127	rc = VINF_SUCCESS;
1128	/** @todo check for incomplete writes */
1129	else
1130	rc = rtSocketError();
1131
1132	RTMemTmpFree(paMsg);
1133	}
1134
1135	#else /* !RT_OS_WINDOWS */
1136	AssertCompileSize(struct iovec, sizeof(RTSGSEG));
1137	AssertCompileMemberSize(struct iovec, iov_base, RT_SIZEOFMEMB(RTSGSEG, pvSeg));
1138	AssertCompileMemberSize(struct iovec, iov_len, RT_SIZEOFMEMB(RTSGSEG, cbSeg));
1139
1140	struct iovec paMsg = (struct iovec )RTMemTmpAllocZ(pSgBuf->cSegs * sizeof(struct iovec));
1141	if (paMsg)
1142	{
1143	for (unsigned i = 0; i < pSgBuf->cSegs; i++)
1144	{
1145	paMsg[i].iov_base = pSgBuf->paSegs[i].pvSeg;
1146	paMsg[i].iov_len = pSgBuf->paSegs[i].cbSeg;
1147	}
1148
1149	struct msghdr msgHdr;
1150	RT_ZERO(msgHdr);
1151	msgHdr.msg_iov = paMsg;
1152	msgHdr.msg_iovlen = pSgBuf->cSegs;
1153	ssize_t cbWritten = sendmsg(pThis->hNative, &msgHdr, MSG_NOSIGNAL);
1154	if (RT_LIKELY(cbWritten >= 0))
1155	rc = VINF_SUCCESS;
1156	/** @todo check for incomplete writes */
1157	else
1158	rc = rtSocketError();
1159
1160	RTMemTmpFree(paMsg);
1161	}
1162	#endif /* !RT_OS_WINDOWS */
1163
1164	rtSocketUnlock(pThis);
1165	return rc;
1166	}
1167
1168
1169	RTDECL(int) RTSocketSgWriteL(RTSOCKET hSocket, size_t cSegs, ...)
1170	{
1171	va_list va;
1172	va_start(va, cSegs);
1173	int rc = RTSocketSgWriteLV(hSocket, cSegs, va);
1174	va_end(va);
1175	return rc;
1176	}
1177
1178
1179	RTDECL(int) RTSocketSgWriteLV(RTSOCKET hSocket, size_t cSegs, va_list va)
1180	{
1181	/*
1182	* Set up a S/G segment array + buffer on the stack and pass it
1183	* on to RTSocketSgWrite.
1184	*/
1185	Assert(cSegs <= 16);
1186	PRTSGSEG paSegs = (PRTSGSEG)alloca(cSegs * sizeof(RTSGSEG));
1187	AssertReturn(paSegs, VERR_NO_TMP_MEMORY);
1188	for (size_t i = 0; i < cSegs; i++)
1189	{
1190	paSegs[i].pvSeg = va_arg(va, void *);
1191	paSegs[i].cbSeg = va_arg(va, size_t);
1192	}
1193
1194	RTSGBUF SgBuf;
1195	RTSgBufInit(&SgBuf, paSegs, cSegs);
1196	return RTSocketSgWrite(hSocket, &SgBuf);
1197	}
1198
1199
1200	RTDECL(int) RTSocketReadNB(RTSOCKET hSocket, void pvBuffer, size_t cbBuffer, size_t pcbRead)
1201	{
1202	/*
1203	* Validate input.
1204	*/
1205	RTSOCKETINT *pThis = hSocket;
1206	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1207	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1208	AssertReturn(cbBuffer > 0, VERR_INVALID_PARAMETER);
1209	AssertPtr(pvBuffer);
1210	AssertPtrReturn(pcbRead, VERR_INVALID_PARAMETER);
1211	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1212
1213	int rc = rtSocketSwitchBlockingMode(pThis, false /* fBlocking */);
1214	if (RT_FAILURE(rc))
1215	return rc;
1216
1217	rtSocketErrorReset();
1218	#ifdef RTSOCKET_MAX_READ
1219	int cbNow = cbBuffer >= RTSOCKET_MAX_WRITE ? RTSOCKET_MAX_WRITE : (int)cbBuffer;
1220	#else
1221	size_t cbNow = cbBuffer;
1222	#endif
1223
1224	#ifdef RT_OS_WINDOWS
1225	int cbRead = recv(pThis->hNative, (char *)pvBuffer, cbNow, MSG_NOSIGNAL);
1226	if (cbRead >= 0)
1227	{
1228	*pcbRead = cbRead;
1229	rc = VINF_SUCCESS;
1230	}
1231	else
1232	rc = rtSocketError();
1233
1234	if (rc == VERR_TRY_AGAIN)
1235	rc = VINF_TRY_AGAIN;
1236	#else
1237	ssize_t cbRead = recv(pThis->hNative, pvBuffer, cbNow, MSG_NOSIGNAL);
1238	if (cbRead >= 0)
1239	*pcbRead = cbRead;
1240	else if (errno == EAGAIN)
1241	{
1242	*pcbRead = 0;
1243	rc = VINF_TRY_AGAIN;
1244	}
1245	else
1246	rc = rtSocketError();
1247	#endif
1248
1249	rtSocketUnlock(pThis);
1250	return rc;
1251	}
1252
1253
1254	RTDECL(int) RTSocketWriteNB(RTSOCKET hSocket, const void pvBuffer, size_t cbBuffer, size_t pcbWritten)
1255	{
1256	/*
1257	* Validate input.
1258	*/
1259	RTSOCKETINT *pThis = hSocket;
1260	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1261	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1262	AssertPtrReturn(pcbWritten, VERR_INVALID_PARAMETER);
1263	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1264
1265	int rc = rtSocketSwitchBlockingMode(pThis, false /* fBlocking */);
1266	if (RT_FAILURE(rc))
1267	return rc;
1268
1269	rtSocketErrorReset();
1270	#ifdef RTSOCKET_MAX_WRITE
1271	int cbNow = cbBuffer >= RTSOCKET_MAX_WRITE ? RTSOCKET_MAX_WRITE : (int)cbBuffer;
1272	#else
1273	size_t cbNow = cbBuffer;
1274	#endif
1275
1276	#ifdef RT_OS_WINDOWS
1277	int cbWritten = send(pThis->hNative, (const char *)pvBuffer, cbNow, MSG_NOSIGNAL);
1278	if (cbWritten >= 0)
1279	{
1280	*pcbWritten = cbWritten;
1281	rc = VINF_SUCCESS;
1282	}
1283	else
1284	rc = rtSocketError();
1285
1286	if (rc == VERR_TRY_AGAIN)
1287	rc = VINF_TRY_AGAIN;
1288	#else
1289	ssize_t cbWritten = send(pThis->hNative, pvBuffer, cbBuffer, MSG_NOSIGNAL);
1290	if (cbWritten >= 0)
1291	*pcbWritten = cbWritten;
1292	else if (errno == EAGAIN)
1293	{
1294	*pcbWritten = 0;
1295	rc = VINF_TRY_AGAIN;
1296	}
1297	else
1298	rc = rtSocketError();
1299	#endif
1300
1301	rtSocketUnlock(pThis);
1302	return rc;
1303	}
1304
1305
1306	RTDECL(int) RTSocketSgWriteNB(RTSOCKET hSocket, PCRTSGBUF pSgBuf, size_t *pcbWritten)
1307	{
1308	/*
1309	* Validate input.
1310	*/
1311	RTSOCKETINT *pThis = hSocket;
1312	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1313	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1314	AssertPtrReturn(pSgBuf, VERR_INVALID_PARAMETER);
1315	AssertPtrReturn(pcbWritten, VERR_INVALID_PARAMETER);
1316	AssertReturn(pSgBuf->cSegs > 0, VERR_INVALID_PARAMETER);
1317	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1318
1319	int rc = rtSocketSwitchBlockingMode(pThis, false /* fBlocking */);
1320	if (RT_FAILURE(rc))
1321	return rc;
1322
1323	unsigned cSegsToSend = 0;
1324	rc = VERR_NO_TMP_MEMORY;
1325	#ifdef RT_OS_WINDOWS
1326	LPWSABUF paMsg = NULL;
1327
1328	RTSgBufMapToNative(paMsg, pSgBuf, WSABUF, buf, char *, len, u_long, cSegsToSend);
1329	if (paMsg)
1330	{
1331	DWORD dwSent = 0;
1332	int hrc = WSASend(pThis->hNative, paMsg, cSegsToSend, &dwSent,
1333	MSG_NOSIGNAL, NULL, NULL);
1334	if (!hrc)
1335	rc = VINF_SUCCESS;
1336	else
1337	rc = rtSocketError();
1338
1339	*pcbWritten = dwSent;
1340
1341	RTMemTmpFree(paMsg);
1342	}
1343
1344	#else /* !RT_OS_WINDOWS */
1345	struct iovec *paMsg = NULL;
1346
1347	RTSgBufMapToNative(paMsg, pSgBuf, struct iovec, iov_base, void *, iov_len, size_t, cSegsToSend);
1348	if (paMsg)
1349	{
1350	struct msghdr msgHdr;
1351	RT_ZERO(msgHdr);
1352	msgHdr.msg_iov = paMsg;
1353	msgHdr.msg_iovlen = cSegsToSend;
1354	ssize_t cbWritten = sendmsg(pThis->hNative, &msgHdr, MSG_NOSIGNAL);
1355	if (RT_LIKELY(cbWritten >= 0))
1356	{
1357	rc = VINF_SUCCESS;
1358	*pcbWritten = cbWritten;
1359	}
1360	else
1361	rc = rtSocketError();
1362
1363	RTMemTmpFree(paMsg);
1364	}
1365	#endif /* !RT_OS_WINDOWS */
1366
1367	rtSocketUnlock(pThis);
1368	return rc;
1369	}
1370
1371
1372	RTDECL(int) RTSocketSgWriteLNB(RTSOCKET hSocket, size_t cSegs, size_t *pcbWritten, ...)
1373	{
1374	va_list va;
1375	va_start(va, pcbWritten);
1376	int rc = RTSocketSgWriteLVNB(hSocket, cSegs, pcbWritten, va);
1377	va_end(va);
1378	return rc;
1379	}
1380
1381
1382	RTDECL(int) RTSocketSgWriteLVNB(RTSOCKET hSocket, size_t cSegs, size_t *pcbWritten, va_list va)
1383	{
1384	/*
1385	* Set up a S/G segment array + buffer on the stack and pass it
1386	* on to RTSocketSgWrite.
1387	*/
1388	Assert(cSegs <= 16);
1389	PRTSGSEG paSegs = (PRTSGSEG)alloca(cSegs * sizeof(RTSGSEG));
1390	AssertReturn(paSegs, VERR_NO_TMP_MEMORY);
1391	for (size_t i = 0; i < cSegs; i++)
1392	{
1393	paSegs[i].pvSeg = va_arg(va, void *);
1394	paSegs[i].cbSeg = va_arg(va, size_t);
1395	}
1396
1397	RTSGBUF SgBuf;
1398	RTSgBufInit(&SgBuf, paSegs, cSegs);
1399	return RTSocketSgWriteNB(hSocket, &SgBuf, pcbWritten);
1400	}
1401
1402
1403	RTDECL(int) RTSocketSelectOne(RTSOCKET hSocket, RTMSINTERVAL cMillies)
1404	{
1405	/*
1406	* Validate input.
1407	*/
1408	RTSOCKETINT *pThis = hSocket;
1409	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1410	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1411	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
1412	int const fdMax = (int)pThis->hNative + 1;
1413	AssertReturn(fdMax - 1 == pThis->hNative, VERR_INTERNAL_ERROR_5);
1414
1415	/*
1416	* Set up the file descriptor sets and do the select.
1417	*/
1418	fd_set fdsetR;
1419	FD_ZERO(&fdsetR);
1420	FD_SET(pThis->hNative, &fdsetR);
1421
1422	fd_set fdsetE = fdsetR;
1423
1424	int rc;
1425	if (cMillies == RT_INDEFINITE_WAIT)
1426	rc = select(fdMax, &fdsetR, NULL, &fdsetE, NULL);
1427	else
1428	{
1429	struct timeval timeout;
1430	timeout.tv_sec = cMillies / 1000;
1431	timeout.tv_usec = (cMillies % 1000) * 1000;
1432	rc = select(fdMax, &fdsetR, NULL, &fdsetE, &timeout);
1433	}
1434	if (rc > 0)
1435	rc = VINF_SUCCESS;
1436	else if (rc == 0)
1437	rc = VERR_TIMEOUT;
1438	else
1439	rc = rtSocketError();
1440
1441	return rc;
1442	}
1443
1444
1445	RTDECL(int) RTSocketSelectOneEx(RTSOCKET hSocket, uint32_t fEvents, uint32_t *pfEvents,
1446	RTMSINTERVAL cMillies)
1447	{
1448	/*
1449	* Validate input.
1450	*/
1451	RTSOCKETINT *pThis = hSocket;
1452	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1453	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1454	AssertPtrReturn(pfEvents, VERR_INVALID_PARAMETER);
1455	AssertReturn(!(fEvents & ~RTSOCKET_EVT_VALID_MASK), VERR_INVALID_PARAMETER);
1456	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
1457	int const fdMax = (int)pThis->hNative + 1;
1458	AssertReturn(fdMax - 1 == pThis->hNative, VERR_INTERNAL_ERROR_5);
1459
1460	*pfEvents = 0;
1461
1462	/*
1463	* Set up the file descriptor sets and do the select.
1464	*/
1465	fd_set fdsetR;
1466	fd_set fdsetW;
1467	fd_set fdsetE;
1468	FD_ZERO(&fdsetR);
1469	FD_ZERO(&fdsetW);
1470	FD_ZERO(&fdsetE);
1471
1472	if (fEvents & RTSOCKET_EVT_READ)
1473	FD_SET(pThis->hNative, &fdsetR);
1474	if (fEvents & RTSOCKET_EVT_WRITE)
1475	FD_SET(pThis->hNative, &fdsetW);
1476	if (fEvents & RTSOCKET_EVT_ERROR)
1477	FD_SET(pThis->hNative, &fdsetE);
1478
1479	int rc;
1480	if (cMillies == RT_INDEFINITE_WAIT)
1481	rc = select(fdMax, &fdsetR, &fdsetW, &fdsetE, NULL);
1482	else
1483	{
1484	struct timeval timeout;
1485	timeout.tv_sec = cMillies / 1000;
1486	timeout.tv_usec = (cMillies % 1000) * 1000;
1487	rc = select(fdMax, &fdsetR, &fdsetW, &fdsetE, &timeout);
1488	}
1489	if (rc > 0)
1490	{
1491	if (FD_ISSET(pThis->hNative, &fdsetR))
1492	*pfEvents \|= RTSOCKET_EVT_READ;
1493	if (FD_ISSET(pThis->hNative, &fdsetW))
1494	*pfEvents \|= RTSOCKET_EVT_WRITE;
1495	if (FD_ISSET(pThis->hNative, &fdsetE))
1496	*pfEvents \|= RTSOCKET_EVT_ERROR;
1497
1498	rc = VINF_SUCCESS;
1499	}
1500	else if (rc == 0)
1501	rc = VERR_TIMEOUT;
1502	else
1503	rc = rtSocketError();
1504
1505	return rc;
1506	}
1507
1508
1509	RTDECL(int) RTSocketShutdown(RTSOCKET hSocket, bool fRead, bool fWrite)
1510	{
1511	/*
1512	* Validate input, don't lock it because we might want to interrupt a call
1513	* active on a different thread.
1514	*/
1515	RTSOCKETINT *pThis = hSocket;
1516	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1517	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1518	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
1519	AssertReturn(fRead \|\| fWrite, VERR_INVALID_PARAMETER);
1520
1521	/*
1522	* Do the job.
1523	*/
1524	int rc = VINF_SUCCESS;
1525	int fHow;
1526	if (fRead && fWrite)
1527	fHow = SHUT_RDWR;
1528	else if (fRead)
1529	fHow = SHUT_RD;
1530	else
1531	fHow = SHUT_WR;
1532	if (shutdown(pThis->hNative, fHow) == -1)
1533	rc = rtSocketError();
1534
1535	return rc;
1536	}
1537
1538
1539	RTDECL(int) RTSocketGetLocalAddress(RTSOCKET hSocket, PRTNETADDR pAddr)
1540	{
1541	/*
1542	* Validate input.
1543	*/
1544	RTSOCKETINT *pThis = hSocket;
1545	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1546	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1547	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
1548
1549	/*
1550	* Get the address and convert it.
1551	*/
1552	int rc;
1553	RTSOCKADDRUNION u;
1554	#ifdef RT_OS_WINDOWS
1555	int cbAddr = sizeof(u);
1556	#else
1557	socklen_t cbAddr = sizeof(u);
1558	#endif
1559	RT_ZERO(u);
1560	if (getsockname(pThis->hNative, &u.Addr, &cbAddr) == 0)
1561	rc = rtSocketNetAddrFromAddr(&u, cbAddr, pAddr);
1562	else
1563	rc = rtSocketError();
1564
1565	return rc;
1566	}
1567
1568
1569	RTDECL(int) RTSocketGetPeerAddress(RTSOCKET hSocket, PRTNETADDR pAddr)
1570	{
1571	/*
1572	* Validate input.
1573	*/
1574	RTSOCKETINT *pThis = hSocket;
1575	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1576	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1577	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
1578
1579	/*
1580	* Get the address and convert it.
1581	*/
1582	int rc;
1583	RTSOCKADDRUNION u;
1584	#ifdef RT_OS_WINDOWS
1585	int cbAddr = sizeof(u);
1586	#else
1587	socklen_t cbAddr = sizeof(u);
1588	#endif
1589	RT_ZERO(u);
1590	if (getpeername(pThis->hNative, &u.Addr, &cbAddr) == 0)
1591	rc = rtSocketNetAddrFromAddr(&u, cbAddr, pAddr);
1592	else
1593	rc = rtSocketError();
1594
1595	return rc;
1596	}
1597
1598
1599
1600	/**
1601	* Wrapper around bind.
1602	*
1603	* @returns IPRT status code.
1604	* @param hSocket The socket handle.
1605	* @param pAddr The address to bind to.
1606	*/
1607	int rtSocketBind(RTSOCKET hSocket, PCRTNETADDR pAddr)
1608	{
1609	/*
1610	* Validate input.
1611	*/
1612	RTSOCKETINT *pThis = hSocket;
1613	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1614	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1615	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1616
1617	RTSOCKADDRUNION u;
1618	int cbAddr;
1619	int rc = rtSocketAddrFromNetAddr(pAddr, &u, sizeof(u), &cbAddr);
1620	if (RT_SUCCESS(rc))
1621	{
1622	if (bind(pThis->hNative, &u.Addr, cbAddr) != 0)
1623	rc = rtSocketError();
1624	}
1625
1626	rtSocketUnlock(pThis);
1627	return rc;
1628	}
1629
1630
1631	/**
1632	* Wrapper around listen.
1633	*
1634	* @returns IPRT status code.
1635	* @param hSocket The socket handle.
1636	* @param cMaxPending The max number of pending connections.
1637	*/
1638	int rtSocketListen(RTSOCKET hSocket, int cMaxPending)
1639	{
1640	/*
1641	* Validate input.
1642	*/
1643	RTSOCKETINT *pThis = hSocket;
1644	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1645	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1646	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1647
1648	int rc = VINF_SUCCESS;
1649	if (listen(pThis->hNative, cMaxPending) != 0)
1650	rc = rtSocketError();
1651
1652	rtSocketUnlock(pThis);
1653	return rc;
1654	}
1655
1656
1657	/**
1658	* Wrapper around accept.
1659	*
1660	* @returns IPRT status code.
1661	* @param hSocket The socket handle.
1662	* @param phClient Where to return the client socket handle on
1663	* success.
1664	* @param pAddr Where to return the client address.
1665	* @param pcbAddr On input this gives the size buffer size of what
1666	* @a pAddr point to. On return this contains the
1667	* size of what's stored at @a pAddr.
1668	*/
1669	int rtSocketAccept(RTSOCKET hSocket, PRTSOCKET phClient, struct sockaddr pAddr, size_t pcbAddr)
1670	{
1671	/*
1672	* Validate input.
1673	* Only lock the socket temporarily while we get the native handle, so that
1674	* we can safely shutdown and destroy the socket from a different thread.
1675	*/
1676	RTSOCKETINT *pThis = hSocket;
1677	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1678	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1679	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1680
1681	/*
1682	* Call accept().
1683	*/
1684	rtSocketErrorReset();
1685	int rc = VINF_SUCCESS;
1686	#ifdef RT_OS_WINDOWS
1687	int cbAddr = (int)*pcbAddr;
1688	#else
1689	socklen_t cbAddr = *pcbAddr;
1690	#endif
1691	RTSOCKETNATIVE hNativeClient = accept(pThis->hNative, pAddr, &cbAddr);
1692	if (hNativeClient != NIL_RTSOCKETNATIVE)
1693	{
1694	*pcbAddr = cbAddr;
1695
1696	/*
1697	* Wrap the client socket.
1698	*/
1699	rc = rtSocketCreateForNative(phClient, hNativeClient);
1700	if (RT_FAILURE(rc))
1701	{
1702	#ifdef RT_OS_WINDOWS
1703	closesocket(hNativeClient);
1704	#else
1705	close(hNativeClient);
1706	#endif
1707	}
1708	}
1709	else
1710	rc = rtSocketError();
1711
1712	rtSocketUnlock(pThis);
1713	return rc;
1714	}
1715
1716
1717	/**
1718	* Wrapper around connect.
1719	*
1720	* @returns IPRT status code.
1721	* @param hSocket The socket handle.
1722	* @param pAddr The socket address to connect to.
1723	*/
1724	int rtSocketConnect(RTSOCKET hSocket, PCRTNETADDR pAddr)
1725	{
1726	/*
1727	* Validate input.
1728	*/
1729	RTSOCKETINT *pThis = hSocket;
1730	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1731	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1732	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1733
1734	RTSOCKADDRUNION u;
1735	int cbAddr;
1736	int rc = rtSocketAddrFromNetAddr(pAddr, &u, sizeof(u), &cbAddr);
1737	if (RT_SUCCESS(rc))
1738	{
1739	if (connect(pThis->hNative, &u.Addr, cbAddr) != 0)
1740	rc = rtSocketError();
1741	}
1742
1743	rtSocketUnlock(pThis);
1744	return rc;
1745	}
1746
1747
1748	/**
1749	* Wrapper around setsockopt.
1750	*
1751	* @returns IPRT status code.
1752	* @param hSocket The socket handle.
1753	* @param iLevel The protocol level, e.g. IPPORTO_TCP.
1754	* @param iOption The option, e.g. TCP_NODELAY.
1755	* @param pvValue The value buffer.
1756	* @param cbValue The size of the value pointed to by pvValue.
1757	*/
1758	int rtSocketSetOpt(RTSOCKET hSocket, int iLevel, int iOption, void const *pvValue, int cbValue)
1759	{
1760	/*
1761	* Validate input.
1762	*/
1763	RTSOCKETINT *pThis = hSocket;
1764	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1765	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1766	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1767
1768	int rc = VINF_SUCCESS;
1769	if (setsockopt(pThis->hNative, iLevel, iOption, (const char *)pvValue, cbValue) != 0)
1770	rc = rtSocketError();
1771
1772	rtSocketUnlock(pThis);
1773	return rc;
1774	}
1775
1776
1777	/**
1778	* Internal RTPollSetAdd helper that returns the handle that should be added to
1779	* the pollset.
1780	*
1781	* @returns Valid handle on success, INVALID_HANDLE_VALUE on failure.
1782	* @param hSocket The socket handle.
1783	* @param fEvents The events we're polling for.
1784	* @param phNative Where to put the primary handle.
1785	*/
1786	int rtSocketPollGetHandle(RTSOCKET hSocket, uint32_t fEvents, PRTHCINTPTR phNative)
1787	{
1788	RTSOCKETINT *pThis = hSocket;
1789	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1790	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1791	#ifdef RT_OS_WINDOWS
1792	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1793
1794	int rc = VINF_SUCCESS;
1795	if (pThis->hEvent != WSA_INVALID_EVENT)
1796	*phNative = (RTHCINTPTR)pThis->hEvent;
1797	else
1798	{
1799	pThis->hEvent = WSACreateEvent();
1800	*phNative = (RTHCINTPTR)pThis->hEvent;
1801	if (pThis->hEvent == WSA_INVALID_EVENT)
1802	rc = rtSocketError();
1803	}
1804
1805	rtSocketUnlock(pThis);
1806	return rc;
1807
1808	#else /* !RT_OS_WINDOWS */
1809	*phNative = (RTHCUINTPTR)pThis->hNative;
1810	return VINF_SUCCESS;
1811	#endif /* !RT_OS_WINDOWS */
1812	}
1813
1814	#ifdef RT_OS_WINDOWS
1815
1816	/**
1817	* Undos the harm done by WSAEventSelect.
1818	*
1819	* @returns IPRT status code.
1820	* @param pThis The socket handle.
1821	*/
1822	static int rtSocketPollClearEventAndRestoreBlocking(RTSOCKETINT *pThis)
1823	{
1824	int rc = VINF_SUCCESS;
1825	if (pThis->fSubscribedEvts)
1826	{
1827	if (WSAEventSelect(pThis->hNative, WSA_INVALID_EVENT, 0) == 0)
1828	{
1829	pThis->fSubscribedEvts = 0;
1830
1831	/*
1832	* Switch back to blocking mode if that was the state before the
1833	* operation.
1834	*/
1835	if (pThis->fBlocking)
1836	{
1837	u_long fNonBlocking = 0;
1838	int rc2 = ioctlsocket(pThis->hNative, FIONBIO, &fNonBlocking);
1839	if (rc2 != 0)
1840	{
1841	rc = rtSocketError();
1842	AssertMsgFailed(("%Rrc; rc2=%d\n", rc, rc2));
1843	}
1844	}
1845	}
1846	else
1847	{
1848	rc = rtSocketError();
1849	AssertMsgFailed(("%Rrc\n", rc));
1850	}
1851	}
1852	return rc;
1853	}
1854
1855
1856	/**
1857	* Updates the mask of events we're subscribing to.
1858	*
1859	* @returns IPRT status code.
1860	* @param pThis The socket handle.
1861	* @param fEvents The events we want to subscribe to.
1862	*/
1863	static int rtSocketPollUpdateEvents(RTSOCKETINT *pThis, uint32_t fEvents)
1864	{
1865	LONG fNetworkEvents = 0;
1866	if (fEvents & RTPOLL_EVT_READ)
1867	fNetworkEvents \|= FD_READ;
1868	if (fEvents & RTPOLL_EVT_WRITE)
1869	fNetworkEvents \|= FD_WRITE;
1870	if (fEvents & RTPOLL_EVT_ERROR)
1871	fNetworkEvents \|= FD_CLOSE;
1872	LogFlowFunc(("fNetworkEvents=%#x\n", fNetworkEvents));
1873	if (WSAEventSelect(pThis->hNative, pThis->hEvent, fNetworkEvents) == 0)
1874	{
1875	pThis->fSubscribedEvts = fEvents;
1876	return VINF_SUCCESS;
1877	}
1878
1879	int rc = rtSocketError();
1880	AssertMsgFailed(("fNetworkEvents=%#x rc=%Rrc\n", fNetworkEvents, rtSocketError()));
1881	return rc;
1882	}
1883
1884	#endif /* RT_OS_WINDOWS */
1885
1886
1887	#if defined(RT_OS_WINDOWS) \|\| defined(RT_OS_OS2)
1888
1889	/**
1890	* Checks for pending events.
1891	*
1892	* @returns Event mask or 0.
1893	* @param pThis The socket handle.
1894	* @param fEvents The desired events.
1895	*/
1896	static uint32_t rtSocketPollCheck(RTSOCKETINT *pThis, uint32_t fEvents)
1897	{
1898	uint32_t fRetEvents = 0;
1899
1900	LogFlowFunc(("pThis=%#p fEvents=%#x\n", pThis, fEvents));
1901
1902	# ifdef RT_OS_WINDOWS
1903	/* Make sure WSAEnumNetworkEvents returns what we want. */
1904	int rc = VINF_SUCCESS;
1905	if ((pThis->fSubscribedEvts & fEvents) != fEvents)
1906	rc = rtSocketPollUpdateEvents(pThis, pThis->fSubscribedEvts \| fEvents);
1907
1908	/* Get the event mask, ASSUMES that WSAEnumNetworkEvents doesn't clear stuff. */
1909	WSANETWORKEVENTS NetEvts;
1910	RT_ZERO(NetEvts);
1911	if (WSAEnumNetworkEvents(pThis->hNative, pThis->hEvent, &NetEvts) == 0)
1912	{
1913	if ( (NetEvts.lNetworkEvents & FD_READ)
1914	&& (fEvents & RTPOLL_EVT_READ)
1915	&& NetEvts.iErrorCode[FD_READ_BIT] == 0)
1916	fRetEvents \|= RTPOLL_EVT_READ;
1917
1918	if ( (NetEvts.lNetworkEvents & FD_WRITE)
1919	&& (fEvents & RTPOLL_EVT_WRITE)
1920	&& NetEvts.iErrorCode[FD_WRITE_BIT] == 0)
1921	fRetEvents \|= RTPOLL_EVT_WRITE;
1922
1923	if (fEvents & RTPOLL_EVT_ERROR)
1924	{
1925	if (NetEvts.lNetworkEvents & FD_CLOSE)
1926	fRetEvents \|= RTPOLL_EVT_ERROR;
1927	else
1928	for (uint32_t i = 0; i < FD_MAX_EVENTS; i++)
1929	if ( (NetEvts.lNetworkEvents & (1L << i))
1930	&& NetEvts.iErrorCode[i] != 0)
1931	fRetEvents \|= RTPOLL_EVT_ERROR;
1932	}
1933	}
1934	else
1935	rc = rtSocketError();
1936
1937	/* Fall back on select if we hit an error above. */
1938	if (RT_FAILURE(rc))
1939	{
1940
1941	}
1942
1943	#else /* RT_OS_OS2 */
1944	int aFds[4] = { pThis->hNative, pThis->hNative, pThis->hNative, -1 };
1945	int rc = os2_select(aFds, 1, 1, 1, 0);
1946	if (rc > 0)
1947	{
1948	if (aFds[0] == pThis->hNative)
1949	fRetEvents \|= RTPOLL_EVT_READ;
1950	if (aFds[1] == pThis->hNative)
1951	fRetEvents \|= RTPOLL_EVT_WRITE;
1952	if (aFds[2] == pThis->hNative)
1953	fRetEvents \|= RTPOLL_EVT_ERROR;
1954	fRetEvents &= fEvents;
1955	}
1956	#endif /* RT_OS_OS2 */
1957
1958	LogFlowFunc(("fRetEvents=%#x\n", fRetEvents));
1959	return fRetEvents;
1960	}
1961
1962
1963	/**
1964	* Internal RTPoll helper that polls the socket handle and, if @a fNoWait is
1965	* clear, starts whatever actions we've got running during the poll call.
1966	*
1967	* @returns 0 if no pending events, actions initiated if @a fNoWait is clear.
1968	* Event mask (in @a fEvents) and no actions if the handle is ready
1969	* already.
1970	* UINT32_MAX (asserted) if the socket handle is busy in I/O or a
1971	* different poll set.
1972	*
1973	* @param hSocket The socket handle.
1974	* @param hPollSet The poll set handle (for access checks).
1975	* @param fEvents The events we're polling for.
1976	* @param fFinalEntry Set if this is the final entry for this handle
1977	* in this poll set. This can be used for dealing
1978	* with duplicate entries.
1979	* @param fNoWait Set if it's a zero-wait poll call. Clear if
1980	* we'll wait for an event to occur.
1981	*
1982	* @remarks There is a potential race wrt duplicate handles when @a fNoWait is
1983	* @c true, we don't currently care about that oddity...
1984	*/
1985	uint32_t rtSocketPollStart(RTSOCKET hSocket, RTPOLLSET hPollSet, uint32_t fEvents, bool fFinalEntry, bool fNoWait)
1986	{
1987	RTSOCKETINT *pThis = hSocket;
1988	AssertPtrReturn(pThis, UINT32_MAX);
1989	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, UINT32_MAX);
1990	/** @todo This isn't quite sane. Replace by critsect and open up concurrent
1991	* reads and writes! */
1992	if (rtSocketTryLock(pThis))
1993	pThis->hPollSet = hPollSet;
1994	else
1995	{
1996	AssertReturn(pThis->hPollSet == hPollSet, UINT32_MAX);
1997	ASMAtomicIncU32(&pThis->cUsers);
1998	}
1999
2000	/* (rtSocketPollCheck will reset the event object). */
2001	# ifdef RT_OS_WINDOWS
2002	uint32_t fRetEvents = pThis->fEventsSaved;
2003	pThis->fEventsSaved = 0; /* Reset */
2004	fRetEvents \|= rtSocketPollCheck(pThis, fEvents);
2005
2006	if ( !fRetEvents
2007	&& !fNoWait)
2008	{
2009	pThis->fPollEvts \|= fEvents;
2010	if ( fFinalEntry
2011	&& pThis->fSubscribedEvts != pThis->fPollEvts)
2012	{
2013	int rc = rtSocketPollUpdateEvents(pThis, pThis->fPollEvts);
2014	if (RT_FAILURE(rc))
2015	{
2016	pThis->fPollEvts = 0;
2017	fRetEvents = UINT32_MAX;
2018	}
2019	}
2020	}
2021	# else
2022	uint32_t fRetEvents = rtSocketPollCheck(pThis, fEvents);
2023	# endif
2024
2025	if (fRetEvents \|\| fNoWait)
2026	{
2027	if (pThis->cUsers == 1)
2028	{
2029	# ifdef RT_OS_WINDOWS
2030	rtSocketPollClearEventAndRestoreBlocking(pThis);
2031	# endif
2032	pThis->hPollSet = NIL_RTPOLLSET;
2033	}
2034	ASMAtomicDecU32(&pThis->cUsers);
2035	}
2036
2037	return fRetEvents;
2038	}
2039
2040
2041	/**
2042	* Called after a WaitForMultipleObjects returned in order to check for pending
2043	* events and stop whatever actions that rtSocketPollStart() initiated.
2044	*
2045	* @returns Event mask or 0.
2046	*
2047	* @param hSocket The socket handle.
2048	* @param fEvents The events we're polling for.
2049	* @param fFinalEntry Set if this is the final entry for this handle
2050	* in this poll set. This can be used for dealing
2051	* with duplicate entries. Only keep in mind that
2052	* this method is called in reverse order, so the
2053	* first call will have this set (when the entire
2054	* set was processed).
2055	* @param fHarvestEvents Set if we should check for pending events.
2056	*/
2057	uint32_t rtSocketPollDone(RTSOCKET hSocket, uint32_t fEvents, bool fFinalEntry, bool fHarvestEvents)
2058	{
2059	RTSOCKETINT *pThis = hSocket;
2060	AssertPtrReturn(pThis, 0);
2061	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, 0);
2062	Assert(pThis->cUsers > 0);
2063	Assert(pThis->hPollSet != NIL_RTPOLLSET);
2064
2065	/* Harvest events and clear the event mask for the next round of polling. */
2066	uint32_t fRetEvents = rtSocketPollCheck(pThis, fEvents);
2067	# ifdef RT_OS_WINDOWS
2068	pThis->fPollEvts = 0;
2069
2070	/*
2071	* Save the write event if required.
2072	* It is only posted once and might get lost if the another source in the
2073	* pollset with a higher priority has pending events.
2074	*/
2075	if ( !fHarvestEvents
2076	&& fRetEvents)
2077	{
2078	pThis->fEventsSaved = fRetEvents;
2079	fRetEvents = 0;
2080	}
2081	# endif
2082
2083	/* Make the socket blocking again and unlock the handle. */
2084	if (pThis->cUsers == 1)
2085	{
2086	# ifdef RT_OS_WINDOWS
2087	rtSocketPollClearEventAndRestoreBlocking(pThis);
2088	# endif
2089	pThis->hPollSet = NIL_RTPOLLSET;
2090	}
2091	ASMAtomicDecU32(&pThis->cUsers);
2092	return fRetEvents;
2093	}
2094
2095	#endif /* RT_OS_WINDOWS \|\| RT_OS_OS2 */
2096

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

source: vbox/trunk/src/VBox/Runtime/r3/socket.cpp@ 49971

Download in other formats: