ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/eio.c
(Generate patch)

Comparing libeio/eio.c (file contents):
Revision 1.31 by root, Sat Jun 6 17:25:13 2009 UTC vs.
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC

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__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
85# define D_INO(de) (de)->d_fileno 88# define D_INO(de) (de)->d_fileno
86# define _DIRENT_HAVE_D_TYPE /* sigh */ 89# define D_NAMLEN(de) (de)->d_namlen
87# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
88# define D_INO(de) (de)->d_ino 91# define D_INO(de) (de)->d_ino
89# endif 92# endif
93
94#ifdef _D_EXACT_NAMLEN
95# undef D_NAMLEN
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
90 98
91# ifdef _DIRENT_HAVE_D_TYPE 99# ifdef _DIRENT_HAVE_D_TYPE
92# define D_TYPE(de) (de)->d_type 100# define D_TYPE(de) (de)->d_type
93# endif 101# endif
94 102
… …
99#endif 107#endif
100 108
101#if HAVE_SENDFILE 109#if HAVE_SENDFILE
102# if __linux 110# if __linux
103# include <sys/sendfile.h> 111# include <sys/sendfile.h>
104# elif __freebsd 112# elif __FreeBSD__ || defined __APPLE__
105# include <sys/socket.h> 113# include <sys/socket.h>
106# include <sys/uio.h> 114# include <sys/uio.h>
107# elif __hpux 115# elif __hpux
108# include <sys/socket.h> 116# include <sys/socket.h>
109# elif __solaris /* not yet */ 117# elif __solaris
110# include <sys/sendfile.h> 118# include <sys/sendfile.h>
111# else 119# else
112# error sendfile support requested but not available 120# error sendfile support requested but not available
113# endif 121# endif
114#endif 122#endif
115 123
116#ifndef D_TYPE 124#ifndef D_TYPE
117# define D_TYPE(de) 0 125# define D_TYPE(de) 0
118#endif 126#endif
119
120#ifndef D_INO 127#ifndef D_INO
121# define D_INO(de) 0 128# define D_INO(de) 0
129#endif
130#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name)
122#endif 132#endif
123 133
124/* number of seconds after which an idle threads exit */ 134/* number of seconds after which an idle threads exit */
125#define IDLE_TIMEOUT 10 135#define IDLE_TIMEOUT 10
126 136
… …
178 if (wrk->dirp) \ 188 if (wrk->dirp) \
179 { \ 189 { \
180 closedir (wrk->dirp); \ 190 closedir (wrk->dirp); \
181 wrk->dirp = 0; \ 191 wrk->dirp = 0; \
182 } 192 }
193
183#define ETP_WORKER_COMMON \ 194#define ETP_WORKER_COMMON \
184 void *dbuf; \ 195 void *dbuf; \
185 DIR *dirp; 196 DIR *dirp;
186 197
187/*****************************************************************************/ 198/*****************************************************************************/
188 199
189#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
190 201
191/* calculcate time difference in ~1/EIO_TICKS of a second */ 202/* calculate time difference in ~1/EIO_TICKS of a second */
192static int tvdiff (struct timeval *tv1, struct timeval *tv2) 203static int tvdiff (struct timeval *tv1, struct timeval *tv2)
193{ 204{
194 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
195 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 206 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
196} 207}
… …
588} 599}
589 600
590static void etp_set_max_poll_time (double nseconds) 601static void etp_set_max_poll_time (double nseconds)
591{ 602{
592 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 603 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
593 max_poll_time = nseconds; 604 max_poll_time = nseconds * EIO_TICKS;
594 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
595} 606}
596 607
597static void etp_set_max_poll_reqs (unsigned int maxreqs) 608static void etp_set_max_poll_reqs (unsigned int maxreqs)
598{ 609{
… …
857 if (!res || errno != ENOSYS) 868 if (!res || errno != ENOSYS)
858 return res; 869 return res;
859#endif 870#endif
860 871
861 /* 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
862 * call fdatasync, as thta matches the expectation of it's users best */ 873 * call fdatasync, as that matches the expectation of its users best */
863 return fdatasync (fd); 874 return fdatasync (fd);
864} 875}
865 876
866#if !HAVE_READAHEAD 877#if !HAVE_READAHEAD
867# undef readahead 878# undef readahead
… …
899 910
900#if HAVE_SENDFILE 911#if HAVE_SENDFILE
901# if __linux 912# if __linux
902 res = sendfile (ofd, ifd, &offset, count); 913 res = sendfile (ofd, ifd, &offset, count);
903 914
904# elif __freebsd 915# elif __FreeBSD__
905 /* 916 /*
906 * 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
907 * wasted on making it similar to other I/O functions. 918 * wasted on making it similar to other I/O functions.
908 */ 919 */
909 { 920 {
910 off_t sbytes; 921 off_t sbytes;
911 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
912 923
913 if (res < 0 && sbytes) 924 #if 0 /* according to the manpage, this is correct, but broken behaviour */
914 /* 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)
915 res = sbytes; 945 res = sbytes;
916 } 946 }
917 947
918# elif __hpux 948# elif __hpux
919 res = sendfile (ofd, ifd, offset, count, 0, 0); 949 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
933 if (res < 0 && sbytes) 963 if (res < 0 && sbytes)
934 res = sbytes; 964 res = sbytes;
935 } 965 }
936 966
937# 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
938#else 978#else
939 res = -1; 979 res = -1;
940 errno = ENOSYS; 980 errno = ENOSYS;
941#endif 981#endif
942 982
943 if (res < 0 983 if (res < 0
944 && (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 */
945#if __solaris 990#if __solaris
946 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 991 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
947#endif 992#endif
948 ) 993 )
949 ) 994 )
… …
980 } 1025 }
981 1026
982 return res; 1027 return res;
983} 1028}
984 1029
985static int 1030static signed char
986eio_dent_cmp (const void *a_, const void *b_) 1031eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
987{ 1032{
988 const eio_dirent *a = (const eio_dirent *)a_; 1033 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 */ 1034 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1035}
1036
1037#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1038
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 */
1041
1042static void
1043eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1044{
1045 unsigned char bits [9 + sizeof (ino_t) * 8];
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;
1062 int i, j;
1063
1064 /* we store the byte offset of byte n into byte n of "endianness" */
1065 for (i = 0; i < sizeof (ino_t); ++i)
1066 ((unsigned char *)&endianness)[i] = i;
1067
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
1143static void
1144eio_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) */
1157 {
1158 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1159 }
1160 }
1161
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)
1167 {
1168 eio_dirent value = *i;
1169
1170 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1171 j [1] = j [0];
1172
1173 j [1] = value;
1174 }
1175 }
1176}
1177
1178static void
1179eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1180{
1181 if (size <= 1)
1182 return; /* our insertion sort relies on size > 0 */
1183
1184 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1185 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1186 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1187
1188 /* use an insertion sort at the end, or for small arrays, */
1189 /* as insertion sort is more efficient for small partitions */
1190 eio_dent_insertion_sort (dents, size);
993} 1191}
994 1192
995/* read a full directory */ 1193/* read a full directory */
996static void 1194static void
997eio__scandir (eio_req *req, etp_worker *self) 1195eio__scandir (eio_req *req, etp_worker *self)
998{ 1196{
999 DIR *dirp; 1197 DIR *dirp;
1000 EIO_STRUCT_DIRENT *entp; 1198 EIO_STRUCT_DIRENT *entp;
1001 unsigned char *name, *names; 1199 char *name, *names;
1002 int namesalloc = 4096; 1200 int namesalloc = 4096;
1003 int namesoffs = 0; 1201 int namesoffs = 0;
1004 int flags = req->int1; 1202 int flags = req->int1;
1005 eio_dirent *dents = 0; 1203 eio_dirent *dents = 0;
1006 int dentalloc = 128; 1204 int dentalloc = 128;
1007 int dentoffs = 0; 1205 int dentoffs = 0;
1206 ino_t inode_bits = 0;
1008 1207
1009 req->result = -1; 1208 req->result = -1;
1010 1209
1011 if (!(flags & EIO_READDIR_DENTS)) 1210 if (!(flags & EIO_READDIR_DENTS))
1012 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1211 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1013 1212
1014 X_LOCK (wrklock); 1213 X_LOCK (wrklock);
1015
1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1214 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1017 self->dirp = dirp = opendir (req->ptr1); 1215 self->dirp = dirp = opendir (req->ptr1);
1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1216 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; 1217 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1021 1218 req->ptr2 = names = malloc (namesalloc);
1022 X_UNLOCK (wrklock); 1219 X_UNLOCK (wrklock);
1023 1220
1024 if (dirp && names && (!flags || dents)) 1221 if (dirp && names && (!flags || dents))
1025 for (;;) 1222 for (;;)
1026 { 1223 {
… …
1034 1231
1035 /* sort etc. */ 1232 /* sort etc. */
1036 req->int1 = flags; 1233 req->int1 = flags;
1037 req->result = dentoffs; 1234 req->result = dentoffs;
1038 1235
1039 if (flags & EIO_READDIR_STAT_ORDER || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)) 1236 if (flags & EIO_READDIR_STAT_ORDER)
1237 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
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
1040 { 1242 {
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 */ 1243 /* in this case, all is known, and we just put dirs first and sort them */
1047 eio_dirent *ent = dents + dentoffs; 1244 eio_dirent *oth = dents + dentoffs;
1048 eio_dirent *dir = dents; 1245 eio_dirent *dir = dents;
1049 1246
1247 /* now partition dirs to the front, and non-dirs to the back */
1248 /* by walking from both sides and swapping if necessary */
1249 /* also clear score, so it doesn't influence sorting */
1050 while (ent > dir) 1250 while (oth > dir)
1051 { 1251 {
1052 if (dir->type == DT_DIR) 1252 if (dir->type == EIO_DT_DIR)
1053 ++dir; 1253 ++dir;
1054 else
1055 {
1056 --ent;
1057
1058 if (ent->type == DT_DIR) 1254 else if ((--oth)->type == EIO_DT_DIR)
1059 { 1255 {
1060 eio_dirent tmp = *dir; 1256 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1061 *dir = *ent;
1062 *ent = tmp;
1063 1257
1064 ++dir; 1258 ++dir;
1065 } 1259 }
1066 } 1260 }
1067 }
1068 1261
1069 /* now sort the dirs only */ 1262 /* now sort the dirs only */
1070 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); 1263 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1071 } 1264 }
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 1265
1075 break; 1266 break;
1076 } 1267 }
1077 1268
1269 /* now add the entry to our list(s) */
1078 name = entp->d_name; 1270 name = entp->d_name;
1079 1271
1272 /* skip . and .. entries */
1080 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1273 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1081 { 1274 {
1082 int len = strlen (name) + 1; 1275 int len = D_NAMLEN (entp) + 1;
1083 1276
1084 while (expect_false (namesoffs + len > namesalloc)) 1277 while (expect_false (namesoffs + len > namesalloc))
1085 { 1278 {
1086 namesalloc *= 2; 1279 namesalloc *= 2;
1087 X_LOCK (wrklock); 1280 X_LOCK (wrklock);
1088 req->ptr1 = names = realloc (names, namesalloc); 1281 req->ptr2 = names = realloc (names, namesalloc);
1089 X_UNLOCK (wrklock); 1282 X_UNLOCK (wrklock);
1090 1283
1091 if (!names) 1284 if (!names)
1092 break; 1285 break;
1093 } 1286 }
… …
1100 1293
1101 if (expect_false (dentoffs == dentalloc)) 1294 if (expect_false (dentoffs == dentalloc))
1102 { 1295 {
1103 dentalloc *= 2; 1296 dentalloc *= 2;
1104 X_LOCK (wrklock); 1297 X_LOCK (wrklock);
1105 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1298 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1106 X_UNLOCK (wrklock); 1299 X_UNLOCK (wrklock);
1107 1300
1108 if (!dents) 1301 if (!dents)
1109 break; 1302 break;
1110 } 1303 }
1111 1304
1112 ent = dents + dentoffs; 1305 ent = dents + dentoffs;
1113 1306
1114 ent->name = names + namesoffs; 1307 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1115 ent->namelen = len - 1; 1308 ent->namelen = len - 1;
1116 ent->inode = D_INO (entp); 1309 ent->inode = D_INO (entp);
1310
1311 inode_bits |= ent->inode;
1117 1312
1118 switch (D_TYPE (entp)) 1313 switch (D_TYPE (entp))
1119 { 1314 {
1120 default: 1315 default:
1121 ent->type = EIO_DT_UNKNOWN; 1316 ent->type = EIO_DT_UNKNOWN;
… …
1126 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1321 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1127 #endif 1322 #endif
1128 #ifdef DT_CHR 1323 #ifdef DT_CHR
1129 case DT_CHR: ent->type = EIO_DT_CHR; break; 1324 case DT_CHR: ent->type = EIO_DT_CHR; break;
1130 #endif 1325 #endif
1326 #ifdef DT_MPC
1327 case DT_MPC: ent->type = EIO_DT_MPC; break;
1328 #endif
1131 #ifdef DT_DIR 1329 #ifdef DT_DIR
1132 case DT_DIR: ent->type = EIO_DT_DIR; break; 1330 case DT_DIR: ent->type = EIO_DT_DIR; break;
1133 #endif 1331 #endif
1332 #ifdef DT_NAM
1333 case DT_NAM: ent->type = EIO_DT_NAM; break;
1334 #endif
1134 #ifdef DT_BLK 1335 #ifdef DT_BLK
1135 case DT_BLK: ent->type = EIO_DT_BLK; break; 1336 case DT_BLK: ent->type = EIO_DT_BLK; break;
1136 #endif 1337 #endif
1338 #ifdef DT_MPB
1339 case DT_MPB: ent->type = EIO_DT_MPB; break;
1340 #endif
1137 #ifdef DT_REG 1341 #ifdef DT_REG
1138 case DT_REG: ent->type = EIO_DT_REG; break; 1342 case DT_REG: ent->type = EIO_DT_REG; break;
1139 #endif 1343 #endif
1344 #ifdef DT_NWK
1345 case DT_NWK: ent->type = EIO_DT_NWK; break;
1346 #endif
1347 #ifdef DT_CMP
1348 case DT_CMP: ent->type = EIO_DT_CMP; break;
1349 #endif
1140 #ifdef DT_LNK 1350 #ifdef DT_LNK
1141 case DT_LNK: ent->type = EIO_DT_LNK; break; 1351 case DT_LNK: ent->type = EIO_DT_LNK; break;
1142 #endif 1352 #endif
1143 #ifdef DT_SOCK 1353 #ifdef DT_SOCK
1144 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1354 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1145 #endif 1355 #endif
1356 #ifdef DT_DOOR
1357 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1358 #endif
1146 #ifdef DT_WHT 1359 #ifdef DT_WHT
1147 case DT_WHT: ent->type = EIO_DT_WHT; break; 1360 case DT_WHT: ent->type = EIO_DT_WHT; break;
1148 #endif 1361 #endif
1149 } 1362 }
1150 1363
1151 ent->score = 0; 1364 ent->score = 7;
1152 1365
1153 if (flags & EIO_READDIR_DIRS_FIRST) 1366 if (flags & EIO_READDIR_DIRS_FIRST)
1154 { 1367 {
1155 if (ent->type == EIO_DT_UNKNOWN) 1368 if (ent->type == EIO_DT_UNKNOWN)
1156 { 1369 {
1157 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 */
1158 ent->score = 98; 1371 ent->score = 1;
1159 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1372 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 */ 1373 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1161 } 1374 }
1162 else if (ent->type == DT_DIR) 1375 else if (ent->type == EIO_DT_DIR)
1163 ent->score = 100; 1376 ent->score = 0;
1164 } 1377 }
1165 } 1378 }
1166 1379
1167 namesoffs += len; 1380 namesoffs += len;
1168 ++dentoffs; 1381 ++dentoffs;
1169 } 1382 }
1383
1384 if (EIO_CANCELLED (req))
1385 {
1386 errno = ECANCELED;
1387 break;
1388 }
1170 } 1389 }
1171 else 1390}
1172 req->result = -1;
1173 1391
1174 /* if user doesn't want the dents, do not provide it */ 1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1175 if (!(flags & EIO_READDIR_DENTS)) 1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1394#else
1395
1396int
1397eio__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)
1176 { 1402 {
1177 X_LOCK (wrklock); 1403 flags = 0
1178 free (dents); 1404 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1179 req->ptr2 = req->ptr1; 1405 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1180 req->ptr1 = 0; 1406 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1181 X_UNLOCK (wrklock);
1182 } 1407 }
1183}
1184 1408
1185#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1409 return msync (mem, len, flags);
1186# undef msync 1410}
1187# define msync(a,b,c) ((errno = ENOSYS), -1) 1411
1188#endif 1412#endif
1189 1413
1190int 1414int
1191eio__mtouch (void *mem, size_t len, int flags) 1415eio__mtouch (void *mem, size_t len, int flags)
1192{ 1416{
… …
1199 1423
1200 if (!page) 1424 if (!page)
1201 page = sysconf (_SC_PAGESIZE); 1425 page = sysconf (_SC_PAGESIZE);
1202#endif 1426#endif
1203 1427
1428 /* round down to start of page, although this is probably useless */
1204 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 */
1205 1430
1206 if (addr < end) 1431 if (addr < end)
1207 if (flags) /* modify */ 1432 if (flags & EIO_MT_MODIFY) /* modify */
1208 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1433 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1209 else 1434 else
1210 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1435 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1211 1436
1212 return 0; 1437 return 0;
… …
1360 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1585 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1361 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1586 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1362 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1587 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1363 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1588 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1364 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
1365 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;
1366 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;
1367 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;
1368 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;
1369 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1599 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1384 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1385 1615
1386 case EIO_SYNC: req->result = 0; sync (); break; 1616 case EIO_SYNC: req->result = 0; sync (); break;
1387 case EIO_FSYNC: req->result = fsync (req->int1); break; 1617 case EIO_FSYNC: req->result = fsync (req->int1); break;
1388 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1389 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;
1390 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;
1391 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;
1392 1622
1393 case EIO_READDIR: eio__scandir (req, self); break; 1623 case EIO_READDIR: eio__scandir (req, self); break;
1394 1624
… …
1516eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1746eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1517{ 1747{
1518 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1748 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1519} 1749}
1520 1750
1751eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1752{
1753 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1754}
1755
1521eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1756eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1522{ 1757{
1523 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;
1524} 1759}
1525 1760
… …
1595} 1830}
1596 1831
1597eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1832eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1598{ 1833{
1599 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1834 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1835}
1836
1837eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1838{
1839 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1600} 1840}
1601 1841
1602eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1842eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1603{ 1843{
1604 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1844 return eio__1path (EIO_UNLINK, path, pri, cb, data);
… …
1708/* misc garbage */ 1948/* misc garbage */
1709 1949
1710ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1950ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1711{ 1951{
1712 etp_worker wrk; 1952 etp_worker wrk;
1953 ssize_t ret;
1713 1954
1714 wrk.dbuf = 0; 1955 wrk.dbuf = 0;
1715 1956
1716 eio__sendfile (ofd, ifd, offset, count, &wrk); 1957 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1717 1958
1718 if (wrk.dbuf) 1959 if (wrk.dbuf)
1719 free (wrk.dbuf); 1960 free (wrk.dbuf);
1720}
1721 1961
1962 return ret;
1963}
1964

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines