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

Comparing libeio/eio.c (file contents):
Revision 1.49 by root, Sat Jan 2 14:24:32 2010 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) 2007,2008,2009,2010 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 __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 __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
… …
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 || defined __APPLE__ 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 129# elif __solaris
… …
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
… …
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))
… …
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;
927 979
928# elif defined (__APPLE__) 980# elif defined (__APPLE__)
929
930 {
931 off_t sbytes = count; 981 off_t sbytes = count;
932 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
933 983
934 if (res < 0 && errno == EAGAIN && sbytes) 984 /* according to the manpage, sbytes is always valid */
985 if (sbytes)
935 res = sbytes; 986 res = sbytes;
936 }
937 987
938# elif __hpux 988# elif __hpux
939 res = sendfile (ofd, ifd, offset, count, 0, 0); 989 res = sendfile (ofd, ifd, offset, count, 0, 0);
940 990
941# elif __solaris 991# elif __solaris
942 {
943 struct sendfilevec vec; 992 struct sendfilevec vec;
944 size_t sbytes; 993 size_t sbytes;
945 994
946 vec.sfv_fd = ifd; 995 vec.sfv_fd = ifd;
947 vec.sfv_flag = 0; 996 vec.sfv_flag = 0;
948 vec.sfv_off = offset; 997 vec.sfv_off = offset;
949 vec.sfv_len = count; 998 vec.sfv_len = count;
950 999
951 res = sendfilev (ofd, &vec, 1, &sbytes); 1000 res = sendfilev (ofd, &vec, 1, &sbytes);
952 1001
953 if (res < 0 && sbytes) 1002 if (res < 0 && sbytes)
954 res = sbytes; 1003 res = sbytes;
955 }
956 1004
957# endif 1005# endif
958 1006
959#elif defined (_WIN32) 1007#elif defined (_WIN32)
960
961 /* does not work, just for documentation of what would need to be done */ 1008 /* does not work, just for documentation of what would need to be done */
962 { 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 */
963 HANDLE h = TO_SOCKET (ifd); 1012 HANDLE h = TO_SOCKET (ifd);
964 SetFilePointer (h, offset, 0, FILE_BEGIN); 1013 SetFilePointer (h, offset, 0, FILE_BEGIN);
965 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1014 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
966 }
967 1015
968#else 1016#else
969 res = -1; 1017 res = -1;
970 errno = ENOSYS; 1018 errno = ENOSYS;
971#endif 1019#endif
972 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
973 if (res < 0 1044 if (res < 0
974 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1045 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
975 /* BSDs */ 1046 /* BSDs */
976#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1047#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
977 || errno == ENOTSUP 1048 || errno == ENOTSUP
978#endif 1049#endif
… …
1018} 1089}
1019 1090
1020static signed char 1091static signed char
1021eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1092eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1022{ 1093{
1023 return a->score - b->score ? a->score - b->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 */
1024 : 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;
1025} 1098}
1026 1099
1027#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1100#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1028 1101
1029#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1102#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
… …
1035 unsigned char bits [9 + sizeof (ino_t) * 8]; 1108 unsigned char bits [9 + sizeof (ino_t) * 8];
1036 unsigned char *bit = bits; 1109 unsigned char *bit = bits;
1037 1110
1038 assert (CHAR_BIT == 8); 1111 assert (CHAR_BIT == 8);
1039 assert (sizeof (eio_dirent) * 8 < 256); 1112 assert (sizeof (eio_dirent) * 8 < 256);
1040 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1113 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1041 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1114 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1042 1115
1043 if (size <= EIO_SORT_FAST) 1116 if (size <= EIO_SORT_FAST)
1044 return; 1117 return;
1045 1118
1046 /* first prepare an array of bits to test in our radix sort */ 1119 /* first prepare an array of bits to test in our radix sort */
… …
1201 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1274 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1202 1275
1203 X_LOCK (wrklock); 1276 X_LOCK (wrklock);
1204 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1277 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1205 self->dirp = dirp = opendir (req->ptr1); 1278 self->dirp = dirp = opendir (req->ptr1);
1279
1206 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1280 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1207 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1281 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1208 req->ptr2 = names = malloc (namesalloc); 1282 req->ptr2 = names = malloc (namesalloc);
1209 X_UNLOCK (wrklock); 1283 X_UNLOCK (wrklock);
1210 1284
… …
1222 /* sort etc. */ 1296 /* sort etc. */
1223 req->int1 = flags; 1297 req->int1 = flags;
1224 req->result = dentoffs; 1298 req->result = dentoffs;
1225 1299
1226 if (flags & EIO_READDIR_STAT_ORDER) 1300 if (flags & EIO_READDIR_STAT_ORDER)
1227 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1301 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1228 else if (flags & EIO_READDIR_DIRS_FIRST) 1302 else if (flags & EIO_READDIR_DIRS_FIRST)
1229 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1303 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1230 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1304 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1231 else 1305 else
1232 { 1306 {
… …
1234 eio_dirent *oth = dents + dentoffs; 1308 eio_dirent *oth = dents + dentoffs;
1235 eio_dirent *dir = dents; 1309 eio_dirent *dir = dents;
1236 1310
1237 /* now partition dirs to the front, and non-dirs to the back */ 1311 /* now partition dirs to the front, and non-dirs to the back */
1238 /* by walking from both sides and swapping if necessary */ 1312 /* by walking from both sides and swapping if necessary */
1239 /* also clear score, so it doesn't influence sorting */
1240 while (oth > dir) 1313 while (oth > dir)
1241 { 1314 {
1242 if (dir->type == EIO_DT_DIR) 1315 if (dir->type == EIO_DT_DIR)
1243 ++dir; 1316 ++dir;
1244 else if ((--oth)->type == EIO_DT_DIR) 1317 else if ((--oth)->type == EIO_DT_DIR)
… …
1247 1320
1248 ++dir; 1321 ++dir;
1249 } 1322 }
1250 } 1323 }
1251 1324
1252 /* now sort the dirs only */ 1325 /* now sort the dirs only (dirs all have the same score) */
1253 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1326 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1254 } 1327 }
1255 1328
1256 break; 1329 break;
1257 } 1330 }
… …
1377 break; 1450 break;
1378 } 1451 }
1379 } 1452 }
1380} 1453}
1381 1454
1455#ifdef PAGESIZE
1456# define eio_pagesize() PAGESIZE
1457#else
1458static intptr_t
1459eio_pagesize (void)
1460{
1461 static intptr_t page;
1462
1463 if (!page)
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 */
1495 #endif
1496
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 }
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
1382#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1523#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1383# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1524# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1384#else 1525#else
1385 1526
1386int 1527int
1387eio__msync (void *mem, size_t len, int flags) 1528eio__msync (void *mem, size_t len, int flags)
1388{ 1529{
1530 eio_page_align (&mem, &len);
1531
1389 if (EIO_MS_ASYNC != MS_SYNC 1532 if (EIO_MS_ASYNC != MS_SYNC
1390 || EIO_MS_INVALIDATE != MS_INVALIDATE 1533 || EIO_MS_INVALIDATE != MS_INVALIDATE
1391 || EIO_MS_SYNC != MS_SYNC) 1534 || EIO_MS_SYNC != MS_SYNC)
1392 { 1535 {
1393 flags = 0 1536 flags = 0
… …
1400} 1543}
1401 1544
1402#endif 1545#endif
1403 1546
1404int 1547int
1405eio__mtouch (void *mem, size_t len, int flags) 1548eio__mtouch (eio_req *req)
1406{ 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 {
1407 intptr_t addr = (intptr_t)mem; 1557 intptr_t addr = (intptr_t)mem;
1408 intptr_t end = addr + len; 1558 intptr_t end = addr + len;
1409#ifdef PAGESIZE 1559 intptr_t page = eio_pagesize ();
1410 const intptr_t page = PAGESIZE;
1411#else
1412 static intptr_t page;
1413 1560
1414 if (!page)
1415 page = sysconf (_SC_PAGESIZE);
1416#endif
1417
1418 /* round down to start of page, although this is probably useless */
1419 addr &= ~(page - 1); /* assume page size is always a power of two */
1420
1421 if (addr < end) 1561 if (addr < end)
1422 if (flags & EIO_MT_MODIFY) /* modify */ 1562 if (flags & EIO_MT_MODIFY) /* modify */
1423 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));
1424 else 1564 else
1425 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 }
1426 1567
1427 return 0; 1568 return 0;
1428} 1569}
1429 1570
1430/*****************************************************************************/ 1571/*****************************************************************************/
… …
1464 if (req) 1605 if (req)
1465 break; 1606 break;
1466 1607
1467 ++idle; 1608 ++idle;
1468 1609
1469 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1610 ts.tv_sec = time (0) + idle_timeout;
1470 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1611 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1471 { 1612 {
1472 if (idle > max_idle) 1613 if (idle > max_idle)
1473 { 1614 {
1474 --idle; 1615 --idle;
… …
1553 return 0; \ 1694 return 0; \
1554 } 1695 }
1555 1696
1556static void eio_execute (etp_worker *self, eio_req *req) 1697static void eio_execute (etp_worker *self, eio_req *req)
1557{ 1698{
1558 errno = 0;
1559
1560 switch (req->type) 1699 switch (req->type)
1561 { 1700 {
1562 case EIO_READ: ALLOC (req->size); 1701 case EIO_READ: ALLOC (req->size);
1563 req->result = req->offs >= 0 1702 req->result = req->offs >= 0
1564 ? pread (req->int1, req->ptr2, req->size, req->offs) 1703 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1575 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1714 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1576 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1715 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1577 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1716 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1578 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1717 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1579 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
1580 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;
1581 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;
1582 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;
1583 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;
1584 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1728 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1591 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1735 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1592 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;
1593 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1737 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1594 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1738 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1595 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1739 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1596 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;
1597 1741
1598 case EIO_READLINK: ALLOC (NAME_MAX); 1742 case EIO_READLINK: ALLOC (PATH_MAX);
1599 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1743 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1600 1744
1601 case EIO_SYNC: req->result = 0; sync (); break; 1745 case EIO_SYNC: req->result = 0; sync (); break;
1602 case EIO_FSYNC: req->result = fsync (req->int1); break; 1746 case EIO_FSYNC: req->result = fsync (req->int1); break;
1603 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1747 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1604 case EIO_MSYNC: req->result = eio__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;
1605 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;
1606 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;
1607 1753
1608 case EIO_READDIR: eio__scandir (req, self); break; 1754 case EIO_READDIR: eio__scandir (req, self); break;
1609 1755
1610 case EIO_BUSY: 1756 case EIO_BUSY:
1611#ifdef _WIN32 1757#ifdef _WIN32
1612 Sleep (req->nv1 * 1000.); 1758 Sleep (req->nv1 * 1e3);
1613#else 1759#else
1614 { 1760 {
1615 struct timeval tv; 1761 struct timeval tv;
1616 1762
1617 tv.tv_sec = req->nv1; 1763 tv.tv_sec = req->nv1;
1618 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1764 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1619 1765
1620 req->result = select (0, 0, 0, 0, &tv); 1766 req->result = select (0, 0, 0, 0, &tv);
1621 } 1767 }
1622#endif 1768#endif
1623 break; 1769 break;
… …
1638 times = tv; 1784 times = tv;
1639 } 1785 }
1640 else 1786 else
1641 times = 0; 1787 times = 0;
1642 1788
1643
1644 req->result = req->type == EIO_FUTIME 1789 req->result = req->type == EIO_FUTIME
1645 ? futimes (req->int1, times) 1790 ? futimes (req->int1, times)
1646 : utimes (req->ptr1, times); 1791 : utimes (req->ptr1, times);
1647 } 1792 }
1648 break; 1793 break;
… …
1653 case EIO_NOP: 1798 case EIO_NOP:
1654 req->result = 0; 1799 req->result = 0;
1655 break; 1800 break;
1656 1801
1657 case EIO_CUSTOM: 1802 case EIO_CUSTOM:
1658 ((void (*)(eio_req *))req->feed) (req); 1803 req->feed (req);
1659 break; 1804 break;
1660 1805
1661 default: 1806 default:
1807 errno = ENOSYS;
1662 req->result = -1; 1808 req->result = -1;
1663 break; 1809 break;
1664 } 1810 }
1665 1811
1666 req->errorno = errno; 1812 req->errorno = errno;
… …
1696eio_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)
1697{ 1843{
1698 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;
1699} 1845}
1700 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
1701eio_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)
1702{ 1858{
1703 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;
1704} 1860}
1705 1861
… …
1729} 1885}
1730 1886
1731eio_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)
1732{ 1888{
1733 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;
1734} 1895}
1735 1896
1736eio_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)
1737{ 1898{
1738 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;
… …
1812eio_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)
1813{ 1974{
1814 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1975 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1815} 1976}
1816 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
1817eio_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)
1818{ 1984{
1819 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1985 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1820} 1986}
1821 1987
… …
1829 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1995 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1830} 1996}
1831 1997
1832eio_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)
1833{ 1999{
1834 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;
1835} 2001}
1836 2002
1837static eio_req * 2003static eio_req *
1838eio__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)
1839{ 2005{
… …
1863eio_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)
1864{ 2030{
1865 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2031 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1866} 2032}
1867 2033
1868eio_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);
1869{ 2035{
1870 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2036 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1871} 2037}
1872 2038
1873#endif 2039#endif
1874 2040
1875eio_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