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

Comparing libeio/eio.c (file contents):
Revision 1.32 by root, Sat Jun 6 18:06:55 2009 UTC vs.
Revision 1.54 by root, Tue Feb 23 14:32:45 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
127/* used for struct dirent, AIX doesn't provide it */ 137/* used for struct dirent, AIX doesn't provide it */
128#ifndef NAME_MAX 138#ifndef NAME_MAX
129# define NAME_MAX 4096 139# define NAME_MAX 4096
140#endif
141
142/* used for readlink etc. */
143#ifndef PATH_MAX
144# define PATH_MAX 4096
130#endif 145#endif
131 146
132/* buffer size for various temporary buffers */ 147/* buffer size for various temporary buffers */
133#define EIO_BUFSIZE 65536 148#define EIO_BUFSIZE 65536
134 149
… …
187 202
188/*****************************************************************************/ 203/*****************************************************************************/
189 204
190#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 205#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
191 206
192/* calculcate time difference in ~1/EIO_TICKS of a second */ 207/* calculate time difference in ~1/EIO_TICKS of a second */
193static int tvdiff (struct timeval *tv1, struct timeval *tv2) 208static int tvdiff (struct timeval *tv1, struct timeval *tv2)
194{ 209{
195 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 210 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
196 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 211 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
197} 212}
… …
589} 604}
590 605
591static void etp_set_max_poll_time (double nseconds) 606static void etp_set_max_poll_time (double nseconds)
592{ 607{
593 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 608 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
594 max_poll_time = nseconds; 609 max_poll_time = nseconds * EIO_TICKS;
595 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
596} 611}
597 612
598static void etp_set_max_poll_reqs (unsigned int maxreqs) 613static void etp_set_max_poll_reqs (unsigned int maxreqs)
599{ 614{
… …
858 if (!res || errno != ENOSYS) 873 if (!res || errno != ENOSYS)
859 return res; 874 return res;
860#endif 875#endif
861 876
862 /* even though we could play tricks with the flags, it's better to always 877 /* even though we could play tricks with the flags, it's better to always
863 * call fdatasync, as thta matches the expectation of it's users best */ 878 * call fdatasync, as that matches the expectation of its users best */
864 return fdatasync (fd); 879 return fdatasync (fd);
865} 880}
866 881
867#if !HAVE_READAHEAD 882#if !HAVE_READAHEAD
868# undef readahead 883# undef readahead
… …
900 915
901#if HAVE_SENDFILE 916#if HAVE_SENDFILE
902# if __linux 917# if __linux
903 res = sendfile (ofd, ifd, &offset, count); 918 res = sendfile (ofd, ifd, &offset, count);
904 919
905# elif __freebsd 920# elif __FreeBSD__
906 /* 921 /*
907 * Of course, the freebsd sendfile is a dire hack with no thoughts 922 * Of course, the freebsd sendfile is a dire hack with no thoughts
908 * wasted on making it similar to other I/O functions. 923 * wasted on making it similar to other I/O functions.
909 */ 924 */
910 { 925 {
911 off_t sbytes; 926 off_t sbytes;
912 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 927 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
913 928
914 if (res < 0 && sbytes) 929 #if 0 /* according to the manpage, this is correct, but broken behaviour */
915 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 930 /* freebsd' sendfile will return 0 on success */
931 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
932 /* not on e.g. EIO or EPIPE - sounds broken */
933 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
934 res = sbytes;
935 #endif
936
937 /* according to source inspection, this is correct, and useful behaviour */
938 if (sbytes)
939 res = sbytes;
940 }
941
942# elif defined (__APPLE__)
943
944 {
945 off_t sbytes = count;
946 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
947
948 /* according to the manpage, sbytes is always valid */
949 if (sbytes)
916 res = sbytes; 950 res = sbytes;
917 } 951 }
918 952
919# elif __hpux 953# elif __hpux
920 res = sendfile (ofd, ifd, offset, count, 0, 0); 954 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
934 if (res < 0 && sbytes) 968 if (res < 0 && sbytes)
935 res = sbytes; 969 res = sbytes;
936 } 970 }
937 971
938# endif 972# endif
973
974#elif defined (_WIN32)
975
976 /* does not work, just for documentation of what would need to be done */
977 {
978 HANDLE h = TO_SOCKET (ifd);
979 SetFilePointer (h, offset, 0, FILE_BEGIN);
980 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
981 }
982
939#else 983#else
940 res = -1; 984 res = -1;
941 errno = ENOSYS; 985 errno = ENOSYS;
942#endif 986#endif
943 987
944 if (res < 0 988 if (res < 0
945 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 989 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
990 /* BSDs */
991#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
992 || errno == ENOTSUP
993#endif
994 || errno == EOPNOTSUPP /* BSDs */
946#if __solaris 995#if __solaris
947 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 996 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
948#endif 997#endif
949 ) 998 )
950 ) 999 )
… …
981 } 1030 }
982 1031
983 return res; 1032 return res;
984} 1033}
985 1034
986static int 1035static signed char
987eio_dent_cmp (const void *a_, const void *b_) 1036eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
988{ 1037{
989 const eio_dirent *a = (const eio_dirent *)a_; 1038 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
990 const eio_dirent *b = (const eio_dirent *)b_;
991
992 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
993 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ 1039 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1040}
1041
1042#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1043
1044#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1045#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1046
1047static void
1048eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1049{
1050 unsigned char bits [9 + sizeof (ino_t) * 8];
1051 unsigned char *bit = bits;
1052
1053 assert (CHAR_BIT == 8);
1054 assert (sizeof (eio_dirent) * 8 < 256);
1055 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1056 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1057
1058 if (size <= EIO_SORT_FAST)
1059 return;
1060
1061 /* first prepare an array of bits to test in our radix sort */
1062 /* try to take endianness into account, as well as differences in ino_t sizes */
1063 /* inode_bits must contain all inodes ORed together */
1064 /* which is used to skip bits that are 0 everywhere, which is very common */
1065 {
1066 ino_t endianness;
1067 int i, j;
1068
1069 /* we store the byte offset of byte n into byte n of "endianness" */
1070 for (i = 0; i < sizeof (ino_t); ++i)
1071 ((unsigned char *)&endianness)[i] = i;
1072
1073 *bit++ = 0;
1074
1075 for (i = 0; i < sizeof (ino_t); ++i)
1076 {
1077 /* shifting off the byte offsets out of "endianness" */
1078 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1079 endianness >>= 8;
1080
1081 for (j = 0; j < 8; ++j)
1082 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1083 *bit++ = offs + j;
1084 }
1085
1086 for (j = 0; j < 8; ++j)
1087 if (score_bits & (1 << j))
1088 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1089 }
1090
1091 /* now actually do the sorting (a variant of MSD radix sort) */
1092 {
1093 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1094 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1095 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1096 int stk_idx = 0;
1097
1098 base_stk [stk_idx] = dents;
1099 end_stk [stk_idx] = dents + size;
1100 bit_stk [stk_idx] = bit - 1;
1101
1102 do
1103 {
1104 base = base_stk [stk_idx];
1105 end = end_stk [stk_idx];
1106 bit = bit_stk [stk_idx];
1107
1108 for (;;)
1109 {
1110 unsigned char O = *bit >> 3;
1111 unsigned char M = 1 << (*bit & 7);
1112
1113 eio_dirent *a = base;
1114 eio_dirent *b = end;
1115
1116 if (b - a < EIO_SORT_CUTOFF)
1117 break;
1118
1119 /* now bit-partition the array on the bit */
1120 /* this ugly asymmetric loop seems to perform much better than typical */
1121 /* partition algos found in the literature */
1122 do
1123 if (!(((unsigned char *)a)[O] & M))
1124 ++a;
1125 else if (!(((unsigned char *)--b)[O] & M))
1126 {
1127 eio_dirent tmp = *a; *a = *b; *b = tmp;
1128 ++a;
1129 }
1130 while (b > a);
1131
1132 /* next bit, or stop, if no bits left in this path */
1133 if (!*--bit)
1134 break;
1135
1136 base_stk [stk_idx] = a;
1137 end_stk [stk_idx] = end;
1138 bit_stk [stk_idx] = bit;
1139 ++stk_idx;
1140
1141 end = a;
1142 }
1143 }
1144 while (stk_idx--);
1145 }
1146}
1147
1148static void
1149eio_dent_insertion_sort (eio_dirent *dents, int size)
1150{
1151 /* first move the smallest element to the front, to act as a sentinel */
1152 {
1153 int i;
1154 eio_dirent *min = dents;
1155
1156 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1157 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1158 if (EIO_DENT_CMP (dents [i], <, *min))
1159 min = &dents [i];
1160
1161 /* swap elements 0 and j (minimum) */
1162 {
1163 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1164 }
1165 }
1166
1167 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1168 {
1169 eio_dirent *i, *j;
1170
1171 for (i = dents + 1; i < dents + size; ++i)
1172 {
1173 eio_dirent value = *i;
1174
1175 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1176 j [1] = j [0];
1177
1178 j [1] = value;
1179 }
1180 }
1181}
1182
1183static void
1184eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1185{
1186 if (size <= 1)
1187 return; /* our insertion sort relies on size > 0 */
1188
1189 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1190 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1191 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1192
1193 /* use an insertion sort at the end, or for small arrays, */
1194 /* as insertion sort is more efficient for small partitions */
1195 eio_dent_insertion_sort (dents, size);
994} 1196}
995 1197
996/* read a full directory */ 1198/* read a full directory */
997static void 1199static void
998eio__scandir (eio_req *req, etp_worker *self) 1200eio__scandir (eio_req *req, etp_worker *self)
999{ 1201{
1000 DIR *dirp; 1202 DIR *dirp;
1001 EIO_STRUCT_DIRENT *entp; 1203 EIO_STRUCT_DIRENT *entp;
1002 unsigned char *name, *names; 1204 char *name, *names;
1003 int namesalloc = 4096; 1205 int namesalloc = 4096;
1004 int namesoffs = 0; 1206 int namesoffs = 0;
1005 int flags = req->int1; 1207 int flags = req->int1;
1006 eio_dirent *dents = 0; 1208 eio_dirent *dents = 0;
1007 int dentalloc = 128; 1209 int dentalloc = 128;
1008 int dentoffs = 0; 1210 int dentoffs = 0;
1211 ino_t inode_bits = 0;
1009 1212
1010 req->result = -1; 1213 req->result = -1;
1011 1214
1012 if (!(flags & EIO_READDIR_DENTS)) 1215 if (!(flags & EIO_READDIR_DENTS))
1013 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1216 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1014 1217
1015 X_LOCK (wrklock); 1218 X_LOCK (wrklock);
1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1219 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1017 self->dirp = dirp = opendir (req->ptr1); 1220 self->dirp = dirp = opendir (req->ptr1);
1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1221 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; 1222 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1223 req->ptr2 = names = malloc (namesalloc);
1021 X_UNLOCK (wrklock); 1224 X_UNLOCK (wrklock);
1022 1225
1023 if (dirp && names && (!flags || dents)) 1226 if (dirp && names && (!flags || dents))
1024 for (;;) 1227 for (;;)
1025 { 1228 {
… …
1033 1236
1034 /* sort etc. */ 1237 /* sort etc. */
1035 req->int1 = flags; 1238 req->int1 = flags;
1036 req->result = dentoffs; 1239 req->result = dentoffs;
1037 1240
1038 if (dents)
1039 {
1040 eio_dirent *ent = dents + dentoffs;
1041
1042 while (ent > dents)
1043 (--ent)->name = names + (size_t)ent->name;
1044 }
1045
1046 if (flags & EIO_READDIR_STAT_ORDER 1241 if (flags & EIO_READDIR_STAT_ORDER)
1047 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1242 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1048 {
1049 /* pray your qsort doesn't use quicksort */
1050 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1051 }
1052 else if (flags & EIO_READDIR_DIRS_FIRST) 1243 else if (flags & EIO_READDIR_DIRS_FIRST)
1244 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1245 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1246 else
1053 { 1247 {
1054 /* in this case, all is known, and we just put dirs first and sort them */ 1248 /* in this case, all is known, and we just put dirs first and sort them */
1055 eio_dirent *ent = dents + dentoffs; 1249 eio_dirent *oth = dents + dentoffs;
1056 eio_dirent *dir = dents; 1250 eio_dirent *dir = dents;
1057 1251
1252 /* now partition dirs to the front, and non-dirs to the back */
1253 /* by walking from both sides and swapping if necessary */
1254 /* also clear score, so it doesn't influence sorting */
1058 while (ent > dir) 1255 while (oth > dir)
1059 { 1256 {
1060 if (dir->type == DT_DIR) 1257 if (dir->type == EIO_DT_DIR)
1061 ++dir; 1258 ++dir;
1062 else
1063 {
1064 --ent;
1065
1066 if (ent->type == DT_DIR) 1259 else if ((--oth)->type == EIO_DT_DIR)
1067 { 1260 {
1068 eio_dirent tmp = *dir; 1261 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1069 *dir = *ent;
1070 *ent = tmp;
1071 1262
1072 ++dir; 1263 ++dir;
1073 } 1264 }
1074 } 1265 }
1075 }
1076 1266
1077 /* now sort the dirs only */ 1267 /* now sort the dirs only */
1078 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); 1268 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1079 } 1269 }
1080
1081 /* only provide the names array unless DENTS is specified */
1082 if (!(flags & EIO_READDIR_DENTS))
1083 {
1084 X_LOCK (wrklock);
1085 assert (!dents);
1086 req->ptr1 = 0;
1087 req->ptr2 = names;
1088 X_UNLOCK (wrklock);
1089 }
1090 1270
1091 break; 1271 break;
1092 } 1272 }
1093 1273
1094 /* now add the entry to our list(s) */ 1274 /* now add the entry to our list(s) */
1095 name = entp->d_name; 1275 name = entp->d_name;
1096 1276
1097 /* skip . and .. entries */ 1277 /* skip . and .. entries */
1098 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1278 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1099 { 1279 {
1100 int len = strlen (name) + 1; 1280 int len = D_NAMLEN (entp) + 1;
1101 1281
1102 while (expect_false (namesoffs + len > namesalloc)) 1282 while (expect_false (namesoffs + len > namesalloc))
1103 { 1283 {
1104 namesalloc *= 2; 1284 namesalloc *= 2;
1105 X_LOCK (wrklock); 1285 X_LOCK (wrklock);
1106 req->ptr1 = names = realloc (names, namesalloc); 1286 req->ptr2 = names = realloc (names, namesalloc);
1107 X_UNLOCK (wrklock); 1287 X_UNLOCK (wrklock);
1108 1288
1109 if (!names) 1289 if (!names)
1110 break; 1290 break;
1111 } 1291 }
… …
1118 1298
1119 if (expect_false (dentoffs == dentalloc)) 1299 if (expect_false (dentoffs == dentalloc))
1120 { 1300 {
1121 dentalloc *= 2; 1301 dentalloc *= 2;
1122 X_LOCK (wrklock); 1302 X_LOCK (wrklock);
1123 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1303 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124 X_UNLOCK (wrklock); 1304 X_UNLOCK (wrklock);
1125 1305
1126 if (!dents) 1306 if (!dents)
1127 break; 1307 break;
1128 } 1308 }
1129 1309
1130 ent = dents + dentoffs; 1310 ent = dents + dentoffs;
1131 1311
1132 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1312 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1133 ent->namelen = len - 1; 1313 ent->namelen = len - 1;
1134 ent->inode = D_INO (entp); 1314 ent->inode = D_INO (entp);
1315
1316 inode_bits |= ent->inode;
1135 1317
1136 switch (D_TYPE (entp)) 1318 switch (D_TYPE (entp))
1137 { 1319 {
1138 default: 1320 default:
1139 ent->type = EIO_DT_UNKNOWN; 1321 ent->type = EIO_DT_UNKNOWN;
… …
1144 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1326 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145 #endif 1327 #endif
1146 #ifdef DT_CHR 1328 #ifdef DT_CHR
1147 case DT_CHR: ent->type = EIO_DT_CHR; break; 1329 case DT_CHR: ent->type = EIO_DT_CHR; break;
1148 #endif 1330 #endif
1331 #ifdef DT_MPC
1332 case DT_MPC: ent->type = EIO_DT_MPC; break;
1333 #endif
