ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/eio.c
(Generate patch)

Comparing libeio/eio.c (file contents):
Revision 1.16 by root, Fri May 30 04:50:21 2008 UTC vs.
Revision 1.69 by root, Fri Jun 10 06:50:42 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45
46#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE
48#endif
41#include "xthread.h" 49#include "xthread.h"
42 50
43#include <errno.h> 51#include <errno.h>
44#include <stddef.h> 52#include <stddef.h>
45#include <stdlib.h> 53#include <stdlib.h>
46#include <string.h> 54#include <string.h>
47#include <errno.h> 55#include <errno.h>
48#include <sys/types.h> 56#include <sys/types.h>
49#include <sys/stat.h> 57#include <sys/stat.h>
58#include <sys/statvfs.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
54#ifndef EIO_FINISH 69#ifndef EIO_FINISH
55# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 70# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
56#endif 71#endif
57 72
58#ifndef EIO_DESTROY 73#ifndef EIO_DESTROY
… …
64#endif 79#endif
65 80
66#ifdef _WIN32 81#ifdef _WIN32
67 82
68 /*doh*/ 83 /*doh*/
69
70#else 84#else
71 85
72# include "config.h"
73# include <sys/time.h> 86# include <sys/time.h>
74# include <sys/select.h> 87# include <sys/select.h>
75# include <unistd.h> 88# include <unistd.h>
76# include <utime.h> 89# include <utime.h>
77# include <signal.h> 90# include <signal.h>
78# include <dirent.h> 91# include <dirent.h>
79 92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
100# define D_INO(de) (de)->d_fileno
101# define D_NAMLEN(de) (de)->d_namlen
102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103# define D_INO(de) (de)->d_ino
104# endif
105
106#ifdef _D_EXACT_NAMLEN
107# undef D_NAMLEN
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
110
111# ifdef _DIRENT_HAVE_D_TYPE
112# define D_TYPE(de) (de)->d_type
113# endif
114
80# ifndef EIO_STRUCT_DIRENT 115# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 116# define EIO_STRUCT_DIRENT struct dirent
82# endif 117# endif
83 118
84#endif 119#endif
85 120
86#if HAVE_SENDFILE 121#if HAVE_SENDFILE
87# if __linux 122# if __linux
88# include <sys/sendfile.h> 123# include <sys/sendfile.h>
89# elif __freebsd 124# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 125# include <sys/socket.h>
91# include <sys/uio.h> 126# include <sys/uio.h>
92# elif __hpux 127# elif __hpux
93# include <sys/socket.h> 128# include <sys/socket.h>
94# elif __solaris /* not yet */ 129# elif __solaris
95# include <sys/sendfile.h> 130# include <sys/sendfile.h>
96# else 131# else
97# error sendfile support requested but not available 132# error sendfile support requested but not available
98# endif 133# endif
99#endif 134#endif
100 135
101/* number of seconds after which an idle threads exit */ 136#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 137# define D_TYPE(de) 0
138#endif
139#ifndef D_INO
140# define D_INO(de) 0
141#endif
142#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name)
144#endif
103 145
104/* used for struct dirent, AIX doesn't provide it */ 146/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 147#ifndef NAME_MAX
106# define NAME_MAX 4096 148# define NAME_MAX 4096
149#endif
150
151/* used for readlink etc. */
152#ifndef PATH_MAX
153# define PATH_MAX 4096
107#endif 154#endif
108 155
109/* buffer size for various temporary buffers */ 156/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 157#define EIO_BUFSIZE 65536
111 158
… …
117 errno = ENOMEM; \ 164 errno = ENOMEM; \
118 if (!eio_buf) \ 165 if (!eio_buf) \
119 return -1; 166 return -1;
120 167
121#define EIO_TICKS ((1000000 + 1023) >> 10) 168#define EIO_TICKS ((1000000 + 1023) >> 10)
169
170/*****************************************************************************/
171
172#if __GNUC__ >= 3
173# define expect(expr,value) __builtin_expect ((expr),(value))
174#else
175# define expect(expr,value) (expr)
176#endif
177
178#define expect_false(expr) expect ((expr) != 0, 0)
179#define expect_true(expr) expect ((expr) != 0, 1)
122 180
123/*****************************************************************************/ 181/*****************************************************************************/
124 182
125#define ETP_PRI_MIN EIO_PRI_MIN 183#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 184#define ETP_PRI_MAX EIO_PRI_MAX
… …
144 if (wrk->dirp) \ 202 if (wrk->dirp) \
145 { \ 203 { \
146 closedir (wrk->dirp); \ 204 closedir (wrk->dirp); \
147 wrk->dirp = 0; \ 205 wrk->dirp = 0; \
148 } 206 }
207
149#define ETP_WORKER_COMMON \ 208#define ETP_WORKER_COMMON \
150 void *dbuf; \ 209 void *dbuf; \
151 DIR *dirp; 210 DIR *dirp;
152 211
153/*****************************************************************************/ 212/*****************************************************************************/
154 213
155#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156 215
157/* calculcate time difference in ~1/EIO_TICKS of a second */ 216/* calculate time difference in ~1/EIO_TICKS of a second */
158static int tvdiff (struct timeval *tv1, struct timeval *tv2) 217static int tvdiff (struct timeval *tv1, struct timeval *tv2)
159{ 218{
160 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 220 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
162} 221}
… …
170static unsigned int max_poll_reqs; /* reslock */ 229static unsigned int max_poll_reqs; /* reslock */
171 230
172static volatile unsigned int nreqs; /* reqlock */ 231static volatile unsigned int nreqs; /* reqlock */
173static volatile unsigned int nready; /* reqlock */ 232static volatile unsigned int nready; /* reqlock */
174static volatile unsigned int npending; /* reqlock */ 233static volatile unsigned int npending; /* reqlock */
175static volatile unsigned int max_idle = 4; 234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
176 236
237static xmutex_t wrklock;
238static xmutex_t reslock;
239static xmutex_t reqlock;
240static xcond_t reqwait;
241
242#if !HAVE_PREADWRITE
243/*
244 * make our pread/pwrite emulation safe against themselves, but not against
245 * normal read/write by using a mutex. slows down execution a lot,
246 * but that's your problem, not mine.
247 */
177static mutex_t wrklock = X_MUTEX_INIT; 248static xmutex_t preadwritelock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT; 249#endif
179static mutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT;
181 250
182typedef struct etp_worker 251typedef struct etp_worker
183{ 252{
184 /* locked by wrklock */ 253 /* locked by wrklock */
185 struct etp_worker *prev, *next; 254 struct etp_worker *prev, *next;
186 255
187 thread_t tid; 256 xthread_t tid;
188 257
189 /* locked by reslock, reqlock or wrklock */ 258 /* locked by reslock, reqlock or wrklock */
190 ETP_REQ *req; /* currently processed request */ 259 ETP_REQ *req; /* currently processed request */
191 260
192 ETP_WORKER_COMMON 261 ETP_WORKER_COMMON
… …
306 } 375 }
307 376
308 abort (); 377 abort ();
309} 378}
310 379
380static void etp_thread_init (void)
381{
382 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait);
386}
387
311static void etp_atfork_prepare (void) 388static void etp_atfork_prepare (void)
312{ 389{
313 X_LOCK (wrklock); 390 X_LOCK (wrklock);
314 X_LOCK (reqlock); 391 X_LOCK (reqlock);
315 X_LOCK (reslock); 392 X_LOCK (reslock);
… …
330 407
331static void etp_atfork_child (void) 408static void etp_atfork_child (void)
332{ 409{
333 ETP_REQ *prv; 410 ETP_REQ *prv;
334 411
335 while (prv = reqq_shift (&req_queue)) 412 while ((prv = reqq_shift (&req_queue)))
336 ETP_DESTROY (prv); 413 ETP_DESTROY (prv);
337 414
338 while (prv = reqq_shift (&res_queue)) 415 while ((prv = reqq_shift (&res_queue)))
339 ETP_DESTROY (prv); 416 ETP_DESTROY (prv);
340 417
341 while (wrk_first.next != &wrk_first) 418 while (wrk_first.next != &wrk_first)
342 { 419 {
343 etp_worker *wrk = wrk_first.next; 420 etp_worker *wrk = wrk_first.next;
… …
353 idle = 0; 430 idle = 0;
354 nreqs = 0; 431 nreqs = 0;
355 nready = 0; 432 nready = 0;
356 npending = 0; 433 npending = 0;
357 434
358 etp_atfork_parent (); 435 etp_thread_init ();
359} 436}
360 437
361static void 438static void
362etp_once_init (void) 439etp_once_init (void)
363{ 440{
441 etp_thread_init ();
364 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
365} 443}
366 444
367static int 445static int
368etp_init (void (*want_poll)(void), void (*done_poll)(void)) 446etp_init (void (*want_poll)(void), void (*done_poll)(void))
… …
371 449
372 pthread_once (&doinit, etp_once_init); 450 pthread_once (&doinit, etp_once_init);
373 451
374 want_poll_cb = want_poll; 452 want_poll_cb = want_poll;
375 done_poll_cb = done_poll; 453 done_poll_cb = done_poll;
454
455 return 0;
376} 456}
377 457
378X_THREAD_PROC (etp_proc); 458X_THREAD_PROC (etp_proc);
379 459
380static void etp_start_thread (void) 460static void etp_start_thread (void)
… …
400 X_UNLOCK (wrklock); 480 X_UNLOCK (wrklock);
401} 481}
402 482
403static void etp_maybe_start_thread (void) 483static void etp_maybe_start_thread (void)
404{ 484{
405 if (etp_nthreads () >= wanted) 485 if (expect_true (etp_nthreads () >= wanted))
406 return; 486 return;
407 487
408 /* todo: maybe use idle here, but might be less exact */ 488 /* todo: maybe use idle here, but might be less exact */
409 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 489 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
410 return; 490 return;
411 491
412 etp_start_thread (); 492 etp_start_thread ();
413} 493}
414 494
… …
467 547
468 X_LOCK (reqlock); 548 X_LOCK (reqlock);
469 --nreqs; 549 --nreqs;
470 X_UNLOCK (reqlock); 550 X_UNLOCK (reqlock);
471 551
472 if (req->type == EIO_GROUP && req->size) 552 if (expect_false (req->type == EIO_GROUP && req->size))
473 { 553 {
474 req->int1 = 1; /* mark request as delayed */ 554 req->int1 = 1; /* mark request as delayed */
475 continue; 555 continue;
476 } 556 }
477 else 557 else
478 { 558 {
479 int res = ETP_FINISH (req); 559 int res = ETP_FINISH (req);
480 if (res) 560 if (expect_false (res))
481 return res; 561 return res;
482 } 562 }
483 563
484 if (maxreqs && !--maxreqs) 564 if (expect_false (maxreqs && !--maxreqs))
485 break; 565 break;
486 566
487 if (maxtime) 567 if (maxtime)
488 { 568 {
489 gettimeofday (&tv_now, 0); 569 gettimeofday (&tv_now, 0);
… …
508 588
509static void etp_submit (ETP_REQ *req) 589static void etp_submit (ETP_REQ *req)
510{ 590{
511 req->pri -= ETP_PRI_MIN; 591 req->pri -= ETP_PRI_MIN;
512 592
513 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 593 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
514 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 594 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
515 595
596 if (expect_false (req->type == EIO_GROUP))
597 {
598 /* I hope this is worth it :/ */
516 X_LOCK (reqlock); 599 X_LOCK (reqlock);
517 ++nreqs; 600 ++nreqs;
601 X_UNLOCK (reqlock);
602
603 X_LOCK (reslock);
604
605 ++npending;
606
607 if (!reqq_push (&res_queue, req) && want_poll_cb)
608 want_poll_cb ();
609
610 X_UNLOCK (reslock);
611 }
612 else
613 {
614 X_LOCK (reqlock);
615 ++nreqs;
518 ++nready; 616 ++nready;
519 reqq_push (&req_queue, req); 617 reqq_push (&req_queue, req);
520 X_COND_SIGNAL (reqwait); 618 X_COND_SIGNAL (reqwait);
521 X_UNLOCK (reqlock); 619 X_UNLOCK (reqlock);
522 620
523 etp_maybe_start_thread (); 621 etp_maybe_start_thread ();
622 }
524} 623}
525 624
526static void etp_set_max_poll_time (double nseconds) 625static void etp_set_max_poll_time (double nseconds)
527{ 626{
528 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 627 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
529 max_poll_time = nseconds; 628 max_poll_time = nseconds * EIO_TICKS;
530 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
531} 630}
532 631
533static void etp_set_max_poll_reqs (unsigned int maxreqs) 632static void etp_set_max_poll_reqs (unsigned int maxreqs)
534{ 633{
… …
538} 637}
539 638
540static void etp_set_max_idle (unsigned int nthreads) 639static void etp_set_max_idle (unsigned int nthreads)
541{ 640{
542 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
543 max_idle = nthreads <= 0 ? 1 : nthreads; 642 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644}
645
646static void etp_set_idle_timeout (unsigned int seconds)
647{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds;
544 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
545} 651}
546 652
547static void etp_set_min_parallel (unsigned int nthreads) 653static void etp_set_min_parallel (unsigned int nthreads)
548{ 654{
… …
563 669
564static void grp_try_feed (eio_req *grp) 670static void grp_try_feed (eio_req *grp)
565{ 671{
566 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
567 { 673 {
568 int old_len = grp->size; 674 grp->flags &= ~EIO_FLAG_GROUPADD;
569 675
570 EIO_FEED (grp); 676 EIO_FEED (grp);
571 677
572 /* stop if no progress has been made */ 678 /* stop if no progress has been made */
573 if (old_len == grp->size) 679 if (!(grp->flags & EIO_FLAG_GROUPADD))
574 { 680 {
575 grp->feed = 0; 681 grp->feed = 0;
576 break; 682 break;
577 } 683 }
578 } 684 }
… …
676void eio_set_max_idle (unsigned int nthreads) 782void eio_set_max_idle (unsigned int nthreads)
677{ 783{
678 etp_set_max_idle (nthreads); 784 etp_set_max_idle (nthreads);
679} 785}
680 786
787void eio_set_idle_timeout (unsigned int seconds)
788{
789 etp_set_idle_timeout (seconds);
790}
791
681void eio_set_min_parallel (unsigned int nthreads) 792void eio_set_min_parallel (unsigned int nthreads)
682{ 793{
683 etp_set_min_parallel (nthreads); 794 etp_set_min_parallel (nthreads);
684} 795}
685 796
… …
695 806
696/*****************************************************************************/ 807/*****************************************************************************/
697/* work around various missing functions */ 808/* work around various missing functions */
698 809
699#if !HAVE_PREADWRITE 810#if !HAVE_PREADWRITE
811# undef pread
812# undef pwrite
700# define pread eio__pread 813# define pread eio__pread
701# define pwrite eio__pwrite 814# define pwrite eio__pwrite
702
703/*
704 * make our pread/pwrite safe against themselves, but not against
705 * normal read/write by using a mutex. slows down execution a lot,
706 * but that's your problem, not mine.
707 */
708static mutex_t preadwritelock = X_MUTEX_INIT;
709 815
710static ssize_t 816static ssize_t
711eio__pread (int fd, void *buf, size_t count, off_t offset) 817eio__pread (int fd, void *buf, size_t count, off_t offset)
712{ 818{
713 ssize_t res; 819 ssize_t res;
… …
731 837
732 X_LOCK (preadwritelock); 838 X_LOCK (preadwritelock);
733 ooffset = lseek (fd, 0, SEEK_CUR); 839 ooffset = lseek (fd, 0, SEEK_CUR);
734 lseek (fd, offset, SEEK_SET); 840 lseek (fd, offset, SEEK_SET);
735 res = write (fd, buf, count); 841 res = write (fd, buf, count);
736 lseek (fd, offset, SEEK_SET); 842 lseek (fd, ooffset, SEEK_SET);
737 X_UNLOCK (preadwritelock); 843 X_UNLOCK (preadwritelock);
738 844
739 return res; 845 return res;
740} 846}
741#endif 847#endif
742 848
743#ifndef HAVE_FUTIMES 849#ifndef HAVE_UTIMES
744 850
851# undef utimes
745# define utimes(path,times) eio__utimes (path, times) 852# define utimes(path,times) eio__utimes (path, times)
746# define futimes(fd,times) eio__futimes (fd, times)
747 853
748static int 854static int
749eio__utimes (const char *filename, const struct timeval times[2]) 855eio__utimes (const char *filename, const struct timeval times[2])
750{ 856{
751 if (times) 857 if (times)
… …
759 } 865 }
760 else 866 else
761 return utime (filename, 0); 867 return utime (filename, 0);
762} 868}
763 869
870#endif
871
872#ifndef HAVE_FUTIMES
873
874# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times)
876
764static int eio__futimes (int fd, const struct timeval tv[2]) 877static int eio__futimes (int fd, const struct timeval tv[2])
765{ 878{
766 errno = ENOSYS; 879 errno = ENOSYS;
767 return -1; 880 return -1;
768} 881}
769 882
770#endif 883#endif
771 884
772#if !HAVE_FDATASYNC 885#if !HAVE_FDATASYNC
886# undef fdatasync
773# define fdatasync fsync 887# define fdatasync(fd) fsync (fd)
774#endif 888#endif
889
890/* sync_file_range always needs emulation */
891int
892eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893{
894#if HAVE_SYNC_FILE_RANGE
895 int res;
896
897 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
898 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
899 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
900 {
901 flags = 0
902 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
903 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
904 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
905 }
906
907 res = sync_file_range (fd, offset, nbytes, flags);
908
909 if (!res || errno != ENOSYS)
910 return res;
911#endif
912
913 /* even though we could play tricks with the flags, it's better to always
914 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd);
916}
775 917
776#if !HAVE_READAHEAD 918#if !HAVE_READAHEAD
919# undef readahead
777# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
778 921
779static ssize_t 922static ssize_t
780eio__readahead (int fd, off_t offset, size_t count, worker *self) 923eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
781{ 924{
782 size_t todo = count; 925 size_t todo = count;
783 dBUF; 926 dBUF;
784 927
785 while (todo > 0) 928 while (todo > 0)
… …
799 942
800/* sendfile always needs emulation */ 943/* sendfile always needs emulation */
801static ssize_t 944static ssize_t
802eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
803{ 946{
947 ssize_t written = 0;
804 ssize_t res; 948 ssize_t res;
805 949
806 if (!count) 950 if (!count)
807 return 0; 951 return 0;
808 952
953 for (;;)
954 {
809#if HAVE_SENDFILE 955#if HAVE_SENDFILE
810# if __linux 956# if __linux
957 off_t soffset = offset;
811 res = sendfile (ofd, ifd, &offset, count); 958 res = sendfile (ofd, ifd, &soffset, count);
812 959
813# elif __freebsd 960# elif __FreeBSD__
814 /* 961 /*
815 * Of course, the freebsd sendfile is a dire hack with no thoughts 962 * Of course, the freebsd sendfile is a dire hack with no thoughts
816 * wasted on making it similar to other I/O functions. 963 * wasted on making it similar to other I/O functions.
817 */ 964 */
818 {
819 off_t sbytes; 965 off_t sbytes;
820 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 966 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
821 967
822 if (res < 0 && sbytes) 968 #if 0 /* according to the manpage, this is correct, but broken behaviour */
823 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 969 /* freebsd' sendfile will return 0 on success */
970 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
971 /* not on e.g. EIO or EPIPE - sounds broken */
972 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
824 res = sbytes; 973 res = sbytes;
825 } 974 #endif
975
976 /* according to source inspection, this is correct, and useful behaviour */
977 if (sbytes)
978 res = sbytes;
979
980# elif defined (__APPLE__)
981 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983
984 /* according to the manpage, sbytes is always valid */
985 if (sbytes)
986 res = sbytes;
826 987
827# elif __hpux 988# elif __hpux
828 res = sendfile (ofd, ifd, offset, count, 0, 0); 989 res = sendfile (ofd, ifd, offset, count, 0, 0);
829 990
830# elif __solaris 991# elif __solaris
831 {
832 struct sendfilevec vec; 992 struct sendfilevec vec;
833 size_t sbytes; 993 size_t sbytes;
834 994
835 vec.sfv_fd = ifd; 995 vec.sfv_fd = ifd;
836 vec.sfv_flag = 0; 996 vec.sfv_flag = 0;
837 vec.sfv_off = offset; 997 vec.sfv_off = offset;
838 vec.sfv_len = count; 998 vec.sfv_len = count;
839 999
840 res = sendfilev (ofd, &vec, 1, &sbytes); 1000 res = sendfilev (ofd, &vec, 1, &sbytes);
841 1001
842 if (res < 0 && sbytes) 1002 if (res < 0 && sbytes)
843 res = sbytes; 1003 res = sbytes;
844 }
845 1004
846# endif 1005# endif
1006
1007#elif defined (_WIN32)
1008 /* does not work, just for documentation of what would need to be done */
1009 HANDLE h = TO_SOCKET (ifd);
1010 SetFilePointer (h, offset, 0, FILE_BEGIN);
1011 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1012
847#else 1013#else
848 res = -1; 1014 res = -1;
849 errno = ENOSYS; 1015 errno = ENOSYS;
850#endif 1016#endif
851 1017
1018 /* we assume sendfile can copy at least 128mb in one go */
1019 if (res <= 128 * 1024 * 1024)
1020 {
1021 if (res > 0)
1022 written += res;
1023
1024 if (written)
1025 return written;
1026
1027 break;
1028 }
1029 else
1030 {
1031 /* if we requested more, then probably the kernel was lazy */
1032 written += res;
1033 offset += res;
1034 count -= res;
1035
1036 if (!count)
1037 return written;
1038 }
1039 }
1040
852 if (res < 0 1041 if (res < 0
853 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1042 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1043 /* BSDs */
1044#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1045 || errno == ENOTSUP
1046#endif
1047 || errno == EOPNOTSUPP /* BSDs */
854#if __solaris 1048#if __solaris
855 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1049 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
856#endif 1050#endif
857 ) 1051 )
858 ) 1052 )
… …
889 } 1083 }
890 1084
891 return res; 1085 return res;
892} 1086}
893 1087
1088static signed char
1089eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1090{
1091 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1092 : a->inode < b->inode ? -1
1093 : a->inode > b->inode ? 1
1094 : 0;
1095}
1096
1097#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1098
1099#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1100#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1101
1102static void
1103eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1104{
1105 unsigned char bits [9 + sizeof (ino_t) * 8];
1106 unsigned char *bit = bits;
1107
1108 assert (CHAR_BIT == 8);
1109 assert (sizeof (eio_dirent) * 8 < 256);
1110 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1111 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1112
1113 if (size <= EIO_SORT_FAST)
1114 return;
1115
1116 /* first prepare an array of bits to test in our radix sort */
1117 /* try to take endianness into account, as well as differences in ino_t sizes */
1118 /* inode_bits must contain all inodes ORed together */
1119 /* which is used to skip bits that are 0 everywhere, which is very common */
1120 {
1121 ino_t endianness;
1122 int i, j;
1123
1124 /* we store the byte offset of byte n into byte n of "endianness" */
1125 for (i = 0; i < sizeof (ino_t); ++i)
1126 ((unsigned char *)&endianness)[i] = i;
1127
1128 *bit++ = 0;
1129
1130 for (i = 0; i < sizeof (ino_t); ++i)
1131 {
1132 /* shifting off the byte offsets out of "endianness" */
1133 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1134 endianness >>= 8;
1135
1136 for (j = 0; j < 8; ++j)
1137 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1138 *bit++ = offs + j;
1139 }
1140
1141 for (j = 0; j < 8; ++j)
1142 if (score_bits & (1 << j))
1143 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1144 }
1145
1146 /* now actually do the sorting (a variant of MSD radix sort) */
1147 {
1148 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1149 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1150 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1151 int stk_idx = 0;
1152
1153 base_stk [stk_idx] = dents;
1154 end_stk [stk_idx] = dents + size;
1155 bit_stk [stk_idx] = bit - 1;
1156
1157 do
1158 {
1159 base = base_stk [stk_idx];
1160 end = end_stk [stk_idx];
1161 bit = bit_stk [stk_idx];
1162
1163 for (;;)
1164 {
1165 unsigned char O = *bit >> 3;
1166 unsigned char M = 1 << (*bit & 7);
1167
1168 eio_dirent *a = base;
1169 eio_dirent *b = end;
1170
1171 if (b - a < EIO_SORT_CUTOFF)
1172 break;
1173
1174 /* now bit-partition the array on the bit */
1175 /* this ugly asymmetric loop seems to perform much better than typical */
1176 /* partition algos found in the literature */
1177 do
1178 if (!(((unsigned char *)a)[O] & M))
1179 ++a;
1180 else if (!(((unsigned char *)--b)[O] & M))
1181 {
1182 eio_dirent tmp = *a; *a = *b; *b = tmp;
1183 ++a;
1184 }
1185 while (b > a);
1186
1187 /* next bit, or stop, if no bits left in this path */
1188 if (!*--bit)
1189 break;
1190
1191 base_stk [stk_idx] = a;
1192 end_stk [stk_idx] = end;
1193 bit_stk [stk_idx] = bit;
1194 ++stk_idx;
1195
1196 end = a;
1197 }
1198 }
1199 while (stk_idx--);
1200 }
1201}
1202
1203static void
1204eio_dent_insertion_sort (eio_dirent *dents, int size)
1205{
1206 /* first move the smallest element to the front, to act as a sentinel */
1207 {
1208 int i;
1209 eio_dirent *min = dents;
1210
1211 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1212 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1213 if (EIO_DENT_CMP (dents [i], <, *min))
1214 min = &dents [i];
1215
1216 /* swap elements 0 and j (minimum) */
1217 {
1218 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1219 }
1220 }
1221
1222 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1223 {
1224 eio_dirent *i, *j;
1225
1226 for (i = dents + 1; i < dents + size; ++i)
1227 {
1228 eio_dirent value = *i;
1229
1230 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1231 j [1] = j [0];
1232
1233 j [1] = value;
1234 }
1235 }
1236}
1237
1238static void
1239eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1240{
1241 if (size <= 1)
1242 return; /* our insertion sort relies on size > 0 */
1243
1244 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1245 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1246 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1247
1248 /* use an insertion sort at the end, or for small arrays, */
1249 /* as insertion sort is more efficient for small partitions */
1250 eio_dent_insertion_sort (dents, size);
1251}
1252
894/* read a full directory */ 1253/* read a full directory */
895static void 1254static void
896eio__scandir (eio_req *req, etp_worker *self) 1255eio__scandir (eio_req *req, etp_worker *self)
897{ 1256{
898 DIR *dirp; 1257 DIR *dirp;
899 EIO_STRUCT_DIRENT *entp; 1258 EIO_STRUCT_DIRENT *entp;
900 char *name, *names; 1259 char *name, *names;
901 int memlen = 4096; 1260 int namesalloc = 4096;
902 int memofs = 0; 1261 int namesoffs = 0;
1262 int flags = req->int1;
1263 eio_dirent *dents = 0;
1264 int dentalloc = 128;
903 int res = 0; 1265 int dentoffs = 0;
1266 ino_t inode_bits = 0;
1267
1268 req->result = -1;
1269
1270 if (!(flags & EIO_READDIR_DENTS))
1271 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
904 1272
905 X_LOCK (wrklock); 1273 X_LOCK (wrklock);
1274 /* the corresponding closedir is in ETP_WORKER_CLEAR */
906 self->dirp = dirp = opendir (req->ptr1); 1275 self->dirp = dirp = opendir (req->ptr1);
1276
907 req->flags |= EIO_FLAG_PTR2_FREE; 1277 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1278 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
908 req->ptr2 = names = malloc (memlen); 1279 req->ptr2 = names = malloc (namesalloc);
909 X_UNLOCK (wrklock); 1280 X_UNLOCK (wrklock);
910 1281
911 if (dirp && names) 1282 if (dirp && names && (!flags || dents))
912 for (;;) 1283 for (;;)
913 { 1284 {
914 errno = 0; 1285 errno = 0;
915 entp = readdir (dirp); 1286 entp = readdir (dirp);
916 1287
917 if (!entp) 1288 if (!entp)
1289 {
1290 if (errno)
1291 break;
1292
1293 /* sort etc. */
1294 req->int1 = flags;
1295 req->result = dentoffs;
1296
1297 if (flags & EIO_READDIR_STAT_ORDER)
1298 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1299 else if (flags & EIO_READDIR_DIRS_FIRST)
1300 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1301 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1302 else
1303 {
1304 /* in this case, all is known, and we just put dirs first and sort them */
1305 eio_dirent *oth = dents + dentoffs;
1306 eio_dirent *dir = dents;
1307
1308 /* now partition dirs to the front, and non-dirs to the back */
1309 /* by walking from both sides and swapping if necessary */
1310 while (oth > dir)
1311 {
1312 if (dir->type == EIO_DT_DIR)
1313 ++dir;
1314 else if ((--oth)->type == EIO_DT_DIR)
1315 {
1316 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1317
1318 ++dir;
1319 }
1320 }
1321
1322 /* now sort the dirs only (dirs all have the same score) */
1323 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1324 }
1325
918 break; 1326 break;
1327 }
919 1328
1329 /* now add the entry to our list(s) */
920 name = entp->d_name; 1330 name = entp->d_name;
921 1331
1332 /* skip . and .. entries */
922 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1333 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
923 { 1334 {
924 int len = strlen (name) + 1; 1335 int len = D_NAMLEN (entp) + 1;
925 1336
926 res++; 1337 while (expect_false (namesoffs + len > namesalloc))
927
928 while (memofs + len > memlen)
929 { 1338 {
930 memlen *= 2; 1339 namesalloc *= 2;
931 X_LOCK (wrklock); 1340 X_LOCK (wrklock);
932 req->ptr2 = names = realloc (names, memlen); 1341 req->ptr2 = names = realloc (names, namesalloc);
933 X_UNLOCK (wrklock); 1342 X_UNLOCK (wrklock);
934 1343
935 if (!names) 1344 if (!names)
936 break; 1345 break;
937 } 1346 }
938 1347
939 memcpy (names + memofs, name, len); 1348 memcpy (names + namesoffs, name, len);
1349
1350 if (dents)
1351 {
1352 struct eio_dirent *ent;
1353
1354 if (expect_false (dentoffs == dentalloc))
1355 {
1356 dentalloc *= 2;
1357 X_LOCK (wrklock);
1358 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1359 X_UNLOCK (wrklock);
1360
1361 if (!dents)
1362 break;
1363 }
1364
1365 ent = dents + dentoffs;
1366
1367 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1368 ent->namelen = len - 1;
1369 ent->inode = D_INO (entp);
1370
1371 inode_bits |= ent->inode;
1372
1373 switch (D_TYPE (entp))
1374 {
1375 default:
1376 ent->type = EIO_DT_UNKNOWN;
1377 flags |= EIO_READDIR_FOUND_UNKNOWN;
1378 break;
1379
1380 #ifdef DT_FIFO
1381 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1382 #endif
1383 #ifdef DT_CHR
1384 case DT_CHR: ent->type = EIO_DT_CHR; break;
1385 #endif
1386 #ifdef DT_MPC
1387 case DT_MPC: ent->type = EIO_DT_MPC; break;
1388 #endif
1389 #ifdef DT_DIR
1390 case DT_DIR: ent->type = EIO_DT_DIR; break;
1391 #endif
1392 #ifdef DT_NAM
1393 case DT_NAM: ent->type = EIO_DT_NAM; break;
1394 #endif
1395 #ifdef DT_BLK
1396 case DT_BLK: ent->type = EIO_DT_BLK; break;
1397 #endif
1398 #ifdef DT_MPB
1399 case DT_MPB: ent->type = EIO_DT_MPB; break;
1400 #endif
1401 #ifdef DT_REG
1402 case DT_REG: ent->type = EIO_DT_REG; break;
1403 #endif
1404 #ifdef DT_NWK
1405 case DT_NWK: ent->type = EIO_DT_NWK; break;
1406 #endif
1407 #ifdef DT_CMP
1408 case DT_CMP: ent->type = EIO_DT_CMP; break;
1409 #endif
1410 #ifdef DT_LNK
1411 case DT_LNK: ent->type = EIO_DT_LNK; break;
1412 #endif
1413 #ifdef DT_SOCK
1414 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1415 #endif
1416 #ifdef DT_DOOR
1417 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1418 #endif
1419 #ifdef DT_WHT
1420 case DT_WHT: ent->type = EIO_DT_WHT; break;
1421 #endif
1422 }
1423
1424 ent->score = 7;
1425
1426 if (flags & EIO_READDIR_DIRS_FIRST)
1427 {
1428 if (ent->type == EIO_DT_UNKNOWN)
1429 {
1430 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1431 ent->score = 1;
1432 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1433 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1434 }
1435 else if (ent->type == EIO_DT_DIR)
1436 ent->score = 0;
1437 }
1438 }
1439
940 memofs += len; 1440 namesoffs += len;
1441 ++dentoffs;
1442 }
1443
1444 if (EIO_CANCELLED (req))
1445 {
1446 errno = ECANCELED;
1447 break;
941 } 1448 }
942 } 1449 }
1450}
943 1451
944 if (errno) 1452#ifdef PAGESIZE
945 res = -1; 1453# define eio_pagesize() PAGESIZE
1454#else
1455static intptr_t
1456eio_pagesize (void)
1457{
1458 static intptr_t page;
1459
1460 if (!page)
1461 page = sysconf (_SC_PAGESIZE);
1462
1463 return page;
1464}
1465#endif
1466
1467static void
1468eio_page_align (void **addr, size_t *length)
1469{
1470 intptr_t mask = eio_pagesize () - 1;
1471
1472 /* round down addr */
1473 intptr_t adj = mask & (intptr_t)*addr;
1474
1475 *addr = (void *)((intptr_t)*addr - adj);
1476 *length += adj;
1477
1478 /* round up length */
1479 *length = (*length + mask) & ~mask;
1480}
1481
1482#if !_POSIX_MEMLOCK
1483# define eio__mlockall(a) ((errno = ENOSYS), -1)
1484#else
1485
1486static int
1487eio__mlockall (int flags)
1488{
1489 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1490 extern int mallopt (int, int);
1491 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1492 #endif
1493
1494 if (EIO_MCL_CURRENT != MCL_CURRENT
1495 || EIO_MCL_FUTURE != MCL_FUTURE)
1496 {
1497 flags = 0
1498 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1499 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1500 }
1501
1502 return mlockall (flags);
1503}
1504#endif
1505
1506#if !_POSIX_MEMLOCK_RANGE
1507# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1508#else
1509
1510static int
1511eio__mlock (void *addr, size_t length)
1512{
1513 eio_page_align (&addr, &length);
1514
1515 return mlock (addr, length);
1516}
1517
1518#endif
1519
1520#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1521# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1522#else
1523
1524int
1525eio__msync (void *mem, size_t len, int flags)
1526{
1527 eio_page_align (&mem, &len);
1528
1529 if (EIO_MS_ASYNC != MS_SYNC
1530 || EIO_MS_INVALIDATE != MS_INVALIDATE
1531 || EIO_MS_SYNC != MS_SYNC)
1532 {
1533 flags = 0
1534 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1535 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1536 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1537 }
1538
1539 return msync (mem, len, flags);
1540}
1541
1542#endif
1543
1544int
1545eio__mtouch (eio_req *req)
1546{
1547 void *mem = req->ptr2;
1548 size_t len = req->size;
1549 int flags = req->int1;
1550
1551 eio_page_align (&mem, &len);
1552
946 1553 {
947 req->result = res; 1554 intptr_t addr = (intptr_t)mem;
1555 intptr_t end = addr + len;
1556 intptr_t page = eio_pagesize ();
1557
1558 if (addr < end)
1559 if (flags & EIO_MT_MODIFY) /* modify */
1560 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1561 else
1562 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1563 }
1564
1565 return 0;
948} 1566}
949 1567
950/*****************************************************************************/ 1568/*****************************************************************************/
951 1569
952#define ALLOC(len) \ 1570#define ALLOC(len) \
… …
984 if (req) 1602 if (req)
985 break; 1603 break;
986 1604
987 ++idle; 1605 ++idle;
988 1606
989 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1607 ts.tv_sec = time (0) + idle_timeout;
990 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1608 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
991 { 1609 {
992 if (idle > max_idle) 1610 if (idle > max_idle)
993 { 1611 {
994 --idle; 1612 --idle;
… …
1039 1657
1040/*****************************************************************************/ 1658/*****************************************************************************/
1041 1659
1042int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1660int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1043{ 1661{
1044 etp_init (want_poll, done_poll); 1662 return etp_init (want_poll, done_poll);
1045} 1663}
1046 1664
1047static void eio_api_destroy (eio_req *req) 1665static void eio_api_destroy (eio_req *req)
1048{ 1666{
1049 free (req); 1667 free (req);
… …
1073 return 0; \ 1691 return 0; \
1074 } 1692 }
1075 1693
1076static void eio_execute (etp_worker *self, eio_req *req) 1694static void eio_execute (etp_worker *self, eio_req *req)
1077{ 1695{
1078 errno = 0;
1079
1080 switch (req->type) 1696 switch (req->type)
1081 { 1697 {
1082 case EIO_READ: ALLOC (req->size); 1698 case EIO_READ: ALLOC (req->size);
1083 req->result = req->offs >= 0 1699 req->result = req->offs >= 0
1084 ? pread (req->int1, req->ptr2, req->size, req->offs) 1700 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1094 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1710 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1095 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1711 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1096 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1712 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1097 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1713 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1714 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1715
1716 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1717 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1718 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1719 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1099 1720
1100 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1721 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1101 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1722 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1102 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1723 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1103 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1724 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
… …
1113 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1734 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1114 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1735 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1115 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1736 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1116 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 1737 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1117 1738
1118 case EIO_READLINK: ALLOC (NAME_MAX); 1739 case EIO_READLINK: ALLOC (PATH_MAX);
1119 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1740 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1120 1741
1121 case EIO_SYNC: req->result = 0; sync (); break; 1742 case EIO_SYNC: req->result = 0; sync (); break;
1122 case EIO_FSYNC: req->result = fsync (req->int1); break; 1743 case EIO_FSYNC: req->result = fsync (req->int1); break;
1123 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1744 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1745 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1746 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1747 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1748 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1749 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1124 1750
1125 case EIO_READDIR: eio__scandir (req, self); break; 1751 case EIO_READDIR: eio__scandir (req, self); break;
1126 1752
1127 case EIO_BUSY: 1753 case EIO_BUSY:
1128#ifdef _WIN32 1754#ifdef _WIN32
1129 Sleep (req->nv1 * 1000.); 1755 Sleep (req->nv1 * 1e3);
1130#else 1756#else
1131 { 1757 {
1132 struct timeval tv; 1758 struct timeval tv;
1133 1759
1134 tv.tv_sec = req->nv1; 1760 tv.tv_sec = req->nv1;
1135 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1761 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1136 1762
1137 req->result = select (0, 0, 0, 0, &tv); 1763 req->result = select (0, 0, 0, 0, &tv);
1138 } 1764 }
1139#endif 1765#endif
1140 break; 1766 break;
… …
1155 times = tv; 1781 times = tv;
1156 } 1782 }
1157 else 1783 else
1158 times = 0; 1784 times = 0;
1159 1785
1160
1161 req->result = req->type == EIO_FUTIME 1786 req->result = req->type == EIO_FUTIME
1162 ? futimes (req->int1, times) 1787 ? futimes (req->int1, times)
1163 : utimes (req->ptr1, times); 1788 : utimes (req->ptr1, times);
1164 } 1789 }
1790 break;
1165 1791
1166 case EIO_GROUP: 1792 case EIO_GROUP:
1793 abort (); /* handled in eio_request */
1794
1167 case EIO_NOP: 1795 case EIO_NOP:
1168 req->result = 0; 1796 req->result = 0;
1169 break; 1797 break;
1170 1798
1171 case EIO_CUSTOM: 1799 case EIO_CUSTOM:
1172 req->feed (req); 1800 req->feed (req);
1173 break; 1801 break;
1174 1802
1175 default: 1803 default:
1804 errno = ENOSYS;
1176 req->result = -1; 1805 req->result = -1;
1177 break; 1806 break;
1178 } 1807 }
1179 1808
1180 req->errorno = errno; 1809 req->errorno = errno;
… …
1200eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1829eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1201{ 1830{
1202 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1831 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1203} 1832}
1204 1833
1834eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1835{
1836 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1837}
1838
1839eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1840{
1841 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1842}
1843
1844eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1845{
1846 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1847}
1848
1849eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1850{
1851 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1852}
1853
1854eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1855{
1856 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1857}
1858
1205eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1859eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1206{ 1860{
1207 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1861 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1208} 1862}
1209 1863
… …
1228} 1882}
1229 1883
1230eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1884eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1231{ 1885{
1232 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1886 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1887}
1888
1889eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1890{
1891 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1233} 1892}
1234 1893
1235eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1894eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1236{ 1895{
1237 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1896 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
… …
1311eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1970eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1312{ 1971{
1313 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1972 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1314} 1973}
1315 1974
1975eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1976{
1977 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1978}
1979
1316eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1980eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1317{ 1981{
1318 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1982 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1319} 1983}
1320 1984
1321eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1985eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1322{ 1986{
1323 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1987 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1324} 1988}
1325 1989
1326eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1990eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1327{ 1991{
1328 return eio__1path (EIO_READDIR, path, pri, cb, data); 1992 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1329} 1993}
1330 1994
1331eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1995eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1332{ 1996{
1333 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1997 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1334} 1998}
1335 1999
1336static eio_req * 2000static eio_req *
1337eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2001eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1338{ 2002{
… …
1362eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2026eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1363{ 2027{
1364 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2028 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1365} 2029}
1366 2030
1367eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2031eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data);
1368{ 2032{
1369 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2033 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1370} 2034}
1371 2035
1372#endif 2036#endif
… …
1402 2066
1403void eio_grp_add (eio_req *grp, eio_req *req) 2067void eio_grp_add (eio_req *grp, eio_req *req)
1404{ 2068{
1405 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2069 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1406 2070
2071 grp->flags |= EIO_FLAG_GROUPADD;
2072
1407 ++grp->size; 2073 ++grp->size;
1408 req->grp = grp; 2074 req->grp = grp;
1409 2075
1410 req->grp_prev = 0; 2076 req->grp_prev = 0;
1411 req->grp_next = grp->grp_first; 2077 req->grp_next = grp->grp_first;
… …
1420/* misc garbage */ 2086/* misc garbage */
1421 2087
1422ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2088ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1423{ 2089{
1424 etp_worker wrk; 2090 etp_worker wrk;
2091 ssize_t ret;
1425 2092
1426 wrk.dbuf = 0; 2093 wrk.dbuf = 0;
1427 2094
1428 eio__sendfile (ofd, ifd, offset, count, &wrk); 2095 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1429 2096
1430 if (wrk.dbuf) 2097 if (wrk.dbuf)
1431 free (wrk.dbuf); 2098 free (wrk.dbuf);
1432}
1433 2099
2100 return ret;
2101}
2102

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines