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 1.19 by root, Tue Jun 17 23:34:20 2008 UTC vs.
Revision 1.41 by root, Sun Jun 14 20:36:59 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;
402 X_UNLOCK (wrklock); 455 X_UNLOCK (wrklock);
403} 456}
404 457
405static void etp_maybe_start_thread (void) 458static void etp_maybe_start_thread (void)
406{ 459{
407 if (etp_nthreads () >= wanted) 460 if (expect_true (etp_nthreads () >= wanted))
408 return; 461 return;
409 462
410 /* todo: maybe use idle here, but might be less exact */ 463 /* todo: maybe use idle here, but might be less exact */
411 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 ()))
412 return; 465 return;
413 466
414 etp_start_thread (); 467 etp_start_thread ();
415} 468}
416 469
469 522
470 X_LOCK (reqlock); 523 X_LOCK (reqlock);
471 --nreqs; 524 --nreqs;
472 X_UNLOCK (reqlock); 525 X_UNLOCK (reqlock);
473 526
474 if (req->type == EIO_GROUP && req->size) 527 if (expect_false (req->type == EIO_GROUP && req->size))
475 { 528 {
476 req->int1 = 1; /* mark request as delayed */ 529 req->int1 = 1; /* mark request as delayed */
477 continue; 530 continue;
478 } 531 }
479 else 532 else
480 { 533 {
481 int res = ETP_FINISH (req); 534 int res = ETP_FINISH (req);
482 if (res) 535 if (expect_false (res))
483 return res; 536 return res;
484 } 537 }
485 538
486 if (maxreqs && !--maxreqs) 539 if (expect_false (maxreqs && !--maxreqs))
487 break; 540 break;
488 541
489 if (maxtime) 542 if (maxtime)
490 { 543 {
491 gettimeofday (&tv_now, 0); 544 gettimeofday (&tv_now, 0);
510 563
511static void etp_submit (ETP_REQ *req) 564static void etp_submit (ETP_REQ *req)
512{ 565{
513 req->pri -= ETP_PRI_MIN; 566 req->pri -= ETP_PRI_MIN;
514 567
515 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;
516 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;
517 570
571 if (expect_false (req->type == EIO_GROUP))
572 {
573 /* I hope this is worth it :/ */
518 X_LOCK (reqlock); 574 X_LOCK (reqlock);
519 ++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;
520 ++nready; 591 ++nready;
521 reqq_push (&req_queue, req); 592 reqq_push (&req_queue, req);
522 X_COND_SIGNAL (reqwait); 593 X_COND_SIGNAL (reqwait);
523 X_UNLOCK (reqlock); 594 X_UNLOCK (reqlock);
524 595
525 etp_maybe_start_thread (); 596 etp_maybe_start_thread ();
597 }
526} 598}
527 599
528static void etp_set_max_poll_time (double nseconds) 600static void etp_set_max_poll_time (double nseconds)
529{ 601{
530 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 602 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
565 637
566static void grp_try_feed (eio_req *grp) 638static void grp_try_feed (eio_req *grp)
567{ 639{
568 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569 { 641 {
570 int old_len = grp->size; 642 grp->flags &= ~EIO_FLAG_GROUPADD;
571 643
572 EIO_FEED (grp); 644 EIO_FEED (grp);
573 645
574 /* stop if no progress has been made */ 646 /* stop if no progress has been made */
575 if (old_len == grp->size) 647 if (!(grp->flags & EIO_FLAG_GROUPADD))
576 { 648 {
577 grp->feed = 0; 649 grp->feed = 0;
578 break; 650 break;
579 } 651 }
580 } 652 }
697 769
698/*****************************************************************************/ 770/*****************************************************************************/
699/* work around various missing functions */ 771/* work around various missing functions */
700 772
701#if !HAVE_PREADWRITE 773#if !HAVE_PREADWRITE
774# undef pread
775# undef pwrite
702# define pread eio__pread 776# define pread eio__pread
703# define pwrite eio__pwrite 777# define pwrite eio__pwrite
704
705/*
706 * make our pread/pwrite safe against themselves, but not against
707 * normal read/write by using a mutex. slows down execution a lot,
708 * but that's your problem, not mine.
709 */
710static mutex_t preadwritelock = X_MUTEX_INIT;
711 778
712static ssize_t 779static ssize_t
713eio__pread (int fd, void *buf, size_t count, off_t offset) 780eio__pread (int fd, void *buf, size_t count, off_t offset)
714{ 781{
715 ssize_t res; 782 ssize_t res;
733 800
734 X_LOCK (preadwritelock); 801 X_LOCK (preadwritelock);
735 ooffset = lseek (fd, 0, SEEK_CUR); 802 ooffset = lseek (fd, 0, SEEK_CUR);
736 lseek (fd, offset, SEEK_SET); 803 lseek (fd, offset, SEEK_SET);
737 res = write (fd, buf, count); 804 res = write (fd, buf, count);
738 lseek (fd, offset, SEEK_SET); 805 lseek (fd, ooffset, SEEK_SET);
739 X_UNLOCK (preadwritelock); 806 X_UNLOCK (preadwritelock);
740 807
741 return res; 808 return res;
742} 809}
743#endif 810#endif
744 811
745#ifndef HAVE_FUTIMES 812#ifndef HAVE_FUTIMES
746 813
814# undef utimes
815# undef futimes
747# define utimes(path,times) eio__utimes (path, times) 816# define utimes(path,times) eio__utimes (path, times)
748# define futimes(fd,times) eio__futimes (fd, times) 817# define futimes(fd,times) eio__futimes (fd, times)
749 818
750static int 819static int
751eio__utimes (const char *filename, const struct timeval times[2]) 820eio__utimes (const char *filename, const struct timeval times[2])
770} 839}
771 840
772#endif 841#endif
773 842
774#if !HAVE_FDATASYNC 843#if !HAVE_FDATASYNC
844# undef fdatasync
775# define fdatasync fsync 845# define fdatasync(fd) fsync (fd)
776#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}
777 875
778#if !HAVE_READAHEAD 876#if !HAVE_READAHEAD
877# undef readahead
779# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780 879
781static ssize_t 880static ssize_t
782eio__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)
783{ 882{
891 } 990 }
892 991
893 return res; 992 return res;
894} 993}
895 994
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_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1003#define EIO_DENT_NUM_SCORES 9
1004
1005#define EIO_QSORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1006#define EIO_QSORT_SKIP 60 /* when to skip qsort completely */
1007
1008static void
1009eio_dent_sort (eio_dirent *dents, int size)
1010{
1011 if (size <= 1)
1012 return; /* our insertion sort relies on size > 0 */
1013
1014 if (size > EIO_QSORT_SKIP) /* skip quicksort for small directories */
1015 {
1016 /* first, use quicksort */
1017 /* should be good for 2**31 entries */
1018 int i;
1019 struct rng { eio_dirent *l, *r; } rng [32];
1020
1021 i = 0;
1022 rng[0].l = dents;
1023 rng[0].r = dents + size;
1024
1025 while (expect_true (i >= 0))
1026 {
1027 eio_dirent *L = rng [i].l;
1028 eio_dirent *R = rng [i].r - 1;
1029
1030 if (expect_false (L + EIO_QSORT_CUTOFF < R))
1031 {
1032 eio_dirent *mid = &dents [((L - dents) + (R - dents)) >> 1];
1033 eio_dirent piv = *mid; *mid = *L; *L = piv;
1034
1035 while (L < R)
1036 {
1037 while (EIO_DENT_CMP (*R, >=, piv) && L < R)
1038 --R;
1039
1040 if (L < R)
1041 *L++ = *R;
1042
1043 while (EIO_DENT_CMP (*L, <=, piv) && L < R)
1044 ++L;
1045
1046 if (L < R)
1047 *R-- = *L;
1048 }
1049
1050 *L = piv;
1051
1052 ++i;
1053 rng [i].l = L + 1;
1054 rng [i].r = rng [i - 1].r;
1055 rng [i - 1].r = L;
1056
1057 if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1058 {
1059 struct rng t;
1060
1061 t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1062 }
1063 }
1064 else
1065 --i;
1066 }
1067 }
1068
1069 /* use an insertion sort after qsort, or for small arrays */
1070 /* first move the smallest element to the front, to act as a sentinel */
1071 {
1072 int i;
1073 eio_dirent *min = dents;
1074
1075 for (i = size > EIO_QSORT_SKIP ? EIO_QSORT_CUTOFF + 1 : size; --i; )
1076 if (EIO_DENT_CMP (dents [i], <, *min))
1077 min = &dents [i];
1078
1079 /* swap elements 0 and j (minimum) */
1080 {
1081 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1082 }
1083 }
1084
1085 {
1086 /* then do standard insertion sort */
1087 eio_dirent *i, *j;
1088
1089 for (i = dents + 1; i < dents + size; ++i)
1090 {
1091 eio_dirent value = *i;
1092
1093 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1094 j [1] = j [0];
1095
1096 j [1] = value;
1097 }
1098 }
1099}
1100
896/* read a full directory */ 1101/* read a full directory */
897static void 1102static void
898eio__scandir (eio_req *req, etp_worker *self) 1103eio__scandir (eio_req *req, etp_worker *self)
899{ 1104{
900 DIR *dirp; 1105 DIR *dirp;
901 EIO_STRUCT_DIRENT *entp; 1106 EIO_STRUCT_DIRENT *entp;
902 char *name, *names; 1107 char *name, *names;
903 int memlen = 4096; 1108 int namesalloc = 4096;
904 int memofs = 0; 1109 int namesoffs = 0;
1110 int flags = req->int1;
1111 eio_dirent *dents = 0;
1112 int dentalloc = 128;
905 int res = 0; 1113 int dentoffs = 0;
1114
1115 req->result = -1;
1116
1117 if (!(flags & EIO_READDIR_DENTS))
1118 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
906 1119
907 X_LOCK (wrklock); 1120 X_LOCK (wrklock);
1121 /* the corresponding closedir is in ETP_WORKER_CLEAR */
908 self->dirp = dirp = opendir (req->ptr1); 1122 self->dirp = dirp = opendir (req->ptr1);
909 req->flags |= EIO_FLAG_PTR2_FREE; 1123 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
910 req->ptr2 = names = malloc (memlen); 1124 req->ptr1 = names = malloc (namesalloc);
1125 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
911 X_UNLOCK (wrklock); 1126 X_UNLOCK (wrklock);
912 1127
913 if (dirp && names) 1128 if (dirp && names && (!flags || dents))
914 for (;;) 1129 for (;;)
915 { 1130 {
916 errno = 0; 1131 errno = 0;
917 entp = readdir (dirp); 1132 entp = readdir (dirp);
918 1133
919 if (!entp) 1134 if (!entp)
1135 {
1136 if (errno)
1137 break;
1138
1139 /* sort etc. */
1140 req->int1 = flags;
1141 req->result = dentoffs;
1142
1143 if (dents)
1144 {
1145 eio_dirent *ent = dents + dentoffs;
1146
1147 while (ent > dents)
1148 (--ent)->name = names + (size_t)ent->name;
1149 }
1150
1151 if (flags & EIO_READDIR_STAT_ORDER
1152 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
1153 eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1154 else if (flags & EIO_READDIR_DIRS_FIRST)
1155 {
1156 /* in this case, all is known, and we just put dirs first and sort them */
1157 eio_dirent *ent = dents + dentoffs;
1158 eio_dirent *dir = dents;
1159
1160 /* now move dirs to the front, and non-dirs to the back */
1161 /* by walking from both sides and swapping if necessary */
1162 while (ent > dir)
1163 {
1164 if (dir->type == DT_DIR)
1165 {
1166 dir->score = 0;
1167 ++dir;
1168 }
1169 else
1170 {
1171 --ent;
1172
1173 if (ent->type == DT_DIR)
1174 {
1175 eio_dirent tmp = *dir;
1176 *dir = *ent;
1177 *ent = tmp;
1178
1179 dir->score = 0;
1180 ++dir;
1181 }
1182 }
1183 }
1184
1185 /* now sort the dirs only */
1186 eio_dent_sort (dents, dir - dents);
1187 }
1188
1189 /* only provide the names array unless DENTS is specified */
1190 if (!(flags & EIO_READDIR_DENTS))
1191 {
1192 X_LOCK (wrklock);
1193 assert (!dents);
1194 req->ptr1 = 0;
1195 req->ptr2 = names;
1196 X_UNLOCK (wrklock);
1197 }
1198
920 break; 1199 break;
1200 }
921 1201
1202 /* now add the entry to our list(s) */
922 name = entp->d_name; 1203 name = entp->d_name;
923 1204
1205 /* skip . and .. entries */
924 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1206 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
925 { 1207 {
926 int len = strlen (name) + 1; 1208 int len = D_NAMLEN (entp) + 1;
927 1209
928 res++; 1210 while (expect_false (namesoffs + len > namesalloc))
929
930 while (memofs + len > memlen)
931 { 1211 {
932 memlen *= 2; 1212 namesalloc *= 2;
933 X_LOCK (wrklock); 1213 X_LOCK (wrklock);
934 req->ptr2 = names = realloc (names, memlen); 1214 req->ptr1 = names = realloc (names, namesalloc);
935 X_UNLOCK (wrklock); 1215 X_UNLOCK (wrklock);
936 1216
937 if (!names) 1217 if (!names)
938 break; 1218 break;
939 } 1219 }
940 1220
941 memcpy (names + memofs, name, len); 1221 memcpy (names + namesoffs, name, len);
1222
1223 if (dents)
1224 {
1225 struct eio_dirent *ent;
1226
1227 if (expect_false (dentoffs == dentalloc))
1228 {
1229 dentalloc *= 2;
1230 X_LOCK (wrklock);
1231 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1232 X_UNLOCK (wrklock);
1233
1234 if (!dents)
1235 break;
1236 }
1237
1238 ent = dents + dentoffs;
1239
1240 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
1241 ent->namelen = len - 1;
1242 ent->inode = D_INO (entp) ? D_INO (entp) : dentoffs;
1243
1244 switch (D_TYPE (entp))
1245 {
1246 default:
1247 ent->type = EIO_DT_UNKNOWN;
1248 flags |= EIO_READDIR_FOUND_UNKNOWN;
1249 break;
1250
1251 #ifdef DT_FIFO
1252 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1253 #endif
1254 #ifdef DT_CHR
1255 case DT_CHR: ent->type = EIO_DT_CHR; break;
1256 #endif
1257 #ifdef DT_MPC
1258 case DT_MPC: ent->type = EIO_DT_MPC; break;
1259 #endif
1260 #ifdef DT_DIR
1261 case DT_DIR: ent->type = EIO_DT_DIR; break;
1262 #endif
1263 #ifdef DT_NAM
1264 case DT_NAM: ent->type = EIO_DT_NAM; break;
1265 #endif
1266 #ifdef DT_BLK
1267 case DT_BLK: ent->type = EIO_DT_BLK; break;
1268 #endif
1269 #ifdef DT_MPB
1270 case DT_MPB: ent->type = EIO_DT_MPB; break;
1271 #endif
1272 #ifdef DT_REG
1273 case DT_REG: ent->type = EIO_DT_REG; break;
1274 #endif
1275 #ifdef DT_NWK
1276 case DT_NWK: ent->type = EIO_DT_NWK; break;
1277 #endif
1278 #ifdef DT_CMP
1279 case DT_CMP: ent->type = EIO_DT_CMP; break;
1280 #endif
1281 #ifdef DT_LNK
1282 case DT_LNK: ent->type = EIO_DT_LNK; break;
1283 #endif
1284 #ifdef DT_SOCK
1285 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1286 #endif
1287 #ifdef DT_DOOR
1288 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1289 #endif
1290 #ifdef DT_WHT
1291 case DT_WHT: ent->type = EIO_DT_WHT; break;
1292 #endif
1293 }
1294
1295 ent->score = 0;
1296
1297 if (flags & EIO_READDIR_DIRS_FIRST)
1298 {
1299 if (ent->type == EIO_DT_UNKNOWN)
1300 {
1301 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1302 ent->score = 7;
1303 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1304 ent->score = len <= 2 ? len + 4 : len <= 4 ? 3 : len <= 7 ? 2 : 1; /* shorter == more likely dir, but avoid too many classes */
1305 }
1306 else if (ent->type == EIO_DT_DIR)
1307 ent->score = 8;
1308 }
1309 }
1310
942 memofs += len; 1311 namesoffs += len;
1312 ++dentoffs;
1313 }
1314
1315 if (EIO_CANCELLED (req))
1316 {
1317 errno = ECANCELED;
1318 break;
943 } 1319 }
944 } 1320 }
1321}
945 1322
946 if (errno) 1323#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
947 res = -1; 1324# undef msync
948 1325# define msync(a,b,c) ((errno = ENOSYS), -1)
949 req->result = res; 1326#endif
1327
1328int
1329eio__mtouch (void *mem, size_t len, int flags)
1330{
1331 intptr_t addr = (intptr_t)mem;
1332 intptr_t end = addr + len;
1333#ifdef PAGESIZE
1334 const intptr_t page = PAGESIZE;
1335#else
1336 static intptr_t page;
1337
1338 if (!page)
1339 page = sysconf (_SC_PAGESIZE);
1340#endif
1341
1342 addr &= ~(page - 1); /* assume page size is always a power of two */
1343
1344 if (addr < end)
1345 if (flags) /* modify */
1346 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1347 else
1348 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1349
1350 return 0;
950} 1351}
951 1352
952/*****************************************************************************/ 1353/*****************************************************************************/
953 1354
954#define ALLOC(len) \ 1355#define ALLOC(len) \
1121 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1522 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1122 1523
1123 case EIO_SYNC: req->result = 0; sync (); break; 1524 case EIO_SYNC: req->result = 0; sync (); break;
1124 case EIO_FSYNC: req->result = fsync (req->int1); break; 1525 case EIO_FSYNC: req->result = fsync (req->int1); break;
1125 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1526 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1527 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1528 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1529 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1126 1530
1127 case EIO_READDIR: eio__scandir (req, self); break; 1531 case EIO_READDIR: eio__scandir (req, self); break;
1128 1532
1129 case EIO_BUSY: 1533 case EIO_BUSY:
1130#ifdef _WIN32 1534#ifdef _WIN32
1162 1566
1163 req->result = req->type == EIO_FUTIME 1567 req->result = req->type == EIO_FUTIME
1164 ? futimes (req->int1, times) 1568 ? futimes (req->int1, times)
1165 : utimes (req->ptr1, times); 1569 : utimes (req->ptr1, times);
1166 } 1570 }
1571 break;
1167 1572
1168 case EIO_GROUP: 1573 case EIO_GROUP:
1574 abort (); /* handled in eio_request */
1575
1169 case EIO_NOP: 1576 case EIO_NOP:
1170 req->result = 0; 1577 req->result = 0;
1171 break; 1578 break;
1172 1579
1173 case EIO_CUSTOM: 1580 case EIO_CUSTOM:
1174 req->feed (req); 1581 ((void (*)(eio_req *))req->feed) (req);
1175 break; 1582 break;
1176 1583
1177 default: 1584 default:
1178 req->result = -1; 1585 req->result = -1;
1179 break; 1586 break;
1202eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1609eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1203{ 1610{
1204 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1611 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1205} 1612}
1206 1613
1614eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1615{
1616 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1617}
1618
1619eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1620{
1621 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1622}
1623
1624eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1625{
1626 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1627}
1628
1207eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1629eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1208{ 1630{
1209 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1631 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1210} 1632}
1211 1633
1323eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1745eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1324{ 1746{
1325 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1747 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1326} 1748}
1327 1749
1328eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1750eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1329{ 1751{
1330 return eio__1path (EIO_READDIR, path, pri, cb, data); 1752 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1331} 1753}
1332 1754
1333eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1755eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1334{ 1756{
1335 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1757 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1366 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1788 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1367} 1789}
1368 1790
1369eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 1791eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1370{ 1792{
1371 REQ (EIO_CUSTOM); req->feed = execute; SEND; 1793 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1372} 1794}
1373 1795
1374#endif 1796#endif
1375 1797
1376eio_req *eio_grp (eio_cb cb, void *data) 1798eio_req *eio_grp (eio_cb cb, void *data)
1404 1826
1405void eio_grp_add (eio_req *grp, eio_req *req) 1827void eio_grp_add (eio_req *grp, eio_req *req)
1406{ 1828{
1407 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1829 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1408 1830
1831 grp->flags |= EIO_FLAG_GROUPADD;
1832
1409 ++grp->size; 1833 ++grp->size;
1410 req->grp = grp; 1834 req->grp = grp;
1411 1835
1412 req->grp_prev = 0; 1836 req->grp_prev = 0;
1413 req->grp_next = grp->grp_first; 1837 req->grp_next = grp->grp_first;
1422/* misc garbage */ 1846/* misc garbage */
1423 1847
1424ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1848ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1425{ 1849{
1426 etp_worker wrk; 1850 etp_worker wrk;
1851 ssize_t ret;
1427 1852
1428 wrk.dbuf = 0; 1853 wrk.dbuf = 0;
1429 1854
1430 eio__sendfile (ofd, ifd, offset, count, &wrk); 1855 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1431 1856
1432 if (wrk.dbuf) 1857 if (wrk.dbuf)
1433 free (wrk.dbuf); 1858 free (wrk.dbuf);
1434}
1435 1859
1860 return ret;
1861}
1862

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines