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 | * |
… | |
… | |
49 | #include <stdlib.h> |
49 | #include <stdlib.h> |
50 | #include <string.h> |
50 | #include <string.h> |
51 | #include <errno.h> |
51 | #include <errno.h> |
52 | #include <sys/types.h> |
52 | #include <sys/types.h> |
53 | #include <sys/stat.h> |
53 | #include <sys/stat.h> |
|
|
54 | #include <sys/statvfs.h> |
54 | #include <limits.h> |
55 | #include <limits.h> |
55 | #include <fcntl.h> |
56 | #include <fcntl.h> |
56 | #include <assert.h> |
57 | #include <assert.h> |
57 | |
58 | |
58 | #ifndef EIO_FINISH |
59 | #ifndef EIO_FINISH |
… | |
… | |
80 | # include <utime.h> |
81 | # include <utime.h> |
81 | # include <signal.h> |
82 | # include <signal.h> |
82 | # include <dirent.h> |
83 | # include <dirent.h> |
83 | |
84 | |
84 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
85 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
85 | # if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
86 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
86 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
87 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
87 | # define D_INO(de) (de)->d_fileno |
88 | # define D_INO(de) (de)->d_fileno |
88 | # define D_NAMLEN(de) (de)->d_namlen |
89 | # define D_NAMLEN(de) (de)->d_namlen |
89 | # elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
90 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
90 | # define D_INO(de) (de)->d_ino |
91 | # define D_INO(de) (de)->d_ino |
91 | # endif |
92 | # endif |
92 | |
93 | |
93 | #ifdef _D_EXACT_NAMLEN |
94 | #ifdef _D_EXACT_NAMLEN |
94 | # undef D_NAMLEN |
95 | # undef D_NAMLEN |
… | |
… | |
106 | #endif |
107 | #endif |
107 | |
108 | |
108 | #if HAVE_SENDFILE |
109 | #if HAVE_SENDFILE |
109 | # if __linux |
110 | # if __linux |
110 | # include <sys/sendfile.h> |
111 | # include <sys/sendfile.h> |
111 | # elif __freebsd |
112 | # elif __FreeBSD__ || defined __APPLE__ |
112 | # include <sys/socket.h> |
113 | # include <sys/socket.h> |
113 | # include <sys/uio.h> |
114 | # include <sys/uio.h> |
114 | # elif __hpux |
115 | # elif __hpux |
115 | # include <sys/socket.h> |
116 | # include <sys/socket.h> |
116 | # elif __solaris /* not yet */ |
117 | # elif __solaris |
117 | # include <sys/sendfile.h> |
118 | # include <sys/sendfile.h> |
118 | # else |
119 | # else |
119 | # error sendfile support requested but not available |
120 | # error sendfile support requested but not available |
120 | # endif |
121 | # endif |
121 | #endif |
122 | #endif |
… | |
… | |
196 | |
197 | |
197 | /*****************************************************************************/ |
198 | /*****************************************************************************/ |
198 | |
199 | |
199 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
200 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
200 | |
201 | |
201 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
202 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
202 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
203 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
203 | { |
204 | { |
204 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
205 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
205 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
206 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
206 | } |
207 | } |
… | |
… | |
598 | } |
599 | } |
599 | |
600 | |
600 | static void etp_set_max_poll_time (double nseconds) |
601 | static void etp_set_max_poll_time (double nseconds) |
601 | { |
602 | { |
602 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
603 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
603 | max_poll_time = nseconds; |
604 | max_poll_time = nseconds * EIO_TICKS; |
604 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
605 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
605 | } |
606 | } |
606 | |
607 | |
607 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
608 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
608 | { |
609 | { |
… | |
… | |
867 | if (!res || errno != ENOSYS) |
868 | if (!res || errno != ENOSYS) |
868 | return res; |
869 | return res; |
869 | #endif |
870 | #endif |
870 | |
871 | |
871 | /* even though we could play tricks with the flags, it's better to always |
872 | /* 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 */ |
873 | * call fdatasync, as that matches the expectation of its users best */ |
873 | return fdatasync (fd); |
874 | return fdatasync (fd); |
874 | } |
875 | } |
875 | |
876 | |
876 | #if !HAVE_READAHEAD |
877 | #if !HAVE_READAHEAD |
877 | # undef readahead |
878 | # undef readahead |
… | |
… | |
909 | |
910 | |
910 | #if HAVE_SENDFILE |
911 | #if HAVE_SENDFILE |
911 | # if __linux |
912 | # if __linux |
912 | res = sendfile (ofd, ifd, &offset, count); |
913 | res = sendfile (ofd, ifd, &offset, count); |
913 | |
914 | |
914 | # elif __freebsd |
915 | # elif __FreeBSD__ |
915 | /* |
916 | /* |
916 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
917 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
917 | * wasted on making it similar to other I/O functions. |
918 | * wasted on making it similar to other I/O functions. |
918 | */ |
919 | */ |
919 | { |
920 | { |
920 | off_t sbytes; |
921 | off_t sbytes; |
921 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
922 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
922 | |
923 | |
923 | if (res < 0 && sbytes) |
924 | /* freebsd' sendfile will return 0 on success */ |
924 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
925 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
926 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
927 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
|
|
928 | res = sbytes; |
|
|
929 | } |
|
|
930 | |
|
|
931 | # elif defined (__APPLE__) |
|
|
932 | |
|
|
933 | { |
|
|
934 | off_t sbytes = count; |
|
|
935 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
936 | |
|
|
937 | /* according to the manpage, sbytes is always valid */ |
|
|
938 | if (sbytes) |
925 | res = sbytes; |
939 | res = sbytes; |
926 | } |
940 | } |
927 | |
941 | |
928 | # elif __hpux |
942 | # elif __hpux |
929 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
943 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
… | |
… | |
943 | if (res < 0 && sbytes) |
957 | if (res < 0 && sbytes) |
944 | res = sbytes; |
958 | res = sbytes; |
945 | } |
959 | } |
946 | |
960 | |
947 | # endif |
961 | # endif |
|
|
962 | |
|
|
963 | #elif defined (_WIN32) |
|
|
964 | |
|
|
965 | /* does not work, just for documentation of what would need to be done */ |
|
|
966 | { |
|
|
967 | HANDLE h = TO_SOCKET (ifd); |
|
|
968 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
969 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
970 | } |
|
|
971 | |
948 | #else |
972 | #else |
949 | res = -1; |
973 | res = -1; |
950 | errno = ENOSYS; |
974 | errno = ENOSYS; |
951 | #endif |
975 | #endif |
952 | |
976 | |
953 | if (res < 0 |
977 | if (res < 0 |
954 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
978 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
979 | /* BSDs */ |
|
|
980 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
981 | || errno == ENOTSUP |
|
|
982 | #endif |
|
|
983 | || errno == EOPNOTSUPP /* BSDs */ |
955 | #if __solaris |
984 | #if __solaris |
956 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
985 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
957 | #endif |
986 | #endif |
958 | ) |
987 | ) |
959 | ) |
988 | ) |
… | |
… | |
993 | } |
1022 | } |
994 | |
1023 | |
995 | static signed char |
1024 | static signed char |
996 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1025 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
997 | { |
1026 | { |
998 | return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ |
1027 | 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; |
1028 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
1000 | } |
1029 | } |
1001 | |
1030 | |
1002 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1031 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1003 | |
1032 | |
1004 | #define EIO_QSORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1033 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1005 | #define EIO_QSORT_SKIP 60 /* when to skip qsort completely */ |
1034 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
1006 | |
1035 | |
1007 | static void |
1036 | static void |
1008 | eio_dent_sort (eio_dirent *dents, int size) |
1037 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1009 | { |
1038 | { |
|
|
1039 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1040 | unsigned char *bit = bits; |
|
|
1041 | |
|
|
1042 | assert (CHAR_BIT == 8); |
|
|
1043 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1044 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
|
|
1045 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
|
|
1046 | |
|
|
1047 | if (size <= EIO_SORT_FAST) |
|
|
1048 | return; |
|
|
1049 | |
|
|
1050 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1051 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1052 | /* inode_bits must contain all inodes ORed together */ |
|
|
1053 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1054 | { |
|
|
1055 | ino_t endianness; |
1010 | int i, j; |
1056 | int i, j; |
1011 | |
1057 | |
|
|
1058 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1059 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1060 | ((unsigned char *)&endianness)[i] = i; |
|
|
1061 | |
|
|
1062 | *bit++ = 0; |
|
|
1063 | |
|
|
1064 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1065 | { |
|
|
1066 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1067 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1068 | endianness >>= 8; |
|
|
1069 | |
|
|
1070 | for (j = 0; j < 8; ++j) |
|
|
1071 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1072 | *bit++ = offs + j; |
|
|
1073 | } |
|
|
1074 | |
|
|
1075 | for (j = 0; j < 8; ++j) |
|
|
1076 | if (score_bits & (1 << j)) |
|
|
1077 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1078 | } |
|
|
1079 | |
|
|
1080 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1081 | { |
|
|
1082 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1083 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1084 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1085 | int stk_idx = 0; |
|
|
1086 | |
|
|
1087 | base_stk [stk_idx] = dents; |
|
|
1088 | end_stk [stk_idx] = dents + size; |
|
|
1089 | bit_stk [stk_idx] = bit - 1; |
|
|
1090 | |
|
|
1091 | do |
|
|
1092 | { |
|
|
1093 | base = base_stk [stk_idx]; |
|
|
1094 | end = end_stk [stk_idx]; |
|
|
1095 | bit = bit_stk [stk_idx]; |
|
|
1096 | |
|
|
1097 | for (;;) |
|
|
1098 | { |
|
|
1099 | unsigned char O = *bit >> 3; |
|
|
1100 | unsigned char M = 1 << (*bit & 7); |
|
|
1101 | |
|
|
1102 | eio_dirent *a = base; |
|
|
1103 | eio_dirent *b = end; |
|
|
1104 | |
|
|
1105 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1106 | break; |
|
|
1107 | |
|
|
1108 | /* now bit-partition the array on the bit */ |
|
|
1109 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1110 | /* partition algos found in the literature */ |
|
|
1111 | do |
|
|
1112 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1113 | ++a; |
|
|
1114 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1115 | { |
|
|
1116 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1117 | ++a; |
|
|
1118 | } |
|
|
1119 | while (b > a); |
|
|
1120 | |
|
|
1121 | /* next bit, or stop, if no bits left in this path */ |
|
|
1122 | if (!*--bit) |
|
|
1123 | break; |
|
|
1124 | |
|
|
1125 | base_stk [stk_idx] = a; |
|
|
1126 | end_stk [stk_idx] = end; |
|
|
1127 | bit_stk [stk_idx] = bit; |
|
|
1128 | ++stk_idx; |
|
|
1129 | |
|
|
1130 | end = a; |
|
|
1131 | } |
|
|
1132 | } |
|
|
1133 | while (stk_idx--); |
|
|
1134 | } |
|
|
1135 | } |
|
|
1136 | |
|
|
1137 | static void |
|
|
1138 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1139 | { |
|
|
1140 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1141 | { |
|
|
1142 | int i; |
|
|
1143 | eio_dirent *min = dents; |
|
|
1144 | |
|
|
1145 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1146 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1147 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1148 | min = &dents [i]; |
|
|
1149 | |
|
|
1150 | /* swap elements 0 and j (minimum) */ |
|
|
1151 | { |
|
|
1152 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1153 | } |
|
|
1154 | } |
|
|
1155 | |
|
|
1156 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1157 | { |
|
|
1158 | eio_dirent *i, *j; |
|
|
1159 | |
|
|
1160 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1161 | { |
|
|
1162 | eio_dirent value = *i; |
|
|
1163 | |
|
|
1164 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1165 | j [1] = j [0]; |
|
|
1166 | |
|
|
1167 | j [1] = value; |
|
|
1168 | } |
|
|
1169 | } |
|
|
1170 | } |
|
|
1171 | |
|
|
1172 | static void |
|
|
1173 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1174 | { |
1012 | if (size <= 1) |
1175 | if (size <= 1) |
1013 | return; /* our insertion sort relies on size > 0 */ |
1176 | return; /* our insertion sort relies on size > 0 */ |
1014 | |
1177 | |
1015 | if (size > EIO_QSORT_SKIP) /* skip quicksort for small directories */ |
1178 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
1016 | { |
1179 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
1017 | /* first, use quicksort */ |
1180 | 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 | |
1181 | |
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 */ |
1182 | /* 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 */ |
1183 | /* as insertion sort is more efficient for small partitions */ |
1070 | { |
1184 | eio_dent_insertion_sort (dents, size); |
1071 | int min = 0; |
|
|
1072 | |
|
|
1073 | for (i = size > EIO_QSORT_SKIP ? EIO_QSORT_CUTOFF : 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 | dents [j + 1] = dents [j]; |
|
|
1090 | |
|
|
1091 | dents [j + 1] = value; |
|
|
1092 | } |
|
|
1093 | } |
1185 | } |
1094 | |
1186 | |
1095 | /* read a full directory */ |
1187 | /* read a full directory */ |
1096 | static void |
1188 | static void |
1097 | eio__scandir (eio_req *req, etp_worker *self) |
1189 | eio__scandir (eio_req *req, etp_worker *self) |
… | |
… | |
1103 | int namesoffs = 0; |
1195 | int namesoffs = 0; |
1104 | int flags = req->int1; |
1196 | int flags = req->int1; |
1105 | eio_dirent *dents = 0; |
1197 | eio_dirent *dents = 0; |
1106 | int dentalloc = 128; |
1198 | int dentalloc = 128; |
1107 | int dentoffs = 0; |
1199 | int dentoffs = 0; |
|
|
1200 | ino_t inode_bits = 0; |
1108 | |
1201 | |
1109 | req->result = -1; |
1202 | req->result = -1; |
1110 | |
1203 | |
1111 | if (!(flags & EIO_READDIR_DENTS)) |
1204 | if (!(flags & EIO_READDIR_DENTS)) |
1112 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1205 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1113 | |
1206 | |
1114 | X_LOCK (wrklock); |
1207 | X_LOCK (wrklock); |
1115 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1208 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1116 | self->dirp = dirp = opendir (req->ptr1); |
1209 | self->dirp = dirp = opendir (req->ptr1); |
1117 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1210 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1118 | req->ptr1 = names = malloc (namesalloc); |
|
|
1119 | req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1211 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
|
|
1212 | req->ptr2 = names = malloc (namesalloc); |
1120 | X_UNLOCK (wrklock); |
1213 | X_UNLOCK (wrklock); |
1121 | |
1214 | |
1122 | if (dirp && names && (!flags || dents)) |
1215 | if (dirp && names && (!flags || dents)) |
1123 | for (;;) |
1216 | for (;;) |
1124 | { |
1217 | { |
… | |
… | |
1132 | |
1225 | |
1133 | /* sort etc. */ |
1226 | /* sort etc. */ |
1134 | req->int1 = flags; |
1227 | req->int1 = flags; |
1135 | req->result = dentoffs; |
1228 | req->result = dentoffs; |
1136 | |
1229 | |
1137 | if (dents) |
|
|
1138 | { |
|
|
1139 | eio_dirent *ent = dents + dentoffs; |
|
|
1140 | |
|
|
1141 | while (ent > dents) |
|
|
1142 | (--ent)->name = names + (size_t)ent->name; |
|
|
1143 | } |
|
|
1144 | |
|
|
1145 | if (flags & EIO_READDIR_STAT_ORDER |
1230 | if (flags & EIO_READDIR_STAT_ORDER) |
1146 | || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) |
1231 | eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ |
1147 | eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */ |
|
|
1148 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1232 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1233 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1234 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1235 | else |
1149 | { |
1236 | { |
1150 | /* in this case, all is known, and we just put dirs first and sort them */ |
1237 | /* in this case, all is known, and we just put dirs first and sort them */ |
1151 | eio_dirent *ent = dents + dentoffs; |
1238 | eio_dirent *oth = dents + dentoffs; |
1152 | eio_dirent *dir = dents; |
1239 | eio_dirent *dir = dents; |
1153 | |
1240 | |
1154 | /* now move dirs to the front, and non-dirs to the back */ |
1241 | /* now partition dirs to the front, and non-dirs to the back */ |
1155 | /* by walking from both sides and swapping if necessary */ |
1242 | /* by walking from both sides and swapping if necessary */ |
|
|
1243 | /* also clear score, so it doesn't influence sorting */ |
1156 | while (ent > dir) |
1244 | while (oth > dir) |
1157 | { |
1245 | { |
1158 | if (dir->type == DT_DIR) |
1246 | if (dir->type == EIO_DT_DIR) |
1159 | ++dir; |
1247 | ++dir; |
1160 | else |
|
|
1161 | { |
|
|
1162 | --ent; |
|
|
1163 | |
|
|
1164 | if (ent->type == DT_DIR) |
1248 | else if ((--oth)->type == EIO_DT_DIR) |
1165 | { |
1249 | { |
1166 | eio_dirent tmp = *dir; |
1250 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
1167 | *dir = *ent; |
|
|
1168 | *ent = tmp; |
|
|
1169 | |
1251 | |
1170 | ++dir; |
1252 | ++dir; |
1171 | } |
1253 | } |
1172 | } |
1254 | } |
1173 | } |
|
|
1174 | |
1255 | |
1175 | /* now sort the dirs only */ |
1256 | /* now sort the dirs only */ |
1176 | eio_dent_sort (dents, dir - dents); |
1257 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1177 | } |
1258 | } |
1178 | |
|
|
1179 | /* only provide the names array unless DENTS is specified */ |
|
|
1180 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1181 | { |
|
|
1182 | X_LOCK (wrklock); |
|
|
1183 | assert (!dents); |
|
|
1184 | req->ptr1 = 0; |
|
|
1185 | req->ptr2 = names; |
|
|
1186 | X_UNLOCK (wrklock); |
|
|
1187 | } |
|
|
1188 | |
1259 | |
1189 | break; |
1260 | break; |
1190 | } |
1261 | } |
1191 | |
1262 | |
1192 | /* now add the entry to our list(s) */ |
1263 | /* now add the entry to our list(s) */ |
… | |
… | |
1199 | |
1270 | |
1200 | while (expect_false (namesoffs + len > namesalloc)) |
1271 | while (expect_false (namesoffs + len > namesalloc)) |
1201 | { |
1272 | { |
1202 | namesalloc *= 2; |
1273 | namesalloc *= 2; |
1203 | X_LOCK (wrklock); |
1274 | X_LOCK (wrklock); |
1204 | req->ptr1 = names = realloc (names, namesalloc); |
1275 | req->ptr2 = names = realloc (names, namesalloc); |
1205 | X_UNLOCK (wrklock); |
1276 | X_UNLOCK (wrklock); |
1206 | |
1277 | |
1207 | if (!names) |
1278 | if (!names) |
1208 | break; |
1279 | break; |
1209 | } |
1280 | } |
… | |
… | |
1216 | |
1287 | |
1217 | if (expect_false (dentoffs == dentalloc)) |
1288 | if (expect_false (dentoffs == dentalloc)) |
1218 | { |
1289 | { |
1219 | dentalloc *= 2; |
1290 | dentalloc *= 2; |
1220 | X_LOCK (wrklock); |
1291 | X_LOCK (wrklock); |
1221 | req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1292 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1222 | X_UNLOCK (wrklock); |
1293 | X_UNLOCK (wrklock); |
1223 | |
1294 | |
1224 | if (!dents) |
1295 | if (!dents) |
1225 | break; |
1296 | break; |
1226 | } |
1297 | } |
1227 | |
1298 | |
1228 | ent = dents + dentoffs; |
1299 | ent = dents + dentoffs; |
1229 | |
1300 | |
1230 | ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ |
1301 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
1231 | ent->namelen = len - 1; |
1302 | ent->namelen = len - 1; |
1232 | ent->inode = D_INO (entp); |
1303 | ent->inode = D_INO (entp); |
|
|
1304 | |
|
|
1305 | inode_bits |= ent->inode; |
1233 | |
1306 | |
1234 | switch (D_TYPE (entp)) |
1307 | switch (D_TYPE (entp)) |
1235 | { |
1308 | { |
1236 | default: |
1309 | default: |
1237 | ent->type = EIO_DT_UNKNOWN; |
1310 | ent->type = EIO_DT_UNKNOWN; |
… | |
… | |
1280 | #ifdef DT_WHT |
1353 | #ifdef DT_WHT |
1281 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1354 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1282 | #endif |
1355 | #endif |
1283 | } |
1356 | } |
1284 | |
1357 | |
1285 | ent->score = 0; |
1358 | ent->score = 7; |
1286 | |
1359 | |
1287 | if (flags & EIO_READDIR_DIRS_FIRST) |
1360 | if (flags & EIO_READDIR_DIRS_FIRST) |
1288 | { |
1361 | { |
1289 | if (ent->type == EIO_DT_UNKNOWN) |
1362 | if (ent->type == EIO_DT_UNKNOWN) |
1290 | { |
1363 | { |
1291 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1364 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1292 | ent->score = 98; |
1365 | ent->score = 1; |
1293 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1366 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1294 | ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ |
1367 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
1295 | } |
1368 | } |
1296 | else if (ent->type == EIO_DT_DIR) |
1369 | else if (ent->type == EIO_DT_DIR) |
1297 | ent->score = 100; |
1370 | ent->score = 0; |
1298 | } |
1371 | } |
1299 | } |
1372 | } |
1300 | |
1373 | |
1301 | namesoffs += len; |
1374 | namesoffs += len; |
1302 | ++dentoffs; |
1375 | ++dentoffs; |
… | |
… | |
1309 | } |
1382 | } |
1310 | } |
1383 | } |
1311 | } |
1384 | } |
1312 | |
1385 | |
1313 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1386 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1314 | # undef msync |
|
|
1315 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
1387 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1388 | #else |
|
|
1389 | |
|
|
1390 | int |
|
|
1391 | eio__msync (void *mem, size_t len, int flags) |
|
|
1392 | { |
|
|
1393 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1394 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1395 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1396 | { |
|
|
1397 | flags = 0 |
|
|
1398 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1399 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1400 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1401 | } |
|
|
1402 | |
|
|
1403 | return msync (mem, len, flags); |
|
|
1404 | } |
|
|
1405 | |
1316 | #endif |
1406 | #endif |
1317 | |
1407 | |
1318 | int |
1408 | int |
1319 | eio__mtouch (void *mem, size_t len, int flags) |
1409 | eio__mtouch (void *mem, size_t len, int flags) |
1320 | { |
1410 | { |
… | |
… | |
1327 | |
1417 | |
1328 | if (!page) |
1418 | if (!page) |
1329 | page = sysconf (_SC_PAGESIZE); |
1419 | page = sysconf (_SC_PAGESIZE); |
1330 | #endif |
1420 | #endif |
1331 | |
1421 | |
|
|
1422 | /* round down to start of page, although this is probably useless */ |
1332 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1423 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1333 | |
1424 | |
1334 | if (addr < end) |
1425 | if (addr < end) |
1335 | if (flags) /* modify */ |
1426 | if (flags & EIO_MT_MODIFY) /* modify */ |
1336 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1427 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1337 | else |
1428 | else |
1338 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1429 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1339 | |
1430 | |
1340 | return 0; |
1431 | return 0; |
… | |
… | |
1488 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1579 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1489 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1580 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1490 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1581 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1491 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1582 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1492 | |
1583 | |
|
|
1584 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1585 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1586 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1587 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1588 | |
1493 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1589 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1494 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1590 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1495 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1591 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1496 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1592 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1497 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1593 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1512 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1608 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1513 | |
1609 | |
1514 | case EIO_SYNC: req->result = 0; sync (); break; |
1610 | case EIO_SYNC: req->result = 0; sync (); break; |
1515 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1611 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1516 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1612 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1517 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1613 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1518 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1614 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1519 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1615 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1520 | |
1616 | |
1521 | case EIO_READDIR: eio__scandir (req, self); break; |
1617 | case EIO_READDIR: eio__scandir (req, self); break; |
1522 | |
1618 | |
… | |
… | |
1644 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1740 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1645 | { |
1741 | { |
1646 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1742 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1647 | } |
1743 | } |
1648 | |
1744 | |
|
|
1745 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
1746 | { |
|
|
1747 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
|
|
1748 | } |
|
|
1749 | |
1649 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1750 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1650 | { |
1751 | { |
1651 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1752 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1652 | } |
1753 | } |
1653 | |
1754 | |
… | |
… | |
1723 | } |
1824 | } |
1724 | |
1825 | |
1725 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1826 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1726 | { |
1827 | { |
1727 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1828 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
|
|
1829 | } |
|
|
1830 | |
|
|
1831 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
1832 | { |
|
|
1833 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
1728 | } |
1834 | } |
1729 | |
1835 | |
1730 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1836 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1731 | { |
1837 | { |
1732 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1838 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |