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

Comparing libeio/eio.c (file contents):
Revision 1.43 by root, Wed Jun 17 18:29:15 2009 UTC vs.
Revision 1.69 by root, Fri Jun 10 06:50:42 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;
… …
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 HANDLE h = TO_SOCKET (ifd);
1010 SetFilePointer (h, offset, 0, FILE_BEGIN);
1011 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1012
948#else 1013#else
949 res = -1; 1014 res = -1;
950 errno = ENOSYS; 1015 errno = ENOSYS;
951#endif 1016#endif
952 1017
1018 /* we assume sendfile can copy at least 128mb in one go */
1019 if (res <= 128 * 1024 * 1024)
1020 {
1021 if (res > 0)
1022 written += res;
1023
1024 if (written)
1025 return written;
1026
1027 break;
1028 }
1029 else
1030 {
1031 /* if we requested more, then probably the kernel was lazy */
1032 written += res;
1033 offset += res;
1034 count -= res;
1035
1036 if (!count)
1037 return written;
1038 }
1039 }
1040
953 if (res < 0 1041 if (res < 0
954 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1042 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1043 /* BSDs */
1044#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1045 || errno == ENOTSUP
1046#endif
1047 || errno == EOPNOTSUPP /* BSDs */
955#if __solaris 1048#if __solaris
956 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1049 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
957#endif 1050#endif
958 ) 1051 )
959 ) 1052 )
… …
993} 1086}
994 1087
995static signed char 1088static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1089eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{ 1090{
998 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1091 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; 1092 : a->inode < b->inode ? -1
1093 : a->inode > b->inode ? 1
1094 : 0;
1000} 1095}
1001 1096
1002#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1097#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1003 1098
1004#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1099#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
… …
1010 unsigned char bits [9 + sizeof (ino_t) * 8]; 1105 unsigned char bits [9 + sizeof (ino_t) * 8];
1011 unsigned char *bit = bits; 1106 unsigned char *bit = bits;
1012 1107
1013 assert (CHAR_BIT == 8); 1108 assert (CHAR_BIT == 8);
1014 assert (sizeof (eio_dirent) * 8 < 256); 1109 assert (sizeof (eio_dirent) * 8 < 256);
1015 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1110 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1016 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1111 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1017 1112
1018 if (size <= EIO_SORT_FAST) 1113 if (size <= EIO_SORT_FAST)
1019 return; 1114 return;
1020 1115
1021 /* first prepare an array of bits to test in our radix sort */ 1116 /* first prepare an array of bits to test in our radix sort */
… …
1176 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1271 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1177 1272
1178 X_LOCK (wrklock); 1273 X_LOCK (wrklock);
1179 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1274 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1180 self->dirp = dirp = opendir (req->ptr1); 1275 self->dirp = dirp = opendir (req->ptr1);
1276
1181 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1277 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1182 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1278 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1183 req->ptr2 = names = malloc (namesalloc); 1279 req->ptr2 = names = malloc (namesalloc);
1184 X_UNLOCK (wrklock); 1280 X_UNLOCK (wrklock);
1185 1281
… …
1197 /* sort etc. */ 1293 /* sort etc. */
1198 req->int1 = flags; 1294 req->int1 = flags;
1199 req->result = dentoffs; 1295 req->result = dentoffs;
1200 1296
1201 if (flags & EIO_READDIR_STAT_ORDER) 1297 if (flags & EIO_READDIR_STAT_ORDER)
1202 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1298 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1203 else if (flags & EIO_READDIR_DIRS_FIRST) 1299 else if (flags & EIO_READDIR_DIRS_FIRST)
1204 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1300 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1205 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1301 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1206 else 1302 else
1207 { 1303 {
… …
1209 eio_dirent *oth = dents + dentoffs; 1305 eio_dirent *oth = dents + dentoffs;
1210 eio_dirent *dir = dents; 1306 eio_dirent *dir = dents;
1211 1307
1212 /* now partition dirs to the front, and non-dirs to the back */ 1308 /* now partition dirs to the front, and non-dirs to the back */
1213 /* by walking from both sides and swapping if necessary */ 1309 /* by walking from both sides and swapping if necessary */
1214 /* also clear score, so it doesn't influence sorting */
1215 while (oth > dir) 1310 while (oth > dir)
1216 { 1311 {
1217 if (dir->type == EIO_DT_DIR) 1312 if (dir->type == EIO_DT_DIR)
1218 ++dir; 1313 ++dir;
1219 else if ((--oth)->type == EIO_DT_DIR) 1314 else if ((--oth)->type == EIO_DT_DIR)
… …
1222 1317
1223 ++dir; 1318 ++dir;
1224 } 1319 }
1225 } 1320 }
1226 1321
1227 /* now sort the dirs only */ 1322 /* now sort the dirs only (dirs all have the same score) */
1228 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1323 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1229 } 1324 }
1230 1325
1231 break; 1326 break;
1232 } 1327 }
… …
1352 break; 1447 break;
1353 } 1448 }
1354 } 1449 }
1355} 1450}
1356 1451
1357#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1358# undef msync
1359# define msync(a,b,c) ((errno = ENOSYS), -1)
1360#endif
1361
1362int
1363eio__mtouch (void *mem, size_t len, int flags)
1364{
1365 intptr_t addr = (intptr_t)mem;
1366 intptr_t end = addr + len;
1367#ifdef PAGESIZE 1452#ifdef PAGESIZE
1368 const intptr_t page = PAGESIZE; 1453# define eio_pagesize() PAGESIZE
1369#else 1454#else
1455static intptr_t
1456eio_pagesize (void)
1457{
1370 static intptr_t page; 1458 static intptr_t page;
1371 1459
1372 if (!page) 1460 if (!page)
1373 page = sysconf (_SC_PAGESIZE); 1461 page = sysconf (_SC_PAGESIZE);
1462
1463 return page;
1464}
1465#endif
1466
1467static void
1468eio_page_align (void **addr, size_t *length)
1469{
1470 intptr_t mask = eio_pagesize () - 1;
1471
1472 /* round down addr */
1473 intptr_t adj = mask & (intptr_t)*addr;
1474
1475 *addr = (void *)((intptr_t)*addr - adj);
1476 *length += adj;
1477
1478 /* round up length */
1479 *length = (*length + mask) & ~mask;
1480}
1481
1482#if !_POSIX_MEMLOCK
1483# define eio__mlockall(a) ((errno = ENOSYS), -1)
1484#else
1485
1486static int
1487eio__mlockall (int flags)
1488{
1489 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1490 extern int mallopt (int, int);
1491 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1374#endif 1492 #endif
1375 1493
1376 addr &= ~(page - 1); /* assume page size is always a power of two */ 1494 if (EIO_MCL_CURRENT != MCL_CURRENT
1495 || EIO_MCL_FUTURE != MCL_FUTURE)
1496 {
1497 flags = 0
1498 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1499 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1500 }
1377 1501
1502 return mlockall (flags);
1503}
1504#endif
1505
1506#if !_POSIX_MEMLOCK_RANGE
1507# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1508#else
1509
1510static int
1511eio__mlock (void *addr, size_t length)
1512{
1513 eio_page_align (&addr, &length);
1514
1515 return mlock (addr, length);
1516}
1517
1518#endif
1519
1520#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1521# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1522#else
1523
1524int
1525eio__msync (void *mem, size_t len, int flags)
1526{
1527 eio_page_align (&mem, &len);
1528
1529 if (EIO_MS_ASYNC != MS_SYNC
1530 || EIO_MS_INVALIDATE != MS_INVALIDATE
1531 || EIO_MS_SYNC != MS_SYNC)
1532 {
1533 flags = 0
1534 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1535 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1536 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1537 }
1538
1539 return msync (mem, len, flags);
1540}
1541
1542#endif
1543
1544int
1545eio__mtouch (eio_req *req)
1546{
1547 void *mem = req->ptr2;
1548 size_t len = req->size;
1549 int flags = req->int1;
1550
1551 eio_page_align (&mem, &len);
1552
1553 {
1554 intptr_t addr = (intptr_t)mem;
1555 intptr_t end = addr + len;
1556 intptr_t page = eio_pagesize ();
1557
1378 if (addr < end) 1558 if (addr < end)
1379 if (flags) /* modify */ 1559 if (flags & EIO_MT_MODIFY) /* modify */
1380 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1560 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1381 else 1561 else
1382 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1562 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1563 }
1383 1564
1384 return 0; 1565 return 0;
1385} 1566}
1386 1567
1387/*****************************************************************************/ 1568/*****************************************************************************/
… …
1421 if (req) 1602 if (req)
1422 break; 1603 break;
1423 1604
1424 ++idle; 1605 ++idle;
1425 1606
1426 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1607 ts.tv_sec = time (0) + idle_timeout;
1427 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1608 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1428 { 1609 {
1429 if (idle > max_idle) 1610 if (idle > max_idle)
1430 { 1611 {
1431 --idle; 1612 --idle;
… …
1510 return 0; \ 1691 return 0; \
1511 } 1692 }
1512 1693
1513static void eio_execute (etp_worker *self, eio_req *req) 1694static void eio_execute (etp_worker *self, eio_req *req)
1514{ 1695{
1515 errno = 0;
1516
1517 switch (req->type) 1696 switch (req->type)
1518 { 1697 {
1519 case EIO_READ: ALLOC (req->size); 1698 case EIO_READ: ALLOC (req->size);
1520 req->result = req->offs >= 0 1699 req->result = req->offs >= 0
1521 ? pread (req->int1, req->ptr2, req->size, req->offs) 1700 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1532 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1711 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1533 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1712 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1534 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1713 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1535 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1714 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1536 1715
1716 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1717 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1718 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1719 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1720
1537 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1721 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; 1722 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; 1723 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; 1724 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; 1725 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1548 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1732 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1549 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1733 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; 1734 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1551 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1735 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1552 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1736 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; 1737 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1554 1738
1555 case EIO_READLINK: ALLOC (NAME_MAX); 1739 case EIO_READLINK: ALLOC (PATH_MAX);
1556 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1740 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1557 1741
1558 case EIO_SYNC: req->result = 0; sync (); break; 1742 case EIO_SYNC: req->result = 0; sync (); break;
1559 case EIO_FSYNC: req->result = fsync (req->int1); break; 1743 case EIO_FSYNC: req->result = fsync (req->int1); break;
1560 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1744 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1561 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1745 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1746 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1562 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1747 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1748 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; 1749 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1564 1750
1565 case EIO_READDIR: eio__scandir (req, self); break; 1751 case EIO_READDIR: eio__scandir (req, self); break;
1566 1752
1567 case EIO_BUSY: 1753 case EIO_BUSY:
1568#ifdef _WIN32 1754#ifdef _WIN32
1569 Sleep (req->nv1 * 1000.); 1755 Sleep (req->nv1 * 1e3);
1570#else 1756#else
1571 { 1757 {
1572 struct timeval tv; 1758 struct timeval tv;
1573 1759
1574 tv.tv_sec = req->nv1; 1760 tv.tv_sec = req->nv1;
1575 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1761 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1576 1762
1577 req->result = select (0, 0, 0, 0, &tv); 1763 req->result = select (0, 0, 0, 0, &tv);
1578 } 1764 }
1579#endif 1765#endif
1580 break; 1766 break;
… …
1595 times = tv; 1781 times = tv;
1596 } 1782 }
1597 else 1783 else
1598 times = 0; 1784 times = 0;
1599 1785
1600
1601 req->result = req->type == EIO_FUTIME 1786 req->result = req->type == EIO_FUTIME
1602 ? futimes (req->int1, times) 1787 ? futimes (req->int1, times)
1603 : utimes (req->ptr1, times); 1788 : utimes (req->ptr1, times);
1604 } 1789 }
1605 break; 1790 break;
… …
1610 case EIO_NOP: 1795 case EIO_NOP:
1611 req->result = 0; 1796 req->result = 0;
1612 break; 1797 break;
1613 1798
1614 case EIO_CUSTOM: 1799 case EIO_CUSTOM:
1615 ((void (*)(eio_req *))req->feed) (req); 1800 req->feed (req);
1616 break; 1801 break;
1617 1802
1618 default: 1803 default:
1804 errno = ENOSYS;
1619 req->result = -1; 1805 req->result = -1;
1620 break; 1806 break;
1621 } 1807 }
1622 1808
1623 req->errorno = errno; 1809 req->errorno = errno;
… …
1653eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1839eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1654{ 1840{
1655 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1841 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1656} 1842}
1657 1843
1844eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1845{
1846 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1847}
1848
1849eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1850{
1851 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1852}
1853
1658eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1854eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1659{ 1855{
1660 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1856 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1661} 1857}
1662 1858
… …
1686} 1882}
1687 1883
1688eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1884eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1689{ 1885{
1690 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1886 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1887}
1888
1889eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1890{
1891 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1691} 1892}
1692 1893
1693eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1894eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1694{ 1895{
1695 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1896 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) 1970eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1770{ 1971{
1771 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1972 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1772} 1973}
1773 1974
1975eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1976{
1977 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1978}
1979
1774eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1980eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1775{ 1981{
1776 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1982 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1777} 1983}
1778 1984
… …
1786 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1992 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1787} 1993}
1788 1994
1789eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1995eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1790{ 1996{
1791 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1997 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1792} 1998}
1793 1999
1794static eio_req * 2000static eio_req *
1795eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2001eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1796{ 2002{
… …
1820eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2026eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1821{ 2027{
1822 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2028 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1823} 2029}
1824 2030
1825eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2031eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data);
1826{ 2032{
1827 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2033 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1828} 2034}
1829 2035
1830#endif 2036#endif
1831 2037
1832eio_req *eio_grp (eio_cb cb, void *data) 2038eio_req *eio_grp (eio_cb cb, void *data)

Diff Legend

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