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 1.37 by root, Fri Jun 12 16:48:08 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;
314 X_LOCK (reqlock); 367 X_LOCK (reqlock);
315 X_LOCK (reslock); 368 X_LOCK (reslock);
316#if !HAVE_PREADWRITE 369#if !HAVE_PREADWRITE
317 X_LOCK (preadwritelock); 370 X_LOCK (preadwritelock);
318#endif 371#endif
319#if !HAVE_READDIR_R
320 X_LOCK (readdirlock);
321#endif
322} 372}
323 373
324static void etp_atfork_parent (void) 374static void etp_atfork_parent (void)
325{ 375{
326#if !HAVE_READDIR_R
327 X_UNLOCK (readdirlock);
328#endif
329#if !HAVE_PREADWRITE 376#if !HAVE_PREADWRITE
330 X_UNLOCK (preadwritelock); 377 X_UNLOCK (preadwritelock);
331#endif 378#endif
332 X_UNLOCK (reslock); 379 X_UNLOCK (reslock);
333 X_UNLOCK (reqlock); 380 X_UNLOCK (reqlock);
336 383
337static void etp_atfork_child (void) 384static void etp_atfork_child (void)
338{ 385{
339 ETP_REQ *prv; 386 ETP_REQ *prv;
340 387
341 while (prv = reqq_shift (&req_queue)) 388 while ((prv = reqq_shift (&req_queue)))
342 ETP_DESTROY (prv); 389 ETP_DESTROY (prv);
343 390
344 while (prv = reqq_shift (&res_queue)) 391 while ((prv = reqq_shift (&res_queue)))
345 ETP_DESTROY (prv); 392 ETP_DESTROY (prv);
346 393
347 while (wrk_first.next != &wrk_first) 394 while (wrk_first.next != &wrk_first)
348 { 395 {
349 etp_worker *wrk = wrk_first.next; 396 etp_worker *wrk = wrk_first.next;
377 424
378 pthread_once (&doinit, etp_once_init); 425 pthread_once (&doinit, etp_once_init);
379 426
380 want_poll_cb = want_poll; 427 want_poll_cb = want_poll;
381 done_poll_cb = done_poll; 428 done_poll_cb = done_poll;
429
430 return 0;
382} 431}
383 432
384X_THREAD_PROC (etp_proc); 433X_THREAD_PROC (etp_proc);
385 434
386static void etp_start_thread (void) 435static void etp_start_thread (void)
406 X_UNLOCK (wrklock); 455 X_UNLOCK (wrklock);
407} 456}
408 457
409static void etp_maybe_start_thread (void) 458static void etp_maybe_start_thread (void)
410{ 459{
411 if (etp_nthreads () >= wanted) 460 if (expect_true (etp_nthreads () >= wanted))
412 return; 461 return;
413 462
414 /* todo: maybe use idle here, but might be less exact */ 463 /* todo: maybe use idle here, but might be less exact */
415 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 ()))
416 return; 465 return;
417 466
418 etp_start_thread (); 467 etp_start_thread ();
419} 468}
420 469
473 522
474 X_LOCK (reqlock); 523 X_LOCK (reqlock);
475 --nreqs; 524 --nreqs;
476 X_UNLOCK (reqlock); 525 X_UNLOCK (reqlock);
477 526
478 if (req->type == EIO_GROUP && req->size) 527 if (expect_false (req->type == EIO_GROUP && req->size))
479 { 528 {
480 req->int1 = 1; /* mark request as delayed */ 529 req->int1 = 1; /* mark request as delayed */
481 continue; 530 continue;
482 } 531 }
483 else 532 else
484 { 533 {
485 int res = ETP_FINISH (req); 534 int res = ETP_FINISH (req);
486 if (res) 535 if (expect_false (res))
487 return res; 536 return res;
488 } 537 }
489 538
490 if (maxreqs && !--maxreqs) 539 if (expect_false (maxreqs && !--maxreqs))
491 break; 540 break;
492 541
493 if (maxtime) 542 if (maxtime)
494 { 543 {
495 gettimeofday (&tv_now, 0); 544 gettimeofday (&tv_now, 0);
514 563
515static void etp_submit (ETP_REQ *req) 564static void etp_submit (ETP_REQ *req)
516{ 565{
517 req->pri -= ETP_PRI_MIN; 566 req->pri -= ETP_PRI_MIN;
518 567
519 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;
520 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;
521 570
571 if (expect_false (req->type == EIO_GROUP))
572 {
573 /* I hope this is worth it :/ */
522 X_LOCK (reqlock); 574 X_LOCK (reqlock);
523 ++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;
524 ++nready; 591 ++nready;
525 reqq_push (&req_queue, req); 592 reqq_push (&req_queue, req);
526 X_COND_SIGNAL (reqwait); 593 X_COND_SIGNAL (reqwait);
527 X_UNLOCK (reqlock); 594 X_UNLOCK (reqlock);
528 595
529 etp_maybe_start_thread (); 596 etp_maybe_start_thread ();
597 }
530} 598}
531 599
532static void etp_set_max_poll_time (double nseconds) 600static void etp_set_max_poll_time (double nseconds)
533{ 601{
534 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 602 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
569 637
570static void grp_try_feed (eio_req *grp) 638static void grp_try_feed (eio_req *grp)
571{ 639{
572 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
573 { 641 {
574 int old_len = grp->size; 642 grp->flags &= ~EIO_FLAG_GROUPADD;
575 643
576 EIO_FEED (grp); 644 EIO_FEED (grp);
577 645
578 /* stop if no progress has been made */ 646 /* stop if no progress has been made */
579 if (old_len == grp->size) 647 if (!(grp->flags & EIO_FLAG_GROUPADD))
580 { 648 {
581 grp->feed = 0; 649 grp->feed = 0;
582 break; 650 break;
583 } 651 }
584 } 652 }
701 769
702/*****************************************************************************/ 770/*****************************************************************************/
703/* work around various missing functions */ 771/* work around various missing functions */
704 772
705#if !HAVE_PREADWRITE 773#if !HAVE_PREADWRITE
774# undef pread
775# undef pwrite
706# define pread eio__pread 776# define pread eio__pread
707# define pwrite eio__pwrite 777# 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;
715 778
716static ssize_t 779static ssize_t
717eio__pread (int fd, void *buf, size_t count, off_t offset) 780eio__pread (int fd, void *buf, size_t count, off_t offset)
718{ 781{
719 ssize_t res; 782 ssize_t res;
737 800
738 X_LOCK (preadwritelock); 801 X_LOCK (preadwritelock);
739 ooffset = lseek (fd, 0, SEEK_CUR); 802 ooffset = lseek (fd, 0, SEEK_CUR);
740 lseek (fd, offset, SEEK_SET); 803 lseek (fd, offset, SEEK_SET);
741 res = write (fd, buf, count); 804 res = write (fd, buf, count);
742 lseek (fd, offset, SEEK_SET); 805 lseek (fd, ooffset, SEEK_SET);
743 X_UNLOCK (preadwritelock); 806 X_UNLOCK (preadwritelock);
744 807
745 return res; 808 return res;
746} 809}
747#endif 810#endif
748 811
749#ifndef HAVE_FUTIMES 812#ifndef HAVE_FUTIMES
750 813
814# undef utimes
815# undef futimes
751# define utimes(path,times) eio__utimes (path, times) 816# define utimes(path,times) eio__utimes (path, times)
752# define futimes(fd,times) eio__futimes (fd, times) 817# define futimes(fd,times) eio__futimes (fd, times)
753 818
754static int 819static int
755eio__utimes (const char *filename, const struct timeval times[2]) 820eio__utimes (const char *filename, const struct timeval times[2])
774} 839}
775 840
776#endif 841#endif
777 842
778#if !HAVE_FDATASYNC 843#if !HAVE_FDATASYNC
844# undef fdatasync
779# define fdatasync fsync 845# define fdatasync(fd) fsync (fd)
780#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}
781 875
782#if !HAVE_READAHEAD 876#if !HAVE_READAHEAD
877# undef readahead
783# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
784 879
785static ssize_t 880static ssize_t
786eio__readahead (int fd, off_t offset, size_t count, worker *self) 881eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
787{ 882{
788 size_t todo = count; 883 size_t todo = count;
789 dBUF; 884 dBUF;
790 885
791 while (todo > 0) 886 while (todo > 0)
799 894
800 errno = 0; 895 errno = 0;
801 return count; 896 return count;
802} 897}
803 898
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 899#endif
836 900
837/* sendfile always needs emulation */ 901/* sendfile always needs emulation */
838static ssize_t 902static ssize_t
839eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 903eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
926 } 990 }
927 991
928 return res; 992 return res;
929} 993}
930 994
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 /* should be good for 2**31 entries */
1008 struct rng { int l, r; } rng [32];
1009 int i, j;
1010
1011 i = 0;
1012 rng[0].l = 0;
1013 rng[0].r = size;
1014
1015 while (expect_true (i >= 0))
1016 {
1017 int L = rng [i].l;
1018 int R = rng [i].r - 1;
1019
1020 if (expect_false (L + EIO_QSORT_CUTOFF < R))
1021 {
1022 eio_dirent piv = dents [L];
1023
1024 while (L < R)
1025 {
1026 while (eio_dent_cmp (&dents [R], &piv) >= 0 && L < R)
1027 --R;
1028
1029 if (L < R)
1030 dents [L++] = dents [R];
1031
1032 while (eio_dent_cmp (&dents [L], &piv) <= 0 && L < R)
1033 ++L;
1034
1035 if (L < R)
1036 dents [R--] = dents [L];
1037 }
1038
1039 dents [L] = piv;
1040
1041 ++i;
1042 rng [i].l = L + 1;
1043 rng [i].r = rng [i - 1].r;
1044 rng [i - 1].r = L;
1045
1046 if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1047 {
1048 struct rng t;
1049
1050 t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1051 }
1052 }
1053 else
1054 --i;
1055 }
1056
1057 /* use a simple insertion sort at the end */
1058 for (i = 1; i < size; ++i)
1059 {
1060 eio_dirent value = dents [i];
1061
1062 for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j)
1063 dents [j + 1] = dents [j];
1064
1065 dents [j + 1] = value;
1066 }
1067}
1068
931/* read a full directory */ 1069/* read a full directory */
932static void 1070static void
933eio__scandir (eio_req *req, etp_worker *self) 1071eio__scandir (eio_req *req, etp_worker *self)
934{ 1072{
935 DIR *dirp; 1073 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; 1074 EIO_STRUCT_DIRENT *entp;
942 char *name, *names; 1075 char *name, *names;
943 int memlen = 4096; 1076 int namesalloc = 4096;
944 int memofs = 0; 1077 int namesoffs = 0;
1078 int flags = req->int1;
1079 eio_dirent *dents = 0;
1080 int dentalloc = 128;
945 int res = 0; 1081 int dentoffs = 0;
1082
1083 req->result = -1;
1084
1085 if (!(flags & EIO_READDIR_DENTS))
1086 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
946 1087
947 X_LOCK (wrklock); 1088 X_LOCK (wrklock);
1089 /* the corresponding closedir is in ETP_WORKER_CLEAR */
948 self->dirp = dirp = opendir (req->ptr1); 1090 self->dirp = dirp = opendir (req->ptr1);
949 self->dbuf = u = malloc (sizeof (*u));
950 req->flags |= EIO_FLAG_PTR2_FREE; 1091 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
951 req->ptr2 = names = malloc (memlen); 1092 req->ptr1 = names = malloc (namesalloc);
1093 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
952 X_UNLOCK (wrklock); 1094 X_UNLOCK (wrklock);
953 1095
954 if (dirp && u && names) 1096 if (dirp && names && (!flags || dents))
955 for (;;) 1097 for (;;)
956 { 1098 {
957 errno = 0; 1099 errno = 0;
958 readdir_r (dirp, &u->d, &entp); 1100 entp = readdir (dirp);
959 1101
960 if (!entp) 1102 if (!entp)
1103 {
1104 if (errno)
1105 break;
1106
1107 /* sort etc. */
1108 req->int1 = flags;
1109 req->result = dentoffs;
1110
1111 if (dents)
1112 {
1113 eio_dirent *ent = dents + dentoffs;
1114
1115 while (ent > dents)
1116 (--ent)->name = names + (size_t)ent->name;
1117 }
1118
1119 if (flags & EIO_READDIR_STAT_ORDER
1120 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
1121 eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1122 else if (flags & EIO_READDIR_DIRS_FIRST)
1123 {
1124 /* in this case, all is known, and we just put dirs first and sort them */
1125 eio_dirent *ent = dents + dentoffs;
1126 eio_dirent *dir = dents;
1127
1128 /* now move dirs to the front, and non-dirs to the back */
1129 /* by walking from both sides and swapping if necessary */
1130 while (ent > dir)
1131 {
1132 if (dir->type == DT_DIR)
1133 ++dir;
1134 else
1135 {
1136 --ent;
1137
1138 if (ent->type == DT_DIR)
1139 {
1140 eio_dirent tmp = *dir;
1141 *dir = *ent;
1142 *ent = tmp;
1143
1144 ++dir;
1145 }
1146 }
1147 }
1148
1149 /* now sort the dirs only */
1150 eio_dent_sort (dents, dir - dents);
1151 }
1152
1153 /* only provide the names array unless DENTS is specified */
1154 if (!(flags & EIO_READDIR_DENTS))
1155 {
1156 X_LOCK (wrklock);
1157 assert (!dents);
1158 req->ptr1 = 0;
1159 req->ptr2 = names;
1160 X_UNLOCK (wrklock);
1161 }
1162
961 break; 1163 break;
1164 }
962 1165
1166 /* now add the entry to our list(s) */
963 name = entp->d_name; 1167 name = entp->d_name;
964 1168
1169 /* skip . and .. entries */
965 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1170 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
966 { 1171 {
967 int len = strlen (name) + 1; 1172 int len = D_NAMLEN (entp) + 1;
968 1173
969 res++; 1174 while (expect_false (namesoffs + len > namesalloc))
970
971 while (memofs + len > memlen)
972 { 1175 {
973 memlen *= 2; 1176 namesalloc *= 2;
974 X_LOCK (wrklock); 1177 X_LOCK (wrklock);
975 req->ptr2 = names = realloc (names, memlen); 1178 req->ptr1 = names = realloc (names, namesalloc);
976 X_UNLOCK (wrklock); 1179 X_UNLOCK (wrklock);
977 1180
978 if (!names) 1181 if (!names)
979 break; 1182 break;
980 } 1183 }
981 1184
982 memcpy (names + memofs, name, len); 1185 memcpy (names + namesoffs, name, len);
1186
1187 if (dents)
1188 {
1189 struct eio_dirent *ent;
1190
1191 if (expect_false (dentoffs == dentalloc))
1192 {
1193 dentalloc *= 2;
1194 X_LOCK (wrklock);
1195 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1196 X_UNLOCK (wrklock);
1197
1198 if (!dents)
1199 break;
1200 }
1201
1202 ent = dents + dentoffs;
1203
1204 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
1205 ent->namelen = len - 1;
1206 ent->inode = D_INO (entp);
1207
1208 switch (D_TYPE (entp))
1209 {
1210 default:
1211 ent->type = EIO_DT_UNKNOWN;
1212 flags |= EIO_READDIR_FOUND_UNKNOWN;
1213 break;
1214
1215 #ifdef DT_FIFO
1216 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1217 #endif
1218 #ifdef DT_CHR
1219 case DT_CHR: ent->type = EIO_DT_CHR; break;
1220 #endif
1221 #ifdef DT_MPC
1222 case DT_MPC: ent->type = EIO_DT_MPC; break;
1223 #endif
1224 #ifdef DT_DIR
1225 case DT_DIR: ent->type = EIO_DT_DIR; break;
1226 #endif
1227 #ifdef DT_NAM
1228 case DT_NAM: ent->type = EIO_DT_NAM; break;
1229 #endif
1230 #ifdef DT_BLK
1231 case DT_BLK: ent->type = EIO_DT_BLK; break;
1232 #endif
1233 #ifdef DT_MPB
1234 case DT_MPB: ent->type = EIO_DT_MPB; break;
1235 #endif
1236 #ifdef DT_REG
1237 case DT_REG: ent->type = EIO_DT_REG; break;
1238 #endif
1239 #ifdef DT_NWK
1240 case DT_NWK: ent->type = EIO_DT_NWK; break;
1241 #endif
1242 #ifdef DT_CMP
1243 case DT_CMP: ent->type = EIO_DT_CMP; break;
1244 #endif
1245 #ifdef DT_LNK
1246 case DT_LNK: ent->type = EIO_DT_LNK; break;
1247 #endif
1248 #ifdef DT_SOCK
1249 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1250 #endif
1251 #ifdef DT_DOOR
1252 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1253 #endif
1254 #ifdef DT_WHT
1255 case DT_WHT: ent->type = EIO_DT_WHT; break;
1256 #endif
1257 }
1258
1259 ent->score = 0;
1260
1261 if (flags & EIO_READDIR_DIRS_FIRST)
1262 {
1263 if (ent->type == EIO_DT_UNKNOWN)
1264 {
1265 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1266 ent->score = 98;
1267 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1268 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
1269 }
1270 else if (ent->type == EIO_DT_DIR)
1271 ent->score = 100;
1272 }
1273 }
1274
983 memofs += len; 1275 namesoffs += len;
1276 ++dentoffs;
1277 }
1278
1279 if (EIO_CANCELLED (req))
1280 {
1281 errno = ECANCELED;
1282 break;
984 } 1283 }
985 } 1284 }
1285}
986 1286
987 if (errno) 1287#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
988 res = -1; 1288# undef msync
989 1289# define msync(a,b,c) ((errno = ENOSYS), -1)
990 req->result = res; 1290#endif
1291
1292int
1293eio__mtouch (void *mem, size_t len, int flags)
1294{
1295 intptr_t addr = (intptr_t)mem;
1296 intptr_t end = addr + len;
1297#ifdef PAGESIZE
1298 const intptr_t page = PAGESIZE;
1299#else
1300 static intptr_t page;
1301
1302 if (!page)
1303 page = sysconf (_SC_PAGESIZE);
1304#endif
1305
1306 addr &= ~(page - 1); /* assume page size is always a power of two */
1307
1308 if (addr < end)
1309 if (flags) /* modify */
1310 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1311 else
1312 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1313
1314 return 0;
991} 1315}
992 1316
993/*****************************************************************************/ 1317/*****************************************************************************/
994 1318
995#define ALLOC(len) \ 1319#define ALLOC(len) \
1082 1406
1083/*****************************************************************************/ 1407/*****************************************************************************/
1084 1408
1085int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1409int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1086{ 1410{
1087 etp_init (want_poll, done_poll); 1411 return etp_init (want_poll, done_poll);
1088} 1412}
1089 1413
1090static void eio_api_destroy (eio_req *req) 1414static void eio_api_destroy (eio_req *req)
1091{ 1415{
1092 free (req); 1416 free (req);
1154 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1478 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1155 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1479 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; 1480 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1157 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1481 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1158 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1482 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; 1483 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1160 1484
1161 case EIO_READLINK: ALLOC (NAME_MAX); 1485 case EIO_READLINK: ALLOC (NAME_MAX);
1162 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1486 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1163 1487
1164 case EIO_SYNC: req->result = 0; sync (); break; 1488 case EIO_SYNC: req->result = 0; sync (); break;
1165 case EIO_FSYNC: req->result = fsync (req->int1); break; 1489 case EIO_FSYNC: req->result = fsync (req->int1); break;
1166 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1490 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1491 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1492 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1493 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1167 1494
1168 case EIO_READDIR: eio__scandir (req, self); break; 1495 case EIO_READDIR: eio__scandir (req, self); break;
1169 1496
1170 case EIO_BUSY: 1497 case EIO_BUSY:
1171#ifdef _WIN32 1498#ifdef _WIN32
1203 1530
1204 req->result = req->type == EIO_FUTIME 1531 req->result = req->type == EIO_FUTIME
1205 ? futimes (req->int1, times) 1532 ? futimes (req->int1, times)
1206 : utimes (req->ptr1, times); 1533 : utimes (req->ptr1, times);
1207 } 1534 }
1535 break;
1208 1536
1209 case EIO_GROUP: 1537 case EIO_GROUP:
1538 abort (); /* handled in eio_request */
1539
1210 case EIO_NOP: 1540 case EIO_NOP:
1211 req->result = 0; 1541 req->result = 0;
1212 break; 1542 break;
1213 1543
1214 case EIO_CUSTOM: 1544 case EIO_CUSTOM:
1215 req->feed (req); 1545 ((void (*)(eio_req *))req->feed) (req);
1216 break; 1546 break;
1217 1547
1218 default: 1548 default:
1219 req->result = -1; 1549 req->result = -1;
1220 break; 1550 break;
1243eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1573eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1244{ 1574{
1245 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1575 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1246} 1576}
1247 1577
1578eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1579{
1580 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1581}
1582
1583eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1584{
1585 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1586}
1587
1588eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1589{
1590 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1591}
1592
1248eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1593eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1249{ 1594{
1250 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1595 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1251} 1596}
1252 1597
1364eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1709eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1365{ 1710{
1366 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1711 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1367} 1712}
1368 1713
1369eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1714eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1370{ 1715{
1371 return eio__1path (EIO_READDIR, path, pri, cb, data); 1716 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1372} 1717}
1373 1718
1374eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1719eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1375{ 1720{
1376 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1721 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1377} 1722}
1378 1723
1379static eio_req * 1724static eio_req *
1380eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1725eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1381{ 1726{
1407 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1752 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1408} 1753}
1409 1754
1410eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 1755eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1411{ 1756{
1412 REQ (EIO_CUSTOM); req->feed = execute; SEND; 1757 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1413} 1758}
1414 1759
1415#endif 1760#endif
1416 1761
1417eio_req *eio_grp (eio_cb cb, void *data) 1762eio_req *eio_grp (eio_cb cb, void *data)
1445 1790
1446void eio_grp_add (eio_req *grp, eio_req *req) 1791void eio_grp_add (eio_req *grp, eio_req *req)
1447{ 1792{
1448 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1793 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1449 1794
1795 grp->flags |= EIO_FLAG_GROUPADD;
1796
1450 ++grp->size; 1797 ++grp->size;
1451 req->grp = grp; 1798 req->grp = grp;
1452 1799
1453 req->grp_prev = 0; 1800 req->grp_prev = 0;
1454 req->grp_next = grp->grp_first; 1801 req->grp_next = grp->grp_first;
1463/* misc garbage */ 1810/* misc garbage */
1464 1811
1465ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1812ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1466{ 1813{
1467 etp_worker wrk; 1814 etp_worker wrk;
1815 ssize_t ret;
1468 1816
1469 wrk.dbuf = 0; 1817 wrk.dbuf = 0;
1470 1818
1471 eio__sendfile (ofd, ifd, offset, count, &wrk); 1819 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1472 1820
1473 if (wrk.dbuf) 1821 if (wrk.dbuf)
1474 free (wrk.dbuf); 1822 free (wrk.dbuf);
1475}
1476 1823
1824 return ret;
1825}
1826

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines