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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.39 by root, Sat Jun 13 03:59:04 2009 UTC vs.
Revision 1.67 by root, Sun Jun 5 20:25:00 2011 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) 2007,2008,2009,2010,2011 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 *
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
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#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
41 45
42#ifdef EIO_STACKSIZE 46#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 47# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif 48#endif
49#include <stdlib.h> 53#include <stdlib.h>
50#include <string.h> 54#include <string.h>
51#include <errno.h> 55#include <errno.h>
52#include <sys/types.h> 56#include <sys/types.h>
53#include <sys/stat.h> 57#include <sys/stat.h>
58#include <sys/statvfs.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
58#ifndef EIO_FINISH 69#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 70# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 71#endif
61 72
62#ifndef EIO_DESTROY 73#ifndef EIO_DESTROY
70#ifdef _WIN32 81#ifdef _WIN32
71 82
72 /*doh*/ 83 /*doh*/
73#else 84#else
74 85
75# include "config.h"
76# include <sys/time.h> 86# include <sys/time.h>
77# include <sys/select.h> 87# include <sys/select.h>
78# include <sys/mman.h>
79# include <unistd.h> 88# include <unistd.h>
80# include <utime.h> 89# include <utime.h>
81# include <signal.h> 90# include <signal.h>
82# include <dirent.h> 91# include <dirent.h>
83 92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */ 99# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno 100# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen 101# define D_NAMLEN(de) (de)->d_namlen
89# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino 103# define D_INO(de) (de)->d_ino
91# endif 104# endif
92 105
93#ifdef _D_EXACT_NAMLEN 106#ifdef _D_EXACT_NAMLEN
94# undef D_NAMLEN 107# undef D_NAMLEN
106#endif 119#endif
107 120
108#if HAVE_SENDFILE 121#if HAVE_SENDFILE
109# if __linux 122# if __linux
110# include <sys/sendfile.h> 123# include <sys/sendfile.h>
111# elif __freebsd 124# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 125# include <sys/socket.h>
113# include <sys/uio.h> 126# include <sys/uio.h>
114# elif __hpux 127# elif __hpux
115# include <sys/socket.h> 128# include <sys/socket.h>
116# elif __solaris /* not yet */ 129# elif __solaris
117# include <sys/sendfile.h> 130# include <sys/sendfile.h>
118# else 131# else
119# error sendfile support requested but not available 132# error sendfile support requested but not available
120# endif 133# endif
121#endif 134#endif
128#endif 141#endif
129#ifndef D_NAMLEN 142#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 143# define D_NAMLEN(de) strlen ((de)->d_name)
131#endif 144#endif
132 145
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135
136/* used for struct dirent, AIX doesn't provide it */ 146/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 147#ifndef NAME_MAX
138# define NAME_MAX 4096 148# define NAME_MAX 4096
149#endif
150
151/* used for readlink etc. */
152#ifndef PATH_MAX
153# define PATH_MAX 4096
139#endif 154#endif
140 155
141/* buffer size for various temporary buffers */ 156/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 157#define EIO_BUFSIZE 65536
143 158
196 211
197/*****************************************************************************/ 212/*****************************************************************************/
198 213
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200 215
201/* calculcate time difference in ~1/EIO_TICKS of a second */ 216/* calculate time difference in ~1/EIO_TICKS of a second */
202static int tvdiff (struct timeval *tv1, struct timeval *tv2) 217static int tvdiff (struct timeval *tv1, struct timeval *tv2)
203{ 218{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 220 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206} 221}
214static unsigned int max_poll_reqs; /* reslock */ 229static unsigned int max_poll_reqs; /* reslock */
215 230
216static volatile unsigned int nreqs; /* reqlock */ 231static volatile unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */ 232static volatile unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */ 233static volatile unsigned int npending; /* reqlock */
219static volatile unsigned int max_idle = 4; 234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220 236
221static mutex_t wrklock = X_MUTEX_INIT; 237static xmutex_t wrklock;
222static mutex_t reslock = X_MUTEX_INIT; 238static xmutex_t reslock;
223static mutex_t reqlock = X_MUTEX_INIT; 239static xmutex_t reqlock;
224static cond_t reqwait = X_COND_INIT; 240static xcond_t reqwait;
225 241
226#if !HAVE_PREADWRITE 242#if !HAVE_PREADWRITE
227/* 243/*
228 * make our pread/pwrite emulation safe against themselves, but not against 244 * make our pread/pwrite emulation safe against themselves, but not against
229 * normal read/write by using a mutex. slows down execution a lot, 245 * normal read/write by using a mutex. slows down execution a lot,
230 * but that's your problem, not mine. 246 * but that's your problem, not mine.
231 */ 247 */
232static mutex_t preadwritelock = X_MUTEX_INIT; 248static xmutex_t preadwritelock = X_MUTEX_INIT;
233#endif 249#endif
234 250
235typedef struct etp_worker 251typedef struct etp_worker
236{ 252{
237 /* locked by wrklock */ 253 /* locked by wrklock */
238 struct etp_worker *prev, *next; 254 struct etp_worker *prev, *next;
239 255
240 thread_t tid; 256 xthread_t tid;
241 257
242 /* locked by reslock, reqlock or wrklock */ 258 /* locked by reslock, reqlock or wrklock */
243 ETP_REQ *req; /* currently processed request */ 259 ETP_REQ *req; /* currently processed request */
244 260
245 ETP_WORKER_COMMON 261 ETP_WORKER_COMMON
359 } 375 }
360 376
361 abort (); 377 abort ();
362} 378}
363 379
380static void etp_thread_init (void)
381{
382 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait);
386}
387
364static void etp_atfork_prepare (void) 388static void etp_atfork_prepare (void)
365{ 389{
366 X_LOCK (wrklock); 390 X_LOCK (wrklock);
367 X_LOCK (reqlock); 391 X_LOCK (reqlock);
368 X_LOCK (reslock); 392 X_LOCK (reslock);
406 idle = 0; 430 idle = 0;
407 nreqs = 0; 431 nreqs = 0;
408 nready = 0; 432 nready = 0;
409 npending = 0; 433 npending = 0;
410 434
411 etp_atfork_parent (); 435 etp_thread_init ();
412} 436}
413 437
414static void 438static void
415etp_once_init (void) 439etp_once_init (void)
416{ 440{
441 etp_thread_init ();
417 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418} 443}
419 444
420static int 445static int
421etp_init (void (*want_poll)(void), void (*done_poll)(void)) 446etp_init (void (*want_poll)(void), void (*done_poll)(void))
598} 623}
599 624
600static void etp_set_max_poll_time (double nseconds) 625static void etp_set_max_poll_time (double nseconds)
601{ 626{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 627 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds; 628 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605} 630}
606 631
607static void etp_set_max_poll_reqs (unsigned int maxreqs) 632static void etp_set_max_poll_reqs (unsigned int maxreqs)
608{ 633{
612} 637}
613 638
614static void etp_set_max_idle (unsigned int nthreads) 639static void etp_set_max_idle (unsigned int nthreads)
615{ 640{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads; 642 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644}
645
646static void etp_set_idle_timeout (unsigned int seconds)
647{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619} 651}
620 652
621static void etp_set_min_parallel (unsigned int nthreads) 653static void etp_set_min_parallel (unsigned int nthreads)
622{ 654{
750void eio_set_max_idle (unsigned int nthreads) 782void eio_set_max_idle (unsigned int nthreads)
751{ 783{
752 etp_set_max_idle (nthreads); 784 etp_set_max_idle (nthreads);
753} 785}
754 786
787void eio_set_idle_timeout (unsigned int seconds)
788{
789 etp_set_idle_timeout (seconds);
790}
791
755void eio_set_min_parallel (unsigned int nthreads) 792void eio_set_min_parallel (unsigned int nthreads)
756{ 793{
757 etp_set_min_parallel (nthreads); 794 etp_set_min_parallel (nthreads);
758} 795}
759 796
807 844
808 return res; 845 return res;
809} 846}
810#endif 847#endif
811 848
812#ifndef HAVE_FUTIMES 849#ifndef HAVE_UTIMES
813 850
814# undef utimes 851# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 852# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 853
819static int 854static int
820eio__utimes (const char *filename, const struct timeval times[2]) 855eio__utimes (const char *filename, const struct timeval times[2])
821{ 856{
822 if (times) 857 if (times)
829 return utime (filename, &buf); 864 return utime (filename, &buf);
830 } 865 }
831 else 866 else
832 return utime (filename, 0); 867 return utime (filename, 0);
833} 868}
869
870#endif
871
872#ifndef HAVE_FUTIMES
873
874# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times)
834 876
835static int eio__futimes (int fd, const struct timeval tv[2]) 877static int eio__futimes (int fd, const struct timeval tv[2])
836{ 878{
837 errno = ENOSYS; 879 errno = ENOSYS;
838 return -1; 880 return -1;
867 if (!res || errno != ENOSYS) 909 if (!res || errno != ENOSYS)
868 return res; 910 return res;
869#endif 911#endif
870 912
871 /* even though we could play tricks with the flags, it's better to always 913 /* even though we could play tricks with the flags, it's better to always
872 * call fdatasync, as thta matches the expectation of it's users best */ 914 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 915 return fdatasync (fd);
874} 916}
875 917
876#if !HAVE_READAHEAD 918#if !HAVE_READAHEAD
877# undef readahead 919# undef readahead
909 951
910#if HAVE_SENDFILE 952#if HAVE_SENDFILE
911# if __linux 953# if __linux
912 res = sendfile (ofd, ifd, &offset, count); 954 res = sendfile (ofd, ifd, &offset, count);
913 955
914# elif __freebsd 956# elif __FreeBSD__
915 /* 957 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 958 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 959 * wasted on making it similar to other I/O functions.
918 */ 960 */
919 { 961 {
920 off_t sbytes; 962 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 963 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 964
923 if (res < 0 && sbytes) 965 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 966 /* freebsd' sendfile will return 0 on success */
967 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968 /* not on e.g. EIO or EPIPE - sounds broken */
969 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970 res = sbytes;
971 #endif
972
973 /* according to source inspection, this is correct, and useful behaviour */
974 if (sbytes)
975 res = sbytes;
976 }
977
978# elif defined (__APPLE__)
979
980 {
981 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983
984 /* according to the manpage, sbytes is always valid */
985 if (sbytes)
925 res = sbytes; 986 res = sbytes;
926 } 987 }
927 988
928# elif __hpux 989# elif __hpux
929 res = sendfile (ofd, ifd, offset, count, 0, 0); 990 res = sendfile (ofd, ifd, offset, count, 0, 0);
943 if (res < 0 && sbytes) 1004 if (res < 0 && sbytes)
944 res = sbytes; 1005 res = sbytes;
945 } 1006 }
946 1007
947# endif 1008# endif
1009
1010#elif defined (_WIN32)
1011
1012 /* does not work, just for documentation of what would need to be done */
1013 {
1014 HANDLE h = TO_SOCKET (ifd);
1015 SetFilePointer (h, offset, 0, FILE_BEGIN);
1016 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017 }
1018
948#else 1019#else
949 res = -1; 1020 res = -1;
950 errno = ENOSYS; 1021 errno = ENOSYS;
951#endif 1022#endif
952 1023
953 if (res < 0 1024 if (res < 0
954 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1025 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026 /* BSDs */
1027#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028 || errno == ENOTSUP
1029#endif
1030 || errno == EOPNOTSUPP /* BSDs */
955#if __solaris 1031#if __solaris
956 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1032 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
957#endif 1033#endif
958 ) 1034 )
959 ) 1035 )
993} 1069}
994 1070
995static signed char 1071static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1072eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{ 1073{
998 return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ 1074 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
999 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1075 : a->inode < b->inode ? -1
1076 : a->inode > b->inode ? 1
1077 : 0;
1000} 1078}
1001 1079
1002#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1080#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1003 1081
1004#define EIO_QSORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1082#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1005#define EIO_QSORT_SKIP 60 /* when to skip qsort completely */ 1083#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1006 1084
1007static void 1085static void
1008eio_dent_sort (eio_dirent *dents, int size) 1086eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1009{ 1087{
1088 unsigned char bits [9 + sizeof (ino_t) * 8];
1089 unsigned char *bit = bits;
1090
1091 assert (CHAR_BIT == 8);
1092 assert (sizeof (eio_dirent) * 8 < 256);
1093 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1094 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1095
1096 if (size <= EIO_SORT_FAST)
1097 return;
1098
1099 /* first prepare an array of bits to test in our radix sort */
1100 /* try to take endianness into account, as well as differences in ino_t sizes */
1101 /* inode_bits must contain all inodes ORed together */
1102 /* which is used to skip bits that are 0 everywhere, which is very common */
1103 {
1104 ino_t endianness;
1010 int i, j; 1105 int i, j;
1011 1106
1107 /* we store the byte offset of byte n into byte n of "endianness" */
1108 for (i = 0; i < sizeof (ino_t); ++i)
1109 ((unsigned char *)&endianness)[i] = i;
1110
1111 *bit++ = 0;
1112
1113 for (i = 0; i < sizeof (ino_t); ++i)
1114 {
1115 /* shifting off the byte offsets out of "endianness" */
1116 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117 endianness >>= 8;
1118
1119 for (j = 0; j < 8; ++j)
1120 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1121 *bit++ = offs + j;
1122 }
1123
1124 for (j = 0; j < 8; ++j)
1125 if (score_bits & (1 << j))
1126 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1127 }
1128
1129 /* now actually do the sorting (a variant of MSD radix sort) */
1130 {
1131 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1132 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1133 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1134 int stk_idx = 0;
1135
1136 base_stk [stk_idx] = dents;
1137 end_stk [stk_idx] = dents + size;
1138 bit_stk [stk_idx] = bit - 1;
1139
1140 do
1141 {
1142 base = base_stk [stk_idx];
1143 end = end_stk [stk_idx];
1144 bit = bit_stk [stk_idx];
1145
1146 for (;;)
1147 {
1148 unsigned char O = *bit >> 3;
1149 unsigned char M = 1 << (*bit & 7);
1150
1151 eio_dirent *a = base;
1152 eio_dirent *b = end;
1153
1154 if (b - a < EIO_SORT_CUTOFF)
1155 break;
1156
1157 /* now bit-partition the array on the bit */
1158 /* this ugly asymmetric loop seems to perform much better than typical */
1159 /* partition algos found in the literature */
1160 do
1161 if (!(((unsigned char *)a)[O] & M))
1162 ++a;
1163 else if (!(((unsigned char *)--b)[O] & M))
1164 {
1165 eio_dirent tmp = *a; *a = *b; *b = tmp;
1166 ++a;
1167 }
1168 while (b > a);
1169
1170 /* next bit, or stop, if no bits left in this path */
1171 if (!*--bit)
1172 break;
1173
1174 base_stk [stk_idx] = a;
1175 end_stk [stk_idx] = end;
1176 bit_stk [stk_idx] = bit;
1177 ++stk_idx;
1178
1179 end = a;
1180 }
1181 }
1182 while (stk_idx--);
1183 }
1184}
1185
1186static void
1187eio_dent_insertion_sort (eio_dirent *dents, int size)
1188{
1189 /* first move the smallest element to the front, to act as a sentinel */
1190 {
1191 int i;
1192 eio_dirent *min = dents;
1193
1194 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1195 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1196 if (EIO_DENT_CMP (dents [i], <, *min))
1197 min = &dents [i];
1198
1199 /* swap elements 0 and j (minimum) */
1200 {
1201 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1202 }
1203 }
1204
1205 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1206 {
1207 eio_dirent *i, *j;
1208
1209 for (i = dents + 1; i < dents + size; ++i)
1210 {
1211 eio_dirent value = *i;
1212
1213 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1214 j [1] = j [0];
1215
1216 j [1] = value;
1217 }
1218 }
1219}
1220
1221static void
1222eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1223{
1012 if (size <= 1) 1224 if (size <= 1)
1013 return; /* our insertion sort relies on size > 0 */ 1225 return; /* our insertion sort relies on size > 0 */
1014 1226
1015 if (size > EIO_QSORT_SKIP) /* skip quicksort for small directories */ 1227 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1016 { 1228 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1017 /* first, use quicksort */ 1229 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1018 /* should be good for 2**31 entries */
1019 struct rng { int l, r; } rng [32];
1020 1230
1021 i = 0;
1022 rng[0].l = 0;
1023 rng[0].r = size;
1024
1025 while (expect_true (i >= 0))
1026 {
1027 int L = rng [i].l;
1028 int R = rng [i].r - 1;
1029
1030 if (expect_false (L + EIO_QSORT_CUTOFF < R))
1031 {
1032 eio_dirent piv = dents [L];
1033
1034 while (L < R)
1035 {
1036 while (EIO_DENT_CMP (dents [R], >=, piv) && L < R)
1037 --R;
1038
1039 if (L < R)
1040 dents [L++] = dents [R];
1041
1042 while (EIO_DENT_CMP (dents [L], <=, piv) && L < R)
1043 ++L;
1044
1045 if (L < R)
1046 dents [R--] = dents [L];
1047 }
1048
1049 dents [L] = piv;
1050
1051 ++i;
1052 rng [i].l = L + 1;
1053 rng [i].r = rng [i - 1].r;
1054 rng [i - 1].r = L;
1055
1056 if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1057 {
1058 struct rng t;
1059
1060 t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1061 }
1062 }
1063 else
1064 --i;
1065 }
1066 }
1067
1068 /* use an insertion sort after qsort, or for small arrays */ 1231 /* use an insertion sort at the end, or for small arrays, */
1069 /* first move the smallest element to the front, to act as a sentinel */ 1232 /* as insertion sort is more efficient for small partitions */
1070 { 1233 eio_dent_insertion_sort (dents, size);
1071 int min = 0;
1072
1073 for (i = size > EIO_QSORT_SKIP ? EIO_QSORT_CUTOFF : size; --i; )
1074 if (EIO_DENT_CMP (dents [i], <, dents [min]))
1075 min = i;
1076
1077 /* swap elements 0 and j (minimum) */
1078 {
1079 eio_dirent tmp = dents [0]; dents [0] = dents [min]; dents [min] = tmp;
1080 }
1081 }
1082
1083 /* then do standard insertion sort */
1084 for (i = 1; i < size; ++i)
1085 {
1086 eio_dirent value = dents [i];
1087
1088 for (j = i - 1; EIO_DENT_CMP (dents [j], >, value); --j)
1089 dents [j + 1] = dents [j];
1090
1091 dents [j + 1] = value;
1092 }
1093} 1234}
1094 1235
1095/* read a full directory */ 1236/* read a full directory */
1096static void 1237static void
1097eio__scandir (eio_req *req, etp_worker *self) 1238eio__scandir (eio_req *req, etp_worker *self)
1103 int namesoffs = 0; 1244 int namesoffs = 0;
1104 int flags = req->int1; 1245 int flags = req->int1;
1105 eio_dirent *dents = 0; 1246 eio_dirent *dents = 0;
1106 int dentalloc = 128; 1247 int dentalloc = 128;
1107 int dentoffs = 0; 1248 int dentoffs = 0;
1249 ino_t inode_bits = 0;
1108 1250
1109 req->result = -1; 1251 req->result = -1;
1110 1252
1111 if (!(flags & EIO_READDIR_DENTS)) 1253 if (!(flags & EIO_READDIR_DENTS))
1112 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1254 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1113 1255
1114 X_LOCK (wrklock); 1256 X_LOCK (wrklock);
1115 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1257 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1116 self->dirp = dirp = opendir (req->ptr1); 1258 self->dirp = dirp = opendir (req->ptr1);
1259
1117 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1260 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1118 req->ptr1 = names = malloc (namesalloc);
1119 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1261 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262 req->ptr2 = names = malloc (namesalloc);
1120 X_UNLOCK (wrklock); 1263 X_UNLOCK (wrklock);
1121 1264
1122 if (dirp && names && (!flags || dents)) 1265 if (dirp && names && (!flags || dents))
1123 for (;;) 1266 for (;;)
1124 { 1267 {
1132 1275
1133 /* sort etc. */ 1276 /* sort etc. */
1134 req->int1 = flags; 1277 req->int1 = flags;
1135 req->result = dentoffs; 1278 req->result = dentoffs;
1136 1279
1137 if (dents)
1138 {
1139 eio_dirent *ent = dents + dentoffs;
1140
1141 while (ent > dents)
1142 (--ent)->name = names + (size_t)ent->name;
1143 }
1144
1145 if (flags & EIO_READDIR_STAT_ORDER 1280 if (flags & EIO_READDIR_STAT_ORDER)
1146 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1281 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1147 eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1148 else if (flags & EIO_READDIR_DIRS_FIRST) 1282 else if (flags & EIO_READDIR_DIRS_FIRST)
1283 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1284 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1285 else
1149 { 1286 {
1150 /* in this case, all is known, and we just put dirs first and sort them */ 1287 /* in this case, all is known, and we just put dirs first and sort them */
1151 eio_dirent *ent = dents + dentoffs; 1288 eio_dirent *oth = dents + dentoffs;
1152 eio_dirent *dir = dents; 1289 eio_dirent *dir = dents;
1153 1290
1154 /* now move dirs to the front, and non-dirs to the back */ 1291 /* now partition dirs to the front, and non-dirs to the back */
1155 /* by walking from both sides and swapping if necessary */ 1292 /* by walking from both sides and swapping if necessary */
1156 while (ent > dir) 1293 while (oth > dir)
1157 { 1294 {
1158 if (dir->type == DT_DIR) 1295 if (dir->type == EIO_DT_DIR)
1159 ++dir; 1296 ++dir;
1160 else
1161 {
1162 --ent;
1163
1164 if (ent->type == DT_DIR) 1297 else if ((--oth)->type == EIO_DT_DIR)
1165 { 1298 {
1166 eio_dirent tmp = *dir; 1299 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1167 *dir = *ent;
1168 *ent = tmp;
1169 1300
1170 ++dir; 1301 ++dir;
1171 } 1302 }
1172 } 1303 }
1304
1305 /* now sort the dirs only (dirs all have the same score) */
1306 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1173 } 1307 }
1174
1175 /* now sort the dirs only */
1176 eio_dent_sort (dents, dir - dents);
1177 }
1178
1179 /* only provide the names array unless DENTS is specified */
1180 if (!(flags & EIO_READDIR_DENTS))
1181 {
1182 X_LOCK (wrklock);
1183 assert (!dents);
1184 req->ptr1 = 0;
1185 req->ptr2 = names;
1186 X_UNLOCK (wrklock);
1187 }
1188 1308
1189 break; 1309 break;
1190 } 1310 }
1191 1311
1192 /* now add the entry to our list(s) */ 1312 /* now add the entry to our list(s) */
1199 1319
1200 while (expect_false (namesoffs + len > namesalloc)) 1320 while (expect_false (namesoffs + len > namesalloc))
1201 { 1321 {
1202 namesalloc *= 2; 1322 namesalloc *= 2;
1203 X_LOCK (wrklock); 1323 X_LOCK (wrklock);
1204 req->ptr1 = names = realloc (names, namesalloc); 1324 req->ptr2 = names = realloc (names, namesalloc);
1205 X_UNLOCK (wrklock); 1325 X_UNLOCK (wrklock);
1206 1326
1207 if (!names) 1327 if (!names)
1208 break; 1328 break;
1209 } 1329 }
1216 1336
1217 if (expect_false (dentoffs == dentalloc)) 1337 if (expect_false (dentoffs == dentalloc))
1218 { 1338 {
1219 dentalloc *= 2; 1339 dentalloc *= 2;
1220 X_LOCK (wrklock); 1340 X_LOCK (wrklock);
1221 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1341 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1222 X_UNLOCK (wrklock); 1342 X_UNLOCK (wrklock);
1223 1343
1224 if (!dents) 1344 if (!dents)
1225 break; 1345 break;
1226 } 1346 }
1227 1347
1228 ent = dents + dentoffs; 1348 ent = dents + dentoffs;
1229 1349
1230 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1350 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1231 ent->namelen = len - 1; 1351 ent->namelen = len - 1;
1232 ent->inode = D_INO (entp); 1352 ent->inode = D_INO (entp);
1353
1354 inode_bits |= ent->inode;
1233 1355
1234 switch (D_TYPE (entp)) 1356 switch (D_TYPE (entp))
1235 { 1357 {
1236 default: 1358 default:
1237 ent->type = EIO_DT_UNKNOWN; 1359 ent->type = EIO_DT_UNKNOWN;
1280 #ifdef DT_WHT 1402 #ifdef DT_WHT
1281 case DT_WHT: ent->type = EIO_DT_WHT; break; 1403 case DT_WHT: ent->type = EIO_DT_WHT; break;
1282 #endif 1404 #endif
1283 } 1405 }
1284 1406
1285 ent->score = 0; 1407 ent->score = 7;
1286 1408
1287 if (flags & EIO_READDIR_DIRS_FIRST) 1409 if (flags & EIO_READDIR_DIRS_FIRST)
1288 { 1410 {
1289 if (ent->type == EIO_DT_UNKNOWN) 1411 if (ent->type == EIO_DT_UNKNOWN)
1290 { 1412 {
1291 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1413 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1292 ent->score = 98; 1414 ent->score = 1;
1293 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1415 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1294 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ 1416 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1295 } 1417 }
1296 else if (ent->type == EIO_DT_DIR) 1418 else if (ent->type == EIO_DT_DIR)
1297 ent->score = 100; 1419 ent->score = 0;
1298 } 1420 }
1299 } 1421 }
1300 1422
1301 namesoffs += len; 1423 namesoffs += len;
1302 ++dentoffs; 1424 ++dentoffs;
1308 break; 1430 break;
1309 } 1431 }
1310 } 1432 }
1311} 1433}
1312 1434
1313#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1314# undef msync
1315# define msync(a,b,c) ((errno = ENOSYS), -1)
1316#endif
1317
1318int
1319eio__mtouch (void *mem, size_t len, int flags)
1320{
1321 intptr_t addr = (intptr_t)mem;
1322 intptr_t end = addr + len;
1323#ifdef PAGESIZE 1435#ifdef PAGESIZE
1324 const intptr_t page = PAGESIZE; 1436# define eio_pagesize() PAGESIZE
1325#else 1437#else
1438static intptr_t
1439eio_pagesize (void)
1440{
1326 static intptr_t page; 1441 static intptr_t page;
1327 1442
1328 if (!page) 1443 if (!page)
1329 page = sysconf (_SC_PAGESIZE); 1444 page = sysconf (_SC_PAGESIZE);
1445
1446 return page;
1447}
1448#endif
1449
1450static void
1451eio_page_align (void **addr, size_t *length)
1452{
1453 intptr_t mask = eio_pagesize () - 1;
1454
1455 /* round down addr */
1456 intptr_t adj = mask & (intptr_t)*addr;
1457
1458 *addr = (void *)((intptr_t)*addr - adj);
1459 *length += adj;
1460
1461 /* round up length */
1462 *length = (*length + mask) & ~mask;
1463}
1464
1465#if !_POSIX_MEMLOCK
1466# define eio__mlockall(a) ((errno = ENOSYS), -1)
1467#else
1468
1469static int
1470eio__mlockall (int flags)
1471{
1472 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1473 extern int mallopt (int, int);
1474 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1330#endif 1475 #endif
1331 1476
1332 addr &= ~(page - 1); /* assume page size is always a power of two */ 1477 if (EIO_MCL_CURRENT != MCL_CURRENT
1478 || EIO_MCL_FUTURE != MCL_FUTURE)
1479 {
1480 flags = 0
1481 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483 }
1333 1484
1485 return mlockall (flags);
1486}
1487#endif
1488
1489#if !_POSIX_MEMLOCK_RANGE
1490# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491#else
1492
1493static int
1494eio__mlock (void *addr, size_t length)
1495{
1496 eio_page_align (&addr, &length);
1497
1498 return mlock (addr, length);
1499}
1500
1501#endif
1502
1503#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1504# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1505#else
1506
1507int
1508eio__msync (void *mem, size_t len, int flags)
1509{
1510 eio_page_align (&mem, &len);
1511
1512 if (EIO_MS_ASYNC != MS_SYNC
1513 || EIO_MS_INVALIDATE != MS_INVALIDATE
1514 || EIO_MS_SYNC != MS_SYNC)
1515 {
1516 flags = 0
1517 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1518 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1519 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1520 }
1521
1522 return msync (mem, len, flags);
1523}
1524
1525#endif
1526
1527int
1528eio__mtouch (eio_req *req)
1529{
1530 void *mem = req->ptr2;
1531 size_t len = req->size;
1532 int flags = req->int1;
1533
1534 eio_page_align (&mem, &len);
1535
1536 {
1537 intptr_t addr = (intptr_t)mem;
1538 intptr_t end = addr + len;
1539 intptr_t page = eio_pagesize ();
1540
1334 if (addr < end) 1541 if (addr < end)
1335 if (flags) /* modify */ 1542 if (flags & EIO_MT_MODIFY) /* modify */
1336 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1543 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1337 else 1544 else
1338 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1545 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1546 }
1339 1547
1340 return 0; 1548 return 0;
1341} 1549}
1342 1550
1343/*****************************************************************************/ 1551/*****************************************************************************/
1377 if (req) 1585 if (req)
1378 break; 1586 break;
1379 1587
1380 ++idle; 1588 ++idle;
1381 1589
1382 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1590 ts.tv_sec = time (0) + idle_timeout;
1383 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1591 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1384 { 1592 {
1385 if (idle > max_idle) 1593 if (idle > max_idle)
1386 { 1594 {
1387 --idle; 1595 --idle;
1466 return 0; \ 1674 return 0; \
1467 } 1675 }
1468 1676
1469static void eio_execute (etp_worker *self, eio_req *req) 1677static void eio_execute (etp_worker *self, eio_req *req)
1470{ 1678{
1471 errno = 0;
1472
1473 switch (req->type) 1679 switch (req->type)
1474 { 1680 {
1475 case EIO_READ: ALLOC (req->size); 1681 case EIO_READ: ALLOC (req->size);
1476 req->result = req->offs >= 0 1682 req->result = req->offs >= 0
1477 ? pread (req->int1, req->ptr2, req->size, req->offs) 1683 ? pread (req->int1, req->ptr2, req->size, req->offs)
1488 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1694 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1489 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1695 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1490 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1696 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1491 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1697 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1492 1698
1699 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1700 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1701 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1702 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1703
1493 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1704 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1494 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1705 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1495 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1706 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1496 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1707 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1497 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1708 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1504 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1715 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1505 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1716 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1506 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1717 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1507 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1718 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1508 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1719 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1509 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1720 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1510 1721
1511 case EIO_READLINK: ALLOC (NAME_MAX); 1722 case EIO_READLINK: ALLOC (PATH_MAX);
1512 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1723 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1513 1724
1514 case EIO_SYNC: req->result = 0; sync (); break; 1725 case EIO_SYNC: req->result = 0; sync (); break;
1515 case EIO_FSYNC: req->result = fsync (req->int1); break; 1726 case EIO_FSYNC: req->result = fsync (req->int1); break;
1516 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1727 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1517 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1728 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1729 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1518 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1730 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1731 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1519 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1732 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1520 1733
1521 case EIO_READDIR: eio__scandir (req, self); break; 1734 case EIO_READDIR: eio__scandir (req, self); break;
1522 1735
1523 case EIO_BUSY: 1736 case EIO_BUSY:
1524#ifdef _WIN32 1737#ifdef _WIN32
1525 Sleep (req->nv1 * 1000.); 1738 Sleep (req->nv1 * 1e3);
1526#else 1739#else
1527 { 1740 {
1528 struct timeval tv; 1741 struct timeval tv;
1529 1742
1530 tv.tv_sec = req->nv1; 1743 tv.tv_sec = req->nv1;
1531 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1744 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1532 1745
1533 req->result = select (0, 0, 0, 0, &tv); 1746 req->result = select (0, 0, 0, 0, &tv);
1534 } 1747 }
1535#endif 1748#endif
1536 break; 1749 break;
1551 times = tv; 1764 times = tv;
1552 } 1765 }
1553 else 1766 else
1554 times = 0; 1767 times = 0;
1555 1768
1556
1557 req->result = req->type == EIO_FUTIME 1769 req->result = req->type == EIO_FUTIME
1558 ? futimes (req->int1, times) 1770 ? futimes (req->int1, times)
1559 : utimes (req->ptr1, times); 1771 : utimes (req->ptr1, times);
1560 } 1772 }
1561 break; 1773 break;
1570 case EIO_CUSTOM: 1782 case EIO_CUSTOM:
1571 ((void (*)(eio_req *))req->feed) (req); 1783 ((void (*)(eio_req *))req->feed) (req);
1572 break; 1784 break;
1573 1785
1574 default: 1786 default:
1787 errno = ENOSYS;
1575 req->result = -1; 1788 req->result = -1;
1576 break; 1789 break;
1577 } 1790 }
1578 1791
1579 req->errorno = errno; 1792 req->errorno = errno;
1609eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1822eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1610{ 1823{
1611 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1824 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1612} 1825}
1613 1826
1827eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1828{
1829 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1830}
1831
1832eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1833{
1834 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1835}
1836
1614eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1837eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1615{ 1838{
1616 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1839 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1617} 1840}
1618 1841
1642} 1865}
1643 1866
1644eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1867eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1645{ 1868{
1646 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1869 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1870}
1871
1872eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1873{
1874 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1647} 1875}
1648 1876
1649eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1877eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1650{ 1878{
1651 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1879 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1725eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1953eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1726{ 1954{
1727 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1955 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1728} 1956}
1729 1957
1958eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1959{
1960 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1961}
1962
1730eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1963eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1731{ 1964{
1732 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1965 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1733} 1966}
1734 1967
1742 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1975 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1743} 1976}
1744 1977
1745eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1978eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1746{ 1979{
1747 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1980 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1748} 1981}
1749 1982
1750static eio_req * 1983static eio_req *
1751eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1984eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1752{ 1985{

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines