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

Comparing libeio/eio.c (file contents):
Revision 1.17 by root, Tue Jun 3 05:12:51 2008 UTC vs.
Revision 1.33 by root, Sat Jun 6 19:44:17 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 _DIRENT_HAVE_D_TYPE /* sigh */
86# define D_INO(de) (de)->d_fileno
87# define D_NAMLEN(de) (de)->d_namlen
88# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
89# define D_INO(de) (de)->d_ino
90# endif
91
92#ifdef _D_EXACT_NAMLEN
93# undef D_NAMLEN
94# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95#endif
96
97# ifdef _DIRENT_HAVE_D_TYPE
98# define D_TYPE(de) (de)->d_type
99# endif
79 100
80# ifndef EIO_STRUCT_DIRENT 101# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 102# define EIO_STRUCT_DIRENT struct dirent
82# endif 103# endif
83 104
96# else 117# else
97# error sendfile support requested but not available 118# error sendfile support requested but not available
98# endif 119# endif
99#endif 120#endif
100 121
122#ifndef D_TYPE
123# define D_TYPE(de) 0
124#endif
125#ifndef D_INO
126# define D_INO(de) 0
127#endif
128#ifndef D_NAMLEN
129# define D_NAMLEN(de) strlen ((de)->d_name)
130#endif
131
101/* number of seconds after which an idle threads exit */ 132/* number of seconds after which an idle threads exit */
102#define IDLE_TIMEOUT 10 133#define IDLE_TIMEOUT 10
103 134
104/* used for struct dirent, AIX doesn't provide it */ 135/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 136#ifndef NAME_MAX
117 errno = ENOMEM; \ 148 errno = ENOMEM; \
118 if (!eio_buf) \ 149 if (!eio_buf) \
119 return -1; 150 return -1;
120 151
121#define EIO_TICKS ((1000000 + 1023) >> 10) 152#define EIO_TICKS ((1000000 + 1023) >> 10)
153
154/*****************************************************************************/
155
156#if __GNUC__ >= 3
157# define expect(expr,value) __builtin_expect ((expr),(value))
158#else
159# define expect(expr,value) (expr)
160#endif
161
162#define expect_false(expr) expect ((expr) != 0, 0)
163#define expect_true(expr) expect ((expr) != 0, 1)
122 164
123/*****************************************************************************/ 165/*****************************************************************************/
124 166
125#define ETP_PRI_MIN EIO_PRI_MIN 167#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 168#define ETP_PRI_MAX EIO_PRI_MAX
144 if (wrk->dirp) \ 186 if (wrk->dirp) \
145 { \ 187 { \
146 closedir (wrk->dirp); \ 188 closedir (wrk->dirp); \
147 wrk->dirp = 0; \ 189 wrk->dirp = 0; \
148 } 190 }
191
149#define ETP_WORKER_COMMON \ 192#define ETP_WORKER_COMMON \
150 void *dbuf; \ 193 void *dbuf; \
151 DIR *dirp; 194 DIR *dirp;
152 195
153/*****************************************************************************/ 196/*****************************************************************************/
176 219
177static mutex_t wrklock = X_MUTEX_INIT; 220static mutex_t wrklock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT; 221static mutex_t reslock = X_MUTEX_INIT;
179static mutex_t reqlock = X_MUTEX_INIT; 222static mutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT; 223static cond_t reqwait = X_COND_INIT;
224
225#if !HAVE_PREADWRITE
226/*
227 * make our pread/pwrite emulation safe against themselves, but not against
228 * normal read/write by using a mutex. slows down execution a lot,
229 * but that's your problem, not mine.
230 */
231static mutex_t preadwritelock = X_MUTEX_INIT;
232#endif
181 233
182typedef struct etp_worker 234typedef struct etp_worker
183{ 235{
184 /* locked by wrklock */ 236 /* locked by wrklock */
185 struct etp_worker *prev, *next; 237 struct etp_worker *prev, *next;
330 382
331static void etp_atfork_child (void) 383static void etp_atfork_child (void)
332{ 384{
333 ETP_REQ *prv; 385 ETP_REQ *prv;
334 386
335 while (prv = reqq_shift (&req_queue)) 387 while ((prv = reqq_shift (&req_queue)))
336 ETP_DESTROY (prv); 388 ETP_DESTROY (prv);
337 389
338 while (prv = reqq_shift (&res_queue)) 390 while ((prv = reqq_shift (&res_queue)))
339 ETP_DESTROY (prv); 391 ETP_DESTROY (prv);
340 392
341 while (wrk_first.next != &wrk_first) 393 while (wrk_first.next != &wrk_first)
342 { 394 {
343 etp_worker *wrk = wrk_first.next; 395 etp_worker *wrk = wrk_first.next;
371 423
372 pthread_once (&doinit, etp_once_init); 424 pthread_once (&doinit, etp_once_init);
373 425
374 want_poll_cb = want_poll; 426 want_poll_cb = want_poll;
375 done_poll_cb = done_poll; 427 done_poll_cb = done_poll;
428
429 return 0;
376} 430}
377 431
378X_THREAD_PROC (etp_proc); 432X_THREAD_PROC (etp_proc);
379 433
380static void etp_start_thread (void) 434static void etp_start_thread (void)
400 X_UNLOCK (wrklock); 454 X_UNLOCK (wrklock);
401} 455}
402 456
403static void etp_maybe_start_thread (void) 457static void etp_maybe_start_thread (void)
404{ 458{
405 if (etp_nthreads () >= wanted) 459 if (expect_true (etp_nthreads () >= wanted))
406 return; 460 return;
407 461
408 /* todo: maybe use idle here, but might be less exact */ 462 /* todo: maybe use idle here, but might be less exact */
409 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 463 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
410 return; 464 return;
411 465
412 etp_start_thread (); 466 etp_start_thread ();
413} 467}
414 468
467 521
468 X_LOCK (reqlock); 522 X_LOCK (reqlock);
469 --nreqs; 523 --nreqs;
470 X_UNLOCK (reqlock); 524 X_UNLOCK (reqlock);
471 525
472 if (req->type == EIO_GROUP && req->size) 526 if (expect_false (req->type == EIO_GROUP && req->size))
473 { 527 {
474 req->int1 = 1; /* mark request as delayed */ 528 req->int1 = 1; /* mark request as delayed */
475 continue; 529 continue;
476 } 530 }
477 else 531 else
478 { 532 {
479 int res = ETP_FINISH (req); 533 int res = ETP_FINISH (req);
480 if (res) 534 if (expect_false (res))
481 return res; 535 return res;
482 } 536 }
483 537
484 if (maxreqs && !--maxreqs) 538 if (expect_false (maxreqs && !--maxreqs))
485 break; 539 break;
486 540
487 if (maxtime) 541 if (maxtime)
488 { 542 {
489 gettimeofday (&tv_now, 0); 543 gettimeofday (&tv_now, 0);
508 562
509static void etp_submit (ETP_REQ *req) 563static void etp_submit (ETP_REQ *req)
510{ 564{
511 req->pri -= ETP_PRI_MIN; 565 req->pri -= ETP_PRI_MIN;
512 566
513 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 567 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; 568 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
515 569
570 if (expect_false (req->type == EIO_GROUP))
571 {
572 /* I hope this is worth it :/ */
516 X_LOCK (reqlock); 573 X_LOCK (reqlock);
517 ++nreqs; 574 ++nreqs;
575 X_UNLOCK (reqlock);
576
577 X_LOCK (reslock);
578
579 ++npending;
580
581 if (!reqq_push (&res_queue, req) && want_poll_cb)
582 want_poll_cb ();
583
584 X_UNLOCK (reslock);
585 }
586 else
587 {
588 X_LOCK (reqlock);
589 ++nreqs;
518 ++nready; 590 ++nready;
519 reqq_push (&req_queue, req); 591 reqq_push (&req_queue, req);
520 X_COND_SIGNAL (reqwait); 592 X_COND_SIGNAL (reqwait);
521 X_UNLOCK (reqlock); 593 X_UNLOCK (reqlock);
522 594
523 etp_maybe_start_thread (); 595 etp_maybe_start_thread ();
596 }
524} 597}
525 598
526static void etp_set_max_poll_time (double nseconds) 599static void etp_set_max_poll_time (double nseconds)
527{ 600{
528 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 601 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
563 636
564static void grp_try_feed (eio_req *grp) 637static void grp_try_feed (eio_req *grp)
565{ 638{
566 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 639 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
567 { 640 {
568 int old_len = grp->size; 641 grp->flags &= ~EIO_FLAG_GROUPADD;
569 642
570 EIO_FEED (grp); 643 EIO_FEED (grp);
571 644
572 /* stop if no progress has been made */ 645 /* stop if no progress has been made */
573 if (old_len == grp->size) 646 if (!(grp->flags & EIO_FLAG_GROUPADD))
574 { 647 {
575 grp->feed = 0; 648 grp->feed = 0;
576 break; 649 break;
577 } 650 }
578 } 651 }
695 768
696/*****************************************************************************/ 769/*****************************************************************************/
697/* work around various missing functions */ 770/* work around various missing functions */
698 771
699#if !HAVE_PREADWRITE 772#if !HAVE_PREADWRITE
773# undef pread
774# undef pwrite
700# define pread eio__pread 775# define pread eio__pread
701# define pwrite eio__pwrite 776# 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;
709 777
710static ssize_t 778static ssize_t
711eio__pread (int fd, void *buf, size_t count, off_t offset) 779eio__pread (int fd, void *buf, size_t count, off_t offset)
712{ 780{
713 ssize_t res; 781 ssize_t res;
731 799
732 X_LOCK (preadwritelock); 800 X_LOCK (preadwritelock);
733 ooffset = lseek (fd, 0, SEEK_CUR); 801 ooffset = lseek (fd, 0, SEEK_CUR);
734 lseek (fd, offset, SEEK_SET); 802 lseek (fd, offset, SEEK_SET);
735 res = write (fd, buf, count); 803 res = write (fd, buf, count);
736 lseek (fd, offset, SEEK_SET); 804 lseek (fd, ooffset, SEEK_SET);
737 X_UNLOCK (preadwritelock); 805 X_UNLOCK (preadwritelock);
738 806
739 return res; 807 return res;
740} 808}
741#endif 809#endif
742 810
743#ifndef HAVE_FUTIMES 811#ifndef HAVE_FUTIMES
744 812
813# undef utimes
814# undef futimes
745# define utimes(path,times) eio__utimes (path, times) 815# define utimes(path,times) eio__utimes (path, times)
746# define futimes(fd,times) eio__futimes (fd, times) 816# define futimes(fd,times) eio__futimes (fd, times)
747 817
748static int 818static int
749eio__utimes (const char *filename, const struct timeval times[2]) 819eio__utimes (const char *filename, const struct timeval times[2])
768} 838}
769 839
770#endif 840#endif
771 841
772#if !HAVE_FDATASYNC 842#if !HAVE_FDATASYNC
843# undef fdatasync
773# define fdatasync fsync 844# define fdatasync(fd) fsync (fd)
774#endif 845#endif
846
847/* sync_file_range always needs emulation */
848int
849eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850{
851#if HAVE_SYNC_FILE_RANGE
852 int res;
853
854 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
855 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
856 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
857 {
858 flags = 0
859 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
860 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
861 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
862 }
863
864 res = sync_file_range (fd, offset, nbytes, flags);
865
866 if (!res || errno != ENOSYS)
867 return res;
868#endif
869
870 /* even though we could play tricks with the flags, it's better to always
871 * call fdatasync, as thta matches the expectation of it's users best */
872 return fdatasync (fd);
873}
775 874
776#if !HAVE_READAHEAD 875#if !HAVE_READAHEAD
876# undef readahead
777# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 877# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
778 878
779static ssize_t 879static ssize_t
780eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 880eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
781{ 881{
889 } 989 }
890 990
891 return res; 991 return res;
892} 992}
893 993
994static int
995eio_dent_cmp (const void *a_, const void *b_)
996{
997 const eio_dirent *a = (const eio_dirent *)a_;
998 const eio_dirent *b = (const eio_dirent *)b_;
999
1000 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
1001 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */
1002}
1003
894/* read a full directory */ 1004/* read a full directory */
895static void 1005static void
896eio__scandir (eio_req *req, etp_worker *self) 1006eio__scandir (eio_req *req, etp_worker *self)
897{ 1007{
898 DIR *dirp; 1008 DIR *dirp;
899 EIO_STRUCT_DIRENT *entp; 1009 EIO_STRUCT_DIRENT *entp;
900 char *name, *names; 1010 unsigned char *name, *names;
901 int memlen = 4096; 1011 int namesalloc = 4096;
902 int memofs = 0; 1012 int namesoffs = 0;
1013 int flags = req->int1;
1014 eio_dirent *dents = 0;
1015 int dentalloc = 128;
903 int res = 0; 1016 int dentoffs = 0;
1017
1018 req->result = -1;
1019
1020 if (!(flags & EIO_READDIR_DENTS))
1021 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
904 1022
905 X_LOCK (wrklock); 1023 X_LOCK (wrklock);
1024 /* the corresponding closedir is in ETP_WORKER_CLEAR */
906 self->dirp = dirp = opendir (req->ptr1); 1025 self->dirp = dirp = opendir (req->ptr1);
907 req->flags |= EIO_FLAG_PTR2_FREE; 1026 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
908 req->ptr2 = names = malloc (memlen); 1027 req->ptr1 = names = malloc (namesalloc);
1028 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
909 X_UNLOCK (wrklock); 1029 X_UNLOCK (wrklock);
910 1030
911 if (dirp && names) 1031 if (dirp && names && (!flags || dents))
912 for (;;) 1032 for (;;)
913 { 1033 {
914 errno = 0; 1034 errno = 0;
915 entp = readdir (dirp); 1035 entp = readdir (dirp);
916 1036
917 if (!entp) 1037 if (!entp)
1038 {
1039 if (errno)
1040 break;
1041
1042 /* sort etc. */
1043 req->int1 = flags;
1044 req->result = dentoffs;
1045
1046 if (dents)
1047 {
1048 eio_dirent *ent = dents + dentoffs;
1049
1050 while (ent > dents)
1051 (--ent)->name = names + (size_t)ent->name;
1052 }
1053
1054 if (flags & EIO_READDIR_STAT_ORDER
1055 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
1056 {
1057 /* pray your qsort doesn't use quicksort */
1058 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059 }
1060 else if (flags & EIO_READDIR_DIRS_FIRST)
1061 {
1062 /* in this case, all is known, and we just put dirs first and sort them */
1063 eio_dirent *ent = dents + dentoffs;
1064 eio_dirent *dir = dents;
1065
1066 while (ent > dir)
1067 {
1068 if (dir->type == DT_DIR)
1069 ++dir;
1070 else
1071 {
1072 --ent;
1073
1074 if (ent->type == DT_DIR)
1075 {
1076 eio_dirent tmp = *dir;
1077 *dir = *ent;
1078 *ent = tmp;
1079
1080 ++dir;
1081 }
1082 }
1083 }
1084
1085 /* now sort the dirs only */
1086 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1087 }
1088
1089 /* only provide the names array unless DENTS is specified */
1090 if (!(flags & EIO_READDIR_DENTS))
1091 {
1092 X_LOCK (wrklock);
1093 assert (!dents);
1094 req->ptr1 = 0;
1095 req->ptr2 = names;
1096 X_UNLOCK (wrklock);
1097 }
1098
918 break; 1099 break;
1100 }
919 1101
1102 /* now add the entry to our list(s) */
920 name = entp->d_name; 1103 name = entp->d_name;
921 1104
1105 /* skip . and .. entries */
922 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1106 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
923 { 1107 {
924 int len = strlen (name) + 1; 1108 int len = D_NAMLEN (entp) + 1;
925 1109
926 res++; 1110 while (expect_false (namesoffs + len > namesalloc))
927
928 while (memofs + len > memlen)
929 { 1111 {
930 memlen *= 2; 1112 namesalloc *= 2;
931 X_LOCK (wrklock); 1113 X_LOCK (wrklock);
932 req->ptr2 = names = realloc (names, memlen); 1114 req->ptr1 = names = realloc (names, namesalloc);
933 X_UNLOCK (wrklock); 1115 X_UNLOCK (wrklock);
934 1116
935 if (!names) 1117 if (!names)
936 break; 1118 break;
937 } 1119 }
938 1120
939 memcpy (names + memofs, name, len); 1121 memcpy (names + namesoffs, name, len);
1122
1123 if (dents)
1124 {
1125 struct eio_dirent *ent;
1126
1127 if (expect_false (dentoffs == dentalloc))
1128 {
1129 dentalloc *= 2;
1130 X_LOCK (wrklock);
1131 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132 X_UNLOCK (wrklock);
1133
1134 if (!dents)
1135 break;
1136 }
1137
1138 ent = dents + dentoffs;
1139
1140 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
1141 ent->namelen = len - 1;
1142 ent->inode = D_INO (entp);
1143
1144 switch (D_TYPE (entp))
1145 {
1146 default:
1147 ent->type = EIO_DT_UNKNOWN;
1148 flags |= EIO_READDIR_FOUND_UNKNOWN;
1149 break;
1150
1151 #ifdef DT_FIFO
1152 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1153 #endif
1154 #ifdef DT_CHR
1155 case DT_CHR: ent->type = EIO_DT_CHR; break;
1156 #endif
1157 #ifdef DT_MPC
1158 case DT_MPC: ent->type = EIO_DT_MPC; break;
1159 #endif
1160 #ifdef DT_DIR
1161 case DT_DIR: ent->type = EIO_DT_DIR; break;
1162 #endif
1163 #ifdef DT_NAM
1164 case DT_NAM: ent->type = EIO_DT_NAM; break;
1165 #endif
1166 #ifdef DT_BLK
1167 case DT_BLK: ent->type = EIO_DT_BLK; break;
1168 #endif
1169 #ifdef DT_MPB
1170 case DT_MPB: ent->type = EIO_DT_MPB; break;
1171 #endif
1172 #ifdef DT_REG
1173 case DT_REG: ent->type = EIO_DT_REG; break;
1174 #endif
1175 #ifdef DT_NWK
1176 case DT_NWK: ent->type = EIO_DT_NWK; break;
1177 #endif
1178 #ifdef DT_CMP
1179 case DT_CMP: ent->type = EIO_DT_CMP; break;
1180 #endif
1181 #ifdef DT_LNK
1182 case DT_LNK: ent->type = EIO_DT_LNK; break;
1183 #endif
1184 #ifdef DT_SOCK
1185 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1186 #endif
1187 #ifdef DT_DOOR
1188 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1189 #endif
1190 #ifdef DT_WHT
1191 case DT_WHT: ent->type = EIO_DT_WHT; break;
1192 #endif
1193 }
1194
1195 ent->score = 0;
1196
1197 if (flags & EIO_READDIR_DIRS_FIRST)
1198 {
1199 if (ent->type == EIO_DT_UNKNOWN)
1200 {
1201 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202 ent->score = 98;
1203 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1204 ent->score = len <= 2 ? len + 60 : len <= 4 ? 50 : len <= 7 ? 40 : 10; /* shorter == more likely dir, but avoid too many classes */
1205 }
1206 else if (ent->type == EIO_DT_DIR)
1207 ent->score = 100;
1208 }
1209 }
1210
940 memofs += len; 1211 namesoffs += len;
1212 ++dentoffs;
941 } 1213 }
942 } 1214 }
943 1215 else
944 if (errno)
945 res = -1;
946
947 req->result = res; 1216 req->result = -1;
1217}
1218
1219#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1220# undef msync
1221# define msync(a,b,c) ((errno = ENOSYS), -1)
1222#endif
1223
1224int
1225eio__mtouch (void *mem, size_t len, int flags)
1226{
1227 intptr_t addr = (intptr_t)mem;
1228 intptr_t end = addr + len;
1229#ifdef PAGESIZE
1230 const intptr_t page = PAGESIZE;
1231#else
1232 static intptr_t page;
1233
1234 if (!page)
1235 page = sysconf (_SC_PAGESIZE);
1236#endif
1237
1238 addr &= ~(page - 1); /* assume page size is always a power of two */
1239
1240 if (addr < end)
1241 if (flags) /* modify */
1242 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1243 else
1244 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1245
1246 return 0;
948} 1247}
949 1248
950/*****************************************************************************/ 1249/*****************************************************************************/
951 1250
952#define ALLOC(len) \ 1251#define ALLOC(len) \
1039 1338
1040/*****************************************************************************/ 1339/*****************************************************************************/
1041 1340
1042int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1341int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1043{ 1342{
1044 etp_init (want_poll, done_poll); 1343 return etp_init (want_poll, done_poll);
1045} 1344}
1046 1345
1047static void eio_api_destroy (eio_req *req) 1346static void eio_api_destroy (eio_req *req)
1048{ 1347{
1049 free (req); 1348 free (req);
1111 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1410 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1112 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1411 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1113 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1412 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1114 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1413 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1115 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1414 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; 1415 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1117 1416
1118 case EIO_READLINK: ALLOC (NAME_MAX); 1417 case EIO_READLINK: ALLOC (NAME_MAX);
1119 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1418 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1120 1419
1121 case EIO_SYNC: req->result = 0; sync (); break; 1420 case EIO_SYNC: req->result = 0; sync (); break;
1122 case EIO_FSYNC: req->result = fsync (req->int1); break; 1421 case EIO_FSYNC: req->result = fsync (req->int1); break;
1123 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1422 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1424 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1425 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1124 1426
1125 case EIO_READDIR: eio__scandir (req, self); break; 1427 case EIO_READDIR: eio__scandir (req, self); break;
1126 1428
1127 case EIO_BUSY: 1429 case EIO_BUSY:
1128#ifdef _WIN32 1430#ifdef _WIN32
1160 1462
1161 req->result = req->type == EIO_FUTIME 1463 req->result = req->type == EIO_FUTIME
1162 ? futimes (req->int1, times) 1464 ? futimes (req->int1, times)
1163 : utimes (req->ptr1, times); 1465 : utimes (req->ptr1, times);
1164 } 1466 }
1467 break;
1165 1468
1166 case EIO_GROUP: 1469 case EIO_GROUP:
1470 abort (); /* handled in eio_request */
1471
1167 case EIO_NOP: 1472 case EIO_NOP:
1168 req->result = 0; 1473 req->result = 0;
1169 break; 1474 break;
1170 1475
1171 case EIO_CUSTOM: 1476 case EIO_CUSTOM:
1172 req->feed (req); 1477 ((void (*)(eio_req *))req->feed) (req);
1173 break; 1478 break;
1174 1479
1175 default: 1480 default:
1176 req->result = -1; 1481 req->result = -1;
1177 break; 1482 break;
1200eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1505eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1201{ 1506{
1202 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1507 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1203} 1508}
1204 1509
1510eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1511{
1512 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1513}
1514
1515eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516{
1517 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518}
1519
1520eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1521{
1522 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1523}
1524
1205eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1525eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1206{ 1526{
1207 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1527 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1208} 1528}
1209 1529
1321eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1641eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1322{ 1642{
1323 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1643 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1324} 1644}
1325 1645
1326eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1646eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1327{ 1647{
1328 return eio__1path (EIO_READDIR, path, pri, cb, data); 1648 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1329} 1649}
1330 1650
1331eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1651eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1332{ 1652{
1333 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1653 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1334} 1654}
1335 1655
1336static eio_req * 1656static eio_req *
1337eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1657eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1338{ 1658{
1364 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1684 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1365} 1685}
1366 1686
1367eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 1687eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1368{ 1688{
1369 REQ (EIO_CUSTOM); req->feed = execute; SEND; 1689 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1370} 1690}
1371 1691
1372#endif 1692#endif
1373 1693
1374eio_req *eio_grp (eio_cb cb, void *data) 1694eio_req *eio_grp (eio_cb cb, void *data)
1402 1722
1403void eio_grp_add (eio_req *grp, eio_req *req) 1723void eio_grp_add (eio_req *grp, eio_req *req)
1404{ 1724{
1405 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1725 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1406 1726
1727 grp->flags |= EIO_FLAG_GROUPADD;
1728
1407 ++grp->size; 1729 ++grp->size;
1408 req->grp = grp; 1730 req->grp = grp;
1409 1731
1410 req->grp_prev = 0; 1732 req->grp_prev = 0;
1411 req->grp_next = grp->grp_first; 1733 req->grp_next = grp->grp_first;

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines