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 1.38 by root, Fri Jun 12 20:01:42 2009 UTC vs.
Revision 1.70 by root, Fri Jun 10 06:55:10 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
900 942
901/* sendfile always needs emulation */ 943/* sendfile always needs emulation */
902static ssize_t 944static ssize_t
903eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
904{ 946{
947 ssize_t written = 0;
905 ssize_t res; 948 ssize_t res;
906 949
907 if (!count) 950 if (!count)
908 return 0; 951 return 0;
909 952
953 for (;;)
954 {
910#if HAVE_SENDFILE 955#if HAVE_SENDFILE
911# if __linux 956# if __linux
957 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 958 res = sendfile (ofd, ifd, &soffset, count);
913 959
914# elif __freebsd 960# elif __FreeBSD__
915 /* 961 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 962 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 963 * wasted on making it similar to other I/O functions.
918 */ 964 */
919 {
920 off_t sbytes; 965 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 966 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 967
923 if (res < 0 && sbytes) 968 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 969 /* freebsd' sendfile will return 0 on success */
970 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
971 /* not on e.g. EIO or EPIPE - sounds broken */
972 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925 res = sbytes; 973 res = sbytes;
926 } 974 #endif
975
976 /* according to source inspection, this is correct, and useful behaviour */
977 if (sbytes)
978 res = sbytes;
979
980# elif defined (__APPLE__)
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)
986 res = sbytes;
927 987
928# elif __hpux 988# elif __hpux
929 res = sendfile (ofd, ifd, offset, count, 0, 0); 989 res = sendfile (ofd, ifd, offset, count, 0, 0);
930 990
931# elif __solaris 991# elif __solaris
932 {
933 struct sendfilevec vec; 992 struct sendfilevec vec;
934 size_t sbytes; 993 size_t sbytes;
935 994
936 vec.sfv_fd = ifd; 995 vec.sfv_fd = ifd;
937 vec.sfv_flag = 0; 996 vec.sfv_flag = 0;
938 vec.sfv_off = offset; 997 vec.sfv_off = offset;
939 vec.sfv_len = count; 998 vec.sfv_len = count;
940 999
941 res = sendfilev (ofd, &vec, 1, &sbytes); 1000 res = sendfilev (ofd, &vec, 1, &sbytes);
942 1001
943 if (res < 0 && sbytes) 1002 if (res < 0 && sbytes)
944 res = sbytes; 1003 res = sbytes;
945 }
946 1004
947# endif 1005# endif
1006
1007#elif defined (_WIN32)
1008 /* does not work, just for documentation of what would need to be done */
1009 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1010 /* libeio guarantees that the file offset does not change, and windows */
1011 /* has no way to get an independent handle to the same file description */
1012 HANDLE h = TO_SOCKET (ifd);
1013 SetFilePointer (h, offset, 0, FILE_BEGIN);
1014 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1015
948#else 1016#else
949 res = -1; 1017 res = -1;
950 errno = ENOSYS; 1018 errno = ENOSYS;
951#endif 1019#endif
952 1020
1021 /* we assume sendfile can copy at least 128mb in one go */
1022 if (res <= 128 * 1024 * 1024)
1023 {
1024 if (res > 0)
1025 written += res;
1026
1027 if (written)
1028 return written;
1029
1030 break;
1031 }
1032 else
1033 {
1034 /* if we requested more, then probably the kernel was lazy */
1035 written += res;
1036 offset += res;
1037 count -= res;
1038
1039 if (!count)
1040 return written;
1041 }
1042 }
1043
953 if (res < 0 1044 if (res < 0
954 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1045 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1046 /* BSDs */
1047#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1048 || errno == ENOTSUP
1049#endif
1050 || errno == EOPNOTSUPP /* BSDs */
955#if __solaris 1051#if __solaris
956 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1052 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
957#endif 1053#endif
958 ) 1054 )
959 ) 1055 )
993} 1089}
994 1090
995static signed char 1091static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1092eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{ 1093{
998 return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ 1094 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; 1095 : a->inode < b->inode ? -1
1096 : a->inode > b->inode ? 1
1097 : 0;
1000} 1098}
1001 1099
1100#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1101
1002#define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */ 1102#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1103#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1003 1104
1004static void 1105static void
1005eio_dent_sort (eio_dirent *dents, int size) 1106eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1006{ 1107{
1108 unsigned char bits [9 + sizeof (ino_t) * 8];
1109 unsigned char *bit = bits;
1110
1111 assert (CHAR_BIT == 8);
1112 assert (sizeof (eio_dirent) * 8 < 256);
1113 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1114 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1115
1116 if (size <= EIO_SORT_FAST)
1117 return;
1118
1119 /* first prepare an array of bits to test in our radix sort */
1120 /* try to take endianness into account, as well as differences in ino_t sizes */
1121 /* inode_bits must contain all inodes ORed together */
1122 /* which is used to skip bits that are 0 everywhere, which is very common */
1123 {
1124 ino_t endianness;
1007 int i, j; 1125 int i, j;
1008 1126
1009 if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */ 1127 /* we store the byte offset of byte n into byte n of "endianness" */
1128 for (i = 0; i < sizeof (ino_t); ++i)
1129 ((unsigned char *)&endianness)[i] = i;
1130
1131 *bit++ = 0;
1132
1133 for (i = 0; i < sizeof (ino_t); ++i)
1134 {
1135 /* shifting off the byte offsets out of "endianness" */
1136 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1137 endianness >>= 8;
1138
1139 for (j = 0; j < 8; ++j)
1140 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1141 *bit++ = offs + j;
1142 }
1143
1144 for (j = 0; j < 8; ++j)
1145 if (score_bits & (1 << j))
1146 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1147 }
1148
1149 /* now actually do the sorting (a variant of MSD radix sort) */
1150 {
1151 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1152 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1153 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1154 int stk_idx = 0;
1155
1156 base_stk [stk_idx] = dents;
1157 end_stk [stk_idx] = dents + size;
1158 bit_stk [stk_idx] = bit - 1;
1159
1160 do
1161 {
1162 base = base_stk [stk_idx];
1163 end = end_stk [stk_idx];
1164 bit = bit_stk [stk_idx];
1165
1166 for (;;)
1167 {
1168 unsigned char O = *bit >> 3;
1169 unsigned char M = 1 << (*bit & 7);
1170
1171 eio_dirent *a = base;
1172 eio_dirent *b = end;
1173
1174 if (b - a < EIO_SORT_CUTOFF)
1175 break;
1176
1177 /* now bit-partition the array on the bit */
1178 /* this ugly asymmetric loop seems to perform much better than typical */
1179 /* partition algos found in the literature */
1180 do
1181 if (!(((unsigned char *)a)[O] & M))
1182 ++a;
1183 else if (!(((unsigned char *)--b)[O] & M))
1184 {
1185 eio_dirent tmp = *a; *a = *b; *b = tmp;
1186 ++a;
1187 }
1188 while (b > a);
1189
1190 /* next bit, or stop, if no bits left in this path */
1191 if (!*--bit)
1192 break;
1193
1194 base_stk [stk_idx] = a;
1195 end_stk [stk_idx] = end;
1196 bit_stk [stk_idx] = bit;
1197 ++stk_idx;
1198
1199 end = a;
1200 }
1201 }
1202 while (stk_idx--);
1203 }
1204}
1205
1206static void
1207eio_dent_insertion_sort (eio_dirent *dents, int size)
1208{
1209 /* first move the smallest element to the front, to act as a sentinel */
1210 {
1211 int i;
1212 eio_dirent *min = dents;
1213
1214 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1215 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1216 if (EIO_DENT_CMP (dents [i], <, *min))
1217 min = &dents [i];
1218
1219 /* swap elements 0 and j (minimum) */
1010 { 1220 {
1011 /* first, use quicksort */ 1221 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1012 /* should be good for 2**31 entries */ 1222 }
1013 struct rng { int l, r; } rng [32]; 1223 }
1014 1224
1015 i = 0; 1225 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1016 rng[0].l = 0; 1226 {
1017 rng[0].r = size; 1227 eio_dirent *i, *j;
1018 1228
1019 while (expect_true (i >= 0)) 1229 for (i = dents + 1; i < dents + size; ++i)
1020 { 1230 {
1021 int L = rng [i].l; 1231 eio_dirent value = *i;
1022 int R = rng [i].r - 1;
1023 1232
1024 if (expect_false (L + EIO_QSORT_CUTOFF < R)) 1233 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1025 { 1234 j [1] = j [0];
1026 eio_dirent piv = dents [L];
1027 1235
1028 while (L < R) 1236 j [1] = value;
1029 {
1030 while (eio_dent_cmp (&dents [R], &piv) >= 0 && L < R)
1031 --R;
1032
1033 if (L < R)
1034 dents [L++] = dents [R];
1035
1036 while (eio_dent_cmp (&dents [L], &piv) <= 0 && L < R)
1037 ++L;
1038
1039 if (L < R)
1040 dents [R--] = dents [L];
1041 }
1042
1043 dents [L] = piv;
1044
1045 ++i;
1046 rng [i].l = L + 1;
1047 rng [i].r = rng [i - 1].r;
1048 rng [i - 1].r = L;
1049
1050 if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1051 {
1052 struct rng t;
1053
1054 t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1055 }
1056 }
1057 else
1058 --i;
1059 } 1237 }
1060 } 1238 }
1239}
1061 1240
1062 /* use a simple insertion sort at the end */ 1241static void
1063 for (i = 1; i < size; ++i) 1242eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1064 { 1243{
1065 eio_dirent value = dents [i]; 1244 if (size <= 1)
1245 return; /* our insertion sort relies on size > 0 */
1066 1246
1067 for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j) 1247 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1068 dents [j + 1] = dents [j]; 1248 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1249 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1069 1250
1070 dents [j + 1] = value; 1251 /* use an insertion sort at the end, or for small arrays, */
1071 } 1252 /* as insertion sort is more efficient for small partitions */
1253 eio_dent_insertion_sort (dents, size);
1072} 1254}
1073 1255
1074/* read a full directory */ 1256/* read a full directory */
1075static void 1257static void
1076eio__scandir (eio_req *req, etp_worker *self) 1258eio__scandir (eio_req *req, etp_worker *self)
1082 int namesoffs = 0; 1264 int namesoffs = 0;
1083 int flags = req->int1; 1265 int flags = req->int1;
1084 eio_dirent *dents = 0; 1266 eio_dirent *dents = 0;
1085 int dentalloc = 128; 1267 int dentalloc = 128;
1086 int dentoffs = 0; 1268 int dentoffs = 0;
1269 ino_t inode_bits = 0;
1087 1270
1088 req->result = -1; 1271 req->result = -1;
1089 1272
1090 if (!(flags & EIO_READDIR_DENTS)) 1273 if (!(flags & EIO_READDIR_DENTS))
1091 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1274 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1092 1275
1093 X_LOCK (wrklock); 1276 X_LOCK (wrklock);
1094 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1277 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1095 self->dirp = dirp = opendir (req->ptr1); 1278 self->dirp = dirp = opendir (req->ptr1);
1279
1096 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1280 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1097 req->ptr1 = names = malloc (namesalloc);
1098 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1281 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1282 req->ptr2 = names = malloc (namesalloc);
1099 X_UNLOCK (wrklock); 1283 X_UNLOCK (wrklock);
1100 1284
1101 if (dirp && names && (!flags || dents)) 1285 if (dirp && names && (!flags || dents))
1102 for (;;) 1286 for (;;)
1103 { 1287 {
1111 1295
1112 /* sort etc. */ 1296 /* sort etc. */
1113 req->int1 = flags; 1297 req->int1 = flags;
1114 req->result = dentoffs; 1298 req->result = dentoffs;
1115 1299
1116 if (dents)
1117 {
1118 eio_dirent *ent = dents + dentoffs;
1119
1120 while (ent > dents)
1121 (--ent)->name = names + (size_t)ent->name;
1122 }
1123
1124 if (flags & EIO_READDIR_STAT_ORDER 1300 if (flags & EIO_READDIR_STAT_ORDER)
1125 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1301 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1126 eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1127 else if (flags & EIO_READDIR_DIRS_FIRST) 1302 else if (flags & EIO_READDIR_DIRS_FIRST)
1303 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1304 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1305 else
1128 { 1306 {
1129 /* in this case, all is known, and we just put dirs first and sort them */ 1307 /* in this case, all is known, and we just put dirs first and sort them */
1130 eio_dirent *ent = dents + dentoffs; 1308 eio_dirent *oth = dents + dentoffs;
1131 eio_dirent *dir = dents; 1309 eio_dirent *dir = dents;
1132 1310
1133 /* now move dirs to the front, and non-dirs to the back */ 1311 /* now partition dirs to the front, and non-dirs to the back */
1134 /* by walking from both sides and swapping if necessary */ 1312 /* by walking from both sides and swapping if necessary */
1135 while (ent > dir) 1313 while (oth > dir)
1136 { 1314 {
1137 if (dir->type == DT_DIR) 1315 if (dir->type == EIO_DT_DIR)
1138 ++dir; 1316 ++dir;
1139 else
1140 {
1141 --ent;
1142
1143 if (ent->type == DT_DIR) 1317 else if ((--oth)->type == EIO_DT_DIR)
1144 { 1318 {
1145 eio_dirent tmp = *dir; 1319 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1146 *dir = *ent;
1147 *ent = tmp;
1148 1320
1149 ++dir; 1321 ++dir;
1150 } 1322 }
1151 } 1323 }
1324
1325 /* now sort the dirs only (dirs all have the same score) */
1326 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1152 } 1327 }
1153
1154 /* now sort the dirs only */
1155 eio_dent_sort (dents, dir - dents);
1156 }
1157
1158 /* only provide the names array unless DENTS is specified */
1159 if (!(flags & EIO_READDIR_DENTS))
1160 {
1161 X_LOCK (wrklock);
1162 assert (!dents);
1163 req->ptr1 = 0;
1164 req->ptr2 = names;
1165 X_UNLOCK (wrklock);
1166 }
1167 1328
1168 break; 1329 break;
1169 } 1330 }
1170 1331
1171 /* now add the entry to our list(s) */ 1332 /* now add the entry to our list(s) */
1178 1339
1179 while (expect_false (namesoffs + len > namesalloc)) 1340 while (expect_false (namesoffs + len > namesalloc))
1180 { 1341 {
1181 namesalloc *= 2; 1342 namesalloc *= 2;
1182 X_LOCK (wrklock); 1343 X_LOCK (wrklock);
1183 req->ptr1 = names = realloc (names, namesalloc); 1344 req->ptr2 = names = realloc (names, namesalloc);
1184 X_UNLOCK (wrklock); 1345 X_UNLOCK (wrklock);
1185 1346
1186 if (!names) 1347 if (!names)
1187 break; 1348 break;
1188 } 1349 }
1195 1356
1196 if (expect_false (dentoffs == dentalloc)) 1357 if (expect_false (dentoffs == dentalloc))
1197 { 1358 {
1198 dentalloc *= 2; 1359 dentalloc *= 2;
1199 X_LOCK (wrklock); 1360 X_LOCK (wrklock);
1200 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1361 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1201 X_UNLOCK (wrklock); 1362 X_UNLOCK (wrklock);
1202 1363
1203 if (!dents) 1364 if (!dents)
1204 break; 1365 break;
1205 } 1366 }
1206 1367
1207 ent = dents + dentoffs; 1368 ent = dents + dentoffs;
1208 1369
1209 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1370 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1210 ent->namelen = len - 1; 1371 ent->namelen = len - 1;
1211 ent->inode = D_INO (entp); 1372 ent->inode = D_INO (entp);
1373
1374 inode_bits |= ent->inode;
1212 1375
1213 switch (D_TYPE (entp)) 1376 switch (D_TYPE (entp))
1214 { 1377 {
1215 default: 1378 default:
1216 ent->type = EIO_DT_UNKNOWN; 1379 ent->type = EIO_DT_UNKNOWN;
1259 #ifdef DT_WHT 1422 #ifdef DT_WHT
1260 case DT_WHT: ent->type = EIO_DT_WHT; break; 1423 case DT_WHT: ent->type = EIO_DT_WHT; break;
1261 #endif 1424 #endif
1262 } 1425 }
1263 1426
1264 ent->score = 0; 1427 ent->score = 7;
1265 1428
1266 if (flags & EIO_READDIR_DIRS_FIRST) 1429 if (flags & EIO_READDIR_DIRS_FIRST)
1267 { 1430 {
1268 if (ent->type == EIO_DT_UNKNOWN) 1431 if (ent->type == EIO_DT_UNKNOWN)
1269 { 1432 {
1270 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1433 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1271 ent->score = 98; 1434 ent->score = 1;
1272 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1435 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1273 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ 1436 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1274 } 1437 }
1275 else if (ent->type == EIO_DT_DIR) 1438 else if (ent->type == EIO_DT_DIR)
1276 ent->score = 100; 1439 ent->score = 0;
1277 } 1440 }
1278 } 1441 }
1279 1442
1280 namesoffs += len; 1443 namesoffs += len;
1281 ++dentoffs; 1444 ++dentoffs;
1287 break; 1450 break;
1288 } 1451 }
1289 } 1452 }
1290} 1453}
1291 1454
1292#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1293# undef msync
1294# define msync(a,b,c) ((errno = ENOSYS), -1)
1295#endif
1296
1297int
1298eio__mtouch (void *mem, size_t len, int flags)
1299{
1300 intptr_t addr = (intptr_t)mem;
1301 intptr_t end = addr + len;
1302#ifdef PAGESIZE 1455#ifdef PAGESIZE
1303 const intptr_t page = PAGESIZE; 1456# define eio_pagesize() PAGESIZE
1304#else 1457#else
1458static intptr_t
1459eio_pagesize (void)
1460{
1305 static intptr_t page; 1461 static intptr_t page;
1306 1462
1307 if (!page) 1463 if (!page)
1308 page = sysconf (_SC_PAGESIZE); 1464 page = sysconf (_SC_PAGESIZE);
1465
1466 return page;
1467}
1468#endif
1469
1470static void
1471eio_page_align (void **addr, size_t *length)
1472{
1473 intptr_t mask = eio_pagesize () - 1;
1474
1475 /* round down addr */
1476 intptr_t adj = mask & (intptr_t)*addr;
1477
1478 *addr = (void *)((intptr_t)*addr - adj);
1479 *length += adj;
1480
1481 /* round up length */
1482 *length = (*length + mask) & ~mask;
1483}
1484
1485#if !_POSIX_MEMLOCK
1486# define eio__mlockall(a) ((errno = ENOSYS), -1)
1487#else
1488
1489static int
1490eio__mlockall (int flags)
1491{
1492 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1493 extern int mallopt (int, int);
1494 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1309#endif 1495 #endif
1310 1496
1311 addr &= ~(page - 1); /* assume page size is always a power of two */ 1497 if (EIO_MCL_CURRENT != MCL_CURRENT
1498 || EIO_MCL_FUTURE != MCL_FUTURE)
1499 {
1500 flags = 0
1501 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1502 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1503 }
1312 1504
1505 return mlockall (flags);
1506}
1507#endif
1508
1509#if !_POSIX_MEMLOCK_RANGE
1510# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1511#else
1512
1513static int
1514eio__mlock (void *addr, size_t length)
1515{
1516 eio_page_align (&addr, &length);
1517
1518 return mlock (addr, length);
1519}
1520
1521#endif
1522
1523#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1524# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1525#else
1526
1527int
1528eio__msync (void *mem, size_t len, int flags)
1529{
1530 eio_page_align (&mem, &len);
1531
1532 if (EIO_MS_ASYNC != MS_SYNC
1533 || EIO_MS_INVALIDATE != MS_INVALIDATE
1534 || EIO_MS_SYNC != MS_SYNC)
1535 {
1536 flags = 0
1537 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1538 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1539 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1540 }
1541
1542 return msync (mem, len, flags);
1543}
1544
1545#endif
1546
1547int
1548eio__mtouch (eio_req *req)
1549{
1550 void *mem = req->ptr2;
1551 size_t len = req->size;
1552 int flags = req->int1;
1553
1554 eio_page_align (&mem, &len);
1555
1556 {
1557 intptr_t addr = (intptr_t)mem;
1558 intptr_t end = addr + len;
1559 intptr_t page = eio_pagesize ();
1560
1313 if (addr < end) 1561 if (addr < end)
1314 if (flags) /* modify */ 1562 if (flags & EIO_MT_MODIFY) /* modify */
1315 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1563 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1316 else 1564 else
1317 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1565 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1566 }
1318 1567
1319 return 0; 1568 return 0;
1320} 1569}
1321 1570
1322/*****************************************************************************/ 1571/*****************************************************************************/
1356 if (req) 1605 if (req)
1357 break; 1606 break;
1358 1607
1359 ++idle; 1608 ++idle;
1360 1609
1361 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1610 ts.tv_sec = time (0) + idle_timeout;
1362 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1611 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1363 { 1612 {
1364 if (idle > max_idle) 1613 if (idle > max_idle)
1365 { 1614 {
1366 --idle; 1615 --idle;
1445 return 0; \ 1694 return 0; \
1446 } 1695 }
1447 1696
1448static void eio_execute (etp_worker *self, eio_req *req) 1697static void eio_execute (etp_worker *self, eio_req *req)
1449{ 1698{
1450 errno = 0;
1451
1452 switch (req->type) 1699 switch (req->type)
1453 { 1700 {
1454 case EIO_READ: ALLOC (req->size); 1701 case EIO_READ: ALLOC (req->size);
1455 req->result = req->offs >= 0 1702 req->result = req->offs >= 0
1456 ? pread (req->int1, req->ptr2, req->size, req->offs) 1703 ? pread (req->int1, req->ptr2, req->size, req->offs)
1467 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1714 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1468 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1715 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1469 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1716 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1470 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1717 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1471 1718
1719 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1720 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1721 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1722 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1723
1472 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1724 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1473 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1725 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1474 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1726 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1475 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1727 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1476 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1728 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1483 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1735 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1484 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1736 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1485 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1737 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1486 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1738 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1487 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1739 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1488 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1740 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1489 1741
1490 case EIO_READLINK: ALLOC (NAME_MAX); 1742 case EIO_READLINK: ALLOC (PATH_MAX);
1491 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1743 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1492 1744
1493 case EIO_SYNC: req->result = 0; sync (); break; 1745 case EIO_SYNC: req->result = 0; sync (); break;
1494 case EIO_FSYNC: req->result = fsync (req->int1); break; 1746 case EIO_FSYNC: req->result = fsync (req->int1); break;
1495 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1747 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1496 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1748 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1749 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1497 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1750 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1751 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1498 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1752 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1499 1753
1500 case EIO_READDIR: eio__scandir (req, self); break; 1754 case EIO_READDIR: eio__scandir (req, self); break;
1501 1755
1502 case EIO_BUSY: 1756 case EIO_BUSY:
1503#ifdef _WIN32 1757#ifdef _WIN32
1504 Sleep (req->nv1 * 1000.); 1758 Sleep (req->nv1 * 1e3);
1505#else 1759#else
1506 { 1760 {
1507 struct timeval tv; 1761 struct timeval tv;
1508 1762
1509 tv.tv_sec = req->nv1; 1763 tv.tv_sec = req->nv1;
1510 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1764 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1511 1765
1512 req->result = select (0, 0, 0, 0, &tv); 1766 req->result = select (0, 0, 0, 0, &tv);
1513 } 1767 }
1514#endif 1768#endif
1515 break; 1769 break;
1530 times = tv; 1784 times = tv;
1531 } 1785 }
1532 else 1786 else
1533 times = 0; 1787 times = 0;
1534 1788
1535
1536 req->result = req->type == EIO_FUTIME 1789 req->result = req->type == EIO_FUTIME
1537 ? futimes (req->int1, times) 1790 ? futimes (req->int1, times)
1538 : utimes (req->ptr1, times); 1791 : utimes (req->ptr1, times);
1539 } 1792 }
1540 break; 1793 break;
1545 case EIO_NOP: 1798 case EIO_NOP:
1546 req->result = 0; 1799 req->result = 0;
1547 break; 1800 break;
1548 1801
1549 case EIO_CUSTOM: 1802 case EIO_CUSTOM:
1550 ((void (*)(eio_req *))req->feed) (req); 1803 req->feed (req);
1551 break; 1804 break;
1552 1805
1553 default: 1806 default:
1807 errno = ENOSYS;
1554 req->result = -1; 1808 req->result = -1;
1555 break; 1809 break;
1556 } 1810 }
1557 1811
1558 req->errorno = errno; 1812 req->errorno = errno;
1588eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1842eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1589{ 1843{
1590 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1844 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1591} 1845}
1592 1846
1847eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1848{
1849 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1850}
1851
1852eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1853{
1854 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1855}
1856
1593eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1857eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1594{ 1858{
1595 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1859 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1596} 1860}
1597 1861
1621} 1885}
1622 1886
1623eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1887eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1624{ 1888{
1625 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1889 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1890}
1891
1892eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1893{
1894 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1626} 1895}
1627 1896
1628eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1897eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1629{ 1898{
1630 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1899 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1704eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1973eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1705{ 1974{
1706 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1975 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1707} 1976}
1708 1977
1978eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1979{
1980 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1981}
1982
1709eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1983eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1710{ 1984{
1711 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1985 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1712} 1986}
1713 1987
1721 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1995 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1722} 1996}
1723 1997
1724eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1998eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1725{ 1999{
1726 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2000 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1727} 2001}
1728 2002
1729static eio_req * 2003static eio_req *
1730eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2004eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1731{ 2005{
1755eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2029eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1756{ 2030{
1757 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2031 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1758} 2032}
1759 2033
1760eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2034eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data);
1761{ 2035{
1762 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2036 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1763} 2037}
1764 2038
1765#endif 2039#endif
1766 2040
1767eio_req *eio_grp (eio_cb cb, void *data) 2041eio_req *eio_grp (eio_cb cb, void *data)

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines