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

Comparing libeio/eio.c (file contents):
Revision 1.30 by root, Wed Jun 3 12:24:49 2009 UTC vs.
Revision 1.52 by root, Fri Jan 8 09:29:56 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 *
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
68#endif 69#endif
69 70
70#ifdef _WIN32 71#ifdef _WIN32
71 72
72 /*doh*/ 73 /*doh*/
73
74#else 74#else
75 75
76# include "config.h" 76# include "config.h"
77# include <sys/time.h> 77# include <sys/time.h>
78# include <sys/select.h> 78# include <sys/select.h>
79# include <sys/mman.h>
79# include <unistd.h> 80# include <unistd.h>
80# include <utime.h> 81# include <utime.h>
81# include <signal.h> 82# include <signal.h>
82# include <dirent.h> 83# include <dirent.h>
83 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
84# ifndef EIO_STRUCT_DIRENT 103# ifndef EIO_STRUCT_DIRENT
85# define EIO_STRUCT_DIRENT struct dirent 104# define EIO_STRUCT_DIRENT struct dirent
86# endif 105# endif
87 106
88#endif 107#endif
89 108
90#if HAVE_SENDFILE 109#if HAVE_SENDFILE
91# if __linux 110# if __linux
92# include <sys/sendfile.h> 111# include <sys/sendfile.h>
93# elif __freebsd 112# elif __FreeBSD__ || defined __APPLE__
94# include <sys/socket.h> 113# include <sys/socket.h>
95# include <sys/uio.h> 114# include <sys/uio.h>
96# elif __hpux 115# elif __hpux
97# include <sys/socket.h> 116# include <sys/socket.h>
98# elif __solaris /* not yet */ 117# elif __solaris
99# include <sys/sendfile.h> 118# include <sys/sendfile.h>
100# else 119# else
101# error sendfile support requested but not available 120# error sendfile support requested but not available
102# 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)
103#endif 132#endif
104 133
105/* number of seconds after which an idle threads exit */ 134/* number of seconds after which an idle threads exit */
106#define IDLE_TIMEOUT 10 135#define IDLE_TIMEOUT 10
107 136
159 if (wrk->dirp) \ 188 if (wrk->dirp) \
160 { \ 189 { \
161 closedir (wrk->dirp); \ 190 closedir (wrk->dirp); \
162 wrk->dirp = 0; \ 191 wrk->dirp = 0; \
163 } 192 }
193
164#define ETP_WORKER_COMMON \ 194#define ETP_WORKER_COMMON \
165 void *dbuf; \ 195 void *dbuf; \
166 DIR *dirp; 196 DIR *dirp;
167 197
168/*****************************************************************************/ 198/*****************************************************************************/
169 199
170#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
171 201
172/* calculcate time difference in ~1/EIO_TICKS of a second */ 202/* calculate time difference in ~1/EIO_TICKS of a second */
173static int tvdiff (struct timeval *tv1, struct timeval *tv2) 203static int tvdiff (struct timeval *tv1, struct timeval *tv2)
174{ 204{
175 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
176 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 206 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
177} 207}
569} 599}
570 600
571static void etp_set_max_poll_time (double nseconds) 601static void etp_set_max_poll_time (double nseconds)
572{ 602{
573 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 603 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
574 max_poll_time = nseconds; 604 max_poll_time = nseconds * EIO_TICKS;
575 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
576} 606}
577 607
578static void etp_set_max_poll_reqs (unsigned int maxreqs) 608static void etp_set_max_poll_reqs (unsigned int maxreqs)
579{ 609{
838 if (!res || errno != ENOSYS) 868 if (!res || errno != ENOSYS)
839 return res; 869 return res;
840#endif 870#endif
841 871
842 /* 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
843 * call fdatasync, as thta matches the expectation of it's users best */ 873 * call fdatasync, as that matches the expectation of its users best */
844 return fdatasync (fd); 874 return fdatasync (fd);
845} 875}
846 876
847#if !HAVE_READAHEAD 877#if !HAVE_READAHEAD
848# undef readahead 878# undef readahead
880 910
881#if HAVE_SENDFILE 911#if HAVE_SENDFILE
882# if __linux 912# if __linux
883 res = sendfile (ofd, ifd, &offset, count); 913 res = sendfile (ofd, ifd, &offset, count);
884 914
885# elif __freebsd 915# elif __FreeBSD__
886 /* 916 /*
887 * 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
888 * wasted on making it similar to other I/O functions. 918 * wasted on making it similar to other I/O functions.
889 */ 919 */
890 { 920 {
891 off_t sbytes; 921 off_t sbytes;
892 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
893 923
894 if (res < 0 && sbytes) 924 /* freebsd' sendfile will return 0 on success */
895 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 925 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
926 /* not on e.g. EIO or EPIPE - sounds broken */
927 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
928 res = sbytes;
929 }
930
931# elif defined (__APPLE__)
932
933 {
934 off_t sbytes = count;
935 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
936
937 /* according to the manpage, sbytes is always valid */
938 if (sbytes)
896 res = sbytes; 939 res = sbytes;
897 } 940 }
898 941
899# elif __hpux 942# elif __hpux
900 res = sendfile (ofd, ifd, offset, count, 0, 0); 943 res = sendfile (ofd, ifd, offset, count, 0, 0);
914 if (res < 0 && sbytes) 957 if (res < 0 && sbytes)
915 res = sbytes; 958 res = sbytes;
916 } 959 }
917 960
918# endif 961# endif
962
963#elif defined (_WIN32)
964
965 /* does not work, just for documentation of what would need to be done */
966 {
967 HANDLE h = TO_SOCKET (ifd);
968 SetFilePointer (h, offset, 0, FILE_BEGIN);
969 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
970 }
971
919#else 972#else
920 res = -1; 973 res = -1;
921 errno = ENOSYS; 974 errno = ENOSYS;
922#endif 975#endif
923 976
924 if (res < 0 977 if (res < 0
925 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 978 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
979 /* BSDs */
980#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
981 || errno == ENOTSUP
982#endif
983 || errno == EOPNOTSUPP /* BSDs */
926#if __solaris 984#if __solaris
927 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 985 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
928#endif 986#endif
929 ) 987 )
930 ) 988 )
961 } 1019 }
962 1020
963 return res; 1021 return res;
964} 1022}
965 1023
1024static signed char
1025eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1026{
1027 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1028 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1029}
1030
1031#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1032
1033#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1034#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1035
1036static void
1037eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1038{
1039 unsigned char bits [9 + sizeof (ino_t) * 8];
1040 unsigned char *bit = bits;
1041
1042 assert (CHAR_BIT == 8);
1043 assert (sizeof (eio_dirent) * 8 < 256);
1044 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1045 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1046
1047 if (size <= EIO_SORT_FAST)
1048 return;
1049
1050 /* first prepare an array of bits to test in our radix sort */
1051 /* try to take endianness into account, as well as differences in ino_t sizes */
1052 /* inode_bits must contain all inodes ORed together */
1053 /* which is used to skip bits that are 0 everywhere, which is very common */
1054 {
1055 ino_t endianness;
1056 int i, j;
1057
1058 /* we store the byte offset of byte n into byte n of "endianness" */
1059 for (i = 0; i < sizeof (ino_t); ++i)
1060 ((unsigned char *)&endianness)[i] = i;
1061
1062 *bit++ = 0;
1063
1064 for (i = 0; i < sizeof (ino_t); ++i)
1065 {
1066 /* shifting off the byte offsets out of "endianness" */
1067 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1068 endianness >>= 8;
1069
1070 for (j = 0; j < 8; ++j)
1071 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1072 *bit++ = offs + j;
1073 }
1074
1075 for (j = 0; j < 8; ++j)
1076 if (score_bits & (1 << j))
1077 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1078 }
1079
1080 /* now actually do the sorting (a variant of MSD radix sort) */
1081 {
1082 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1083 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1084 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1085 int stk_idx = 0;
1086
1087 base_stk [stk_idx] = dents;
1088 end_stk [stk_idx] = dents + size;
1089 bit_stk [stk_idx] = bit - 1;
1090
1091 do
1092 {
1093 base = base_stk [stk_idx];
1094 end = end_stk [stk_idx];
1095 bit = bit_stk [stk_idx];
1096
1097 for (;;)
1098 {
1099 unsigned char O = *bit >> 3;
1100 unsigned char M = 1 << (*bit & 7);
1101
1102 eio_dirent *a = base;
1103 eio_dirent *b = end;
1104
1105 if (b - a < EIO_SORT_CUTOFF)
1106 break;
1107
1108 /* now bit-partition the array on the bit */
1109 /* this ugly asymmetric loop seems to perform much better than typical */
1110 /* partition algos found in the literature */
1111 do
1112 if (!(((unsigned char *)a)[O] & M))
1113 ++a;
1114 else if (!(((unsigned char *)--b)[O] & M))
1115 {
1116 eio_dirent tmp = *a; *a = *b; *b = tmp;
1117 ++a;
1118 }
1119 while (b > a);
1120
1121 /* next bit, or stop, if no bits left in this path */
1122 if (!*--bit)
1123 break;
1124
1125 base_stk [stk_idx] = a;
1126 end_stk [stk_idx] = end;
1127 bit_stk [stk_idx] = bit;
1128 ++stk_idx;
1129
1130 end = a;
1131 }
1132 }
1133 while (stk_idx--);
1134 }
1135}
1136
1137static void
1138eio_dent_insertion_sort (eio_dirent *dents, int size)
1139{
1140 /* first move the smallest element to the front, to act as a sentinel */
1141 {
1142 int i;
1143 eio_dirent *min = dents;
1144
1145 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1146 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1147 if (EIO_DENT_CMP (dents [i], <, *min))
1148 min = &dents [i];
1149
1150 /* swap elements 0 and j (minimum) */
1151 {
1152 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1153 }
1154 }
1155
1156 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1157 {
1158 eio_dirent *i, *j;
1159
1160 for (i = dents + 1; i < dents + size; ++i)
1161 {
1162 eio_dirent value = *i;
1163
1164 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1165 j [1] = j [0];
1166
1167 j [1] = value;
1168 }
1169 }
1170}
1171
1172static void
1173eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1174{
1175 if (size <= 1)
1176 return; /* our insertion sort relies on size > 0 */
1177
1178 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1179 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1180 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1181
1182 /* use an insertion sort at the end, or for small arrays, */
1183 /* as insertion sort is more efficient for small partitions */
1184 eio_dent_insertion_sort (dents, size);
1185}
1186
966/* read a full directory */ 1187/* read a full directory */
967static void 1188static void
968eio__scandir (eio_req *req, etp_worker *self) 1189eio__scandir (eio_req *req, etp_worker *self)
969{ 1190{
970 DIR *dirp; 1191 DIR *dirp;
971 EIO_STRUCT_DIRENT *entp; 1192 EIO_STRUCT_DIRENT *entp;
972 char *name, *names; 1193 char *name, *names;
973 int memlen = 4096; 1194 int namesalloc = 4096;
974 int memofs = 0; 1195 int namesoffs = 0;
1196 int flags = req->int1;
1197 eio_dirent *dents = 0;
1198 int dentalloc = 128;
975 int res = 0; 1199 int dentoffs = 0;
1200 ino_t inode_bits = 0;
1201
1202 req->result = -1;
1203
1204 if (!(flags & EIO_READDIR_DENTS))
1205 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
976 1206
977 X_LOCK (wrklock); 1207 X_LOCK (wrklock);
978 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1208 /* the corresponding closedir is in ETP_WORKER_CLEAR */
979 self->dirp = dirp = opendir (req->ptr1); 1209 self->dirp = dirp = opendir (req->ptr1);
980 req->flags |= EIO_FLAG_PTR2_FREE; 1210 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1211 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
981 req->ptr2 = names = malloc (memlen); 1212 req->ptr2 = names = malloc (namesalloc);
982 X_UNLOCK (wrklock); 1213 X_UNLOCK (wrklock);
983 1214
984 if (dirp && names) 1215 if (dirp && names && (!flags || dents))
985 for (;;) 1216 for (;;)
986 { 1217 {
987 errno = 0; 1218 errno = 0;
988 entp = readdir (dirp); 1219 entp = readdir (dirp);
989 1220
990 if (!entp) 1221 if (!entp)
1222 {
1223 if (errno)
1224 break;
1225
1226 /* sort etc. */
1227 req->int1 = flags;
1228 req->result = dentoffs;
1229
1230 if (flags & EIO_READDIR_STAT_ORDER)
1231 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1232 else if (flags & EIO_READDIR_DIRS_FIRST)
1233 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1234 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1235 else
1236 {
1237 /* in this case, all is known, and we just put dirs first and sort them */
1238 eio_dirent *oth = dents + dentoffs;
1239 eio_dirent *dir = dents;
1240
1241 /* now partition dirs to the front, and non-dirs to the back */
1242 /* by walking from both sides and swapping if necessary */
1243 /* also clear score, so it doesn't influence sorting */
1244 while (oth > dir)
1245 {
1246 if (dir->type == EIO_DT_DIR)
1247 ++dir;
1248 else if ((--oth)->type == EIO_DT_DIR)
1249 {
1250 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1251
1252 ++dir;
1253 }
1254 }
1255
1256 /* now sort the dirs only */
1257 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1258 }
1259
991 break; 1260 break;
1261 }
992 1262
1263 /* now add the entry to our list(s) */
993 name = entp->d_name; 1264 name = entp->d_name;
994 1265
1266 /* skip . and .. entries */
995 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1267 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
996 { 1268 {
997 int len = strlen (name) + 1; 1269 int len = D_NAMLEN (entp) + 1;
998 1270
999 res++; 1271 while (expect_false (namesoffs + len > namesalloc))
1000
1001 while (memofs + len > memlen)
1002 { 1272 {
1003 memlen *= 2; 1273 namesalloc *= 2;
1004 X_LOCK (wrklock); 1274 X_LOCK (wrklock);
1005 req->ptr2 = names = realloc (names, memlen); 1275 req->ptr2 = names = realloc (names, namesalloc);
1006 X_UNLOCK (wrklock); 1276 X_UNLOCK (wrklock);
1007 1277
1008 if (!names) 1278 if (!names)
1009 break; 1279 break;
1010 } 1280 }
1011 1281
1012 memcpy (names + memofs, name, len); 1282 memcpy (names + namesoffs, name, len);
1283
1284 if (dents)
1285 {
1286 struct eio_dirent *ent;
1287
1288 if (expect_false (dentoffs == dentalloc))
1289 {
1290 dentalloc *= 2;
1291 X_LOCK (wrklock);
1292 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1293 X_UNLOCK (wrklock);
1294
1295 if (!dents)
1296 break;
1297 }
1298
1299 ent = dents + dentoffs;
1300
1301 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1302 ent->namelen = len - 1;
1303 ent->inode = D_INO (entp);
1304
1305 inode_bits |= ent->inode;
1306
1307 switch (D_TYPE (entp))
1308 {
1309 default:
1310 ent->type = EIO_DT_UNKNOWN;
1311 flags |= EIO_READDIR_FOUND_UNKNOWN;
1312 break;
1313
1314 #ifdef DT_FIFO
1315 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1316 #endif
1317 #ifdef DT_CHR
1318 case DT_CHR: ent->type = EIO_DT_CHR; break;
1319 #endif
1320 #ifdef DT_MPC
1321 case DT_MPC: ent->type = EIO_DT_MPC; break;
1322 #endif
1323 #ifdef DT_DIR
1324 case DT_DIR: ent->type = EIO_DT_DIR; break;
1325 #endif
1326 #ifdef DT_NAM
1327 case DT_NAM: ent->type = EIO_DT_NAM; break;
1328 #endif
1329 #ifdef DT_BLK
1330 case DT_BLK: ent->type = EIO_DT_BLK; break;
1331 #endif
1332 #ifdef DT_MPB
1333 case DT_MPB: ent->type = EIO_DT_MPB; break;
1334 #endif
1335 #ifdef DT_REG
1336 case DT_REG: ent->type = EIO_DT_REG; break;
1337 #endif
1338 #ifdef DT_NWK
1339 case DT_NWK: ent->type = EIO_DT_NWK; break;
1340 #endif
1341 #ifdef DT_CMP
1342 case DT_CMP: ent->type = EIO_DT_CMP; break;
1343 #endif
1344 #ifdef DT_LNK
1345 case DT_LNK: ent->type = EIO_DT_LNK; break;
1346 #endif
1347 #ifdef DT_SOCK
1348 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1349 #endif
1350 #ifdef DT_DOOR
1351 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1352 #endif
1353 #ifdef DT_WHT
1354 case DT_WHT: ent->type = EIO_DT_WHT; break;
1355 #endif
1356 }
1357
1358 ent->score = 7;
1359
1360 if (flags & EIO_READDIR_DIRS_FIRST)
1361 {
1362 if (ent->type == EIO_DT_UNKNOWN)
1363 {
1364 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1365 ent->score = 1;
1366 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1367 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1368 }
1369 else if (ent->type == EIO_DT_DIR)
1370 ent->score = 0;
1371 }
1372 }
1373
1013 memofs += len; 1374 namesoffs += len;
1375 ++dentoffs;
1376 }
1377
1378 if (EIO_CANCELLED (req))
1379 {
1380 errno = ECANCELED;
1381 break;
1014 } 1382 }
1015 } 1383 }
1016
1017 if (errno)
1018 res = -1;
1019
1020 req->result = res;
1021} 1384}
1022 1385
1023#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1386#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1024# undef msync
1025# define msync(a,b,c) ((errno = ENOSYS), -1) 1387# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1388#else
1389
1390int
1391eio__msync (void *mem, size_t len, int flags)
1392{
1393 if (EIO_MS_ASYNC != MS_SYNC
1394 || EIO_MS_INVALIDATE != MS_INVALIDATE
1395 || EIO_MS_SYNC != MS_SYNC)
1396 {
1397 flags = 0
1398 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1399 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1400 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1401 }
1402
1403 return msync (mem, len, flags);
1404}
1405
1026#endif 1406#endif
1027 1407
1028int 1408int
1029eio__mtouch (void *mem, size_t len, int flags) 1409eio__mtouch (void *mem, size_t len, int flags)
1030{ 1410{
1037 1417
1038 if (!page) 1418 if (!page)
1039 page = sysconf (_SC_PAGESIZE); 1419 page = sysconf (_SC_PAGESIZE);
1040#endif 1420#endif
1041 1421
1422 /* round down to start of page, although this is probably useless */
1042 addr &= ~(page - 1); /* assume page size is always a power of two */ 1423 addr &= ~(page - 1); /* assume page size is always a power of two */
1043 1424
1044 if (addr < end) 1425 if (addr < end)
1045 if (flags) /* modify */ 1426 if (flags & EIO_MT_MODIFY) /* modify */
1046 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1427 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1047 else 1428 else
1048 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1429 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1049 1430
1050 return 0; 1431 return 0;
1198 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1579 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1199 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1580 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1200 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1581 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1201 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1582 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1202 1583
1584 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1585 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1586 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1587 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1588
1203 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1589 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1204 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1590 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1205 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1591 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1206 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1592 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1207 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1593 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1222 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1608 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1223 1609
1224 case EIO_SYNC: req->result = 0; sync (); break; 1610 case EIO_SYNC: req->result = 0; sync (); break;
1225 case EIO_FSYNC: req->result = fsync (req->int1); break; 1611 case EIO_FSYNC: req->result = fsync (req->int1); break;
1226 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1612 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1227 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1613 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1228 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1614 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1229 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1615 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1230 1616
1231 case EIO_READDIR: eio__scandir (req, self); break; 1617 case EIO_READDIR: eio__scandir (req, self); break;
1232 1618
1354eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1740eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1355{ 1741{
1356 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1742 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1357} 1743}
1358 1744
1745eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1746{
1747 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1748}
1749
1359eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1750eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1360{ 1751{
1361 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1752 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1362} 1753}
1363 1754
1435eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1826eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1436{ 1827{
1437 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1828 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1438} 1829}
1439 1830
1831eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1832{
1833 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1834}
1835
1440eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1836eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1441{ 1837{
1442 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1838 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1443} 1839}
1444 1840
1445eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1841eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1446{ 1842{
1447 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1843 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1448} 1844}
1449 1845
1450eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1846eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1451{ 1847{
1452 return eio__1path (EIO_READDIR, path, pri, cb, data); 1848 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1453} 1849}
1454 1850
1455eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1851eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1456{ 1852{
1457 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1853 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1546/* misc garbage */ 1942/* misc garbage */
1547 1943
1548ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1944ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1549{ 1945{
1550 etp_worker wrk; 1946 etp_worker wrk;
1947 ssize_t ret;
1551 1948
1552 wrk.dbuf = 0; 1949 wrk.dbuf = 0;
1553 1950
1554 eio__sendfile (ofd, ifd, offset, count, &wrk); 1951 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1555 1952
1556 if (wrk.dbuf) 1953 if (wrk.dbuf)
1557 free (wrk.dbuf); 1954 free (wrk.dbuf);
1558}
1559 1955
1956 return ret;
1957}
1958

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines