socket.cpp@ 53487

Last change on this file since 53487 was 53487, checked in by vboxsync, 10 years ago
Split rtSocketBindRawAddr from rtSocketBind.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 59.4 KB

Line
1	/* $Id: socket.cpp 53487 2014-12-09 12:34:02Z 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	cchIpAddress = RTStrPrintf(szIpAddress, sizeof(szIpAddress),
786	"%RTnaipv4", pgrSa->sin_addr.s_addr);
787	Assert(cchIpAddress >= 7 && cchIpAddress < sizeof(szIpAddress) - 1);
788	enmAddrType = RTNETADDRTYPE_IPV4;
789	rc = VINF_SUCCESS;
790	}
791	else if (pgrResult->ai_family == AF_INET6)
792	{
793	struct sockaddr_in6 const pgrSa6 = (struct sockaddr_in6 const )pgrResult->ai_addr;
794	cchIpAddress = RTStrPrintf(szIpAddress, sizeof(szIpAddress),
795	"%RTnaipv6", (PRTNETADDRIPV6)&pgrSa6->sin6_addr);
796	enmAddrType = RTNETADDRTYPE_IPV6;
797	rc = VINF_SUCCESS;
798	}
799	else
800	{
801	rc = VERR_NET_ADDRESS_NOT_AVAILABLE;
802	szIpAddress[0] = '\0';
803	cchIpAddress = 0;
804	}
805	freeaddrinfo(pgrResults);
806
807	/*
808	* Copy out the result.
809	*/
810	size_t const cbResult = *pcbResult;
811	*pcbResult = cchIpAddress + 1;
812	if (cchIpAddress < cbResult)
813	memcpy(pszResult, szIpAddress, cchIpAddress + 1);
814	else
815	{
816	RT_BZERO(pszResult, cbResult);
817	if (RT_SUCCESS(rc))
818	rc = VERR_BUFFER_OVERFLOW;
819	}
820	if (penmAddrType && RT_SUCCESS(rc))
821	*penmAddrType = enmAddrType;
822	return rc;
823	#endif /* !RT_OS_OS2 */
824	}
825
826
827	RTDECL(int) RTSocketRead(RTSOCKET hSocket, void pvBuffer, size_t cbBuffer, size_t pcbRead)
828	{
829	/*
830	* Validate input.
831	*/
832	RTSOCKETINT *pThis = hSocket;
833	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
834	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
835	AssertReturn(cbBuffer > 0, VERR_INVALID_PARAMETER);
836	AssertPtr(pvBuffer);
837	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
838
839	int rc = rtSocketSwitchBlockingMode(pThis, true /* fBlocking */);
840	if (RT_FAILURE(rc))
841	return rc;
842
843	/*
844	* Read loop.
845	* If pcbRead is NULL we have to fill the entire buffer!
846	*/
847	size_t cbRead = 0;
848	size_t cbToRead = cbBuffer;
849	for (;;)
850	{
851	rtSocketErrorReset();
852	#ifdef RTSOCKET_MAX_READ
853	int cbNow = cbToRead >= RTSOCKET_MAX_READ ? RTSOCKET_MAX_READ : (int)cbToRead;
854	#else
855	size_t cbNow = cbToRead;
856	#endif
857	ssize_t cbBytesRead = recv(pThis->hNative, (char *)pvBuffer + cbRead, cbNow, MSG_NOSIGNAL);
858	if (cbBytesRead <= 0)
859	{
860	rc = rtSocketError();
861	Assert(RT_FAILURE_NP(rc) \|\| cbBytesRead == 0);
862	if (RT_SUCCESS_NP(rc))
863	{
864	if (!pcbRead)
865	rc = VERR_NET_SHUTDOWN;
866	else
867	{
868	*pcbRead = 0;
869	rc = VINF_SUCCESS;
870	}
871	}
872	break;
873	}
874	if (pcbRead)
875	{
876	/* return partial data */
877	*pcbRead = cbBytesRead;
878	break;
879	}
880
881	/* read more? */
882	cbRead += cbBytesRead;
883	if (cbRead == cbBuffer)
884	break;
885
886	/* next */
887	cbToRead = cbBuffer - cbRead;
888	}
889
890	rtSocketUnlock(pThis);
891	return rc;
892	}
893
894
895	RTDECL(int) RTSocketReadFrom(RTSOCKET hSocket, void pvBuffer, size_t cbBuffer, size_t pcbRead, PRTNETADDR pSrcAddr)
896	{
897	/*
898	* Validate input.
899	*/
900	RTSOCKETINT *pThis = hSocket;
901	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
902	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
903	AssertReturn(cbBuffer > 0, VERR_INVALID_PARAMETER);
904	AssertPtr(pvBuffer);
905	AssertPtr(pcbRead);
906	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
907
908	int rc = rtSocketSwitchBlockingMode(pThis, true /* fBlocking */);
909	if (RT_FAILURE(rc))
910	return rc;
911
912	/*
913	* Read data.
914	*/
915	size_t cbRead = 0;
916	size_t cbToRead = cbBuffer;
917	rtSocketErrorReset();
918	RTSOCKADDRUNION u;
919	#ifdef RTSOCKET_MAX_READ
920	int cbNow = cbToRead >= RTSOCKET_MAX_READ ? RTSOCKET_MAX_READ : (int)cbToRead;
921	int cbAddr = sizeof(u);
922	#else
923	size_t cbNow = cbToRead;
924	socklen_t cbAddr = sizeof(u);
925	#endif
926	ssize_t cbBytesRead = recvfrom(pThis->hNative, (char *)pvBuffer + cbRead, cbNow, MSG_NOSIGNAL, &u.Addr, &cbAddr);
927	if (cbBytesRead <= 0)
928	{
929	rc = rtSocketError();
930	Assert(RT_FAILURE_NP(rc) \|\| cbBytesRead == 0);
931	if (RT_SUCCESS_NP(rc))
932	{
933	*pcbRead = 0;
934	rc = VINF_SUCCESS;
935	}
936	}
937	else
938	{
939	if (pSrcAddr)
940	rc = rtSocketNetAddrFromAddr(&u, cbAddr, pSrcAddr);
941	*pcbRead = cbBytesRead;
942	}
943
944	rtSocketUnlock(pThis);
945	return rc;
946	}
947
948
949	RTDECL(int) RTSocketWrite(RTSOCKET hSocket, const void *pvBuffer, size_t cbBuffer)
950	{
951	/*
952	* Validate input.
953	*/
954	RTSOCKETINT *pThis = hSocket;
955	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
956	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
957	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
958
959	int rc = rtSocketSwitchBlockingMode(pThis, true /* fBlocking */);
960	if (RT_FAILURE(rc))
961	return rc;
962
963	/*
964	* Try write all at once.
965	*/
966	#ifdef RTSOCKET_MAX_WRITE
967	int cbNow = cbBuffer >= RTSOCKET_MAX_WRITE ? RTSOCKET_MAX_WRITE : (int)cbBuffer;
968	#else
969	size_t cbNow = cbBuffer >= SSIZE_MAX ? SSIZE_MAX : cbBuffer;
970	#endif
971	ssize_t cbWritten = send(pThis->hNative, (const char *)pvBuffer, cbNow, MSG_NOSIGNAL);
972	if (RT_LIKELY((size_t)cbWritten == cbBuffer && cbWritten >= 0))
973	rc = VINF_SUCCESS;
974	else if (cbWritten < 0)
975	rc = rtSocketError();
976	else
977	{
978	/*
979	* Unfinished business, write the remainder of the request. Must ignore
980	* VERR_INTERRUPTED here if we've managed to send something.
981	*/
982	size_t cbSentSoFar = 0;
983	for (;;)
984	{
985	/* advance */
986	cbBuffer -= (size_t)cbWritten;
987	if (!cbBuffer)
988	break;
989	cbSentSoFar += (size_t)cbWritten;
990	pvBuffer = (char const *)pvBuffer + cbWritten;
991
992	/* send */
993	#ifdef RTSOCKET_MAX_WRITE
994	cbNow = cbBuffer >= RTSOCKET_MAX_WRITE ? RTSOCKET_MAX_WRITE : (int)cbBuffer;
995	#else
996	cbNow = cbBuffer >= SSIZE_MAX ? SSIZE_MAX : cbBuffer;
997	#endif
998	cbWritten = send(pThis->hNative, (const char *)pvBuffer, cbNow, MSG_NOSIGNAL);
999	if (cbWritten >= 0)
1000	AssertMsg(cbBuffer >= (size_t)cbWritten, ("Wrote more than we requested!!! cbWritten=%zu cbBuffer=%zu rtSocketError()=%d\n",
1001	cbWritten, cbBuffer, rtSocketError()));
1002	else
1003	{
1004	rc = rtSocketError();
1005	if (rc != VERR_INTERNAL_ERROR \|\| cbSentSoFar == 0)
1006	break;
1007	cbWritten = 0;
1008	rc = VINF_SUCCESS;
1009	}
1010	}
1011	}
1012
1013	rtSocketUnlock(pThis);
1014	return rc;
1015	}
1016
1017
1018	RTDECL(int) RTSocketWriteTo(RTSOCKET hSocket, const void *pvBuffer, size_t cbBuffer, PCRTNETADDR pAddr)
1019	{
1020	/*
1021	* Validate input.
1022	*/
1023	RTSOCKETINT *pThis = hSocket;
1024	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1025	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1026
1027	/* no locking since UDP reads may be done concurrently to writes, and
1028	* this is the normal use case of this code. */
1029
1030	int rc = rtSocketSwitchBlockingMode(pThis, true /* fBlocking */);
1031	if (RT_FAILURE(rc))
1032	return rc;
1033
1034	/* Figure out destination address. */
1035	struct sockaddr *pSA = NULL;
1036	#ifdef RT_OS_WINDOWS
1037	int cbSA = 0;
1038	#else
1039	socklen_t cbSA = 0;
1040	#endif
1041	RTSOCKADDRUNION u;
1042	if (pAddr)
1043	{
1044	rc = rtSocketAddrFromNetAddr(pAddr, &u, sizeof(u), NULL);
1045	if (RT_FAILURE(rc))
1046	return rc;
1047	pSA = &u.Addr;
1048	cbSA = sizeof(u);
1049	}
1050
1051	/*
1052	* Must write all at once, otherwise it is a failure.
1053	*/
1054	#ifdef RT_OS_WINDOWS
1055	int cbNow = cbBuffer >= RTSOCKET_MAX_WRITE ? RTSOCKET_MAX_WRITE : (int)cbBuffer;
1056	#else
1057	size_t cbNow = cbBuffer >= SSIZE_MAX ? SSIZE_MAX : cbBuffer;
1058	#endif
1059	ssize_t cbWritten = sendto(pThis->hNative, (const char *)pvBuffer, cbNow, MSG_NOSIGNAL, pSA, cbSA);
1060	if (RT_LIKELY((size_t)cbWritten == cbBuffer && cbWritten >= 0))
1061	rc = VINF_SUCCESS;
1062	else if (cbWritten < 0)
1063	rc = rtSocketError();
1064	else
1065	rc = VERR_TOO_MUCH_DATA;
1066
1067	rtSocketUnlock(pThis);
1068	return rc;
1069	}
1070
1071
1072	RTDECL(int) RTSocketSgWrite(RTSOCKET hSocket, PCRTSGBUF pSgBuf)
1073	{
1074	/*
1075	* Validate input.
1076	*/
1077	RTSOCKETINT *pThis = hSocket;
1078	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1079	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1080	AssertPtrReturn(pSgBuf, VERR_INVALID_PARAMETER);
1081	AssertReturn(pSgBuf->cSegs > 0, VERR_INVALID_PARAMETER);
1082	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1083
1084	int rc = rtSocketSwitchBlockingMode(pThis, true /* fBlocking */);
1085	if (RT_FAILURE(rc))
1086	return rc;
1087
1088	/*
1089	* Construct message descriptor (translate pSgBuf) and send it.
1090	*/
1091	rc = VERR_NO_TMP_MEMORY;
1092	#ifdef RT_OS_WINDOWS
1093	AssertCompileSize(WSABUF, sizeof(RTSGSEG));
1094	AssertCompileMemberSize(WSABUF, buf, RT_SIZEOFMEMB(RTSGSEG, pvSeg));
1095
1096	LPWSABUF paMsg = (LPWSABUF)RTMemTmpAllocZ(pSgBuf->cSegs * sizeof(WSABUF));
1097	if (paMsg)
1098	{
1099	for (unsigned i = 0; i < pSgBuf->cSegs; i++)
1100	{
1101	paMsg[i].buf = (char *)pSgBuf->paSegs[i].pvSeg;
1102	paMsg[i].len = (u_long)pSgBuf->paSegs[i].cbSeg;
1103	}
1104
1105	DWORD dwSent;
1106	int hrc = WSASend(pThis->hNative, paMsg, pSgBuf->cSegs, &dwSent,
1107	MSG_NOSIGNAL, NULL, NULL);
1108	if (!hrc)
1109	rc = VINF_SUCCESS;
1110	/** @todo check for incomplete writes */
1111	else
1112	rc = rtSocketError();
1113
1114	RTMemTmpFree(paMsg);
1115	}
1116
1117	#else /* !RT_OS_WINDOWS */
1118	AssertCompileSize(struct iovec, sizeof(RTSGSEG));
1119	AssertCompileMemberSize(struct iovec, iov_base, RT_SIZEOFMEMB(RTSGSEG, pvSeg));
1120	AssertCompileMemberSize(struct iovec, iov_len, RT_SIZEOFMEMB(RTSGSEG, cbSeg));
1121
1122	struct iovec paMsg = (struct iovec )RTMemTmpAllocZ(pSgBuf->cSegs * sizeof(struct iovec));
1123	if (paMsg)
1124	{
1125	for (unsigned i = 0; i < pSgBuf->cSegs; i++)
1126	{
1127	paMsg[i].iov_base = pSgBuf->paSegs[i].pvSeg;
1128	paMsg[i].iov_len = pSgBuf->paSegs[i].cbSeg;
1129	}
1130
1131	struct msghdr msgHdr;
1132	RT_ZERO(msgHdr);
1133	msgHdr.msg_iov = paMsg;
1134	msgHdr.msg_iovlen = pSgBuf->cSegs;
1135	ssize_t cbWritten = sendmsg(pThis->hNative, &msgHdr, MSG_NOSIGNAL);
1136	if (RT_LIKELY(cbWritten >= 0))
1137	rc = VINF_SUCCESS;
1138	/** @todo check for incomplete writes */
1139	else
1140	rc = rtSocketError();
1141
1142	RTMemTmpFree(paMsg);
1143	}
1144	#endif /* !RT_OS_WINDOWS */
1145
1146	rtSocketUnlock(pThis);
1147	return rc;
1148	}
1149
1150
1151	RTDECL(int) RTSocketSgWriteL(RTSOCKET hSocket, size_t cSegs, ...)
1152	{
1153	va_list va;
1154	va_start(va, cSegs);
1155	int rc = RTSocketSgWriteLV(hSocket, cSegs, va);
1156	va_end(va);
1157	return rc;
1158	}
1159
1160
1161	RTDECL(int) RTSocketSgWriteLV(RTSOCKET hSocket, size_t cSegs, va_list va)
1162	{
1163	/*
1164	* Set up a S/G segment array + buffer on the stack and pass it
1165	* on to RTSocketSgWrite.
1166	*/
1167	Assert(cSegs <= 16);
1168	PRTSGSEG paSegs = (PRTSGSEG)alloca(cSegs * sizeof(RTSGSEG));
1169	AssertReturn(paSegs, VERR_NO_TMP_MEMORY);
1170	for (size_t i = 0; i < cSegs; i++)
1171	{
1172	paSegs[i].pvSeg = va_arg(va, void *);
1173	paSegs[i].cbSeg = va_arg(va, size_t);
1174	}
1175
1176	RTSGBUF SgBuf;
1177	RTSgBufInit(&SgBuf, paSegs, cSegs);
1178	return RTSocketSgWrite(hSocket, &SgBuf);
1179	}
1180
1181
1182	RTDECL(int) RTSocketReadNB(RTSOCKET hSocket, void pvBuffer, size_t cbBuffer, size_t pcbRead)
1183	{
1184	/*
1185	* Validate input.
1186	*/
1187	RTSOCKETINT *pThis = hSocket;
1188	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1189	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1190	AssertReturn(cbBuffer > 0, VERR_INVALID_PARAMETER);
1191	AssertPtr(pvBuffer);
1192	AssertPtrReturn(pcbRead, VERR_INVALID_PARAMETER);
1193	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1194
1195	int rc = rtSocketSwitchBlockingMode(pThis, false /* fBlocking */);
1196	if (RT_FAILURE(rc))
1197	return rc;
1198
1199	rtSocketErrorReset();
1200	#ifdef RTSOCKET_MAX_READ
1201	int cbNow = cbBuffer >= RTSOCKET_MAX_WRITE ? RTSOCKET_MAX_WRITE : (int)cbBuffer;
1202	#else
1203	size_t cbNow = cbBuffer;
1204	#endif
1205
1206	#ifdef RT_OS_WINDOWS
1207	int cbRead = recv(pThis->hNative, (char *)pvBuffer, cbNow, MSG_NOSIGNAL);
1208	if (cbRead >= 0)
1209	{
1210	*pcbRead = cbRead;
1211	rc = VINF_SUCCESS;
1212	}
1213	else
1214	rc = rtSocketError();
1215
1216	if (rc == VERR_TRY_AGAIN)
1217	rc = VINF_TRY_AGAIN;
1218	#else
1219	ssize_t cbRead = recv(pThis->hNative, pvBuffer, cbNow, MSG_NOSIGNAL);
1220	if (cbRead >= 0)
1221	*pcbRead = cbRead;
1222	else if (errno == EAGAIN)
1223	{
1224	*pcbRead = 0;
1225	rc = VINF_TRY_AGAIN;
1226	}
1227	else
1228	rc = rtSocketError();
1229	#endif
1230
1231	rtSocketUnlock(pThis);
1232	return rc;
1233	}
1234
1235
1236	RTDECL(int) RTSocketWriteNB(RTSOCKET hSocket, const void pvBuffer, size_t cbBuffer, size_t pcbWritten)
1237	{
1238	/*
1239	* Validate input.
1240	*/
1241	RTSOCKETINT *pThis = hSocket;
1242	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1243	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1244	AssertPtrReturn(pcbWritten, VERR_INVALID_PARAMETER);
1245	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1246
1247	int rc = rtSocketSwitchBlockingMode(pThis, false /* fBlocking */);
1248	if (RT_FAILURE(rc))
1249	return rc;
1250
1251	rtSocketErrorReset();
1252	#ifdef RTSOCKET_MAX_WRITE
1253	int cbNow = cbBuffer >= RTSOCKET_MAX_WRITE ? RTSOCKET_MAX_WRITE : (int)cbBuffer;
1254	#else
1255	size_t cbNow = cbBuffer;
1256	#endif
1257
1258	#ifdef RT_OS_WINDOWS
1259	int cbWritten = send(pThis->hNative, (const char *)pvBuffer, cbNow, MSG_NOSIGNAL);
1260	if (cbWritten >= 0)
1261	{
1262	*pcbWritten = cbWritten;
1263	rc = VINF_SUCCESS;
1264	}
1265	else
1266	rc = rtSocketError();
1267
1268	if (rc == VERR_TRY_AGAIN)
1269	rc = VINF_TRY_AGAIN;
1270	#else
1271	ssize_t cbWritten = send(pThis->hNative, pvBuffer, cbBuffer, MSG_NOSIGNAL);
1272	if (cbWritten >= 0)
1273	*pcbWritten = cbWritten;
1274	else if (errno == EAGAIN)
1275	{
1276	*pcbWritten = 0;
1277	rc = VINF_TRY_AGAIN;
1278	}
1279	else
1280	rc = rtSocketError();
1281	#endif
1282
1283	rtSocketUnlock(pThis);
1284	return rc;
1285	}
1286
1287
1288	RTDECL(int) RTSocketSgWriteNB(RTSOCKET hSocket, PCRTSGBUF pSgBuf, size_t *pcbWritten)
1289	{
1290	/*
1291	* Validate input.
1292	*/
1293	RTSOCKETINT *pThis = hSocket;
1294	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1295	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1296	AssertPtrReturn(pSgBuf, VERR_INVALID_PARAMETER);
1297	AssertPtrReturn(pcbWritten, VERR_INVALID_PARAMETER);
1298	AssertReturn(pSgBuf->cSegs > 0, VERR_INVALID_PARAMETER);
1299	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1300
1301	int rc = rtSocketSwitchBlockingMode(pThis, false /* fBlocking */);
1302	if (RT_FAILURE(rc))
1303	return rc;
1304
1305	unsigned cSegsToSend = 0;
1306	rc = VERR_NO_TMP_MEMORY;
1307	#ifdef RT_OS_WINDOWS
1308	LPWSABUF paMsg = NULL;
1309
1310	RTSgBufMapToNative(paMsg, pSgBuf, WSABUF, buf, char *, len, u_long, cSegsToSend);
1311	if (paMsg)
1312	{
1313	DWORD dwSent = 0;
1314	int hrc = WSASend(pThis->hNative, paMsg, cSegsToSend, &dwSent,
1315	MSG_NOSIGNAL, NULL, NULL);
1316	if (!hrc)
1317	rc = VINF_SUCCESS;
1318	else
1319	rc = rtSocketError();
1320
1321	*pcbWritten = dwSent;
1322
1323	RTMemTmpFree(paMsg);
1324	}
1325
1326	#else /* !RT_OS_WINDOWS */
1327	struct iovec *paMsg = NULL;
1328
1329	RTSgBufMapToNative(paMsg, pSgBuf, struct iovec, iov_base, void *, iov_len, size_t, cSegsToSend);
1330	if (paMsg)
1331	{
1332	struct msghdr msgHdr;
1333	RT_ZERO(msgHdr);
1334	msgHdr.msg_iov = paMsg;
1335	msgHdr.msg_iovlen = cSegsToSend;
1336	ssize_t cbWritten = sendmsg(pThis->hNative, &msgHdr, MSG_NOSIGNAL);
1337	if (RT_LIKELY(cbWritten >= 0))
1338	{
1339	rc = VINF_SUCCESS;
1340	*pcbWritten = cbWritten;
1341	}
1342	else
1343	rc = rtSocketError();
1344
1345	RTMemTmpFree(paMsg);
1346	}
1347	#endif /* !RT_OS_WINDOWS */
1348
1349	rtSocketUnlock(pThis);
1350	return rc;
1351	}
1352
1353
1354	RTDECL(int) RTSocketSgWriteLNB(RTSOCKET hSocket, size_t cSegs, size_t *pcbWritten, ...)
1355	{
1356	va_list va;
1357	va_start(va, pcbWritten);
1358	int rc = RTSocketSgWriteLVNB(hSocket, cSegs, pcbWritten, va);
1359	va_end(va);
1360	return rc;
1361	}
1362
1363
1364	RTDECL(int) RTSocketSgWriteLVNB(RTSOCKET hSocket, size_t cSegs, size_t *pcbWritten, va_list va)
1365	{
1366	/*
1367	* Set up a S/G segment array + buffer on the stack and pass it
1368	* on to RTSocketSgWrite.
1369	*/
1370	Assert(cSegs <= 16);
1371	PRTSGSEG paSegs = (PRTSGSEG)alloca(cSegs * sizeof(RTSGSEG));
1372	AssertReturn(paSegs, VERR_NO_TMP_MEMORY);
1373	for (size_t i = 0; i < cSegs; i++)
1374	{
1375	paSegs[i].pvSeg = va_arg(va, void *);
1376	paSegs[i].cbSeg = va_arg(va, size_t);
1377	}
1378
1379	RTSGBUF SgBuf;
1380	RTSgBufInit(&SgBuf, paSegs, cSegs);
1381	return RTSocketSgWriteNB(hSocket, &SgBuf, pcbWritten);
1382	}
1383
1384
1385	RTDECL(int) RTSocketSelectOne(RTSOCKET hSocket, RTMSINTERVAL cMillies)
1386	{
1387	/*
1388	* Validate input.
1389	*/
1390	RTSOCKETINT *pThis = hSocket;
1391	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1392	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1393	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
1394	int const fdMax = (int)pThis->hNative + 1;
1395	AssertReturn(fdMax - 1 == pThis->hNative, VERR_INTERNAL_ERROR_5);
1396
1397	/*
1398	* Set up the file descriptor sets and do the select.
1399	*/
1400	fd_set fdsetR;
1401	FD_ZERO(&fdsetR);
1402	FD_SET(pThis->hNative, &fdsetR);
1403
1404	fd_set fdsetE = fdsetR;
1405
1406	int rc;
1407	if (cMillies == RT_INDEFINITE_WAIT)
1408	rc = select(fdMax, &fdsetR, NULL, &fdsetE, NULL);
1409	else
1410	{
1411	struct timeval timeout;
1412	timeout.tv_sec = cMillies / 1000;
1413	timeout.tv_usec = (cMillies % 1000) * 1000;
1414	rc = select(fdMax, &fdsetR, NULL, &fdsetE, &timeout);
1415	}
1416	if (rc > 0)
1417	rc = VINF_SUCCESS;
1418	else if (rc == 0)
1419	rc = VERR_TIMEOUT;
1420	else
1421	rc = rtSocketError();
1422
1423	return rc;
1424	}
1425
1426
1427	RTDECL(int) RTSocketSelectOneEx(RTSOCKET hSocket, uint32_t fEvents, uint32_t *pfEvents,
1428	RTMSINTERVAL cMillies)
1429	{
1430	/*
1431	* Validate input.
1432	*/
1433	RTSOCKETINT *pThis = hSocket;
1434	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1435	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1436	AssertPtrReturn(pfEvents, VERR_INVALID_PARAMETER);
1437	AssertReturn(!(fEvents & ~RTSOCKET_EVT_VALID_MASK), VERR_INVALID_PARAMETER);
1438	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
1439	int const fdMax = (int)pThis->hNative + 1;
1440	AssertReturn(fdMax - 1 == pThis->hNative, VERR_INTERNAL_ERROR_5);
1441
1442	*pfEvents = 0;
1443
1444	/*
1445	* Set up the file descriptor sets and do the select.
1446	*/
1447	fd_set fdsetR;
1448	fd_set fdsetW;
1449	fd_set fdsetE;
1450	FD_ZERO(&fdsetR);
1451	FD_ZERO(&fdsetW);
1452	FD_ZERO(&fdsetE);
1453
1454	if (fEvents & RTSOCKET_EVT_READ)
1455	FD_SET(pThis->hNative, &fdsetR);
1456	if (fEvents & RTSOCKET_EVT_WRITE)
1457	FD_SET(pThis->hNative, &fdsetW);
1458	if (fEvents & RTSOCKET_EVT_ERROR)
1459	FD_SET(pThis->hNative, &fdsetE);
1460
1461	int rc;
1462	if (cMillies == RT_INDEFINITE_WAIT)
1463	rc = select(fdMax, &fdsetR, &fdsetW, &fdsetE, NULL);
1464	else
1465	{
1466	struct timeval timeout;
1467	timeout.tv_sec = cMillies / 1000;
1468	timeout.tv_usec = (cMillies % 1000) * 1000;
1469	rc = select(fdMax, &fdsetR, &fdsetW, &fdsetE, &timeout);
1470	}
1471	if (rc > 0)
1472	{
1473	if (FD_ISSET(pThis->hNative, &fdsetR))
1474	*pfEvents \|= RTSOCKET_EVT_READ;
1475	if (FD_ISSET(pThis->hNative, &fdsetW))
1476	*pfEvents \|= RTSOCKET_EVT_WRITE;
1477	if (FD_ISSET(pThis->hNative, &fdsetE))
1478	*pfEvents \|= RTSOCKET_EVT_ERROR;
1479
1480	rc = VINF_SUCCESS;
1481	}
1482	else if (rc == 0)
1483	rc = VERR_TIMEOUT;
1484	else
1485	rc = rtSocketError();
1486
1487	return rc;
1488	}
1489
1490
1491	RTDECL(int) RTSocketShutdown(RTSOCKET hSocket, bool fRead, bool fWrite)
1492	{
1493	/*
1494	* Validate input, don't lock it because we might want to interrupt a call
1495	* active on a different thread.
1496	*/
1497	RTSOCKETINT *pThis = hSocket;
1498	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1499	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1500	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
1501	AssertReturn(fRead \|\| fWrite, VERR_INVALID_PARAMETER);
1502
1503	/*
1504	* Do the job.
1505	*/
1506	int rc = VINF_SUCCESS;
1507	int fHow;
1508	if (fRead && fWrite)
1509	fHow = SHUT_RDWR;
1510	else if (fRead)
1511	fHow = SHUT_RD;
1512	else
1513	fHow = SHUT_WR;
1514	if (shutdown(pThis->hNative, fHow) == -1)
1515	rc = rtSocketError();
1516
1517	return rc;
1518	}
1519
1520
1521	RTDECL(int) RTSocketGetLocalAddress(RTSOCKET hSocket, PRTNETADDR pAddr)
1522	{
1523	/*
1524	* Validate input.
1525	*/
1526	RTSOCKETINT *pThis = hSocket;
1527	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1528	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1529	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
1530
1531	/*
1532	* Get the address and convert it.
1533	*/
1534	int rc;
1535	RTSOCKADDRUNION u;
1536	#ifdef RT_OS_WINDOWS
1537	int cbAddr = sizeof(u);
1538	#else
1539	socklen_t cbAddr = sizeof(u);
1540	#endif
1541	RT_ZERO(u);
1542	if (getsockname(pThis->hNative, &u.Addr, &cbAddr) == 0)
1543	rc = rtSocketNetAddrFromAddr(&u, cbAddr, pAddr);
1544	else
1545	rc = rtSocketError();
1546
1547	return rc;
1548	}
1549
1550
1551	RTDECL(int) RTSocketGetPeerAddress(RTSOCKET hSocket, PRTNETADDR pAddr)
1552	{
1553	/*
1554	* Validate input.
1555	*/
1556	RTSOCKETINT *pThis = hSocket;
1557	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1558	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1559	AssertReturn(RTMemPoolRefCount(pThis) >= (pThis->cUsers ? 2U : 1U), VERR_CALLER_NO_REFERENCE);
1560
1561	/*
1562	* Get the address and convert it.
1563	*/
1564	int rc;
1565	RTSOCKADDRUNION u;
1566	#ifdef RT_OS_WINDOWS
1567	int cbAddr = sizeof(u);
1568	#else
1569	socklen_t cbAddr = sizeof(u);
1570	#endif
1571	RT_ZERO(u);
1572	if (getpeername(pThis->hNative, &u.Addr, &cbAddr) == 0)
1573	rc = rtSocketNetAddrFromAddr(&u, cbAddr, pAddr);
1574	else
1575	rc = rtSocketError();
1576
1577	return rc;
1578	}
1579
1580
1581
1582	/**
1583	* Wrapper around bind.
1584	*
1585	* @returns IPRT status code.
1586	* @param hSocket The socket handle.
1587	* @param pAddr The address to bind to.
1588	*/
1589	int rtSocketBind(RTSOCKET hSocket, PCRTNETADDR pAddr)
1590	{
1591	RTSOCKADDRUNION u;
1592	int cbAddr;
1593	int rc = rtSocketAddrFromNetAddr(pAddr, &u, sizeof(u), &cbAddr);
1594	if (RT_SUCCESS(rc))
1595	rc = rtSocketBindRawAddr(hSocket, &u.Addr, cbAddr);
1596	return rc;
1597	}
1598
1599
1600	/**
1601	* Very thin wrapper around bind.
1602	*
1603	* @returns IPRT status code.
1604	* @param hSocket The socket handle.
1605	* @param pvAddr The address to bind to (struct sockaddr and
1606	* friends).
1607	* @param cbAddr The size of the address.
1608	*/
1609	int rtSocketBindRawAddr(RTSOCKET hSocket, void const *pvAddr, size_t cbAddr)
1610	{
1611	/*
1612	* Validate input.
1613	*/
1614	RTSOCKETINT *pThis = hSocket;
1615	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1616	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1617	AssertPtrReturn(pvAddr, VERR_INVALID_POINTER);
1618	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1619
1620	int rc;
1621	if (bind(pThis->hNative, (struct sockaddr const *)pvAddr, cbAddr) == 0)
1622	rc = VINF_SUCCESS;
1623	else
1624	rc = rtSocketError();
1625
1626	rtSocketUnlock(pThis);
1627	return rc;
1628	}
1629
1630
1631
1632	/**
1633	* Wrapper around listen.
1634	*
1635	* @returns IPRT status code.
1636	* @param hSocket The socket handle.
1637	* @param cMaxPending The max number of pending connections.
1638	*/
1639	int rtSocketListen(RTSOCKET hSocket, int cMaxPending)
1640	{
1641	/*
1642	* Validate input.
1643	*/
1644	RTSOCKETINT *pThis = hSocket;
1645	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1646	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1647	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1648
1649	int rc = VINF_SUCCESS;
1650	if (listen(pThis->hNative, cMaxPending) != 0)
1651	rc = rtSocketError();
1652
1653	rtSocketUnlock(pThis);
1654	return rc;
1655	}
1656
1657
1658	/**
1659	* Wrapper around accept.
1660	*
1661	* @returns IPRT status code.
1662	* @param hSocket The socket handle.
1663	* @param phClient Where to return the client socket handle on
1664	* success.
1665	* @param pAddr Where to return the client address.
1666	* @param pcbAddr On input this gives the size buffer size of what
1667	* @a pAddr point to. On return this contains the
1668	* size of what's stored at @a pAddr.
1669	*/
1670	int rtSocketAccept(RTSOCKET hSocket, PRTSOCKET phClient, struct sockaddr pAddr, size_t pcbAddr)
1671	{
1672	/*
1673	* Validate input.
1674	* Only lock the socket temporarily while we get the native handle, so that
1675	* we can safely shutdown and destroy the socket from a different thread.
1676	*/
1677	RTSOCKETINT *pThis = hSocket;
1678	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1679	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1680	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1681
1682	/*
1683	* Call accept().
1684	*/
1685	rtSocketErrorReset();
1686	int rc = VINF_SUCCESS;
1687	#ifdef RT_OS_WINDOWS
1688	int cbAddr = (int)*pcbAddr;
1689	#else
1690	socklen_t cbAddr = *pcbAddr;
1691	#endif
1692	RTSOCKETNATIVE hNativeClient = accept(pThis->hNative, pAddr, &cbAddr);
1693	if (hNativeClient != NIL_RTSOCKETNATIVE)
1694	{
1695	*pcbAddr = cbAddr;
1696
1697	/*
1698	* Wrap the client socket.
1699	*/
1700	rc = rtSocketCreateForNative(phClient, hNativeClient);
1701	if (RT_FAILURE(rc))
1702	{
1703	#ifdef RT_OS_WINDOWS
1704	closesocket(hNativeClient);
1705	#else
1706	close(hNativeClient);
1707	#endif
1708	}
1709	}
1710	else
1711	rc = rtSocketError();
1712
1713	rtSocketUnlock(pThis);
1714	return rc;
1715	}
1716
1717
1718	/**
1719	* Wrapper around connect.
1720	*
1721	* @returns IPRT status code.
1722	* @param hSocket The socket handle.
1723	* @param pAddr The socket address to connect to.
1724	*/
1725	int rtSocketConnect(RTSOCKET hSocket, PCRTNETADDR pAddr)
1726	{
1727	/*
1728	* Validate input.
1729	*/
1730	RTSOCKETINT *pThis = hSocket;
1731	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1732	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1733	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1734
1735	RTSOCKADDRUNION u;
1736	int cbAddr;
1737	int rc = rtSocketAddrFromNetAddr(pAddr, &u, sizeof(u), &cbAddr);
1738	if (RT_SUCCESS(rc))
1739	{
1740	if (connect(pThis->hNative, &u.Addr, cbAddr) != 0)
1741	rc = rtSocketError();
1742	}
1743
1744	rtSocketUnlock(pThis);
1745	return rc;
1746	}
1747
1748
1749	/**
1750	* Wrapper around setsockopt.
1751	*
1752	* @returns IPRT status code.
1753	* @param hSocket The socket handle.
1754	* @param iLevel The protocol level, e.g. IPPORTO_TCP.
1755	* @param iOption The option, e.g. TCP_NODELAY.
1756	* @param pvValue The value buffer.
1757	* @param cbValue The size of the value pointed to by pvValue.
1758	*/
1759	int rtSocketSetOpt(RTSOCKET hSocket, int iLevel, int iOption, void const *pvValue, int cbValue)
1760	{
1761	/*
1762	* Validate input.
1763	*/
1764	RTSOCKETINT *pThis = hSocket;
1765	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1766	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1767	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1768
1769	int rc = VINF_SUCCESS;
1770	if (setsockopt(pThis->hNative, iLevel, iOption, (const char *)pvValue, cbValue) != 0)
1771	rc = rtSocketError();
1772
1773	rtSocketUnlock(pThis);
1774	return rc;
1775	}
1776
1777
1778	/**
1779	* Internal RTPollSetAdd helper that returns the handle that should be added to
1780	* the pollset.
1781	*
1782	* @returns Valid handle on success, INVALID_HANDLE_VALUE on failure.
1783	* @param hSocket The socket handle.
1784	* @param fEvents The events we're polling for.
1785	* @param phNative Where to put the primary handle.
1786	*/
1787	int rtSocketPollGetHandle(RTSOCKET hSocket, uint32_t fEvents, PRTHCINTPTR phNative)
1788	{
1789	RTSOCKETINT *pThis = hSocket;
1790	AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
1791	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, VERR_INVALID_HANDLE);
1792	#ifdef RT_OS_WINDOWS
1793	AssertReturn(rtSocketTryLock(pThis), VERR_CONCURRENT_ACCESS);
1794
1795	int rc = VINF_SUCCESS;
1796	if (pThis->hEvent != WSA_INVALID_EVENT)
1797	*phNative = (RTHCINTPTR)pThis->hEvent;
1798	else
1799	{
1800	pThis->hEvent = WSACreateEvent();
1801	*phNative = (RTHCINTPTR)pThis->hEvent;
1802	if (pThis->hEvent == WSA_INVALID_EVENT)
1803	rc = rtSocketError();
1804	}
1805
1806	rtSocketUnlock(pThis);
1807	return rc;
1808
1809	#else /* !RT_OS_WINDOWS */
1810	*phNative = (RTHCUINTPTR)pThis->hNative;
1811	return VINF_SUCCESS;
1812	#endif /* !RT_OS_WINDOWS */
1813	}
1814
1815	#ifdef RT_OS_WINDOWS
1816
1817	/**
1818	* Undos the harm done by WSAEventSelect.
1819	*
1820	* @returns IPRT status code.
1821	* @param pThis The socket handle.
1822	*/
1823	static int rtSocketPollClearEventAndRestoreBlocking(RTSOCKETINT *pThis)
1824	{
1825	int rc = VINF_SUCCESS;
1826	if (pThis->fSubscribedEvts)
1827	{
1828	if (WSAEventSelect(pThis->hNative, WSA_INVALID_EVENT, 0) == 0)
1829	{
1830	pThis->fSubscribedEvts = 0;
1831
1832	/*
1833	* Switch back to blocking mode if that was the state before the
1834	* operation.
1835	*/
1836	if (pThis->fBlocking)
1837	{
1838	u_long fNonBlocking = 0;
1839	int rc2 = ioctlsocket(pThis->hNative, FIONBIO, &fNonBlocking);
1840	if (rc2 != 0)
1841	{
1842	rc = rtSocketError();
1843	AssertMsgFailed(("%Rrc; rc2=%d\n", rc, rc2));
1844	}
1845	}
1846	}
1847	else
1848	{
1849	rc = rtSocketError();
1850	AssertMsgFailed(("%Rrc\n", rc));
1851	}
1852	}
1853	return rc;
1854	}
1855
1856
1857	/**
1858	* Updates the mask of events we're subscribing to.
1859	*
1860	* @returns IPRT status code.
1861	* @param pThis The socket handle.
1862	* @param fEvents The events we want to subscribe to.
1863	*/
1864	static int rtSocketPollUpdateEvents(RTSOCKETINT *pThis, uint32_t fEvents)
1865	{
1866	LONG fNetworkEvents = 0;
1867	if (fEvents & RTPOLL_EVT_READ)
1868	fNetworkEvents \|= FD_READ;
1869	if (fEvents & RTPOLL_EVT_WRITE)
1870	fNetworkEvents \|= FD_WRITE;
1871	if (fEvents & RTPOLL_EVT_ERROR)
1872	fNetworkEvents \|= FD_CLOSE;
1873	LogFlowFunc(("fNetworkEvents=%#x\n", fNetworkEvents));
1874	if (WSAEventSelect(pThis->hNative, pThis->hEvent, fNetworkEvents) == 0)
1875	{
1876	pThis->fSubscribedEvts = fEvents;
1877	return VINF_SUCCESS;
1878	}
1879
1880	int rc = rtSocketError();
1881	AssertMsgFailed(("fNetworkEvents=%#x rc=%Rrc\n", fNetworkEvents, rtSocketError()));
1882	return rc;
1883	}
1884
1885	#endif /* RT_OS_WINDOWS */
1886
1887
1888	#if defined(RT_OS_WINDOWS) \|\| defined(RT_OS_OS2)
1889
1890	/**
1891	* Checks for pending events.
1892	*
1893	* @returns Event mask or 0.
1894	* @param pThis The socket handle.
1895	* @param fEvents The desired events.
1896	*/
1897	static uint32_t rtSocketPollCheck(RTSOCKETINT *pThis, uint32_t fEvents)
1898	{
1899	uint32_t fRetEvents = 0;
1900
1901	LogFlowFunc(("pThis=%#p fEvents=%#x\n", pThis, fEvents));
1902
1903	# ifdef RT_OS_WINDOWS
1904	/* Make sure WSAEnumNetworkEvents returns what we want. */
1905	int rc = VINF_SUCCESS;
1906	if ((pThis->fSubscribedEvts & fEvents) != fEvents)
1907	rc = rtSocketPollUpdateEvents(pThis, pThis->fSubscribedEvts \| fEvents);
1908
1909	/* Get the event mask, ASSUMES that WSAEnumNetworkEvents doesn't clear stuff. */
1910	WSANETWORKEVENTS NetEvts;
1911	RT_ZERO(NetEvts);
1912	if (WSAEnumNetworkEvents(pThis->hNative, pThis->hEvent, &NetEvts) == 0)
1913	{
1914	if ( (NetEvts.lNetworkEvents & FD_READ)
1915	&& (fEvents & RTPOLL_EVT_READ)
1916	&& NetEvts.iErrorCode[FD_READ_BIT] == 0)
1917	fRetEvents \|= RTPOLL_EVT_READ;
1918
1919	if ( (NetEvts.lNetworkEvents & FD_WRITE)
1920	&& (fEvents & RTPOLL_EVT_WRITE)
1921	&& NetEvts.iErrorCode[FD_WRITE_BIT] == 0)
1922	fRetEvents \|= RTPOLL_EVT_WRITE;
1923
1924	if (fEvents & RTPOLL_EVT_ERROR)
1925	{
1926	if (NetEvts.lNetworkEvents & FD_CLOSE)
1927	fRetEvents \|= RTPOLL_EVT_ERROR;
1928	else
1929	for (uint32_t i = 0; i < FD_MAX_EVENTS; i++)
1930	if ( (NetEvts.lNetworkEvents & (1L << i))
1931	&& NetEvts.iErrorCode[i] != 0)
1932	fRetEvents \|= RTPOLL_EVT_ERROR;
1933	}
1934	}
1935	else
1936	rc = rtSocketError();
1937
1938	/* Fall back on select if we hit an error above. */
1939	if (RT_FAILURE(rc))
1940	{
1941
1942	}
1943
1944	#else /* RT_OS_OS2 */
1945	int aFds[4] = { pThis->hNative, pThis->hNative, pThis->hNative, -1 };
1946	int rc = os2_select(aFds, 1, 1, 1, 0);
1947	if (rc > 0)
1948	{
1949	if (aFds[0] == pThis->hNative)
1950	fRetEvents \|= RTPOLL_EVT_READ;
1951	if (aFds[1] == pThis->hNative)
1952	fRetEvents \|= RTPOLL_EVT_WRITE;
1953	if (aFds[2] == pThis->hNative)
1954	fRetEvents \|= RTPOLL_EVT_ERROR;
1955	fRetEvents &= fEvents;
1956	}
1957	#endif /* RT_OS_OS2 */
1958
1959	LogFlowFunc(("fRetEvents=%#x\n", fRetEvents));
1960	return fRetEvents;
1961	}
1962
1963
1964	/**
1965	* Internal RTPoll helper that polls the socket handle and, if @a fNoWait is
1966	* clear, starts whatever actions we've got running during the poll call.
1967	*
1968	* @returns 0 if no pending events, actions initiated if @a fNoWait is clear.
1969	* Event mask (in @a fEvents) and no actions if the handle is ready
1970	* already.
1971	* UINT32_MAX (asserted) if the socket handle is busy in I/O or a
1972	* different poll set.
1973	*
1974	* @param hSocket The socket handle.
1975	* @param hPollSet The poll set handle (for access checks).
1976	* @param fEvents The events we're polling for.
1977	* @param fFinalEntry Set if this is the final entry for this handle
1978	* in this poll set. This can be used for dealing
1979	* with duplicate entries.
1980	* @param fNoWait Set if it's a zero-wait poll call. Clear if
1981	* we'll wait for an event to occur.
1982	*
1983	* @remarks There is a potential race wrt duplicate handles when @a fNoWait is
1984	* @c true, we don't currently care about that oddity...
1985	*/
1986	uint32_t rtSocketPollStart(RTSOCKET hSocket, RTPOLLSET hPollSet, uint32_t fEvents, bool fFinalEntry, bool fNoWait)
1987	{
1988	RTSOCKETINT *pThis = hSocket;
1989	AssertPtrReturn(pThis, UINT32_MAX);
1990	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, UINT32_MAX);
1991	/** @todo This isn't quite sane. Replace by critsect and open up concurrent
1992	* reads and writes! */
1993	if (rtSocketTryLock(pThis))
1994	pThis->hPollSet = hPollSet;
1995	else
1996	{
1997	AssertReturn(pThis->hPollSet == hPollSet, UINT32_MAX);
1998	ASMAtomicIncU32(&pThis->cUsers);
1999	}
2000
2001	/* (rtSocketPollCheck will reset the event object). */
2002	# ifdef RT_OS_WINDOWS
2003	uint32_t fRetEvents = pThis->fEventsSaved;
2004	pThis->fEventsSaved = 0; /* Reset */
2005	fRetEvents \|= rtSocketPollCheck(pThis, fEvents);
2006
2007	if ( !fRetEvents
2008	&& !fNoWait)
2009	{
2010	pThis->fPollEvts \|= fEvents;
2011	if ( fFinalEntry
2012	&& pThis->fSubscribedEvts != pThis->fPollEvts)
2013	{
2014	int rc = rtSocketPollUpdateEvents(pThis, pThis->fPollEvts);
2015	if (RT_FAILURE(rc))
2016	{
2017	pThis->fPollEvts = 0;
2018	fRetEvents = UINT32_MAX;
2019	}
2020	}
2021	}
2022	# else
2023	uint32_t fRetEvents = rtSocketPollCheck(pThis, fEvents);
2024	# endif
2025
2026	if (fRetEvents \|\| fNoWait)
2027	{
2028	if (pThis->cUsers == 1)
2029	{
2030	# ifdef RT_OS_WINDOWS
2031	rtSocketPollClearEventAndRestoreBlocking(pThis);
2032	# endif
2033	pThis->hPollSet = NIL_RTPOLLSET;
2034	}
2035	ASMAtomicDecU32(&pThis->cUsers);
2036	}
2037
2038	return fRetEvents;
2039	}
2040
2041
2042	/**
2043	* Called after a WaitForMultipleObjects returned in order to check for pending
2044	* events and stop whatever actions that rtSocketPollStart() initiated.
2045	*
2046	* @returns Event mask or 0.
2047	*
2048	* @param hSocket The socket handle.
2049	* @param fEvents The events we're polling for.
2050	* @param fFinalEntry Set if this is the final entry for this handle
2051	* in this poll set. This can be used for dealing
2052	* with duplicate entries. Only keep in mind that
2053	* this method is called in reverse order, so the
2054	* first call will have this set (when the entire
2055	* set was processed).
2056	* @param fHarvestEvents Set if we should check for pending events.
2057	*/
2058	uint32_t rtSocketPollDone(RTSOCKET hSocket, uint32_t fEvents, bool fFinalEntry, bool fHarvestEvents)
2059	{
2060	RTSOCKETINT *pThis = hSocket;
2061	AssertPtrReturn(pThis, 0);
2062	AssertReturn(pThis->u32Magic == RTSOCKET_MAGIC, 0);
2063	Assert(pThis->cUsers > 0);
2064	Assert(pThis->hPollSet != NIL_RTPOLLSET);
2065
2066	/* Harvest events and clear the event mask for the next round of polling. */
2067	uint32_t fRetEvents = rtSocketPollCheck(pThis, fEvents);
2068	# ifdef RT_OS_WINDOWS
2069	pThis->fPollEvts = 0;
2070
2071	/*
2072	* Save the write event if required.
2073	* It is only posted once and might get lost if the another source in the
2074	* pollset with a higher priority has pending events.
2075	*/
2076	if ( !fHarvestEvents
2077	&& fRetEvents)
2078	{
2079	pThis->fEventsSaved = fRetEvents;
2080	fRetEvents = 0;
2081	}
2082	# endif
2083
2084	/* Make the socket blocking again and unlock the handle. */
2085	if (pThis->cUsers == 1)
2086	{
2087	# ifdef RT_OS_WINDOWS
2088	rtSocketPollClearEventAndRestoreBlocking(pThis);
2089	# endif
2090	pThis->hPollSet = NIL_RTPOLLSET;
2091	}
2092	ASMAtomicDecU32(&pThis->cUsers);
2093	return fRetEvents;
2094	}
2095
2096	#endif /* RT_OS_WINDOWS \|\| RT_OS_OS2 */
2097

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

source: vbox/trunk/src/VBox/Runtime/r3/socket.cpp@ 53487

Download in other formats: