-
Notifications
You must be signed in to change notification settings - Fork 10
/
Copy pathio_wait.h
1325 lines (1232 loc) · 37.6 KB
/
io_wait.h
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
/*
* $Id$
*
* Copyright (C) 2005 iptelorg GmbH
*
* Permission to use, copy, modify, and distribute this software for any
* purpose with or without fee is hereby granted, provided that the above
* copyright notice and this permission notice appear in all copies.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
*/
/*
* tcp io wait common stuff used by tcp_main.c & tcp_read.c
* All the functions are inline because of speed reasons and because they are
* used only from 2 places.
* You also have to define:
* int handle_io(struct fd_map* fm, short events, int idx) (see below)
* (this could be trivially replaced by a callback pointer entry attached
* to the io_wait handler if more flexibility rather then performance
* is needed)
* fd_type - define to some enum of you choice and define also
* FD_TYPE_DEFINED (if you don't do it fd_type will be defined
* to int). 0 has a special not set/not init. meaning
* (a lot of sanity checks and the sigio_rt code are based on
* this assumption)
* local_malloc (defaults to pkg_malloc)
* local_free (defaults to pkg_free)
*
*/
/*
* History:
* --------
* 2005-06-13 created by andrei
* 2005-06-26 added kqueue (andrei)
* 2005-07-01 added /dev/poll (andrei)
* 2006-05-30 sigio 64 bit workarround enabled for kernels < 2.6.5 (andrei)
* 2007-11-22 when handle_io() is called in a loop check & stop if the fd was
* removed inside handle_io() (andrei)
* 2007-11-29 support for write (POLLOUT); added io_watch_chg() (andrei)
* 2008-02-04 POLLRDHUP & EPOLLRDHUP support (automatically enabled if POLLIN
* is set) (andrei)
*/
#ifndef _io_wait_h
#define _io_wait_h
#include <errno.h>
#include <string.h>
#ifdef HAVE_SIGIO_RT
#define __USE_GNU /* or else F_SETSIG won't be included */
#include <sys/types.h> /* recv */
#include <sys/socket.h> /* recv */
#include <signal.h> /* sigprocmask, sigwait a.s.o */
#endif
#define _GNU_SOURCE /* for POLLRDHUP on linux */
#include <sys/poll.h>
#include <fcntl.h>
#ifdef HAVE_EPOLL
#include <sys/epoll.h>
#endif
#ifdef HAVE_KQUEUE
#include <sys/types.h> /* needed on freebsd */
#include <sys/event.h>
#include <sys/time.h>
#endif
#ifdef HAVE_DEVPOLL
#include <sys/devpoll.h>
#endif
#ifdef HAVE_SELECT
/* needed on openbsd for select*/
#include <sys/time.h>
#include <sys/types.h>
#include <unistd.h>
/* needed according to POSIX for select*/
#include <sys/select.h>
#endif
#include "dprint.h"
#include "poll_types.h" /* poll_types*/
#ifdef HAVE_SIGIO_RT
#include "pt.h" /* mypid() */
#endif
#include "compiler_opt.h"
#ifdef HAVE_EPOLL
/* fix defines for EPOLL */
#if defined POLLRDHUP && ! defined EPOLLRDHUP
#define EPOLLRDHUP POLLRDHUP /* should work on all linuxes */
#endif /* POLLRDHUP && EPOLLRDHUP */
#endif /* HAVE_EPOLL */
extern int _os_ver; /* os version number, needed to select bugs workarrounds */
#if 0
enum fd_types; /* this should be defined from the including file,
see tcp_main.c for an example,
0 has a special meaning: not used/empty*/
#endif
#ifndef FD_TYPE_DEFINED
typedef int fd_type;
#define FD_TYPE_DEFINED
#endif
/* maps a fd to some other structure; used in almost all cases
* except epoll and maybe kqueue or /dev/poll */
struct fd_map{
int fd; /* fd no */
fd_type type; /* "data" type */
void* data; /* pointer to the corresponding structure */
short events; /* events we are interested int */
};
#ifdef HAVE_KQUEUE
#ifndef KQ_CHANGES_ARRAY_SIZE
#define KQ_CHANGES_ARRAY_SIZE 256
#ifdef __OS_netbsd
#define KEV_UDATA_CAST (intptr_t)
#else
#define KEV_UDATA_CAST
#endif
#endif
#endif
/* handler structure */
struct io_wait_handler{
enum poll_types poll_method;
int flags;
struct fd_map* fd_hash;
int fd_no; /* current index used in fd_array and the passed size for
ep_array & kq_array*/
int max_fd_no; /* maximum fd no, is also the size of fd_array,
fd_hash and ep_array*/
/* common stuff for POLL, SIGIO_RT and SELECT
* since poll support is always compiled => this will always be compiled */
struct pollfd* fd_array; /* used also by devpoll as devpoll array */
int crt_fd_array_idx; /* crt idx for which handle_io is called
(updated also by del -> internal optimization) */
/* end of common stuff */
#ifdef HAVE_EPOLL
int epfd; /* epoll ctrl fd */
struct epoll_event* ep_array;
#endif
#ifdef HAVE_SIGIO_RT
sigset_t sset; /* signal mask for sigio & sigrtmin */
int signo; /* real time signal used */
#endif
#ifdef HAVE_KQUEUE
int kq_fd;
struct kevent* kq_array; /* used for the eventlist*/
struct kevent* kq_changes; /* used for the changelist */
size_t kq_nchanges;
size_t kq_changes_size; /* size of the changes array */
#endif
#ifdef HAVE_DEVPOLL
int dpoll_fd;
#endif
#ifdef HAVE_SELECT
fd_set master_rset; /* read set */
fd_set master_wset; /* write set */
int max_fd_select; /* maximum select used fd */
#endif
};
typedef struct io_wait_handler io_wait_h;
/* get the corresponding fd_map structure pointer */
#define get_fd_map(h, fd) (&(h)->fd_hash[(fd)])
/* remove a fd_map structure from the hash; the pointer must be returned
* by get_fd_map or hash_fd_map*/
#define unhash_fd_map(pfm) \
do{ \
(pfm)->type=0 /*F_NONE */; \
(pfm)->fd=-1; \
}while(0)
/* add a fd_map structure to the fd hash */
static inline struct fd_map* hash_fd_map( io_wait_h* h,
int fd,
short events,
fd_type type,
void* data)
{
h->fd_hash[fd].fd=fd;
h->fd_hash[fd].events=events;
h->fd_hash[fd].type=type;
h->fd_hash[fd].data=data;
return &h->fd_hash[fd];
}
#ifdef HANDLE_IO_INLINE
/* generic handle io routine, this must be defined in the including file
* (faster then registering a callback pointer)
*
* params: fm - pointer to a fd hash entry
* events - combinations of POLLIN, POLLOUT, POLLERR & POLLHUP
* idx - index in the fd_array (or -1 if not known)
* return: -1 on error
* 0 on EAGAIN or when by some other way it is known that no more
* io events are queued on the fd (the receive buffer is empty).
* Usefull to detect when there are no more io events queued for
* sigio_rt, epoll_et, kqueue.
* >0 on successfull read from the fd (when there might be more io
* queued -- the receive buffer might still be non-empty)
*/
inline static int handle_io(struct fd_map* fm, short events, int idx);
#else
int handle_io(struct fd_map* fm, short events, int idx);
#endif
#ifdef HAVE_KQUEUE
/*
* kqueue specific function: register a change
* (adds a change to the kevent change array, and if full flushes it first)
*
* TODO: check if the event already exists in the change list or if it's
* complementary to an event in the list (e.g. EVFILT_WRITE, EV_DELETE
* and EVFILT_WRITE, EV_ADD for the same fd).
* returns: -1 on error, 0 on success
*/
static inline int kq_ev_change(io_wait_h* h, int fd, int filter, int flag,
void* data)
{
int n;
struct timespec tspec;
if (h->kq_nchanges>=h->kq_changes_size){
/* changes array full ! */
LOG(L_WARN, "WARNING: kq_ev_change: kqueue changes array full"
" trying to flush...\n");
tspec.tv_sec=0;
tspec.tv_nsec=0;
again:
n=kevent(h->kq_fd, h->kq_changes, h->kq_nchanges, 0, 0, &tspec);
if (n==-1){
if (errno==EINTR) goto again;
LOG(L_ERR, "ERROR: io_watch_add: kevent flush changes "
" failed: %s [%d]\n", strerror(errno), errno);
return -1;
}
h->kq_nchanges=0; /* changes array is empty */
}
EV_SET(&h->kq_changes[h->kq_nchanges], fd, filter, flag, 0, 0,
KEV_UDATA_CAST data);
h->kq_nchanges++;
return 0;
}
#endif
/* generic io_watch_add function
* Params:
* h - pointer to initialized io_wait handle
* fd - fd to watch
* events - bitmap with the fd events for which the fd should be watched
* (combination of POLLIN and POLLOUT)
* type - fd type (non 0 value, returned in the call to handle_io)
* data - pointer/private data returned in the handle_io call
* returns 0 on success, -1 on error
*
* WARNING: handle_io() can be called immediately (from io_watch_add()) so
* make sure that any dependent init. (e.g. data stuff) is made before
* calling io_watch_add
*
* this version should be faster than pointers to poll_method specific
* functions (it avoids functions calls, the overhead being only an extra
* switch())*/
inline static int io_watch_add( io_wait_h* h,
int fd,
short events,
fd_type type,
void* data)
{
/* helper macros */
#define fd_array_setup(ev) \
do{ \
h->fd_array[h->fd_no].fd=fd; \
h->fd_array[h->fd_no].events=(ev); /* useless for select */ \
h->fd_array[h->fd_no].revents=0; /* useless for select */ \
}while(0)
#define set_fd_flags(f) \
do{ \
flags=fcntl(fd, F_GETFL); \
if (flags==-1){ \
LOG(L_ERR, "ERROR: io_watch_add: fnctl: GETFL failed:" \
" %s [%d]\n", strerror(errno), errno); \
goto error; \
} \
if (fcntl(fd, F_SETFL, flags|(f))==-1){ \
LOG(L_ERR, "ERROR: io_watch_add: fnctl: SETFL" \
" failed: %s [%d]\n", strerror(errno), errno); \
goto error; \
} \
}while(0)
struct fd_map* e;
int flags;
#ifdef HAVE_EPOLL
struct epoll_event ep_event;
#endif
#ifdef HAVE_DEVPOLL
struct pollfd pfd;
#endif
#if defined(HAVE_SIGIO_RT) || defined (HAVE_EPOLL)
int n;
#endif
#if defined(HAVE_SIGIO_RT)
int idx;
int check_io;
struct pollfd pf;
check_io=0; /* set to 1 if we need to check for pre-existing queued
io/data on the fd */
idx=-1;
#endif
e=0;
/* sanity checks */
if (unlikely(fd==-1)){
LOG(L_CRIT, "BUG: io_watch_add: fd is -1!\n");
goto error;
}
if (unlikely((events&(POLLIN|POLLOUT))==0)){
LOG(L_CRIT, "BUG: io_watch_add: invalid events: 0x%0x\n", events);
goto error;
}
/* check if not too big */
if (unlikely(h->fd_no>=h->max_fd_no)){
LOG(L_CRIT, "ERROR: io_watch_add: maximum fd number exceeded:"
" %d/%d\n", h->fd_no, h->max_fd_no);
goto error;
}
DBG("DBG: io_watch_add(%p, %d, %d, %p), fd_no=%d\n",
h, fd, type, data, h->fd_no);
/* hash sanity check */
e=get_fd_map(h, fd);
if (unlikely(e && (e->type!=0 /*F_NONE*/))){
LOG(L_ERR, "ERROR: io_watch_add: trying to overwrite entry %d"
" watched for %x in the hash(%d, %d, %p) with (%d, %d, %p)\n",
fd, events, e->fd, e->type, e->data, fd, type, data);
e=0;
goto error;
}
if (unlikely((e=hash_fd_map(h, fd, events, type, data))==0)){
LOG(L_ERR, "ERROR: io_watch_add: failed to hash the fd %d\n", fd);
goto error;
}
switch(h->poll_method){ /* faster then pointer to functions */
case POLL_POLL:
#ifdef POLLRDHUP
/* listen to POLLRDHUP by default (if POLLIN) */
events|=((int)!(events & POLLIN) - 1) & POLLRDHUP;
#endif /* POLLRDHUP */
fd_array_setup(events);
set_fd_flags(O_NONBLOCK);
break;
#ifdef HAVE_SELECT
case POLL_SELECT:
fd_array_setup(events);
if (likely(events & POLLIN))
FD_SET(fd, &h->master_rset);
if (unlikely(events & POLLOUT))
FD_SET(fd, &h->master_wset);
if (h->max_fd_select<fd) h->max_fd_select=fd;
break;
#endif
#ifdef HAVE_SIGIO_RT
case POLL_SIGIO_RT:
fd_array_setup(events);
/* re-set O_ASYNC might be needed, if not done from
* io_watch_del (or if somebody wants to add a fd which has
* already O_ASYNC/F_SETSIG set on a duplicate)
*/
/* set async & signal */
if (fcntl(fd, F_SETOWN, my_pid())==-1){
LOG(L_ERR, "ERROR: io_watch_add: fnctl: SETOWN"
" failed: %s [%d]\n", strerror(errno), errno);
goto error;
}
if (fcntl(fd, F_SETSIG, h->signo)==-1){
LOG(L_ERR, "ERROR: io_watch_add: fnctl: SETSIG"
" failed: %s [%d]\n", strerror(errno), errno);
goto error;
}
/* set both non-blocking and async */
set_fd_flags(O_ASYNC| O_NONBLOCK);
#ifdef EXTRA_DEBUG
DBG("io_watch_add: sigio_rt on f %d, signal %d to pid %d\n",
fd, h->signo, my_pid());
#endif
/* empty socket receive buffer, if buffer is already full
* no more space to put packets
* => no more signals are ever generated
* also when moving fds, the freshly moved fd might have
* already some bytes queued, we want to get them now
* and not later -- andrei */
idx=h->fd_no;
check_io=1;
break;
#endif
#ifdef HAVE_EPOLL
case POLL_EPOLL_LT:
ep_event.events=
#ifdef POLLRDHUP
/* listen for EPOLLRDHUP too */
((EPOLLIN|EPOLLRDHUP) & ((int)!(events & POLLIN)-1) ) |
#else /* POLLRDHUP */
(EPOLLIN & ((int)!(events & POLLIN)-1) ) |
#endif /* POLLRDHUP */
(EPOLLOUT & ((int)!(events & POLLOUT)-1) );
ep_event.data.ptr=e;
again1:
n=epoll_ctl(h->epfd, EPOLL_CTL_ADD, fd, &ep_event);
if (unlikely(n==-1)){
if (errno==EAGAIN) goto again1;
LOG(L_ERR, "ERROR: io_watch_add: epoll_ctl failed: %s [%d]\n",
strerror(errno), errno);
goto error;
}
break;
case POLL_EPOLL_ET:
set_fd_flags(O_NONBLOCK);
ep_event.events=
#ifdef POLLRDHUP
/* listen for EPOLLRDHUP too */
((EPOLLIN|EPOLLRDHUP) & ((int)!(events & POLLIN)-1) ) |
#else /* POLLRDHUP */
(EPOLLIN & ((int)!(events & POLLIN)-1) ) |
#endif /* POLLRDHUP */
(EPOLLOUT & ((int)!(events & POLLOUT)-1) ) |
EPOLLET;
ep_event.data.ptr=e;
again2:
n=epoll_ctl(h->epfd, EPOLL_CTL_ADD, fd, &ep_event);
if (unlikely(n==-1)){
if (errno==EAGAIN) goto again2;
LOG(L_ERR, "ERROR: io_watch_add: epoll_ctl failed: %s [%d]\n",
strerror(errno), errno);
goto error;
}
break;
#endif
#ifdef HAVE_KQUEUE
case POLL_KQUEUE:
if (likely( events & POLLIN)){
if (unlikely(kq_ev_change(h, fd, EVFILT_READ, EV_ADD, e)==-1))
goto error;
}
if (unlikely( events & POLLOUT)){
if (unlikely(kq_ev_change(h, fd, EVFILT_WRITE, EV_ADD, e)==-1))
{
if (likely(events & POLLIN)){
kq_ev_change(h, fd, EVFILT_READ, EV_DELETE, 0);
}
}
goto error;
}
break;
#endif
#ifdef HAVE_DEVPOLL
case POLL_DEVPOLL:
pfd.fd=fd;
pfd.events=events;
pfd.revents=0;
again_devpoll:
if (write(h->dpoll_fd, &pfd, sizeof(pfd))==-1){
if (errno==EAGAIN) goto again_devpoll;
LOG(L_ERR, "ERROR: io_watch_add: /dev/poll write failed:"
"%s [%d]\n", strerror(errno), errno);
goto error;
}
break;
#endif
default:
LOG(L_CRIT, "BUG: io_watch_add: no support for poll method "
" %s (%d)\n", poll_method_str[h->poll_method],
h->poll_method);
goto error;
}
h->fd_no++; /* "activate" changes, for epoll/kqueue/devpoll it
has only informative value */
#if defined(HAVE_SIGIO_RT)
if (check_io){
/* handle possible pre-existing events */
pf.fd=fd;
pf.events=events;
check_io_again:
n=0;
while(e->type && ((n=poll(&pf, 1, 0))>0) &&
(handle_io(e, pf.revents, idx)>0) &&
(pf.revents & (e->events|POLLERR|POLLHUP)));
if (unlikely(e->type && (n==-1))){
if (errno==EINTR) goto check_io_again;
LOG(L_ERR, "ERROR: io_watch_add: check_io poll: %s [%d]\n",
strerror(errno), errno);
}
}
#endif
return 0;
error:
if (e) unhash_fd_map(e);
return -1;
#undef fd_array_setup
#undef set_fd_flags
}
#define IO_FD_CLOSING 16
/* parameters: h - handler
* fd - file descriptor
* index - index in the fd_array if known, -1 if not
* (if index==-1 fd_array will be searched for the
* corresponding fd* entry -- slower but unavoidable in
* some cases). index is not used (no fd_array) for epoll,
* /dev/poll and kqueue
* flags - optimization flags, e.g. IO_FD_CLOSING, the fd was
* or will shortly be closed, in some cases we can avoid
* extra remove operations (e.g.: epoll, kqueue, sigio)
* returns 0 if ok, -1 on error */
inline static int io_watch_del(io_wait_h* h, int fd, int idx, int flags)
{
#define fix_fd_array \
do{\
if (unlikely(idx==-1)){ \
/* fix idx if -1 and needed */ \
for (idx=0; (idx<h->fd_no) && \
(h->fd_array[idx].fd!=fd); idx++); \
} \
if (likely(idx<h->fd_no)){ \
memmove(&h->fd_array[idx], &h->fd_array[idx+1], \
(h->fd_no-(idx+1))*sizeof(*(h->fd_array))); \
if ((idx<=h->crt_fd_array_idx) && (h->crt_fd_array_idx>=0)) \
h->crt_fd_array_idx--; \
} \
}while(0)
struct fd_map* e;
int events;
#ifdef HAVE_EPOLL
int n;
struct epoll_event ep_event;
#endif
#ifdef HAVE_DEVPOLL
struct pollfd pfd;
#endif
#ifdef HAVE_SIGIO_RT
int fd_flags;
#endif
if (unlikely((fd<0) || (fd>=h->max_fd_no))){
LOG(L_CRIT, "BUG: io_watch_del: invalid fd %d, not in [0, %d) \n",
fd, h->fd_no);
goto error;
}
DBG("DBG: io_watch_del (%p, %d, %d, 0x%x) fd_no=%d called\n",
h, fd, idx, flags, h->fd_no);
e=get_fd_map(h, fd);
/* more sanity checks */
if (unlikely(e==0)){
LOG(L_CRIT, "BUG: io_watch_del: no corresponding hash entry for %d\n",
fd);
goto error;
}
if (unlikely(e->type==0 /*F_NONE*/)){
LOG(L_ERR, "ERROR: io_watch_del: trying to delete already erased"
" entry %d in the hash(%d, %d, %p) flags %x)\n",
fd, e->fd, e->type, e->data, flags);
goto error;
}
events=e->events;
unhash_fd_map(e);
switch(h->poll_method){
case POLL_POLL:
fix_fd_array;
break;
#ifdef HAVE_SELECT
case POLL_SELECT:
if (likely(events & POLLIN))
FD_CLR(fd, &h->master_rset);
if (unlikely(events & POLLOUT))
FD_CLR(fd, &h->master_wset);
if (unlikely(h->max_fd_select && (h->max_fd_select==fd)))
/* we don't know the prev. max, so we just decrement it */
h->max_fd_select--;
fix_fd_array;
break;
#endif
#ifdef HAVE_SIGIO_RT
case POLL_SIGIO_RT:
fix_fd_array;
/* the O_ASYNC flag must be reset all the time, the fd
* can be changed only if O_ASYNC is reset (if not and
* the fd is a duplicate, you will get signals from the dup. fd
* and not from the original, even if the dup. fd was closed
* and the signals re-set on the original) -- andrei
*/
/*if (!(flags & IO_FD_CLOSING)){*/
/* reset ASYNC */
fd_flags=fcntl(fd, F_GETFL);
if (unlikely(fd_flags==-1)){
LOG(L_ERR, "ERROR: io_watch_del: fnctl: GETFL failed:"
" %s [%d]\n", strerror(errno), errno);
goto error;
}
if (unlikely(fcntl(fd, F_SETFL, fd_flags&(~O_ASYNC))==-1)){
LOG(L_ERR, "ERROR: io_watch_del: fnctl: SETFL"
" failed: %s [%d]\n", strerror(errno), errno);
goto error;
}
break;
#endif
#ifdef HAVE_EPOLL
case POLL_EPOLL_LT:
case POLL_EPOLL_ET:
/* epoll doesn't seem to automatically remove sockets,
* if the socket is a duplicate/moved and the original
* is still open. The fd is removed from the epoll set
* only when the original (and all the copies?) is/are
* closed. This is probably a bug in epoll. --andrei */
#ifdef EPOLL_NO_CLOSE_BUG
if (!(flags & IO_FD_CLOSING)){
#endif
again_epoll:
n=epoll_ctl(h->epfd, EPOLL_CTL_DEL, fd, &ep_event);
if (unlikely(n==-1)){
if (errno==EAGAIN) goto again_epoll;
LOG(L_ERR, "ERROR: io_watch_del: removing fd from epoll "
"list failed: %s [%d]\n", strerror(errno), errno);
goto error;
}
#ifdef EPOLL_NO_CLOSE_BUG
}
#endif
break;
#endif
#ifdef HAVE_KQUEUE
case POLL_KQUEUE:
if (!(flags & IO_FD_CLOSING)){
if (likely(events & POLLIN)){
if (unlikely(kq_ev_change(h, fd, EVFILT_READ,
EV_DELETE, 0) ==-1)){
/* try to delete the write filter anyway */
if (events & POLLOUT){
kq_ev_change(h, fd, EVFILT_WRITE, EV_DELETE, 0);
}
goto error;
}
}
if (unlikely(events & POLLOUT)){
if (unlikely(kq_ev_change(h, fd, EVFILT_WRITE,
EV_DELETE, 0) ==-1))
goto error;
}
}
break;
#endif
#ifdef HAVE_DEVPOLL
case POLL_DEVPOLL:
/* for /dev/poll the closed fds _must_ be removed
(they are not removed automatically on close()) */
pfd.fd=fd;
pfd.events=POLLREMOVE;
pfd.revents=0;
again_devpoll:
if (write(h->dpoll_fd, &pfd, sizeof(pfd))==-1){
if (errno==EINTR) goto again_devpoll;
LOG(L_ERR, "ERROR: io_watch_del: removing fd from "
"/dev/poll failed: %s [%d]\n",
strerror(errno), errno);
goto error;
}
break;
#endif
default:
LOG(L_CRIT, "BUG: io_watch_del: no support for poll method "
" %s (%d)\n", poll_method_str[h->poll_method],
h->poll_method);
goto error;
}
h->fd_no--;
return 0;
error:
return -1;
#undef fix_fd_array
}
/* parameters: h - handler
* fd - file descriptor
* events - new events to watch for
* idx - index in the fd_array if known, -1 if not
* (if index==-1 fd_array will be searched for the
* corresponding fd* entry -- slower but unavoidable in
* some cases). index is not used (no fd_array) for epoll,
* /dev/poll and kqueue
* returns 0 if ok, -1 on error */
inline static int io_watch_chg(io_wait_h* h, int fd, short events, int idx )
{
#define fd_array_chg(ev) \
do{\
if (unlikely(idx==-1)){ \
/* fix idx if -1 and needed */ \
for (idx=0; (idx<h->fd_no) && \
(h->fd_array[idx].fd!=fd); idx++); \
} \
if (likely(idx<h->fd_no)){ \
h->fd_array[idx].events=(ev); \
} \
}while(0)
struct fd_map* e;
int add_events;
int del_events;
#ifdef HAVE_DEVPOLL
struct pollfd pfd;
#endif
#ifdef HAVE_EPOLL
int n;
struct epoll_event ep_event;
#endif
if (unlikely((fd<0) || (fd>=h->max_fd_no))){
LOG(L_CRIT, "BUG: io_watch_chg: invalid fd %d, not in [0, %d) \n",
fd, h->fd_no);
goto error;
}
if (unlikely((events&(POLLIN|POLLOUT))==0)){
LOG(L_CRIT, "BUG: io_watch_chg: invalid events: 0x%0x\n", events);
goto error;
}
DBG("DBG: io_watch_chg (%p, %d, 0x%x, 0x%x) fd_no=%d called\n",
h, fd, events, idx, h->fd_no);
e=get_fd_map(h, fd);
/* more sanity checks */
if (unlikely(e==0)){
LOG(L_CRIT, "BUG: io_watch_chg: no corresponding hash entry for %d\n",
fd);
goto error;
}
if (unlikely(e->type==0 /*F_NONE*/)){
LOG(L_ERR, "ERROR: io_watch_chg: trying to change an already erased"
" entry %d in the hash(%d, %d, %p) )\n",
fd, e->fd, e->type, e->data);
goto error;
}
add_events=events & ~e->events;
del_events=e->events & ~events;
e->events=events;
switch(h->poll_method){
case POLL_POLL:
#ifdef POLLRDHUP
/* listen to POLLRDHUP by default (if POLLIN) */
events|=((int)!(events & POLLIN) - 1) & POLLRDHUP;
#endif /* POLLRDHUP */
fd_array_chg(events);
break;
#ifdef HAVE_SELECT
case POLL_SELECT:
fd_array_chg(events);
if (unlikely(del_events & POLLIN))
FD_CLR(fd, &h->master_rset);
else if (unlikely(add_events & POLLIN))
FD_SET(fd, &h->master_rset);
if (likely(del_events & POLLOUT))
FD_CLR(fd, &h->master_wset);
else if (likely(add_events & POLLOUT))
FD_SET(fd, &h->master_wset);
break;
#endif
#ifdef HAVE_SIGIO_RT
case POLL_SIGIO_RT:
fd_array_chg(events);
/* no need for check_io, since SIGIO_RT listens by default for all
* the events */
break;
#endif
#ifdef HAVE_EPOLL
case POLL_EPOLL_LT:
ep_event.events=
#ifdef POLLRDHUP
/* listen for EPOLLRDHUP too */
((EPOLLIN|EPOLLRDHUP) & ((int)!(events & POLLIN)-1) ) |
#else /* POLLRDHUP */
(EPOLLIN & ((int)!(events & POLLIN)-1) ) |
#endif /* POLLRDHUP */
(EPOLLOUT & ((int)!(events & POLLOUT)-1) );
ep_event.data.ptr=e;
again_epoll_lt:
n=epoll_ctl(h->epfd, EPOLL_CTL_MOD, fd, &ep_event);
if (unlikely(n==-1)){
if (errno==EAGAIN) goto again_epoll_lt;
LOG(L_ERR, "ERROR: io_watch_chg: modifying epoll events"
" failed: %s [%d]\n", strerror(errno), errno);
goto error;
}
break;
case POLL_EPOLL_ET:
ep_event.events=
#ifdef POLLRDHUP
/* listen for EPOLLRDHUP too */
((EPOLLIN|EPOLLRDHUP) & ((int)!(events & POLLIN)-1) ) |
#else /* POLLRDHUP */
(EPOLLIN & ((int)!(events & POLLIN)-1) ) |
#endif /* POLLRDHUP */
(EPOLLOUT & ((int)!(events & POLLOUT)-1) ) |
EPOLLET;
ep_event.data.ptr=e;
again_epoll_et:
n=epoll_ctl(h->epfd, EPOLL_CTL_MOD, fd, &ep_event);
if (unlikely(n==-1)){
if (errno==EAGAIN) goto again_epoll_et;
LOG(L_ERR, "ERROR: io_watch_chg: modifying epoll events"
" failed: %s [%d]\n", strerror(errno), errno);
goto error;
}
break;
#endif
#ifdef HAVE_KQUEUE
case POLL_KQUEUE:
if (unlikely(del_events & POLLIN)){
if (unlikely(kq_ev_change(h, fd, EVFILT_READ,
EV_DELETE, 0) ==-1))
goto error;
}else if (unlikely(add_events & POLLIN)){
if (unlikely(kq_ev_change(h, fd, EVFILT_READ, EV_ADD, e) ==-1))
goto error;
}
if (likely(del_events & POLLOUT)){
if (unlikely(kq_ev_change(h, fd, EVFILT_WRITE,
EV_DELETE, 0) ==-1))
goto error;
}else if (likely(add_events & POLLOUT)){
if (unlikely(kq_ev_change(h, fd, EVFILT_WRITE, EV_ADD, e)==-1))
goto error;
}
break;
#endif
#ifdef HAVE_DEVPOLL
case POLL_DEVPOLL:
/* for /dev/poll the closed fds _must_ be removed
(they are not removed automatically on close()) */
pfd.fd=fd;
pfd.events=POLLREMOVE;
pfd.revents=0;
again_devpoll1:
if (unlikely(write(h->dpoll_fd, &pfd, sizeof(pfd))==-1)){
if (errno==EINTR) goto again_devpoll1;
LOG(L_ERR, "ERROR: io_watch_chg: removing fd from "
"/dev/poll failed: %s [%d]\n",
strerror(errno), errno);
goto error;
}
again_devpoll2:
pfd.events=events;
pfd.revents=0;
if (unlikely(write(h->dpoll_fd, &pfd, sizeof(pfd))==-1)){
if (errno==EINTR) goto again_devpoll2;
LOG(L_ERR, "ERROR: io_watch_chg: re-adding fd to "
"/dev/poll failed: %s [%d]\n",
strerror(errno), errno);
goto error;
}
break;
#endif
default:
LOG(L_CRIT, "BUG: io_watch_chg: no support for poll method "
" %s (%d)\n", poll_method_str[h->poll_method],
h->poll_method);
goto error;
}
return 0;
error:
return -1;
#undef fix_fd_array
}
/* io_wait_loop_x style function
* wait for io using poll()
* params: h - io_wait handle
* t - timeout in s
* repeat - if !=0 handle_io will be called until it returns <=0
* returns: number of IO events handled on success (can be 0), -1 on error
*/
inline static int io_wait_loop_poll(io_wait_h* h, int t, int repeat)
{
int n, r;
int ret;
struct fd_map* fm;
again:
ret=n=poll(h->fd_array, h->fd_no, t*1000);
if (n==-1){
if (errno==EINTR) goto again; /* signal, ignore it */
else{
LOG(L_ERR, "ERROR:io_wait_loop_poll: poll: %s [%d]\n",
strerror(errno), errno);
goto error;
}
}
for (r=0; (r<h->fd_no) && n; r++){
fm=get_fd_map(h, h->fd_array[r].fd);
if (h->fd_array[r].revents & (fm->events|POLLERR|POLLHUP)){
n--;
/* sanity checks */
if (unlikely((h->fd_array[r].fd >= h->max_fd_no)||
(h->fd_array[r].fd < 0))){
LOG(L_CRIT, "BUG: io_wait_loop_poll: bad fd %d "
"(no in the 0 - %d range)\n",
h->fd_array[r].fd, h->max_fd_no);
/* try to continue anyway */
h->fd_array[r].events=0; /* clear the events */
continue;
}
h->crt_fd_array_idx=r;
/* repeat handle_io if repeat, fd still watched (not deleted
* inside handle_io), handle_io returns that there's still
* IO and the fd is still watched for the triggering event */
while(fm->type &&
(handle_io(fm, h->fd_array[r].revents, r) > 0) &&
repeat && ((fm->events|POLLERR|POLLHUP) &
h->fd_array[r].revents));
r=h->crt_fd_array_idx; /* can change due to io_watch_del(fd)
array shifting */
}
}
error:
return ret;
}
#ifdef HAVE_SELECT
/* wait for io using select */
inline static int io_wait_loop_select(io_wait_h* h, int t, int repeat)
{
fd_set sel_rset;
fd_set sel_wset;
int n, ret;
struct timeval timeout;
int r;
struct fd_map* fm;
int revents;
again:
sel_rset=h->master_rset;
sel_wset=h->master_wset;
timeout.tv_sec=t;
timeout.tv_usec=0;
ret=n=select(h->max_fd_select+1, &sel_rset, &sel_wset, 0, &timeout);
if (n<0){
if (errno==EINTR) goto again; /* just a signal */
LOG(L_ERR, "ERROR: io_wait_loop_select: select: %s [%d]\n",
strerror(errno), errno);
n=0;
/* continue */
}
/* use poll fd array */
for(r=0; (r<h->fd_no) && n; r++){
revents=0;
if (likely(FD_ISSET(h->fd_array[r].fd, &sel_rset)))
revents|=POLLIN;
if (unlikely(FD_ISSET(h->fd_array[r].fd, &sel_wset)))