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 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
924 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
925 | /* freebsd' sendfile will return 0 on success */ |
|
|
926 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
927 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
928 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
|
|
929 | res = sbytes; |
|
|
930 | #endif |
|
|
931 | |
|
|
932 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
933 | if (sbytes) |
|
|
934 | res = sbytes; |
|
|
935 | } |
|
|
936 | |
|
|
937 | # elif defined (__APPLE__) |
|
|
938 | |
|
|
939 | { |
|
|
940 | off_t sbytes = count; |
|
|
941 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
942 | |
|
|
943 | /* according to the manpage, sbytes is always valid */ |
|
|
944 | if (sbytes) |
925 | res = sbytes; |
945 | res = sbytes; |
926 | } |
946 | } |
927 | |
947 | |
928 | # elif __hpux |
948 | # elif __hpux |
929 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
949 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
… | |
… | |
943 | if (res < 0 && sbytes) |
963 | if (res < 0 && sbytes) |
944 | res = sbytes; |
964 | res = sbytes; |
945 | } |
965 | } |
946 | |
966 | |
947 | # endif |
967 | # endif |
|
|
968 | |
|
|
969 | #elif defined (_WIN32) |
|
|
970 | |
|
|
971 | /* does not work, just for documentation of what would need to be done */ |
|
|
972 | { |
|
|
973 | HANDLE h = TO_SOCKET (ifd); |
|
|
974 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
975 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
976 | } |
|
|
977 | |
948 | #else |
978 | #else |
949 | res = -1; |
979 | res = -1; |
950 | errno = ENOSYS; |
980 | errno = ENOSYS; |
951 | #endif |
981 | #endif |
952 | |
982 | |
953 | if (res < 0 |
983 | if (res < 0 |
954 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
984 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
985 | /* BSDs */ |
|
|
986 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
987 | || errno == ENOTSUP |
|
|
988 | #endif |
|
|
989 | || errno == EOPNOTSUPP /* BSDs */ |
955 | #if __solaris |
990 | #if __solaris |
956 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
991 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
957 | #endif |
992 | #endif |
958 | ) |
993 | ) |
959 | ) |
994 | ) |
… | |
… | |
993 | } |
1028 | } |
994 | |
1029 | |
995 | static signed char |
1030 | static signed char |
996 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1031 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
997 | { |
1032 | { |
998 | return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ |
1033 | 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; |
1034 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
1000 | } |
1035 | } |
1001 | |
1036 | |
|
|
1037 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1038 | |
1002 | #define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */ |
1039 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1040 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
1003 | |
1041 | |
1004 | static void |
1042 | static void |
1005 | eio_dent_sort (eio_dirent *dents, int size) |
1043 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1006 | { |
1044 | { |
1007 | /* should be good for 2**31 entries */ |
1045 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
1008 | struct rng { int l, r; } rng [32]; |
1046 | unsigned char *bit = bits; |
|
|
1047 | |
|
|
1048 | assert (CHAR_BIT == 8); |
|
|
1049 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1050 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
|
|
1051 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
|
|
1052 | |
|
|
1053 | if (size <= EIO_SORT_FAST) |
|
|
1054 | return; |
|
|
1055 | |
|
|
1056 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1057 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1058 | /* inode_bits must contain all inodes ORed together */ |
|
|
1059 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1060 | { |
|
|
1061 | ino_t endianness; |
1009 | int i, j; |
1062 | int i, j; |
1010 | |
1063 | |
1011 | i = 0; |
1064 | /* we store the byte offset of byte n into byte n of "endianness" */ |
1012 | rng[0].l = 0; |
1065 | for (i = 0; i < sizeof (ino_t); ++i) |
1013 | rng[0].r = size; |
1066 | ((unsigned char *)&endianness)[i] = i; |
1014 | |
1067 | |
1015 | while (expect_true (i >= 0)) |
1068 | *bit++ = 0; |
|
|
1069 | |
|
|
1070 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1071 | { |
|
|
1072 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1073 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1074 | endianness >>= 8; |
|
|
1075 | |
|
|
1076 | for (j = 0; j < 8; ++j) |
|
|
1077 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1078 | *bit++ = offs + j; |
|
|
1079 | } |
|
|
1080 | |
|
|
1081 | for (j = 0; j < 8; ++j) |
|
|
1082 | if (score_bits & (1 << j)) |
|
|
1083 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1084 | } |
|
|
1085 | |
|
|
1086 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1087 | { |
|
|
1088 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1089 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1090 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1091 | int stk_idx = 0; |
|
|
1092 | |
|
|
1093 | base_stk [stk_idx] = dents; |
|
|
1094 | end_stk [stk_idx] = dents + size; |
|
|
1095 | bit_stk [stk_idx] = bit - 1; |
|
|
1096 | |
|
|
1097 | do |
|
|
1098 | { |
|
|
1099 | base = base_stk [stk_idx]; |
|
|
1100 | end = end_stk [stk_idx]; |
|
|
1101 | bit = bit_stk [stk_idx]; |
|
|
1102 | |
|
|
1103 | for (;;) |
|
|
1104 | { |
|
|
1105 | unsigned char O = *bit >> 3; |
|
|
1106 | unsigned char M = 1 << (*bit & 7); |
|
|
1107 | |
|
|
1108 | eio_dirent *a = base; |
|
|
1109 | eio_dirent *b = end; |
|
|
1110 | |
|
|
1111 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1112 | break; |
|
|
1113 | |
|
|
1114 | /* now bit-partition the array on the bit */ |
|
|
1115 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1116 | /* partition algos found in the literature */ |
|
|
1117 | do |
|
|
1118 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1119 | ++a; |
|
|
1120 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1121 | { |
|
|
1122 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1123 | ++a; |
|
|
1124 | } |
|
|
1125 | while (b > a); |
|
|
1126 | |
|
|
1127 | /* next bit, or stop, if no bits left in this path */ |
|
|
1128 | if (!*--bit) |
|
|
1129 | break; |
|
|
1130 | |
|
|
1131 | base_stk [stk_idx] = a; |
|
|
1132 | end_stk [stk_idx] = end; |
|
|
1133 | bit_stk [stk_idx] = bit; |
|
|
1134 | ++stk_idx; |
|
|
1135 | |
|
|
1136 | end = a; |
|
|
1137 | } |
|
|
1138 | } |
|
|
1139 | while (stk_idx--); |
|
|
1140 | } |
|
|
1141 | } |
|
|
1142 | |
|
|
1143 | static void |
|
|
1144 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1145 | { |
|
|
1146 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1147 | { |
|
|
1148 | int i; |
|
|
1149 | eio_dirent *min = dents; |
|
|
1150 | |
|
|
1151 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1152 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1153 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1154 | min = &dents [i]; |
|
|
1155 | |
|
|
1156 | /* swap elements 0 and j (minimum) */ |
1016 | { |
1157 | { |
1017 | int L = rng [i].l; |
1158 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
1018 | int R = rng [i].r - 1; |
1159 | } |
|
|
1160 | } |
1019 | |
1161 | |
1020 | if (expect_false (L + EIO_QSORT_CUTOFF < R)) |
1162 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1163 | { |
|
|
1164 | eio_dirent *i, *j; |
|
|
1165 | |
|
|
1166 | for (i = dents + 1; i < dents + size; ++i) |
1021 | { |
1167 | { |
1022 | eio_dirent piv = dents [L]; |
1168 | eio_dirent value = *i; |
1023 | |
1169 | |
1024 | while (L < R) |
1170 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
1025 | { |
1171 | j [1] = j [0]; |
1026 | while (eio_dent_cmp (&dents [R], &piv) >= 0 && L < R) |
|
|
1027 | --R; |
|
|
1028 | |
1172 | |
1029 | if (L < R) |
1173 | j [1] = value; |
1030 | dents [L++] = dents [R]; |
|
|
1031 | |
|
|
1032 | while (eio_dent_cmp (&dents [L], &piv) <= 0 && L < R) |
|
|
1033 | ++L; |
|
|
1034 | |
|
|
1035 | if (L < R) |
|
|
1036 | dents [R--] = dents [L]; |
|
|
1037 | } |
|
|
1038 | |
|
|
1039 | dents [L] = piv; |
|
|
1040 | |
|
|
1041 | ++i; |
|
|
1042 | rng [i].l = L + 1; |
|
|
1043 | rng [i].r = rng [i - 1].r; |
|
|
1044 | rng [i - 1].r = L; |
|
|
1045 | |
|
|
1046 | if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l) |
|
|
1047 | { |
|
|
1048 | struct rng t; |
|
|
1049 | |
|
|
1050 | t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t; |
|
|
1051 | } |
|
|
1052 | } |
1174 | } |
1053 | else |
|
|
1054 | --i; |
|
|
1055 | } |
1175 | } |
|
|
1176 | } |
1056 | |
1177 | |
1057 | /* use a simple insertion sort at the end */ |
1178 | static void |
1058 | for (i = 1; i < size; ++i) |
1179 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1059 | { |
1180 | { |
1060 | eio_dirent value = dents [i]; |
1181 | if (size <= 1) |
|
|
1182 | return; /* our insertion sort relies on size > 0 */ |
1061 | |
1183 | |
1062 | for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j) |
1184 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
1063 | dents [j + 1] = dents [j]; |
1185 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1186 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
1064 | |
1187 | |
1065 | dents [j + 1] = value; |
1188 | /* use an insertion sort at the end, or for small arrays, */ |
1066 | } |
1189 | /* as insertion sort is more efficient for small partitions */ |
|
|
1190 | eio_dent_insertion_sort (dents, size); |
1067 | } |
1191 | } |
1068 | |
1192 | |
1069 | /* read a full directory */ |
1193 | /* read a full directory */ |
1070 | static void |
1194 | static void |
1071 | eio__scandir (eio_req *req, etp_worker *self) |
1195 | eio__scandir (eio_req *req, etp_worker *self) |
… | |
… | |
1077 | int namesoffs = 0; |
1201 | int namesoffs = 0; |
1078 | int flags = req->int1; |
1202 | int flags = req->int1; |
1079 | eio_dirent *dents = 0; |
1203 | eio_dirent *dents = 0; |
1080 | int dentalloc = 128; |
1204 | int dentalloc = 128; |
1081 | int dentoffs = 0; |
1205 | int dentoffs = 0; |
|
|
1206 | ino_t inode_bits = 0; |
1082 | |
1207 | |
1083 | req->result = -1; |
1208 | req->result = -1; |
1084 | |
1209 | |
1085 | if (!(flags & EIO_READDIR_DENTS)) |
1210 | if (!(flags & EIO_READDIR_DENTS)) |
1086 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1211 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1087 | |
1212 | |
1088 | X_LOCK (wrklock); |
1213 | X_LOCK (wrklock); |
1089 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1214 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1090 | self->dirp = dirp = opendir (req->ptr1); |
1215 | self->dirp = dirp = opendir (req->ptr1); |
1091 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1216 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1092 | req->ptr1 = names = malloc (namesalloc); |
|
|
1093 | req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1217 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
|
|
1218 | req->ptr2 = names = malloc (namesalloc); |
1094 | X_UNLOCK (wrklock); |
1219 | X_UNLOCK (wrklock); |
1095 | |
1220 | |
1096 | if (dirp && names && (!flags || dents)) |
1221 | if (dirp && names && (!flags || dents)) |
1097 | for (;;) |
1222 | for (;;) |
1098 | { |
1223 | { |
… | |
… | |
1106 | |
1231 | |
1107 | /* sort etc. */ |
1232 | /* sort etc. */ |
1108 | req->int1 = flags; |
1233 | req->int1 = flags; |
1109 | req->result = dentoffs; |
1234 | req->result = dentoffs; |
1110 | |
1235 | |
1111 | if (dents) |
|
|
1112 | { |
|
|
1113 | eio_dirent *ent = dents + dentoffs; |
|
|
1114 | |
|
|
1115 | while (ent > dents) |
|
|
1116 | (--ent)->name = names + (size_t)ent->name; |
|
|
1117 | } |
|
|
1118 | |
|
|
1119 | if (flags & EIO_READDIR_STAT_ORDER |
1236 | if (flags & EIO_READDIR_STAT_ORDER) |
1120 | || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) |
1237 | eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ |
1121 | eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */ |
|
|
1122 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1238 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1239 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1240 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1241 | else |
1123 | { |
1242 | { |
1124 | /* in this case, all is known, and we just put dirs first and sort them */ |
1243 | /* in this case, all is known, and we just put dirs first and sort them */ |
1125 | eio_dirent *ent = dents + dentoffs; |
1244 | eio_dirent *oth = dents + dentoffs; |
1126 | eio_dirent *dir = dents; |
1245 | eio_dirent *dir = dents; |
1127 | |
1246 | |
1128 | /* now move dirs to the front, and non-dirs to the back */ |
1247 | /* now partition dirs to the front, and non-dirs to the back */ |
1129 | /* by walking from both sides and swapping if necessary */ |
1248 | /* by walking from both sides and swapping if necessary */ |
|
|
1249 | /* also clear score, so it doesn't influence sorting */ |
1130 | while (ent > dir) |
1250 | while (oth > dir) |
1131 | { |
1251 | { |
1132 | if (dir->type == DT_DIR) |
1252 | if (dir->type == EIO_DT_DIR) |
1133 | ++dir; |
1253 | ++dir; |
1134 | else |
|
|
1135 | { |
|
|
1136 | --ent; |
|
|
1137 | |
|
|
1138 | if (ent->type == DT_DIR) |
1254 | else if ((--oth)->type == EIO_DT_DIR) |
1139 | { |
1255 | { |
1140 | eio_dirent tmp = *dir; |
1256 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
1141 | *dir = *ent; |
|
|
1142 | *ent = tmp; |
|
|
1143 | |
1257 | |
1144 | ++dir; |
1258 | ++dir; |
1145 | } |
1259 | } |
1146 | } |
1260 | } |
1147 | } |
|
|
1148 | |
1261 | |
1149 | /* now sort the dirs only */ |
1262 | /* now sort the dirs only */ |
1150 | eio_dent_sort (dents, dir - dents); |
1263 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1151 | } |
1264 | } |
1152 | |
|
|
1153 | /* only provide the names array unless DENTS is specified */ |
|
|
1154 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1155 | { |
|
|
1156 | X_LOCK (wrklock); |
|
|
1157 | assert (!dents); |
|
|
1158 | req->ptr1 = 0; |
|
|
1159 | req->ptr2 = names; |
|
|
1160 | X_UNLOCK (wrklock); |
|
|
1161 | } |
|
|
1162 | |
1265 | |
1163 | break; |
1266 | break; |
1164 | } |
1267 | } |
1165 | |
1268 | |
1166 | /* now add the entry to our list(s) */ |
1269 | /* now add the entry to our list(s) */ |
… | |
… | |
1173 | |
1276 | |
1174 | while (expect_false (namesoffs + len > namesalloc)) |
1277 | while (expect_false (namesoffs + len > namesalloc)) |
1175 | { |
1278 | { |
1176 | namesalloc *= 2; |
1279 | namesalloc *= 2; |
1177 | X_LOCK (wrklock); |
1280 | X_LOCK (wrklock); |
1178 | req->ptr1 = names = realloc (names, namesalloc); |
1281 | req->ptr2 = names = realloc (names, namesalloc); |
1179 | X_UNLOCK (wrklock); |
1282 | X_UNLOCK (wrklock); |
1180 | |
1283 | |
1181 | if (!names) |
1284 | if (!names) |
1182 | break; |
1285 | break; |
1183 | } |
1286 | } |
… | |
… | |
1190 | |
1293 | |
1191 | if (expect_false (dentoffs == dentalloc)) |
1294 | if (expect_false (dentoffs == dentalloc)) |
1192 | { |
1295 | { |
1193 | dentalloc *= 2; |
1296 | dentalloc *= 2; |
1194 | X_LOCK (wrklock); |
1297 | X_LOCK (wrklock); |
1195 | req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1298 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1196 | X_UNLOCK (wrklock); |
1299 | X_UNLOCK (wrklock); |
1197 | |
1300 | |
1198 | if (!dents) |
1301 | if (!dents) |
1199 | break; |
1302 | break; |
1200 | } |
1303 | } |
1201 | |
1304 | |
1202 | ent = dents + dentoffs; |
1305 | ent = dents + dentoffs; |
1203 | |
1306 | |
1204 | ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ |
1307 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
1205 | ent->namelen = len - 1; |
1308 | ent->namelen = len - 1; |
1206 | ent->inode = D_INO (entp); |
1309 | ent->inode = D_INO (entp); |
|
|
1310 | |
|
|
1311 | inode_bits |= ent->inode; |
1207 | |
1312 | |
1208 | switch (D_TYPE (entp)) |
1313 | switch (D_TYPE (entp)) |
1209 | { |
1314 | { |
1210 | default: |
1315 | default: |
1211 | ent->type = EIO_DT_UNKNOWN; |
1316 | ent->type = EIO_DT_UNKNOWN; |
… | |
… | |
1254 | #ifdef DT_WHT |
1359 | #ifdef DT_WHT |
1255 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1360 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1256 | #endif |
1361 | #endif |
1257 | } |
1362 | } |
1258 | |
1363 | |
1259 | ent->score = 0; |
1364 | ent->score = 7; |
1260 | |
1365 | |
1261 | if (flags & EIO_READDIR_DIRS_FIRST) |
1366 | if (flags & EIO_READDIR_DIRS_FIRST) |
1262 | { |
1367 | { |
1263 | if (ent->type == EIO_DT_UNKNOWN) |
1368 | if (ent->type == EIO_DT_UNKNOWN) |
1264 | { |
1369 | { |
1265 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1370 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1266 | ent->score = 98; |
1371 | ent->score = 1; |
1267 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1372 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1268 | ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ |
1373 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
1269 | } |
1374 | } |
1270 | else if (ent->type == EIO_DT_DIR) |
1375 | else if (ent->type == EIO_DT_DIR) |
1271 | ent->score = 100; |
1376 | ent->score = 0; |
1272 | } |
1377 | } |
1273 | } |
1378 | } |
1274 | |
1379 | |
1275 | namesoffs += len; |
1380 | namesoffs += len; |
1276 | ++dentoffs; |
1381 | ++dentoffs; |
… | |
… | |
1283 | } |
1388 | } |
1284 | } |
1389 | } |
1285 | } |
1390 | } |
1286 | |
1391 | |
1287 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1392 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1288 | # undef msync |
|
|
1289 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
1393 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1394 | #else |
|
|
1395 | |
|
|
1396 | int |
|
|
1397 | eio__msync (void *mem, size_t len, int flags) |
|
|
1398 | { |
|
|
1399 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1400 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1401 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1402 | { |
|
|
1403 | flags = 0 |
|
|
1404 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1405 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1406 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1407 | } |
|
|
1408 | |
|
|
1409 | return msync (mem, len, flags); |
|
|
1410 | } |
|
|
1411 | |
1290 | #endif |
1412 | #endif |
1291 | |
1413 | |
1292 | int |
1414 | int |
1293 | eio__mtouch (void *mem, size_t len, int flags) |
1415 | eio__mtouch (void *mem, size_t len, int flags) |
1294 | { |
1416 | { |
… | |
… | |
1301 | |
1423 | |
1302 | if (!page) |
1424 | if (!page) |
1303 | page = sysconf (_SC_PAGESIZE); |
1425 | page = sysconf (_SC_PAGESIZE); |
1304 | #endif |
1426 | #endif |
1305 | |
1427 | |
|
|
1428 | /* round down to start of page, although this is probably useless */ |
1306 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1429 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1307 | |
1430 | |
1308 | if (addr < end) |
1431 | if (addr < end) |
1309 | if (flags) /* modify */ |
1432 | if (flags & EIO_MT_MODIFY) /* modify */ |
1310 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1433 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1311 | else |
1434 | else |
1312 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1435 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1313 | |
1436 | |
1314 | return 0; |
1437 | return 0; |
… | |
… | |
1462 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1585 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1463 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1586 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1464 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1587 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1465 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1588 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1466 | |
1589 | |
|
|
1590 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1591 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1592 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1593 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1594 | |
1467 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1595 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1468 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1596 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1469 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1597 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1470 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1598 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1471 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1599 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1486 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1614 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1487 | |
1615 | |
1488 | case EIO_SYNC: req->result = 0; sync (); break; |
1616 | case EIO_SYNC: req->result = 0; sync (); break; |
1489 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1617 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1490 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1618 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1491 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1619 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1492 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1620 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1493 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1621 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1494 | |
1622 | |
1495 | case EIO_READDIR: eio__scandir (req, self); break; |
1623 | case EIO_READDIR: eio__scandir (req, self); break; |
1496 | |
1624 | |
… | |
… | |
1618 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1746 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1619 | { |
1747 | { |
1620 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1748 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1621 | } |
1749 | } |
1622 | |
1750 | |
|
|
1751 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
1752 | { |
|
|
1753 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
|
|
1754 | } |
|
|
1755 | |
1623 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1756 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1624 | { |
1757 | { |
1625 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1758 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1626 | } |
1759 | } |
1627 | |
1760 | |
… | |
… | |
1697 | } |
1830 | } |
1698 | |
1831 | |
1699 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1832 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1700 | { |
1833 | { |
1701 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1834 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
|
|
1835 | } |
|
|
1836 | |
|
|
1837 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
1838 | { |
|
|
1839 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
1702 | } |
1840 | } |
1703 | |
1841 | |
1704 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1842 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1705 | { |
1843 | { |
1706 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1844 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |