summaryrefslogtreecommitdiff
path: root/libusb/os/windows_compat.c
blob: 180ff0c9560c317dbda813e588a3eff84e222580 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
/*
 * Windows compat: POSIX compatibility wrapper
 * Copyright (C) 2009 Pete Batard <pbatard@gmail.com>
 *
 * Parts of poll implementation from libusb-win32, by Stephan Meyer et al.
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 *
 */

/*
 * Posix poll() and pipe() Windows compatibility layer for libusb 1.0
 *
 * The way this layer works is by using OVERLAPPED with async I/O transfers, as
 * OVERLAPPED have an associated event which is flagged for I/O completion.
 * 
 * For USB pollable async I/O, you would typically:
 * - obtain a Windows HANDLE to a file or device that has been opened in
 *   OVERLAPPED mode
 * - call create_fd_for_poll with this handle to obtain a custom fd.
 *   Note that if you need simultaneous R/W access, you need to call create_fd
 *   twice, once in _O_RDONLY and once in _O_WRONLY mode to obtain 2 separate
 *   pollable fds
 * - leave the core functions call the poll routine and flag POLLIN/POLLOUT
 * 
 * For pipe pollable synchronous I/O (read end polling only), you would:
 * - create an anonymous pipe with pipe_for_poll to obtain 2 fds (r & w)
 * - use write_for_poll / read_for_poll to write to either end of the pipe
 * - use poll to check for data to read
 * Note that the read_for_poll/write_for_poll function actually perform 
 * asynchronous I/O internally, and could potentially be modified to support
 * O_NON_BLOCK
 *
 * The way the polling on read_for_poll works is by splitting all read I/O
 * into a dual 1 byte/n-1 bytes asynchronous read operation.
 * The 1 byte data (called the marker), is always armed for asynchronous
 * readout, so that as soon as data becomes available, an OVERLAPPED event
 * will be flagged, which poll can report.
 * Then during the read_for_poll routine itself, this 1 byte marker is copied
 * to the buffer, along with the rest of the data.
 *
 * Note that, since most I/O is buffered, being notified when only the first
 * byte of data is available is unlikely to delay read operations, since the
 * rest of the data should be available in system buffers by the time read
 * is called.
 *
 * Also note that if you don't use read_for_poll to read inbound data, but
 * use the OVERLAPPED directly (which is what we do in the USB async I/O 
 * functions), the marker is not used at all.
 */

/*
 * TODO: Once MinGW supports it (or for other compilation platforms) use
 * CancelIoEx instead of CancelIo for Vista and later platforms
 */
#include <windows.h>
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <pthread.h>
#include <io.h>

#include "windows_compat.h"

// Uncomment to debug the polling layer
//#define DEBUG_WINDOWS_COMPAT
#if defined(DEBUG_WINDOWS_COMPAT)
#define printb printf
#else
// MSVC6 cannot use a variadic argument and non MSVC
// compilers produce warnings if parenthesis are ommitted.
#if defined(_MSC_VER)
#define printb
#else
#define printb(...)
#endif
#endif

#if defined(_PREFAST_)
#pragma warning(disable:28719)
#endif

#if defined(__CYGWIN__ )
// cygwin produces a warning unless these prototypes are defined
extern int _close(int fd);
extern int _snprintf(char *buffer, size_t count, const char *format, ...);
extern int cygwin_attach_handle_to_fd(char *name, int fd, HANDLE handle, int bin, int access); 
// _open_osfhandle() is not available on cygwin, but we can emulate
// it for our needs with cygwin_attach_handle_to_fd()
static inline int _open_osfhandle(intptr_t osfhandle, int flags)
{
	int access;
	switch (flags) {
	case _O_RDONLY:
		access = GENERIC_READ;
		break;
	case _O_WRONLY:
		access = GENERIC_WRITE;
		break;
	case _O_RDWR:
		access = GENERIC_READ|GENERIC_WRITE;
		break;
	default:
		printb("_open_osfhandle (emulated): unuspported access mode\n");
		return -1;
	}
	return cygwin_attach_handle_to_fd("/dev/null", -1, (HANDLE)osfhandle, -1, access);
}
#endif

#define CHECK_INIT_POLLING do {if(!is_polling_set) init_polling();} while(0)

// public fd data
const struct winfd INVALID_WINFD = {-1, NULL, NULL, RW_NONE, FALSE};
struct winfd poll_fd[MAX_FDS];
// internal fd data
struct {
	pthread_mutex_t mutex;  // thread mutex lock for fds
	BYTE marker;            // 1st byte of a read_for_poll operation gets stored here

} _poll_fd[MAX_FDS];

// globals
BOOLEAN is_polling_set = FALSE;
LONG pipe_number = 0;

// Init
void init_polling(void)
{
	int i;
	if (is_polling_set) {
		// The sleep 1 sec is to give enough time for our initialization
		// below to finish, should concurrent simultaneous inits be issued 
		// from multiple threads - we don't want the second init to return
		// before the first has had time to complete its job.
		Sleep(1000);
		return;
	}
	is_polling_set = TRUE;
	for (i=0; i<MAX_FDS; i++) {
		poll_fd[i] = INVALID_WINFD;
		_poll_fd[i].marker = 0;
		pthread_mutex_init(&_poll_fd[i].mutex, NULL);
	}
}

// Internal function to retrieve the table index (and lock the fd mutex)
int _fd_to_index_and_lock(int fd)
{
	int i;

	if (fd <= 0)
		return -1;

	for (i=0; i<MAX_FDS; i++) {
		if (poll_fd[i].fd == fd) {
			pthread_mutex_lock(&_poll_fd[i].mutex);
			return i;
		}
	}
	return -1;
}

OVERLAPPED *create_overlapped(void) 
{
	OVERLAPPED *overlapped = calloc(1, sizeof(OVERLAPPED));
	if (overlapped == NULL) {
		return NULL;
	}
	overlapped->hEvent = CreateEvent(NULL, TRUE, FALSE, NULL);
	if(overlapped->hEvent == NULL) {
		free (overlapped);
		return NULL;
	}
	return overlapped;
}

void free_overlapped(OVERLAPPED *overlapped)
{
	if (overlapped == NULL)
		return;

	if ( (overlapped->hEvent != 0) 
	  && (overlapped->hEvent != INVALID_HANDLE_VALUE) ) {
		CloseHandle(overlapped->hEvent);
	}
	free(overlapped);
}

void reset_overlapped(OVERLAPPED *overlapped)
{
	HANDLE event_handle;
	if (overlapped == NULL)
		return;

	event_handle = overlapped->hEvent;
	if (event_handle != NULL) {
		ResetEvent(event_handle);
	}
	memset(overlapped, 0, sizeof(OVERLAPPED));
	overlapped->hEvent = event_handle;
}

void exit_polling(void)
{
	int i;
	if (!is_polling_set)
		return;
	is_polling_set = FALSE;

	for (i=0; i<MAX_FDS; i++) {
		// Cancel any async I/O (handle can be invalid)
		CancelIo(poll_fd[i].handle);
		// If anything was pending on that I/O, it should be
		// terminating, and we should be able to access the fd
		// mutex lock before too long
		pthread_mutex_lock(&_poll_fd[i].mutex);
		if ( (poll_fd[i].fd > 0) && (poll_fd[i].handle != INVALID_HANDLE_VALUE) && (poll_fd[i].handle != 0)
		  && (GetFileType(poll_fd[i].handle) == FILE_TYPE_UNKNOWN) ) {
			_close(poll_fd[i].fd);
		}
		free_overlapped(poll_fd[i].overlapped);
		poll_fd[i] = INVALID_WINFD;
		pthread_mutex_unlock(&_poll_fd[i].mutex);
		pthread_mutex_destroy(&_poll_fd[i].mutex);
	}
}

/*
 * sets the async I/O read on our 1 byte marker
 *
 * requires a valid index
 */
__inline void _init_read_marker(int index) 
{
	// Cancel any read operation in progress
	CancelIo(poll_fd[index].handle);
	// Setup a new async read on our marker
	reset_overlapped(poll_fd[index].overlapped);
	if (!ReadFile(poll_fd[index].handle, &_poll_fd[index].marker, 1, NULL, poll_fd[index].overlapped)) {
		if(GetLastError() != ERROR_IO_PENDING) {
			printb("_init_read_marker: didn't get IO_PENDING!\n");
			reset_overlapped(poll_fd[index].overlapped);
		}
	} else {
		// We got some sync I/O. We'll pretend it's async and set overlapped manually
		printb("_init_read_marker: marker readout completed before exit!\n");
		// TODO: check IO_PENDING cleared?
		SetEvent(poll_fd[index].overlapped->hEvent);
		poll_fd[index].overlapped->InternalHigh = 1;
	}
}

/*
 * Create an async I/O anonymous pipe (that can be used for sync as well)
 */
int pipe_for_poll(int filedes[2])
{
	int i, j;
	HANDLE handle[2];
	OVERLAPPED *overlapped0, *overlapped1;
	char pipe_name[] = "\\\\.\\pipe\\libusb000000000000";
	LONG our_pipe_number;

	CHECK_INIT_POLLING;

	overlapped0 = calloc(1, sizeof(OVERLAPPED));
	if (overlapped0 == NULL) {
		return -1;
	}

	overlapped1 = calloc(1, sizeof(OVERLAPPED));
	if (overlapped1 == NULL) {
		free(overlapped0);
		return -1;
	}

	our_pipe_number  = InterlockedIncrement(&pipe_number) - 1; // - 1 to mirror postfix operation inside _snprintf
	our_pipe_number &= 0xFFFF; // Could warn if this was necessary (ToDo)
	_snprintf(pipe_name, sizeof(pipe_name), "\\\\.\\pipe\\libusb%08x%04x", (unsigned)GetCurrentProcessId(), our_pipe_number);

	// Read end of the pipe
	handle[0] = CreateNamedPipeA(pipe_name, PIPE_ACCESS_INBOUND|FILE_FLAG_OVERLAPPED,
		PIPE_TYPE_MESSAGE|PIPE_READMODE_MESSAGE, 1, 4096, 4096, 0, NULL);
	if (handle[0] == INVALID_HANDLE_VALUE) {
		printb("Could not create pipe (read end): errcode %d\n", (int)GetLastError());
		goto out1;
	}
	filedes[0] = _open_osfhandle((intptr_t)handle[0], _O_RDONLY);
	printb("filedes[0] = %d\n", filedes[0]);

	// Write end of the pipe
	handle[1] = CreateFileA(pipe_name, GENERIC_WRITE, 0, NULL, OPEN_EXISTING,
		FILE_ATTRIBUTE_NORMAL|FILE_FLAG_OVERLAPPED, NULL);
	if (handle[1] == INVALID_HANDLE_VALUE) {
		printb("Could not create pipe (write end): errcode %d\n", (int)GetLastError());
		goto out2;
	}
	filedes[1] = _open_osfhandle((intptr_t)handle[1], _O_WRONLY);
	printb("filedes[1] = %d\n", filedes[1]);

	// Create an OVERLAPPED for each end
	overlapped0->hEvent = CreateEvent(NULL, TRUE, FALSE, NULL);
	if(!overlapped0->hEvent) {
		goto out3;
	}
	overlapped1->hEvent = CreateEvent(NULL, TRUE, FALSE, NULL);
	if(!overlapped1->hEvent) {
		goto out4;
	}

	for (i=0, j=0; i<MAX_FDS; i++) {
		if (poll_fd[i].fd < 0) {
			pthread_mutex_lock(&_poll_fd[i].mutex);
			poll_fd[i].fd = filedes[j];
			poll_fd[i].handle = handle[j];
			poll_fd[i].overlapped = (j==0)?overlapped0:overlapped1;
			poll_fd[i].rw = RW_READ+j;
			poll_fd[i].completed_synchronously = FALSE;
			j++;
			if (j==1) {
				// Start a 1 byte nonblocking read operation
				// so that we get read event notifications
				_init_read_marker(i);
			}
			pthread_mutex_unlock(&_poll_fd[i].mutex);
			if (j>=2) {
				return 0;
			}
		}
	}

	CloseHandle(overlapped1->hEvent);
out4:
	CloseHandle(overlapped0->hEvent);
out3:
	CloseHandle(handle[1]);
out2:
	CloseHandle(handle[0]);
out1:
	free(overlapped1);
	free(overlapped0);
	return -1;
}

/*
 * Create both an fd and an OVERLAPPED from an open Windows handle, so that 
 * it can be used with our polling function
 * The handle MUST support overlapped transfers (usually requires CreateFile
 * with FILE_FLAG_OVERLAPPED)
 * Return a pollable file descriptor struct, or INVALID_WINFD on error
 *
 * Note that the fd returned by this function is a per-transfer fd, rather
 * than a per-session fd and cannot be used for anything else but our 
 * custom functions (the fd itself points to the NUL: device)
 * if you plan to do R/W on the same handle, you MUST create 2 fds: one for
 * read and one for write. Using a single R/W fd is unsupported and will
 * produce unexpected results
 */
struct winfd create_fd_for_poll(HANDLE handle, int access_mode)
{
	int i, fd;
	struct winfd wfd = INVALID_WINFD;
	OVERLAPPED* overlapped = NULL;

	CHECK_INIT_POLLING;

	if ((handle == 0) || (handle == INVALID_HANDLE_VALUE)) {
		return INVALID_WINFD;
	}

	if ((access_mode != _O_RDONLY) && (access_mode != _O_WRONLY)) {
		printb("create_fd_for_poll: only one of _O_RDONLY or _O_WRONLY are supported.\n"
			"If you want to poll for R/W simultaneously, create multiple fds from the same handle.\n");
		return INVALID_WINFD;
	}
	if (access_mode == _O_RDONLY) {
		wfd.rw = RW_READ;
	} else {
		wfd.rw = RW_WRITE;
	}

	// Ensure that we get a non system conflicting unique fd
	fd = _open_osfhandle((intptr_t)CreateFileA("NUL", 0, 0,
		NULL, OPEN_EXISTING, 0, NULL), _O_RDWR);
	if (fd < 0) {
		return INVALID_WINFD;
	}

	overlapped = create_overlapped();
	if(overlapped == NULL) {
		_close(fd);
		return INVALID_WINFD;
	}

	for (i=0; i<MAX_FDS; i++) {
		if (poll_fd[i].fd < 0) {
			pthread_mutex_lock(&_poll_fd[i].mutex);
			wfd.fd = fd;
			wfd.handle = handle;
			wfd.overlapped = overlapped;
			memcpy(&poll_fd[i], &wfd, sizeof(struct winfd));
			pthread_mutex_unlock(&_poll_fd[i].mutex);
			return wfd;
		}
	}
	free_overlapped(overlapped);
	_close(fd);
	return INVALID_WINFD;
}

void _free_index(int index)
{
	// Cancel any async IO (Don't care about the validity of our handles for this)
	CancelIo(poll_fd[index].handle);
	// close fake handle for devices
	if ( (poll_fd[index].handle != INVALID_HANDLE_VALUE) && (poll_fd[index].handle != 0)
	  && (GetFileType(poll_fd[index].handle) == FILE_TYPE_UNKNOWN) ) {
		_close(poll_fd[index].fd);
	}
	free_overlapped(poll_fd[index].overlapped);
	poll_fd[index] = INVALID_WINFD;
}

/*
 * Release a pollable file descriptor. 
 *
 * Note that the associated Windows handle is not closed by this call
 */
void free_fd_for_poll(int fd)
{
	int index;

	CHECK_INIT_POLLING;

	index = _fd_to_index_and_lock(fd);
	if (index < 0) {
		return;
	}
	_free_index(index);
	pthread_mutex_unlock(&_poll_fd[index].mutex);
}

/*
 * The functions below perform various conversions between fd, handle and OVERLAPPED
 */
struct winfd fd_to_winfd(int fd)
{
	int i;
	struct winfd wfd;

	CHECK_INIT_POLLING;

	if (fd <= 0)
		return INVALID_WINFD;

	for (i=0; i<MAX_FDS; i++) {
		if (poll_fd[i].fd == fd) {
			pthread_mutex_lock(&_poll_fd[i].mutex);
			memcpy(&wfd, &poll_fd[i], sizeof(struct winfd));
			pthread_mutex_unlock(&_poll_fd[i].mutex);
			return wfd;
		}
	}
	return INVALID_WINFD;
}

struct winfd handle_to_winfd(HANDLE handle)
{
	int i;
	struct winfd wfd;

	CHECK_INIT_POLLING;

	if ((handle == 0) || (handle == INVALID_HANDLE_VALUE))
		return INVALID_WINFD;

	for (i=0; i<MAX_FDS; i++) {
		if (poll_fd[i].handle == handle) {
			pthread_mutex_lock(&_poll_fd[i].mutex);
			memcpy(&wfd, &poll_fd[i], sizeof(struct winfd));
			pthread_mutex_unlock(&_poll_fd[i].mutex);
			return wfd;
		}
	}
	return INVALID_WINFD;
}

struct winfd overlapped_to_winfd(OVERLAPPED* overlapped)
{
	int i;
	struct winfd wfd;

	CHECK_INIT_POLLING;

	if (overlapped == NULL)
		return INVALID_WINFD;

	for (i=0; i<MAX_FDS; i++) {
		if (poll_fd[i].overlapped == overlapped) {
			pthread_mutex_lock(&_poll_fd[i].mutex);
			memcpy(&wfd, &poll_fd[i], sizeof(struct winfd));
			pthread_mutex_unlock(&_poll_fd[i].mutex);
			return wfd;
		}
	}
	return INVALID_WINFD;
}

/*
 * POSIX poll equivalent, using Windows OVERLAPPED
 * Currently, this function only accepts one of POLLIN or POLLOUT per fd
 * (but you can create multiple fds from the same handle for read and write)
 */
int poll(struct pollfd *fds, unsigned int nfds, int timeout)
{
	unsigned i, triggered = 0;
	int index, object_index;
	HANDLE *handles_to_wait_on = malloc(nfds*sizeof(HANDLE));
	int *handle_to_index = malloc(nfds*sizeof(int));
	DWORD nb_handles_to_wait_on = 0;
	DWORD ret;

	CHECK_INIT_POLLING;

	if ((handles_to_wait_on == NULL) || (handle_to_index == NULL)) {
		errno = ENOMEM;
		return -1;
	}

	for (i = 0; i < nfds; ++i) {
		fds[i].revents = 0;

		// Only one of POLLIN or POLLOUT can be selected with this version of poll (not both)
		if ((fds[i].events & ~POLLIN) && (!(fds[i].events & POLLOUT))) {
			fds[i].revents |= POLLERR;
			errno = EACCES;
			printb("poll: unsupported set of events\n");
			return -1;
		}

		index = _fd_to_index_and_lock(fds[i].fd);
		if ( (index < 0) || (poll_fd[index].handle == INVALID_HANDLE_VALUE)
		  || (poll_fd[index].handle == 0) || (poll_fd[index].overlapped == NULL)) {
			fds[i].revents |= POLLNVAL | POLLERR;
			errno = EBADF;
			if (index >= 0) {
				pthread_mutex_unlock(&_poll_fd[index].mutex);
			}
			printb("poll: invalid fd\n");
			return -1;
		}

		// IN or OUT must match our fd direction
		if ((fds[i].events & POLLIN) && (poll_fd[index].rw != RW_READ)) {
			fds[i].revents |= POLLNVAL | POLLERR;
			errno = EBADF;
			printb("poll: attempted POLLIN on fd[%d] without READ access\n", i);
			pthread_mutex_unlock(&_poll_fd[index].mutex);
			return -1;
		}

		if ((fds[i].events & POLLOUT) && (poll_fd[index].rw != RW_WRITE)) {
			fds[i].revents |= POLLNVAL | POLLERR;
			errno = EBADF;
			printb("poll: attempted POLLOUT on fd[%d] without WRITE access\n", i);
			pthread_mutex_unlock(&_poll_fd[index].mutex);
			return -1;
		}
		
		printb("poll: fd[%d]=%d (overlapped = %p) got events %04X\n", i, poll_fd[index].fd, poll_fd[index].overlapped, fds[i].events);

		// The following macro only works if overlapped I/O was reported pending
		if ( (HasOverlappedIoCompleted(poll_fd[index].overlapped))
		  || (poll_fd[index].completed_synchronously) ) {
			printb("  completed\n");
			// checks above should ensure this works:
			fds[i].revents = fds[i].events;
			triggered++;
		} else {
			handles_to_wait_on[nb_handles_to_wait_on] = poll_fd[index].overlapped->hEvent;
			handle_to_index[nb_handles_to_wait_on] = i;
			nb_handles_to_wait_on++;
		}
		pthread_mutex_unlock(&_poll_fd[index].mutex);
	}

	if (triggered != 0)
		return triggered;

	// If nothing was triggered, wait on all fds that require it
	if (nb_handles_to_wait_on != 0)	{
		printb("poll: starting %d ms wait for %d handles...\n", timeout, (int)nb_handles_to_wait_on);
		ret = WaitForMultipleObjects(nb_handles_to_wait_on, handles_to_wait_on, 
			FALSE, (timeout==-1)?INFINITE:(DWORD)timeout);

		object_index = ret-WAIT_OBJECT_0;
		if ((object_index >= 0) && ((DWORD)object_index < nb_handles_to_wait_on)) {
			printb("  completed after wait\n");
			i = handle_to_index[object_index];
			index = _fd_to_index_and_lock(fds[i].fd);
			fds[i].revents = fds[i].events;
			triggered++;
			if (index >= 0) {
				pthread_mutex_unlock(&_poll_fd[index].mutex);
			}
		} else if (ret == WAIT_TIMEOUT) {
			printb("  timed out\n");
			return 0;	// 0 = timeout
		} else {
			errno = EIO;
			return -1;	// error
		}
	}

	return triggered;
}

/*
 * close a pollable fd
 *
 * Note that this function will also close the associated handle
 */
int close_for_poll(int fd)
{
	int index;
	HANDLE handle;

	CHECK_INIT_POLLING;

	index = _fd_to_index_and_lock(fd);

	if (index < 0) {
		errno = EBADF;
	} else {
		handle = poll_fd[index].handle;
		_free_index(index);
		if (CloseHandle(handle) == 0) {
			errno = EIO;
		} else {
			errno = 0;
		}
		pthread_mutex_unlock(&_poll_fd[index].mutex);
	}
	return errno?-1:0;
}

/*
 * synchronous write for custom poll (works on Windows file handles that
 * have been opened with the FILE_FLAG_OVERLAPPED flag)
 *
 * Current restrictions: 
 * - binary mode only
 * - no append mode
 */
ssize_t write_for_poll(int fd, const void *buf, size_t count)
{
	int index;
	DWORD wr_count;

	CHECK_INIT_POLLING;

	index = _fd_to_index_and_lock(fd);

	if (count == 0) {
		return 0;
	}

	if ( (index < 0) || (poll_fd[index].overlapped == NULL) 
	  || (poll_fd[index].rw != RW_WRITE) ) {
		errno = EBADF;
		if (index >= 0) {
			pthread_mutex_unlock(&_poll_fd[index].mutex);
		}
		return -1;
	}

	// For sync mode, we shouldn't get pending async write I/O
	if (!HasOverlappedIoCompleted(poll_fd[index].overlapped)) {
		printb("write_for_poll: previous write I/O was flagged pending!\n");
		CancelIo(poll_fd[index].handle);
	}

	printb("write_for_poll: writing %d bytes to fd=%d\n", count, poll_fd[index].fd);

	reset_overlapped(poll_fd[index].overlapped);
	if (!WriteFile(poll_fd[index].handle, buf, (DWORD)count, &wr_count, poll_fd[index].overlapped)) {
		if(GetLastError() == ERROR_IO_PENDING) {
			// I/O started but is not completed => wait till completion
			switch(WaitForSingleObject(poll_fd[index].overlapped->hEvent, INFINITE)) 
			{
			case WAIT_OBJECT_0:
				if (GetOverlappedResult(poll_fd[index].handle, 
					poll_fd[index].overlapped, &wr_count, FALSE)) {
					errno = 0;
					goto out;
				} else {
					printb("write_for_poll: GetOverlappedResult failed with error %d\n", (int)GetLastError());
					errno = EIO;
					goto out;
				}
			default:
				errno = EIO;
				goto out;
			}
		} else {
			// I/O started and failed
			printb("write_for_poll: WriteFile failed with error %d\n", (int)GetLastError());
			errno = EIO;
			goto out;
		}
	}

	// I/O started and completed synchronously
	errno = 0;

out:
	if (errno) {
		reset_overlapped(poll_fd[index].overlapped);
		pthread_mutex_unlock(&_poll_fd[index].mutex);
		return -1;
	} else {
		pthread_mutex_unlock(&_poll_fd[index].mutex);
		return (ssize_t)wr_count;
	}
}

/*
 * synchronous read for custom poll (works on Windows file handles that
 * have been opened with the FILE_FLAG_OVERLAPPED flag)
 */
ssize_t read_for_poll(int fd, void *buf, size_t count)
{
	int index;
	DWORD rd_count;

	CHECK_INIT_POLLING;
	errno = 0;

	if (count == 0) {
		return 0;
	}

	index = _fd_to_index_and_lock(fd);

	if (index < 0) {
		errno = EBADF;
		return -1;
	}

	if (poll_fd[index].rw != RW_READ) {
		errno = EBADF;
		goto out;
	}


	// still waiting for completion => force completion
	if (!HasOverlappedIoCompleted(poll_fd[index].overlapped)) {
		if (WaitForSingleObject(poll_fd[index].overlapped->hEvent, INFINITE) != WAIT_OBJECT_0) {
			printb("read_for_poll: waiting for marker failed: %d\n", (int)GetLastError());
			errno = EIO;
			goto out;
		}
	}

	// Find out if we've read the first byte
	if (!GetOverlappedResult(poll_fd[index].handle,	poll_fd[index].overlapped, &rd_count, FALSE)) {
		if (GetLastError() != ERROR_MORE_DATA) {
			printb("read_for_poll: readout of marker failed: %d\n", (int)GetLastError());
			errno = EIO;
			goto out;
		} else {
			printb("read_for_poll: readout of marker reported more data\n");
		}
	}

	printb("read_for_poll: count = %d, rd_count(marker) = %d\n", count, (int)rd_count);

	// We should have our marker by now
	if (rd_count != 1) {
		printb("read_for_poll: unexpected number of bytes for marker (%d)\n", (int)rd_count);
		errno = EIO;
		goto out;
	}

	((BYTE*)buf)[0] = _poll_fd[index].marker;

	// Read supplementary bytes if needed (blocking)
	if (count > 1) {
		reset_overlapped(poll_fd[index].overlapped);
		if (!ReadFile(poll_fd[index].handle, (char*)buf+1, (DWORD)(count-1), &rd_count, poll_fd[index].overlapped)) {
			if(GetLastError() == ERROR_IO_PENDING) {
				if (!GetOverlappedResult(poll_fd[index].handle,	poll_fd[index].overlapped, &rd_count, TRUE)) {
					// TODO: handle more data!
					printb("read_for_poll: readout of supplementary data failed: %d\n", (int)GetLastError());
					errno = EIO;
					goto out;
				}
			} else {
				printb("read_for_poll: could not start blocking read of supplementary: %d\n", (int)GetLastError());
				errno = EIO;
				goto out;
			}
		}
		// If ReadFile completed synchronously, we're fine too

		printb("read_for_poll: rd_count(supplementary ) = %d\n", (int)rd_count);

		if ((rd_count+1) != count) {
			printb("read_for_poll: wanted %d-1, got %d\n", count, (int)rd_count);
			errno = EIO;
			goto out;
		}
	}

out:
	// Setup pending read I/O for the marker
	_init_read_marker(index);
	pthread_mutex_unlock(&_poll_fd[index].mutex);
	if (errno)
		return -1;
	else
		return count;
}