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 */ |
202 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
202 | static 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 | |
600 | static void etp_set_max_poll_time (double nseconds) |
600 | static 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 | |
607 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
607 | static 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 |
… | |
… | |
923 | if (res < 0 && sbytes) |
923 | if (res < 0 && sbytes) |
924 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
924 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
925 | res = sbytes; |
925 | res = sbytes; |
926 | } |
926 | } |
927 | |
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 | |
928 | # elif __hpux |
938 | # elif __hpux |
929 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
939 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
930 | |
940 | |
931 | # elif __solaris |
941 | # elif __solaris |
932 | { |
942 | { |
… | |
… | |
943 | if (res < 0 && sbytes) |
953 | if (res < 0 && sbytes) |
944 | res = sbytes; |
954 | res = sbytes; |
945 | } |
955 | } |
946 | |
956 | |
947 | # 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 | |
948 | #else |
968 | #else |
949 | res = -1; |
969 | res = -1; |
950 | errno = ENOSYS; |
970 | errno = ENOSYS; |
951 | #endif |
971 | #endif |
952 | |
972 | |
953 | if (res < 0 |
973 | if (res < 0 |
954 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
974 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
975 | || errno == ENOTSUP || errno == EOPNOTSUPP /* BSDs */ |
955 | #if __solaris |
976 | #if __solaris |
956 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
977 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
957 | #endif |
978 | #endif |
958 | ) |
979 | ) |
959 | ) |
980 | ) |
… | |
… | |
993 | } |
1014 | } |
994 | |
1015 | |
995 | static signed char |
1016 | static signed char |
996 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1017 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
997 | { |
1018 | { |
998 | return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ |
1019 | 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; |
1020 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
1000 | } |
1021 | } |
1001 | |
1022 | |
1002 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1023 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1003 | |
1024 | |
1004 | #define EIO_QSORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1025 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1005 | #define EIO_QSORT_SKIP 60 /* when to skip qsort completely */ |
1026 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
1006 | |
1027 | |
1007 | static void |
1028 | static void |
1008 | eio_dent_sort (eio_dirent *dents, int size) |
1029 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1009 | { |
1030 | { |
|
|
1031 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1032 | unsigned char *bit = bits; |
|
|
1033 | |
|
|
1034 | assert (CHAR_BIT == 8); |
|
|
1035 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1036 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
|
|
1037 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
|
|
1038 | |
|
|
1039 | if (size <= EIO_SORT_FAST) |
|
|
1040 | return; |
|
|
1041 | |
|
|
1042 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1043 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1044 | /* inode_bits must contain all inodes ORed together */ |
|
|
1045 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1046 | { |
|
|
1047 | ino_t endianness; |
1010 | int i, j; |
1048 | int i, j; |
1011 | |
1049 | |
|
|
1050 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1051 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1052 | ((unsigned char *)&endianness)[i] = i; |
|
|
1053 | |
|
|
1054 | *bit++ = 0; |
|
|
1055 | |
|
|
1056 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1057 | { |
|
|
1058 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1059 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1060 | endianness >>= 8; |
|
|
1061 | |
|
|
1062 | for (j = 0; j < 8; ++j) |
|
|
1063 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1064 | *bit++ = offs + j; |
|
|
1065 | } |
|
|
1066 | |
|
|
1067 | for (j = 0; j < 8; ++j) |
|
|
1068 | if (score_bits & (1 << j)) |
|
|
1069 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1070 | } |
|
|
1071 | |
|
|
1072 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1073 | { |
|
|
1074 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1075 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1076 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1077 | int stk_idx = 0; |
|
|
1078 | |
|
|
1079 | base_stk [stk_idx] = dents; |
|
|
1080 | end_stk [stk_idx] = dents + size; |
|
|
1081 | bit_stk [stk_idx] = bit - 1; |
|
|
1082 | |
|
|
1083 | do |
|
|
1084 | { |
|
|
1085 | base = base_stk [stk_idx]; |
|
|
1086 | end = end_stk [stk_idx]; |
|
|
1087 | bit = bit_stk [stk_idx]; |
|
|
1088 | |
|
|
1089 | for (;;) |
|
|
1090 | { |
|
|
1091 | unsigned char O = *bit >> 3; |
|
|
1092 | unsigned char M = 1 << (*bit & 7); |
|
|
1093 | |
|
|
1094 | eio_dirent *a = base; |
|
|
1095 | eio_dirent *b = end; |
|
|
1096 | |
|
|
1097 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1098 | break; |
|
|
1099 | |
|
|
1100 | /* now bit-partition the array on the bit */ |
|
|
1101 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1102 | /* partition algos found in the literature */ |
|
|
1103 | do |
|
|
1104 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1105 | ++a; |
|
|
1106 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1107 | { |
|
|
1108 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1109 | ++a; |
|
|
1110 | } |
|
|
1111 | while (b > a); |
|
|
1112 | |
|
|
1113 | /* next bit, or stop, if no bits left in this path */ |
|
|
1114 | if (!*--bit) |
|
|
1115 | break; |
|
|
1116 | |
|
|
1117 | base_stk [stk_idx] = a; |
|
|
1118 | end_stk [stk_idx] = end; |
|
|
1119 | bit_stk [stk_idx] = bit; |
|
|
1120 | ++stk_idx; |
|
|
1121 | |
|
|
1122 | end = a; |
|
|
1123 | } |
|
|
1124 | } |
|
|
1125 | while (stk_idx--); |
|
|
1126 | } |
|
|
1127 | } |
|
|
1128 | |
|
|
1129 | static void |
|
|
1130 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1131 | { |
|
|
1132 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1133 | { |
|
|
1134 | int i; |
|
|
1135 | eio_dirent *min = dents; |
|
|
1136 | |
|
|
1137 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1138 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1139 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1140 | min = &dents [i]; |
|
|
1141 | |
|
|
1142 | /* swap elements 0 and j (minimum) */ |
|
|
1143 | { |
|
|
1144 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1145 | } |
|
|
1146 | } |
|
|
1147 | |
|
|
1148 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1149 | { |
|
|
1150 | eio_dirent *i, *j; |
|
|
1151 | |
|
|
1152 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1153 | { |
|
|
1154 | eio_dirent value = *i; |
|
|
1155 | |
|
|
1156 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1157 | j [1] = j [0]; |
|
|
1158 | |
|
|
1159 | j [1] = value; |
|
|
1160 | } |
|
|
1161 | } |
|
|
1162 | } |
|
|
1163 | |
|
|
1164 | static void |
|
|
1165 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1166 | { |
1012 | if (size <= 1) |
1167 | if (size <= 1) |
1013 | return; /* our insertion sort relies on size > 0 */ |
1168 | return; /* our insertion sort relies on size > 0 */ |
1014 | |
1169 | |
1015 | if (size > EIO_QSORT_SKIP) /* skip quicksort for small directories */ |
1170 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
1016 | { |
1171 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
1017 | /* first, use quicksort */ |
1172 | 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 | |
1173 | |
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 */ |
1174 | /* 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 */ |
1175 | /* as insertion sort is more efficient for small partitions */ |
1070 | { |
1176 | 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 | } |
1177 | } |
1097 | |
1178 | |
1098 | /* read a full directory */ |
1179 | /* read a full directory */ |
1099 | static void |
1180 | static void |
1100 | eio__scandir (eio_req *req, etp_worker *self) |
1181 | eio__scandir (eio_req *req, etp_worker *self) |
… | |
… | |
1106 | int namesoffs = 0; |
1187 | int namesoffs = 0; |
1107 | int flags = req->int1; |
1188 | int flags = req->int1; |
1108 | eio_dirent *dents = 0; |
1189 | eio_dirent *dents = 0; |
1109 | int dentalloc = 128; |
1190 | int dentalloc = 128; |
1110 | int dentoffs = 0; |
1191 | int dentoffs = 0; |
|
|
1192 | ino_t inode_bits = 0; |
1111 | |
1193 | |
1112 | req->result = -1; |
1194 | req->result = -1; |
1113 | |
1195 | |
1114 | if (!(flags & EIO_READDIR_DENTS)) |
1196 | if (!(flags & EIO_READDIR_DENTS)) |
1115 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1197 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1116 | |
1198 | |
1117 | X_LOCK (wrklock); |
1199 | X_LOCK (wrklock); |
1118 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1200 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1119 | self->dirp = dirp = opendir (req->ptr1); |
1201 | self->dirp = dirp = opendir (req->ptr1); |
1120 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1202 | 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; |
1203 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
|
|
1204 | req->ptr2 = names = malloc (namesalloc); |
1123 | X_UNLOCK (wrklock); |
1205 | X_UNLOCK (wrklock); |
1124 | |
1206 | |
1125 | if (dirp && names && (!flags || dents)) |
1207 | if (dirp && names && (!flags || dents)) |
1126 | for (;;) |
1208 | for (;;) |
1127 | { |
1209 | { |
… | |
… | |
1135 | |
1217 | |
1136 | /* sort etc. */ |
1218 | /* sort etc. */ |
1137 | req->int1 = flags; |
1219 | req->int1 = flags; |
1138 | req->result = dentoffs; |
1220 | req->result = dentoffs; |
1139 | |
1221 | |
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 |
1222 | if (flags & EIO_READDIR_STAT_ORDER) |
1149 | || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) |
1223 | 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) |
1224 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1225 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1226 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1227 | else |
1152 | { |
1228 | { |
1153 | /* in this case, all is known, and we just put dirs first and sort them */ |
1229 | /* in this case, all is known, and we just put dirs first and sort them */ |
1154 | eio_dirent *ent = dents + dentoffs; |
1230 | eio_dirent *oth = dents + dentoffs; |
1155 | eio_dirent *dir = dents; |
1231 | eio_dirent *dir = dents; |
1156 | |
1232 | |
1157 | /* now move dirs to the front, and non-dirs to the back */ |
1233 | /* now partition dirs to the front, and non-dirs to the back */ |
1158 | /* by walking from both sides and swapping if necessary */ |
1234 | /* by walking from both sides and swapping if necessary */ |
|
|
1235 | /* also clear score, so it doesn't influence sorting */ |
1159 | while (ent > dir) |
1236 | while (oth > dir) |
1160 | { |
1237 | { |
1161 | if (dir->type == DT_DIR) |
1238 | if (dir->type == EIO_DT_DIR) |
1162 | ++dir; |
1239 | ++dir; |
1163 | else |
|
|
1164 | { |
|
|
1165 | --ent; |
|
|
1166 | |
|
|
1167 | if (ent->type == DT_DIR) |
1240 | else if ((--oth)->type == EIO_DT_DIR) |
1168 | { |
1241 | { |
1169 | eio_dirent tmp = *dir; |
1242 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
1170 | *dir = *ent; |
|
|
1171 | *ent = tmp; |
|
|
1172 | |
1243 | |
1173 | ++dir; |
1244 | ++dir; |
1174 | } |
1245 | } |
1175 | } |
1246 | } |
1176 | } |
|
|
1177 | |
1247 | |
1178 | /* now sort the dirs only */ |
1248 | /* now sort the dirs only */ |
1179 | eio_dent_sort (dents, dir - dents); |
1249 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1180 | } |
1250 | } |
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 | |
1251 | |
1192 | break; |
1252 | break; |
1193 | } |
1253 | } |
1194 | |
1254 | |
1195 | /* now add the entry to our list(s) */ |
1255 | /* now add the entry to our list(s) */ |
… | |
… | |
1202 | |
1262 | |
1203 | while (expect_false (namesoffs + len > namesalloc)) |
1263 | while (expect_false (namesoffs + len > namesalloc)) |
1204 | { |
1264 | { |
1205 | namesalloc *= 2; |
1265 | namesalloc *= 2; |
1206 | X_LOCK (wrklock); |
1266 | X_LOCK (wrklock); |
1207 | req->ptr1 = names = realloc (names, namesalloc); |
1267 | req->ptr2 = names = realloc (names, namesalloc); |
1208 | X_UNLOCK (wrklock); |
1268 | X_UNLOCK (wrklock); |
1209 | |
1269 | |
1210 | if (!names) |
1270 | if (!names) |
1211 | break; |
1271 | break; |
1212 | } |
1272 | } |
… | |
… | |
1219 | |
1279 | |
1220 | if (expect_false (dentoffs == dentalloc)) |
1280 | if (expect_false (dentoffs == dentalloc)) |
1221 | { |
1281 | { |
1222 | dentalloc *= 2; |
1282 | dentalloc *= 2; |
1223 | X_LOCK (wrklock); |
1283 | X_LOCK (wrklock); |
1224 | req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1284 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1225 | X_UNLOCK (wrklock); |
1285 | X_UNLOCK (wrklock); |
1226 | |
1286 | |
1227 | if (!dents) |
1287 | if (!dents) |
1228 | break; |
1288 | break; |
1229 | } |
1289 | } |
1230 | |
1290 | |
1231 | ent = dents + dentoffs; |
1291 | ent = dents + dentoffs; |
1232 | |
1292 | |
1233 | ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ |
1293 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
1234 | ent->namelen = len - 1; |
1294 | ent->namelen = len - 1; |
1235 | ent->inode = D_INO (entp); |
1295 | ent->inode = D_INO (entp); |
|
|
1296 | |
|
|
1297 | inode_bits |= ent->inode; |
1236 | |
1298 | |
1237 | switch (D_TYPE (entp)) |
1299 | switch (D_TYPE (entp)) |
1238 | { |
1300 | { |
1239 | default: |
1301 | default: |
1240 | ent->type = EIO_DT_UNKNOWN; |
1302 | ent->type = EIO_DT_UNKNOWN; |
… | |
… | |
1283 | #ifdef DT_WHT |
1345 | #ifdef DT_WHT |
1284 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1346 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1285 | #endif |
1347 | #endif |
1286 | } |
1348 | } |
1287 | |
1349 | |
1288 | ent->score = 0; |
1350 | ent->score = 7; |
1289 | |
1351 | |
1290 | if (flags & EIO_READDIR_DIRS_FIRST) |
1352 | if (flags & EIO_READDIR_DIRS_FIRST) |
1291 | { |
1353 | { |
1292 | if (ent->type == EIO_DT_UNKNOWN) |
1354 | if (ent->type == EIO_DT_UNKNOWN) |
1293 | { |
1355 | { |
1294 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1356 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1295 | ent->score = 98; |
1357 | ent->score = 1; |
1296 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1358 | 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 */ |
1359 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
1298 | } |
1360 | } |
1299 | else if (ent->type == EIO_DT_DIR) |
1361 | else if (ent->type == EIO_DT_DIR) |
1300 | ent->score = 100; |
1362 | ent->score = 0; |
1301 | } |
1363 | } |
1302 | } |
1364 | } |
1303 | |
1365 | |
1304 | namesoffs += len; |
1366 | namesoffs += len; |
1305 | ++dentoffs; |
1367 | ++dentoffs; |
… | |
… | |
1312 | } |
1374 | } |
1313 | } |
1375 | } |
1314 | } |
1376 | } |
1315 | |
1377 | |
1316 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1378 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1317 | # undef msync |
|
|
1318 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
1379 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1380 | #else |
|
|
1381 | |
|
|
1382 | int |
|
|
1383 | eio__msync (void *mem, size_t len, int flags) |
|
|
1384 | { |
|
|
1385 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1386 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1387 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1388 | { |
|
|
1389 | flags = 0 |
|
|
1390 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1391 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1392 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1393 | } |
|
|
1394 | |
|
|
1395 | return msync (mem, len, flags); |
|
|
1396 | } |
|
|
1397 | |
1319 | #endif |
1398 | #endif |
1320 | |
1399 | |
1321 | int |
1400 | int |
1322 | eio__mtouch (void *mem, size_t len, int flags) |
1401 | eio__mtouch (void *mem, size_t len, int flags) |
1323 | { |
1402 | { |
… | |
… | |
1330 | |
1409 | |
1331 | if (!page) |
1410 | if (!page) |
1332 | page = sysconf (_SC_PAGESIZE); |
1411 | page = sysconf (_SC_PAGESIZE); |
1333 | #endif |
1412 | #endif |
1334 | |
1413 | |
|
|
1414 | /* round down to start of page, although this is probably useless */ |
1335 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1415 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1336 | |
1416 | |
1337 | if (addr < end) |
1417 | if (addr < end) |
1338 | if (flags) /* modify */ |
1418 | if (flags & EIO_MT_MODIFY) /* modify */ |
1339 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1419 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1340 | else |
1420 | else |
1341 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1421 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1342 | |
1422 | |
1343 | return 0; |
1423 | return 0; |
… | |
… | |
1515 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1595 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1516 | |
1596 | |
1517 | case EIO_SYNC: req->result = 0; sync (); break; |
1597 | case EIO_SYNC: req->result = 0; sync (); break; |
1518 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1598 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1519 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1599 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1520 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1600 | 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; |
1601 | 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; |
1602 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1523 | |
1603 | |
1524 | case EIO_READDIR: eio__scandir (req, self); break; |
1604 | case EIO_READDIR: eio__scandir (req, self); break; |
1525 | |
1605 | |