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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.15 by root, Sat May 17 12:26:57 2008 UTC vs.
Revision 1.32 by root, Sat Jun 6 18:06:55 2009 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 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 *
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#include "eio.h" 40#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
41#include "xthread.h" 45#include "xthread.h"
42 46
43#include <errno.h> 47#include <errno.h>
44#include <stddef.h> 48#include <stddef.h>
45#include <stdlib.h> 49#include <stdlib.h>
64#endif 68#endif
65 69
66#ifdef _WIN32 70#ifdef _WIN32
67 71
68 /*doh*/ 72 /*doh*/
69
70#else 73#else
71 74
72# include "config.h" 75# include "config.h"
73# include <sys/time.h> 76# include <sys/time.h>
74# include <sys/select.h> 77# include <sys/select.h>
75# include <unistd.h> 78# include <unistd.h>
76# include <utime.h> 79# include <utime.h>
77# include <signal.h> 80# include <signal.h>
78# include <dirent.h> 81# include <dirent.h>
82
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
85# define D_INO(de) (de)->d_fileno
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
88# define D_INO(de) (de)->d_ino
89# endif
90
91# ifdef _DIRENT_HAVE_D_TYPE
92# define D_TYPE(de) (de)->d_type
93# endif
79 94
80# ifndef EIO_STRUCT_DIRENT 95# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 96# define EIO_STRUCT_DIRENT struct dirent
82# endif 97# endif
83 98
96# else 111# else
97# error sendfile support requested but not available 112# error sendfile support requested but not available
98# endif 113# endif
99#endif 114#endif
100 115
116#ifndef D_TYPE
117# define D_TYPE(de) 0
118#endif
119
120#ifndef D_INO
121# define D_INO(de) 0
122#endif
123
101/* number of seconds after which an idle threads exit */ 124/* number of seconds after which an idle threads exit */
102#define IDLE_TIMEOUT 10 125#define IDLE_TIMEOUT 10
103 126
104/* used for struct dirent, AIX doesn't provide it */ 127/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 128#ifndef NAME_MAX
117 errno = ENOMEM; \ 140 errno = ENOMEM; \
118 if (!eio_buf) \ 141 if (!eio_buf) \
119 return -1; 142 return -1;
120 143
121#define EIO_TICKS ((1000000 + 1023) >> 10) 144#define EIO_TICKS ((1000000 + 1023) >> 10)
145
146/*****************************************************************************/
147
148#if __GNUC__ >= 3
149# define expect(expr,value) __builtin_expect ((expr),(value))
150#else
151# define expect(expr,value) (expr)
152#endif
153
154#define expect_false(expr) expect ((expr) != 0, 0)
155#define expect_true(expr) expect ((expr) != 0, 1)
122 156
123/*****************************************************************************/ 157/*****************************************************************************/
124 158
125#define ETP_PRI_MIN EIO_PRI_MIN 159#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 160#define ETP_PRI_MAX EIO_PRI_MAX
144 if (wrk->dirp) \ 178 if (wrk->dirp) \
145 { \ 179 { \
146 closedir (wrk->dirp); \ 180 closedir (wrk->dirp); \
147 wrk->dirp = 0; \ 181 wrk->dirp = 0; \
148 } 182 }
183
149#define ETP_WORKER_COMMON \ 184#define ETP_WORKER_COMMON \
150 void *dbuf; \ 185 void *dbuf; \
151 DIR *dirp; 186 DIR *dirp;
152 187
153/*****************************************************************************/ 188/*****************************************************************************/
176 211
177static mutex_t wrklock = X_MUTEX_INIT; 212static mutex_t wrklock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT; 213static mutex_t reslock = X_MUTEX_INIT;
179static mutex_t reqlock = X_MUTEX_INIT; 214static mutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT; 215static cond_t reqwait = X_COND_INIT;
216
217#if !HAVE_PREADWRITE
218/*
219 * make our pread/pwrite emulation safe against themselves, but not against
220 * normal read/write by using a mutex. slows down execution a lot,
221 * but that's your problem, not mine.
222 */
223static mutex_t preadwritelock = X_MUTEX_INIT;
224#endif
181 225
182typedef struct etp_worker 226typedef struct etp_worker
183{ 227{
184 /* locked by wrklock */ 228 /* locked by wrklock */
185 struct etp_worker *prev, *next; 229 struct etp_worker *prev, *next;
314 X_LOCK (reqlock); 358 X_LOCK (reqlock);
315 X_LOCK (reslock); 359 X_LOCK (reslock);
316#if !HAVE_PREADWRITE 360#if !HAVE_PREADWRITE
317 X_LOCK (preadwritelock); 361 X_LOCK (preadwritelock);
318#endif 362#endif
319#if !HAVE_READDIR_R
320 X_LOCK (readdirlock);
321#endif
322} 363}
323 364
324static void etp_atfork_parent (void) 365static void etp_atfork_parent (void)
325{ 366{
326#if !HAVE_READDIR_R
327 X_UNLOCK (readdirlock);
328#endif
329#if !HAVE_PREADWRITE 367#if !HAVE_PREADWRITE
330 X_UNLOCK (preadwritelock); 368 X_UNLOCK (preadwritelock);
331#endif 369#endif
332 X_UNLOCK (reslock); 370 X_UNLOCK (reslock);
333 X_UNLOCK (reqlock); 371 X_UNLOCK (reqlock);
336 374
337static void etp_atfork_child (void) 375static void etp_atfork_child (void)
338{ 376{
339 ETP_REQ *prv; 377 ETP_REQ *prv;
340 378
341 while (prv = reqq_shift (&req_queue)) 379 while ((prv = reqq_shift (&req_queue)))
342 ETP_DESTROY (prv); 380 ETP_DESTROY (prv);
343 381
344 while (prv = reqq_shift (&res_queue)) 382 while ((prv = reqq_shift (&res_queue)))
345 ETP_DESTROY (prv); 383 ETP_DESTROY (prv);
346 384
347 while (wrk_first.next != &wrk_first) 385 while (wrk_first.next != &wrk_first)
348 { 386 {
349 etp_worker *wrk = wrk_first.next; 387 etp_worker *wrk = wrk_first.next;
377 415
378 pthread_once (&doinit, etp_once_init); 416 pthread_once (&doinit, etp_once_init);
379 417
380 want_poll_cb = want_poll; 418 want_poll_cb = want_poll;
381 done_poll_cb = done_poll; 419 done_poll_cb = done_poll;
420
421 return 0;
382} 422}
383 423
384X_THREAD_PROC (etp_proc); 424X_THREAD_PROC (etp_proc);
385 425
386static void etp_start_thread (void) 426static void etp_start_thread (void)
406 X_UNLOCK (wrklock); 446 X_UNLOCK (wrklock);
407} 447}
408 448
409static void etp_maybe_start_thread (void) 449static void etp_maybe_start_thread (void)
410{ 450{
411 if (etp_nthreads () >= wanted) 451 if (expect_true (etp_nthreads () >= wanted))
412 return; 452 return;
413 453
414 /* todo: maybe use idle here, but might be less exact */ 454 /* todo: maybe use idle here, but might be less exact */
415 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 455 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
416 return; 456 return;
417 457
418 etp_start_thread (); 458 etp_start_thread ();
419} 459}
420 460
473 513
474 X_LOCK (reqlock); 514 X_LOCK (reqlock);
475 --nreqs; 515 --nreqs;
476 X_UNLOCK (reqlock); 516 X_UNLOCK (reqlock);
477 517
478 if (req->type == EIO_GROUP && req->size) 518 if (expect_false (req->type == EIO_GROUP && req->size))
479 { 519 {
480 req->int1 = 1; /* mark request as delayed */ 520 req->int1 = 1; /* mark request as delayed */
481 continue; 521 continue;
482 } 522 }
483 else 523 else
484 { 524 {
485 int res = ETP_FINISH (req); 525 int res = ETP_FINISH (req);
486 if (res) 526 if (expect_false (res))
487 return res; 527 return res;
488 } 528 }
489 529
490 if (maxreqs && !--maxreqs) 530 if (expect_false (maxreqs && !--maxreqs))
491 break; 531 break;
492 532
493 if (maxtime) 533 if (maxtime)
494 { 534 {
495 gettimeofday (&tv_now, 0); 535 gettimeofday (&tv_now, 0);
514 554
515static void etp_submit (ETP_REQ *req) 555static void etp_submit (ETP_REQ *req)
516{ 556{
517 req->pri -= ETP_PRI_MIN; 557 req->pri -= ETP_PRI_MIN;
518 558
519 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 559 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
520 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 560 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
521 561
562 if (expect_false (req->type == EIO_GROUP))
563 {
564 /* I hope this is worth it :/ */
522 X_LOCK (reqlock); 565 X_LOCK (reqlock);
523 ++nreqs; 566 ++nreqs;
567 X_UNLOCK (reqlock);
568
569 X_LOCK (reslock);
570
571 ++npending;
572
573 if (!reqq_push (&res_queue, req) && want_poll_cb)
574 want_poll_cb ();
575
576 X_UNLOCK (reslock);
577 }
578 else
579 {
580 X_LOCK (reqlock);
581 ++nreqs;
524 ++nready; 582 ++nready;
525 reqq_push (&req_queue, req); 583 reqq_push (&req_queue, req);
526 X_COND_SIGNAL (reqwait); 584 X_COND_SIGNAL (reqwait);
527 X_UNLOCK (reqlock); 585 X_UNLOCK (reqlock);
528 586
529 etp_maybe_start_thread (); 587 etp_maybe_start_thread ();
588 }
530} 589}
531 590
532static void etp_set_max_poll_time (double nseconds) 591static void etp_set_max_poll_time (double nseconds)
533{ 592{
534 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 593 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
569 628
570static void grp_try_feed (eio_req *grp) 629static void grp_try_feed (eio_req *grp)
571{ 630{
572 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 631 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
573 { 632 {
574 int old_len = grp->size; 633 grp->flags &= ~EIO_FLAG_GROUPADD;
575 634
576 EIO_FEED (grp); 635 EIO_FEED (grp);
577 636
578 /* stop if no progress has been made */ 637 /* stop if no progress has been made */
579 if (old_len == grp->size) 638 if (!(grp->flags & EIO_FLAG_GROUPADD))
580 { 639 {
581 grp->feed = 0; 640 grp->feed = 0;
582 break; 641 break;
583 } 642 }
584 } 643 }
701 760
702/*****************************************************************************/ 761/*****************************************************************************/
703/* work around various missing functions */ 762/* work around various missing functions */
704 763
705#if !HAVE_PREADWRITE 764#if !HAVE_PREADWRITE
765# undef pread
766# undef pwrite
706# define pread eio__pread 767# define pread eio__pread
707# define pwrite eio__pwrite 768# define pwrite eio__pwrite
708
709/*
710 * make our pread/pwrite safe against themselves, but not against
711 * normal read/write by using a mutex. slows down execution a lot,
712 * but that's your problem, not mine.
713 */
714static mutex_t preadwritelock = X_MUTEX_INIT;
715 769
716static ssize_t 770static ssize_t
717eio__pread (int fd, void *buf, size_t count, off_t offset) 771eio__pread (int fd, void *buf, size_t count, off_t offset)
718{ 772{
719 ssize_t res; 773 ssize_t res;
737 791
738 X_LOCK (preadwritelock); 792 X_LOCK (preadwritelock);
739 ooffset = lseek (fd, 0, SEEK_CUR); 793 ooffset = lseek (fd, 0, SEEK_CUR);
740 lseek (fd, offset, SEEK_SET); 794 lseek (fd, offset, SEEK_SET);
741 res = write (fd, buf, count); 795 res = write (fd, buf, count);
742 lseek (fd, offset, SEEK_SET); 796 lseek (fd, ooffset, SEEK_SET);
743 X_UNLOCK (preadwritelock); 797 X_UNLOCK (preadwritelock);
744 798
745 return res; 799 return res;
746} 800}
747#endif 801#endif
748 802
749#ifndef HAVE_FUTIMES 803#ifndef HAVE_FUTIMES
750 804
805# undef utimes
806# undef futimes
751# define utimes(path,times) eio__utimes (path, times) 807# define utimes(path,times) eio__utimes (path, times)
752# define futimes(fd,times) eio__futimes (fd, times) 808# define futimes(fd,times) eio__futimes (fd, times)
753 809
754static int 810static int
755eio__utimes (const char *filename, const struct timeval times[2]) 811eio__utimes (const char *filename, const struct timeval times[2])
774} 830}
775 831
776#endif 832#endif
777 833
778#if !HAVE_FDATASYNC 834#if !HAVE_FDATASYNC
835# undef fdatasync
779# define fdatasync fsync 836# define fdatasync(fd) fsync (fd)
780#endif 837#endif
838
839/* sync_file_range always needs emulation */
840int
841eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
842{
843#if HAVE_SYNC_FILE_RANGE
844 int res;
845
846 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
847 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
848 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
849 {
850 flags = 0
851 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
852 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
853 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
854 }
855
856 res = sync_file_range (fd, offset, nbytes, flags);
857
858 if (!res || errno != ENOSYS)
859 return res;
860#endif
861
862 /* even though we could play tricks with the flags, it's better to always
863 * call fdatasync, as thta matches the expectation of it's users best */
864 return fdatasync (fd);
865}
781 866
782#if !HAVE_READAHEAD 867#if !HAVE_READAHEAD
868# undef readahead
783# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 869# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
784 870
785static ssize_t 871static ssize_t
786eio__readahead (int fd, off_t offset, size_t count, worker *self) 872eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
787{ 873{
788 size_t todo = count; 874 size_t todo = count;
789 dBUF; 875 dBUF;
790 876
791 while (todo > 0) 877 while (todo > 0)
799 885
800 errno = 0; 886 errno = 0;
801 return count; 887 return count;
802} 888}
803 889
804#endif
805
806#if !HAVE_READDIR_R
807# define readdir_r eio__readdir_r
808
809static mutex_t readdirlock = X_MUTEX_INIT;
810
811static int
812eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
813{
814 EIO_STRUCT_DIRENT *e;
815 int errorno;
816
817 X_LOCK (readdirlock);
818
819 e = readdir (dirp);
820 errorno = errno;
821
822 if (e)
823 {
824 *res = ent;
825 strcpy (ent->d_name, e->d_name);
826 }
827 else
828 *res = 0;
829
830 X_UNLOCK (readdirlock);
831
832 errno = errorno;
833 return e ? 0 : -1;
834}
835#endif 890#endif
836 891
837/* sendfile always needs emulation */ 892/* sendfile always needs emulation */
838static ssize_t 893static ssize_t
839eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 894eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
926 } 981 }
927 982
928 return res; 983 return res;
929} 984}
930 985
986static int
987eio_dent_cmp (const void *a_, const void *b_)
988{
989 const eio_dirent *a = (const eio_dirent *)a_;
990 const eio_dirent *b = (const eio_dirent *)b_;
991
992 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
993 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */
994}
995
931/* read a full directory */ 996/* read a full directory */
932static void 997static void
933eio__scandir (eio_req *req, etp_worker *self) 998eio__scandir (eio_req *req, etp_worker *self)
934{ 999{
935 DIR *dirp; 1000 DIR *dirp;
936 union
937 {
938 EIO_STRUCT_DIRENT d;
939 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
940 } *u;
941 EIO_STRUCT_DIRENT *entp; 1001 EIO_STRUCT_DIRENT *entp;
942 char *name, *names; 1002 unsigned char *name, *names;
943 int memlen = 4096; 1003 int namesalloc = 4096;
944 int memofs = 0; 1004 int namesoffs = 0;
1005 int flags = req->int1;
1006 eio_dirent *dents = 0;
1007 int dentalloc = 128;
945 int res = 0; 1008 int dentoffs = 0;
1009
1010 req->result = -1;
1011
1012 if (!(flags & EIO_READDIR_DENTS))
1013 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
946 1014
947 X_LOCK (wrklock); 1015 X_LOCK (wrklock);
1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */
948 self->dirp = dirp = opendir (req->ptr1); 1017 self->dirp = dirp = opendir (req->ptr1);
949 self->dbuf = u = malloc (sizeof (*u));
950 req->flags |= EIO_FLAG_PTR2_FREE; 1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
951 req->ptr2 = names = malloc (memlen); 1019 req->ptr1 = names = malloc (namesalloc);
1020 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
952 X_UNLOCK (wrklock); 1021 X_UNLOCK (wrklock);
953 1022
954 if (dirp && u && names) 1023 if (dirp && names && (!flags || dents))
955 for (;;) 1024 for (;;)
956 { 1025 {
957 errno = 0; 1026 errno = 0;
958 readdir_r (dirp, &u->d, &entp); 1027 entp = readdir (dirp);
959 1028
960 if (!entp) 1029 if (!entp)
1030 {
1031 if (errno)
1032 break;
1033
1034 /* sort etc. */
1035 req->int1 = flags;
1036 req->result = dentoffs;
1037
1038 if (dents)
1039 {
1040 eio_dirent *ent = dents + dentoffs;
1041
1042 while (ent > dents)
1043 (--ent)->name = names + (size_t)ent->name;
1044 }
1045
1046 if (flags & EIO_READDIR_STAT_ORDER
1047 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
1048 {
1049 /* pray your qsort doesn't use quicksort */
1050 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1051 }
1052 else if (flags & EIO_READDIR_DIRS_FIRST)
1053 {
1054 /* in this case, all is known, and we just put dirs first and sort them */
1055 eio_dirent *ent = dents + dentoffs;
1056 eio_dirent *dir = dents;
1057
1058 while (ent > dir)
1059 {
1060 if (dir->type == DT_DIR)
1061 ++dir;
1062 else
1063 {
1064 --ent;
1065
1066 if (ent->type == DT_DIR)
1067 {
1068 eio_dirent tmp = *dir;
1069 *dir = *ent;
1070 *ent = tmp;
1071
1072 ++dir;
1073 }
1074 }
1075 }
1076
1077 /* now sort the dirs only */
1078 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1079 }
1080
1081 /* only provide the names array unless DENTS is specified */
1082 if (!(flags & EIO_READDIR_DENTS))
1083 {
1084 X_LOCK (wrklock);
1085 assert (!dents);
1086 req->ptr1 = 0;
1087 req->ptr2 = names;
1088 X_UNLOCK (wrklock);
1089 }
1090
961 break; 1091 break;
1092 }
962 1093
1094 /* now add the entry to our list(s) */
963 name = entp->d_name; 1095 name = entp->d_name;
964 1096
1097 /* skip . and .. entries */
965 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1098 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
966 { 1099 {
967 int len = strlen (name) + 1; 1100 int len = strlen (name) + 1;
968 1101
969 res++; 1102 while (expect_false (namesoffs + len > namesalloc))
970
971 while (memofs + len > memlen)
972 { 1103 {
973 memlen *= 2; 1104 namesalloc *= 2;
974 X_LOCK (wrklock); 1105 X_LOCK (wrklock);
975 req->ptr2 = names = realloc (names, memlen); 1106 req->ptr1 = names = realloc (names, namesalloc);
976 X_UNLOCK (wrklock); 1107 X_UNLOCK (wrklock);
977 1108
978 if (!names) 1109 if (!names)
979 break; 1110 break;
980 } 1111 }
981 1112
982 memcpy (names + memofs, name, len); 1113 memcpy (names + namesoffs, name, len);
1114
1115 if (dents)
1116 {
1117 struct eio_dirent *ent;
1118
1119 if (expect_false (dentoffs == dentalloc))
1120 {
1121 dentalloc *= 2;
1122 X_LOCK (wrklock);
1123 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124 X_UNLOCK (wrklock);
1125
1126 if (!dents)
1127 break;
1128 }
1129
1130 ent = dents + dentoffs;
1131
1132 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
1133 ent->namelen = len - 1;
1134 ent->inode = D_INO (entp);
1135
1136 switch (D_TYPE (entp))
1137 {
1138 default:
1139 ent->type = EIO_DT_UNKNOWN;
1140 flags |= EIO_READDIR_FOUND_UNKNOWN;
1141 break;
1142
1143 #ifdef DT_FIFO
1144 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145 #endif
1146 #ifdef DT_CHR
1147 case DT_CHR: ent->type = EIO_DT_CHR; break;
1148 #endif
1149 #ifdef DT_DIR
1150 case DT_DIR: ent->type = EIO_DT_DIR; break;
1151 #endif
1152 #ifdef DT_BLK
1153 case DT_BLK: ent->type = EIO_DT_BLK; break;
1154 #endif
1155 #ifdef DT_REG
1156 case DT_REG: ent->type = EIO_DT_REG; break;
1157 #endif
1158 #ifdef DT_LNK
1159 case DT_LNK: ent->type = EIO_DT_LNK; break;
1160 #endif
1161 #ifdef DT_SOCK
1162 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163 #endif
1164 #ifdef DT_WHT
1165 case DT_WHT: ent->type = EIO_DT_WHT; break;
1166 #endif
1167 }
1168
1169 ent->score = 0;
1170
1171 if (flags & EIO_READDIR_DIRS_FIRST)
1172 {
1173 if (ent->type == EIO_DT_UNKNOWN)
1174 {
1175 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176 ent->score = 98;
1177 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1178 ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
1179 }
1180 else if (ent->type == DT_DIR)
1181 ent->score = 100;
1182 }
1183 }
1184
983 memofs += len; 1185 namesoffs += len;
1186 ++dentoffs;
984 } 1187 }
985 } 1188 }
986 1189 else
987 if (errno)
988 res = -1;
989
990 req->result = res; 1190 req->result = -1;
1191}
1192
1193#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1194# undef msync
1195# define msync(a,b,c) ((errno = ENOSYS), -1)
1196#endif
1197
1198int
1199eio__mtouch (void *mem, size_t len, int flags)
1200{
1201 intptr_t addr = (intptr_t)mem;
1202 intptr_t end = addr + len;
1203#ifdef PAGESIZE
1204 const intptr_t page = PAGESIZE;
1205#else
1206 static intptr_t page;
1207
1208 if (!page)
1209 page = sysconf (_SC_PAGESIZE);
1210#endif
1211
1212 addr &= ~(page - 1); /* assume page size is always a power of two */
1213
1214 if (addr < end)
1215 if (flags) /* modify */
1216 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1217 else
1218 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1219
1220 return 0;
991} 1221}
992 1222
993/*****************************************************************************/ 1223/*****************************************************************************/
994 1224
995#define ALLOC(len) \ 1225#define ALLOC(len) \
1082 1312
1083/*****************************************************************************/ 1313/*****************************************************************************/
1084 1314
1085int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1315int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1086{ 1316{
1087 etp_init (want_poll, done_poll); 1317 return etp_init (want_poll, done_poll);
1088} 1318}
1089 1319
1090static void eio_api_destroy (eio_req *req) 1320static void eio_api_destroy (eio_req *req)
1091{ 1321{
1092 free (req); 1322 free (req);
1154 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1384 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1155 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1385 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1156 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1386 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1157 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1387 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1158 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1388 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1159 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 1389 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1160 1390
1161 case EIO_READLINK: ALLOC (NAME_MAX); 1391 case EIO_READLINK: ALLOC (NAME_MAX);
1162 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1392 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1163 1393
1164 case EIO_SYNC: req->result = 0; sync (); break; 1394 case EIO_SYNC: req->result = 0; sync (); break;
1165 case EIO_FSYNC: req->result = fsync (req->int1); break; 1395 case EIO_FSYNC: req->result = fsync (req->int1); break;
1166 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1396 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1397 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1398 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1399 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1167 1400
1168 case EIO_READDIR: eio__scandir (req, self); break; 1401 case EIO_READDIR: eio__scandir (req, self); break;
1169 1402
1170 case EIO_BUSY: 1403 case EIO_BUSY:
1171#ifdef _WIN32 1404#ifdef _WIN32
1203 1436
1204 req->result = req->type == EIO_FUTIME 1437 req->result = req->type == EIO_FUTIME
1205 ? futimes (req->int1, times) 1438 ? futimes (req->int1, times)
1206 : utimes (req->ptr1, times); 1439 : utimes (req->ptr1, times);
1207 } 1440 }
1441 break;
1208 1442
1209 case EIO_GROUP: 1443 case EIO_GROUP:
1444 abort (); /* handled in eio_request */
1445
1210 case EIO_NOP: 1446 case EIO_NOP:
1211 req->result = 0; 1447 req->result = 0;
1212 break; 1448 break;
1213 1449
1214 case EIO_CUSTOM: 1450 case EIO_CUSTOM:
1215 req->feed (req); 1451 ((void (*)(eio_req *))req->feed) (req);
1216 break; 1452 break;
1217 1453
1218 default: 1454 default:
1219 req->result = -1; 1455 req->result = -1;
1220 break; 1456 break;
1243eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1479eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1244{ 1480{
1245 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1481 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1246} 1482}
1247 1483
1484eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1485{
1486 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1487}
1488
1489eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1490{
1491 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1492}
1493
1494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1495{
1496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1497}
1498
1248eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1499eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1249{ 1500{
1250 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1501 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1251} 1502}
1252 1503
1364eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1615eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1365{ 1616{
1366 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1617 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1367} 1618}
1368 1619
1369eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1620eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1370{ 1621{
1371 return eio__1path (EIO_READDIR, path, pri, cb, data); 1622 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1372} 1623}
1373 1624
1374eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1625eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1375{ 1626{
1376 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1627 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1377} 1628}
1378 1629
1379static eio_req * 1630static eio_req *
1380eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1631eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1381{ 1632{
1407 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1658 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1408} 1659}
1409 1660
1410eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 1661eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1411{ 1662{
1412 REQ (EIO_CUSTOM); req->feed = execute; SEND; 1663 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1413} 1664}
1414 1665
1415#endif 1666#endif
1416 1667
1417eio_req *eio_grp (eio_cb cb, void *data) 1668eio_req *eio_grp (eio_cb cb, void *data)
1445 1696
1446void eio_grp_add (eio_req *grp, eio_req *req) 1697void eio_grp_add (eio_req *grp, eio_req *req)
1447{ 1698{
1448 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1699 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1449 1700
1701 grp->flags |= EIO_FLAG_GROUPADD;
1702
1450 ++grp->size; 1703 ++grp->size;
1451 req->grp = grp; 1704 req->grp = grp;
1452 1705
1453 req->grp_prev = 0; 1706 req->grp_prev = 0;
1454 req->grp_next = grp->grp_first; 1707 req->grp_next = grp->grp_first;

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines