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

Comparing libeio/eio.c (file contents):
Revision 1.14 by root, Sat May 17 12:17:25 2008 UTC vs.
Revision 1.31 by root, Sat Jun 6 17:25:13 2009 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#include "eio.h" 40#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
41#include "xthread.h" 45#include "xthread.h"
42 46
43#include <errno.h> 47#include <errno.h>
44#include <stddef.h> 48#include <stddef.h>
45#include <stdlib.h> 49#include <stdlib.h>
64#endif 68#endif
65 69
66#ifdef _WIN32 70#ifdef _WIN32
67 71
68 /*doh*/ 72 /*doh*/
69
70#else 73#else
71 74
72# include "config.h" 75# include "config.h"
73# include <sys/time.h> 76# include <sys/time.h>
74# include <sys/select.h> 77# include <sys/select.h>
75# include <unistd.h> 78# include <unistd.h>
76# include <utime.h> 79# include <utime.h>
77# include <signal.h> 80# include <signal.h>
78# include <dirent.h> 81# include <dirent.h>
82
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
85# define D_INO(de) (de)->d_fileno
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
88# define D_INO(de) (de)->d_ino
89# endif
90
91# ifdef _DIRENT_HAVE_D_TYPE
92# define D_TYPE(de) (de)->d_type
93# endif
79 94
80# ifndef EIO_STRUCT_DIRENT 95# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 96# define EIO_STRUCT_DIRENT struct dirent
82# endif 97# endif
83 98
96# else 111# else
97# error sendfile support requested but not available 112# error sendfile support requested but not available
98# endif 113# endif
99#endif 114#endif
100 115
116#ifndef D_TYPE
117# define D_TYPE(de) 0
118#endif
119
120#ifndef D_INO
121# define D_INO(de) 0
122#endif
123
101/* number of seconds after which an idle threads exit */ 124/* number of seconds after which an idle threads exit */
102#define IDLE_TIMEOUT 10 125#define IDLE_TIMEOUT 10
103 126
104/* used for struct dirent, AIX doesn't provide it */ 127/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 128#ifndef NAME_MAX
120 143
121#define EIO_TICKS ((1000000 + 1023) >> 10) 144#define EIO_TICKS ((1000000 + 1023) >> 10)
122 145
123/*****************************************************************************/ 146/*****************************************************************************/
124 147
148#if __GNUC__ >= 3
149# define expect(expr,value) __builtin_expect ((expr),(value))
150#else
151# define expect(expr,value) (expr)
152#endif
153
154#define expect_false(expr) expect ((expr) != 0, 0)
155#define expect_true(expr) expect ((expr) != 0, 1)
156
157/*****************************************************************************/
158
125#define ETP_PRI_MIN EIO_PRI_MIN 159#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 160#define ETP_PRI_MAX EIO_PRI_MAX
161
162struct etp_worker;
127 163
128#define ETP_REQ eio_req 164#define ETP_REQ eio_req
129#define ETP_DESTROY(req) eio_destroy (req) 165#define ETP_DESTROY(req) eio_destroy (req)
130static int eio_finish (eio_req *req); 166static int eio_finish (eio_req *req);
131#define ETP_FINISH(req) eio_finish (req) 167#define ETP_FINISH(req) eio_finish (req)
168static void eio_execute (struct etp_worker *self, eio_req *req);
169#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
132 170
133#define ETP_WORKER_CLEAR(req) \ 171#define ETP_WORKER_CLEAR(req) \
134 if (wrk->dbuf) \ 172 if (wrk->dbuf) \
135 { \ 173 { \
136 free (wrk->dbuf); \ 174 free (wrk->dbuf); \
173static mutex_t wrklock = X_MUTEX_INIT; 211static mutex_t wrklock = X_MUTEX_INIT;
174static mutex_t reslock = X_MUTEX_INIT; 212static mutex_t reslock = X_MUTEX_INIT;
175static mutex_t reqlock = X_MUTEX_INIT; 213static mutex_t reqlock = X_MUTEX_INIT;
176static cond_t reqwait = X_COND_INIT; 214static cond_t reqwait = X_COND_INIT;
177 215
216#if !HAVE_PREADWRITE
217/*
218 * make our pread/pwrite emulation safe against themselves, but not against
219 * normal read/write by using a mutex. slows down execution a lot,
220 * but that's your problem, not mine.
221 */
222static mutex_t preadwritelock = X_MUTEX_INIT;
223#endif
224
178typedef struct etp_worker 225typedef struct etp_worker
179{ 226{
180 /* locked by wrklock */ 227 /* locked by wrklock */
181 struct etp_worker *prev, *next; 228 struct etp_worker *prev, *next;
182 229
310 X_LOCK (reqlock); 357 X_LOCK (reqlock);
311 X_LOCK (reslock); 358 X_LOCK (reslock);
312#if !HAVE_PREADWRITE 359#if !HAVE_PREADWRITE
313 X_LOCK (preadwritelock); 360 X_LOCK (preadwritelock);
314#endif 361#endif
315#if !HAVE_READDIR_R
316 X_LOCK (readdirlock);
317#endif
318} 362}
319 363
320static void etp_atfork_parent (void) 364static void etp_atfork_parent (void)
321{ 365{
322#if !HAVE_READDIR_R
323 X_UNLOCK (readdirlock);
324#endif
325#if !HAVE_PREADWRITE 366#if !HAVE_PREADWRITE
326 X_UNLOCK (preadwritelock); 367 X_UNLOCK (preadwritelock);
327#endif 368#endif
328 X_UNLOCK (reslock); 369 X_UNLOCK (reslock);
329 X_UNLOCK (reqlock); 370 X_UNLOCK (reqlock);
332 373
333static void etp_atfork_child (void) 374static void etp_atfork_child (void)
334{ 375{
335 ETP_REQ *prv; 376 ETP_REQ *prv;
336 377
337 while (prv = reqq_shift (&req_queue)) 378 while ((prv = reqq_shift (&req_queue)))
338 ETP_DESTROY (prv); 379 ETP_DESTROY (prv);
339 380
340 while (prv = reqq_shift (&res_queue)) 381 while ((prv = reqq_shift (&res_queue)))
341 ETP_DESTROY (prv); 382 ETP_DESTROY (prv);
342 383
343 while (wrk_first.next != &wrk_first) 384 while (wrk_first.next != &wrk_first)
344 { 385 {
345 etp_worker *wrk = wrk_first.next; 386 etp_worker *wrk = wrk_first.next;
373 414
374 pthread_once (&doinit, etp_once_init); 415 pthread_once (&doinit, etp_once_init);
375 416
376 want_poll_cb = want_poll; 417 want_poll_cb = want_poll;
377 done_poll_cb = done_poll; 418 done_poll_cb = done_poll;
419
420 return 0;
378} 421}
379 422
380X_THREAD_PROC (etp_proc); 423X_THREAD_PROC (etp_proc);
381 424
382static void etp_start_thread (void) 425static void etp_start_thread (void)
402 X_UNLOCK (wrklock); 445 X_UNLOCK (wrklock);
403} 446}
404 447
405static void etp_maybe_start_thread (void) 448static void etp_maybe_start_thread (void)
406{ 449{
407 if (etp_nthreads () >= wanted) 450 if (expect_true (etp_nthreads () >= wanted))
408 return; 451 return;
409 452
410 /* todo: maybe use idle here, but might be less exact */ 453 /* todo: maybe use idle here, but might be less exact */
411 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 454 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
412 return; 455 return;
413 456
414 etp_start_thread (); 457 etp_start_thread ();
415} 458}
416 459
469 512
470 X_LOCK (reqlock); 513 X_LOCK (reqlock);
471 --nreqs; 514 --nreqs;
472 X_UNLOCK (reqlock); 515 X_UNLOCK (reqlock);
473 516
474 if (req->type == EIO_GROUP && req->size) 517 if (expect_false (req->type == EIO_GROUP && req->size))
475 { 518 {
476 req->int1 = 1; /* mark request as delayed */ 519 req->int1 = 1; /* mark request as delayed */
477 continue; 520 continue;
478 } 521 }
479 else 522 else
480 { 523 {
481 int res = ETP_FINISH (req); 524 int res = ETP_FINISH (req);
482 if (res) 525 if (expect_false (res))
483 return res; 526 return res;
484 } 527 }
485 528
486 if (maxreqs && !--maxreqs) 529 if (expect_false (maxreqs && !--maxreqs))
487 break; 530 break;
488 531
489 if (maxtime) 532 if (maxtime)
490 { 533 {
491 gettimeofday (&tv_now, 0); 534 gettimeofday (&tv_now, 0);
510 553
511static void etp_submit (ETP_REQ *req) 554static void etp_submit (ETP_REQ *req)
512{ 555{
513 req->pri -= ETP_PRI_MIN; 556 req->pri -= ETP_PRI_MIN;
514 557
515 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 558 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; 559 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
517 560
561 if (expect_false (req->type == EIO_GROUP))
562 {
563 /* I hope this is worth it :/ */
518 X_LOCK (reqlock); 564 X_LOCK (reqlock);
519 ++nreqs; 565 ++nreqs;
566 X_UNLOCK (reqlock);
567
568 X_LOCK (reslock);
569
570 ++npending;
571
572 if (!reqq_push (&res_queue, req) && want_poll_cb)
573 want_poll_cb ();
574
575 X_UNLOCK (reslock);
576 }
577 else
578 {
579 X_LOCK (reqlock);
580 ++nreqs;
520 ++nready; 581 ++nready;
521 reqq_push (&req_queue, req); 582 reqq_push (&req_queue, req);
522 X_COND_SIGNAL (reqwait); 583 X_COND_SIGNAL (reqwait);
523 X_UNLOCK (reqlock); 584 X_UNLOCK (reqlock);
524 585
525 etp_maybe_start_thread (); 586 etp_maybe_start_thread ();
587 }
526} 588}
527 589
528static void etp_set_max_poll_time (double nseconds) 590static void etp_set_max_poll_time (double nseconds)
529{ 591{
530 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 592 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
565 627
566static void grp_try_feed (eio_req *grp) 628static void grp_try_feed (eio_req *grp)
567{ 629{
568 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 630 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569 { 631 {
570 int old_len = grp->size; 632 grp->flags &= ~EIO_FLAG_GROUPADD;
571 633
572 EIO_FEED (grp); 634 EIO_FEED (grp);
573 635
574 /* stop if no progress has been made */ 636 /* stop if no progress has been made */
575 if (old_len == grp->size) 637 if (!(grp->flags & EIO_FLAG_GROUPADD))
576 { 638 {
577 grp->feed = 0; 639 grp->feed = 0;
578 break; 640 break;
579 } 641 }
580 } 642 }
697 759
698/*****************************************************************************/ 760/*****************************************************************************/
699/* work around various missing functions */ 761/* work around various missing functions */
700 762
701#if !HAVE_PREADWRITE 763#if !HAVE_PREADWRITE
764# undef pread
765# undef pwrite
702# define pread eio__pread 766# define pread eio__pread
703# define pwrite eio__pwrite 767# 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;
711 768
712static ssize_t 769static ssize_t
713eio__pread (int fd, void *buf, size_t count, off_t offset) 770eio__pread (int fd, void *buf, size_t count, off_t offset)
714{ 771{
715 ssize_t res; 772 ssize_t res;
733 790
734 X_LOCK (preadwritelock); 791 X_LOCK (preadwritelock);
735 ooffset = lseek (fd, 0, SEEK_CUR); 792 ooffset = lseek (fd, 0, SEEK_CUR);
736 lseek (fd, offset, SEEK_SET); 793 lseek (fd, offset, SEEK_SET);
737 res = write (fd, buf, count); 794 res = write (fd, buf, count);
738 lseek (fd, offset, SEEK_SET); 795 lseek (fd, ooffset, SEEK_SET);
739 X_UNLOCK (preadwritelock); 796 X_UNLOCK (preadwritelock);
740 797
741 return res; 798 return res;
742} 799}
743#endif 800#endif
744 801
745#ifndef HAVE_FUTIMES 802#ifndef HAVE_FUTIMES
746 803
804# undef utimes
805# undef futimes
747# define utimes(path,times) eio__utimes (path, times) 806# define utimes(path,times) eio__utimes (path, times)
748# define futimes(fd,times) eio__futimes (fd, times) 807# define futimes(fd,times) eio__futimes (fd, times)
749 808
750static int 809static int
751eio__utimes (const char *filename, const struct timeval times[2]) 810eio__utimes (const char *filename, const struct timeval times[2])
770} 829}
771 830
772#endif 831#endif
773 832
774#if !HAVE_FDATASYNC 833#if !HAVE_FDATASYNC
834# undef fdatasync
775# define fdatasync fsync 835# define fdatasync(fd) fsync (fd)
776#endif 836#endif
837
838/* sync_file_range always needs emulation */
839int
840eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
841{
842#if HAVE_SYNC_FILE_RANGE
843 int res;
844
845 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
846 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
847 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
848 {
849 flags = 0
850 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
851 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
852 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
853 }
854
855 res = sync_file_range (fd, offset, nbytes, flags);
856
857 if (!res || errno != ENOSYS)
858 return res;
859#endif
860
861 /* even though we could play tricks with the flags, it's better to always
862 * call fdatasync, as thta matches the expectation of it's users best */
863 return fdatasync (fd);
864}
777 865
778#if !HAVE_READAHEAD 866#if !HAVE_READAHEAD
867# undef readahead
779# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 868# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780 869
781static ssize_t 870static ssize_t
782eio__readahead (int fd, off_t offset, size_t count, worker *self) 871eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783{ 872{
784 size_t todo = count; 873 size_t todo = count;
785 dBUF; 874 dBUF;
786 875
787 while (todo > 0) 876 while (todo > 0)
795 884
796 errno = 0; 885 errno = 0;
797 return count; 886 return count;
798} 887}
799 888
800#endif
801
802#if !HAVE_READDIR_R
803# define readdir_r eio__readdir_r
804
805static mutex_t readdirlock = X_MUTEX_INIT;
806
807static int
808eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
809{
810 EIO_STRUCT_DIRENT *e;
811 int errorno;
812
813 X_LOCK (readdirlock);
814
815 e = readdir (dirp);
816 errorno = errno;
817
818 if (e)
819 {
820 *res = ent;
821 strcpy (ent->d_name, e->d_name);
822 }
823 else
824 *res = 0;
825
826 X_UNLOCK (readdirlock);
827
828 errno = errorno;
829 return e ? 0 : -1;
830}
831#endif 889#endif
832 890
833/* sendfile always needs emulation */ 891/* sendfile always needs emulation */
834static ssize_t 892static ssize_t
835eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 893eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
922 } 980 }
923 981
924 return res; 982 return res;
925} 983}
926 984
985static int
986eio_dent_cmp (const void *a_, const void *b_)
987{
988 const eio_dirent *a = (const eio_dirent *)a_;
989 const eio_dirent *b = (const eio_dirent *)b_;
990
991 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
992 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */
993}
994
927/* read a full directory */ 995/* read a full directory */
928static void 996static void
929eio__scandir (eio_req *req, etp_worker *self) 997eio__scandir (eio_req *req, etp_worker *self)
930{ 998{
931 DIR *dirp; 999 DIR *dirp;
932 union
933 {
934 EIO_STRUCT_DIRENT d;
935 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
936 } *u;
937 EIO_STRUCT_DIRENT *entp; 1000 EIO_STRUCT_DIRENT *entp;
938 char *name, *names; 1001 unsigned char *name, *names;
939 int memlen = 4096; 1002 int namesalloc = 4096;
940 int memofs = 0; 1003 int namesoffs = 0;
1004 int flags = req->int1;
1005 eio_dirent *dents = 0;
1006 int dentalloc = 128;
941 int res = 0; 1007 int dentoffs = 0;
1008
1009 req->result = -1;
1010
1011 if (!(flags & EIO_READDIR_DENTS))
1012 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
942 1013
943 X_LOCK (wrklock); 1014 X_LOCK (wrklock);
1015
1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */
944 self->dirp = dirp = opendir (req->ptr1); 1017 self->dirp = dirp = opendir (req->ptr1);
945 self->dbuf = u = malloc (sizeof (*u));
946 req->flags |= EIO_FLAG_PTR2_FREE; 1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
947 req->ptr2 = names = malloc (memlen); 1019 req->ptr1 = names = malloc (namesalloc);
1020 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1021
948 X_UNLOCK (wrklock); 1022 X_UNLOCK (wrklock);
949 1023
950 if (dirp && u && names) 1024 if (dirp && names && (!flags || dents))
951 for (;;) 1025 for (;;)
952 { 1026 {
953 errno = 0; 1027 errno = 0;
954 readdir_r (dirp, &u->d, &entp); 1028 entp = readdir (dirp);
955 1029
956 if (!entp) 1030 if (!entp)
1031 {
1032 if (errno)
1033 break;
1034
1035 /* sort etc. */
1036 req->int1 = flags;
1037 req->result = dentoffs;
1038
1039 if (flags & EIO_READDIR_STAT_ORDER || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))
1040 {
1041 /* pray your qsort doesn't use quicksort */
1042 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1043 }
1044 else if (flags & EIO_READDIR_DIRS_FIRST && !(flags & EIO_READDIR_FOUND_UNKNOWN))
1045 {
1046 /* in this case, all is known, and we just put dirs first and sort them */
1047 eio_dirent *ent = dents + dentoffs;
1048 eio_dirent *dir = dents;
1049
1050 while (ent > dir)
1051 {
1052 if (dir->type == DT_DIR)
1053 ++dir;
1054 else
1055 {
1056 --ent;
1057
1058 if (ent->type == DT_DIR)
1059 {
1060 eio_dirent tmp = *dir;
1061 *dir = *ent;
1062 *ent = tmp;
1063
1064 ++dir;
1065 }
1066 }
1067 }
1068
1069 /* now sort the dirs only */
1070 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1071 }
1072
1073 {int i; for(i=0;i<dentoffs;++i){eio_dirent *e=dents+i; printf ("%9ld %3d %s\n", e->inode,e->score,e->name);}}//D
1074
957 break; 1075 break;
1076 }
958 1077
959 name = entp->d_name; 1078 name = entp->d_name;
960 1079
961 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1080 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
962 { 1081 {
963 int len = strlen (name) + 1; 1082 int len = strlen (name) + 1;
964 1083
965 res++; 1084 while (expect_false (namesoffs + len > namesalloc))
966
967 while (memofs + len > memlen)
968 { 1085 {
969 memlen *= 2; 1086 namesalloc *= 2;
970 X_LOCK (wrklock); 1087 X_LOCK (wrklock);
971 req->ptr2 = names = realloc (names, memlen); 1088 req->ptr1 = names = realloc (names, namesalloc);
972 X_UNLOCK (wrklock); 1089 X_UNLOCK (wrklock);
973 1090
974 if (!names) 1091 if (!names)
975 break; 1092 break;
976 } 1093 }
977 1094
978 memcpy (names + memofs, name, len); 1095 memcpy (names + namesoffs, name, len);
1096
1097 if (dents)
1098 {
1099 struct eio_dirent *ent;
1100
1101 if (expect_false (dentoffs == dentalloc))
1102 {
1103 dentalloc *= 2;
1104 X_LOCK (wrklock);
1105 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1106 X_UNLOCK (wrklock);
1107
1108 if (!dents)
1109 break;
1110 }
1111
1112 ent = dents + dentoffs;
1113
1114 ent->name = names + namesoffs;
1115 ent->namelen = len - 1;
1116 ent->inode = D_INO (entp);
1117
1118 switch (D_TYPE (entp))
1119 {
1120 default:
1121 ent->type = EIO_DT_UNKNOWN;
1122 flags |= EIO_READDIR_FOUND_UNKNOWN;
1123 break;
1124
1125 #ifdef DT_FIFO
1126 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1127 #endif
1128 #ifdef DT_CHR
1129 case DT_CHR: ent->type = EIO_DT_CHR; break;
1130 #endif
1131 #ifdef DT_DIR
1132 case DT_DIR: ent->type = EIO_DT_DIR; break;
1133 #endif
1134 #ifdef DT_BLK
1135 case DT_BLK: ent->type = EIO_DT_BLK; break;
1136 #endif
1137 #ifdef DT_REG
1138 case DT_REG: ent->type = EIO_DT_REG; break;
1139 #endif
1140 #ifdef DT_LNK
1141 case DT_LNK: ent->type = EIO_DT_LNK; break;
1142 #endif
1143 #ifdef DT_SOCK
1144 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1145 #endif
1146 #ifdef DT_WHT
1147 case DT_WHT: ent->type = EIO_DT_WHT; break;
1148 #endif
1149 }
1150
1151 ent->score = 0;
1152
1153 if (flags & EIO_READDIR_DIRS_FIRST)
1154 {
1155 if (ent->type == EIO_DT_UNKNOWN)
1156 {
1157 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1158 ent->score = 98;
1159 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1160 ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
1161 }
1162 else if (ent->type == DT_DIR)
1163 ent->score = 100;
1164 }
1165 }
1166
979 memofs += len; 1167 namesoffs += len;
1168 ++dentoffs;
980 } 1169 }
981 } 1170 }
982 1171 else
983 if (errno)
984 res = -1;
985
986 req->result = res; 1172 req->result = -1;
1173
1174 /* if user doesn't want the dents, do not provide it */
1175 if (!(flags & EIO_READDIR_DENTS))
1176 {
1177 X_LOCK (wrklock);
1178 free (dents);
1179 req->ptr2 = req->ptr1;
1180 req->ptr1 = 0;
1181 X_UNLOCK (wrklock);
1182 }
1183}
1184
1185#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1186# undef msync
1187# define msync(a,b,c) ((errno = ENOSYS), -1)
1188#endif
1189
1190int
1191eio__mtouch (void *mem, size_t len, int flags)
1192{
1193 intptr_t addr = (intptr_t)mem;
1194 intptr_t end = addr + len;
1195#ifdef PAGESIZE
1196 const intptr_t page = PAGESIZE;
1197#else
1198 static intptr_t page;
1199
1200 if (!page)
1201 page = sysconf (_SC_PAGESIZE);
1202#endif
1203
1204 addr &= ~(page - 1); /* assume page size is always a power of two */
1205
1206 if (addr < end)
1207 if (flags) /* modify */
1208 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1209 else
1210 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1211
1212 return 0;
987} 1213}
988 1214
989/*****************************************************************************/ 1215/*****************************************************************************/
990 1216
991#define ALLOC(len) \ 1217#define ALLOC(len) \
1047 1273
1048 --nready; 1274 --nready;
1049 1275
1050 X_UNLOCK (reqlock); 1276 X_UNLOCK (reqlock);
1051 1277
1052 errno = 0; /* strictly unnecessary */ 1278 if (req->type < 0)
1279 goto quit;
1053 1280
1054 if (!EIO_CANCELLED (req)) 1281 if (!EIO_CANCELLED (req))
1055 switch (req->type) 1282 ETP_EXECUTE (self, req);
1056 {
1057 case EIO_READ: ALLOC (req->size);
1058 req->result = req->offs >= 0
1059 ? pread (req->int1, req->ptr2, req->size, req->offs)
1060 : read (req->int1, req->ptr2, req->size); break;
1061 case EIO_WRITE: req->result = req->offs >= 0
1062 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1063 : write (req->int1, req->ptr2, req->size); break;
1064
1065 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1066 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1067
1068 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1069 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1070 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1071 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1072 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1073 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1074
1075 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1076 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1077 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1078 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1079 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1080 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1081
1082 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1083 case EIO_CLOSE: req->result = close (req->int1); break;
1084 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1085 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1086 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1087 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1088 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1089 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1090 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1091 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1092
1093 case EIO_READLINK: ALLOC (NAME_MAX);
1094 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1095
1096 case EIO_SYNC: req->result = 0; sync (); break;
1097 case EIO_FSYNC: req->result = fsync (req->int1); break;
1098 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1099
1100 case EIO_READDIR: eio__scandir (req, self); break;
1101
1102 case EIO_BUSY:
1103#ifdef _WIN32
1104 Sleep (req->nv1 * 1000.);
1105#else
1106 {
1107 struct timeval tv;
1108
1109 tv.tv_sec = req->nv1;
1110 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1111
1112 req->result = select (0, 0, 0, 0, &tv);
1113 }
1114#endif
1115 break;
1116
1117 case EIO_UTIME:
1118 case EIO_FUTIME:
1119 {
1120 struct timeval tv[2];
1121 struct timeval *times;
1122
1123 if (req->nv1 != -1. || req->nv2 != -1.)
1124 {
1125 tv[0].tv_sec = req->nv1;
1126 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1127 tv[1].tv_sec = req->nv2;
1128 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1129
1130 times = tv;
1131 }
1132 else
1133 times = 0;
1134
1135
1136 req->result = req->type == EIO_FUTIME
1137 ? futimes (req->int1, times)
1138 : utimes (req->ptr1, times);
1139 }
1140
1141 case EIO_GROUP:
1142 case EIO_NOP:
1143 req->result = 0;
1144 break;
1145
1146 case -1:
1147 goto quit;
1148
1149 default:
1150 req->result = -1;
1151 break;
1152 }
1153
1154 req->errorno = errno;
1155 1283
1156 X_LOCK (reslock); 1284 X_LOCK (reslock);
1157 1285
1158 ++npending; 1286 ++npending;
1159 1287
1176 1304
1177/*****************************************************************************/ 1305/*****************************************************************************/
1178 1306
1179int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1307int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1180{ 1308{
1181 etp_init (want_poll, done_poll); 1309 return etp_init (want_poll, done_poll);
1182} 1310}
1183 1311
1184static void eio_api_destroy (eio_req *req) 1312static void eio_api_destroy (eio_req *req)
1185{ 1313{
1186 free (req); 1314 free (req);
1208 { \ 1336 { \
1209 eio_api_destroy (req); \ 1337 eio_api_destroy (req); \
1210 return 0; \ 1338 return 0; \
1211 } 1339 }
1212 1340
1341static void eio_execute (etp_worker *self, eio_req *req)
1342{
1343 errno = 0;
1344
1345 switch (req->type)
1346 {
1347 case EIO_READ: ALLOC (req->size);
1348 req->result = req->offs >= 0
1349 ? pread (req->int1, req->ptr2, req->size, req->offs)
1350 : read (req->int1, req->ptr2, req->size); break;
1351 case EIO_WRITE: req->result = req->offs >= 0
1352 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1353 : write (req->int1, req->ptr2, req->size); break;
1354
1355 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1356 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1357
1358 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1359 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1360 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1361 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1362 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1363 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1364
1365 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1366 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1367 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1368 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1369 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1370 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1371
1372 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1373 case EIO_CLOSE: req->result = close (req->int1); break;
1374 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1375 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1376 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1377 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1378 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1379 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1380 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1381 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1382
1383 case EIO_READLINK: ALLOC (NAME_MAX);
1384 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1385
1386 case EIO_SYNC: req->result = 0; sync (); break;
1387 case EIO_FSYNC: req->result = fsync (req->int1); break;
1388 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1389 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1390 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1391 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1392
1393 case EIO_READDIR: eio__scandir (req, self); break;
1394
1395 case EIO_BUSY:
1396#ifdef _WIN32
1397 Sleep (req->nv1 * 1000.);
1398#else
1399 {
1400 struct timeval tv;
1401
1402 tv.tv_sec = req->nv1;
1403 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1404
1405 req->result = select (0, 0, 0, 0, &tv);
1406 }
1407#endif
1408 break;
1409
1410 case EIO_UTIME:
1411 case EIO_FUTIME:
1412 {
1413 struct timeval tv[2];
1414 struct timeval *times;
1415
1416 if (req->nv1 != -1. || req->nv2 != -1.)
1417 {
1418 tv[0].tv_sec = req->nv1;
1419 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1420 tv[1].tv_sec = req->nv2;
1421 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1422
1423 times = tv;
1424 }
1425 else
1426 times = 0;
1427
1428
1429 req->result = req->type == EIO_FUTIME
1430 ? futimes (req->int1, times)
1431 : utimes (req->ptr1, times);
1432 }
1433 break;
1434
1435 case EIO_GROUP:
1436 abort (); /* handled in eio_request */
1437
1438 case EIO_NOP:
1439 req->result = 0;
1440 break;
1441
1442 case EIO_CUSTOM:
1443 ((void (*)(eio_req *))req->feed) (req);
1444 break;
1445
1446 default:
1447 req->result = -1;
1448 break;
1449 }
1450
1451 req->errorno = errno;
1452}
1453
1213#ifndef EIO_NO_WRAPPERS 1454#ifndef EIO_NO_WRAPPERS
1214 1455
1215eio_req *eio_nop (int pri, eio_cb cb, void *data) 1456eio_req *eio_nop (int pri, eio_cb cb, void *data)
1216{ 1457{
1217 REQ (EIO_NOP); SEND; 1458 REQ (EIO_NOP); SEND;
1228} 1469}
1229 1470
1230eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1471eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1231{ 1472{
1232 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1473 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1474}
1475
1476eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1477{
1478 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1479}
1480
1481eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1482{
1483 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1484}
1485
1486eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1487{
1488 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1233} 1489}
1234 1490
1235eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1491eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1236{ 1492{
1237 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1493 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1351eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1607eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1352{ 1608{
1353 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1609 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1354} 1610}
1355 1611
1356eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1612eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1357{ 1613{
1358 return eio__1path (EIO_READDIR, path, pri, cb, data); 1614 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1359} 1615}
1360 1616
1361eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1617eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1362{ 1618{
1363 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1619 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1364} 1620}
1365 1621
1366static eio_req * 1622static eio_req *
1367eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1623eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1368{ 1624{
1392eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1648eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1393{ 1649{
1394 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1650 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1395} 1651}
1396 1652
1653eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1654{
1655 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1656}
1657
1397#endif 1658#endif
1398 1659
1399eio_req *eio_grp (eio_cb cb, void *data) 1660eio_req *eio_grp (eio_cb cb, void *data)
1400{ 1661{
1401 const int pri = EIO_PRI_MAX; 1662 const int pri = EIO_PRI_MAX;
1427 1688
1428void eio_grp_add (eio_req *grp, eio_req *req) 1689void eio_grp_add (eio_req *grp, eio_req *req)
1429{ 1690{
1430 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1691 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1431 1692
1693 grp->flags |= EIO_FLAG_GROUPADD;
1694
1432 ++grp->size; 1695 ++grp->size;
1433 req->grp = grp; 1696 req->grp = grp;
1434 1697
1435 req->grp_prev = 0; 1698 req->grp_prev = 0;
1436 req->grp_next = grp->grp_first; 1699 req->grp_next = grp->grp_first;

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines