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 1.42 by root, Mon Jun 15 05:34:49 2009 UTC vs.
Revision 1.65 by root, Mon May 30 12:56:50 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
58#ifndef EIO_FINISH 63#ifndef EIO_FINISH
70#ifdef _WIN32 75#ifdef _WIN32
71 76
72 /*doh*/ 77 /*doh*/
73#else 78#else
74 79
75# include "config.h"
76# include <sys/time.h> 80# include <sys/time.h>
77# include <sys/select.h> 81# include <sys/select.h>
78# include <sys/mman.h>
79# include <unistd.h> 82# include <unistd.h>
80# include <utime.h> 83# include <utime.h>
81# include <signal.h> 84# include <signal.h>
82# include <dirent.h> 85# include <dirent.h>
83 86
87#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
88# include <sys/mman.h>
89#endif
90
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 91/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 92# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */ 93# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno 94# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen 95# define D_NAMLEN(de) (de)->d_namlen
89# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 96# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino 97# define D_INO(de) (de)->d_ino
91# endif 98# endif
92 99
93#ifdef _D_EXACT_NAMLEN 100#ifdef _D_EXACT_NAMLEN
94# undef D_NAMLEN 101# undef D_NAMLEN
106#endif 113#endif
107 114
108#if HAVE_SENDFILE 115#if HAVE_SENDFILE
109# if __linux 116# if __linux
110# include <sys/sendfile.h> 117# include <sys/sendfile.h>
111# elif __freebsd 118# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 119# include <sys/socket.h>
113# include <sys/uio.h> 120# include <sys/uio.h>
114# elif __hpux 121# elif __hpux
115# include <sys/socket.h> 122# include <sys/socket.h>
116# elif __solaris /* not yet */ 123# elif __solaris
117# include <sys/sendfile.h> 124# include <sys/sendfile.h>
118# else 125# else
119# error sendfile support requested but not available 126# error sendfile support requested but not available
120# endif 127# endif
121#endif 128#endif
128#endif 135#endif
129#ifndef D_NAMLEN 136#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 137# define D_NAMLEN(de) strlen ((de)->d_name)
131#endif 138#endif
132 139
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 */ 140/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 141#ifndef NAME_MAX
138# define NAME_MAX 4096 142# define NAME_MAX 4096
143#endif
144
145/* used for readlink etc. */
146#ifndef PATH_MAX
147# define PATH_MAX 4096
139#endif 148#endif
140 149
141/* buffer size for various temporary buffers */ 150/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 151#define EIO_BUFSIZE 65536
143 152
196 205
197/*****************************************************************************/ 206/*****************************************************************************/
198 207
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200 209
201/* calculcate time difference in ~1/EIO_TICKS of a second */ 210/* calculate time difference in ~1/EIO_TICKS of a second */
202static int tvdiff (struct timeval *tv1, struct timeval *tv2) 211static int tvdiff (struct timeval *tv1, struct timeval *tv2)
203{ 212{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 214 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206} 215}
214static unsigned int max_poll_reqs; /* reslock */ 223static unsigned int max_poll_reqs; /* reslock */
215 224
216static volatile unsigned int nreqs; /* reqlock */ 225static volatile unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */ 226static volatile unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */ 227static volatile unsigned int npending; /* reqlock */
219static volatile unsigned int max_idle = 4; 228static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
229static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220 230
221static mutex_t wrklock = X_MUTEX_INIT; 231static xmutex_t wrklock;
222static mutex_t reslock = X_MUTEX_INIT; 232static xmutex_t reslock;
223static mutex_t reqlock = X_MUTEX_INIT; 233static xmutex_t reqlock;
224static cond_t reqwait = X_COND_INIT; 234static xcond_t reqwait;
225 235
226#if !HAVE_PREADWRITE 236#if !HAVE_PREADWRITE
227/* 237/*
228 * make our pread/pwrite emulation safe against themselves, but not against 238 * make our pread/pwrite emulation safe against themselves, but not against
229 * normal read/write by using a mutex. slows down execution a lot, 239 * normal read/write by using a mutex. slows down execution a lot,
230 * but that's your problem, not mine. 240 * but that's your problem, not mine.
231 */ 241 */
232static mutex_t preadwritelock = X_MUTEX_INIT; 242static xmutex_t preadwritelock = X_MUTEX_INIT;
233#endif 243#endif
234 244
235typedef struct etp_worker 245typedef struct etp_worker
236{ 246{
237 /* locked by wrklock */ 247 /* locked by wrklock */
238 struct etp_worker *prev, *next; 248 struct etp_worker *prev, *next;
239 249
240 thread_t tid; 250 xthread_t tid;
241 251
242 /* locked by reslock, reqlock or wrklock */ 252 /* locked by reslock, reqlock or wrklock */
243 ETP_REQ *req; /* currently processed request */ 253 ETP_REQ *req; /* currently processed request */
244 254
245 ETP_WORKER_COMMON 255 ETP_WORKER_COMMON
359 } 369 }
360 370
361 abort (); 371 abort ();
362} 372}
363 373
374static void etp_thread_init (void)
375{
376 X_MUTEX_CREATE (wrklock);
377 X_MUTEX_CREATE (reslock);
378 X_MUTEX_CREATE (reqlock);
379 X_COND_CREATE (reqwait);
380}
381
364static void etp_atfork_prepare (void) 382static void etp_atfork_prepare (void)
365{ 383{
366 X_LOCK (wrklock); 384 X_LOCK (wrklock);
367 X_LOCK (reqlock); 385 X_LOCK (reqlock);
368 X_LOCK (reslock); 386 X_LOCK (reslock);
406 idle = 0; 424 idle = 0;
407 nreqs = 0; 425 nreqs = 0;
408 nready = 0; 426 nready = 0;
409 npending = 0; 427 npending = 0;
410 428
411 etp_atfork_parent (); 429 etp_thread_init ();
412} 430}
413 431
414static void 432static void
415etp_once_init (void) 433etp_once_init (void)
416{ 434{
435 etp_thread_init ();
417 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 436 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418} 437}
419 438
420static int 439static int
421etp_init (void (*want_poll)(void), void (*done_poll)(void)) 440etp_init (void (*want_poll)(void), void (*done_poll)(void))
598} 617}
599 618
600static void etp_set_max_poll_time (double nseconds) 619static void etp_set_max_poll_time (double nseconds)
601{ 620{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 621 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds; 622 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 623 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605} 624}
606 625
607static void etp_set_max_poll_reqs (unsigned int maxreqs) 626static void etp_set_max_poll_reqs (unsigned int maxreqs)
608{ 627{
612} 631}
613 632
614static void etp_set_max_idle (unsigned int nthreads) 633static void etp_set_max_idle (unsigned int nthreads)
615{ 634{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 635 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads; 636 max_idle = nthreads;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
638}
639
640static void etp_set_idle_timeout (unsigned int seconds)
641{
642 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
643 idle_timeout = seconds;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 644 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619} 645}
620 646
621static void etp_set_min_parallel (unsigned int nthreads) 647static void etp_set_min_parallel (unsigned int nthreads)
622{ 648{
750void eio_set_max_idle (unsigned int nthreads) 776void eio_set_max_idle (unsigned int nthreads)
751{ 777{
752 etp_set_max_idle (nthreads); 778 etp_set_max_idle (nthreads);
753} 779}
754 780
781void eio_set_idle_timeout (unsigned int seconds)
782{
783 etp_set_idle_timeout (seconds);
784}
785
755void eio_set_min_parallel (unsigned int nthreads) 786void eio_set_min_parallel (unsigned int nthreads)
756{ 787{
757 etp_set_min_parallel (nthreads); 788 etp_set_min_parallel (nthreads);
758} 789}
759 790
807 838
808 return res; 839 return res;
809} 840}
810#endif 841#endif
811 842
812#ifndef HAVE_FUTIMES 843#ifndef HAVE_UTIMES
813 844
814# undef utimes 845# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 846# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 847
819static int 848static int
820eio__utimes (const char *filename, const struct timeval times[2]) 849eio__utimes (const char *filename, const struct timeval times[2])
821{ 850{
822 if (times) 851 if (times)
829 return utime (filename, &buf); 858 return utime (filename, &buf);
830 } 859 }
831 else 860 else
832 return utime (filename, 0); 861 return utime (filename, 0);
833} 862}
863
864#endif
865
866#ifndef HAVE_FUTIMES
867
868# undef futimes
869# define futimes(fd,times) eio__futimes (fd, times)
834 870
835static int eio__futimes (int fd, const struct timeval tv[2]) 871static int eio__futimes (int fd, const struct timeval tv[2])
836{ 872{
837 errno = ENOSYS; 873 errno = ENOSYS;
838 return -1; 874 return -1;
867 if (!res || errno != ENOSYS) 903 if (!res || errno != ENOSYS)
868 return res; 904 return res;
869#endif 905#endif
870 906
871 /* even though we could play tricks with the flags, it's better to always 907 /* 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 */ 908 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 909 return fdatasync (fd);
874} 910}
875 911
876#if !HAVE_READAHEAD 912#if !HAVE_READAHEAD
877# undef readahead 913# undef readahead
909 945
910#if HAVE_SENDFILE 946#if HAVE_SENDFILE
911# if __linux 947# if __linux
912 res = sendfile (ofd, ifd, &offset, count); 948 res = sendfile (ofd, ifd, &offset, count);
913 949
914# elif __freebsd 950# elif __FreeBSD__
915 /* 951 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 952 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 953 * wasted on making it similar to other I/O functions.
918 */ 954 */
919 { 955 {
920 off_t sbytes; 956 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 957 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 958
923 if (res < 0 && sbytes) 959 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 960 /* freebsd' sendfile will return 0 on success */
961 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
962 /* not on e.g. EIO or EPIPE - sounds broken */
963 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
964 res = sbytes;
965 #endif
966
967 /* according to source inspection, this is correct, and useful behaviour */
968 if (sbytes)
969 res = sbytes;
970 }
971
972# elif defined (__APPLE__)
973
974 {
975 off_t sbytes = count;
976 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
977
978 /* according to the manpage, sbytes is always valid */
979 if (sbytes)
925 res = sbytes; 980 res = sbytes;
926 } 981 }
927 982
928# elif __hpux 983# elif __hpux
929 res = sendfile (ofd, ifd, offset, count, 0, 0); 984 res = sendfile (ofd, ifd, offset, count, 0, 0);
943 if (res < 0 && sbytes) 998 if (res < 0 && sbytes)
944 res = sbytes; 999 res = sbytes;
945 } 1000 }
946 1001
947# endif 1002# endif
1003
1004#elif defined (_WIN32)
1005
1006 /* does not work, just for documentation of what would need to be done */
1007 {
1008 HANDLE h = TO_SOCKET (ifd);
1009 SetFilePointer (h, offset, 0, FILE_BEGIN);
1010 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1011 }
1012
948#else 1013#else
949 res = -1; 1014 res = -1;
950 errno = ENOSYS; 1015 errno = ENOSYS;
951#endif 1016#endif
952 1017
953 if (res < 0 1018 if (res < 0
954 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1019 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1020 /* BSDs */
1021#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1022 || errno == ENOTSUP
1023#endif
1024 || errno == EOPNOTSUPP /* BSDs */
955#if __solaris 1025#if __solaris
956 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1026 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
957#endif 1027#endif
958 ) 1028 )
959 ) 1029 )
993} 1063}
994 1064
995static signed char 1065static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1066eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{ 1067{
998 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1068 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; 1069 : a->inode < b->inode ? -1
1070 : a->inode > b->inode ? 1
1071 : 0;
1000} 1072}
1001 1073
1002#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1074#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1003 1075
1004#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1076#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1010 unsigned char bits [9 + sizeof (ino_t) * 8]; 1082 unsigned char bits [9 + sizeof (ino_t) * 8];
1011 unsigned char *bit = bits; 1083 unsigned char *bit = bits;
1012 1084
1013 assert (CHAR_BIT == 8); 1085 assert (CHAR_BIT == 8);
1014 assert (sizeof (eio_dirent) * 8 < 256); 1086 assert (sizeof (eio_dirent) * 8 < 256);
1015 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1087 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1016 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1088 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1017 1089
1018 if (size <= EIO_SORT_FAST) 1090 if (size <= EIO_SORT_FAST)
1019 return; 1091 return;
1020 1092
1021 /* first prepare an array of bits to test in our radix sort */ 1093 /* first prepare an array of bits to test in our radix sort */
1176 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1248 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1177 1249
1178 X_LOCK (wrklock); 1250 X_LOCK (wrklock);
1179 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1251 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1180 self->dirp = dirp = opendir (req->ptr1); 1252 self->dirp = dirp = opendir (req->ptr1);
1253
1181 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1254 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1182 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1255 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1183 req->ptr2 = names = malloc (namesalloc); 1256 req->ptr2 = names = malloc (namesalloc);
1184 X_UNLOCK (wrklock); 1257 X_UNLOCK (wrklock);
1185 1258
1197 /* sort etc. */ 1270 /* sort etc. */
1198 req->int1 = flags; 1271 req->int1 = flags;
1199 req->result = dentoffs; 1272 req->result = dentoffs;
1200 1273
1201 if (flags & EIO_READDIR_STAT_ORDER) 1274 if (flags & EIO_READDIR_STAT_ORDER)
1202 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1275 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1203 else if (flags & EIO_READDIR_DIRS_FIRST) 1276 else if (flags & EIO_READDIR_DIRS_FIRST)
1204 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1277 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1205 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1278 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1206 else 1279 else
1207 { 1280 {
1209 eio_dirent *oth = dents + dentoffs; 1282 eio_dirent *oth = dents + dentoffs;
1210 eio_dirent *dir = dents; 1283 eio_dirent *dir = dents;
1211 1284
1212 /* now partition dirs to the front, and non-dirs to the back */ 1285 /* now partition dirs to the front, and non-dirs to the back */
1213 /* by walking from both sides and swapping if necessary */ 1286 /* by walking from both sides and swapping if necessary */
1214 /* also clear score, so it doesn't influence sorting */
1215 while (oth > dir) 1287 while (oth > dir)
1216 { 1288 {
1217 if (dir->type == EIO_DT_DIR) 1289 if (dir->type == EIO_DT_DIR)
1218 ++dir; 1290 ++dir;
1219 else if ((--oth)->type == EIO_DT_DIR) 1291 else if ((--oth)->type == EIO_DT_DIR)
1222 1294
1223 ++dir; 1295 ++dir;
1224 } 1296 }
1225 } 1297 }
1226 1298
1227 /* now sort the dirs only */ 1299 /* now sort the dirs only (dirs all have the same score) */
1228 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1300 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1229 } 1301 }
1230 1302
1231 break; 1303 break;
1232 } 1304 }
1352 break; 1424 break;
1353 } 1425 }
1354 } 1426 }
1355} 1427}
1356 1428
1429#ifdef PAGESIZE
1430# define eio_pagesize() PAGESIZE
1431#else
1432static intptr_t
1433eio_pagesize (void)
1434{
1435 static intptr_t page;
1436
1437 if (!page)
1438 page = sysconf (_SC_PAGESIZE);
1439
1440 return page;
1441}
1442#endif
1443
1444static void
1445eio_page_align (void **addr, size_t *length)
1446{
1447 intptr_t mask = eio_pagesize () - 1;
1448
1449 /* round down addr */
1450 intptr_t adj = mask & (intptr_t)*addr;
1451
1452 *addr = (void *)((intptr_t)*addr - adj);
1453 *length += adj;
1454
1455 /* round up length */
1456 *length = (*length + mask) & ~mask;
1457}
1458
1459#if !_POSIX_MEMLOCK
1460# define eio__mlockall(a) ((errno = ENOSYS), -1)
1461#else
1462
1463static int
1464eio__mlockall (int flags)
1465{
1466 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1467 extern int mallopt (int, int);
1468 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1469 #endif
1470
1471 if (EIO_MCL_CURRENT != MCL_CURRENT
1472 || EIO_MCL_FUTURE != MCL_FUTURE)
1473 {
1474 flags = 0
1475 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1476 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1477 }
1478
1479 return mlockall (flags);
1480}
1481#endif
1482
1483#if !_POSIX_MEMLOCK_RANGE
1484# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1485#else
1486
1487static int
1488eio__mlock (void *addr, size_t length)
1489{
1490 eio_page_align (&addr, &length);
1491
1492 return mlock (addr, length);
1493}
1494
1495#endif
1496
1357#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1497#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1358# undef msync
1359# define msync(a,b,c) ((errno = ENOSYS), -1) 1498# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1499#else
1500
1501int
1502eio__msync (void *mem, size_t len, int flags)
1503{
1504 eio_page_align (&mem, &len);
1505
1506 if (EIO_MS_ASYNC != MS_SYNC
1507 || EIO_MS_INVALIDATE != MS_INVALIDATE
1508 || EIO_MS_SYNC != MS_SYNC)
1509 {
1510 flags = 0
1511 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1512 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1513 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1514 }
1515
1516 return msync (mem, len, flags);
1517}
1518
1360#endif 1519#endif
1361 1520
1362int 1521int
1363eio__mtouch (void *mem, size_t len, int flags) 1522eio__mtouch (void *mem, size_t len, int flags)
1364{ 1523{
1524 eio_page_align (&mem, &len);
1525
1526 {
1365 intptr_t addr = (intptr_t)mem; 1527 intptr_t addr = (intptr_t)mem;
1366 intptr_t end = addr + len; 1528 intptr_t end = addr + len;
1367#ifdef PAGESIZE 1529 intptr_t page = eio_pagesize ();
1368 const intptr_t page = PAGESIZE;
1369#else
1370 static intptr_t page;
1371 1530
1372 if (!page)
1373 page = sysconf (_SC_PAGESIZE);
1374#endif
1375
1376 addr &= ~(page - 1); /* assume page size is always a power of two */
1377
1378 if (addr < end) 1531 if (addr < end)
1379 if (flags) /* modify */ 1532 if (flags & EIO_MT_MODIFY) /* modify */
1380 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1533 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1381 else 1534 else
1382 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1535 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1536 }
1383 1537
1384 return 0; 1538 return 0;
1385} 1539}
1386 1540
1387/*****************************************************************************/ 1541/*****************************************************************************/
1421 if (req) 1575 if (req)
1422 break; 1576 break;
1423 1577
1424 ++idle; 1578 ++idle;
1425 1579
1426 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1580 ts.tv_sec = time (0) + idle_timeout;
1427 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1581 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1428 { 1582 {
1429 if (idle > max_idle) 1583 if (idle > max_idle)
1430 { 1584 {
1431 --idle; 1585 --idle;
1510 return 0; \ 1664 return 0; \
1511 } 1665 }
1512 1666
1513static void eio_execute (etp_worker *self, eio_req *req) 1667static void eio_execute (etp_worker *self, eio_req *req)
1514{ 1668{
1515 errno = 0;
1516
1517 switch (req->type) 1669 switch (req->type)
1518 { 1670 {
1519 case EIO_READ: ALLOC (req->size); 1671 case EIO_READ: ALLOC (req->size);
1520 req->result = req->offs >= 0 1672 req->result = req->offs >= 0
1521 ? pread (req->int1, req->ptr2, req->size, req->offs) 1673 ? pread (req->int1, req->ptr2, req->size, req->offs)
1532 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1684 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1533 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1685 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1534 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1686 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1535 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1687 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1536 1688
1689 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1690 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1691 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1692 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1693
1537 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1694 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1538 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1695 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1539 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1696 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1540 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1697 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1541 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1698 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1548 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1705 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1549 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1706 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1550 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1707 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1551 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1708 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1552 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1709 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1553 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1710 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1554 1711
1555 case EIO_READLINK: ALLOC (NAME_MAX); 1712 case EIO_READLINK: ALLOC (PATH_MAX);
1556 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1713 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1557 1714
1558 case EIO_SYNC: req->result = 0; sync (); break; 1715 case EIO_SYNC: req->result = 0; sync (); break;
1559 case EIO_FSYNC: req->result = fsync (req->int1); break; 1716 case EIO_FSYNC: req->result = fsync (req->int1); break;
1560 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1717 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1561 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1718 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1562 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1719 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1720 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1721 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1563 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1722 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1564 1723
1565 case EIO_READDIR: eio__scandir (req, self); break; 1724 case EIO_READDIR: eio__scandir (req, self); break;
1566 1725
1567 case EIO_BUSY: 1726 case EIO_BUSY:
1568#ifdef _WIN32 1727#ifdef _WIN32
1569 Sleep (req->nv1 * 1000.); 1728 Sleep (req->nv1 * 1e3);
1570#else 1729#else
1571 { 1730 {
1572 struct timeval tv; 1731 struct timeval tv;
1573 1732
1574 tv.tv_sec = req->nv1; 1733 tv.tv_sec = req->nv1;
1575 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1734 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1576 1735
1577 req->result = select (0, 0, 0, 0, &tv); 1736 req->result = select (0, 0, 0, 0, &tv);
1578 } 1737 }
1579#endif 1738#endif
1580 break; 1739 break;
1595 times = tv; 1754 times = tv;
1596 } 1755 }
1597 else 1756 else
1598 times = 0; 1757 times = 0;
1599 1758
1600
1601 req->result = req->type == EIO_FUTIME 1759 req->result = req->type == EIO_FUTIME
1602 ? futimes (req->int1, times) 1760 ? futimes (req->int1, times)
1603 : utimes (req->ptr1, times); 1761 : utimes (req->ptr1, times);
1604 } 1762 }
1605 break; 1763 break;
1614 case EIO_CUSTOM: 1772 case EIO_CUSTOM:
1615 ((void (*)(eio_req *))req->feed) (req); 1773 ((void (*)(eio_req *))req->feed) (req);
1616 break; 1774 break;
1617 1775
1618 default: 1776 default:
1777 errno = ENOSYS;
1619 req->result = -1; 1778 req->result = -1;
1620 break; 1779 break;
1621 } 1780 }
1622 1781
1623 req->errorno = errno; 1782 req->errorno = errno;
1653eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1812eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1654{ 1813{
1655 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1814 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1656} 1815}
1657 1816
1817eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1818{
1819 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1820}
1821
1822eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1823{
1824 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1825}
1826
1658eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1827eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1659{ 1828{
1660 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1829 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1661} 1830}
1662 1831
1686} 1855}
1687 1856
1688eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1857eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1689{ 1858{
1690 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1859 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1860}
1861
1862eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1863{
1864 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1691} 1865}
1692 1866
1693eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1867eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1694{ 1868{
1695 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1869 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1769eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1943eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1770{ 1944{
1771 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1945 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1772} 1946}
1773 1947
1948eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1949{
1950 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1951}
1952
1774eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1953eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1775{ 1954{
1776 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1955 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1777} 1956}
1778 1957
1786 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1965 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1787} 1966}
1788 1967
1789eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1968eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1790{ 1969{
1791 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1970 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1792} 1971}
1793 1972
1794static eio_req * 1973static eio_req *
1795eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1974eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1796{ 1975{

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines