ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/IO-AIO/AIO.xs
Revision: 1.33
Committed: Mon Aug 22 23:21:57 2005 UTC (18 years, 8 months ago) by root
Branch: MAIN
Changes since 1.32: +1 -0 lines
Log Message:
*** empty log message ***

File Contents

# Content
1 #define _REENTRANT 1
2 #include <errno.h>
3
4 #include "EXTERN.h"
5 #include "perl.h"
6 #include "XSUB.h"
7
8 #include "autoconf/config.h"
9
10 #include <pthread.h>
11
12 #include <sys/types.h>
13 #include <sys/stat.h>
14
15 #include <unistd.h>
16 #include <fcntl.h>
17 #include <signal.h>
18 #include <sched.h>
19
20 #if HAVE_SENDFILE
21 # if __linux
22 # include <sys/sendfile.h>
23 # elif __freebsd
24 # include <sys/socket.h>
25 # include <sys/uio.h>
26 # elif __hpux
27 # include <sys/socket.h>
28 # endif
29 #endif
30
31 #if __ia64
32 # define STACKSIZE 65536
33 #else
34 # define STACKSIZE 4096
35 #endif
36
37 enum {
38 REQ_QUIT,
39 REQ_OPEN, REQ_CLOSE,
40 REQ_READ, REQ_WRITE, REQ_READAHEAD,
41 REQ_SENDFILE,
42 REQ_STAT, REQ_LSTAT, REQ_FSTAT,
43 REQ_FSYNC, REQ_FDATASYNC,
44 REQ_UNLINK, REQ_RMDIR,
45 REQ_SYMLINK,
46 };
47
48 typedef struct aio_cb {
49 struct aio_cb *volatile next;
50
51 int type;
52
53 int fd, fd2;
54 off_t offset;
55 size_t length;
56 ssize_t result;
57 mode_t mode; /* open */
58 int errorno;
59 SV *data, *callback;
60 SV *fh, *fh2;
61 void *dataptr, *data2ptr;
62 STRLEN dataoffset;
63
64 Stat_t *statdata;
65 } aio_cb;
66
67 typedef aio_cb *aio_req;
68
69 static int started, wanted;
70 static volatile int nreqs;
71 static int max_outstanding = 1<<30;
72 static int respipe [2];
73
74 static pthread_mutex_t reslock = PTHREAD_MUTEX_INITIALIZER;
75 static pthread_mutex_t reqlock = PTHREAD_MUTEX_INITIALIZER;
76 static pthread_cond_t reqwait = PTHREAD_COND_INITIALIZER;
77
78 static volatile aio_req reqs, reqe; /* queue start, queue end */
79 static volatile aio_req ress, rese; /* queue start, queue end */
80
81 static void free_req (aio_req req)
82 {
83 if (req->data)
84 SvREFCNT_dec (req->data);
85
86 if (req->fh)
87 SvREFCNT_dec (req->fh);
88
89 if (req->fh2)
90 SvREFCNT_dec (req->fh2);
91
92 if (req->statdata)
93 Safefree (req->statdata);
94
95 if (req->callback)
96 SvREFCNT_dec (req->callback);
97
98 Safefree (req);
99 }
100
101 static void
102 poll_wait ()
103 {
104 if (nreqs && !ress)
105 {
106 fd_set rfd;
107 FD_ZERO(&rfd);
108 FD_SET(respipe [0], &rfd);
109
110 select (respipe [0] + 1, &rfd, 0, 0, 0);
111 }
112 }
113
114 static int
115 poll_cb ()
116 {
117 dSP;
118 int count = 0;
119 int do_croak = 0;
120 aio_req req;
121
122 for (;;)
123 {
124 pthread_mutex_lock (&reslock);
125 req = ress;
126
127 if (req)
128 {
129 ress = req->next;
130
131 if (!ress)
132 {
133 /* read any signals sent by the worker threads */
134 char buf [32];
135 while (read (respipe [0], buf, 32) == 32)
136 ;
137
138 rese = 0;
139 }
140 }
141
142 pthread_mutex_unlock (&reslock);
143
144 if (!req)
145 break;
146
147 nreqs--;
148
149 if (req->type == REQ_QUIT)
150 started--;
151 else
152 {
153 int errorno = errno;
154 errno = req->errorno;
155
156 if (req->type == REQ_READ)
157 SvCUR_set (req->data, req->dataoffset + (req->result > 0 ? req->result : 0));
158
159 if (req->data2ptr && (req->type == REQ_READ || req->type == REQ_WRITE))
160 SvREADONLY_off (req->data);
161
162 if (req->statdata)
163 {
164 PL_laststype = req->type == REQ_LSTAT ? OP_LSTAT : OP_STAT;
165 PL_laststatval = req->result;
166 PL_statcache = *(req->statdata);
167 }
168
169 ENTER;
170 PUSHMARK (SP);
171 XPUSHs (sv_2mortal (newSViv (req->result)));
172
173 if (req->type == REQ_OPEN)
174 {
175 /* convert fd to fh */
176 SV *fh;
177
178 PUTBACK;
179 call_pv ("IO::AIO::_fd2fh", G_SCALAR | G_EVAL);
180 SPAGAIN;
181
182 fh = SvREFCNT_inc (POPs);
183
184 PUSHMARK (SP);
185 XPUSHs (sv_2mortal (fh));
186 }
187
188 if (SvOK (req->callback))
189 {
190 PUTBACK;
191 call_sv (req->callback, G_VOID | G_EVAL);
192 SPAGAIN;
193
194 if (SvTRUE (ERRSV))
195 {
196 free_req (req);
197 croak (0);
198 }
199 }
200
201 LEAVE;
202
203 errno = errorno;
204 count++;
205 }
206
207 free_req (req);
208 }
209
210 return count;
211 }
212
213 static void *aio_proc(void *arg);
214
215 static void
216 start_thread (void)
217 {
218 sigset_t fullsigset, oldsigset;
219 pthread_t tid;
220 pthread_attr_t attr;
221
222 pthread_attr_init (&attr);
223 pthread_attr_setstacksize (&attr, STACKSIZE);
224 pthread_attr_setdetachstate (&attr, PTHREAD_CREATE_DETACHED);
225
226 sigfillset (&fullsigset);
227 sigprocmask (SIG_SETMASK, &fullsigset, &oldsigset);
228
229 if (pthread_create (&tid, &attr, aio_proc, 0) == 0)
230 started++;
231
232 sigprocmask (SIG_SETMASK, &oldsigset, 0);
233 }
234
235 static void
236 send_req (aio_req req)
237 {
238 while (started < wanted && nreqs >= started)
239 start_thread ();
240
241 nreqs++;
242
243 pthread_mutex_lock (&reqlock);
244
245 req->next = 0;
246
247 if (reqe)
248 {
249 reqe->next = req;
250 reqe = req;
251 }
252 else
253 reqe = reqs = req;
254
255 pthread_cond_signal (&reqwait);
256 pthread_mutex_unlock (&reqlock);
257
258 if (nreqs > max_outstanding)
259 for (;;)
260 {
261 poll_cb ();
262
263 if (nreqs <= max_outstanding)
264 break;
265
266 poll_wait ();
267 }
268 }
269
270 static void
271 end_thread (void)
272 {
273 aio_req req;
274 Newz (0, req, 1, aio_cb);
275 req->type = REQ_QUIT;
276
277 send_req (req);
278 }
279
280 static void min_parallel (int nthreads)
281 {
282 if (wanted < nthreads)
283 wanted = nthreads;
284 }
285
286 static void max_parallel (int nthreads)
287 {
288 int cur = started;
289
290 if (wanted > nthreads)
291 wanted = nthreads;
292
293 while (cur > wanted)
294 {
295 end_thread ();
296 cur--;
297 }
298
299 while (started > wanted)
300 {
301 poll_wait ();
302 poll_cb ();
303 }
304 }
305
306 static void create_pipe ()
307 {
308 if (pipe (respipe))
309 croak ("unable to initialize result pipe");
310
311 if (fcntl (respipe [0], F_SETFL, O_NONBLOCK))
312 croak ("cannot set result pipe to nonblocking mode");
313
314 if (fcntl (respipe [1], F_SETFL, O_NONBLOCK))
315 croak ("cannot set result pipe to nonblocking mode");
316 }
317
318 static void atfork_prepare (void)
319 {
320 pthread_mutex_lock (&reqlock);
321 pthread_mutex_lock (&reslock);
322 }
323
324 static void atfork_parent (void)
325 {
326 pthread_mutex_unlock (&reslock);
327 pthread_mutex_unlock (&reqlock);
328 }
329
330 static void atfork_child (void)
331 {
332 aio_req prv;
333
334 started = 0;
335
336 while (reqs)
337 {
338 prv = reqs;
339 reqs = prv->next;
340 free_req (prv);
341 }
342
343 reqs = reqe = 0;
344
345 while (ress)
346 {
347 prv = ress;
348 ress = prv->next;
349 free_req (prv);
350 }
351
352 ress = rese = 0;
353
354 close (respipe [0]);
355 close (respipe [1]);
356 create_pipe ();
357
358 atfork_parent ();
359 }
360
361 /*****************************************************************************/
362 /* work around various missing functions */
363
364 #if !HAVE_PREADWRITE
365 # define pread aio_pread
366 # define pwrite aio_pwrite
367
368 /*
369 * make our pread/pwrite safe against themselves, but not against
370 * normal read/write by using a mutex. slows down execution a lot,
371 * but that's your problem, not mine.
372 */
373 static pthread_mutex_t iolock = PTHREAD_MUTEX_INITIALIZER;
374
375 static ssize_t
376 pread (int fd, void *buf, size_t count, off_t offset)
377 {
378 ssize_t res;
379 off_t ooffset;
380
381 pthread_mutex_lock (&iolock);
382 ooffset = lseek (fd, 0, SEEK_CUR);
383 lseek (fd, offset, SEEK_SET);
384 res = read (fd, buf, count);
385 lseek (fd, ooffset, SEEK_SET);
386 pthread_mutex_unlock (&iolock);
387
388 return res;
389 }
390
391 static ssize_t
392 pwrite (int fd, void *buf, size_t count, off_t offset)
393 {
394 ssize_t res;
395 off_t ooffset;
396
397 pthread_mutex_lock (&iolock);
398 ooffset = lseek (fd, 0, SEEK_CUR);
399 lseek (fd, offset, SEEK_SET);
400 res = write (fd, buf, count);
401 lseek (fd, offset, SEEK_SET);
402 pthread_mutex_unlock (&iolock);
403
404 return res;
405 }
406 #endif
407
408 #if !HAVE_FDATASYNC
409 # define fdatasync fsync
410 #endif
411
412 #if !HAVE_READAHEAD
413 # define readahead aio_readahead
414
415 static char readahead_buf[4096];
416
417 static ssize_t
418 readahead (int fd, off_t offset, size_t count)
419 {
420 while (count > 0)
421 {
422 size_t len = count < sizeof (readahead_buf) ? count : sizeof (readahead_buf);
423
424 pread (fd, readahead_buf, len, offset);
425 offset += len;
426 count -= len;
427 }
428
429 errno = 0;
430 }
431 #endif
432
433 /* sendfile always needs emulation */
434 static ssize_t
435 sendfile_ (int ofd, int ifd, off_t offset, size_t count)
436 {
437 ssize_t res;
438
439 if (!count)
440 return 0;
441
442 #if __linux
443 res = sendfile (ofd, ifd, &offset, count);
444
445 #elif __freebsd
446 /*
447 * Of course, the freebsd sendfile is a dire hack with no thoughts
448 * wasted on making it similar to other i/o functions.
449 */
450 {
451 off_t sbytes;
452 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
453
454 if (!res && errno == EAGAIN)
455 /* maybe on others, too, as usual, the manpage leaves you guessing */
456 res = sbytes;
457 }
458
459 #elif __hpux
460 res = sendfile (ofd, ifd, offset, count, 0, 0);
461
462 #else
463 res = -1;
464 errno = ENOSYS;
465 #endif
466
467 if (res < 0 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK))
468 {
469 /* emulate sendfile. this is a major pain in the ass */
470 char *buf = malloc (4096);
471 res = 0;
472
473 for (;;)
474 {
475 ssize_t cnt;
476
477 cnt = pread (ifd, buf, 4096, offset);
478
479 if (cnt <= 0)
480 {
481 if (cnt && !res) res = -1;
482 break;
483 }
484
485 cnt = write (ofd, buf, cnt);
486
487 if (cnt <= 0)
488 {
489 if (cnt && !res) res = -1;
490 break;
491 }
492
493 offset += cnt;
494 res += cnt;
495 }
496
497 {
498 int errorno = errno;
499 free (buf);
500 errno = errorno;
501 }
502 }
503
504 return res;
505 }
506
507 /*****************************************************************************/
508
509 static void *
510 aio_proc (void *thr_arg)
511 {
512 aio_req req;
513 int type;
514
515 do
516 {
517 pthread_mutex_lock (&reqlock);
518
519 for (;;)
520 {
521 req = reqs;
522
523 if (reqs)
524 {
525 reqs = reqs->next;
526 if (!reqs) reqe = 0;
527 }
528
529 if (req)
530 break;
531
532 pthread_cond_wait (&reqwait, &reqlock);
533 }
534
535 pthread_mutex_unlock (&reqlock);
536
537 errno = 0; /* strictly unnecessary */
538
539 type = req->type;
540
541 switch (type)
542 {
543 case REQ_READ: req->result = pread (req->fd, req->dataptr, req->length, req->offset); break;
544 case REQ_WRITE: req->result = pwrite (req->fd, req->dataptr, req->length, req->offset); break;
545
546 case REQ_READAHEAD: req->result = readahead (req->fd, req->offset, req->length); break;
547 case REQ_SENDFILE: req->result = sendfile_ (req->fd, req->fd2, req->offset, req->length); break;
548
549 case REQ_STAT: req->result = stat (req->dataptr, req->statdata); break;
550 case REQ_LSTAT: req->result = lstat (req->dataptr, req->statdata); break;
551 case REQ_FSTAT: req->result = fstat (req->fd , req->statdata); break;
552
553 case REQ_OPEN: req->result = open (req->dataptr, req->fd, req->mode); break;
554 case REQ_CLOSE: req->result = close (req->fd); break;
555 case REQ_UNLINK: req->result = unlink (req->dataptr); break;
556 case REQ_RMDIR: req->result = rmdir (req->dataptr); break;
557 case REQ_SYMLINK: req->result = symlink (req->data2ptr, req->dataptr); break;
558
559 case REQ_FDATASYNC: req->result = fdatasync (req->fd); break;
560 case REQ_FSYNC: req->result = fsync (req->fd); break;
561
562 case REQ_QUIT:
563 break;
564
565 default:
566 req->result = ENOSYS;
567 break;
568 }
569
570 req->errorno = errno;
571
572 pthread_mutex_lock (&reslock);
573
574 req->next = 0;
575
576 if (rese)
577 {
578 rese->next = req;
579 rese = req;
580 }
581 else
582 {
583 rese = ress = req;
584
585 /* write a dummy byte to the pipe so fh becomes ready */
586 write (respipe [1], &respipe, 1);
587 }
588
589 pthread_mutex_unlock (&reslock);
590 }
591 while (type != REQ_QUIT);
592
593 return 0;
594 }
595
596 #define dREQ \
597 aio_req req; \
598 \
599 if (SvOK (callback) && !SvROK (callback)) \
600 croak ("clalback must be undef or of reference type"); \
601 \
602 Newz (0, req, 1, aio_cb); \
603 if (!req) \
604 croak ("out of memory during aio_req allocation"); \
605 \
606 req->callback = newSVsv (callback);
607
608 MODULE = IO::AIO PACKAGE = IO::AIO
609
610 PROTOTYPES: ENABLE
611
612 BOOT:
613 {
614 create_pipe ();
615 pthread_atfork (atfork_prepare, atfork_parent, atfork_child);
616 }
617
618 void
619 min_parallel(nthreads)
620 int nthreads
621 PROTOTYPE: $
622
623 void
624 max_parallel(nthreads)
625 int nthreads
626 PROTOTYPE: $
627
628 int
629 max_outstanding(nreqs)
630 int nreqs
631 PROTOTYPE: $
632 CODE:
633 RETVAL = max_outstanding;
634 max_outstanding = nreqs;
635
636 void
637 aio_open(pathname,flags,mode,callback=&PL_sv_undef)
638 SV * pathname
639 int flags
640 int mode
641 SV * callback
642 PROTOTYPE: $$$;$
643 CODE:
644 {
645 dREQ;
646
647 req->type = REQ_OPEN;
648 req->data = newSVsv (pathname);
649 req->dataptr = SvPVbyte_nolen (req->data);
650 req->fd = flags;
651 req->mode = mode;
652
653 send_req (req);
654 }
655
656 void
657 aio_close(fh,callback=&PL_sv_undef)
658 SV * fh
659 SV * callback
660 PROTOTYPE: $;$
661 ALIAS:
662 aio_close = REQ_CLOSE
663 aio_fsync = REQ_FSYNC
664 aio_fdatasync = REQ_FDATASYNC
665 CODE:
666 {
667 dREQ;
668
669 req->type = ix;
670 req->fh = newSVsv (fh);
671 req->fd = PerlIO_fileno (IoIFP (sv_2io (fh)));
672
673 send_req (req);
674 }
675
676 void
677 aio_read(fh,offset,length,data,dataoffset,callback=&PL_sv_undef)
678 SV * fh
679 UV offset
680 UV length
681 SV * data
682 UV dataoffset
683 SV * callback
684 ALIAS:
685 aio_read = REQ_READ
686 aio_write = REQ_WRITE
687 PROTOTYPE: $$$$$;$
688 CODE:
689 {
690 aio_req req;
691 STRLEN svlen;
692 char *svptr = SvPVbyte (data, svlen);
693
694 SvUPGRADE (data, SVt_PV);
695 SvPOK_on (data);
696
697 if (dataoffset < 0)
698 dataoffset += svlen;
699
700 if (dataoffset < 0 || dataoffset > svlen)
701 croak ("data offset outside of string");
702
703 if (ix == REQ_WRITE)
704 {
705 /* write: check length and adjust. */
706 if (length < 0 || length + dataoffset > svlen)
707 length = svlen - dataoffset;
708 }
709 else
710 {
711 /* read: grow scalar as necessary */
712 svptr = SvGROW (data, length + dataoffset);
713 }
714
715 if (length < 0)
716 croak ("length must not be negative");
717
718 {
719 dREQ;
720
721 req->type = ix;
722 req->fh = newSVsv (fh);
723 req->fd = PerlIO_fileno (ix == REQ_READ ? IoIFP (sv_2io (fh))
724 : IoOFP (sv_2io (fh)));
725 req->offset = offset;
726 req->length = length;
727 req->data = SvREFCNT_inc (data);
728 req->dataptr = (char *)svptr + dataoffset;
729
730 if (!SvREADONLY (data))
731 {
732 SvREADONLY_on (data);
733 req->data2ptr = (void *)data;
734 }
735
736 send_req (req);
737 }
738 }
739
740 void
741 aio_sendfile(out_fh,in_fh,in_offset,length,callback=&PL_sv_undef)
742 SV * out_fh
743 SV * in_fh
744 UV in_offset
745 UV length
746 SV * callback
747 PROTOTYPE: $$$$;$
748 CODE:
749 {
750 dREQ;
751
752 req->type = REQ_SENDFILE;
753 req->fh = newSVsv (out_fh);
754 req->fd = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
755 req->fh2 = newSVsv (in_fh);
756 req->fd2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
757 req->offset = in_offset;
758 req->length = length;
759
760 send_req (req);
761 }
762
763 void
764 aio_readahead(fh,offset,length,callback=&PL_sv_undef)
765 SV * fh
766 UV offset
767 IV length
768 SV * callback
769 PROTOTYPE: $$$;$
770 CODE:
771 {
772 dREQ;
773
774 req->type = REQ_READAHEAD;
775 req->fh = newSVsv (fh);
776 req->fd = PerlIO_fileno (IoIFP (sv_2io (fh)));
777 req->offset = offset;
778 req->length = length;
779
780 send_req (req);
781 }
782
783 void
784 aio_stat(fh_or_path,callback=&PL_sv_undef)
785 SV * fh_or_path
786 SV * callback
787 ALIAS:
788 aio_stat = REQ_STAT
789 aio_lstat = REQ_LSTAT
790 CODE:
791 {
792 dREQ;
793
794 New (0, req->statdata, 1, Stat_t);
795 if (!req->statdata)
796 {
797 free_req (req);
798 croak ("out of memory during aio_req->statdata allocation");
799 }
800
801 if (SvPOK (fh_or_path))
802 {
803 req->type = ix;
804 req->data = newSVsv (fh_or_path);
805 req->dataptr = SvPVbyte_nolen (req->data);
806 }
807 else
808 {
809 req->type = REQ_FSTAT;
810 req->fh = newSVsv (fh_or_path);
811 req->fd = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
812 }
813
814 send_req (req);
815 }
816
817 void
818 aio_unlink(pathname,callback=&PL_sv_undef)
819 SV * pathname
820 SV * callback
821 ALIAS:
822 aio_unlink = REQ_UNLINK
823 aio_rmdir = REQ_RMDIR
824 CODE:
825 {
826 dREQ;
827
828 req->type = ix;
829 req->data = newSVsv (pathname);
830 req->dataptr = SvPVbyte_nolen (req->data);
831
832 send_req (req);
833 }
834
835 void
836 aio_symlink(oldpath,newpath,callback=&PL_sv_undef)
837 SV * oldpath
838 SV * newpath
839 SV * callback
840 CODE:
841 {
842 dREQ;
843
844 req->type = REQ_SYMLINK;
845 req->fh = newSVsv (oldpath);
846 req->data2ptr = SvPVbyte_nolen (req->fh);
847 req->data = newSVsv (newpath);
848 req->dataptr = SvPVbyte_nolen (req->data);
849
850 send_req (req);
851 }
852
853 void
854 flush()
855 PROTOTYPE:
856 CODE:
857 while (nreqs)
858 {
859 poll_wait ();
860 poll_cb ();
861 }
862
863 void
864 poll()
865 PROTOTYPE:
866 CODE:
867 if (nreqs)
868 {
869 poll_wait ();
870 poll_cb ();
871 }
872
873 int
874 poll_fileno()
875 PROTOTYPE:
876 CODE:
877 RETVAL = respipe [0];
878 OUTPUT:
879 RETVAL
880
881 int
882 poll_cb(...)
883 PROTOTYPE:
884 CODE:
885 RETVAL = poll_cb ();
886 OUTPUT:
887 RETVAL
888
889 void
890 poll_wait()
891 PROTOTYPE:
892 CODE:
893 if (nreqs)
894 poll_wait ();
895
896 int
897 nreqs()
898 PROTOTYPE:
899 CODE:
900 RETVAL = nreqs;
901 OUTPUT:
902 RETVAL
903