aboutsummaryrefslogtreecommitdiffhomepage
path: root/src/core/lib/iomgr/ev_poll_posix.c
blob: fafb3b4b6f3b99f5b2df5b74f49fdb38f682c811 (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
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
/*
 *
 * Copyright 2015-2016, Google Inc.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are
 * met:
 *
 *     * Redistributions of source code must retain the above copyright
 * notice, this list of conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above
 * copyright notice, this list of conditions and the following disclaimer
 * in the documentation and/or other materials provided with the
 * distribution.
 *     * Neither the name of Google Inc. nor the names of its
 * contributors may be used to endorse or promote products derived from
 * this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 */

#include <grpc/support/port_platform.h>

#ifdef GPR_POSIX_SOCKET

#include "src/core/lib/iomgr/ev_poll_posix.h"

#include <assert.h>
#include <errno.h>
#include <poll.h>
#include <string.h>
#include <sys/socket.h>
#include <unistd.h>

#include <grpc/support/alloc.h>
#include <grpc/support/log.h>
#include <grpc/support/string_util.h>
#include <grpc/support/tls.h>
#include <grpc/support/useful.h>

#include "src/core/lib/iomgr/iomgr_internal.h"
#include "src/core/lib/iomgr/wakeup_fd_posix.h"
#include "src/core/lib/profiling/timers.h"
#include "src/core/lib/support/block_annotate.h"

/*******************************************************************************
 * FD declarations
 */

typedef struct grpc_fd_watcher {
  struct grpc_fd_watcher *next;
  struct grpc_fd_watcher *prev;
  grpc_pollset *pollset;
  grpc_pollset_worker *worker;
  grpc_fd *fd;
} grpc_fd_watcher;

struct grpc_fd {
  int fd;
  /* refst format:
     bit0:   1=active/0=orphaned
     bit1-n: refcount
     meaning that mostly we ref by two to avoid altering the orphaned bit,
     and just unref by 1 when we're ready to flag the object as orphaned */
  gpr_atm refst;

  gpr_mu mu;
  int shutdown;
  int closed;
  int released;

  /* The watcher list.

     The following watcher related fields are protected by watcher_mu.

     An fd_watcher is an ephemeral object created when an fd wants to
     begin polling, and destroyed after the poll.

     It denotes the fd's interest in whether to read poll or write poll
     or both or neither on this fd.

     If a watcher is asked to poll for reads or writes, the read_watcher
     or write_watcher fields are set respectively. A watcher may be asked
     to poll for both, in which case both fields will be set.

     read_watcher and write_watcher may be NULL if no watcher has been
     asked to poll for reads or writes.

     If an fd_watcher is not asked to poll for reads or writes, it's added
     to a linked list of inactive watchers, rooted at inactive_watcher_root.
     If at a later time there becomes need of a poller to poll, one of
     the inactive pollers may be kicked out of their poll loops to take
     that responsibility. */
  grpc_fd_watcher inactive_watcher_root;
  grpc_fd_watcher *read_watcher;
  grpc_fd_watcher *write_watcher;

  grpc_closure *read_closure;
  grpc_closure *write_closure;

  grpc_closure *on_done_closure;

  grpc_iomgr_object iomgr_object;

  /* The pollset that last noticed and notified that the fd is readable */
  grpc_pollset *read_notifier_pollset;
};

/* Begin polling on an fd.
   Registers that the given pollset is interested in this fd - so that if read
   or writability interest changes, the pollset can be kicked to pick up that
   new interest.
   Return value is:
     (fd_needs_read? read_mask : 0) | (fd_needs_write? write_mask : 0)
   i.e. a combination of read_mask and write_mask determined by the fd's current
   interest in said events.
   Polling strategies that do not need to alter their behavior depending on the
   fd's current interest (such as epoll) do not need to call this function.
   MUST NOT be called with a pollset lock taken */
static uint32_t fd_begin_poll(grpc_fd *fd, grpc_pollset *pollset,
                              grpc_pollset_worker *worker, uint32_t read_mask,
                              uint32_t write_mask, grpc_fd_watcher *rec);
/* Complete polling previously started with fd_begin_poll
   MUST NOT be called with a pollset lock taken
   if got_read or got_write are 1, also does the become_{readable,writable} as
   appropriate. */
static void fd_end_poll(grpc_exec_ctx *exec_ctx, grpc_fd_watcher *rec,
                        int got_read, int got_write,
                        grpc_pollset *read_notifier_pollset);

/* Return 1 if this fd is orphaned, 0 otherwise */
static bool fd_is_orphaned(grpc_fd *fd);

/* Reference counting for fds */
/*#define GRPC_FD_REF_COUNT_DEBUG*/
#ifdef GRPC_FD_REF_COUNT_DEBUG
static void fd_ref(grpc_fd *fd, const char *reason, const char *file, int line);
static void fd_unref(grpc_fd *fd, const char *reason, const char *file,
                     int line);
#define GRPC_FD_REF(fd, reason) fd_ref(fd, reason, __FILE__, __LINE__)
#define GRPC_FD_UNREF(fd, reason) fd_unref(fd, reason, __FILE__, __LINE__)
#else
static void fd_ref(grpc_fd *fd);
static void fd_unref(grpc_fd *fd);
#define GRPC_FD_REF(fd, reason) fd_ref(fd)
#define GRPC_FD_UNREF(fd, reason) fd_unref(fd)
#endif

#define CLOSURE_NOT_READY ((grpc_closure *)0)
#define CLOSURE_READY ((grpc_closure *)1)

/*******************************************************************************
 * pollset declarations
 */

typedef struct grpc_cached_wakeup_fd {
  grpc_wakeup_fd fd;
  struct grpc_cached_wakeup_fd *next;
} grpc_cached_wakeup_fd;

struct grpc_pollset_worker {
  grpc_cached_wakeup_fd *wakeup_fd;
  int reevaluate_polling_on_wakeup;
  int kicked_specifically;
  struct grpc_pollset_worker *next;
  struct grpc_pollset_worker *prev;
};

struct grpc_pollset {
  gpr_mu mu;
  grpc_pollset_worker root_worker;
  int in_flight_cbs;
  int shutting_down;
  int called_shutdown;
  int kicked_without_pollers;
  grpc_closure *shutdown_done;
  grpc_closure_list idle_jobs;
  /* all polled fds */
  size_t fd_count;
  size_t fd_capacity;
  grpc_fd **fds;
  /* fds that have been removed from the pollset explicitly */
  size_t del_count;
  size_t del_capacity;
  grpc_fd **dels;
  /* Local cache of eventfds for workers */
  grpc_cached_wakeup_fd *local_wakeup_cache;
};

/* Add an fd to a pollset */
static void pollset_add_fd(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
                           struct grpc_fd *fd);

static void pollset_set_add_fd(grpc_exec_ctx *exec_ctx,
                               grpc_pollset_set *pollset_set, grpc_fd *fd);

/* Convert a timespec to milliseconds:
   - very small or negative poll times are clamped to zero to do a
     non-blocking poll (which becomes spin polling)
   - other small values are rounded up to one millisecond
   - longer than a millisecond polls are rounded up to the next nearest
     millisecond to avoid spinning
   - infinite timeouts are converted to -1 */
static int poll_deadline_to_millis_timeout(gpr_timespec deadline,
                                           gpr_timespec now);

/* Allow kick to wakeup the currently polling worker */
#define GRPC_POLLSET_CAN_KICK_SELF 1
/* Force the wakee to repoll when awoken */
#define GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP 2
/* As per pollset_kick, with an extended set of flags (defined above)
   -- mostly for fd_posix's use. */
static void pollset_kick_ext(grpc_pollset *p,
                             grpc_pollset_worker *specific_worker,
                             uint32_t flags);

/* Return 1 if the pollset has active threads in pollset_work (pollset must
 * be locked) */
static int pollset_has_workers(grpc_pollset *pollset);

/*******************************************************************************
 * pollset_set definitions
 */

struct grpc_pollset_set {
  gpr_mu mu;

  size_t pollset_count;
  size_t pollset_capacity;
  grpc_pollset **pollsets;

  size_t pollset_set_count;
  size_t pollset_set_capacity;
  struct grpc_pollset_set **pollset_sets;

  size_t fd_count;
  size_t fd_capacity;
  grpc_fd **fds;
};

/*******************************************************************************
 * fd_posix.c
 */

#ifdef GRPC_FD_REF_COUNT_DEBUG
#define REF_BY(fd, n, reason) ref_by(fd, n, reason, __FILE__, __LINE__)
#define UNREF_BY(fd, n, reason) unref_by(fd, n, reason, __FILE__, __LINE__)
static void ref_by(grpc_fd *fd, int n, const char *reason, const char *file,
                   int line) {
  gpr_log(GPR_DEBUG, "FD %d %p   ref %d %d -> %d [%s; %s:%d]", fd->fd, fd, n,
          gpr_atm_no_barrier_load(&fd->refst),
          gpr_atm_no_barrier_load(&fd->refst) + n, reason, file, line);
#else
#define REF_BY(fd, n, reason) ref_by(fd, n)
#define UNREF_BY(fd, n, reason) unref_by(fd, n)
static void ref_by(grpc_fd *fd, int n) {
#endif
  GPR_ASSERT(gpr_atm_no_barrier_fetch_add(&fd->refst, n) > 0);
}

#ifdef GRPC_FD_REF_COUNT_DEBUG
static void unref_by(grpc_fd *fd, int n, const char *reason, const char *file,
                     int line) {
  gpr_atm old;
  gpr_log(GPR_DEBUG, "FD %d %p unref %d %d -> %d [%s; %s:%d]", fd->fd, fd, n,
          gpr_atm_no_barrier_load(&fd->refst),
          gpr_atm_no_barrier_load(&fd->refst) - n, reason, file, line);
#else
static void unref_by(grpc_fd *fd, int n) {
  gpr_atm old;
#endif
  old = gpr_atm_full_fetch_add(&fd->refst, -n);
  if (old == n) {
    gpr_mu_destroy(&fd->mu);
    grpc_iomgr_unregister_object(&fd->iomgr_object);
    gpr_free(fd);
  } else {
    GPR_ASSERT(old > n);
  }
}

static grpc_fd *fd_create(int fd, const char *name) {
  grpc_fd *r = gpr_malloc(sizeof(*r));
  gpr_mu_init(&r->mu);
  gpr_atm_rel_store(&r->refst, 1);
  r->shutdown = 0;
  r->read_closure = CLOSURE_NOT_READY;
  r->write_closure = CLOSURE_NOT_READY;
  r->fd = fd;
  r->inactive_watcher_root.next = r->inactive_watcher_root.prev =
      &r->inactive_watcher_root;
  r->read_watcher = r->write_watcher = NULL;
  r->on_done_closure = NULL;
  r->closed = 0;
  r->released = 0;
  r->read_notifier_pollset = NULL;

  char *name2;
  gpr_asprintf(&name2, "%s fd=%d", name, fd);
  grpc_iomgr_register_object(&r->iomgr_object, name2);
  gpr_free(name2);
#ifdef GRPC_FD_REF_COUNT_DEBUG
  gpr_log(GPR_DEBUG, "FD %d %p create %s", fd, r, name);
#endif
  return r;
}

static bool fd_is_orphaned(grpc_fd *fd) {
  return (gpr_atm_acq_load(&fd->refst) & 1) == 0;
}

/* Return the read-notifier pollset */
static grpc_pollset *fd_get_read_notifier_pollset(grpc_exec_ctx *exec_ctx,
                                                  grpc_fd *fd) {
  grpc_pollset *notifier = NULL;

  gpr_mu_lock(&fd->mu);
  notifier = fd->read_notifier_pollset;
  gpr_mu_unlock(&fd->mu);

  return notifier;
}

static void pollset_kick_locked(grpc_fd_watcher *watcher) {
  gpr_mu_lock(&watcher->pollset->mu);
  GPR_ASSERT(watcher->worker);
  pollset_kick_ext(watcher->pollset, watcher->worker,
                   GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP);
  gpr_mu_unlock(&watcher->pollset->mu);
}

static void maybe_wake_one_watcher_locked(grpc_fd *fd) {
  if (fd->inactive_watcher_root.next != &fd->inactive_watcher_root) {
    pollset_kick_locked(fd->inactive_watcher_root.next);
  } else if (fd->read_watcher) {
    pollset_kick_locked(fd->read_watcher);
  } else if (fd->write_watcher) {
    pollset_kick_locked(fd->write_watcher);
  }
}

static void wake_all_watchers_locked(grpc_fd *fd) {
  grpc_fd_watcher *watcher;
  for (watcher = fd->inactive_watcher_root.next;
       watcher != &fd->inactive_watcher_root; watcher = watcher->next) {
    pollset_kick_locked(watcher);
  }
  if (fd->read_watcher) {
    pollset_kick_locked(fd->read_watcher);
  }
  if (fd->write_watcher && fd->write_watcher != fd->read_watcher) {
    pollset_kick_locked(fd->write_watcher);
  }
}

static int has_watchers(grpc_fd *fd) {
  return fd->read_watcher != NULL || fd->write_watcher != NULL ||
         fd->inactive_watcher_root.next != &fd->inactive_watcher_root;
}

static void close_fd_locked(grpc_exec_ctx *exec_ctx, grpc_fd *fd) {
  fd->closed = 1;
  if (!fd->released) {
    close(fd->fd);
  }
  grpc_exec_ctx_enqueue(exec_ctx, fd->on_done_closure, true, NULL);
}

static int fd_wrapped_fd(grpc_fd *fd) {
  if (fd->released || fd->closed) {
    return -1;
  } else {
    return fd->fd;
  }
}

static void fd_orphan(grpc_exec_ctx *exec_ctx, grpc_fd *fd,
                      grpc_closure *on_done, int *release_fd,
                      const char *reason) {
  fd->on_done_closure = on_done;
  fd->released = release_fd != NULL;
  if (!fd->released) {
    shutdown(fd->fd, SHUT_RDWR);
  } else {
    *release_fd = fd->fd;
  }
  gpr_mu_lock(&fd->mu);
  REF_BY(fd, 1, reason); /* remove active status, but keep referenced */
  if (!has_watchers(fd)) {
    close_fd_locked(exec_ctx, fd);
  } else {
    wake_all_watchers_locked(fd);
  }
  gpr_mu_unlock(&fd->mu);
  UNREF_BY(fd, 2, reason); /* drop the reference */
}

/* increment refcount by two to avoid changing the orphan bit */
#ifdef GRPC_FD_REF_COUNT_DEBUG
static void fd_ref(grpc_fd *fd, const char *reason, const char *file,
                   int line) {
  ref_by(fd, 2, reason, file, line);
}

static void fd_unref(grpc_fd *fd, const char *reason, const char *file,
                     int line) {
  unref_by(fd, 2, reason, file, line);
}
#else
static void fd_ref(grpc_fd *fd) { ref_by(fd, 2); }

static void fd_unref(grpc_fd *fd) { unref_by(fd, 2); }
#endif

static void notify_on_locked(grpc_exec_ctx *exec_ctx, grpc_fd *fd,
                             grpc_closure **st, grpc_closure *closure) {
  if (*st == CLOSURE_NOT_READY) {
    /* not ready ==> switch to a waiting state by setting the closure */
    *st = closure;
  } else if (*st == CLOSURE_READY) {
    /* already ready ==> queue the closure to run immediately */
    *st = CLOSURE_NOT_READY;
    grpc_exec_ctx_enqueue(exec_ctx, closure, !fd->shutdown, NULL);
    maybe_wake_one_watcher_locked(fd);
  } else {
    /* upcallptr was set to a different closure.  This is an error! */
    gpr_log(GPR_ERROR,
            "User called a notify_on function with a previous callback still "
            "pending");
    abort();
  }
}

/* returns 1 if state becomes not ready */
static int set_ready_locked(grpc_exec_ctx *exec_ctx, grpc_fd *fd,
                            grpc_closure **st) {
  if (*st == CLOSURE_READY) {
    /* duplicate ready ==> ignore */
    return 0;
  } else if (*st == CLOSURE_NOT_READY) {
    /* not ready, and not waiting ==> flag ready */
    *st = CLOSURE_READY;
    return 0;
  } else {
    /* waiting ==> queue closure */
    grpc_exec_ctx_enqueue(exec_ctx, *st, !fd->shutdown, NULL);
    *st = CLOSURE_NOT_READY;
    return 1;
  }
}

static void set_read_notifier_pollset_locked(
    grpc_exec_ctx *exec_ctx, grpc_fd *fd, grpc_pollset *read_notifier_pollset) {
  fd->read_notifier_pollset = read_notifier_pollset;
}

static void fd_shutdown(grpc_exec_ctx *exec_ctx, grpc_fd *fd) {
  gpr_mu_lock(&fd->mu);
  GPR_ASSERT(!fd->shutdown);
  fd->shutdown = 1;
  set_ready_locked(exec_ctx, fd, &fd->read_closure);
  set_ready_locked(exec_ctx, fd, &fd->write_closure);
  gpr_mu_unlock(&fd->mu);
}

static void fd_notify_on_read(grpc_exec_ctx *exec_ctx, grpc_fd *fd,
                              grpc_closure *closure) {
  gpr_mu_lock(&fd->mu);
  notify_on_locked(exec_ctx, fd, &fd->read_closure, closure);
  gpr_mu_unlock(&fd->mu);
}

static void fd_notify_on_write(grpc_exec_ctx *exec_ctx, grpc_fd *fd,
                               grpc_closure *closure) {
  gpr_mu_lock(&fd->mu);
  notify_on_locked(exec_ctx, fd, &fd->write_closure, closure);
  gpr_mu_unlock(&fd->mu);
}

static uint32_t fd_begin_poll(grpc_fd *fd, grpc_pollset *pollset,
                              grpc_pollset_worker *worker, uint32_t read_mask,
                              uint32_t write_mask, grpc_fd_watcher *watcher) {
  uint32_t mask = 0;
  grpc_closure *cur;
  int requested;
  /* keep track of pollers that have requested our events, in case they change
   */
  GRPC_FD_REF(fd, "poll");

  gpr_mu_lock(&fd->mu);

  /* if we are shutdown, then don't add to the watcher set */
  if (fd->shutdown) {
    watcher->fd = NULL;
    watcher->pollset = NULL;
    watcher->worker = NULL;
    gpr_mu_unlock(&fd->mu);
    GRPC_FD_UNREF(fd, "poll");
    return 0;
  }

  /* if there is nobody polling for read, but we need to, then start doing so */
  cur = fd->read_closure;
  requested = cur != CLOSURE_READY;
  if (read_mask && fd->read_watcher == NULL && requested) {
    fd->read_watcher = watcher;
    mask |= read_mask;
  }
  /* if there is nobody polling for write, but we need to, then start doing so
   */
  cur = fd->write_closure;
  requested = cur != CLOSURE_READY;
  if (write_mask && fd->write_watcher == NULL && requested) {
    fd->write_watcher = watcher;
    mask |= write_mask;
  }
  /* if not polling, remember this watcher in case we need someone to later */
  if (mask == 0 && worker != NULL) {
    watcher->next = &fd->inactive_watcher_root;
    watcher->prev = watcher->next->prev;
    watcher->next->prev = watcher->prev->next = watcher;
  }
  watcher->pollset = pollset;
  watcher->worker = worker;
  watcher->fd = fd;
  gpr_mu_unlock(&fd->mu);

  return mask;
}

static void fd_end_poll(grpc_exec_ctx *exec_ctx, grpc_fd_watcher *watcher,
                        int got_read, int got_write,
                        grpc_pollset *read_notifier_pollset) {
  int was_polling = 0;
  int kick = 0;
  grpc_fd *fd = watcher->fd;

  if (fd == NULL) {
    return;
  }

  gpr_mu_lock(&fd->mu);

  if (watcher == fd->read_watcher) {
    /* remove read watcher, kick if we still need a read */
    was_polling = 1;
    if (!got_read) {
      kick = 1;
    }
    fd->read_watcher = NULL;
  }
  if (watcher == fd->write_watcher) {
    /* remove write watcher, kick if we still need a write */
    was_polling = 1;
    if (!got_write) {
      kick = 1;
    }
    fd->write_watcher = NULL;
  }
  if (!was_polling && watcher->worker != NULL) {
    /* remove from inactive list */
    watcher->next->prev = watcher->prev;
    watcher->prev->next = watcher->next;
  }
  if (got_read) {
    if (set_ready_locked(exec_ctx, fd, &fd->read_closure)) {
      kick = 1;
    }
    if (read_notifier_pollset != NULL) {
      set_read_notifier_pollset_locked(exec_ctx, fd, read_notifier_pollset);
    }
  }
  if (got_write) {
    if (set_ready_locked(exec_ctx, fd, &fd->write_closure)) {
      kick = 1;
    }
  }
  if (kick) {
    maybe_wake_one_watcher_locked(fd);
  }
  if (fd_is_orphaned(fd) && !has_watchers(fd) && !fd->closed) {
    close_fd_locked(exec_ctx, fd);
  }
  gpr_mu_unlock(&fd->mu);

  GRPC_FD_UNREF(fd, "poll");
}

/*******************************************************************************
 * pollset_posix.c
 */

GPR_TLS_DECL(g_current_thread_poller);
GPR_TLS_DECL(g_current_thread_worker);

static void remove_worker(grpc_pollset *p, grpc_pollset_worker *worker) {
  worker->prev->next = worker->next;
  worker->next->prev = worker->prev;
}

static int pollset_has_workers(grpc_pollset *p) {
  return p->root_worker.next != &p->root_worker;
}

static grpc_pollset_worker *pop_front_worker(grpc_pollset *p) {
  if (pollset_has_workers(p)) {
    grpc_pollset_worker *w = p->root_worker.next;
    remove_worker(p, w);
    return w;
  } else {
    return NULL;
  }
}

static void push_back_worker(grpc_pollset *p, grpc_pollset_worker *worker) {
  worker->next = &p->root_worker;
  worker->prev = worker->next->prev;
  worker->prev->next = worker->next->prev = worker;
}

static void push_front_worker(grpc_pollset *p, grpc_pollset_worker *worker) {
  worker->prev = &p->root_worker;
  worker->next = worker->prev->next;
  worker->prev->next = worker->next->prev = worker;
}

static void pollset_kick_ext(grpc_pollset *p,
                             grpc_pollset_worker *specific_worker,
                             uint32_t flags) {
  GPR_TIMER_BEGIN("pollset_kick_ext", 0);

  /* pollset->mu already held */
  if (specific_worker != NULL) {
    if (specific_worker == GRPC_POLLSET_KICK_BROADCAST) {
      GPR_TIMER_BEGIN("pollset_kick_ext.broadcast", 0);
      GPR_ASSERT((flags & GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP) == 0);
      for (specific_worker = p->root_worker.next;
           specific_worker != &p->root_worker;
           specific_worker = specific_worker->next) {
        grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd);
      }
      p->kicked_without_pollers = 1;
      GPR_TIMER_END("pollset_kick_ext.broadcast", 0);
    } else if (gpr_tls_get(&g_current_thread_worker) !=
               (intptr_t)specific_worker) {
      GPR_TIMER_MARK("different_thread_worker", 0);
      if ((flags & GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP) != 0) {
        specific_worker->reevaluate_polling_on_wakeup = 1;
      }
      specific_worker->kicked_specifically = 1;
      grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd);
    } else if ((flags & GRPC_POLLSET_CAN_KICK_SELF) != 0) {
      GPR_TIMER_MARK("kick_yoself", 0);
      if ((flags & GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP) != 0) {
        specific_worker->reevaluate_polling_on_wakeup = 1;
      }
      specific_worker->kicked_specifically = 1;
      grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd);
    }
  } else if (gpr_tls_get(&g_current_thread_poller) != (intptr_t)p) {
    GPR_ASSERT((flags & GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP) == 0);
    GPR_TIMER_MARK("kick_anonymous", 0);
    specific_worker = pop_front_worker(p);
    if (specific_worker != NULL) {
      if (gpr_tls_get(&g_current_thread_worker) == (intptr_t)specific_worker) {
        GPR_TIMER_MARK("kick_anonymous_not_self", 0);
        push_back_worker(p, specific_worker);
        specific_worker = pop_front_worker(p);
        if ((flags & GRPC_POLLSET_CAN_KICK_SELF) == 0 &&
            gpr_tls_get(&g_current_thread_worker) ==
                (intptr_t)specific_worker) {
          push_back_worker(p, specific_worker);
          specific_worker = NULL;
        }
      }
      if (specific_worker != NULL) {
        GPR_TIMER_MARK("finally_kick", 0);
        push_back_worker(p, specific_worker);
        grpc_wakeup_fd_wakeup(&specific_worker->wakeup_fd->fd);
      }
    } else {
      GPR_TIMER_MARK("kicked_no_pollers", 0);
      p->kicked_without_pollers = 1;
    }
  }

  GPR_TIMER_END("pollset_kick_ext", 0);
}

static void pollset_kick(grpc_pollset *p,
                         grpc_pollset_worker *specific_worker) {
  pollset_kick_ext(p, specific_worker, 0);
}

/* global state management */

static void pollset_global_init(void) {
  gpr_tls_init(&g_current_thread_poller);
  gpr_tls_init(&g_current_thread_worker);
  grpc_wakeup_fd_init(&grpc_global_wakeup_fd);
}

static void pollset_global_shutdown(void) {
  grpc_wakeup_fd_destroy(&grpc_global_wakeup_fd);
  gpr_tls_destroy(&g_current_thread_poller);
  gpr_tls_destroy(&g_current_thread_worker);
}

static void kick_poller(void) { grpc_wakeup_fd_wakeup(&grpc_global_wakeup_fd); }

/* main interface */

static void pollset_init(grpc_pollset *pollset, gpr_mu **mu) {
  gpr_mu_init(&pollset->mu);
  *mu = &pollset->mu;
  pollset->root_worker.next = pollset->root_worker.prev = &pollset->root_worker;
  pollset->in_flight_cbs = 0;
  pollset->shutting_down = 0;
  pollset->called_shutdown = 0;
  pollset->kicked_without_pollers = 0;
  pollset->idle_jobs.head = pollset->idle_jobs.tail = NULL;
  pollset->local_wakeup_cache = NULL;
  pollset->kicked_without_pollers = 0;
  pollset->fd_count = 0;
  pollset->fd_capacity = 0;
  pollset->del_count = 0;
  pollset->del_capacity = 0;
  pollset->fds = NULL;
  pollset->dels = NULL;
}

static void pollset_destroy(grpc_pollset *pollset) {
  GPR_ASSERT(pollset->in_flight_cbs == 0);
  GPR_ASSERT(!pollset_has_workers(pollset));
  GPR_ASSERT(pollset->idle_jobs.head == pollset->idle_jobs.tail);
  while (pollset->local_wakeup_cache) {
    grpc_cached_wakeup_fd *next = pollset->local_wakeup_cache->next;
    grpc_wakeup_fd_destroy(&pollset->local_wakeup_cache->fd);
    gpr_free(pollset->local_wakeup_cache);
    pollset->local_wakeup_cache = next;
  }
  gpr_free(pollset->fds);
  gpr_free(pollset->dels);
  gpr_mu_destroy(&pollset->mu);
}

static void pollset_reset(grpc_pollset *pollset) {
  GPR_ASSERT(pollset->shutting_down);
  GPR_ASSERT(pollset->in_flight_cbs == 0);
  GPR_ASSERT(!pollset_has_workers(pollset));
  GPR_ASSERT(pollset->idle_jobs.head == pollset->idle_jobs.tail);
  GPR_ASSERT(pollset->fd_count == 0);
  GPR_ASSERT(pollset->del_count == 0);
  pollset->shutting_down = 0;
  pollset->called_shutdown = 0;
  pollset->kicked_without_pollers = 0;
}

static void pollset_add_fd(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
                           grpc_fd *fd) {
  gpr_mu_lock(&pollset->mu);
  size_t i;
  /* TODO(ctiller): this is O(num_fds^2); maybe switch to a hash set here */
  for (i = 0; i < pollset->fd_count; i++) {
    if (pollset->fds[i] == fd) goto exit;
  }
  if (pollset->fd_count == pollset->fd_capacity) {
    pollset->fd_capacity =
        GPR_MAX(pollset->fd_capacity + 8, pollset->fd_count * 3 / 2);
    pollset->fds =
        gpr_realloc(pollset->fds, sizeof(grpc_fd *) * pollset->fd_capacity);
  }
  pollset->fds[pollset->fd_count++] = fd;
  GRPC_FD_REF(fd, "multipoller");
  pollset_kick(pollset, NULL);
exit:
  gpr_mu_unlock(&pollset->mu);
}

static void finish_shutdown(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset) {
  GPR_ASSERT(grpc_closure_list_empty(pollset->idle_jobs));
  size_t i;
  for (i = 0; i < pollset->fd_count; i++) {
    GRPC_FD_UNREF(pollset->fds[i], "multipoller");
  }
  for (i = 0; i < pollset->del_count; i++) {
    GRPC_FD_UNREF(pollset->dels[i], "multipoller_del");
  }
  pollset->fd_count = 0;
  pollset->del_count = 0;
  grpc_exec_ctx_enqueue(exec_ctx, pollset->shutdown_done, true, NULL);
}

static void pollset_work(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
                         grpc_pollset_worker **worker_hdl, gpr_timespec now,
                         gpr_timespec deadline) {
  grpc_pollset_worker worker;
  *worker_hdl = &worker;

  /* pollset->mu already held */
  int added_worker = 0;
  int locked = 1;
  int queued_work = 0;
  int keep_polling = 0;
  GPR_TIMER_BEGIN("pollset_work", 0);
  /* this must happen before we (potentially) drop pollset->mu */
  worker.next = worker.prev = NULL;
  worker.reevaluate_polling_on_wakeup = 0;
  if (pollset->local_wakeup_cache != NULL) {
    worker.wakeup_fd = pollset->local_wakeup_cache;
    pollset->local_wakeup_cache = worker.wakeup_fd->next;
  } else {
    worker.wakeup_fd = gpr_malloc(sizeof(*worker.wakeup_fd));
    grpc_wakeup_fd_init(&worker.wakeup_fd->fd);
  }
  worker.kicked_specifically = 0;
  /* If there's work waiting for the pollset to be idle, and the
     pollset is idle, then do that work */
  if (!pollset_has_workers(pollset) &&
      !grpc_closure_list_empty(pollset->idle_jobs)) {
    GPR_TIMER_MARK("pollset_work.idle_jobs", 0);
    grpc_exec_ctx_enqueue_list(exec_ctx, &pollset->idle_jobs, NULL);
    goto done;
  }
  /* If we're shutting down then we don't execute any extended work */
  if (pollset->shutting_down) {
    GPR_TIMER_MARK("pollset_work.shutting_down", 0);
    goto done;
  }
  /* Give do_promote priority so we don't starve it out */
  if (pollset->in_flight_cbs) {
    GPR_TIMER_MARK("pollset_work.in_flight_cbs", 0);
    gpr_mu_unlock(&pollset->mu);
    locked = 0;
    goto done;
  }
  /* Start polling, and keep doing so while we're being asked to
     re-evaluate our pollers (this allows poll() based pollers to
     ensure they don't miss wakeups) */
  keep_polling = 1;
  gpr_tls_set(&g_current_thread_poller, (intptr_t)pollset);
  while (keep_polling) {
    keep_polling = 0;
    if (!pollset->kicked_without_pollers) {
      if (!added_worker) {
        push_front_worker(pollset, &worker);
        added_worker = 1;
        gpr_tls_set(&g_current_thread_worker, (intptr_t)&worker);
      }
      GPR_TIMER_BEGIN("maybe_work_and_unlock", 0);
#define POLLOUT_CHECK (POLLOUT | POLLHUP | POLLERR)
#define POLLIN_CHECK (POLLIN | POLLHUP | POLLERR)

      int timeout;
      int r;
      size_t i, j, fd_count;
      nfds_t pfd_count;
      /* TODO(ctiller): inline some elements to avoid an allocation */
      grpc_fd_watcher *watchers;
      struct pollfd *pfds;

      timeout = poll_deadline_to_millis_timeout(deadline, now);
      /* TODO(ctiller): perform just one malloc here if we exceed the inline
       * case */
      pfds = gpr_malloc(sizeof(*pfds) * (pollset->fd_count + 2));
      watchers = gpr_malloc(sizeof(*watchers) * (pollset->fd_count + 2));
      fd_count = 0;
      pfd_count = 2;
      pfds[0].fd = GRPC_WAKEUP_FD_GET_READ_FD(&grpc_global_wakeup_fd);
      pfds[0].events = POLLIN;
      pfds[0].revents = 0;
      pfds[1].fd = GRPC_WAKEUP_FD_GET_READ_FD(&worker.wakeup_fd->fd);
      pfds[1].events = POLLIN;
      pfds[1].revents = 0;
      for (i = 0; i < pollset->fd_count; i++) {
        int remove = fd_is_orphaned(pollset->fds[i]);
        for (j = 0; !remove && j < pollset->del_count; j++) {
          if (pollset->fds[i] == pollset->dels[j]) remove = 1;
        }
        if (remove) {
          GRPC_FD_UNREF(pollset->fds[i], "multipoller");
        } else {
          pollset->fds[fd_count++] = pollset->fds[i];
          watchers[pfd_count].fd = pollset->fds[i];
          GRPC_FD_REF(watchers[pfd_count].fd, "multipoller_start");
          pfds[pfd_count].fd = pollset->fds[i]->fd;
          pfds[pfd_count].revents = 0;
          pfd_count++;
        }
      }
      for (j = 0; j < pollset->del_count; j++) {
        GRPC_FD_UNREF(pollset->dels[j], "multipoller_del");
      }
      pollset->del_count = 0;
      pollset->fd_count = fd_count;
      gpr_mu_unlock(&pollset->mu);

      for (i = 2; i < pfd_count; i++) {
        grpc_fd *fd = watchers[i].fd;
        pfds[i].events = (short)fd_begin_poll(fd, pollset, &worker, POLLIN,
                                              POLLOUT, &watchers[i]);
        GRPC_FD_UNREF(fd, "multipoller_start");
      }

      /* TODO(vpai): Consider first doing a 0 timeout poll here to avoid
         even going into the blocking annotation if possible */
      GRPC_SCHEDULING_START_BLOCKING_REGION;
      r = grpc_poll_function(pfds, pfd_count, timeout);
      GRPC_SCHEDULING_END_BLOCKING_REGION;

      if (r < 0) {
        if (errno != EINTR) {
          gpr_log(GPR_ERROR, "poll() failed: %s", strerror(errno));
        }
        for (i = 2; i < pfd_count; i++) {
          fd_end_poll(exec_ctx, &watchers[i], 0, 0, NULL);
        }
      } else if (r == 0) {
        for (i = 2; i < pfd_count; i++) {
          fd_end_poll(exec_ctx, &watchers[i], 0, 0, NULL);
        }
      } else {
        if (pfds[0].revents & POLLIN_CHECK) {
          grpc_wakeup_fd_consume_wakeup(&grpc_global_wakeup_fd);
        }
        if (pfds[1].revents & POLLIN_CHECK) {
          grpc_wakeup_fd_consume_wakeup(&worker.wakeup_fd->fd);
        }
        for (i = 2; i < pfd_count; i++) {
          if (watchers[i].fd == NULL) {
            fd_end_poll(exec_ctx, &watchers[i], 0, 0, NULL);
          } else {
            fd_end_poll(exec_ctx, &watchers[i], pfds[i].revents & POLLIN_CHECK,
                        pfds[i].revents & POLLOUT_CHECK, pollset);
          }
        }
      }

      gpr_free(pfds);
      gpr_free(watchers);
      GPR_TIMER_END("maybe_work_and_unlock", 0);
      locked = 0;
    } else {
      GPR_TIMER_MARK("pollset_work.kicked_without_pollers", 0);
      pollset->kicked_without_pollers = 0;
    }
  /* Finished execution - start cleaning up.
     Note that we may arrive here from outside the enclosing while() loop.
     In that case we won't loop though as we haven't added worker to the
     worker list, which means nobody could ask us to re-evaluate polling). */
  done:
    if (!locked) {
      queued_work |= grpc_exec_ctx_flush(exec_ctx);
      gpr_mu_lock(&pollset->mu);
      locked = 1;
    }
    /* If we're forced to re-evaluate polling (via pollset_kick with
       GRPC_POLLSET_REEVALUATE_POLLING_ON_WAKEUP) then we land here and force
       a loop */
    if (worker.reevaluate_polling_on_wakeup) {
      worker.reevaluate_polling_on_wakeup = 0;
      pollset->kicked_without_pollers = 0;
      if (queued_work || worker.kicked_specifically) {
        /* If there's queued work on the list, then set the deadline to be
           immediate so we get back out of the polling loop quickly */
        deadline = gpr_inf_past(GPR_CLOCK_MONOTONIC);
      }
      keep_polling = 1;
    }
    if (keep_polling) {
      now = gpr_now(now.clock_type);
    }
  }
  gpr_tls_set(&g_current_thread_poller, 0);
  if (added_worker) {
    remove_worker(pollset, &worker);
    gpr_tls_set(&g_current_thread_worker, 0);
  }
  /* release wakeup fd to the local pool */
  worker.wakeup_fd->next = pollset->local_wakeup_cache;
  pollset->local_wakeup_cache = worker.wakeup_fd;
  /* check shutdown conditions */
  if (pollset->shutting_down) {
    if (pollset_has_workers(pollset)) {
      pollset_kick(pollset, NULL);
    } else if (!pollset->called_shutdown && pollset->in_flight_cbs == 0) {
      pollset->called_shutdown = 1;
      gpr_mu_unlock(&pollset->mu);
      finish_shutdown(exec_ctx, pollset);
      grpc_exec_ctx_flush(exec_ctx);
      /* Continuing to access pollset here is safe -- it is the caller's
       * responsibility to not destroy when it has outstanding calls to
       * pollset_work.
       * TODO(dklempner): Can we refactor the shutdown logic to avoid this? */
      gpr_mu_lock(&pollset->mu);
    } else if (!grpc_closure_list_empty(pollset->idle_jobs)) {
      grpc_exec_ctx_enqueue_list(exec_ctx, &pollset->idle_jobs, NULL);
      gpr_mu_unlock(&pollset->mu);
      grpc_exec_ctx_flush(exec_ctx);
      gpr_mu_lock(&pollset->mu);
    }
  }
  *worker_hdl = NULL;
  GPR_TIMER_END("pollset_work", 0);
}

static void pollset_shutdown(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
                             grpc_closure *closure) {
  GPR_ASSERT(!pollset->shutting_down);
  pollset->shutting_down = 1;
  pollset->shutdown_done = closure;
  pollset_kick(pollset, GRPC_POLLSET_KICK_BROADCAST);
  if (!pollset_has_workers(pollset)) {
    grpc_exec_ctx_enqueue_list(exec_ctx, &pollset->idle_jobs, NULL);
  }
  if (!pollset->called_shutdown && pollset->in_flight_cbs == 0 &&
      !pollset_has_workers(pollset)) {
    pollset->called_shutdown = 1;
    finish_shutdown(exec_ctx, pollset);
  }
}

static int poll_deadline_to_millis_timeout(gpr_timespec deadline,
                                           gpr_timespec now) {
  gpr_timespec timeout;
  static const int64_t max_spin_polling_us = 10;
  if (gpr_time_cmp(deadline, gpr_inf_future(deadline.clock_type)) == 0) {
    return -1;
  }
  if (gpr_time_cmp(deadline, gpr_time_add(now, gpr_time_from_micros(
                                                   max_spin_polling_us,
                                                   GPR_TIMESPAN))) <= 0) {
    return 0;
  }
  timeout = gpr_time_sub(deadline, now);
  return gpr_time_to_millis(gpr_time_add(
      timeout, gpr_time_from_nanos(GPR_NS_PER_MS - 1, GPR_TIMESPAN)));
}

/*******************************************************************************
 * pollset_set_posix.c
 */

static grpc_pollset_set *pollset_set_create(void) {
  grpc_pollset_set *pollset_set = gpr_malloc(sizeof(*pollset_set));
  memset(pollset_set, 0, sizeof(*pollset_set));
  gpr_mu_init(&pollset_set->mu);
  return pollset_set;
}

static void pollset_set_destroy(grpc_pollset_set *pollset_set) {
  size_t i;
  gpr_mu_destroy(&pollset_set->mu);
  for (i = 0; i < pollset_set->fd_count; i++) {
    GRPC_FD_UNREF(pollset_set->fds[i], "pollset_set");
  }
  gpr_free(pollset_set->pollsets);
  gpr_free(pollset_set->pollset_sets);
  gpr_free(pollset_set->fds);
  gpr_free(pollset_set);
}

static void pollset_set_add_pollset(grpc_exec_ctx *exec_ctx,
                                    grpc_pollset_set *pollset_set,
                                    grpc_pollset *pollset) {
  size_t i, j;
  gpr_mu_lock(&pollset_set->mu);
  if (pollset_set->pollset_count == pollset_set->pollset_capacity) {
    pollset_set->pollset_capacity =
        GPR_MAX(8, 2 * pollset_set->pollset_capacity);
    pollset_set->pollsets =
        gpr_realloc(pollset_set->pollsets, pollset_set->pollset_capacity *
                                               sizeof(*pollset_set->pollsets));
  }
  pollset_set->pollsets[pollset_set->pollset_count++] = pollset;
  for (i = 0, j = 0; i < pollset_set->fd_count; i++) {
    if (fd_is_orphaned(pollset_set->fds[i])) {
      GRPC_FD_UNREF(pollset_set->fds[i], "pollset_set");
    } else {
      pollset_add_fd(exec_ctx, pollset, pollset_set->fds[i]);
      pollset_set->fds[j++] = pollset_set->fds[i];
    }
  }
  pollset_set->fd_count = j;
  gpr_mu_unlock(&pollset_set->mu);
}

static void pollset_set_del_pollset(grpc_exec_ctx *exec_ctx,
                                    grpc_pollset_set *pollset_set,
                                    grpc_pollset *pollset) {
  size_t i;
  gpr_mu_lock(&pollset_set->mu);
  for (i = 0; i < pollset_set->pollset_count; i++) {
    if (pollset_set->pollsets[i] == pollset) {
      pollset_set->pollset_count--;
      GPR_SWAP(grpc_pollset *, pollset_set->pollsets[i],
               pollset_set->pollsets[pollset_set->pollset_count]);
      break;
    }
  }
  gpr_mu_unlock(&pollset_set->mu);
}

static void pollset_set_add_pollset_set(grpc_exec_ctx *exec_ctx,
                                        grpc_pollset_set *bag,
                                        grpc_pollset_set *item) {
  size_t i, j;
  gpr_mu_lock(&bag->mu);
  if (bag->pollset_set_count == bag->pollset_set_capacity) {
    bag->pollset_set_capacity = GPR_MAX(8, 2 * bag->pollset_set_capacity);
    bag->pollset_sets =
        gpr_realloc(bag->pollset_sets,
                    bag->pollset_set_capacity * sizeof(*bag->pollset_sets));
  }
  bag->pollset_sets[bag->pollset_set_count++] = item;
  for (i = 0, j = 0; i < bag->fd_count; i++) {
    if (fd_is_orphaned(bag->fds[i])) {
      GRPC_FD_UNREF(bag->fds[i], "pollset_set");
    } else {
      pollset_set_add_fd(exec_ctx, item, bag->fds[i]);
      bag->fds[j++] = bag->fds[i];
    }
  }
  bag->fd_count = j;
  gpr_mu_unlock(&bag->mu);
}

static void pollset_set_del_pollset_set(grpc_exec_ctx *exec_ctx,
                                        grpc_pollset_set *bag,
                                        grpc_pollset_set *item) {
  size_t i;
  gpr_mu_lock(&bag->mu);
  for (i = 0; i < bag->pollset_set_count; i++) {
    if (bag->pollset_sets[i] == item) {
      bag->pollset_set_count--;
      GPR_SWAP(grpc_pollset_set *, bag->pollset_sets[i],
               bag->pollset_sets[bag->pollset_set_count]);
      break;
    }
  }
  gpr_mu_unlock(&bag->mu);
}

static void pollset_set_add_fd(grpc_exec_ctx *exec_ctx,
                               grpc_pollset_set *pollset_set, grpc_fd *fd) {
  size_t i;
  gpr_mu_lock(&pollset_set->mu);
  if (pollset_set->fd_count == pollset_set->fd_capacity) {
    pollset_set->fd_capacity = GPR_MAX(8, 2 * pollset_set->fd_capacity);
    pollset_set->fds = gpr_realloc(
        pollset_set->fds, pollset_set->fd_capacity * sizeof(*pollset_set->fds));
  }
  GRPC_FD_REF(fd, "pollset_set");
  pollset_set->fds[pollset_set->fd_count++] = fd;
  for (i = 0; i < pollset_set->pollset_count; i++) {
    pollset_add_fd(exec_ctx, pollset_set->pollsets[i], fd);
  }
  for (i = 0; i < pollset_set->pollset_set_count; i++) {
    pollset_set_add_fd(exec_ctx, pollset_set->pollset_sets[i], fd);
  }
  gpr_mu_unlock(&pollset_set->mu);
}

static void pollset_set_del_fd(grpc_exec_ctx *exec_ctx,
                               grpc_pollset_set *pollset_set, grpc_fd *fd) {
  size_t i;
  gpr_mu_lock(&pollset_set->mu);
  for (i = 0; i < pollset_set->fd_count; i++) {
    if (pollset_set->fds[i] == fd) {
      pollset_set->fd_count--;
      GPR_SWAP(grpc_fd *, pollset_set->fds[i],
               pollset_set->fds[pollset_set->fd_count]);
      GRPC_FD_UNREF(fd, "pollset_set");
      break;
    }
  }
  for (i = 0; i < pollset_set->pollset_set_count; i++) {
    pollset_set_del_fd(exec_ctx, pollset_set->pollset_sets[i], fd);
  }
  gpr_mu_unlock(&pollset_set->mu);
}

/*******************************************************************************
 * event engine binding
 */

static void shutdown_engine(void) { pollset_global_shutdown(); }

static const grpc_event_engine_vtable vtable = {
    .pollset_size = sizeof(grpc_pollset),

    .fd_create = fd_create,
    .fd_wrapped_fd = fd_wrapped_fd,
    .fd_orphan = fd_orphan,
    .fd_shutdown = fd_shutdown,
    .fd_notify_on_read = fd_notify_on_read,
    .fd_notify_on_write = fd_notify_on_write,
    .fd_get_read_notifier_pollset = fd_get_read_notifier_pollset,

    .pollset_init = pollset_init,
    .pollset_shutdown = pollset_shutdown,
    .pollset_reset = pollset_reset,
    .pollset_destroy = pollset_destroy,
    .pollset_work = pollset_work,
    .pollset_kick = pollset_kick,
    .pollset_add_fd = pollset_add_fd,

    .pollset_set_create = pollset_set_create,
    .pollset_set_destroy = pollset_set_destroy,
    .pollset_set_add_pollset = pollset_set_add_pollset,
    .pollset_set_del_pollset = pollset_set_del_pollset,
    .pollset_set_add_pollset_set = pollset_set_add_pollset_set,
    .pollset_set_del_pollset_set = pollset_set_del_pollset_set,
    .pollset_set_add_fd = pollset_set_add_fd,
    .pollset_set_del_fd = pollset_set_del_fd,

    .kick_poller = kick_poller,

    .shutdown_engine = shutdown_engine,
};

const grpc_event_engine_vtable *grpc_init_poll_posix(void) {
  pollset_global_init();
  return &vtable;
}

#endif