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 | * |
… | |
… | |
73 | #else |
73 | #else |
74 | |
74 | |
75 | # include "config.h" |
75 | # include "config.h" |
76 | # include <sys/time.h> |
76 | # include <sys/time.h> |
77 | # include <sys/select.h> |
77 | # include <sys/select.h> |
|
|
78 | # include <sys/mman.h> |
78 | # include <unistd.h> |
79 | # include <unistd.h> |
79 | # include <utime.h> |
80 | # include <utime.h> |
80 | # include <signal.h> |
81 | # include <signal.h> |
81 | # include <dirent.h> |
82 | # include <dirent.h> |
82 | |
83 | |
83 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
84 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
84 | # if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
85 | # if __freebsd || defined __NetBSD__ || defined __OpenBSD__ |
|
|
86 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
85 | # define D_INO(de) (de)->d_fileno |
87 | # define D_INO(de) (de)->d_fileno |
86 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
88 | # define D_NAMLEN(de) (de)->d_namlen |
87 | # elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
89 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
88 | # define D_INO(de) (de)->d_ino |
90 | # define D_INO(de) (de)->d_ino |
89 | # endif |
91 | # endif |
|
|
92 | |
|
|
93 | #ifdef _D_EXACT_NAMLEN |
|
|
94 | # undef D_NAMLEN |
|
|
95 | # define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
96 | #endif |
90 | |
97 | |
91 | # ifdef _DIRENT_HAVE_D_TYPE |
98 | # ifdef _DIRENT_HAVE_D_TYPE |
92 | # define D_TYPE(de) (de)->d_type |
99 | # define D_TYPE(de) (de)->d_type |
93 | # endif |
100 | # endif |
94 | |
101 | |
… | |
… | |
99 | #endif |
106 | #endif |
100 | |
107 | |
101 | #if HAVE_SENDFILE |
108 | #if HAVE_SENDFILE |
102 | # if __linux |
109 | # if __linux |
103 | # include <sys/sendfile.h> |
110 | # include <sys/sendfile.h> |
104 | # elif __freebsd |
111 | # elif __freebsd || defined __APPLE__ |
105 | # include <sys/socket.h> |
112 | # include <sys/socket.h> |
106 | # include <sys/uio.h> |
113 | # include <sys/uio.h> |
107 | # elif __hpux |
114 | # elif __hpux |
108 | # include <sys/socket.h> |
115 | # include <sys/socket.h> |
109 | # elif __solaris /* not yet */ |
116 | # elif __solaris |
110 | # include <sys/sendfile.h> |
117 | # include <sys/sendfile.h> |
111 | # else |
118 | # else |
112 | # error sendfile support requested but not available |
119 | # error sendfile support requested but not available |
113 | # endif |
120 | # endif |
114 | #endif |
121 | #endif |
115 | |
122 | |
116 | #ifndef D_TYPE |
123 | #ifndef D_TYPE |
117 | # define D_TYPE(de) 0 |
124 | # define D_TYPE(de) 0 |
118 | #endif |
125 | #endif |
119 | |
|
|
120 | #ifndef D_INO |
126 | #ifndef D_INO |
121 | # define D_INO(de) 0 |
127 | # define D_INO(de) 0 |
|
|
128 | #endif |
|
|
129 | #ifndef D_NAMLEN |
|
|
130 | # define D_NAMLEN(de) strlen ((de)->d_name) |
122 | #endif |
131 | #endif |
123 | |
132 | |
124 | /* number of seconds after which an idle threads exit */ |
133 | /* number of seconds after which an idle threads exit */ |
125 | #define IDLE_TIMEOUT 10 |
134 | #define IDLE_TIMEOUT 10 |
126 | |
135 | |
… | |
… | |
178 | if (wrk->dirp) \ |
187 | if (wrk->dirp) \ |
179 | { \ |
188 | { \ |
180 | closedir (wrk->dirp); \ |
189 | closedir (wrk->dirp); \ |
181 | wrk->dirp = 0; \ |
190 | wrk->dirp = 0; \ |
182 | } |
191 | } |
|
|
192 | |
183 | #define ETP_WORKER_COMMON \ |
193 | #define ETP_WORKER_COMMON \ |
184 | void *dbuf; \ |
194 | void *dbuf; \ |
185 | DIR *dirp; |
195 | DIR *dirp; |
186 | |
196 | |
187 | /*****************************************************************************/ |
197 | /*****************************************************************************/ |
188 | |
198 | |
189 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
199 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
190 | |
200 | |
191 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
201 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
192 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
202 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
193 | { |
203 | { |
194 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
204 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
195 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
205 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
196 | } |
206 | } |
… | |
… | |
588 | } |
598 | } |
589 | |
599 | |
590 | static void etp_set_max_poll_time (double nseconds) |
600 | static void etp_set_max_poll_time (double nseconds) |
591 | { |
601 | { |
592 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
602 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
593 | max_poll_time = nseconds; |
603 | max_poll_time = nseconds * EIO_TICKS; |
594 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
604 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
595 | } |
605 | } |
596 | |
606 | |
597 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
607 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
598 | { |
608 | { |
… | |
… | |
857 | if (!res || errno != ENOSYS) |
867 | if (!res || errno != ENOSYS) |
858 | return res; |
868 | return res; |
859 | #endif |
869 | #endif |
860 | |
870 | |
861 | /* even though we could play tricks with the flags, it's better to always |
871 | /* even though we could play tricks with the flags, it's better to always |
862 | * call fdatasync, as thta matches the expectation of it's users best */ |
872 | * call fdatasync, as that matches the expectation of its users best */ |
863 | return fdatasync (fd); |
873 | return fdatasync (fd); |
864 | } |
874 | } |
865 | |
875 | |
866 | #if !HAVE_READAHEAD |
876 | #if !HAVE_READAHEAD |
867 | # undef readahead |
877 | # undef readahead |
… | |
… | |
913 | if (res < 0 && sbytes) |
923 | if (res < 0 && sbytes) |
914 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
924 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
915 | res = sbytes; |
925 | res = sbytes; |
916 | } |
926 | } |
917 | |
927 | |
|
|
928 | # elif defined (__APPLE__) |
|
|
929 | |
|
|
930 | { |
|
|
931 | off_t sbytes = count; |
|
|
932 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
933 | |
|
|
934 | if (res < 0 && errno == EAGAIN && sbytes) |
|
|
935 | res = sbytes; |
|
|
936 | } |
|
|
937 | |
918 | # elif __hpux |
938 | # elif __hpux |
919 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
939 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
920 | |
940 | |
921 | # elif __solaris |
941 | # elif __solaris |
922 | { |
942 | { |
… | |
… | |
933 | if (res < 0 && sbytes) |
953 | if (res < 0 && sbytes) |
934 | res = sbytes; |
954 | res = sbytes; |
935 | } |
955 | } |
936 | |
956 | |
937 | # endif |
957 | # endif |
|
|
958 | |
|
|
959 | #elif defined (_WIN32) |
|
|
960 | |
|
|
961 | /* does not work, just for documentation of what would need to be done */ |
|
|
962 | { |
|
|
963 | HANDLE h = TO_SOCKET (ifd); |
|
|
964 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
965 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
966 | } |
|
|
967 | |
938 | #else |
968 | #else |
939 | res = -1; |
969 | res = -1; |
940 | errno = ENOSYS; |
970 | errno = ENOSYS; |
941 | #endif |
971 | #endif |
942 | |
972 | |
943 | if (res < 0 |
973 | if (res < 0 |
944 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
974 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
975 | /* BSDs */ |
|
|
976 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
977 | || errno == ENOTSUP |
|
|
978 | #endif |
|
|
979 | || errno == EOPNOTSUPP /* BSDs */ |
945 | #if __solaris |
980 | #if __solaris |
946 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
981 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
947 | #endif |
982 | #endif |
948 | ) |
983 | ) |
949 | ) |
984 | ) |
… | |
… | |
980 | } |
1015 | } |
981 | |
1016 | |
982 | return res; |
1017 | return res; |
983 | } |
1018 | } |
984 | |
1019 | |
985 | static int |
1020 | static signed char |
986 | eio_dent_cmp (const void *a_, const void *b_) |
1021 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
987 | { |
1022 | { |
988 | const eio_dirent *a = (const eio_dirent *)a_; |
1023 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
989 | const eio_dirent *b = (const eio_dirent *)b_; |
|
|
990 | |
|
|
991 | return (int)b->score - (int)a->score ? (int)b->score - (int)a->score |
|
|
992 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ |
1024 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
|
|
1025 | } |
|
|
1026 | |
|
|
1027 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1028 | |
|
|
1029 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1030 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
|
|
1031 | |
|
|
1032 | static void |
|
|
1033 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1034 | { |
|
|
1035 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1036 | unsigned char *bit = bits; |
|
|
1037 | |
|
|
1038 | assert (CHAR_BIT == 8); |
|
|
1039 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1040 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
|
|
1041 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
|
|
1042 | |
|
|
1043 | if (size <= EIO_SORT_FAST) |
|
|
1044 | return; |
|
|
1045 | |
|
|
1046 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1047 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1048 | /* inode_bits must contain all inodes ORed together */ |
|
|
1049 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1050 | { |
|
|
1051 | ino_t endianness; |
|
|
1052 | int i, j; |
|
|
1053 | |
|
|
1054 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1055 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1056 | ((unsigned char *)&endianness)[i] = i; |
|
|
1057 | |
|
|
1058 | *bit++ = 0; |
|
|
1059 | |
|
|
1060 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1061 | { |
|
|
1062 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1063 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1064 | endianness >>= 8; |
|
|
1065 | |
|
|
1066 | for (j = 0; j < 8; ++j) |
|
|
1067 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1068 | *bit++ = offs + j; |
|
|
1069 | } |
|
|
1070 | |
|
|
1071 | for (j = 0; j < 8; ++j) |
|
|
1072 | if (score_bits & (1 << j)) |
|
|
1073 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1074 | } |
|
|
1075 | |
|
|
1076 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1077 | { |
|
|
1078 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1079 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1080 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1081 | int stk_idx = 0; |
|
|
1082 | |
|
|
1083 | base_stk [stk_idx] = dents; |
|
|
1084 | end_stk [stk_idx] = dents + size; |
|
|
1085 | bit_stk [stk_idx] = bit - 1; |
|
|
1086 | |
|
|
1087 | do |
|
|
1088 | { |
|
|
1089 | base = base_stk [stk_idx]; |
|
|
1090 | end = end_stk [stk_idx]; |
|
|
1091 | bit = bit_stk [stk_idx]; |
|
|
1092 | |
|
|
1093 | for (;;) |
|
|
1094 | { |
|
|
1095 | unsigned char O = *bit >> 3; |
|
|
1096 | unsigned char M = 1 << (*bit & 7); |
|
|
1097 | |
|
|
1098 | eio_dirent *a = base; |
|
|
1099 | eio_dirent *b = end; |
|
|
1100 | |
|
|
1101 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1102 | break; |
|
|
1103 | |
|
|
1104 | /* now bit-partition the array on the bit */ |
|
|
1105 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1106 | /* partition algos found in the literature */ |
|
|
1107 | do |
|
|
1108 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1109 | ++a; |
|
|
1110 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1111 | { |
|
|
1112 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1113 | ++a; |
|
|
1114 | } |
|
|
1115 | while (b > a); |
|
|
1116 | |
|
|
1117 | /* next bit, or stop, if no bits left in this path */ |
|
|
1118 | if (!*--bit) |
|
|
1119 | break; |
|
|
1120 | |
|
|
1121 | base_stk [stk_idx] = a; |
|
|
1122 | end_stk [stk_idx] = end; |
|
|
1123 | bit_stk [stk_idx] = bit; |
|
|
1124 | ++stk_idx; |
|
|
1125 | |
|
|
1126 | end = a; |
|
|
1127 | } |
|
|
1128 | } |
|
|
1129 | while (stk_idx--); |
|
|
1130 | } |
|
|
1131 | } |
|
|
1132 | |
|
|
1133 | static void |
|
|
1134 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1135 | { |
|
|
1136 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1137 | { |
|
|
1138 | int i; |
|
|
1139 | eio_dirent *min = dents; |
|
|
1140 | |
|
|
1141 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1142 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1143 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1144 | min = &dents [i]; |
|
|
1145 | |
|
|
1146 | /* swap elements 0 and j (minimum) */ |
|
|
1147 | { |
|
|
1148 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1149 | } |
|
|
1150 | } |
|
|
1151 | |
|
|
1152 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1153 | { |
|
|
1154 | eio_dirent *i, *j; |
|
|
1155 | |
|
|
1156 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1157 | { |
|
|
1158 | eio_dirent value = *i; |
|
|
1159 | |
|
|
1160 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1161 | j [1] = j [0]; |
|
|
1162 | |
|
|
1163 | j [1] = value; |
|
|
1164 | } |
|
|
1165 | } |
|
|
1166 | } |
|
|
1167 | |
|
|
1168 | static void |
|
|
1169 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1170 | { |
|
|
1171 | if (size <= 1) |
|
|
1172 | return; /* our insertion sort relies on size > 0 */ |
|
|
1173 | |
|
|
1174 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
|
|
1175 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1176 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
|
|
1177 | |
|
|
1178 | /* use an insertion sort at the end, or for small arrays, */ |
|
|
1179 | /* as insertion sort is more efficient for small partitions */ |
|
|
1180 | eio_dent_insertion_sort (dents, size); |
993 | } |
1181 | } |
994 | |
1182 | |
995 | /* read a full directory */ |
1183 | /* read a full directory */ |
996 | static void |
1184 | static void |
997 | eio__scandir (eio_req *req, etp_worker *self) |
1185 | eio__scandir (eio_req *req, etp_worker *self) |
998 | { |
1186 | { |
999 | DIR *dirp; |
1187 | DIR *dirp; |
1000 | EIO_STRUCT_DIRENT *entp; |
1188 | EIO_STRUCT_DIRENT *entp; |
1001 | unsigned char *name, *names; |
1189 | char *name, *names; |
1002 | int namesalloc = 4096; |
1190 | int namesalloc = 4096; |
1003 | int namesoffs = 0; |
1191 | int namesoffs = 0; |
1004 | int flags = req->int1; |
1192 | int flags = req->int1; |
1005 | eio_dirent *dents = 0; |
1193 | eio_dirent *dents = 0; |
1006 | int dentalloc = 128; |
1194 | int dentalloc = 128; |
1007 | int dentoffs = 0; |
1195 | int dentoffs = 0; |
|
|
1196 | ino_t inode_bits = 0; |
1008 | |
1197 | |
1009 | req->result = -1; |
1198 | req->result = -1; |
1010 | |
1199 | |
1011 | if (!(flags & EIO_READDIR_DENTS)) |
1200 | if (!(flags & EIO_READDIR_DENTS)) |
1012 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1201 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1013 | |
1202 | |
1014 | X_LOCK (wrklock); |
1203 | X_LOCK (wrklock); |
1015 | |
|
|
1016 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1204 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1017 | self->dirp = dirp = opendir (req->ptr1); |
1205 | self->dirp = dirp = opendir (req->ptr1); |
1018 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1206 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1019 | req->ptr1 = names = malloc (namesalloc); |
|
|
1020 | req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1207 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1021 | |
1208 | req->ptr2 = names = malloc (namesalloc); |
1022 | X_UNLOCK (wrklock); |
1209 | X_UNLOCK (wrklock); |
1023 | |
1210 | |
1024 | if (dirp && names && (!flags || dents)) |
1211 | if (dirp && names && (!flags || dents)) |
1025 | for (;;) |
1212 | for (;;) |
1026 | { |
1213 | { |
… | |
… | |
1034 | |
1221 | |
1035 | /* sort etc. */ |
1222 | /* sort etc. */ |
1036 | req->int1 = flags; |
1223 | req->int1 = flags; |
1037 | req->result = dentoffs; |
1224 | req->result = dentoffs; |
1038 | |
1225 | |
1039 | if (flags & EIO_READDIR_STAT_ORDER || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)) |
1226 | if (flags & EIO_READDIR_STAT_ORDER) |
|
|
1227 | eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ |
|
|
1228 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1229 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1230 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1231 | else |
1040 | { |
1232 | { |
1041 | /* pray your qsort doesn't use quicksort */ |
|
|
1042 | qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */ |
|
|
1043 | } |
|
|
1044 | else if (flags & EIO_READDIR_DIRS_FIRST && !(flags & EIO_READDIR_FOUND_UNKNOWN)) |
|
|
1045 | { |
|
|
1046 | /* in this case, all is known, and we just put dirs first and sort them */ |
1233 | /* in this case, all is known, and we just put dirs first and sort them */ |
1047 | eio_dirent *ent = dents + dentoffs; |
1234 | eio_dirent *oth = dents + dentoffs; |
1048 | eio_dirent *dir = dents; |
1235 | eio_dirent *dir = dents; |
1049 | |
1236 | |
|
|
1237 | /* now partition dirs to the front, and non-dirs to the back */ |
|
|
1238 | /* by walking from both sides and swapping if necessary */ |
|
|
1239 | /* also clear score, so it doesn't influence sorting */ |
1050 | while (ent > dir) |
1240 | while (oth > dir) |
1051 | { |
1241 | { |
1052 | if (dir->type == DT_DIR) |
1242 | if (dir->type == EIO_DT_DIR) |
1053 | ++dir; |
1243 | ++dir; |
1054 | else |
|
|
1055 | { |
|
|
1056 | --ent; |
|
|
1057 | |
|
|
1058 | if (ent->type == DT_DIR) |
1244 | else if ((--oth)->type == EIO_DT_DIR) |
1059 | { |
1245 | { |
1060 | eio_dirent tmp = *dir; |
1246 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
1061 | *dir = *ent; |
|
|
1062 | *ent = tmp; |
|
|
1063 | |
1247 | |
1064 | ++dir; |
1248 | ++dir; |
1065 | } |
1249 | } |
1066 | } |
1250 | } |
1067 | } |
|
|
1068 | |
1251 | |
1069 | /* now sort the dirs only */ |
1252 | /* now sort the dirs only */ |
1070 | qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); |
1253 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1071 | } |
1254 | } |
1072 | |
|
|
1073 | {int i; for(i=0;i<dentoffs;++i){eio_dirent *e=dents+i; printf ("%9ld %3d %s\n", e->inode,e->score,e->name);}}//D |
|
|
1074 | |
1255 | |
1075 | break; |
1256 | break; |
1076 | } |
1257 | } |
1077 | |
1258 | |
|
|
1259 | /* now add the entry to our list(s) */ |
1078 | name = entp->d_name; |
1260 | name = entp->d_name; |
1079 | |
1261 | |
|
|
1262 | /* skip . and .. entries */ |
1080 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1263 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1081 | { |
1264 | { |
1082 | int len = strlen (name) + 1; |
1265 | int len = D_NAMLEN (entp) + 1; |
1083 | |
1266 | |
1084 | while (expect_false (namesoffs + len > namesalloc)) |
1267 | while (expect_false (namesoffs + len > namesalloc)) |
1085 | { |
1268 | { |
1086 | namesalloc *= 2; |
1269 | namesalloc *= 2; |
1087 | X_LOCK (wrklock); |
1270 | X_LOCK (wrklock); |
1088 | req->ptr1 = names = realloc (names, namesalloc); |
1271 | req->ptr2 = names = realloc (names, namesalloc); |
1089 | X_UNLOCK (wrklock); |
1272 | X_UNLOCK (wrklock); |
1090 | |
1273 | |
1091 | if (!names) |
1274 | if (!names) |
1092 | break; |
1275 | break; |
1093 | } |
1276 | } |
… | |
… | |
1100 | |
1283 | |
1101 | if (expect_false (dentoffs == dentalloc)) |
1284 | if (expect_false (dentoffs == dentalloc)) |
1102 | { |
1285 | { |
1103 | dentalloc *= 2; |
1286 | dentalloc *= 2; |
1104 | X_LOCK (wrklock); |
1287 | X_LOCK (wrklock); |
1105 | req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1288 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1106 | X_UNLOCK (wrklock); |
1289 | X_UNLOCK (wrklock); |
1107 | |
1290 | |
1108 | if (!dents) |
1291 | if (!dents) |
1109 | break; |
1292 | break; |
1110 | } |
1293 | } |
1111 | |
1294 | |
1112 | ent = dents + dentoffs; |
1295 | ent = dents + dentoffs; |
1113 | |
1296 | |
1114 | ent->name = names + namesoffs; |
1297 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
1115 | ent->namelen = len - 1; |
1298 | ent->namelen = len - 1; |
1116 | ent->inode = D_INO (entp); |
1299 | ent->inode = D_INO (entp); |
|
|
1300 | |
|
|
1301 | inode_bits |= ent->inode; |
1117 | |
1302 | |
1118 | switch (D_TYPE (entp)) |
1303 | switch (D_TYPE (entp)) |
1119 | { |
1304 | { |
1120 | default: |
1305 | default: |
1121 | ent->type = EIO_DT_UNKNOWN; |
1306 | ent->type = EIO_DT_UNKNOWN; |
… | |
… | |
1126 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
1311 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
1127 | #endif |
1312 | #endif |
1128 | #ifdef DT_CHR |
1313 | #ifdef DT_CHR |
1129 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
1314 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
1130 | #endif |
1315 | #endif |
|
|
1316 | #ifdef DT_MPC |
|
|
1317 | case DT_MPC: ent->type = EIO_DT_MPC; break; |
|
|
1318 | #endif |
1131 | #ifdef DT_DIR |
1319 | #ifdef DT_DIR |
1132 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
1320 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
1133 | #endif |
1321 | #endif |
|
|
1322 | #ifdef DT_NAM |
|
|
1323 | case DT_NAM: ent->type = EIO_DT_NAM; break; |
|
|
1324 | #endif |
1134 | #ifdef DT_BLK |
1325 | #ifdef DT_BLK |
1135 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
1326 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
1136 | #endif |
1327 | #endif |
|
|
1328 | #ifdef DT_MPB |
|
|
1329 | case DT_MPB: ent->type = EIO_DT_MPB; break; |
|
|
1330 | #endif |
1137 | #ifdef DT_REG |
1331 | #ifdef DT_REG |
1138 | case DT_REG: ent->type = EIO_DT_REG; break; |
1332 | case DT_REG: ent->type = EIO_DT_REG; break; |
1139 | #endif |
1333 | #endif |
|
|
1334 | #ifdef DT_NWK |
|
|
1335 | case DT_NWK: ent->type = EIO_DT_NWK; break; |
|
|
1336 | #endif |
|
|
1337 | #ifdef DT_CMP |
|
|
1338 | case DT_CMP: ent->type = EIO_DT_CMP; break; |
|
|
1339 | #endif |
1140 | #ifdef DT_LNK |
1340 | #ifdef DT_LNK |
1141 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
1341 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
1142 | #endif |
1342 | #endif |
1143 | #ifdef DT_SOCK |
1343 | #ifdef DT_SOCK |
1144 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
1344 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
1145 | #endif |
1345 | #endif |
|
|
1346 | #ifdef DT_DOOR |
|
|
1347 | case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
|
|
1348 | #endif |
1146 | #ifdef DT_WHT |
1349 | #ifdef DT_WHT |
1147 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1350 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1148 | #endif |
1351 | #endif |
1149 | } |
1352 | } |
1150 | |
1353 | |
1151 | ent->score = 0; |
1354 | ent->score = 7; |
1152 | |
1355 | |
1153 | if (flags & EIO_READDIR_DIRS_FIRST) |
1356 | if (flags & EIO_READDIR_DIRS_FIRST) |
1154 | { |
1357 | { |
1155 | if (ent->type == EIO_DT_UNKNOWN) |
1358 | if (ent->type == EIO_DT_UNKNOWN) |
1156 | { |
1359 | { |
1157 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1360 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1158 | ent->score = 98; |
1361 | ent->score = 1; |
1159 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1362 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1160 | ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ |
1363 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
1161 | } |
1364 | } |
1162 | else if (ent->type == DT_DIR) |
1365 | else if (ent->type == EIO_DT_DIR) |
1163 | ent->score = 100; |
1366 | ent->score = 0; |
1164 | } |
1367 | } |
1165 | } |
1368 | } |
1166 | |
1369 | |
1167 | namesoffs += len; |
1370 | namesoffs += len; |
1168 | ++dentoffs; |
1371 | ++dentoffs; |
1169 | } |
1372 | } |
|
|
1373 | |
|
|
1374 | if (EIO_CANCELLED (req)) |
|
|
1375 | { |
|
|
1376 | errno = ECANCELED; |
|
|
1377 | break; |
|
|
1378 | } |
1170 | } |
1379 | } |
1171 | else |
1380 | } |
1172 | req->result = -1; |
|
|
1173 | |
1381 | |
1174 | /* if user doesn't want the dents, do not provide it */ |
1382 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1175 | if (!(flags & EIO_READDIR_DENTS)) |
1383 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1384 | #else |
|
|
1385 | |
|
|
1386 | int |
|
|
1387 | eio__msync (void *mem, size_t len, int flags) |
|
|
1388 | { |
|
|
1389 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1390 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1391 | || EIO_MS_SYNC != MS_SYNC) |
1176 | { |
1392 | { |
1177 | X_LOCK (wrklock); |
1393 | flags = 0 |
1178 | free (dents); |
1394 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
1179 | req->ptr2 = req->ptr1; |
1395 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
1180 | req->ptr1 = 0; |
1396 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
1181 | X_UNLOCK (wrklock); |
|
|
1182 | } |
1397 | } |
1183 | } |
|
|
1184 | |
1398 | |
1185 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1399 | return msync (mem, len, flags); |
1186 | # undef msync |
1400 | } |
1187 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
1401 | |
1188 | #endif |
1402 | #endif |
1189 | |
1403 | |
1190 | int |
1404 | int |
1191 | eio__mtouch (void *mem, size_t len, int flags) |
1405 | eio__mtouch (void *mem, size_t len, int flags) |
1192 | { |
1406 | { |
… | |
… | |
1199 | |
1413 | |
1200 | if (!page) |
1414 | if (!page) |
1201 | page = sysconf (_SC_PAGESIZE); |
1415 | page = sysconf (_SC_PAGESIZE); |
1202 | #endif |
1416 | #endif |
1203 | |
1417 | |
|
|
1418 | /* round down to start of page, although this is probably useless */ |
1204 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1419 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1205 | |
1420 | |
1206 | if (addr < end) |
1421 | if (addr < end) |
1207 | if (flags) /* modify */ |
1422 | if (flags & EIO_MT_MODIFY) /* modify */ |
1208 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1423 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1209 | else |
1424 | else |
1210 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1425 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1211 | |
1426 | |
1212 | return 0; |
1427 | return 0; |
… | |
… | |
1384 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1599 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1385 | |
1600 | |
1386 | case EIO_SYNC: req->result = 0; sync (); break; |
1601 | case EIO_SYNC: req->result = 0; sync (); break; |
1387 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1602 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1388 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1603 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1389 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1604 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1390 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1605 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1391 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1606 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1392 | |
1607 | |
1393 | case EIO_READDIR: eio__scandir (req, self); break; |
1608 | case EIO_READDIR: eio__scandir (req, self); break; |
1394 | |
1609 | |
… | |
… | |
1708 | /* misc garbage */ |
1923 | /* misc garbage */ |
1709 | |
1924 | |
1710 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1925 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1711 | { |
1926 | { |
1712 | etp_worker wrk; |
1927 | etp_worker wrk; |
|
|
1928 | ssize_t ret; |
1713 | |
1929 | |
1714 | wrk.dbuf = 0; |
1930 | wrk.dbuf = 0; |
1715 | |
1931 | |
1716 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
1932 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1717 | |
1933 | |
1718 | if (wrk.dbuf) |
1934 | if (wrk.dbuf) |
1719 | free (wrk.dbuf); |
1935 | free (wrk.dbuf); |
1720 | } |
|
|
1721 | |
1936 | |
|
|
1937 | return ret; |
|
|
1938 | } |
|
|
1939 | |