1149 #ifdef DT_DIR 1334 #ifdef DT_DIR
1150 case DT_DIR: ent->type = EIO_DT_DIR; break; 1335 case DT_DIR: ent->type = EIO_DT_DIR; break;
1336 #endif
1337 #ifdef DT_NAM
1338 case DT_NAM: ent->type = EIO_DT_NAM; break;
1151 #endif 1339 #endif
1152 #ifdef DT_BLK 1340 #ifdef DT_BLK
1153 case DT_BLK: ent->type = EIO_DT_BLK; break; 1341 case DT_BLK: ent->type = EIO_DT_BLK; break;
1154 #endif 1342 #endif
1343 #ifdef DT_MPB
1344 case DT_MPB: ent->type = EIO_DT_MPB; break;
1345 #endif
1155 #ifdef DT_REG 1346 #ifdef DT_REG
1156 case DT_REG: ent->type = EIO_DT_REG; break; 1347 case DT_REG: ent->type = EIO_DT_REG; break;
1348 #endif
1349 #ifdef DT_NWK
1350 case DT_NWK: ent->type = EIO_DT_NWK; break;
1351 #endif
1352 #ifdef DT_CMP
1353 case DT_CMP: ent->type = EIO_DT_CMP; break;
1157 #endif 1354 #endif
1158 #ifdef DT_LNK 1355 #ifdef DT_LNK
1159 case DT_LNK: ent->type = EIO_DT_LNK; break; 1356 case DT_LNK: ent->type = EIO_DT_LNK; break;
1160 #endif 1357 #endif
1161 #ifdef DT_SOCK 1358 #ifdef DT_SOCK
1162 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1359 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163 #endif 1360 #endif
1361 #ifdef DT_DOOR
1362 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1363 #endif
1164 #ifdef DT_WHT 1364 #ifdef DT_WHT
1165 case DT_WHT: ent->type = EIO_DT_WHT; break; 1365 case DT_WHT: ent->type = EIO_DT_WHT; break;
1166 #endif 1366 #endif
1167 } 1367 }
1168 1368
1169 ent->score = 0; 1369 ent->score = 7;
1170 1370
1171 if (flags & EIO_READDIR_DIRS_FIRST) 1371 if (flags & EIO_READDIR_DIRS_FIRST)
1172 { 1372 {
1173 if (ent->type == EIO_DT_UNKNOWN) 1373 if (ent->type == EIO_DT_UNKNOWN)
1174 { 1374 {
1175 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1375 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176 ent->score = 98; 1376 ent->score = 1;
1177 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1377 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1178 ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ 1378 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1179 } 1379 }
1180 else if (ent->type == DT_DIR) 1380 else if (ent->type == EIO_DT_DIR)
1181 ent->score = 100; 1381 ent->score = 0;
1182 } 1382 }
1183 } 1383 }
1184 1384
1185 namesoffs += len; 1385 namesoffs += len;
1186 ++dentoffs; 1386 ++dentoffs;
1187 } 1387 }
1388
1389 if (EIO_CANCELLED (req))
1390 {
1391 errno = ECANCELED;
1392 break;
1393 }
1188 } 1394 }
1189 else
1190 req->result = -1;
1191} 1395}
1192 1396
1193#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1397#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1194# undef msync
1195# define msync(a,b,c) ((errno = ENOSYS), -1) 1398# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1399#else
1400
1401int
1402eio__msync (void *mem, size_t len, int flags)
1403{
1404 if (EIO_MS_ASYNC != MS_SYNC
1405 || EIO_MS_INVALIDATE != MS_INVALIDATE
1406 || EIO_MS_SYNC != MS_SYNC)
1407 {
1408 flags = 0
1409 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1410 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1411 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1412 }
1413
1414 return msync (mem, len, flags);
1415}
1416
1196#endif 1417#endif
1197 1418
1198int 1419int
1199eio__mtouch (void *mem, size_t len, int flags) 1420eio__mtouch (void *mem, size_t len, int flags)
1200{ 1421{
… …
1207 1428
1208 if (!page) 1429 if (!page)
1209 page = sysconf (_SC_PAGESIZE); 1430 page = sysconf (_SC_PAGESIZE);
1210#endif 1431#endif
1211 1432
1433 /* round down to start of page, although this is probably useless */
1212 addr &= ~(page - 1); /* assume page size is always a power of two */ 1434 addr &= ~(page - 1); /* assume page size is always a power of two */
1213 1435
1214 if (addr < end) 1436 if (addr < end)
1215 if (flags) /* modify */ 1437 if (flags & EIO_MT_MODIFY) /* modify */
1216 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1438 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1217 else 1439 else
1218 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1440 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1219 1441
1220 return 0; 1442 return 0;
… …
1368 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1590 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1369 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1591 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1370 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1592 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1371 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1593 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1372 1594
1595 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1596 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1597 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1598 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1599
1373 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1600 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1374 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1601 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1375 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1602 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1376 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1603 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1377 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1604 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1386 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1613 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1387 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1614 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1388 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1615 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1389 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1616 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1390 1617
1391 case EIO_READLINK: ALLOC (NAME_MAX); 1618 case EIO_READLINK: ALLOC (PATH_MAX);
1392 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1619 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1393 1620
1394 case EIO_SYNC: req->result = 0; sync (); break; 1621 case EIO_SYNC: req->result = 0; sync (); break;
1395 case EIO_FSYNC: req->result = fsync (req->int1); break; 1622 case EIO_FSYNC: req->result = fsync (req->int1); break;
1396 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1623 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1397 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1624 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1398 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1625 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1399 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1626 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1400 1627
1401 case EIO_READDIR: eio__scandir (req, self); break; 1628 case EIO_READDIR: eio__scandir (req, self); break;
1402 1629
… …
1524eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1751eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1525{ 1752{
1526 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1753 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1527} 1754}
1528 1755
1756eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1757{
1758 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1759}
1760
1529eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1761eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1530{ 1762{
1531 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1763 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1532} 1764}
1533 1765
… …
1603} 1835}
1604 1836
1605eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1837eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1606{ 1838{
1607 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1839 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1840}
1841
1842eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1843{
1844 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1608} 1845}
1609 1846
1610eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1847eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1611{ 1848{
1612 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1849 return eio__1path (EIO_UNLINK, path, pri, cb, data);
… …
1716/* misc garbage */ 1953/* misc garbage */
1717 1954
1718ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1955ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1719{ 1956{
1720 etp_worker wrk; 1957 etp_worker wrk;
1958 ssize_t ret;
1721 1959
1722 wrk.dbuf = 0; 1960 wrk.dbuf = 0;
1723 1961
1724 eio__sendfile (ofd, ifd, offset, count, &wrk); 1962 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1725 1963
1726 if (wrk.dbuf) 1964 if (wrk.dbuf)
1727 free (wrk.dbuf); 1965 free (wrk.dbuf);
1728}
1729 1966
1967 return ret;
1968}
1969

Diff Legend

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