aboutsummaryrefslogtreecommitdiff
path: root/winsup/cygwin/aio.cc
blob: c4b3389c46aebcdd79693689cfdd3275fbe8437b (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
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
/* aio.cc: Posix asynchronous i/o functions.

This file is part of Cygwin.

This software is a copyrighted work licensed under the terms of the
Cygwin license.  Please consult the file "CYGWIN_LICENSE" for
details. */

#include "winsup.h"
#include "path.h"
#include "fhandler.h"
#include "dtable.h"
#include "cygheap.h"
#include "sigproc.h"
#include <aio.h>
#include <fcntl.h>
#include <semaphore.h>
#include <unistd.h>

#ifdef __cplusplus
extern "C" {
#endif

/* 'aioinitialized' is a thread-safe status of AIO feature initialization:
 * 0 means uninitialized, >0 means initializing, <0 means initialized
 */
static NO_COPY volatile LONG    aioinitialized = 0;

/* This implementation supports two flavors of asynchronous operation:
 * "inline" and "queued".  Inline AIOs are used when:
 *     (1) fd refers to a local non-locked disk file opened in binary mode,
 *     (2) no more than AIO_MAX inline AIOs will be in progress at same time.
 * In all other cases queued AIOs will be used.
 *
 * An inline AIO is performed by the calling app's thread as a pread|pwrite on
 * a shadow fd that permits Windows asynchronous i/o, with event notification
 * on completion.  Event arrival causes AIO context for the fd to be updated.
 *
 * A queued AIO is performed in a similar manner, but by an AIO worker thread
 * rather than the calling app's thread.  The queued flavor can also operate
 * on sockets, pipes, non-binary files, mandatory-locked files, and files
 * that don't support pread|pwrite.  Generally all these cases are handled as
 * synchronous read|write operations, but still don't delay the app because
 * they're taken care of by AIO worker threads.
 */

/* These variables support inline AIO operations */
static NO_COPY HANDLE            evt_handles[AIO_MAX];
static NO_COPY struct aiocb     *evt_aiocbs[AIO_MAX];
static NO_COPY CRITICAL_SECTION  evt_locks[AIO_MAX]; /* per-slot locks */
static NO_COPY CRITICAL_SECTION  slotcrit; /* lock for slot variables in toto */

/* These variables support queued AIO operations */
static NO_COPY sem_t             worksem;   /* tells whether AIOs are queued */
static NO_COPY CRITICAL_SECTION  workcrit;        /* lock for AIO work queue */
TAILQ_HEAD(queue, aiocb) worklist = TAILQ_HEAD_INITIALIZER(worklist);

static int
aiochkslot (struct aiocb *aio)
{
  EnterCriticalSection (&slotcrit);

  /* Sanity check.. make sure this AIO is not already busy */
  for (int slot = 0; slot < AIO_MAX; ++slot)
    if (evt_aiocbs[slot] == aio)
      {
        debug_printf ("aio %p is already busy in slot %d", aio, slot);
        LeaveCriticalSection (&slotcrit);
        return slot;
      }

  LeaveCriticalSection (&slotcrit);
  return -1;
}

static int
aiogetslot (struct aiocb *aio)
{
  EnterCriticalSection (&slotcrit);

  /* Find free slot for this inline AIO; if none available AIO will be queued */
  for (int slot = 0; slot < AIO_MAX; ++slot)
    if (evt_aiocbs[slot] == NULL)
      {
        /* If aio is NULL this is just an availability check.. no change made */
        if (aio)
          evt_aiocbs[slot] = aio;
        LeaveCriticalSection (&slotcrit);
        return slot;
      }

  LeaveCriticalSection (&slotcrit);
  return -1;
}

static int
aiorelslot (struct aiocb *aio)
{
  EnterCriticalSection (&slotcrit);

  /* Find slot associated with this inline AIO and free it */
  for (int slot = 0; slot < AIO_MAX; ++slot)
    if (evt_aiocbs[slot] == aio)
      {
        evt_aiocbs[slot] = NULL;
        LeaveCriticalSection (&slotcrit);
        return slot;
      }

  LeaveCriticalSection (&slotcrit);
  return -1;
}

static void
aionotify_on_pthread (struct sigevent *evp)
{
  pthread_attr_t *attr;
  pthread_attr_t  default_attr;
  int             rc;
  pthread_t       vaquita; /* == "little porpoise", endangered, see below */

  if (evp->sigev_notify_attributes)
    attr = evp->sigev_notify_attributes;
  else
    {
      pthread_attr_init (attr = &default_attr);
      pthread_attr_setdetachstate (attr, PTHREAD_CREATE_DETACHED);
    }

  /* A "vaquita" thread is a temporary pthread created to deliver a signal to
   * the application.  We don't wait around for the thread to return from the
   * app.  There's some symbolism here of sending a little creature off to tell
   * the app something important.  If all the vaquitas end up wiped out in the
   * wild, a distinct near-term possibility, at least this code remembers them.
   */
  rc = pthread_create (&vaquita, attr,
                       (void * (*) (void *)) evp->sigev_notify_function,
                       evp->sigev_value.sival_ptr);

  /* The following error is not expected. If seen often, develop a recovery. */
  if (rc)
    debug_printf ("aio vaquita thread creation failed, %E");

  /* Should we wait for the signal delivery thread to finish?  We can't: Who
   * knows what mischief the app coder may have in their handler?  Worst case
   * is they accidentally used non-signal-safe functions in their handler.  We
   * return hoping for the best and finish cleaning up our end of notification.
   */
  return;
}

static void
aionotify (struct aiocb *aio)
{
  siginfo_t si = {0};
  si.si_code = SI_ASYNCIO;

  /* If signal notification wanted, send AIO-complete signal */
  switch (aio->aio_sigevent.sigev_notify) {
  case SIGEV_NONE:
    break;

  case SIGEV_SIGNAL:
    si.si_signo = aio->aio_sigevent.sigev_signo;
    si.si_value = aio->aio_sigevent.sigev_value;
    if (si.si_signo)
      sig_send (myself, si);
    break;

  case SIGEV_THREAD:
    aionotify_on_pthread (&aio->aio_sigevent);
    break;
  }

  /* If this op is on LIO list and is last op, send LIO-complete signal */
  if (aio->aio_liocb)
    {
      if (1 == InterlockedExchangeAdd (&aio->aio_liocb->lio_count, -1))
        {
          /* LIO's count has decremented to zero */
          switch (aio->aio_liocb->lio_sigevent->sigev_notify) {
          case SIGEV_NONE:
            break;

          case SIGEV_SIGNAL:
            si.si_signo = aio->aio_liocb->lio_sigevent->sigev_signo;
            si.si_value = aio->aio_liocb->lio_sigevent->sigev_value;
            if (si.si_signo)
              sig_send (myself, si);
            break;

          case SIGEV_THREAD:
            aionotify_on_pthread (aio->aio_liocb->lio_sigevent);
            break;
          }

          free (aio->aio_liocb);
          aio->aio_liocb = NULL;
        }
    }
}

static DWORD __attribute__ ((noreturn))
aiowaiter (void *unused)
{ /* One instance, called on its own cygthread; runs until program exits */
  struct aiocb *aio;

  while (1)
    {
      /* Wait forever for at least one event to be set */
      DWORD res = WaitForMultipleObjects(AIO_MAX, evt_handles, FALSE, INFINITE);
      switch (res)
        {
          case WAIT_FAILED:
            api_fatal ("aiowaiter fatal error, %E");

          default:
            if (res < WAIT_OBJECT_0 || res >= WAIT_OBJECT_0 + AIO_MAX)
              api_fatal ("aiowaiter unexpected WFMO result %d", res);
            int slot = res - WAIT_OBJECT_0;

            /* Guard against "saw completion before request finished" gotcha */
            EnterCriticalSection (&evt_locks[slot]);
            LeaveCriticalSection (&evt_locks[slot]);

            aio = evt_aiocbs[slot];
            debug_printf ("WFMO returns %d, aio %p", res, aio);

            if (aio->aio_errno == EBUSY)
              {
                /* Capture Windows status and convert to Cygwin status */
                NTSTATUS status = (NTSTATUS) aio->aio_wincb.status;
                if (NT_SUCCESS (status))
                  {
                    aio->aio_rbytes = (ssize_t) aio->aio_wincb.info;
                    aio->aio_errno = 0;
                  }
                else
                  {
                    aio->aio_rbytes = -1;
                    aio->aio_errno = geterrno_from_nt_status (status);
                  }
              }
            else
              {
                /* Async operation was simulated; AIO status already updated */
              }

            /* Send completion signal if user requested it */
            aionotify (aio);

            /* Free up the slot used for this inline AIO.  We do this
             * manually rather than calling aiorelslot() because we
             * already have the slot number handy.
             */
            EnterCriticalSection (&slotcrit);
            evt_aiocbs[slot] = NULL;
            LeaveCriticalSection (&slotcrit);
            debug_printf ("retired aio %p; slot %d released", aio, slot);

            /* Notify workers that a slot has opened up */
            sem_post (&worksem);
        }
    }
}

static ssize_t
asyncread (struct aiocb *aio)
{ /* Try to initiate an asynchronous read, either from app or worker thread */
  ssize_t       res = 0;

  cygheap_fdget cfd (aio->aio_fildes);
  if (cfd < 0)
    res = -1; /* errno has been set to EBADF */
  else
    {
      int slot = aiogetslot (aio);
      debug_printf ("slot %d%s", slot, slot >= 0 ? " acquired" : "");
      if (slot >= 0)
        {
          EnterCriticalSection (&evt_locks[slot]);
          aio->aio_errno = EBUSY; /* Mark AIO as physically underway now */
          aio->aio_wincb.event = (void *) evt_handles[slot];
          res = cfd->pread ((void *) aio->aio_buf, aio->aio_nbytes,
                            aio->aio_offset, (void *) aio);
          LeaveCriticalSection (&evt_locks[slot]);
        }
      else
        {
          set_errno (ENOBUFS); /* Internal use only */
          res = -1;
        }
    }

  return res;
}

static ssize_t
asyncwrite (struct aiocb *aio)
{ /* Try to initiate an asynchronous write, either from app or worker thread */
  ssize_t       res = 0;

  cygheap_fdget cfd (aio->aio_fildes);
  if (cfd < 0)
    res = -1; /* errno has been set to EBADF */
  else
    {
      int slot = aiogetslot (aio);
      debug_printf ("slot %d%s", slot, slot >= 0 ? " acquired" : "");
      if (slot >= 0)
        {
          EnterCriticalSection (&evt_locks[slot]);
          aio->aio_errno = EBUSY; /* Mark AIO as physically underway now */
          aio->aio_wincb.event = (void *) evt_handles[slot];
          res = cfd->pwrite ((void *) aio->aio_buf, aio->aio_nbytes,
                             aio->aio_offset, (void *) aio);
          LeaveCriticalSection (&evt_locks[slot]);
        }
      else
        {
          set_errno (ENOBUFS); /* Internal use only */
          res = -1;
        }
    }

  return res;
}

/* Have to forward ref because of chicken v. egg situation */
static DWORD __attribute__ ((noreturn)) aioworker (void *);

static void
aioinit (void)
{
  /* First a cheap test to speed processing after initialization completes */
  if (aioinitialized >= 0)
    {
      /* Guard against multiple threads initializing at same time */
      if (0 == InterlockedExchangeAdd (&aioinitialized, 1))
        {
          int       i = AIO_MAX;
          char     *tnames = (char *) malloc (AIO_MAX * 8);

          if (!tnames)
            api_fatal ("couldn't create aioworker tname table");

          InitializeCriticalSection (&slotcrit);
          InitializeCriticalSection (&workcrit);
          sem_init (&worksem, 0, 0);
          TAILQ_INIT(&worklist);

          /* Create AIO_MAX number of aioworker threads for queued AIOs */
          while (i--)
            {
              __small_sprintf (&tnames[i * 8], "aio%d", AIO_MAX - i);
              if (!new cygthread (aioworker, NULL, &tnames[i * 8]))
                api_fatal ("couldn't create an aioworker thread, %E");
            }

          /* Initialize event handles and slot locks arrays for inline AIOs */
          for (i = 0; i < AIO_MAX; ++i)
            {
              /* Events are non-inheritable, auto-reset, init unset, unnamed */
              evt_handles[i] = CreateEvent (NULL, FALSE, FALSE, NULL);
              if (!evt_handles[i])
                api_fatal ("couldn't create an event, %E");

              InitializeCriticalSection (&evt_locks[i]);
            }

          /* Create aiowaiter thread; waits for inline AIO completion events */
          if (!new cygthread (aiowaiter, NULL, "aio"))
            api_fatal ("couldn't create aiowaiter thread, %E");

          /* Indicate we have completed initialization */
          InterlockedExchange (&aioinitialized, -1);
        }
      else
        /* If 'aioinitialized' is greater than zero, another thread is
         * initializing for us; wait until 'aioinitialized' goes negative
         */
        while (InterlockedExchangeAdd (&aioinitialized, 0) >= 0)
          yield ();
    }
}

static int
aioqueue (struct aiocb *aio)
{ /* Add an AIO to the worklist, to be serviced by a worker thread */
  if (aioinitialized >= 0)
    aioinit ();

  EnterCriticalSection (&workcrit);
  TAILQ_INSERT_TAIL(&worklist, aio, aio_chain);
  LeaveCriticalSection (&workcrit);

  debug_printf ("queued aio %p", aio);
  sem_post (&worksem);

  return 0;
}

static DWORD __attribute__ ((noreturn))
aioworker (void *unused)
{ /* Multiple instances; called on own cygthreads; runs 'til program exits */
  struct aiocb *aio;

  while (1)
    {
      /* Park here until there's work to do or a slot becomes available */
      sem_wait (&worksem);

look4work:
      EnterCriticalSection (&workcrit);
      if (TAILQ_EMPTY(&worklist))
        {
          /* Another aioworker picked up the work already */
          LeaveCriticalSection (&workcrit);
          continue;
        }

      /* Make sure a slot is available before starting this AIO */
      aio = TAILQ_FIRST(&worklist);
      int slot = aiogetslot (NULL);
      if (slot >= 0) // a slot is available
        TAILQ_REMOVE(&worklist, aio, aio_chain);
      LeaveCriticalSection (&workcrit);
      if (slot < 0) // no slot is available, so worklist unchanged and we park
        continue;

      debug_printf ("starting aio %p", aio);
      switch (aio->aio_lio_opcode)
        {
          case LIO_NOP:
            aio->aio_rbytes = 0;
            break;

          case LIO_READ:
            aio->aio_rbytes = asyncread (aio);
            break;

          case LIO_WRITE:
            aio->aio_rbytes = asyncwrite (aio);
            break;

          default:
            errno = EINVAL;
            aio->aio_rbytes = -1;
            break;
        }

      /* If operation still underway, let aiowaiter hear about its finish */
      if (aio->aio_rbytes == 0 && aio->aio_nbytes != 0) // not racy
        continue;

      /* If operation errored, save error number, else clear it */
      if (aio->aio_rbytes == -1)
        aio->aio_errno = get_errno ();
      else
        aio->aio_errno = 0;

      /* If a slot for this queued async AIO was available, but we lost out */
      if (aio->aio_errno == ENOBUFS)
        {
          aio->aio_errno = EINPROGRESS;
          aioqueue (aio); /* Re-queue the AIO */

          /* Another option would be to fail the AIO with error EAGAIN, but
           * experience with iozone showed apps might not expect to see a
           * deferred EAGAIN.  I.e. they should expect EAGAIN on their call to
           * aio_read() or aio_write() but probably not expect to see EAGAIN
           * on an aio_error() query after they'd previously seen EINPROGRESS
           * on the initial AIO call.
           */
          continue;
        }

      /* If seeks aren't permitted on given fd, or pread|pwrite not legal */
      if (aio->aio_errno == ESPIPE)
        {
          ssize_t res = 0;
          off_t curpos;

          cygheap_fdget cfd (aio->aio_fildes);
          if (cfd < 0)
            {
              res = -1;
              goto done; /* errno has been set to EBADF */
            }

          /* If we can get current file position, seek to aio_offset */
          curpos = cfd->lseek (0, SEEK_CUR);
          if (curpos < 0 || cfd->lseek (aio->aio_offset, SEEK_SET) < 0)
            {
              /* Can't seek */
              res = curpos;
              set_errno (0); /* Get rid of ESPIPE we've incurred */
              aio->aio_errno = 0; /* Here too */
            }

          /* Do the requested AIO operation manually, synchronously */
          switch (aio->aio_lio_opcode)
            {
              case LIO_READ:
                /* 2nd argument to cfd->read() is passed by reference... */
                cfd->read ((void *) aio->aio_buf, aio->aio_nbytes);
                /* ...so on return it contains the number of bytes read */
                res = aio->aio_nbytes;
                break;

              case LIO_WRITE:
                res = cfd->write ((void *) aio->aio_buf, aio->aio_nbytes);
                break;
            }

          /* If we had seeked successfully, restore original file position */
          if (curpos >= 0)
            if (cfd->lseek (curpos, SEEK_SET) < 0)
              res = -1;

done:
          /* Update AIO to reflect final result */
          aio->aio_rbytes = res;
          aio->aio_errno = res == -1 ? get_errno () : 0;

          /* Make like the requested async operation completed normally */
          for (int i = 0; i < AIO_MAX; i++)
            if (evt_aiocbs[i] == aio)
              {
                SetEvent (evt_handles[i]);
                goto truly_done;
              }

          /* Free up the slot we ended up not using */
          int slot = aiorelslot (aio);
          debug_printf ("slot %d released", slot);
        }

      /* Send completion signal if user requested it */
      aionotify (aio);

truly_done:
      debug_printf ("completed aio %p", aio);
      goto look4work;
    }
}

int
aio_cancel (int fildes, struct aiocb *aio)
{
  int           aiocount = 0;
  struct aiocb *ptr;
  siginfo_t     si = {0};
  si.si_code = SI_ASYNCIO;

  /* Note 'aio' is allowed to be NULL here; it's used as a wildcard */
restart:
  EnterCriticalSection (&workcrit);
  TAILQ_FOREACH(ptr, &worklist, aio_chain)
    {
      if (ptr->aio_fildes == fildes && (!aio || ptr == aio))
        {
          /* This queued AIO qualifies for cancellation */
          TAILQ_REMOVE(&worklist, ptr, aio_chain);
          LeaveCriticalSection (&workcrit);

          ptr->aio_errno = ECANCELED;
          ptr->aio_rbytes = -1;

          /* If signal notification wanted, send AIO-canceled signal */
          switch (ptr->aio_sigevent.sigev_notify) {
          case SIGEV_NONE:
            break;

          case SIGEV_SIGNAL:
            si.si_signo = ptr->aio_sigevent.sigev_signo;
            si.si_value = ptr->aio_sigevent.sigev_value;
            if (si.si_signo)
              sig_send (myself, si);
            break;

          case SIGEV_THREAD:
            aionotify_on_pthread (&ptr->aio_sigevent);
            break;
          }

          ++aiocount;
          goto restart;
        }
    }
  LeaveCriticalSection (&workcrit);

  /* Note that AIO_NOTCANCELED is not possible in this implementation.  That's
   * because AIOs are dequeued to execute; the worklist search above won't
   * find an AIO that's been dequeued from the worklist.
   */
  if (aiocount)
    return AIO_CANCELED;
  else
    return AIO_ALLDONE;
}

int
aio_error (const struct aiocb *aio)
{
  int res;

  if (!aio)
    {
      set_errno (EINVAL);
      return -1;
    }

  switch (aio->aio_errno)
    {
      case EBUSY:   /* This state for internal use only; not visible to app */
      case ENOBUFS: /* This state for internal use only; not visible to app */
        res = EINPROGRESS;
        break;

      default:
        res = aio->aio_errno;
    }

  return res;
}

int
aio_fsync (int mode, struct aiocb *aio)
{
  if (!aio)
    {
      set_errno (EINVAL);
      return -1;
    }

  switch (mode)
    {
#if defined(O_SYNC)
      case O_SYNC:
        aio->aio_rbytes = fsync (aio->aio_fildes);
        break;

#if defined(O_DSYNC) && O_DSYNC != O_SYNC
      case O_DSYNC:
        aio->aio_rbytes = fdatasync (aio->aio_fildes);
        break;
#endif
#endif

      default:
        set_errno (EINVAL);
        return -1;
    }

  if (aio->aio_rbytes == -1)
    aio->aio_errno = get_errno ();

  return aio->aio_rbytes;
}

int
aio_read (struct aiocb *aio)
{
  ssize_t       res = 0;

  if (!aio)
    {
      set_errno (EINVAL);
      return -1;
    }
  if (aioinitialized >= 0)
    aioinit ();
  if (aio->aio_errno == EINPROGRESS || -1 != aiochkslot (aio))
    {
      set_errno (EAGAIN);
      return -1;
    }

  aio->aio_lio_opcode = LIO_READ;
  aio->aio_errno = EINPROGRESS;
  aio->aio_rbytes = -1;

  /* Ensure zeroed (i.e. initialized but unused) aio_sigevent doesn't signal */
  if (aio->aio_sigevent.sigev_signo == 0)
    aio->aio_sigevent.sigev_notify = SIGEV_NONE;

  /* Try to launch inline async read; only on ESPIPE/ENOBUFS is it queued */
  pthread_testcancel ();
  res = asyncread (aio);

  /* If async read couldn't be launched, queue the AIO for a worker thread */
  if (res == -1)
    switch (get_errno ()) {
    case ESPIPE:
      {
        int slot = aiorelslot (aio);
        if (slot >= 0)
          debug_printf ("slot %d released", slot);
      }
      fallthrough;

    case ENOBUFS:
      aio->aio_errno = EINPROGRESS;
      aio->aio_rbytes = -1;

      res = aioqueue (aio);
      break;

    default:
      ; /* I think this is not possible */
    }

  return res < 0 ? (int) res : 0; /* return 0 on success */
}

ssize_t
aio_return (struct aiocb *aio)
{
  if (!aio)
    {
      set_errno (EINVAL);
      return -1;
    }

  switch (aio->aio_errno)
    {
      case EBUSY:       /* AIO is currently underway (internal state) */
      case ENOBUFS:     /* AIO is currently underway (internal state) */
      case EINPROGRESS: /* AIO has been queued successfully */
        set_errno (EINPROGRESS);
        return -1;

      case EINVAL:      /* aio_return() has already been called on this AIO */
        set_errno (aio->aio_errno);
        return -1;

      default:          /* AIO has completed, successfully or not */
        ;
    }

  /* This AIO has completed so grab any error status if present */
  if (aio->aio_rbytes == -1)
    set_errno (aio->aio_errno);

  /* Set this AIO's errno so later aio_return() calls on this AIO fail */
  aio->aio_errno = EINVAL;

  return aio->aio_rbytes;
}

static int
aiosuspend (const struct aiocb *const aiolist[],
         int nent, const struct timespec *timeout)
{
  /* Returns lowest list index of completed aios, else 'nent' if all completed.
   * If none completed on entry, wait for interval specified by 'timeout'.
   */
  int       res;
  sigset_t  sigmask;
  siginfo_t si;
  ULONGLONG nsecs = 0;
  ULONGLONG time0, time1;
  struct timespec to = {0};

  if (timeout)
    {
      to = *timeout;
      if (!valid_timespec (to))
        {
          set_errno (EINVAL);
          return -1;
        }
      nsecs = (NSPERSEC * to.tv_sec) + to.tv_nsec;
    }

retry:
  sigemptyset (&sigmask);
  int aiocount = 0;
  for (int i = 0; i < nent; ++i)
    if (aiolist[i] && aiolist[i]->aio_liocb)
      {
        if (aiolist[i]->aio_errno == EINPROGRESS ||
            aiolist[i]->aio_errno == ENOBUFS ||
            aiolist[i]->aio_errno == EBUSY)
          {
            ++aiocount;
            if (aiolist[i]->aio_sigevent.sigev_notify == SIGEV_SIGNAL ||
                aiolist[i]->aio_sigevent.sigev_notify == SIGEV_THREAD)
              sigaddset (&sigmask, aiolist[i]->aio_sigevent.sigev_signo);
          }
        else
          return i;
      }

  if (aiocount == 0)
    return nent;

  if (timeout && nsecs == 0)
    {
      set_errno (EAGAIN);
      return -1;
    }

  time0 = get_clock (CLOCK_MONOTONIC)->nsecs ();
  /* Note wait below is abortable even w/ empty sigmask and infinite timeout */
  res = sigtimedwait (&sigmask, &si, timeout ? &to : NULL);
  if (res == -1)
    return -1; /* Return with errno set by failed sigtimedwait() */
  time1 = get_clock (CLOCK_MONOTONIC)->nsecs ();

  /* Adjust timeout to account for time just waited */
  time1 -= time0;
  if (time1 > nsecs)
    nsecs = 0; // just in case we didn't get rescheduled very quickly
  else
    nsecs -= time1;
  to.tv_sec = nsecs / NSPERSEC;
  to.tv_nsec = nsecs % NSPERSEC;

  goto retry;
}

int
aio_suspend (const struct aiocb *const aiolist[],
             int nent, const struct timespec *timeout)
{
  int res;

  if (nent < 0)
    {
      set_errno (EINVAL);
      return -1;
    }

  pthread_testcancel ();
  res = aiosuspend (aiolist, nent, timeout);

  /* If there was an error, or no AIOs completed before or during timeout */
  if (res == -1)
    return res; /* If no AIOs completed, errno has been set to EAGAIN */

  /* Else if all AIOs have completed */
  else if (res == nent)
    return 0;

  /* Else at least one of the AIOs completed */
  else
    return 0;
}

int
aio_write (struct aiocb *aio)
{
  ssize_t       res = 0;

  if (!aio)
    {
      set_errno (EINVAL);
      return -1;
    }
  if (aioinitialized >= 0)
    aioinit ();
  if (aio->aio_errno == EINPROGRESS || -1 != aiochkslot (aio))
    {
      set_errno (EAGAIN);
      return -1;
    }

  aio->aio_lio_opcode = LIO_WRITE;
  aio->aio_errno = EINPROGRESS;
  aio->aio_rbytes = -1;

  /* Ensure zeroed (i.e. initialized but unused) aio_sigevent doesn't signal */
  if (aio->aio_sigevent.sigev_signo == 0)
    aio->aio_sigevent.sigev_notify = SIGEV_NONE;

  /* Try to launch inline async write; only on ESPIPE/ENOBUFS is it queued */
  pthread_testcancel ();
  res = asyncwrite (aio);

  /* If async write couldn't be launched, queue the AIO for a worker thread */
  if (res == -1)
    switch (get_errno ()) {
    case ESPIPE:
      {
        int slot = aiorelslot (aio);
        if (slot >= 0)
          debug_printf ("slot %d released", slot);
      }
      fallthrough;

    case ENOBUFS:
      aio->aio_errno = EINPROGRESS;
      aio->aio_rbytes = -1;

      res = aioqueue (aio);
      break;

    default:
      ; /* I think this is not possible */
    }

  return res < 0 ? (int) res : 0; /* return 0 on success */
}

int
lio_listio (int mode, struct aiocb *__restrict const aiolist[__restrict_arr],
            int nent, struct sigevent *__restrict sig)
{
  struct aiocb *aio;
  struct __liocb *lio;

  pthread_testcancel ();

  if ((mode != LIO_WAIT && mode != LIO_NOWAIT) ||
      (nent < 0 || nent > AIO_LISTIO_MAX))
    {
      set_errno (EINVAL);
      return -1;
    }

  if (sig && nent && mode == LIO_NOWAIT)
    {
      lio = (struct __liocb *) malloc (sizeof (struct __liocb));
      if (!lio)
        {
          set_errno (ENOMEM);
          return -1;
        }

      lio->lio_count = nent;
      lio->lio_sigevent = sig;
    }
  else
    lio = NULL;

  int aiocount = 0;
  for (int i = 0; i < nent; ++i)
    {
      aio = (struct aiocb *) aiolist[i];
      if (!aio)
        {
          if (lio)
            InterlockedDecrement (&lio->lio_count);
          continue;
        }

      aio->aio_liocb = lio;
      switch (aio->aio_lio_opcode)
        {
          case LIO_NOP:
            if (lio)
              InterlockedDecrement (&lio->lio_count);
            continue;

          case LIO_READ:
            aio_read (aio);
            ++aiocount;
            continue;

          case LIO_WRITE:
            aio_write (aio);
            ++aiocount;
            continue;

          default:
            break;
        }

      if (lio)
        InterlockedDecrement (&lio->lio_count);
      aio->aio_errno = EINVAL;
      aio->aio_rbytes = -1;
    }

  /* mode is LIO_NOWAIT so return some kind of answer immediately */
  if (mode == LIO_NOWAIT)
    {
      /* At least one AIO has been launched or queued */
      if (aiocount)
        return 0;

      /* No AIOs have been launched or queued */
      set_errno (EAGAIN);
      return -1;
    }

  /* Else mode is LIO_WAIT so wait for all AIOs to complete or error */
  while (nent)
    {
      int i = aiosuspend ((const struct aiocb *const *) aiolist, nent, NULL);
      if (i >= nent)
        break;
      else
        aiolist[i]->aio_liocb = NULL; /* Avoids repeating notify on this AIO */
    }

  return 0;
}

#ifdef __cplusplus
}
#endif