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

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 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 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 *
… …
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#include "eio.h" 40#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
41#include "xthread.h" 45#include "xthread.h"
42 46
43#include <errno.h> 47#include <errno.h>
44#include <stddef.h> 48#include <stddef.h>
45#include <stdlib.h> 49#include <stdlib.h>
46#include <string.h> 50#include <string.h>
47#include <errno.h> 51#include <errno.h>
48#include <sys/types.h> 52#include <sys/types.h>
49#include <sys/stat.h> 53#include <sys/stat.h>
54#include <sys/statvfs.h>
50#include <limits.h> 55#include <limits.h>
51#include <fcntl.h> 56#include <fcntl.h>
52#include <assert.h> 57#include <assert.h>
53 58
54#ifndef EIO_FINISH 59#ifndef EIO_FINISH
… …
64#endif 69#endif
65 70
66#ifdef _WIN32 71#ifdef _WIN32
67 72
68 /*doh*/ 73 /*doh*/
69
70#else 74#else
71 75
72# include "config.h" 76# include "config.h"
73# include <sys/time.h> 77# include <sys/time.h>
74# include <sys/select.h> 78# include <sys/select.h>
79# include <sys/mman.h>
75# include <unistd.h> 80# include <unistd.h>
76# include <utime.h> 81# include <utime.h>
77# include <signal.h> 82# include <signal.h>
78# include <dirent.h> 83# include <dirent.h>
79 84
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino
92# endif
93
94#ifdef _D_EXACT_NAMLEN
95# undef D_NAMLEN
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
98
99# ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type
101# endif
102
80# ifndef EIO_STRUCT_DIRENT 103# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 104# define EIO_STRUCT_DIRENT struct dirent
82# endif 105# endif
83 106
84#endif 107#endif
85 108
86#if HAVE_SENDFILE 109#if HAVE_SENDFILE
87# if __linux 110# if __linux
88# include <sys/sendfile.h> 111# include <sys/sendfile.h>
89# elif __freebsd 112# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 113# include <sys/socket.h>
91# include <sys/uio.h> 114# include <sys/uio.h>
92# elif __hpux 115# elif __hpux
93# include <sys/socket.h> 116# include <sys/socket.h>
94# elif __solaris /* not yet */ 117# elif __solaris
95# include <sys/sendfile.h> 118# include <sys/sendfile.h>
96# else 119# else
97# error sendfile support requested but not available 120# error sendfile support requested but not available
98# endif 121# endif
122#endif
123
124#ifndef D_TYPE
125# define D_TYPE(de) 0
126#endif
127#ifndef D_INO
128# define D_INO(de) 0
129#endif
130#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name)
99#endif 132#endif
100 133
101/* number of seconds after which an idle threads exit */ 134/* number of seconds after which an idle threads exit */
102#define IDLE_TIMEOUT 10 135#define IDLE_TIMEOUT 10
103 136
… …
155 if (wrk->dirp) \ 188 if (wrk->dirp) \
156 { \ 189 { \
157 closedir (wrk->dirp); \ 190 closedir (wrk->dirp); \
158 wrk->dirp = 0; \ 191 wrk->dirp = 0; \
159 } 192 }
193
160#define ETP_WORKER_COMMON \ 194#define ETP_WORKER_COMMON \
161 void *dbuf; \ 195 void *dbuf; \
162 DIR *dirp; 196 DIR *dirp;
163 197
164/*****************************************************************************/ 198/*****************************************************************************/
165 199
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167 201
168/* calculcate time difference in ~1/EIO_TICKS of a second */ 202/* calculate time difference in ~1/EIO_TICKS of a second */
169static int tvdiff (struct timeval *tv1, struct timeval *tv2) 203static int tvdiff (struct timeval *tv1, struct timeval *tv2)
170{ 204{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 206 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173} 207}
… …
565} 599}
566 600
567static void etp_set_max_poll_time (double nseconds) 601static void etp_set_max_poll_time (double nseconds)
568{ 602{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 603 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds; 604 max_poll_time = nseconds * EIO_TICKS;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572} 606}
573 607
574static void etp_set_max_poll_reqs (unsigned int maxreqs) 608static void etp_set_max_poll_reqs (unsigned int maxreqs)
575{ 609{
… …
767 801
768 X_LOCK (preadwritelock); 802 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR); 803 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET); 804 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count); 805 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET); 806 lseek (fd, ooffset, SEEK_SET);
773 X_UNLOCK (preadwritelock); 807 X_UNLOCK (preadwritelock);
774 808
775 return res; 809 return res;
776} 810}
777#endif 811#endif
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 863 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 864 }
831 865
832 res = sync_file_range (fd, offset, nbytes, flags); 866 res = sync_file_range (fd, offset, nbytes, flags);
833 867
834 if (res != ENOSYS) 868 if (!res || errno != ENOSYS)
835 return res; 869 return res;
836#endif 870#endif
837 871
838 /* 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
839 * call fdatasync, as thta matches the expectation of it's users best */ 873 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 874 return fdatasync (fd);
841} 875}
842 876
843#if !HAVE_READAHEAD 877#if !HAVE_READAHEAD
844# undef readahead 878# undef readahead
… …
876 910
877#if HAVE_SENDFILE 911#if HAVE_SENDFILE
878# if __linux 912# if __linux
879 res = sendfile (ofd, ifd, &offset, count); 913 res = sendfile (ofd, ifd, &offset, count);
880 914
881# elif __freebsd 915# elif __FreeBSD__
882 /* 916 /*
883 * 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
884 * wasted on making it similar to other I/O functions. 918 * wasted on making it similar to other I/O functions.
885 */ 919 */
886 { 920 {
887 off_t sbytes; 921 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 923
890 if (res < 0 && sbytes) 924 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* 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)
892 res = sbytes; 945 res = sbytes;
893 } 946 }
894 947
895# elif __hpux 948# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 949 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
910 if (res < 0 && sbytes) 963 if (res < 0 && sbytes)
911 res = sbytes; 964 res = sbytes;
912 } 965 }
913 966
914# 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
915#else 978#else
916 res = -1; 979 res = -1;
917 errno = ENOSYS; 980 errno = ENOSYS;
918#endif 981#endif
919 982
920 if (res < 0 983 if (res < 0
921 && (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 */
922#if __solaris 990#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 991 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 992#endif
925 ) 993 )
926 ) 994 )
… …
957 } 1025 }
958 1026
959 return res; 1027 return res;
960} 1028}
961 1029
1030static signed char
1031eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1032{
1033 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
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);
1191}
1192
962/* read a full directory */ 1193/* read a full directory */
963static void 1194static void
964eio__scandir (eio_req *req, etp_worker *self) 1195eio__scandir (eio_req *req, etp_worker *self)
965{ 1196{
966 DIR *dirp; 1197 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1198 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1199 char *name, *names;
969 int memlen = 4096; 1200 int namesalloc = 4096;
970 int memofs = 0; 1201 int namesoffs = 0;
1202 int flags = req->int1;
1203 eio_dirent *dents = 0;
1204 int dentalloc = 128;
971 int res = 0; 1205 int dentoffs = 0;
1206 ino_t inode_bits = 0;
1207
1208 req->result = -1;
1209
1210 if (!(flags & EIO_READDIR_DENTS))
1211 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
972 1212
973 X_LOCK (wrklock); 1213 X_LOCK (wrklock);
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1214 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1); 1215 self->dirp = dirp = opendir (req->ptr1);
976 req->flags |= EIO_FLAG_PTR2_FREE; 1216 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1217 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
977 req->ptr2 = names = malloc (memlen); 1218 req->ptr2 = names = malloc (namesalloc);
978 X_UNLOCK (wrklock); 1219 X_UNLOCK (wrklock);
979 1220
980 if (dirp && names) 1221 if (dirp && names && (!flags || dents))
981 for (;;) 1222 for (;;)
982 { 1223 {
983 errno = 0; 1224 errno = 0;
984 entp = readdir (dirp); 1225 entp = readdir (dirp);
985 1226
986 if (!entp) 1227 if (!entp)
1228 {
1229 if (errno)
1230 break;
1231
1232 /* sort etc. */
1233 req->int1 = flags;
1234 req->result = dentoffs;
1235
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
1242 {
1243 /* in this case, all is known, and we just put dirs first and sort them */
1244 eio_dirent *oth = dents + dentoffs;
1245 eio_dirent *dir = dents;
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 */
1250 while (oth > dir)
1251 {
1252 if (dir->type == EIO_DT_DIR)
1253 ++dir;
1254 else if ((--oth)->type == EIO_DT_DIR)
1255 {
1256 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1257
1258 ++dir;
1259 }
1260 }
1261
1262 /* now sort the dirs only */
1263 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1264 }
1265
987 break; 1266 break;
1267 }
988 1268
1269 /* now add the entry to our list(s) */
989 name = entp->d_name; 1270 name = entp->d_name;
990 1271
1272 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1273 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1274 {
993 int len = strlen (name) + 1; 1275 int len = D_NAMLEN (entp) + 1;
994 1276
995 res++; 1277 while (expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1278 {
999 memlen *= 2; 1279 namesalloc *= 2;
1000 X_LOCK (wrklock); 1280 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1281 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock); 1282 X_UNLOCK (wrklock);
1003 1283
1004 if (!names) 1284 if (!names)
1005 break; 1285 break;
1006 } 1286 }
1007 1287
1008 memcpy (names + memofs, name, len); 1288 memcpy (names + namesoffs, name, len);
1289
1290 if (dents)
1291 {
1292 struct eio_dirent *ent;
1293
1294 if (expect_false (dentoffs == dentalloc))
1295 {
1296 dentalloc *= 2;
1297 X_LOCK (wrklock);
1298 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1299 X_UNLOCK (wrklock);
1300
1301 if (!dents)
1302 break;
1303 }
1304
1305 ent = dents + dentoffs;
1306
1307 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1308 ent->namelen = len - 1;
1309 ent->inode = D_INO (entp);
1310
1311 inode_bits |= ent->inode;
1312
1313 switch (D_TYPE (entp))
1314 {
1315 default:
1316 ent->type = EIO_DT_UNKNOWN;
1317 flags |= EIO_READDIR_FOUND_UNKNOWN;
1318 break;
1319
1320 #ifdef DT_FIFO
1321 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1322 #endif
1323 #ifdef DT_CHR
1324 case DT_CHR: ent->type = EIO_DT_CHR; break;
1325 #endif
1326 #ifdef DT_MPC
1327 case DT_MPC: ent->type = EIO_DT_MPC; break;
1328 #endif
1329 #ifdef DT_DIR
1330 case DT_DIR: ent->type = EIO_DT_DIR; break;
1331 #endif
1332 #ifdef DT_NAM
1333 case DT_NAM: ent->type = EIO_DT_NAM; break;
1334 #endif
1335 #ifdef DT_BLK
1336 case DT_BLK: ent->type = EIO_DT_BLK; break;
1337 #endif
1338 #ifdef DT_MPB
1339 case DT_MPB: ent->type = EIO_DT_MPB; break;
1340 #endif
1341 #ifdef DT_REG
1342 case DT_REG: ent->type = EIO_DT_REG; break;
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
1350 #ifdef DT_LNK
1351 case DT_LNK: ent->type = EIO_DT_LNK; break;
1352 #endif
1353 #ifdef DT_SOCK
1354 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1355 #endif
1356 #ifdef DT_DOOR
1357 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1358 #endif
1359 #ifdef DT_WHT
1360 case DT_WHT: ent->type = EIO_DT_WHT; break;
1361 #endif
1362 }
1363
1364 ent->score = 7;
1365
1366 if (flags & EIO_READDIR_DIRS_FIRST)
1367 {
1368 if (ent->type == EIO_DT_UNKNOWN)
1369 {
1370 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1371 ent->score = 1;
1372 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1373 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1374 }
1375 else if (ent->type == EIO_DT_DIR)
1376 ent->score = 0;
1377 }
1378 }
1379
1009 memofs += len; 1380 namesoffs += len;
1381 ++dentoffs;
1382 }
1383
1384 if (EIO_CANCELLED (req))
1385 {
1386 errno = ECANCELED;
1387 break;
1010 } 1388 }
1011 } 1389 }
1012
1013 if (errno)
1014 res = -1;
1015
1016 req->result = res;
1017} 1390}
1018 1391
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1020# undef msync 1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1021# define msync(a,b,c) ENOSYS 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)
1402 {
1403 flags = 0
1404 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1405 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1406 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1407 }
1408
1409 return msync (mem, len, flags);
1410}
1411
1022#endif 1412#endif
1023 1413
1024int 1414int
1025eio__mtouch (void *mem, size_t len, int flags) 1415eio__mtouch (void *mem, size_t len, int flags)
1026{ 1416{
… …
1033 1423
1034 if (!page) 1424 if (!page)
1035 page = sysconf (_SC_PAGESIZE); 1425 page = sysconf (_SC_PAGESIZE);
1036#endif 1426#endif
1037 1427
1428 /* round down to start of page, although this is probably useless */
1038 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 */
1039 1430
1040 if (addr < end) 1431 if (addr < end)
1041 if (flags) /* modify */ 1432 if (flags & EIO_MT_MODIFY) /* modify */
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1433 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1043 else 1434 else
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1435 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1045 1436
1046 return 0; 1437 return 0;
… …
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1585 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1586 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1587 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1588 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198 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
1199 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;
1200 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;
1201 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;
1202 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;
1203 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1599 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1219 1615
1220 case EIO_SYNC: req->result = 0; sync (); break; 1616 case EIO_SYNC: req->result = 0; sync (); break;
1221 case EIO_FSYNC: req->result = fsync (req->int1); break; 1617 case EIO_FSYNC: req->result = fsync (req->int1); break;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1223 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;
1224 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;
1225 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;
1226 1622
1227 case EIO_READDIR: eio__scandir (req, self); break; 1623 case EIO_READDIR: eio__scandir (req, self); break;
1228 1624
… …
1350eio_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)
1351{ 1747{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1748 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 1749}
1354 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
1355eio_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)
1356{ 1757{
1357 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;
1358} 1759}
1359 1760
… …
1431eio_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)
1432{ 1833{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1834 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 1835}
1435 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);
1840}
1841
1436eio_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)
1437{ 1843{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1844 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 1845}
1440 1846
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1847eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 1848{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1849 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 1850}
1445 1851
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1852eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 1853{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 1854 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 1855}
1450 1856
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1857eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 1858{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1859 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
… …
1542/* misc garbage */ 1948/* misc garbage */
1543 1949
1544ssize_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)
1545{ 1951{
1546 etp_worker wrk; 1952 etp_worker wrk;
1953 ssize_t ret;
1547 1954
1548 wrk.dbuf = 0; 1955 wrk.dbuf = 0;
1549 1956
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 1957 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1551 1958
1552 if (wrk.dbuf) 1959 if (wrk.dbuf)
1553 free (wrk.dbuf); 1960 free (wrk.dbuf);
1554}
1555 1961
1962 return ret;
1963}
1964

Diff Legend

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