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 1.34 by root, Sat Jun 6 19:57:22 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 _DIRENT_HAVE_D_TYPE /* sigh */
86# define D_INO(de) (de)->d_fileno
87# define D_NAMLEN(de) (de)->d_namlen
88# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
89# define D_INO(de) (de)->d_ino
90# endif
91
92#ifdef _D_EXACT_NAMLEN
93# undef D_NAMLEN
94# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95#endif
96
97# ifdef _DIRENT_HAVE_D_TYPE
98# define D_TYPE(de) (de)->d_type
99# endif
79 100
80# ifndef EIO_STRUCT_DIRENT 101# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 102# define EIO_STRUCT_DIRENT struct dirent
82# endif 103# endif
83 104
96# else 117# else
97# error sendfile support requested but not available 118# error sendfile support requested but not available
98# endif 119# endif
99#endif 120#endif
100 121
122#ifndef D_TYPE
123# define D_TYPE(de) 0
124#endif
125#ifndef D_INO
126# define D_INO(de) 0
127#endif
128#ifndef D_NAMLEN
129# define D_NAMLEN(de) strlen ((de)->d_name)
130#endif
131
101/* number of seconds after which an idle threads exit */ 132/* number of seconds after which an idle threads exit */
102#define IDLE_TIMEOUT 10 133#define IDLE_TIMEOUT 10
103 134
104/* used for struct dirent, AIX doesn't provide it */ 135/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 136#ifndef NAME_MAX
155 if (wrk->dirp) \ 186 if (wrk->dirp) \
156 { \ 187 { \
157 closedir (wrk->dirp); \ 188 closedir (wrk->dirp); \
158 wrk->dirp = 0; \ 189 wrk->dirp = 0; \
159 } 190 }
191
160#define ETP_WORKER_COMMON \ 192#define ETP_WORKER_COMMON \
161 void *dbuf; \ 193 void *dbuf; \
162 DIR *dirp; 194 DIR *dirp;
163 195
164/*****************************************************************************/ 196/*****************************************************************************/
604 636
605static void grp_try_feed (eio_req *grp) 637static void grp_try_feed (eio_req *grp)
606{ 638{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 639 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 640 {
609 int old_len = grp->size; 641 grp->flags &= ~EIO_FLAG_GROUPADD;
610 642
611 EIO_FEED (grp); 643 EIO_FEED (grp);
612 644
613 /* stop if no progress has been made */ 645 /* stop if no progress has been made */
614 if (old_len == grp->size) 646 if (!(grp->flags & EIO_FLAG_GROUPADD))
615 { 647 {
616 grp->feed = 0; 648 grp->feed = 0;
617 break; 649 break;
618 } 650 }
619 } 651 }
736 768
737/*****************************************************************************/ 769/*****************************************************************************/
738/* work around various missing functions */ 770/* work around various missing functions */
739 771
740#if !HAVE_PREADWRITE 772#if !HAVE_PREADWRITE
773# undef pread
774# undef pwrite
741# define pread eio__pread 775# define pread eio__pread
742# define pwrite eio__pwrite 776# define pwrite eio__pwrite
743 777
744static ssize_t 778static ssize_t
745eio__pread (int fd, void *buf, size_t count, off_t offset) 779eio__pread (int fd, void *buf, size_t count, off_t offset)
765 799
766 X_LOCK (preadwritelock); 800 X_LOCK (preadwritelock);
767 ooffset = lseek (fd, 0, SEEK_CUR); 801 ooffset = lseek (fd, 0, SEEK_CUR);
768 lseek (fd, offset, SEEK_SET); 802 lseek (fd, offset, SEEK_SET);
769 res = write (fd, buf, count); 803 res = write (fd, buf, count);
770 lseek (fd, offset, SEEK_SET); 804 lseek (fd, ooffset, SEEK_SET);
771 X_UNLOCK (preadwritelock); 805 X_UNLOCK (preadwritelock);
772 806
773 return res; 807 return res;
774} 808}
775#endif 809#endif
776 810
777#ifndef HAVE_FUTIMES 811#ifndef HAVE_FUTIMES
778 812
813# undef utimes
814# undef futimes
779# define utimes(path,times) eio__utimes (path, times) 815# define utimes(path,times) eio__utimes (path, times)
780# define futimes(fd,times) eio__futimes (fd, times) 816# define futimes(fd,times) eio__futimes (fd, times)
781 817
782static int 818static int
783eio__utimes (const char *filename, const struct timeval times[2]) 819eio__utimes (const char *filename, const struct timeval times[2])
802} 838}
803 839
804#endif 840#endif
805 841
806#if !HAVE_FDATASYNC 842#if !HAVE_FDATASYNC
843# undef fdatasync
807# define fdatasync fsync 844# define fdatasync(fd) fsync (fd)
808#endif 845#endif
846
847/* sync_file_range always needs emulation */
848int
849eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850{
851#if HAVE_SYNC_FILE_RANGE
852 int res;
853
854 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
855 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
856 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
857 {
858 flags = 0
859 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
860 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
861 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
862 }
863
864 res = sync_file_range (fd, offset, nbytes, flags);
865
866 if (!res || errno != ENOSYS)
867 return res;
868#endif
869
870 /* even though we could play tricks with the flags, it's better to always
871 * call fdatasync, as thta matches the expectation of it's users best */
872 return fdatasync (fd);
873}
809 874
810#if !HAVE_READAHEAD 875#if !HAVE_READAHEAD
876# undef readahead
811# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 877# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
812 878
813static ssize_t 879static ssize_t
814eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 880eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
815{ 881{
923 } 989 }
924 990
925 return res; 991 return res;
926} 992}
927 993
994static int
995eio_dent_cmp (const void *a_, const void *b_)
996{
997 const eio_dirent *a = (const eio_dirent *)a_;
998 const eio_dirent *b = (const eio_dirent *)b_;
999
1000 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
1001 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */
1002}
1003
928/* read a full directory */ 1004/* read a full directory */
929static void 1005static void
930eio__scandir (eio_req *req, etp_worker *self) 1006eio__scandir (eio_req *req, etp_worker *self)
931{ 1007{
932 DIR *dirp; 1008 DIR *dirp;
933 EIO_STRUCT_DIRENT *entp; 1009 EIO_STRUCT_DIRENT *entp;
934 char *name, *names; 1010 unsigned char *name, *names;
935 int memlen = 4096; 1011 int namesalloc = 4096;
936 int memofs = 0; 1012 int namesoffs = 0;
1013 int flags = req->int1;
1014 eio_dirent *dents = 0;
1015 int dentalloc = 128;
937 int res = 0; 1016 int dentoffs = 0;
1017
1018 req->result = -1;
1019
1020 if (!(flags & EIO_READDIR_DENTS))
1021 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
938 1022
939 X_LOCK (wrklock); 1023 X_LOCK (wrklock);
940 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1024 /* the corresponding closedir is in ETP_WORKER_CLEAR */
941 self->dirp = dirp = opendir (req->ptr1); 1025 self->dirp = dirp = opendir (req->ptr1);
942 req->flags |= EIO_FLAG_PTR2_FREE; 1026 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
943 req->ptr2 = names = malloc (memlen); 1027 req->ptr1 = names = malloc (namesalloc);
1028 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
944 X_UNLOCK (wrklock); 1029 X_UNLOCK (wrklock);
945 1030
946 if (dirp && names) 1031 if (dirp && names && (!flags || dents))
947 for (;;) 1032 for (;;)
948 { 1033 {
949 errno = 0; 1034 errno = 0;
950 entp = readdir (dirp); 1035 entp = readdir (dirp);
951 1036
952 if (!entp) 1037 if (!entp)
1038 {
1039 if (errno)
1040 break;
1041
1042 /* sort etc. */
1043 req->int1 = flags;
1044 req->result = dentoffs;
1045
1046 if (dents)
1047 {
1048 eio_dirent *ent = dents + dentoffs;
1049
1050 while (ent > dents)
1051 (--ent)->name = names + (size_t)ent->name;
1052 }
1053
1054 if (flags & EIO_READDIR_STAT_ORDER
1055 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
1056 {
1057 /* pray your qsort doesn't use quicksort */
1058 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059 }
1060 else if (flags & EIO_READDIR_DIRS_FIRST)
1061 {
1062 /* in this case, all is known, and we just put dirs first and sort them */
1063 eio_dirent *ent = dents + dentoffs;
1064 eio_dirent *dir = dents;
1065
1066 while (ent > dir)
1067 {
1068 if (dir->type == DT_DIR)
1069 ++dir;
1070 else
1071 {
1072 --ent;
1073
1074 if (ent->type == DT_DIR)
1075 {
1076 eio_dirent tmp = *dir;
1077 *dir = *ent;
1078 *ent = tmp;
1079
1080 ++dir;
1081 }
1082 }
1083 }
1084
1085 /* now sort the dirs only */
1086 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1087 }
1088
1089 /* only provide the names array unless DENTS is specified */
1090 if (!(flags & EIO_READDIR_DENTS))
1091 {
1092 X_LOCK (wrklock);
1093 assert (!dents);
1094 req->ptr1 = 0;
1095 req->ptr2 = names;
1096 X_UNLOCK (wrklock);
1097 }
1098
953 break; 1099 break;
1100 }
954 1101
1102 /* now add the entry to our list(s) */
955 name = entp->d_name; 1103 name = entp->d_name;
956 1104
1105 /* skip . and .. entries */
957 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1106 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
958 { 1107 {
959 int len = strlen (name) + 1; 1108 int len = D_NAMLEN (entp) + 1;
960 1109
961 res++; 1110 while (expect_false (namesoffs + len > namesalloc))
962
963 while (memofs + len > memlen)
964 { 1111 {
965 memlen *= 2; 1112 namesalloc *= 2;
966 X_LOCK (wrklock); 1113 X_LOCK (wrklock);
967 req->ptr2 = names = realloc (names, memlen); 1114 req->ptr1 = names = realloc (names, namesalloc);
968 X_UNLOCK (wrklock); 1115 X_UNLOCK (wrklock);
969 1116
970 if (!names) 1117 if (!names)
971 break; 1118 break;
972 } 1119 }
973 1120
974 memcpy (names + memofs, name, len); 1121 memcpy (names + namesoffs, name, len);
1122
1123 if (dents)
1124 {
1125 struct eio_dirent *ent;
1126
1127 if (expect_false (dentoffs == dentalloc))
1128 {
1129 dentalloc *= 2;
1130 X_LOCK (wrklock);
1131 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132 X_UNLOCK (wrklock);
1133
1134 if (!dents)
1135 break;
1136 }
1137
1138 ent = dents + dentoffs;
1139
1140 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
1141 ent->namelen = len - 1;
1142 ent->inode = D_INO (entp);
1143
1144 switch (D_TYPE (entp))
1145 {
1146 default:
1147 ent->type = EIO_DT_UNKNOWN;
1148 flags |= EIO_READDIR_FOUND_UNKNOWN;
1149 break;
1150
1151 #ifdef DT_FIFO
1152 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1153 #endif
1154 #ifdef DT_CHR
1155 case DT_CHR: ent->type = EIO_DT_CHR; break;
1156 #endif
1157 #ifdef DT_MPC
1158 case DT_MPC: ent->type = EIO_DT_MPC; break;
1159 #endif
1160 #ifdef DT_DIR
1161 case DT_DIR: ent->type = EIO_DT_DIR; break;
1162 #endif
1163 #ifdef DT_NAM
1164 case DT_NAM: ent->type = EIO_DT_NAM; break;
1165 #endif
1166 #ifdef DT_BLK
1167 case DT_BLK: ent->type = EIO_DT_BLK; break;
1168 #endif
1169 #ifdef DT_MPB
1170 case DT_MPB: ent->type = EIO_DT_MPB; break;
1171 #endif
1172 #ifdef DT_REG
1173 case DT_REG: ent->type = EIO_DT_REG; break;
1174 #endif
1175 #ifdef DT_NWK
1176 case DT_NWK: ent->type = EIO_DT_NWK; break;
1177 #endif
1178 #ifdef DT_CMP
1179 case DT_CMP: ent->type = EIO_DT_CMP; break;
1180 #endif
1181 #ifdef DT_LNK
1182 case DT_LNK: ent->type = EIO_DT_LNK; break;
1183 #endif
1184 #ifdef DT_SOCK
1185 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1186 #endif
1187 #ifdef DT_DOOR
1188 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1189 #endif
1190 #ifdef DT_WHT
1191 case DT_WHT: ent->type = EIO_DT_WHT; break;
1192 #endif
1193 }
1194
1195 ent->score = 0;
1196
1197 if (flags & EIO_READDIR_DIRS_FIRST)
1198 {
1199 if (ent->type == EIO_DT_UNKNOWN)
1200 {
1201 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202 ent->score = 98;
1203 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1204 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
1205 }
1206 else if (ent->type == EIO_DT_DIR)
1207 ent->score = 100;
1208 }
1209 }
1210
975 memofs += len; 1211 namesoffs += len;
1212 ++dentoffs;
976 } 1213 }
977 } 1214 }
978 1215 else
979 if (errno)
980 res = -1;
981
982 req->result = res; 1216 req->result = -1;
983} 1217}
984 1218
985#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1219#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1220# undef msync
986# define msync(a,b,c) ENOSYS 1221# define msync(a,b,c) ((errno = ENOSYS), -1)
987#endif 1222#endif
988 1223
989int 1224int
990eio__mtouch (void *mem, size_t len, int flags) 1225eio__mtouch (void *mem, size_t len, int flags)
991{ 1226{
1185 case EIO_SYNC: req->result = 0; sync (); break; 1420 case EIO_SYNC: req->result = 0; sync (); break;
1186 case EIO_FSYNC: req->result = fsync (req->int1); break; 1421 case EIO_FSYNC: req->result = fsync (req->int1); break;
1187 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1422 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1188 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1423 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; 1424 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1425 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1190 1426
1191 case EIO_READDIR: eio__scandir (req, self); break; 1427 case EIO_READDIR: eio__scandir (req, self); break;
1192 1428
1193 case EIO_BUSY: 1429 case EIO_BUSY:
1194#ifdef _WIN32 1430#ifdef _WIN32
1226 1462
1227 req->result = req->type == EIO_FUTIME 1463 req->result = req->type == EIO_FUTIME
1228 ? futimes (req->int1, times) 1464 ? futimes (req->int1, times)
1229 : utimes (req->ptr1, times); 1465 : utimes (req->ptr1, times);
1230 } 1466 }
1467 break;
1231 1468
1232 case EIO_GROUP: 1469 case EIO_GROUP:
1233 abort (); /* handled in eio_request */ 1470 abort (); /* handled in eio_request */
1234 1471
1235 case EIO_NOP: 1472 case EIO_NOP:
1278eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1515eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1279{ 1516{
1280 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1517 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1281} 1518}
1282 1519
1520eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1521{
1522 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1523}
1524
1283eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1525eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1284{ 1526{
1285 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1527 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1286} 1528}
1287 1529
1399eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1641eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1400{ 1642{
1401 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1643 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1402} 1644}
1403 1645
1404eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1646eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1405{ 1647{
1406 return eio__1path (EIO_READDIR, path, pri, cb, data); 1648 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1407} 1649}
1408 1650
1409eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1651eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1410{ 1652{
1411 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1653 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1480 1722
1481void eio_grp_add (eio_req *grp, eio_req *req) 1723void eio_grp_add (eio_req *grp, eio_req *req)
1482{ 1724{
1483 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1725 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1484 1726
1727 grp->flags |= EIO_FLAG_GROUPADD;
1728
1485 ++grp->size; 1729 ++grp->size;
1486 req->grp = grp; 1730 req->grp = grp;
1487 1731
1488 req->grp_prev = 0; 1732 req->grp_prev = 0;
1489 req->grp_next = grp->grp_first; 1733 req->grp_next = grp->grp_first;

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines