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 1.40 by root, Sat Jun 13 14:58:33 2009 UTC vs.
Revision 1.50 by root, Thu Jan 7 17:18:08 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 *
… …
80# include <utime.h> 80# include <utime.h>
81# include <signal.h> 81# include <signal.h>
82# include <dirent.h> 82# include <dirent.h>
83 83
84/* 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 */
85# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 85# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */ 86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno 87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen 88# define D_NAMLEN(de) (de)->d_namlen
89# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 89# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino 90# define D_INO(de) (de)->d_ino
91# endif 91# endif
92 92
93#ifdef _D_EXACT_NAMLEN 93#ifdef _D_EXACT_NAMLEN
94# undef D_NAMLEN 94# undef D_NAMLEN
… …
106#endif 106#endif
107 107
108#if HAVE_SENDFILE 108#if HAVE_SENDFILE
109# if __linux 109# if __linux
110# include <sys/sendfile.h> 110# include <sys/sendfile.h>
111# elif __freebsd 111# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 112# include <sys/socket.h>
113# include <sys/uio.h> 113# include <sys/uio.h>
114# elif __hpux 114# elif __hpux
115# include <sys/socket.h> 115# include <sys/socket.h>
116# elif __solaris /* not yet */ 116# elif __solaris
117# include <sys/sendfile.h> 117# include <sys/sendfile.h>
118# else 118# else
119# error sendfile support requested but not available 119# error sendfile support requested but not available
120# endif 120# endif
121#endif 121#endif
… …
196 196
197/*****************************************************************************/ 197/*****************************************************************************/
198 198
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200 200
201/* calculcate time difference in ~1/EIO_TICKS of a second */ 201/* calculate time difference in ~1/EIO_TICKS of a second */
202static int tvdiff (struct timeval *tv1, struct timeval *tv2) 202static int tvdiff (struct timeval *tv1, struct timeval *tv2)
203{ 203{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 205 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206} 206}
… …
598} 598}
599 599
600static void etp_set_max_poll_time (double nseconds) 600static void etp_set_max_poll_time (double nseconds)
601{ 601{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 602 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds; 603 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605} 605}
606 606
607static void etp_set_max_poll_reqs (unsigned int maxreqs) 607static void etp_set_max_poll_reqs (unsigned int maxreqs)
608{ 608{
… …
867 if (!res || errno != ENOSYS) 867 if (!res || errno != ENOSYS)
868 return res; 868 return res;
869#endif 869#endif
870 870
871 /* 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
872 * call fdatasync, as thta matches the expectation of it's users best */ 872 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 873 return fdatasync (fd);
874} 874}
875 875
876#if !HAVE_READAHEAD 876#if !HAVE_READAHEAD
877# undef readahead 877# undef readahead
… …
909 909
910#if HAVE_SENDFILE 910#if HAVE_SENDFILE
911# if __linux 911# if __linux
912 res = sendfile (ofd, ifd, &offset, count); 912 res = sendfile (ofd, ifd, &offset, count);
913 913
914# elif __freebsd 914# elif __FreeBSD__
915 /* 915 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 916 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 917 * wasted on making it similar to other I/O functions.
918 */ 918 */
919 { 919 {
920 off_t sbytes; 920 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 922
923 /* freebsd' sendfile will return 0 when success */
923 if (res < 0 && sbytes) 924 if (res == 0 && sbytes)
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 925 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */
926 res = sbytes;
927 }
928
929# elif defined (__APPLE__)
930
931 {
932 off_t sbytes = count;
933 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
934
935 if (res == 0 && errno == EAGAIN && sbytes)
925 res = sbytes; 936 res = sbytes;
926 } 937 }
927 938
928# elif __hpux 939# elif __hpux
929 res = sendfile (ofd, ifd, offset, count, 0, 0); 940 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
943 if (res < 0 && sbytes) 954 if (res < 0 && sbytes)
944 res = sbytes; 955 res = sbytes;
945 } 956 }
946 957
947# endif 958# endif
959
960#elif defined (_WIN32)
961
962 /* does not work, just for documentation of what would need to be done */
963 {
964 HANDLE h = TO_SOCKET (ifd);
965 SetFilePointer (h, offset, 0, FILE_BEGIN);
966 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
967 }
968
948#else 969#else
949 res = -1; 970 res = -1;
950 errno = ENOSYS; 971 errno = ENOSYS;
951#endif 972#endif
952 973
953 if (res < 0 974 if (res < 0
954 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 975 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
976 /* BSDs */
977#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
978 || errno == ENOTSUP
979#endif
980 || errno == EOPNOTSUPP /* BSDs */
955#if __solaris 981#if __solaris
956 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 982 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
957#endif 983#endif
958 ) 984 )
959 ) 985 )
… …
993} 1019}
994 1020
995static signed char 1021static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1022eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{ 1023{
998 return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ 1024 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
999 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1025 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1000} 1026}
1001 1027
1002#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1028#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1003 1029
1004#define EIO_QSORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1030#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1005#define EIO_QSORT_SKIP 60 /* when to skip qsort completely */ 1031#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1006 1032
1007static void 1033static void
1008eio_dent_sort (eio_dirent *dents, int size) 1034eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1009{ 1035{
1036 unsigned char bits [9 + sizeof (ino_t) * 8];
1037 unsigned char *bit = bits;
1038
1039 assert (CHAR_BIT == 8);
1040 assert (sizeof (eio_dirent) * 8 < 256);
1041 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1042 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1043
1044 if (size <= EIO_SORT_FAST)
1045 return;
1046
1047 /* first prepare an array of bits to test in our radix sort */
1048 /* try to take endianness into account, as well as differences in ino_t sizes */
1049 /* inode_bits must contain all inodes ORed together */
1050 /* which is used to skip bits that are 0 everywhere, which is very common */
1051 {
1052 ino_t endianness;
1010 int i, j; 1053 int i, j;
1011 1054
1055 /* we store the byte offset of byte n into byte n of "endianness" */
1056 for (i = 0; i < sizeof (ino_t); ++i)
1057 ((unsigned char *)&endianness)[i] = i;
1058
1059 *bit++ = 0;
1060
1061 for (i = 0; i < sizeof (ino_t); ++i)
1062 {
1063 /* shifting off the byte offsets out of "endianness" */
1064 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1065 endianness >>= 8;
1066
1067 for (j = 0; j < 8; ++j)
1068 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1069 *bit++ = offs + j;
1070 }
1071
1072 for (j = 0; j < 8; ++j)
1073 if (score_bits & (1 << j))
1074 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1075 }
1076
1077 /* now actually do the sorting (a variant of MSD radix sort) */
1078 {
1079 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1080 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1081 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1082 int stk_idx = 0;
1083
1084 base_stk [stk_idx] = dents;
1085 end_stk [stk_idx] = dents + size;
1086 bit_stk [stk_idx] = bit - 1;
1087
1088 do
1089 {
1090 base = base_stk [stk_idx];
1091 end = end_stk [stk_idx];
1092 bit = bit_stk [stk_idx];
1093
1094 for (;;)
1095 {
1096 unsigned char O = *bit >> 3;
1097 unsigned char M = 1 << (*bit & 7);
1098
1099 eio_dirent *a = base;
1100 eio_dirent *b = end;
1101
1102 if (b - a < EIO_SORT_CUTOFF)
1103 break;
1104
1105 /* now bit-partition the array on the bit */
1106 /* this ugly asymmetric loop seems to perform much better than typical */
1107 /* partition algos found in the literature */
1108 do
1109 if (!(((unsigned char *)a)[O] & M))
1110 ++a;
1111 else if (!(((unsigned char *)--b)[O] & M))
1112 {
1113 eio_dirent tmp = *a; *a = *b; *b = tmp;
1114 ++a;
1115 }
1116 while (b > a);
1117
1118 /* next bit, or stop, if no bits left in this path */
1119 if (!*--bit)
1120 break;
1121
1122 base_stk [stk_idx] = a;
1123 end_stk [stk_idx] = end;
1124 bit_stk [stk_idx] = bit;
1125 ++stk_idx;
1126
1127 end = a;
1128 }
1129 }
1130 while (stk_idx--);
1131 }
1132}
1133
1134static void
1135eio_dent_insertion_sort (eio_dirent *dents, int size)
1136{
1137 /* first move the smallest element to the front, to act as a sentinel */
1138 {
1139 int i;
1140 eio_dirent *min = dents;
1141
1142 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1143 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1144 if (EIO_DENT_CMP (dents [i], <, *min))
1145 min = &dents [i];
1146
1147 /* swap elements 0 and j (minimum) */
1148 {
1149 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1150 }
1151 }
1152
1153 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1154 {
1155 eio_dirent *i, *j;
1156
1157 for (i = dents + 1; i < dents + size; ++i)
1158 {
1159 eio_dirent value = *i;
1160
1161 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1162 j [1] = j [0];
1163
1164 j [1] = value;
1165 }
1166 }
1167}
1168
1169static void
1170eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1171{
1012 if (size <= 1) 1172 if (size <= 1)
1013 return; /* our insertion sort relies on size > 0 */ 1173 return; /* our insertion sort relies on size > 0 */
1014 1174
1015 if (size > EIO_QSORT_SKIP) /* skip quicksort for small directories */ 1175 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1016 { 1176 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1017 /* first, use quicksort */ 1177 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1018 /* should be good for 2**31 entries */
1019 struct rng { int l, r; } rng [32];
1020 1178
1021 i = 0;
1022 rng[0].l = 0;
1023 rng[0].r = size;
1024
1025 while (expect_true (i >= 0))
1026 {
1027 int L = rng [i].l;
1028 int R = rng [i].r - 1;
1029
1030 if (expect_false (L + EIO_QSORT_CUTOFF < R))
1031 {
1032 eio_dirent piv = dents [L];
1033
1034 while (L < R)
1035 {
1036 while (EIO_DENT_CMP (dents [R], >=, piv) && L < R)
1037 --R;
1038
1039 if (L < R)
1040 dents [L++] = dents [R];
1041
1042 while (EIO_DENT_CMP (dents [L], <=, piv) && L < R)
1043 ++L;
1044
1045 if (L < R)
1046 dents [R--] = dents [L];
1047 }
1048
1049 dents [L] = piv;
1050
1051 ++i;
1052 rng [i].l = L + 1;
1053 rng [i].r = rng [i - 1].r;
1054 rng [i - 1].r = L;
1055
1056 if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1057 {
1058 struct rng t;
1059
1060 t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1061 }
1062 }
1063 else
1064 --i;
1065 }
1066 }
1067
1068 /* use an insertion sort after qsort, or for small arrays */ 1179 /* use an insertion sort at the end, or for small arrays, */
1069 /* first move the smallest element to the front, to act as a sentinel */ 1180 /* as insertion sort is more efficient for small partitions */
1070 { 1181 eio_dent_insertion_sort (dents, size);
1071 int min = 0;
1072
1073 for (i = size > EIO_QSORT_SKIP ? EIO_QSORT_CUTOFF + 1 : size; --i; )
1074 if (EIO_DENT_CMP (dents [i], <, dents [min]))
1075 min = i;
1076
1077 /* swap elements 0 and j (minimum) */
1078 {
1079 eio_dirent tmp = dents [0]; dents [0] = dents [min]; dents [min] = tmp;
1080 }
1081 }
1082
1083 /* then do standard insertion sort */
1084 for (i = 1; i < size; ++i)
1085 {
1086 eio_dirent value = dents [i];
1087
1088 for (j = i - 1; EIO_DENT_CMP (dents [j], >, value); --j)
1089 {
1090 assert (j >= 0);
1091 dents [j + 1] = dents [j];
1092 }
1093
1094 dents [j + 1] = value;
1095 }
1096} 1182}
1097 1183
1098/* read a full directory */ 1184/* read a full directory */
1099static void 1185static void
1100eio__scandir (eio_req *req, etp_worker *self) 1186eio__scandir (eio_req *req, etp_worker *self)
… …
1106 int namesoffs = 0; 1192 int namesoffs = 0;
1107 int flags = req->int1; 1193 int flags = req->int1;
1108 eio_dirent *dents = 0; 1194 eio_dirent *dents = 0;
1109 int dentalloc = 128; 1195 int dentalloc = 128;
1110 int dentoffs = 0; 1196 int dentoffs = 0;
1197 ino_t inode_bits = 0;
1111 1198
1112 req->result = -1; 1199 req->result = -1;
1113 1200
1114 if (!(flags & EIO_READDIR_DENTS)) 1201 if (!(flags & EIO_READDIR_DENTS))
1115 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1202 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1116 1203
1117 X_LOCK (wrklock); 1204 X_LOCK (wrklock);
1118 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1205 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1119 self->dirp = dirp = opendir (req->ptr1); 1206 self->dirp = dirp = opendir (req->ptr1);
1120 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1207 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1121 req->ptr1 = names = malloc (namesalloc);
1122 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1208 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1209 req->ptr2 = names = malloc (namesalloc);
1123 X_UNLOCK (wrklock); 1210 X_UNLOCK (wrklock);
1124 1211
1125 if (dirp && names && (!flags || dents)) 1212 if (dirp && names && (!flags || dents))
1126 for (;;) 1213 for (;;)
1127 { 1214 {
… …
1135 1222
1136 /* sort etc. */ 1223 /* sort etc. */
1137 req->int1 = flags; 1224 req->int1 = flags;
1138 req->result = dentoffs; 1225 req->result = dentoffs;
1139 1226
1140 if (dents)
1141 {
1142 eio_dirent *ent = dents + dentoffs;
1143
1144 while (ent > dents)
1145 (--ent)->name = names + (size_t)ent->name;
1146 }
1147
1148 if (flags & EIO_READDIR_STAT_ORDER 1227 if (flags & EIO_READDIR_STAT_ORDER)
1149 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1228 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1150 eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1151 else if (flags & EIO_READDIR_DIRS_FIRST) 1229 else if (flags & EIO_READDIR_DIRS_FIRST)
1230 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1231 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1232 else
1152 { 1233 {
1153 /* in this case, all is known, and we just put dirs first and sort them */ 1234 /* in this case, all is known, and we just put dirs first and sort them */
1154 eio_dirent *ent = dents + dentoffs; 1235 eio_dirent *oth = dents + dentoffs;
1155 eio_dirent *dir = dents; 1236 eio_dirent *dir = dents;
1156 1237
1157 /* now move dirs to the front, and non-dirs to the back */ 1238 /* now partition dirs to the front, and non-dirs to the back */
1158 /* by walking from both sides and swapping if necessary */ 1239 /* by walking from both sides and swapping if necessary */
1240 /* also clear score, so it doesn't influence sorting */
1159 while (ent > dir) 1241 while (oth > dir)
1160 { 1242 {
1161 if (dir->type == DT_DIR) 1243 if (dir->type == EIO_DT_DIR)
1162 ++dir; 1244 ++dir;
1163 else
1164 {
1165 --ent;
1166
1167 if (ent->type == DT_DIR) 1245 else if ((--oth)->type == EIO_DT_DIR)
1168 { 1246 {
1169 eio_dirent tmp = *dir; 1247 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1170 *dir = *ent;
1171 *ent = tmp;
1172 1248
1173 ++dir; 1249 ++dir;
1174 } 1250 }
1175 } 1251 }
1176 }
1177 1252
1178 /* now sort the dirs only */ 1253 /* now sort the dirs only */
1179 eio_dent_sort (dents, dir - dents); 1254 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1180 } 1255 }
1181
1182 /* only provide the names array unless DENTS is specified */
1183 if (!(flags & EIO_READDIR_DENTS))
1184 {
1185 X_LOCK (wrklock);
1186 assert (!dents);
1187 req->ptr1 = 0;
1188 req->ptr2 = names;
1189 X_UNLOCK (wrklock);
1190 }
1191 1256
1192 break; 1257 break;
1193 } 1258 }
1194 1259
1195 /* now add the entry to our list(s) */ 1260 /* now add the entry to our list(s) */
… …
1202 1267
1203 while (expect_false (namesoffs + len > namesalloc)) 1268 while (expect_false (namesoffs + len > namesalloc))
1204 { 1269 {
1205 namesalloc *= 2; 1270 namesalloc *= 2;
1206 X_LOCK (wrklock); 1271 X_LOCK (wrklock);
1207 req->ptr1 = names = realloc (names, namesalloc); 1272 req->ptr2 = names = realloc (names, namesalloc);
1208 X_UNLOCK (wrklock); 1273 X_UNLOCK (wrklock);
1209 1274
1210 if (!names) 1275 if (!names)
1211 break; 1276 break;
1212 } 1277 }
… …
1219 1284
1220 if (expect_false (dentoffs == dentalloc)) 1285 if (expect_false (dentoffs == dentalloc))
1221 { 1286 {
1222 dentalloc *= 2; 1287 dentalloc *= 2;
1223 X_LOCK (wrklock); 1288 X_LOCK (wrklock);
1224 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1289 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1225 X_UNLOCK (wrklock); 1290 X_UNLOCK (wrklock);
1226 1291
1227 if (!dents) 1292 if (!dents)
1228 break; 1293 break;
1229 } 1294 }
1230 1295
1231 ent = dents + dentoffs; 1296 ent = dents + dentoffs;
1232 1297
1233 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1298 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1234 ent->namelen = len - 1; 1299 ent->namelen = len - 1;
1235 ent->inode = D_INO (entp); 1300 ent->inode = D_INO (entp);
1301
1302 inode_bits |= ent->inode;
1236 1303
1237 switch (D_TYPE (entp)) 1304 switch (D_TYPE (entp))
1238 { 1305 {
1239 default: 1306 default:
1240 ent->type = EIO_DT_UNKNOWN; 1307 ent->type = EIO_DT_UNKNOWN;
… …
1283 #ifdef DT_WHT 1350 #ifdef DT_WHT
1284 case DT_WHT: ent->type = EIO_DT_WHT; break; 1351 case DT_WHT: ent->type = EIO_DT_WHT; break;
1285 #endif 1352 #endif
1286 } 1353 }
1287 1354
1288 ent->score = 0; 1355 ent->score = 7;
1289 1356
1290 if (flags & EIO_READDIR_DIRS_FIRST) 1357 if (flags & EIO_READDIR_DIRS_FIRST)
1291 { 1358 {
1292 if (ent->type == EIO_DT_UNKNOWN) 1359 if (ent->type == EIO_DT_UNKNOWN)
1293 { 1360 {
1294 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1361 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1295 ent->score = 98; 1362 ent->score = 1;
1296 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1363 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1297 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ 1364 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1298 } 1365 }
1299 else if (ent->type == EIO_DT_DIR) 1366 else if (ent->type == EIO_DT_DIR)
1300 ent->score = 100; 1367 ent->score = 0;
1301 } 1368 }
1302 } 1369 }
1303 1370
1304 namesoffs += len; 1371 namesoffs += len;
1305 ++dentoffs; 1372 ++dentoffs;
… …
1312 } 1379 }
1313 } 1380 }
1314} 1381}
1315 1382
1316#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1383#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1317# undef msync
1318# define msync(a,b,c) ((errno = ENOSYS), -1) 1384# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1385#else
1386
1387int
1388eio__msync (void *mem, size_t len, int flags)
1389{
1390 if (EIO_MS_ASYNC != MS_SYNC
1391 || EIO_MS_INVALIDATE != MS_INVALIDATE
1392 || EIO_MS_SYNC != MS_SYNC)
1393 {
1394 flags = 0
1395 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1396 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1397 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1398 }
1399
1400 return msync (mem, len, flags);
1401}
1402
1319#endif 1403#endif
1320 1404
1321int 1405int
1322eio__mtouch (void *mem, size_t len, int flags) 1406eio__mtouch (void *mem, size_t len, int flags)
1323{ 1407{
… …
1330 1414
1331 if (!page) 1415 if (!page)
1332 page = sysconf (_SC_PAGESIZE); 1416 page = sysconf (_SC_PAGESIZE);
1333#endif 1417#endif
1334 1418
1419 /* round down to start of page, although this is probably useless */
1335 addr &= ~(page - 1); /* assume page size is always a power of two */ 1420 addr &= ~(page - 1); /* assume page size is always a power of two */
1336 1421
1337 if (addr < end) 1422 if (addr < end)
1338 if (flags) /* modify */ 1423 if (flags & EIO_MT_MODIFY) /* modify */
1339 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1424 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1340 else 1425 else
1341 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1426 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1342 1427
1343 return 0; 1428 return 0;
… …
1515 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1600 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1516 1601
1517 case EIO_SYNC: req->result = 0; sync (); break; 1602 case EIO_SYNC: req->result = 0; sync (); break;
1518 case EIO_FSYNC: req->result = fsync (req->int1); break; 1603 case EIO_FSYNC: req->result = fsync (req->int1); break;
1519 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1604 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1520 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1605 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1521 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1606 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1522 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1607 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1523 1608
1524 case EIO_READDIR: eio__scandir (req, self); break; 1609 case EIO_READDIR: eio__scandir (req, self); break;
1525 1610

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines