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 |
… | |
… | |
73 | #else |
74 | #else |
74 | |
75 | |
75 | # include "config.h" |
76 | # include "config.h" |
76 | # include <sys/time.h> |
77 | # include <sys/time.h> |
77 | # include <sys/select.h> |
78 | # include <sys/select.h> |
|
|
79 | # include <sys/mman.h> |
78 | # include <unistd.h> |
80 | # include <unistd.h> |
79 | # include <utime.h> |
81 | # include <utime.h> |
80 | # include <signal.h> |
82 | # include <signal.h> |
81 | # include <dirent.h> |
83 | # include <dirent.h> |
82 | |
84 | |
83 | /* 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 */ |
84 | # if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
86 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
85 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
87 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
86 | # define D_INO(de) (de)->d_fileno |
88 | # define D_INO(de) (de)->d_fileno |
87 | # define D_NAMLEN(de) (de)->d_namlen |
89 | # define D_NAMLEN(de) (de)->d_namlen |
88 | # elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
90 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
89 | # define D_INO(de) (de)->d_ino |
91 | # define D_INO(de) (de)->d_ino |
90 | # endif |
92 | # endif |
91 | |
93 | |
92 | #ifdef _D_EXACT_NAMLEN |
94 | #ifdef _D_EXACT_NAMLEN |
93 | # undef D_NAMLEN |
95 | # undef D_NAMLEN |
… | |
… | |
105 | #endif |
107 | #endif |
106 | |
108 | |
107 | #if HAVE_SENDFILE |
109 | #if HAVE_SENDFILE |
108 | # if __linux |
110 | # if __linux |
109 | # include <sys/sendfile.h> |
111 | # include <sys/sendfile.h> |
110 | # elif __freebsd |
112 | # elif __FreeBSD__ || defined __APPLE__ |
111 | # include <sys/socket.h> |
113 | # include <sys/socket.h> |
112 | # include <sys/uio.h> |
114 | # include <sys/uio.h> |
113 | # elif __hpux |
115 | # elif __hpux |
114 | # include <sys/socket.h> |
116 | # include <sys/socket.h> |
115 | # elif __solaris /* not yet */ |
117 | # elif __solaris |
116 | # include <sys/sendfile.h> |
118 | # include <sys/sendfile.h> |
117 | # else |
119 | # else |
118 | # error sendfile support requested but not available |
120 | # error sendfile support requested but not available |
119 | # endif |
121 | # endif |
120 | #endif |
122 | #endif |
… | |
… | |
195 | |
197 | |
196 | /*****************************************************************************/ |
198 | /*****************************************************************************/ |
197 | |
199 | |
198 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
200 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
199 | |
201 | |
200 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
202 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
201 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
203 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
202 | { |
204 | { |
203 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
205 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
204 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
206 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
205 | } |
207 | } |
… | |
… | |
597 | } |
599 | } |
598 | |
600 | |
599 | static void etp_set_max_poll_time (double nseconds) |
601 | static void etp_set_max_poll_time (double nseconds) |
600 | { |
602 | { |
601 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
603 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
602 | max_poll_time = nseconds; |
604 | max_poll_time = nseconds * EIO_TICKS; |
603 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
605 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
604 | } |
606 | } |
605 | |
607 | |
606 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
608 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
607 | { |
609 | { |
… | |
… | |
866 | if (!res || errno != ENOSYS) |
868 | if (!res || errno != ENOSYS) |
867 | return res; |
869 | return res; |
868 | #endif |
870 | #endif |
869 | |
871 | |
870 | /* 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 |
871 | * call fdatasync, as thta matches the expectation of it's users best */ |
873 | * call fdatasync, as that matches the expectation of its users best */ |
872 | return fdatasync (fd); |
874 | return fdatasync (fd); |
873 | } |
875 | } |
874 | |
876 | |
875 | #if !HAVE_READAHEAD |
877 | #if !HAVE_READAHEAD |
876 | # undef readahead |
878 | # undef readahead |
… | |
… | |
908 | |
910 | |
909 | #if HAVE_SENDFILE |
911 | #if HAVE_SENDFILE |
910 | # if __linux |
912 | # if __linux |
911 | res = sendfile (ofd, ifd, &offset, count); |
913 | res = sendfile (ofd, ifd, &offset, count); |
912 | |
914 | |
913 | # elif __freebsd |
915 | # elif __FreeBSD__ |
914 | /* |
916 | /* |
915 | * 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 |
916 | * wasted on making it similar to other I/O functions. |
918 | * wasted on making it similar to other I/O functions. |
917 | */ |
919 | */ |
918 | { |
920 | { |
919 | off_t sbytes; |
921 | off_t sbytes; |
920 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
922 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
921 | |
923 | |
|
|
924 | /* freebsd' sendfile will return 0 when success */ |
922 | if (res < 0 && sbytes) |
925 | if (res == 0 && sbytes) |
923 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
926 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
|
|
927 | res = sbytes; |
|
|
928 | } |
|
|
929 | |
|
|
930 | # elif defined (__APPLE__) |
|
|
931 | |
|
|
932 | { |
|
|
933 | off_t sbytes = count; |
|
|
934 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
935 | |
|
|
936 | if (res == 0 && errno == EAGAIN && sbytes) |
924 | res = sbytes; |
937 | res = sbytes; |
925 | } |
938 | } |
926 | |
939 | |
927 | # elif __hpux |
940 | # elif __hpux |
928 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
941 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
… | |
… | |
942 | if (res < 0 && sbytes) |
955 | if (res < 0 && sbytes) |
943 | res = sbytes; |
956 | res = sbytes; |
944 | } |
957 | } |
945 | |
958 | |
946 | # endif |
959 | # endif |
|
|
960 | |
|
|
961 | #elif defined (_WIN32) |
|
|
962 | |
|
|
963 | /* does not work, just for documentation of what would need to be done */ |
|
|
964 | { |
|
|
965 | HANDLE h = TO_SOCKET (ifd); |
|
|
966 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
967 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
968 | } |
|
|
969 | |
947 | #else |
970 | #else |
948 | res = -1; |
971 | res = -1; |
949 | errno = ENOSYS; |
972 | errno = ENOSYS; |
950 | #endif |
973 | #endif |
951 | |
974 | |
952 | if (res < 0 |
975 | if (res < 0 |
953 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
976 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
977 | /* BSDs */ |
|
|
978 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
979 | || errno == ENOTSUP |
|
|
980 | #endif |
|
|
981 | || errno == EOPNOTSUPP /* BSDs */ |
954 | #if __solaris |
982 | #if __solaris |
955 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
983 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
956 | #endif |
984 | #endif |
957 | ) |
985 | ) |
958 | ) |
986 | ) |
… | |
… | |
989 | } |
1017 | } |
990 | |
1018 | |
991 | return res; |
1019 | return res; |
992 | } |
1020 | } |
993 | |
1021 | |
994 | static int |
1022 | static signed char |
995 | eio_dent_cmp (const void *a_, const void *b_) |
1023 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
996 | { |
1024 | { |
997 | const eio_dirent *a = (const eio_dirent *)a_; |
1025 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
998 | const eio_dirent *b = (const eio_dirent *)b_; |
|
|
999 | |
|
|
1000 | return (int)b->score - (int)a->score ? (int)b->score - (int)a->score |
|
|
1001 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ |
1026 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
|
|
1027 | } |
|
|
1028 | |
|
|
1029 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1030 | |
|
|
1031 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1032 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
|
|
1033 | |
|
|
1034 | static void |
|
|
1035 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1036 | { |
|
|
1037 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1038 | unsigned char *bit = bits; |
|
|
1039 | |
|
|
1040 | assert (CHAR_BIT == 8); |
|
|
1041 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1042 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
|
|
1043 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
|
|
1044 | |
|
|
1045 | if (size <= EIO_SORT_FAST) |
|
|
1046 | return; |
|
|
1047 | |
|
|
1048 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1049 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1050 | /* inode_bits must contain all inodes ORed together */ |
|
|
1051 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1052 | { |
|
|
1053 | ino_t endianness; |
|
|
1054 | int i, j; |
|
|
1055 | |
|
|
1056 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1057 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1058 | ((unsigned char *)&endianness)[i] = i; |
|
|
1059 | |
|
|
1060 | *bit++ = 0; |
|
|
1061 | |
|
|
1062 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1063 | { |
|
|
1064 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1065 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1066 | endianness >>= 8; |
|
|
1067 | |
|
|
1068 | for (j = 0; j < 8; ++j) |
|
|
1069 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1070 | *bit++ = offs + j; |
|
|
1071 | } |
|
|
1072 | |
|
|
1073 | for (j = 0; j < 8; ++j) |
|
|
1074 | if (score_bits & (1 << j)) |
|
|
1075 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1076 | } |
|
|
1077 | |
|
|
1078 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1079 | { |
|
|
1080 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1081 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1082 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1083 | int stk_idx = 0; |
|
|
1084 | |
|
|
1085 | base_stk [stk_idx] = dents; |
|
|
1086 | end_stk [stk_idx] = dents + size; |
|
|
1087 | bit_stk [stk_idx] = bit - 1; |
|
|
1088 | |
|
|
1089 | do |
|
|
1090 | { |
|
|
1091 | base = base_stk [stk_idx]; |
|
|
1092 | end = end_stk [stk_idx]; |
|
|
1093 | bit = bit_stk [stk_idx]; |
|
|
1094 | |
|
|
1095 | for (;;) |
|
|
1096 | { |
|
|
1097 | unsigned char O = *bit >> 3; |
|
|
1098 | unsigned char M = 1 << (*bit & 7); |
|
|
1099 | |
|
|
1100 | eio_dirent *a = base; |
|
|
1101 | eio_dirent *b = end; |
|
|
1102 | |
|
|
1103 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1104 | break; |
|
|
1105 | |
|
|
1106 | /* now bit-partition the array on the bit */ |
|
|
1107 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1108 | /* partition algos found in the literature */ |
|
|
1109 | do |
|
|
1110 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1111 | ++a; |
|
|
1112 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1113 | { |
|
|
1114 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1115 | ++a; |
|
|
1116 | } |
|
|
1117 | while (b > a); |
|
|
1118 | |
|
|
1119 | /* next bit, or stop, if no bits left in this path */ |
|
|
1120 | if (!*--bit) |
|
|
1121 | break; |
|
|
1122 | |
|
|
1123 | base_stk [stk_idx] = a; |
|
|
1124 | end_stk [stk_idx] = end; |
|
|
1125 | bit_stk [stk_idx] = bit; |
|
|
1126 | ++stk_idx; |
|
|
1127 | |
|
|
1128 | end = a; |
|
|
1129 | } |
|
|
1130 | } |
|
|
1131 | while (stk_idx--); |
|
|
1132 | } |
|
|
1133 | } |
|
|
1134 | |
|
|
1135 | static void |
|
|
1136 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1137 | { |
|
|
1138 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1139 | { |
|
|
1140 | int i; |
|
|
1141 | eio_dirent *min = dents; |
|
|
1142 | |
|
|
1143 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1144 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1145 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1146 | min = &dents [i]; |
|
|
1147 | |
|
|
1148 | /* swap elements 0 and j (minimum) */ |
|
|
1149 | { |
|
|
1150 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1151 | } |
|
|
1152 | } |
|
|
1153 | |
|
|
1154 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1155 | { |
|
|
1156 | eio_dirent *i, *j; |
|
|
1157 | |
|
|
1158 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1159 | { |
|
|
1160 | eio_dirent value = *i; |
|
|
1161 | |
|
|
1162 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1163 | j [1] = j [0]; |
|
|
1164 | |
|
|
1165 | j [1] = value; |
|
|
1166 | } |
|
|
1167 | } |
|
|
1168 | } |
|
|
1169 | |
|
|
1170 | static void |
|
|
1171 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1172 | { |
|
|
1173 | if (size <= 1) |
|
|
1174 | return; /* our insertion sort relies on size > 0 */ |
|
|
1175 | |
|
|
1176 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
|
|
1177 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1178 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
|
|
1179 | |
|
|
1180 | /* use an insertion sort at the end, or for small arrays, */ |
|
|
1181 | /* as insertion sort is more efficient for small partitions */ |
|
|
1182 | eio_dent_insertion_sort (dents, size); |
1002 | } |
1183 | } |
1003 | |
1184 | |
1004 | /* read a full directory */ |
1185 | /* read a full directory */ |
1005 | static void |
1186 | static void |
1006 | eio__scandir (eio_req *req, etp_worker *self) |
1187 | eio__scandir (eio_req *req, etp_worker *self) |
1007 | { |
1188 | { |
1008 | DIR *dirp; |
1189 | DIR *dirp; |
1009 | EIO_STRUCT_DIRENT *entp; |
1190 | EIO_STRUCT_DIRENT *entp; |
1010 | unsigned char *name, *names; |
1191 | char *name, *names; |
1011 | int namesalloc = 4096; |
1192 | int namesalloc = 4096; |
1012 | int namesoffs = 0; |
1193 | int namesoffs = 0; |
1013 | int flags = req->int1; |
1194 | int flags = req->int1; |
1014 | eio_dirent *dents = 0; |
1195 | eio_dirent *dents = 0; |
1015 | int dentalloc = 128; |
1196 | int dentalloc = 128; |
1016 | int dentoffs = 0; |
1197 | int dentoffs = 0; |
|
|
1198 | ino_t inode_bits = 0; |
1017 | |
1199 | |
1018 | req->result = -1; |
1200 | req->result = -1; |
1019 | |
1201 | |
1020 | if (!(flags & EIO_READDIR_DENTS)) |
1202 | if (!(flags & EIO_READDIR_DENTS)) |
1021 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1203 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1022 | |
1204 | |
1023 | X_LOCK (wrklock); |
1205 | X_LOCK (wrklock); |
1024 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1206 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1025 | self->dirp = dirp = opendir (req->ptr1); |
1207 | self->dirp = dirp = opendir (req->ptr1); |
1026 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1208 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1027 | req->ptr1 = names = malloc (namesalloc); |
|
|
1028 | req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1209 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
|
|
1210 | req->ptr2 = names = malloc (namesalloc); |
1029 | X_UNLOCK (wrklock); |
1211 | X_UNLOCK (wrklock); |
1030 | |
1212 | |
1031 | if (dirp && names && (!flags || dents)) |
1213 | if (dirp && names && (!flags || dents)) |
1032 | for (;;) |
1214 | for (;;) |
1033 | { |
1215 | { |
… | |
… | |
1041 | |
1223 | |
1042 | /* sort etc. */ |
1224 | /* sort etc. */ |
1043 | req->int1 = flags; |
1225 | req->int1 = flags; |
1044 | req->result = dentoffs; |
1226 | req->result = dentoffs; |
1045 | |
1227 | |
1046 | if (dents) |
|
|
1047 | { |
|
|
1048 | eio_dirent *ent = dents + dentoffs; |
|
|
1049 | |
|
|
1050 | while (ent > dents) |
|
|
1051 | (--ent)->name = names + (size_t)ent->name; |
|
|
1052 | } |
|
|
1053 | |
|
|
1054 | if (flags & EIO_READDIR_STAT_ORDER |
1228 | if (flags & EIO_READDIR_STAT_ORDER) |
1055 | || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) |
1229 | eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ |
1056 | { |
|
|
1057 | /* pray your qsort doesn't use quicksort */ |
|
|
1058 | qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */ |
|
|
1059 | } |
|
|
1060 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1230 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1231 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1232 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1233 | else |
1061 | { |
1234 | { |
1062 | /* in this case, all is known, and we just put dirs first and sort them */ |
1235 | /* in this case, all is known, and we just put dirs first and sort them */ |
1063 | eio_dirent *ent = dents + dentoffs; |
1236 | eio_dirent *oth = dents + dentoffs; |
1064 | eio_dirent *dir = dents; |
1237 | eio_dirent *dir = dents; |
1065 | |
1238 | |
|
|
1239 | /* now partition dirs to the front, and non-dirs to the back */ |
|
|
1240 | /* by walking from both sides and swapping if necessary */ |
|
|
1241 | /* also clear score, so it doesn't influence sorting */ |
1066 | while (ent > dir) |
1242 | while (oth > dir) |
1067 | { |
1243 | { |
1068 | if (dir->type == DT_DIR) |
1244 | if (dir->type == EIO_DT_DIR) |
1069 | ++dir; |
1245 | ++dir; |
1070 | else |
|
|
1071 | { |
|
|
1072 | --ent; |
|
|
1073 | |
|
|
1074 | if (ent->type == DT_DIR) |
1246 | else if ((--oth)->type == EIO_DT_DIR) |
1075 | { |
1247 | { |
1076 | eio_dirent tmp = *dir; |
1248 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
1077 | *dir = *ent; |
|
|
1078 | *ent = tmp; |
|
|
1079 | |
1249 | |
1080 | ++dir; |
1250 | ++dir; |
1081 | } |
1251 | } |
1082 | } |
1252 | } |
1083 | } |
|
|
1084 | |
1253 | |
1085 | /* now sort the dirs only */ |
1254 | /* now sort the dirs only */ |
1086 | qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); |
1255 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1087 | } |
1256 | } |
1088 | |
|
|
1089 | /* only provide the names array unless DENTS is specified */ |
|
|
1090 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1091 | { |
|
|
1092 | X_LOCK (wrklock); |
|
|
1093 | assert (!dents); |
|
|
1094 | req->ptr1 = 0; |
|
|
1095 | req->ptr2 = names; |
|
|
1096 | X_UNLOCK (wrklock); |
|
|
1097 | } |
|
|
1098 | |
1257 | |
1099 | break; |
1258 | break; |
1100 | } |
1259 | } |
1101 | |
1260 | |
1102 | /* now add the entry to our list(s) */ |
1261 | /* now add the entry to our list(s) */ |
… | |
… | |
1109 | |
1268 | |
1110 | while (expect_false (namesoffs + len > namesalloc)) |
1269 | while (expect_false (namesoffs + len > namesalloc)) |
1111 | { |
1270 | { |
1112 | namesalloc *= 2; |
1271 | namesalloc *= 2; |
1113 | X_LOCK (wrklock); |
1272 | X_LOCK (wrklock); |
1114 | req->ptr1 = names = realloc (names, namesalloc); |
1273 | req->ptr2 = names = realloc (names, namesalloc); |
1115 | X_UNLOCK (wrklock); |
1274 | X_UNLOCK (wrklock); |
1116 | |
1275 | |
1117 | if (!names) |
1276 | if (!names) |
1118 | break; |
1277 | break; |
1119 | } |
1278 | } |
… | |
… | |
1126 | |
1285 | |
1127 | if (expect_false (dentoffs == dentalloc)) |
1286 | if (expect_false (dentoffs == dentalloc)) |
1128 | { |
1287 | { |
1129 | dentalloc *= 2; |
1288 | dentalloc *= 2; |
1130 | X_LOCK (wrklock); |
1289 | X_LOCK (wrklock); |
1131 | req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1290 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1132 | X_UNLOCK (wrklock); |
1291 | X_UNLOCK (wrklock); |
1133 | |
1292 | |
1134 | if (!dents) |
1293 | if (!dents) |
1135 | break; |
1294 | break; |
1136 | } |
1295 | } |
1137 | |
1296 | |
1138 | ent = dents + dentoffs; |
1297 | ent = dents + dentoffs; |
1139 | |
1298 | |
1140 | ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ |
1299 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
1141 | ent->namelen = len - 1; |
1300 | ent->namelen = len - 1; |
1142 | ent->inode = D_INO (entp); |
1301 | ent->inode = D_INO (entp); |
|
|
1302 | |
|
|
1303 | inode_bits |= ent->inode; |
1143 | |
1304 | |
1144 | switch (D_TYPE (entp)) |
1305 | switch (D_TYPE (entp)) |
1145 | { |
1306 | { |
1146 | default: |
1307 | default: |
1147 | ent->type = EIO_DT_UNKNOWN; |
1308 | ent->type = EIO_DT_UNKNOWN; |
… | |
… | |
1190 | #ifdef DT_WHT |
1351 | #ifdef DT_WHT |
1191 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1352 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1192 | #endif |
1353 | #endif |
1193 | } |
1354 | } |
1194 | |
1355 | |
1195 | ent->score = 0; |
1356 | ent->score = 7; |
1196 | |
1357 | |
1197 | if (flags & EIO_READDIR_DIRS_FIRST) |
1358 | if (flags & EIO_READDIR_DIRS_FIRST) |
1198 | { |
1359 | { |
1199 | if (ent->type == EIO_DT_UNKNOWN) |
1360 | if (ent->type == EIO_DT_UNKNOWN) |
1200 | { |
1361 | { |
1201 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1362 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1202 | ent->score = 98; |
1363 | ent->score = 1; |
1203 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1364 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1204 | ent->score = len <= 2 ? len + 60 : len <= 4 ? 50 : len <= 7 ? 40 : 10; /* shorter == more likely dir, but avoid too many classes */ |
1365 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
1205 | } |
1366 | } |
1206 | else if (ent->type == EIO_DT_DIR) |
1367 | else if (ent->type == EIO_DT_DIR) |
1207 | ent->score = 100; |
1368 | ent->score = 0; |
1208 | } |
1369 | } |
1209 | } |
1370 | } |
1210 | |
1371 | |
1211 | namesoffs += len; |
1372 | namesoffs += len; |
1212 | ++dentoffs; |
1373 | ++dentoffs; |
1213 | } |
1374 | } |
|
|
1375 | |
|
|
1376 | if (EIO_CANCELLED (req)) |
|
|
1377 | { |
|
|
1378 | errno = ECANCELED; |
|
|
1379 | break; |
|
|
1380 | } |
1214 | } |
1381 | } |
1215 | else |
|
|
1216 | req->result = -1; |
|
|
1217 | } |
1382 | } |
1218 | |
1383 | |
1219 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1384 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1220 | # undef msync |
|
|
1221 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
1385 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1386 | #else |
|
|
1387 | |
|
|
1388 | int |
|
|
1389 | eio__msync (void *mem, size_t len, int flags) |
|
|
1390 | { |
|
|
1391 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1392 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1393 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1394 | { |
|
|
1395 | flags = 0 |
|
|
1396 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1397 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1398 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1399 | } |
|
|
1400 | |
|
|
1401 | return msync (mem, len, flags); |
|
|
1402 | } |
|
|
1403 | |
1222 | #endif |
1404 | #endif |
1223 | |
1405 | |
1224 | int |
1406 | int |
1225 | eio__mtouch (void *mem, size_t len, int flags) |
1407 | eio__mtouch (void *mem, size_t len, int flags) |
1226 | { |
1408 | { |
… | |
… | |
1233 | |
1415 | |
1234 | if (!page) |
1416 | if (!page) |
1235 | page = sysconf (_SC_PAGESIZE); |
1417 | page = sysconf (_SC_PAGESIZE); |
1236 | #endif |
1418 | #endif |
1237 | |
1419 | |
|
|
1420 | /* round down to start of page, although this is probably useless */ |
1238 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1421 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1239 | |
1422 | |
1240 | if (addr < end) |
1423 | if (addr < end) |
1241 | if (flags) /* modify */ |
1424 | if (flags & EIO_MT_MODIFY) /* modify */ |
1242 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1425 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1243 | else |
1426 | else |
1244 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1427 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1245 | |
1428 | |
1246 | return 0; |
1429 | return 0; |
… | |
… | |
1394 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1577 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1395 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1578 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1396 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1579 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1397 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1580 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1398 | |
1581 | |
|
|
1582 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1583 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1584 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1585 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1586 | |
1399 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1587 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1400 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1588 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1401 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1589 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1402 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1590 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1403 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1591 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1418 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1606 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1419 | |
1607 | |
1420 | case EIO_SYNC: req->result = 0; sync (); break; |
1608 | case EIO_SYNC: req->result = 0; sync (); break; |
1421 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1609 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1422 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1610 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1423 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1611 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1424 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1612 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1425 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1613 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1426 | |
1614 | |
1427 | case EIO_READDIR: eio__scandir (req, self); break; |
1615 | case EIO_READDIR: eio__scandir (req, self); break; |
1428 | |
1616 | |
… | |
… | |
1550 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1738 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1551 | { |
1739 | { |
1552 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1740 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1553 | } |
1741 | } |
1554 | |
1742 | |
|
|
1743 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
1744 | { |
|
|
1745 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
|
|
1746 | } |
|
|
1747 | |
1555 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1748 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1556 | { |
1749 | { |
1557 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1750 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1558 | } |
1751 | } |
1559 | |
1752 | |
… | |
… | |
1629 | } |
1822 | } |
1630 | |
1823 | |
1631 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1824 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1632 | { |
1825 | { |
1633 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1826 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
|
|
1827 | } |
|
|
1828 | |
|
|
1829 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
1830 | { |
|
|
1831 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
1634 | } |
1832 | } |
1635 | |
1833 | |
1636 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1834 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1637 | { |
1835 | { |
1638 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1836 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
… | |
… | |
1742 | /* misc garbage */ |
1940 | /* misc garbage */ |
1743 | |
1941 | |
1744 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1942 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1745 | { |
1943 | { |
1746 | etp_worker wrk; |
1944 | etp_worker wrk; |
|
|
1945 | ssize_t ret; |
1747 | |
1946 | |
1748 | wrk.dbuf = 0; |
1947 | wrk.dbuf = 0; |
1749 | |
1948 | |
1750 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
1949 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1751 | |
1950 | |
1752 | if (wrk.dbuf) |
1951 | if (wrk.dbuf) |
1753 | free (wrk.dbuf); |
1952 | free (wrk.dbuf); |
1754 | } |
|
|
1755 | |
1953 | |
|
|
1954 | return ret; |
|
|
1955 | } |
|
|
1956 | |