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

Comparing libeio/eio.c (file contents):
Revision 1.32 by root, Sat Jun 6 18:06:55 2009 UTC vs.
Revision 1.49 by root, Sat Jan 2 14:24:32 2010 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010 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 *
73#else 73#else
74 74
75# include "config.h" 75# include "config.h"
76# include <sys/time.h> 76# include <sys/time.h>
77# include <sys/select.h> 77# include <sys/select.h>
78# include <sys/mman.h>
78# include <unistd.h> 79# include <unistd.h>
79# include <utime.h> 80# include <utime.h>
80# include <signal.h> 81# include <signal.h>
81# include <dirent.h> 82# include <dirent.h>
82 83
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 85# if __freebsd || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */
85# define D_INO(de) (de)->d_fileno 87# define D_INO(de) (de)->d_fileno
86# define _DIRENT_HAVE_D_TYPE /* sigh */ 88# define D_NAMLEN(de) (de)->d_namlen
87# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 89# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
88# define D_INO(de) (de)->d_ino 90# define D_INO(de) (de)->d_ino
89# endif 91# endif
92
93#ifdef _D_EXACT_NAMLEN
94# undef D_NAMLEN
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
90 97
91# ifdef _DIRENT_HAVE_D_TYPE 98# ifdef _DIRENT_HAVE_D_TYPE
92# define D_TYPE(de) (de)->d_type 99# define D_TYPE(de) (de)->d_type
93# endif 100# endif
94 101
99#endif 106#endif
100 107
101#if HAVE_SENDFILE 108#if HAVE_SENDFILE
102# if __linux 109# if __linux
103# include <sys/sendfile.h> 110# include <sys/sendfile.h>
104# elif __freebsd 111# elif __freebsd || defined __APPLE__
105# include <sys/socket.h> 112# include <sys/socket.h>
106# include <sys/uio.h> 113# include <sys/uio.h>
107# elif __hpux 114# elif __hpux
108# include <sys/socket.h> 115# include <sys/socket.h>
109# elif __solaris /* not yet */ 116# elif __solaris
110# include <sys/sendfile.h> 117# include <sys/sendfile.h>
111# else 118# else
112# error sendfile support requested but not available 119# error sendfile support requested but not available
113# endif 120# endif
114#endif 121#endif
115 122
116#ifndef D_TYPE 123#ifndef D_TYPE
117# define D_TYPE(de) 0 124# define D_TYPE(de) 0
118#endif 125#endif
119
120#ifndef D_INO 126#ifndef D_INO
121# define D_INO(de) 0 127# define D_INO(de) 0
128#endif
129#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name)
122#endif 131#endif
123 132
124/* number of seconds after which an idle threads exit */ 133/* number of seconds after which an idle threads exit */
125#define IDLE_TIMEOUT 10 134#define IDLE_TIMEOUT 10
126 135
187 196
188/*****************************************************************************/ 197/*****************************************************************************/
189 198
190#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
191 200
192/* calculcate time difference in ~1/EIO_TICKS of a second */ 201/* calculate time difference in ~1/EIO_TICKS of a second */
193static int tvdiff (struct timeval *tv1, struct timeval *tv2) 202static int tvdiff (struct timeval *tv1, struct timeval *tv2)
194{ 203{
195 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
196 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 205 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
197} 206}
589} 598}
590 599
591static void etp_set_max_poll_time (double nseconds) 600static void etp_set_max_poll_time (double nseconds)
592{ 601{
593 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 602 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
594 max_poll_time = nseconds; 603 max_poll_time = nseconds * EIO_TICKS;
595 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
596} 605}
597 606
598static void etp_set_max_poll_reqs (unsigned int maxreqs) 607static void etp_set_max_poll_reqs (unsigned int maxreqs)
599{ 608{
858 if (!res || errno != ENOSYS) 867 if (!res || errno != ENOSYS)
859 return res; 868 return res;
860#endif 869#endif
861 870
862 /* even though we could play tricks with the flags, it's better to always 871 /* 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 */ 872 * call fdatasync, as that matches the expectation of its users best */
864 return fdatasync (fd); 873 return fdatasync (fd);
865} 874}
866 875
867#if !HAVE_READAHEAD 876#if !HAVE_READAHEAD
868# undef readahead 877# undef readahead
914 if (res < 0 && sbytes) 923 if (res < 0 && sbytes)
915 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */
916 res = sbytes; 925 res = sbytes;
917 } 926 }
918 927
928# elif defined (__APPLE__)
929
930 {
931 off_t sbytes = count;
932 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
933
934 if (res < 0 && errno == EAGAIN && sbytes)
935 res = sbytes;
936 }
937
919# elif __hpux 938# elif __hpux
920 res = sendfile (ofd, ifd, offset, count, 0, 0); 939 res = sendfile (ofd, ifd, offset, count, 0, 0);
921 940
922# elif __solaris 941# elif __solaris
923 { 942 {
934 if (res < 0 && sbytes) 953 if (res < 0 && sbytes)
935 res = sbytes; 954 res = sbytes;
936 } 955 }
937 956
938# endif 957# endif
958
959#elif defined (_WIN32)
960
961 /* does not work, just for documentation of what would need to be done */
962 {
963 HANDLE h = TO_SOCKET (ifd);
964 SetFilePointer (h, offset, 0, FILE_BEGIN);
965 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
966 }
967
939#else 968#else
940 res = -1; 969 res = -1;
941 errno = ENOSYS; 970 errno = ENOSYS;
942#endif 971#endif
943 972
944 if (res < 0 973 if (res < 0
945 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 974 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
975 /* BSDs */
976#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
977 || errno == ENOTSUP
978#endif
979 || errno == EOPNOTSUPP /* BSDs */
946#if __solaris 980#if __solaris
947 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 981 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
948#endif 982#endif
949 ) 983 )
950 ) 984 )
981 } 1015 }
982 1016
983 return res; 1017 return res;
984} 1018}
985 1019
986static int 1020static signed char
987eio_dent_cmp (const void *a_, const void *b_) 1021eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
988{ 1022{
989 const eio_dirent *a = (const eio_dirent *)a_; 1023 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
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 */ 1024 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1025}
1026
1027#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1028
1029#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1030#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1031
1032static void
1033eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1034{
1035 unsigned char bits [9 + sizeof (ino_t) * 8];
1036 unsigned char *bit = bits;
1037
1038 assert (CHAR_BIT == 8);
1039 assert (sizeof (eio_dirent) * 8 < 256);
1040 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1041 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1042
1043 if (size <= EIO_SORT_FAST)
1044 return;
1045
1046 /* first prepare an array of bits to test in our radix sort */
1047 /* try to take endianness into account, as well as differences in ino_t sizes */
1048 /* inode_bits must contain all inodes ORed together */
1049 /* which is used to skip bits that are 0 everywhere, which is very common */
1050 {
1051 ino_t endianness;
1052 int i, j;
1053
1054 /* we store the byte offset of byte n into byte n of "endianness" */
1055 for (i = 0; i < sizeof (ino_t); ++i)
1056 ((unsigned char *)&endianness)[i] = i;
1057
1058 *bit++ = 0;
1059
1060 for (i = 0; i < sizeof (ino_t); ++i)
1061 {
1062 /* shifting off the byte offsets out of "endianness" */
1063 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1064 endianness >>= 8;
1065
1066 for (j = 0; j < 8; ++j)
1067 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1068 *bit++ = offs + j;
1069 }
1070
1071 for (j = 0; j < 8; ++j)
1072 if (score_bits & (1 << j))
1073 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1074 }
1075
1076 /* now actually do the sorting (a variant of MSD radix sort) */
1077 {
1078 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1079 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1080 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1081 int stk_idx = 0;
1082
1083 base_stk [stk_idx] = dents;
1084 end_stk [stk_idx] = dents + size;
1085 bit_stk [stk_idx] = bit - 1;
1086
1087 do
1088 {
1089 base = base_stk [stk_idx];
1090 end = end_stk [stk_idx];
1091 bit = bit_stk [stk_idx];
1092
1093 for (;;)
1094 {
1095 unsigned char O = *bit >> 3;
1096 unsigned char M = 1 << (*bit & 7);
1097
1098 eio_dirent *a = base;
1099 eio_dirent *b = end;
1100
1101 if (b - a < EIO_SORT_CUTOFF)
1102 break;
1103
1104 /* now bit-partition the array on the bit */
1105 /* this ugly asymmetric loop seems to perform much better than typical */
1106 /* partition algos found in the literature */
1107 do
1108 if (!(((unsigned char *)a)[O] & M))
1109 ++a;
1110 else if (!(((unsigned char *)--b)[O] & M))
1111 {
1112 eio_dirent tmp = *a; *a = *b; *b = tmp;
1113 ++a;
1114 }
1115 while (b > a);
1116
1117 /* next bit, or stop, if no bits left in this path */
1118 if (!*--bit)
1119 break;
1120
1121 base_stk [stk_idx] = a;
1122 end_stk [stk_idx] = end;
1123 bit_stk [stk_idx] = bit;
1124 ++stk_idx;
1125
1126 end = a;
1127 }
1128 }
1129 while (stk_idx--);
1130 }
1131}
1132
1133static void
1134eio_dent_insertion_sort (eio_dirent *dents, int size)
1135{
1136 /* first move the smallest element to the front, to act as a sentinel */
1137 {
1138 int i;
1139 eio_dirent *min = dents;
1140
1141 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1142 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1143 if (EIO_DENT_CMP (dents [i], <, *min))
1144 min = &dents [i];
1145
1146 /* swap elements 0 and j (minimum) */
1147 {
1148 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1149 }
1150 }
1151
1152 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1153 {
1154 eio_dirent *i, *j;
1155
1156 for (i = dents + 1; i < dents + size; ++i)
1157 {
1158 eio_dirent value = *i;
1159
1160 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1161 j [1] = j [0];
1162
1163 j [1] = value;
1164 }
1165 }
1166}
1167
1168static void
1169eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1170{
1171 if (size <= 1)
1172 return; /* our insertion sort relies on size > 0 */
1173
1174 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1175 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1176 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1177
1178 /* use an insertion sort at the end, or for small arrays, */
1179 /* as insertion sort is more efficient for small partitions */
1180 eio_dent_insertion_sort (dents, size);
994} 1181}
995 1182
996/* read a full directory */ 1183/* read a full directory */
997static void 1184static void
998eio__scandir (eio_req *req, etp_worker *self) 1185eio__scandir (eio_req *req, etp_worker *self)
999{ 1186{
1000 DIR *dirp; 1187 DIR *dirp;
1001 EIO_STRUCT_DIRENT *entp; 1188 EIO_STRUCT_DIRENT *entp;
1002 unsigned char *name, *names; 1189 char *name, *names;
1003 int namesalloc = 4096; 1190 int namesalloc = 4096;
1004 int namesoffs = 0; 1191 int namesoffs = 0;
1005 int flags = req->int1; 1192 int flags = req->int1;
1006 eio_dirent *dents = 0; 1193 eio_dirent *dents = 0;
1007 int dentalloc = 128; 1194 int dentalloc = 128;
1008 int dentoffs = 0; 1195 int dentoffs = 0;
1196 ino_t inode_bits = 0;
1009 1197
1010 req->result = -1; 1198 req->result = -1;
1011 1199
1012 if (!(flags & EIO_READDIR_DENTS)) 1200 if (!(flags & EIO_READDIR_DENTS))
1013 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1201 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1014 1202
1015 X_LOCK (wrklock); 1203 X_LOCK (wrklock);
1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1204 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1017 self->dirp = dirp = opendir (req->ptr1); 1205 self->dirp = dirp = opendir (req->ptr1);
1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1206 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1019 req->ptr1 = names = malloc (namesalloc);
1020 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1207 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1208 req->ptr2 = names = malloc (namesalloc);
1021 X_UNLOCK (wrklock); 1209 X_UNLOCK (wrklock);
1022 1210
1023 if (dirp && names && (!flags || dents)) 1211 if (dirp && names && (!flags || dents))
1024 for (;;) 1212 for (;;)
1025 { 1213 {
1033 1221
1034 /* sort etc. */ 1222 /* sort etc. */
1035 req->int1 = flags; 1223 req->int1 = flags;
1036 req->result = dentoffs; 1224 req->result = dentoffs;
1037 1225
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 1226 if (flags & EIO_READDIR_STAT_ORDER)
1047 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1227 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
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) 1228 else if (flags & EIO_READDIR_DIRS_FIRST)
1229 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1230 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1231 else
1053 { 1232 {
1054 /* in this case, all is known, and we just put dirs first and sort them */ 1233 /* in this case, all is known, and we just put dirs first and sort them */
1055 eio_dirent *ent = dents + dentoffs; 1234 eio_dirent *oth = dents + dentoffs;
1056 eio_dirent *dir = dents; 1235 eio_dirent *dir = dents;
1057 1236
1237 /* now partition dirs to the front, and non-dirs to the back */
1238 /* by walking from both sides and swapping if necessary */
1239 /* also clear score, so it doesn't influence sorting */
1058 while (ent > dir) 1240 while (oth > dir)
1059 { 1241 {
1060 if (dir->type == DT_DIR) 1242 if (dir->type == EIO_DT_DIR)
1061 ++dir; 1243 ++dir;
1062 else
1063 {
1064 --ent;
1065
1066 if (ent->type == DT_DIR) 1244 else if ((--oth)->type == EIO_DT_DIR)
1067 { 1245 {
1068 eio_dirent tmp = *dir; 1246 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1069 *dir = *ent;
1070 *ent = tmp;
1071 1247
1072 ++dir; 1248 ++dir;
1073 } 1249 }
1074 } 1250 }
1075 }
1076 1251
1077 /* now sort the dirs only */ 1252 /* now sort the dirs only */
1078 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); 1253 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1079 } 1254 }
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 1255
1091 break; 1256 break;
1092 } 1257 }
1093 1258
1094 /* now add the entry to our list(s) */ 1259 /* now add the entry to our list(s) */
1095 name = entp->d_name; 1260 name = entp->d_name;
1096 1261
1097 /* skip . and .. entries */ 1262 /* skip . and .. entries */
1098 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1263 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1099 { 1264 {
1100 int len = strlen (name) + 1; 1265 int len = D_NAMLEN (entp) + 1;
1101 1266
1102 while (expect_false (namesoffs + len > namesalloc)) 1267 while (expect_false (namesoffs + len > namesalloc))
1103 { 1268 {
1104 namesalloc *= 2; 1269 namesalloc *= 2;
1105 X_LOCK (wrklock); 1270 X_LOCK (wrklock);
1106 req->ptr1 = names = realloc (names, namesalloc); 1271 req->ptr2 = names = realloc (names, namesalloc);
1107 X_UNLOCK (wrklock); 1272 X_UNLOCK (wrklock);
1108 1273
1109 if (!names) 1274 if (!names)
1110 break; 1275 break;
1111 } 1276 }
1118 1283
1119 if (expect_false (dentoffs == dentalloc)) 1284 if (expect_false (dentoffs == dentalloc))
1120 { 1285 {
1121 dentalloc *= 2; 1286 dentalloc *= 2;
1122 X_LOCK (wrklock); 1287 X_LOCK (wrklock);
1123 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1288 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124 X_UNLOCK (wrklock); 1289 X_UNLOCK (wrklock);
1125 1290
1126 if (!dents) 1291 if (!dents)
1127 break; 1292 break;
1128 } 1293 }
1129 1294
1130 ent = dents + dentoffs; 1295 ent = dents + dentoffs;
1131 1296
1132 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1297 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1133 ent->namelen = len - 1; 1298 ent->namelen = len - 1;
1134 ent->inode = D_INO (entp); 1299 ent->inode = D_INO (entp);
1300
1301 inode_bits |= ent->inode;
1135 1302
1136 switch (D_TYPE (entp)) 1303 switch (D_TYPE (entp))
1137 { 1304 {
1138 default: 1305 default:
1139 ent->type = EIO_DT_UNKNOWN; 1306 ent->type = EIO_DT_UNKNOWN;
1144 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1311 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145 #endif 1312 #endif
1146 #ifdef DT_CHR 1313 #ifdef DT_CHR
1147 case DT_CHR: ent->type = EIO_DT_CHR; break; 1314 case DT_CHR: ent->type = EIO_DT_CHR; break;
1148 #endif 1315 #endif
1316 #ifdef DT_MPC
1317 case DT_MPC: ent->type = EIO_DT_MPC; break;
1318 #endif
1149 #ifdef DT_DIR 1319 #ifdef DT_DIR
1150 case DT_DIR: ent->type = EIO_DT_DIR; break; 1320 case DT_DIR: ent->type = EIO_DT_DIR; break;
1321 #endif
1322 #ifdef DT_NAM
1323 case DT_NAM: ent->type = EIO_DT_NAM; break;
1151 #endif 1324 #endif
1152 #ifdef DT_BLK 1325 #ifdef DT_BLK
1153 case DT_BLK: ent->type = EIO_DT_BLK; break; 1326 case DT_BLK: ent->type = EIO_DT_BLK; break;
1154 #endif 1327 #endif
1328 #ifdef DT_MPB
1329 case DT_MPB: ent->type = EIO_DT_MPB; break;
1330 #endif
1155 #ifdef DT_REG 1331 #ifdef DT_REG
1156 case DT_REG: ent->type = EIO_DT_REG; break; 1332 case DT_REG: ent->type = EIO_DT_REG; break;
1333 #endif
1334 #ifdef DT_NWK
1335 case DT_NWK: ent->type = EIO_DT_NWK; break;
1336 #endif
1337 #ifdef DT_CMP
1338 case DT_CMP: ent->type = EIO_DT_CMP; break;
1157 #endif 1339 #endif
1158 #ifdef DT_LNK 1340 #ifdef DT_LNK
1159 case DT_LNK: ent->type = EIO_DT_LNK; break; 1341 case DT_LNK: ent->type = EIO_DT_LNK; break;
1160 #endif 1342 #endif
1161 #ifdef DT_SOCK 1343 #ifdef DT_SOCK
1162 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1344 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163 #endif 1345 #endif
1346 #ifdef DT_DOOR
1347 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1348 #endif
1164 #ifdef DT_WHT 1349 #ifdef DT_WHT
1165 case DT_WHT: ent->type = EIO_DT_WHT; break; 1350 case DT_WHT: ent->type = EIO_DT_WHT; break;
1166 #endif 1351 #endif
1167 } 1352 }
1168 1353
1169 ent->score = 0; 1354 ent->score = 7;
1170 1355
1171 if (flags & EIO_READDIR_DIRS_FIRST) 1356 if (flags & EIO_READDIR_DIRS_FIRST)
1172 { 1357 {
1173 if (ent->type == EIO_DT_UNKNOWN) 1358 if (ent->type == EIO_DT_UNKNOWN)
1174 { 1359 {
1175 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1360 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176 ent->score = 98; 1361 ent->score = 1;
1177 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1362 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 */ 1363 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1179 } 1364 }
1180 else if (ent->type == DT_DIR) 1365 else if (ent->type == EIO_DT_DIR)
1181 ent->score = 100; 1366 ent->score = 0;
1182 } 1367 }
1183 } 1368 }
1184 1369
1185 namesoffs += len; 1370 namesoffs += len;
1186 ++dentoffs; 1371 ++dentoffs;
1187 } 1372 }
1373
1374 if (EIO_CANCELLED (req))
1375 {
1376 errno = ECANCELED;
1377 break;
1378 }
1188 } 1379 }
1189 else
1190 req->result = -1;
1191} 1380}
1192 1381
1193#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1382#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1194# undef msync
1195# define msync(a,b,c) ((errno = ENOSYS), -1) 1383# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1384#else
1385
1386int
1387eio__msync (void *mem, size_t len, int flags)
1388{
1389 if (EIO_MS_ASYNC != MS_SYNC
1390 || EIO_MS_INVALIDATE != MS_INVALIDATE
1391 || EIO_MS_SYNC != MS_SYNC)
1392 {
1393 flags = 0
1394 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1395 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1396 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1397 }
1398
1399 return msync (mem, len, flags);
1400}
1401
1196#endif 1402#endif
1197 1403
1198int 1404int
1199eio__mtouch (void *mem, size_t len, int flags) 1405eio__mtouch (void *mem, size_t len, int flags)
1200{ 1406{
1207 1413
1208 if (!page) 1414 if (!page)
1209 page = sysconf (_SC_PAGESIZE); 1415 page = sysconf (_SC_PAGESIZE);
1210#endif 1416#endif
1211 1417
1418 /* round down to start of page, although this is probably useless */
1212 addr &= ~(page - 1); /* assume page size is always a power of two */ 1419 addr &= ~(page - 1); /* assume page size is always a power of two */
1213 1420
1214 if (addr < end) 1421 if (addr < end)
1215 if (flags) /* modify */ 1422 if (flags & EIO_MT_MODIFY) /* modify */
1216 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1423 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1217 else 1424 else
1218 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1425 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1219 1426
1220 return 0; 1427 return 0;
1392 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1599 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1393 1600
1394 case EIO_SYNC: req->result = 0; sync (); break; 1601 case EIO_SYNC: req->result = 0; sync (); break;
1395 case EIO_FSYNC: req->result = fsync (req->int1); break; 1602 case EIO_FSYNC: req->result = fsync (req->int1); break;
1396 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1603 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1397 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1604 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1398 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1605 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; 1606 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1400 1607
1401 case EIO_READDIR: eio__scandir (req, self); break; 1608 case EIO_READDIR: eio__scandir (req, self); break;
1402 1609
1716/* misc garbage */ 1923/* misc garbage */
1717 1924
1718ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1925ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1719{ 1926{
1720 etp_worker wrk; 1927 etp_worker wrk;
1928 ssize_t ret;
1721 1929
1722 wrk.dbuf = 0; 1930 wrk.dbuf = 0;
1723 1931
1724 eio__sendfile (ofd, ifd, offset, count, &wrk); 1932 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1725 1933
1726 if (wrk.dbuf) 1934 if (wrk.dbuf)
1727 free (wrk.dbuf); 1935 free (wrk.dbuf);
1728}
1729 1936
1937 return ret;
1938}
1939

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines