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 1.38 by root, Fri Jun 12 20:01:42 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>
78# include <sys/mman.h>
75# include <unistd.h> 79# include <unistd.h>
76# include <utime.h> 80# include <utime.h>
77# include <signal.h> 81# include <signal.h>
78# include <dirent.h> 82# include <dirent.h>
83
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen
89# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino
91# endif
92
93#ifdef _D_EXACT_NAMLEN
94# undef D_NAMLEN
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
97
98# ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type
100# endif
79 101
80# ifndef EIO_STRUCT_DIRENT 102# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 103# define EIO_STRUCT_DIRENT struct dirent
82# endif 104# endif
83 105
96# else 118# else
97# error sendfile support requested but not available 119# error sendfile support requested but not available
98# endif 120# endif
99#endif 121#endif
100 122
123#ifndef D_TYPE
124# define D_TYPE(de) 0
125#endif
126#ifndef D_INO
127# define D_INO(de) 0
128#endif
129#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name)
131#endif
132
101/* number of seconds after which an idle threads exit */ 133/* number of seconds after which an idle threads exit */
102#define IDLE_TIMEOUT 10 134#define IDLE_TIMEOUT 10
103 135
104/* used for struct dirent, AIX doesn't provide it */ 136/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 137#ifndef NAME_MAX
117 errno = ENOMEM; \ 149 errno = ENOMEM; \
118 if (!eio_buf) \ 150 if (!eio_buf) \
119 return -1; 151 return -1;
120 152
121#define EIO_TICKS ((1000000 + 1023) >> 10) 153#define EIO_TICKS ((1000000 + 1023) >> 10)
154
155/*****************************************************************************/
156
157#if __GNUC__ >= 3
158# define expect(expr,value) __builtin_expect ((expr),(value))
159#else
160# define expect(expr,value) (expr)
161#endif
162
163#define expect_false(expr) expect ((expr) != 0, 0)
164#define expect_true(expr) expect ((expr) != 0, 1)
122 165
123/*****************************************************************************/ 166/*****************************************************************************/
124 167
125#define ETP_PRI_MIN EIO_PRI_MIN 168#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 169#define ETP_PRI_MAX EIO_PRI_MAX
144 if (wrk->dirp) \ 187 if (wrk->dirp) \
145 { \ 188 { \
146 closedir (wrk->dirp); \ 189 closedir (wrk->dirp); \
147 wrk->dirp = 0; \ 190 wrk->dirp = 0; \
148 } 191 }
192
149#define ETP_WORKER_COMMON \ 193#define ETP_WORKER_COMMON \
150 void *dbuf; \ 194 void *dbuf; \
151 DIR *dirp; 195 DIR *dirp;
152 196
153/*****************************************************************************/ 197/*****************************************************************************/
176 220
177static mutex_t wrklock = X_MUTEX_INIT; 221static mutex_t wrklock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT; 222static mutex_t reslock = X_MUTEX_INIT;
179static mutex_t reqlock = X_MUTEX_INIT; 223static mutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT; 224static cond_t reqwait = X_COND_INIT;
225
226#if !HAVE_PREADWRITE
227/*
228 * make our pread/pwrite emulation safe against themselves, but not against
229 * normal read/write by using a mutex. slows down execution a lot,
230 * but that's your problem, not mine.
231 */
232static mutex_t preadwritelock = X_MUTEX_INIT;
233#endif
181 234
182typedef struct etp_worker 235typedef struct etp_worker
183{ 236{
184 /* locked by wrklock */ 237 /* locked by wrklock */
185 struct etp_worker *prev, *next; 238 struct etp_worker *prev, *next;
330 383
331static void etp_atfork_child (void) 384static void etp_atfork_child (void)
332{ 385{
333 ETP_REQ *prv; 386 ETP_REQ *prv;
334 387
335 while (prv = reqq_shift (&req_queue)) 388 while ((prv = reqq_shift (&req_queue)))
336 ETP_DESTROY (prv); 389 ETP_DESTROY (prv);
337 390
338 while (prv = reqq_shift (&res_queue)) 391 while ((prv = reqq_shift (&res_queue)))
339 ETP_DESTROY (prv); 392 ETP_DESTROY (prv);
340 393
341 while (wrk_first.next != &wrk_first) 394 while (wrk_first.next != &wrk_first)
342 { 395 {
343 etp_worker *wrk = wrk_first.next; 396 etp_worker *wrk = wrk_first.next;
371 424
372 pthread_once (&doinit, etp_once_init); 425 pthread_once (&doinit, etp_once_init);
373 426
374 want_poll_cb = want_poll; 427 want_poll_cb = want_poll;
375 done_poll_cb = done_poll; 428 done_poll_cb = done_poll;
429
430 return 0;
376} 431}
377 432
378X_THREAD_PROC (etp_proc); 433X_THREAD_PROC (etp_proc);
379 434
380static void etp_start_thread (void) 435static void etp_start_thread (void)
400 X_UNLOCK (wrklock); 455 X_UNLOCK (wrklock);
401} 456}
402 457
403static void etp_maybe_start_thread (void) 458static void etp_maybe_start_thread (void)
404{ 459{
405 if (etp_nthreads () >= wanted) 460 if (expect_true (etp_nthreads () >= wanted))
406 return; 461 return;
407 462
408 /* todo: maybe use idle here, but might be less exact */ 463 /* todo: maybe use idle here, but might be less exact */
409 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 464 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
410 return; 465 return;
411 466
412 etp_start_thread (); 467 etp_start_thread ();
413} 468}
414 469
467 522
468 X_LOCK (reqlock); 523 X_LOCK (reqlock);
469 --nreqs; 524 --nreqs;
470 X_UNLOCK (reqlock); 525 X_UNLOCK (reqlock);
471 526
472 if (req->type == EIO_GROUP && req->size) 527 if (expect_false (req->type == EIO_GROUP && req->size))
473 { 528 {
474 req->int1 = 1; /* mark request as delayed */ 529 req->int1 = 1; /* mark request as delayed */
475 continue; 530 continue;
476 } 531 }
477 else 532 else
478 { 533 {
479 int res = ETP_FINISH (req); 534 int res = ETP_FINISH (req);
480 if (res) 535 if (expect_false (res))
481 return res; 536 return res;
482 } 537 }
483 538
484 if (maxreqs && !--maxreqs) 539 if (expect_false (maxreqs && !--maxreqs))
485 break; 540 break;
486 541
487 if (maxtime) 542 if (maxtime)
488 { 543 {
489 gettimeofday (&tv_now, 0); 544 gettimeofday (&tv_now, 0);
508 563
509static void etp_submit (ETP_REQ *req) 564static void etp_submit (ETP_REQ *req)
510{ 565{
511 req->pri -= ETP_PRI_MIN; 566 req->pri -= ETP_PRI_MIN;
512 567
513 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 568 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; 569 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
515 570
571 if (expect_false (req->type == EIO_GROUP))
572 {
573 /* I hope this is worth it :/ */
516 X_LOCK (reqlock); 574 X_LOCK (reqlock);
517 ++nreqs; 575 ++nreqs;
576 X_UNLOCK (reqlock);
577
578 X_LOCK (reslock);
579
580 ++npending;
581
582 if (!reqq_push (&res_queue, req) && want_poll_cb)
583 want_poll_cb ();
584
585 X_UNLOCK (reslock);
586 }
587 else
588 {
589 X_LOCK (reqlock);
590 ++nreqs;
518 ++nready; 591 ++nready;
519 reqq_push (&req_queue, req); 592 reqq_push (&req_queue, req);
520 X_COND_SIGNAL (reqwait); 593 X_COND_SIGNAL (reqwait);
521 X_UNLOCK (reqlock); 594 X_UNLOCK (reqlock);
522 595
523 etp_maybe_start_thread (); 596 etp_maybe_start_thread ();
597 }
524} 598}
525 599
526static void etp_set_max_poll_time (double nseconds) 600static void etp_set_max_poll_time (double nseconds)
527{ 601{
528 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 602 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
563 637
564static void grp_try_feed (eio_req *grp) 638static void grp_try_feed (eio_req *grp)
565{ 639{
566 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
567 { 641 {
568 int old_len = grp->size; 642 grp->flags &= ~EIO_FLAG_GROUPADD;
569 643
570 EIO_FEED (grp); 644 EIO_FEED (grp);
571 645
572 /* stop if no progress has been made */ 646 /* stop if no progress has been made */
573 if (old_len == grp->size) 647 if (!(grp->flags & EIO_FLAG_GROUPADD))
574 { 648 {
575 grp->feed = 0; 649 grp->feed = 0;
576 break; 650 break;
577 } 651 }
578 } 652 }
695 769
696/*****************************************************************************/ 770/*****************************************************************************/
697/* work around various missing functions */ 771/* work around various missing functions */
698 772
699#if !HAVE_PREADWRITE 773#if !HAVE_PREADWRITE
774# undef pread
775# undef pwrite
700# define pread eio__pread 776# define pread eio__pread
701# define pwrite eio__pwrite 777# 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;
709 778
710static ssize_t 779static ssize_t
711eio__pread (int fd, void *buf, size_t count, off_t offset) 780eio__pread (int fd, void *buf, size_t count, off_t offset)
712{ 781{
713 ssize_t res; 782 ssize_t res;
731 800
732 X_LOCK (preadwritelock); 801 X_LOCK (preadwritelock);
733 ooffset = lseek (fd, 0, SEEK_CUR); 802 ooffset = lseek (fd, 0, SEEK_CUR);
734 lseek (fd, offset, SEEK_SET); 803 lseek (fd, offset, SEEK_SET);
735 res = write (fd, buf, count); 804 res = write (fd, buf, count);
736 lseek (fd, offset, SEEK_SET); 805 lseek (fd, ooffset, SEEK_SET);
737 X_UNLOCK (preadwritelock); 806 X_UNLOCK (preadwritelock);
738 807
739 return res; 808 return res;
740} 809}
741#endif 810#endif
742 811
743#ifndef HAVE_FUTIMES 812#ifndef HAVE_FUTIMES
744 813
814# undef utimes
815# undef futimes
745# define utimes(path,times) eio__utimes (path, times) 816# define utimes(path,times) eio__utimes (path, times)
746# define futimes(fd,times) eio__futimes (fd, times) 817# define futimes(fd,times) eio__futimes (fd, times)
747 818
748static int 819static int
749eio__utimes (const char *filename, const struct timeval times[2]) 820eio__utimes (const char *filename, const struct timeval times[2])
768} 839}
769 840
770#endif 841#endif
771 842
772#if !HAVE_FDATASYNC 843#if !HAVE_FDATASYNC
844# undef fdatasync
773# define fdatasync fsync 845# define fdatasync(fd) fsync (fd)
774#endif 846#endif
847
848/* sync_file_range always needs emulation */
849int
850eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851{
852#if HAVE_SYNC_FILE_RANGE
853 int res;
854
855 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
856 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
857 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
858 {
859 flags = 0
860 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
861 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
862 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
863 }
864
865 res = sync_file_range (fd, offset, nbytes, flags);
866
867 if (!res || errno != ENOSYS)
868 return res;
869#endif
870
871 /* even though we could play tricks with the flags, it's better to always
872 * call fdatasync, as thta matches the expectation of it's users best */
873 return fdatasync (fd);
874}
775 875
776#if !HAVE_READAHEAD 876#if !HAVE_READAHEAD
877# undef readahead
777# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
778 879
779static ssize_t 880static ssize_t
780eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 881eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
781{ 882{
889 } 990 }
890 991
891 return res; 992 return res;
892} 993}
893 994
995static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{
998 return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */
999 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1000}
1001
1002#define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */
1003
1004static void
1005eio_dent_sort (eio_dirent *dents, int size)
1006{
1007 int i, j;
1008
1009 if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */
1010 {
1011 /* first, use quicksort */
1012 /* should be good for 2**31 entries */
1013 struct rng { int l, r; } rng [32];
1014
1015 i = 0;
1016 rng[0].l = 0;
1017 rng[0].r = size;
1018
1019 while (expect_true (i >= 0))
1020 {
1021 int L = rng [i].l;
1022 int R = rng [i].r - 1;
1023
1024 if (expect_false (L + EIO_QSORT_CUTOFF < R))
1025 {
1026 eio_dirent piv = dents [L];
1027
1028 while (L < R)
1029 {
1030 while (eio_dent_cmp (&dents [R], &piv) >= 0 && L < R)
1031 --R;
1032
1033 if (L < R)
1034 dents [L++] = dents [R];
1035
1036 while (eio_dent_cmp (&dents [L], &piv) <= 0 && L < R)
1037 ++L;
1038
1039 if (L < R)
1040 dents [R--] = dents [L];
1041 }
1042
1043 dents [L] = piv;
1044
1045 ++i;
1046 rng [i].l = L + 1;
1047 rng [i].r = rng [i - 1].r;
1048 rng [i - 1].r = L;
1049
1050 if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1051 {
1052 struct rng t;
1053
1054 t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1055 }
1056 }
1057 else
1058 --i;
1059 }
1060 }
1061
1062 /* use a simple insertion sort at the end */
1063 for (i = 1; i < size; ++i)
1064 {
1065 eio_dirent value = dents [i];
1066
1067 for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j)
1068 dents [j + 1] = dents [j];
1069
1070 dents [j + 1] = value;
1071 }
1072}
1073
894/* read a full directory */ 1074/* read a full directory */
895static void 1075static void
896eio__scandir (eio_req *req, etp_worker *self) 1076eio__scandir (eio_req *req, etp_worker *self)
897{ 1077{
898 DIR *dirp; 1078 DIR *dirp;
899 EIO_STRUCT_DIRENT *entp; 1079 EIO_STRUCT_DIRENT *entp;
900 char *name, *names; 1080 char *name, *names;
901 int memlen = 4096; 1081 int namesalloc = 4096;
902 int memofs = 0; 1082 int namesoffs = 0;
1083 int flags = req->int1;
1084 eio_dirent *dents = 0;
1085 int dentalloc = 128;
903 int res = 0; 1086 int dentoffs = 0;
1087
1088 req->result = -1;
1089
1090 if (!(flags & EIO_READDIR_DENTS))
1091 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
904 1092
905 X_LOCK (wrklock); 1093 X_LOCK (wrklock);
1094 /* the corresponding closedir is in ETP_WORKER_CLEAR */
906 self->dirp = dirp = opendir (req->ptr1); 1095 self->dirp = dirp = opendir (req->ptr1);
907 req->flags |= EIO_FLAG_PTR2_FREE; 1096 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
908 req->ptr2 = names = malloc (memlen); 1097 req->ptr1 = names = malloc (namesalloc);
1098 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
909 X_UNLOCK (wrklock); 1099 X_UNLOCK (wrklock);
910 1100
911 if (dirp && names) 1101 if (dirp && names && (!flags || dents))
912 for (;;) 1102 for (;;)
913 { 1103 {
914 errno = 0; 1104 errno = 0;
915 entp = readdir (dirp); 1105 entp = readdir (dirp);
916 1106
917 if (!entp) 1107 if (!entp)
1108 {
1109 if (errno)
1110 break;
1111
1112 /* sort etc. */
1113 req->int1 = flags;
1114 req->result = dentoffs;
1115
1116 if (dents)
1117 {
1118 eio_dirent *ent = dents + dentoffs;
1119
1120 while (ent > dents)
1121 (--ent)->name = names + (size_t)ent->name;
1122 }
1123
1124 if (flags & EIO_READDIR_STAT_ORDER
1125 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
1126 eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1127 else if (flags & EIO_READDIR_DIRS_FIRST)
1128 {
1129 /* in this case, all is known, and we just put dirs first and sort them */
1130 eio_dirent *ent = dents + dentoffs;
1131 eio_dirent *dir = dents;
1132
1133 /* now move dirs to the front, and non-dirs to the back */
1134 /* by walking from both sides and swapping if necessary */
1135 while (ent > dir)
1136 {
1137 if (dir->type == DT_DIR)
1138 ++dir;
1139 else
1140 {
1141 --ent;
1142
1143 if (ent->type == DT_DIR)
1144 {
1145 eio_dirent tmp = *dir;
1146 *dir = *ent;
1147 *ent = tmp;
1148
1149 ++dir;
1150 }
1151 }
1152 }
1153
1154 /* now sort the dirs only */
1155 eio_dent_sort (dents, dir - dents);
1156 }
1157
1158 /* only provide the names array unless DENTS is specified */
1159 if (!(flags & EIO_READDIR_DENTS))
1160 {
1161 X_LOCK (wrklock);
1162 assert (!dents);
1163 req->ptr1 = 0;
1164 req->ptr2 = names;
1165 X_UNLOCK (wrklock);
1166 }
1167
918 break; 1168 break;
1169 }
919 1170
1171 /* now add the entry to our list(s) */
920 name = entp->d_name; 1172 name = entp->d_name;
921 1173
1174 /* skip . and .. entries */
922 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1175 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
923 { 1176 {
924 int len = strlen (name) + 1; 1177 int len = D_NAMLEN (entp) + 1;
925 1178
926 res++; 1179 while (expect_false (namesoffs + len > namesalloc))
927
928 while (memofs + len > memlen)
929 { 1180 {
930 memlen *= 2; 1181 namesalloc *= 2;
931 X_LOCK (wrklock); 1182 X_LOCK (wrklock);
932 req->ptr2 = names = realloc (names, memlen); 1183 req->ptr1 = names = realloc (names, namesalloc);
933 X_UNLOCK (wrklock); 1184 X_UNLOCK (wrklock);
934 1185
935 if (!names) 1186 if (!names)
936 break; 1187 break;
937 } 1188 }
938 1189
939 memcpy (names + memofs, name, len); 1190 memcpy (names + namesoffs, name, len);
1191
1192 if (dents)
1193 {
1194 struct eio_dirent *ent;
1195
1196 if (expect_false (dentoffs == dentalloc))
1197 {
1198 dentalloc *= 2;
1199 X_LOCK (wrklock);
1200 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1201 X_UNLOCK (wrklock);
1202
1203 if (!dents)
1204 break;
1205 }
1206
1207 ent = dents + dentoffs;
1208
1209 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
1210 ent->namelen = len - 1;
1211 ent->inode = D_INO (entp);
1212
1213 switch (D_TYPE (entp))
1214 {
1215 default:
1216 ent->type = EIO_DT_UNKNOWN;
1217 flags |= EIO_READDIR_FOUND_UNKNOWN;
1218 break;
1219
1220 #ifdef DT_FIFO
1221 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1222 #endif
1223 #ifdef DT_CHR
1224 case DT_CHR: ent->type = EIO_DT_CHR; break;
1225 #endif
1226 #ifdef DT_MPC
1227 case DT_MPC: ent->type = EIO_DT_MPC; break;
1228 #endif
1229 #ifdef DT_DIR
1230 case DT_DIR: ent->type = EIO_DT_DIR; break;
1231 #endif
1232 #ifdef DT_NAM
1233 case DT_NAM: ent->type = EIO_DT_NAM; break;
1234 #endif
1235 #ifdef DT_BLK
1236 case DT_BLK: ent->type = EIO_DT_BLK; break;
1237 #endif
1238 #ifdef DT_MPB
1239 case DT_MPB: ent->type = EIO_DT_MPB; break;
1240 #endif
1241 #ifdef DT_REG
1242 case DT_REG: ent->type = EIO_DT_REG; break;
1243 #endif
1244 #ifdef DT_NWK
1245 case DT_NWK: ent->type = EIO_DT_NWK; break;
1246 #endif
1247 #ifdef DT_CMP
1248 case DT_CMP: ent->type = EIO_DT_CMP; break;
1249 #endif
1250 #ifdef DT_LNK
1251 case DT_LNK: ent->type = EIO_DT_LNK; break;
1252 #endif
1253 #ifdef DT_SOCK
1254 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1255 #endif
1256 #ifdef DT_DOOR
1257 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1258 #endif
1259 #ifdef DT_WHT
1260 case DT_WHT: ent->type = EIO_DT_WHT; break;
1261 #endif
1262 }
1263
1264 ent->score = 0;
1265
1266 if (flags & EIO_READDIR_DIRS_FIRST)
1267 {
1268 if (ent->type == EIO_DT_UNKNOWN)
1269 {
1270 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1271 ent->score = 98;
1272 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1273 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
1274 }
1275 else if (ent->type == EIO_DT_DIR)
1276 ent->score = 100;
1277 }
1278 }
1279
940 memofs += len; 1280 namesoffs += len;
1281 ++dentoffs;
1282 }
1283
1284 if (EIO_CANCELLED (req))
1285 {
1286 errno = ECANCELED;
1287 break;
941 } 1288 }
942 } 1289 }
1290}
943 1291
944 if (errno) 1292#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
945 res = -1; 1293# undef msync
946 1294# define msync(a,b,c) ((errno = ENOSYS), -1)
947 req->result = res; 1295#endif
1296
1297int
1298eio__mtouch (void *mem, size_t len, int flags)
1299{
1300 intptr_t addr = (intptr_t)mem;
1301 intptr_t end = addr + len;
1302#ifdef PAGESIZE
1303 const intptr_t page = PAGESIZE;
1304#else
1305 static intptr_t page;
1306
1307 if (!page)
1308 page = sysconf (_SC_PAGESIZE);
1309#endif
1310
1311 addr &= ~(page - 1); /* assume page size is always a power of two */
1312
1313 if (addr < end)
1314 if (flags) /* modify */
1315 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1316 else
1317 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1318
1319 return 0;
948} 1320}
949 1321
950/*****************************************************************************/ 1322/*****************************************************************************/
951 1323
952#define ALLOC(len) \ 1324#define ALLOC(len) \
1039 1411
1040/*****************************************************************************/ 1412/*****************************************************************************/
1041 1413
1042int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1414int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1043{ 1415{
1044 etp_init (want_poll, done_poll); 1416 return etp_init (want_poll, done_poll);
1045} 1417}
1046 1418
1047static void eio_api_destroy (eio_req *req) 1419static void eio_api_destroy (eio_req *req)
1048{ 1420{
1049 free (req); 1421 free (req);
1111 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1483 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1112 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1484 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; 1485 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1114 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1486 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1115 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1487 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; 1488 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1117 1489
1118 case EIO_READLINK: ALLOC (NAME_MAX); 1490 case EIO_READLINK: ALLOC (NAME_MAX);
1119 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1491 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1120 1492
1121 case EIO_SYNC: req->result = 0; sync (); break; 1493 case EIO_SYNC: req->result = 0; sync (); break;
1122 case EIO_FSYNC: req->result = fsync (req->int1); break; 1494 case EIO_FSYNC: req->result = fsync (req->int1); break;
1123 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1495 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1496 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1497 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1498 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1124 1499
1125 case EIO_READDIR: eio__scandir (req, self); break; 1500 case EIO_READDIR: eio__scandir (req, self); break;
1126 1501
1127 case EIO_BUSY: 1502 case EIO_BUSY:
1128#ifdef _WIN32 1503#ifdef _WIN32
1160 1535
1161 req->result = req->type == EIO_FUTIME 1536 req->result = req->type == EIO_FUTIME
1162 ? futimes (req->int1, times) 1537 ? futimes (req->int1, times)
1163 : utimes (req->ptr1, times); 1538 : utimes (req->ptr1, times);
1164 } 1539 }
1540 break;
1165 1541
1166 case EIO_GROUP: 1542 case EIO_GROUP:
1543 abort (); /* handled in eio_request */
1544
1167 case EIO_NOP: 1545 case EIO_NOP:
1168 req->result = 0; 1546 req->result = 0;
1169 break; 1547 break;
1170 1548
1171 case EIO_CUSTOM: 1549 case EIO_CUSTOM:
1172 req->feed (req); 1550 ((void (*)(eio_req *))req->feed) (req);
1173 break; 1551 break;
1174 1552
1175 default: 1553 default:
1176 req->result = -1; 1554 req->result = -1;
1177 break; 1555 break;
1200eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1578eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1201{ 1579{
1202 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1580 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1203} 1581}
1204 1582
1583eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1584{
1585 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1586}
1587
1588eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1589{
1590 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1591}
1592
1593eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1594{
1595 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1596}
1597
1205eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1598eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1206{ 1599{
1207 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1600 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1208} 1601}
1209 1602
1321eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1714eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1322{ 1715{
1323 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1716 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1324} 1717}
1325 1718
1326eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1719eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1327{ 1720{
1328 return eio__1path (EIO_READDIR, path, pri, cb, data); 1721 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1329} 1722}
1330 1723
1331eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1724eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1332{ 1725{
1333 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1726 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1334} 1727}
1335 1728
1336static eio_req * 1729static eio_req *
1337eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1730eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1338{ 1731{
1364 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1757 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1365} 1758}
1366 1759
1367eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 1760eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1368{ 1761{
1369 REQ (EIO_CUSTOM); req->feed = execute; SEND; 1762 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1370} 1763}
1371 1764
1372#endif 1765#endif
1373 1766
1374eio_req *eio_grp (eio_cb cb, void *data) 1767eio_req *eio_grp (eio_cb cb, void *data)
1402 1795
1403void eio_grp_add (eio_req *grp, eio_req *req) 1796void eio_grp_add (eio_req *grp, eio_req *req)
1404{ 1797{
1405 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1798 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1406 1799
1800 grp->flags |= EIO_FLAG_GROUPADD;
1801
1407 ++grp->size; 1802 ++grp->size;
1408 req->grp = grp; 1803 req->grp = grp;
1409 1804
1410 req->grp_prev = 0; 1805 req->grp_prev = 0;
1411 req->grp_next = grp->grp_first; 1806 req->grp_next = grp->grp_first;
1420/* misc garbage */ 1815/* misc garbage */
1421 1816
1422ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1817ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1423{ 1818{
1424 etp_worker wrk; 1819 etp_worker wrk;
1820 ssize_t ret;
1425 1821
1426 wrk.dbuf = 0; 1822 wrk.dbuf = 0;
1427 1823
1428 eio__sendfile (ofd, ifd, offset, count, &wrk); 1824 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1429 1825
1430 if (wrk.dbuf) 1826 if (wrk.dbuf)
1431 free (wrk.dbuf); 1827 free (wrk.dbuf);
1432}
1433 1828
1829 return ret;
1830}
1831

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines