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 1.24 by root, Tue Sep 30 13:17:32 2008 UTC vs.
Revision 1.32 by root, Sat Jun 6 18:06:55 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
155 if (wrk->dirp) \ 178 if (wrk->dirp) \
156 { \ 179 { \
157 closedir (wrk->dirp); \ 180 closedir (wrk->dirp); \
158 wrk->dirp = 0; \ 181 wrk->dirp = 0; \
159 } 182 }
183
160#define ETP_WORKER_COMMON \ 184#define ETP_WORKER_COMMON \
161 void *dbuf; \ 185 void *dbuf; \
162 DIR *dirp; 186 DIR *dirp;
163 187
164/*****************************************************************************/ 188/*****************************************************************************/
604 628
605static void grp_try_feed (eio_req *grp) 629static void grp_try_feed (eio_req *grp)
606{ 630{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 631 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 632 {
609 int old_len = grp->size; 633 grp->flags &= ~EIO_FLAG_GROUPADD;
610 634
611 EIO_FEED (grp); 635 EIO_FEED (grp);
612 636
613 /* stop if no progress has been made */ 637 /* stop if no progress has been made */
614 if (old_len == grp->size) 638 if (!(grp->flags & EIO_FLAG_GROUPADD))
615 { 639 {
616 grp->feed = 0; 640 grp->feed = 0;
617 break; 641 break;
618 } 642 }
619 } 643 }
736 760
737/*****************************************************************************/ 761/*****************************************************************************/
738/* work around various missing functions */ 762/* work around various missing functions */
739 763
740#if !HAVE_PREADWRITE 764#if !HAVE_PREADWRITE
765# undef pread
766# undef pwrite
741# define pread eio__pread 767# define pread eio__pread
742# define pwrite eio__pwrite 768# define pwrite eio__pwrite
743 769
744static ssize_t 770static ssize_t
745eio__pread (int fd, void *buf, size_t count, off_t offset) 771eio__pread (int fd, void *buf, size_t count, off_t offset)
765 791
766 X_LOCK (preadwritelock); 792 X_LOCK (preadwritelock);
767 ooffset = lseek (fd, 0, SEEK_CUR); 793 ooffset = lseek (fd, 0, SEEK_CUR);
768 lseek (fd, offset, SEEK_SET); 794 lseek (fd, offset, SEEK_SET);
769 res = write (fd, buf, count); 795 res = write (fd, buf, count);
770 lseek (fd, offset, SEEK_SET); 796 lseek (fd, ooffset, SEEK_SET);
771 X_UNLOCK (preadwritelock); 797 X_UNLOCK (preadwritelock);
772 798
773 return res; 799 return res;
774} 800}
775#endif 801#endif
776 802
777#ifndef HAVE_FUTIMES 803#ifndef HAVE_FUTIMES
778 804
805# undef utimes
806# undef futimes
779# define utimes(path,times) eio__utimes (path, times) 807# define utimes(path,times) eio__utimes (path, times)
780# define futimes(fd,times) eio__futimes (fd, times) 808# define futimes(fd,times) eio__futimes (fd, times)
781 809
782static int 810static int
783eio__utimes (const char *filename, const struct timeval times[2]) 811eio__utimes (const char *filename, const struct timeval times[2])
802} 830}
803 831
804#endif 832#endif
805 833
806#if !HAVE_FDATASYNC 834#if !HAVE_FDATASYNC
835# undef fdatasync
807# define fdatasync fsync 836# define fdatasync(fd) fsync (fd)
808#endif 837#endif
838
839/* sync_file_range always needs emulation */
840int
841eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
842{
843#if HAVE_SYNC_FILE_RANGE
844 int res;
845
846 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
847 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
848 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
849 {
850 flags = 0
851 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
852 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
853 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
854 }
855
856 res = sync_file_range (fd, offset, nbytes, flags);
857
858 if (!res || errno != ENOSYS)
859 return res;
860#endif
861
862 /* even though we could play tricks with the flags, it's better to always
863 * call fdatasync, as thta matches the expectation of it's users best */
864 return fdatasync (fd);
865}
809 866
810#if !HAVE_READAHEAD 867#if !HAVE_READAHEAD
868# undef readahead
811# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 869# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
812 870
813static ssize_t 871static ssize_t
814eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 872eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
815{ 873{
923 } 981 }
924 982
925 return res; 983 return res;
926} 984}
927 985
986static int
987eio_dent_cmp (const void *a_, const void *b_)
988{
989 const eio_dirent *a = (const eio_dirent *)a_;
990 const eio_dirent *b = (const eio_dirent *)b_;
991
992 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
993 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */
994}
995
928/* read a full directory */ 996/* read a full directory */
929static void 997static void
930eio__scandir (eio_req *req, etp_worker *self) 998eio__scandir (eio_req *req, etp_worker *self)
931{ 999{
932 DIR *dirp; 1000 DIR *dirp;
933 EIO_STRUCT_DIRENT *entp; 1001 EIO_STRUCT_DIRENT *entp;
934 char *name, *names; 1002 unsigned char *name, *names;
935 int memlen = 4096; 1003 int namesalloc = 4096;
936 int memofs = 0; 1004 int namesoffs = 0;
1005 int flags = req->int1;
1006 eio_dirent *dents = 0;
1007 int dentalloc = 128;
937 int res = 0; 1008 int dentoffs = 0;
1009
1010 req->result = -1;
1011
1012 if (!(flags & EIO_READDIR_DENTS))
1013 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
938 1014
939 X_LOCK (wrklock); 1015 X_LOCK (wrklock);
940 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */
941 self->dirp = dirp = opendir (req->ptr1); 1017 self->dirp = dirp = opendir (req->ptr1);
942 req->flags |= EIO_FLAG_PTR2_FREE; 1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
943 req->ptr2 = names = malloc (memlen); 1019 req->ptr1 = names = malloc (namesalloc);
1020 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
944 X_UNLOCK (wrklock); 1021 X_UNLOCK (wrklock);
945 1022
946 if (dirp && names) 1023 if (dirp && names && (!flags || dents))
947 for (;;) 1024 for (;;)
948 { 1025 {
949 errno = 0; 1026 errno = 0;
950 entp = readdir (dirp); 1027 entp = readdir (dirp);
951 1028
952 if (!entp) 1029 if (!entp)
1030 {
1031 if (errno)
1032 break;
1033
1034 /* sort etc. */
1035 req->int1 = flags;
1036 req->result = dentoffs;
1037
1038 if (dents)
1039 {
1040 eio_dirent *ent = dents + dentoffs;
1041
1042 while (ent > dents)
1043 (--ent)->name = names + (size_t)ent->name;
1044 }
1045
1046 if (flags & EIO_READDIR_STAT_ORDER
1047 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
1048 {
1049 /* pray your qsort doesn't use quicksort */
1050 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1051 }
1052 else if (flags & EIO_READDIR_DIRS_FIRST)
1053 {
1054 /* in this case, all is known, and we just put dirs first and sort them */
1055 eio_dirent *ent = dents + dentoffs;
1056 eio_dirent *dir = dents;
1057
1058 while (ent > dir)
1059 {
1060 if (dir->type == DT_DIR)
1061 ++dir;
1062 else
1063 {
1064 --ent;
1065
1066 if (ent->type == DT_DIR)
1067 {
1068 eio_dirent tmp = *dir;
1069 *dir = *ent;
1070 *ent = tmp;
1071
1072 ++dir;
1073 }
1074 }
1075 }
1076
1077 /* now sort the dirs only */
1078 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1079 }
1080
1081 /* only provide the names array unless DENTS is specified */
1082 if (!(flags & EIO_READDIR_DENTS))
1083 {
1084 X_LOCK (wrklock);
1085 assert (!dents);
1086 req->ptr1 = 0;
1087 req->ptr2 = names;
1088 X_UNLOCK (wrklock);
1089 }
1090
953 break; 1091 break;
1092 }
954 1093
1094 /* now add the entry to our list(s) */
955 name = entp->d_name; 1095 name = entp->d_name;
956 1096
1097 /* skip . and .. entries */
957 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1098 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
958 { 1099 {
959 int len = strlen (name) + 1; 1100 int len = strlen (name) + 1;
960 1101
961 res++; 1102 while (expect_false (namesoffs + len > namesalloc))
962
963 while (memofs + len > memlen)
964 { 1103 {
965 memlen *= 2; 1104 namesalloc *= 2;
966 X_LOCK (wrklock); 1105 X_LOCK (wrklock);
967 req->ptr2 = names = realloc (names, memlen); 1106 req->ptr1 = names = realloc (names, namesalloc);
968 X_UNLOCK (wrklock); 1107 X_UNLOCK (wrklock);
969 1108
970 if (!names) 1109 if (!names)
971 break; 1110 break;
972 } 1111 }
973 1112
974 memcpy (names + memofs, name, len); 1113 memcpy (names + namesoffs, name, len);
1114
1115 if (dents)
1116 {
1117 struct eio_dirent *ent;
1118
1119 if (expect_false (dentoffs == dentalloc))
1120 {
1121 dentalloc *= 2;
1122 X_LOCK (wrklock);
1123 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124 X_UNLOCK (wrklock);
1125
1126 if (!dents)
1127 break;
1128 }
1129
1130 ent = dents + dentoffs;
1131
1132 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
1133 ent->namelen = len - 1;
1134 ent->inode = D_INO (entp);
1135
1136 switch (D_TYPE (entp))
1137 {
1138 default:
1139 ent->type = EIO_DT_UNKNOWN;
1140 flags |= EIO_READDIR_FOUND_UNKNOWN;
1141 break;
1142
1143 #ifdef DT_FIFO
1144 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145 #endif
1146 #ifdef DT_CHR
1147 case DT_CHR: ent->type = EIO_DT_CHR; break;
1148 #endif
1149 #ifdef DT_DIR
1150 case DT_DIR: ent->type = EIO_DT_DIR; break;
1151 #endif
1152 #ifdef DT_BLK
1153 case DT_BLK: ent->type = EIO_DT_BLK; break;
1154 #endif
1155 #ifdef DT_REG
1156 case DT_REG: ent->type = EIO_DT_REG; break;
1157 #endif
1158 #ifdef DT_LNK
1159 case DT_LNK: ent->type = EIO_DT_LNK; break;
1160 #endif
1161 #ifdef DT_SOCK
1162 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163 #endif
1164 #ifdef DT_WHT
1165 case DT_WHT: ent->type = EIO_DT_WHT; break;
1166 #endif
1167 }
1168
1169 ent->score = 0;
1170
1171 if (flags & EIO_READDIR_DIRS_FIRST)
1172 {
1173 if (ent->type == EIO_DT_UNKNOWN)
1174 {
1175 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176 ent->score = 98;
1177 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1178 ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
1179 }
1180 else if (ent->type == DT_DIR)
1181 ent->score = 100;
1182 }
1183 }
1184
975 memofs += len; 1185 namesoffs += len;
1186 ++dentoffs;
976 } 1187 }
977 } 1188 }
978 1189 else
979 if (errno)
980 res = -1;
981
982 req->result = res; 1190 req->result = -1;
983} 1191}
984 1192
985#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1193#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1194# undef msync
986# define msync(a,b,c) ENOSYS 1195# define msync(a,b,c) ((errno = ENOSYS), -1)
987#endif 1196#endif
988 1197
989int 1198int
990eio__mtouch (void *mem, size_t len, int flags) 1199eio__mtouch (void *mem, size_t len, int flags)
991{ 1200{
1185 case EIO_SYNC: req->result = 0; sync (); break; 1394 case EIO_SYNC: req->result = 0; sync (); break;
1186 case EIO_FSYNC: req->result = fsync (req->int1); break; 1395 case EIO_FSYNC: req->result = fsync (req->int1); break;
1187 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1396 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1188 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1397 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1189 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1398 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1399 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1190 1400
1191 case EIO_READDIR: eio__scandir (req, self); break; 1401 case EIO_READDIR: eio__scandir (req, self); break;
1192 1402
1193 case EIO_BUSY: 1403 case EIO_BUSY:
1194#ifdef _WIN32 1404#ifdef _WIN32
1226 1436
1227 req->result = req->type == EIO_FUTIME 1437 req->result = req->type == EIO_FUTIME
1228 ? futimes (req->int1, times) 1438 ? futimes (req->int1, times)
1229 : utimes (req->ptr1, times); 1439 : utimes (req->ptr1, times);
1230 } 1440 }
1441 break;
1231 1442
1232 case EIO_GROUP: 1443 case EIO_GROUP:
1233 abort (); /* handled in eio_request */ 1444 abort (); /* handled in eio_request */
1234 1445
1235 case EIO_NOP: 1446 case EIO_NOP:
1278eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1489eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1279{ 1490{
1280 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1491 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1281} 1492}
1282 1493
1494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1495{
1496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1497}
1498
1283eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1499eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1284{ 1500{
1285 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1501 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1286} 1502}
1287 1503
1399eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1615eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1400{ 1616{
1401 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1617 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1402} 1618}
1403 1619
1404eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1620eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1405{ 1621{
1406 return eio__1path (EIO_READDIR, path, pri, cb, data); 1622 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1407} 1623}
1408 1624
1409eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1625eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1410{ 1626{
1411 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1627 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1480 1696
1481void eio_grp_add (eio_req *grp, eio_req *req) 1697void eio_grp_add (eio_req *grp, eio_req *req)
1482{ 1698{
1483 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1699 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1484 1700
1701 grp->flags |= EIO_FLAG_GROUPADD;
1702
1485 ++grp->size; 1703 ++grp->size;
1486 req->grp = grp; 1704 req->grp = grp;
1487 1705
1488 req->grp_prev = 0; 1706 req->grp_prev = 0;
1489 req->grp_next = grp->grp_first; 1707 req->grp_next = grp->grp_first;

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines