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

Comparing libeio/eio.c (file contents):
Revision 1.53 by root, Sat Jan 9 10:03:09 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
… …
54#include <sys/statvfs.h> 58#include <sys/statvfs.h>
55#include <limits.h> 59#include <limits.h>
56#include <fcntl.h> 60#include <fcntl.h>
57#include <assert.h> 61#include <assert.h>
58 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
59#ifndef EIO_FINISH 69#ifndef EIO_FINISH
60# 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
61#endif 71#endif
62 72
63#ifndef EIO_DESTROY 73#ifndef EIO_DESTROY
… …
71#ifdef _WIN32 81#ifdef _WIN32
72 82
73 /*doh*/ 83 /*doh*/
74#else 84#else
75 85
76# include "config.h"
77# include <sys/time.h> 86# include <sys/time.h>
78# include <sys/select.h> 87# include <sys/select.h>
79# include <sys/mman.h>
80# include <unistd.h> 88# include <unistd.h>
81# include <utime.h> 89# include <utime.h>
82# include <signal.h> 90# include <signal.h>
83# include <dirent.h> 91# include <dirent.h>
92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
84 96
85/* 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 */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */ 99# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno 100# define D_INO(de) (de)->d_fileno
… …
129#endif 141#endif
130#ifndef D_NAMLEN 142#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 143# define D_NAMLEN(de) strlen ((de)->d_name)
132#endif 144#endif
133 145
134/* number of seconds after which an idle threads exit */
135#define IDLE_TIMEOUT 10
136
137/* used for struct dirent, AIX doesn't provide it */ 146/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 147#ifndef NAME_MAX
139# 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
140#endif 154#endif
141 155
142/* buffer size for various temporary buffers */ 156/* buffer size for various temporary buffers */
143#define EIO_BUFSIZE 65536 157#define EIO_BUFSIZE 65536
144 158
… …
215static unsigned int max_poll_reqs; /* reslock */ 229static unsigned int max_poll_reqs; /* reslock */
216 230
217static volatile unsigned int nreqs; /* reqlock */ 231static volatile unsigned int nreqs; /* reqlock */
218static volatile unsigned int nready; /* reqlock */ 232static volatile unsigned int nready; /* reqlock */
219static volatile unsigned int npending; /* reqlock */ 233static volatile unsigned int npending; /* reqlock */
220static 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 */
221 236
222static mutex_t wrklock = X_MUTEX_INIT; 237static xmutex_t wrklock;
223static mutex_t reslock = X_MUTEX_INIT; 238static xmutex_t reslock;
224static mutex_t reqlock = X_MUTEX_INIT; 239static xmutex_t reqlock;
225static cond_t reqwait = X_COND_INIT; 240static xcond_t reqwait;
226 241
227#if !HAVE_PREADWRITE 242#if !HAVE_PREADWRITE
228/* 243/*
229 * make our pread/pwrite emulation safe against themselves, but not against 244 * make our pread/pwrite emulation safe against themselves, but not against
230 * 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,
231 * but that's your problem, not mine. 246 * but that's your problem, not mine.
232 */ 247 */
233static mutex_t preadwritelock = X_MUTEX_INIT; 248static xmutex_t preadwritelock = X_MUTEX_INIT;
234#endif 249#endif
235 250
236typedef struct etp_worker 251typedef struct etp_worker
237{ 252{
238 /* locked by wrklock */ 253 /* locked by wrklock */
239 struct etp_worker *prev, *next; 254 struct etp_worker *prev, *next;
240 255
241 thread_t tid; 256 xthread_t tid;
242 257
243 /* locked by reslock, reqlock or wrklock */ 258 /* locked by reslock, reqlock or wrklock */
244 ETP_REQ *req; /* currently processed request */ 259 ETP_REQ *req; /* currently processed request */
245 260
246 ETP_WORKER_COMMON 261 ETP_WORKER_COMMON
… …
360 } 375 }
361 376
362 abort (); 377 abort ();
363} 378}
364 379
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
365static void etp_atfork_prepare (void) 388static void etp_atfork_prepare (void)
366{ 389{
367 X_LOCK (wrklock); 390 X_LOCK (wrklock);
368 X_LOCK (reqlock); 391 X_LOCK (reqlock);
369 X_LOCK (reslock); 392 X_LOCK (reslock);
… …
407 idle = 0; 430 idle = 0;
408 nreqs = 0; 431 nreqs = 0;
409 nready = 0; 432 nready = 0;
410 npending = 0; 433 npending = 0;
411 434
412 etp_atfork_parent (); 435 etp_thread_init ();
413} 436}
414 437
415static void 438static void
416etp_once_init (void) 439etp_once_init (void)
417{ 440{
441 etp_thread_init ();
418 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);
419} 443}
420 444
421static int 445static int
422etp_init (void (*want_poll)(void), void (*done_poll)(void)) 446etp_init (void (*want_poll)(void), void (*done_poll)(void))
… …
613} 637}
614 638
615static void etp_set_max_idle (unsigned int nthreads) 639static void etp_set_max_idle (unsigned int nthreads)
616{ 640{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 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;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620} 651}
621 652
622static void etp_set_min_parallel (unsigned int nthreads) 653static void etp_set_min_parallel (unsigned int nthreads)
623{ 654{
… …
751void eio_set_max_idle (unsigned int nthreads) 782void eio_set_max_idle (unsigned int nthreads)
752{ 783{
753 etp_set_max_idle (nthreads); 784 etp_set_max_idle (nthreads);
754} 785}
755 786
787void eio_set_idle_timeout (unsigned int seconds)
788{
789 etp_set_idle_timeout (seconds);
790}
791
756void eio_set_min_parallel (unsigned int nthreads) 792void eio_set_min_parallel (unsigned int nthreads)
757{ 793{
758 etp_set_min_parallel (nthreads); 794 etp_set_min_parallel (nthreads);
759} 795}
760 796
… …
808 844
809 return res; 845 return res;
810} 846}
811#endif 847#endif
812 848
813#ifndef HAVE_FUTIMES 849#ifndef HAVE_UTIMES
814 850
815# undef utimes 851# undef utimes
816# undef futimes
817# define utimes(path,times) eio__utimes (path, times) 852# define utimes(path,times) eio__utimes (path, times)
818# define futimes(fd,times) eio__futimes (fd, times)
819 853
820static int 854static int
821eio__utimes (const char *filename, const struct timeval times[2]) 855eio__utimes (const char *filename, const struct timeval times[2])
822{ 856{
823 if (times) 857 if (times)
… …
830 return utime (filename, &buf); 864 return utime (filename, &buf);
831 } 865 }
832 else 866 else
833 return utime (filename, 0); 867 return utime (filename, 0);
834} 868}
869
870#endif
871
872#ifndef HAVE_FUTIMES
873
874# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times)
835 876
836static int eio__futimes (int fd, const struct timeval tv[2]) 877static int eio__futimes (int fd, const struct timeval tv[2])
837{ 878{
838 errno = ENOSYS; 879 errno = ENOSYS;
839 return -1; 880 return -1;
… …
901 942
902/* sendfile always needs emulation */ 943/* sendfile always needs emulation */
903static ssize_t 944static ssize_t
904eio__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)
905{ 946{
947 ssize_t written = 0;
906 ssize_t res; 948 ssize_t res;
907 949
908 if (!count) 950 if (!count)
909 return 0; 951 return 0;
910 952
953 for (;;)
954 {
911#if HAVE_SENDFILE 955#if HAVE_SENDFILE
912# if __linux 956# if __linux
957 off_t soffset = offset;
913 res = sendfile (ofd, ifd, &offset, count); 958 res = sendfile (ofd, ifd, &soffset, count);
914 959
915# elif __FreeBSD__ 960# elif __FreeBSD__
916 /* 961 /*
917 * 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
918 * wasted on making it similar to other I/O functions. 963 * wasted on making it similar to other I/O functions.
919 */ 964 */
920 {
921 off_t sbytes; 965 off_t sbytes;
922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 966 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
923 967
924 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 968 #if 0 /* according to the manpage, this is correct, but broken behaviour */
925 /* freebsd' sendfile will return 0 on success */ 969 /* freebsd' sendfile will return 0 on success */
926 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 970 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
927 /* not on e.g. EIO or EPIPE - sounds broken */ 971 /* not on e.g. EIO or EPIPE - sounds broken */
928 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 972 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
929 res = sbytes; 973 res = sbytes;
930 #endif 974 #endif
931 975
932 /* according to source inspection, this is correct, and useful behaviour */ 976 /* according to source inspection, this is correct, and useful behaviour */
933 if (sbytes) 977 if (sbytes)
934 res = sbytes; 978 res = sbytes;
935 }
936 979
937# elif defined (__APPLE__) 980# elif defined (__APPLE__)
938
939 {
940 off_t sbytes = count; 981 off_t sbytes = count;
941 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
942 983
943 /* according to the manpage, sbytes is always valid */ 984 /* according to the manpage, sbytes is always valid */
944 if (sbytes) 985 if (sbytes)
945 res = sbytes; 986 res = sbytes;
946 }
947 987
948# elif __hpux 988# elif __hpux
949 res = sendfile (ofd, ifd, offset, count, 0, 0); 989 res = sendfile (ofd, ifd, offset, count, 0, 0);
950 990
951# elif __solaris 991# elif __solaris
952 {
953 struct sendfilevec vec; 992 struct sendfilevec vec;
954 size_t sbytes; 993 size_t sbytes;
955 994
956 vec.sfv_fd = ifd; 995 vec.sfv_fd = ifd;
957 vec.sfv_flag = 0; 996 vec.sfv_flag = 0;
958 vec.sfv_off = offset; 997 vec.sfv_off = offset;
959 vec.sfv_len = count; 998 vec.sfv_len = count;
960 999
961 res = sendfilev (ofd, &vec, 1, &sbytes); 1000 res = sendfilev (ofd, &vec, 1, &sbytes);
962 1001
963 if (res < 0 && sbytes) 1002 if (res < 0 && sbytes)
964 res = sbytes; 1003 res = sbytes;
965 }
966 1004
967# endif 1005# endif
968 1006
969#elif defined (_WIN32) 1007#elif defined (_WIN32)
970
971 /* 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 */
972 { 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 */
973 HANDLE h = TO_SOCKET (ifd); 1012 HANDLE h = TO_SOCKET (ifd);
974 SetFilePointer (h, offset, 0, FILE_BEGIN); 1013 SetFilePointer (h, offset, 0, FILE_BEGIN);
975 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1014 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
976 }
977 1015
978#else 1016#else
979 res = -1; 1017 res = -1;
980 errno = ENOSYS; 1018 errno = ENOSYS;
981#endif 1019#endif
982 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
983 if (res < 0 1044 if (res < 0
984 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1045 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
985 /* BSDs */ 1046 /* BSDs */
986#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... */
987 || errno == ENOTSUP 1048 || errno == ENOTSUP
988#endif 1049#endif
… …
1028} 1089}
1029 1090
1030static signed char 1091static signed char
1031eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1092eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1032{ 1093{
1033 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 */
1034 : 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;
1035} 1098}
1036 1099
1037#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
1038 1101
1039#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 */
… …
1045 unsigned char bits [9 + sizeof (ino_t) * 8]; 1108 unsigned char bits [9 + sizeof (ino_t) * 8];
1046 unsigned char *bit = bits; 1109 unsigned char *bit = bits;
1047 1110
1048 assert (CHAR_BIT == 8); 1111 assert (CHAR_BIT == 8);
1049 assert (sizeof (eio_dirent) * 8 < 256); 1112 assert (sizeof (eio_dirent) * 8 < 256);
1050 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1113 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1051 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1114 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1052 1115
1053 if (size <= EIO_SORT_FAST) 1116 if (size <= EIO_SORT_FAST)
1054 return; 1117 return;
1055 1118
1056 /* 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 */
… …
1211 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1274 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1212 1275
1213 X_LOCK (wrklock); 1276 X_LOCK (wrklock);
1214 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1277 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1215 self->dirp = dirp = opendir (req->ptr1); 1278 self->dirp = dirp = opendir (req->ptr1);
1279
1216 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1280 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1217 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1281 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1218 req->ptr2 = names = malloc (namesalloc); 1282 req->ptr2 = names = malloc (namesalloc);
1219 X_UNLOCK (wrklock); 1283 X_UNLOCK (wrklock);
1220 1284
… …
1232 /* sort etc. */ 1296 /* sort etc. */
1233 req->int1 = flags; 1297 req->int1 = flags;
1234 req->result = dentoffs; 1298 req->result = dentoffs;
1235 1299
1236 if (flags & EIO_READDIR_STAT_ORDER) 1300 if (flags & EIO_READDIR_STAT_ORDER)
1237 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);
1238 else if (flags & EIO_READDIR_DIRS_FIRST) 1302 else if (flags & EIO_READDIR_DIRS_FIRST)
1239 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1303 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1240 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 */
1241 else 1305 else
1242 { 1306 {
… …
1244 eio_dirent *oth = dents + dentoffs; 1308 eio_dirent *oth = dents + dentoffs;
1245 eio_dirent *dir = dents; 1309 eio_dirent *dir = dents;
1246 1310
1247 /* 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 */
1248 /* by walking from both sides and swapping if necessary */ 1312 /* by walking from both sides and swapping if necessary */
1249 /* also clear score, so it doesn't influence sorting */
1250 while (oth > dir) 1313 while (oth > dir)
1251 { 1314 {
1252 if (dir->type == EIO_DT_DIR) 1315 if (dir->type == EIO_DT_DIR)
1253 ++dir; 1316 ++dir;
1254 else if ((--oth)->type == EIO_DT_DIR) 1317 else if ((--oth)->type == EIO_DT_DIR)
… …
1257 1320
1258 ++dir; 1321 ++dir;
1259 } 1322 }
1260 } 1323 }
1261 1324
1262 /* now sort the dirs only */ 1325 /* now sort the dirs only (dirs all have the same score) */
1263 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1326 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1264 } 1327 }
1265 1328
1266 break; 1329 break;
1267 } 1330 }
… …
1387 break; 1450 break;
1388 } 1451 }
1389 } 1452 }
1390} 1453}
1391 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
1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1523#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1524# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1394#else 1525#else
1395 1526
1396int 1527int
1397eio__msync (void *mem, size_t len, int flags) 1528eio__msync (void *mem, size_t len, int flags)
1398{ 1529{
1530 eio_page_align (&mem, &len);
1531
1399 if (EIO_MS_ASYNC != MS_SYNC 1532 if (EIO_MS_ASYNC != MS_SYNC
1400 || EIO_MS_INVALIDATE != MS_INVALIDATE 1533 || EIO_MS_INVALIDATE != MS_INVALIDATE
1401 || EIO_MS_SYNC != MS_SYNC) 1534 || EIO_MS_SYNC != MS_SYNC)
1402 { 1535 {
1403 flags = 0 1536 flags = 0
… …
1410} 1543}
1411 1544
1412#endif 1545#endif
1413 1546
1414int 1547int
1415eio__mtouch (void *mem, size_t len, int flags) 1548eio__mtouch (eio_req *req)
1416{ 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 {
1417 intptr_t addr = (intptr_t)mem; 1557 intptr_t addr = (intptr_t)mem;
1418 intptr_t end = addr + len; 1558 intptr_t end = addr + len;
1419#ifdef PAGESIZE 1559 intptr_t page = eio_pagesize ();
1420 const intptr_t page = PAGESIZE;
1421#else
1422 static intptr_t page;
1423 1560
1424 if (!page)
1425 page = sysconf (_SC_PAGESIZE);
1426#endif
1427
1428 /* round down to start of page, although this is probably useless */
1429 addr &= ~(page - 1); /* assume page size is always a power of two */
1430
1431 if (addr < end) 1561 if (addr < end)
1432 if (flags & EIO_MT_MODIFY) /* modify */ 1562 if (flags & EIO_MT_MODIFY) /* modify */
1433 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));
1434 else 1564 else
1435 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 }
1436 1567
1437 return 0; 1568 return 0;
1438} 1569}
1439 1570
1440/*****************************************************************************/ 1571/*****************************************************************************/
… …
1474 if (req) 1605 if (req)
1475 break; 1606 break;
1476 1607
1477 ++idle; 1608 ++idle;
1478 1609
1479 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1610 ts.tv_sec = time (0) + idle_timeout;
1480 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1611 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1481 { 1612 {
1482 if (idle > max_idle) 1613 if (idle > max_idle)
1483 { 1614 {
1484 --idle; 1615 --idle;
… …
1563 return 0; \ 1694 return 0; \
1564 } 1695 }
1565 1696
1566static void eio_execute (etp_worker *self, eio_req *req) 1697static void eio_execute (etp_worker *self, eio_req *req)
1567{ 1698{
1568 errno = 0;
1569
1570 switch (req->type) 1699 switch (req->type)
1571 { 1700 {
1572 case EIO_READ: ALLOC (req->size); 1701 case EIO_READ: ALLOC (req->size);
1573 req->result = req->offs >= 0 1702 req->result = req->offs >= 0
1574 ? pread (req->int1, req->ptr2, req->size, req->offs) 1703 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1606 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1735 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1607 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;
1608 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1737 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1609 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1738 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1610 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1739 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1611 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;
1612 1741
1613 case EIO_READLINK: ALLOC (NAME_MAX); 1742 case EIO_READLINK: ALLOC (PATH_MAX);
1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1743 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1615 1744
1616 case EIO_SYNC: req->result = 0; sync (); break; 1745 case EIO_SYNC: req->result = 0; sync (); break;
1617 case EIO_FSYNC: req->result = fsync (req->int1); break; 1746 case EIO_FSYNC: req->result = fsync (req->int1); break;
1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1747 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1619 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;
1620 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;
1621 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;
1622 1753
1623 case EIO_READDIR: eio__scandir (req, self); break; 1754 case EIO_READDIR: eio__scandir (req, self); break;
1624 1755
1625 case EIO_BUSY: 1756 case EIO_BUSY:
1626#ifdef _WIN32 1757#ifdef _WIN32
1627 Sleep (req->nv1 * 1000.); 1758 Sleep (req->nv1 * 1e3);
1628#else 1759#else
1629 { 1760 {
1630 struct timeval tv; 1761 struct timeval tv;
1631 1762
1632 tv.tv_sec = req->nv1; 1763 tv.tv_sec = req->nv1;
1633 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1764 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1634 1765
1635 req->result = select (0, 0, 0, 0, &tv); 1766 req->result = select (0, 0, 0, 0, &tv);
1636 } 1767 }
1637#endif 1768#endif
1638 break; 1769 break;
… …
1653 times = tv; 1784 times = tv;
1654 } 1785 }
1655 else 1786 else
1656 times = 0; 1787 times = 0;
1657 1788
1658
1659 req->result = req->type == EIO_FUTIME 1789 req->result = req->type == EIO_FUTIME
1660 ? futimes (req->int1, times) 1790 ? futimes (req->int1, times)
1661 : utimes (req->ptr1, times); 1791 : utimes (req->ptr1, times);
1662 } 1792 }
1663 break; 1793 break;
… …
1668 case EIO_NOP: 1798 case EIO_NOP:
1669 req->result = 0; 1799 req->result = 0;
1670 break; 1800 break;
1671 1801
1672 case EIO_CUSTOM: 1802 case EIO_CUSTOM:
1673 ((void (*)(eio_req *))req->feed) (req); 1803 req->feed (req);
1674 break; 1804 break;
1675 1805
1676 default: 1806 default:
1807 errno = ENOSYS;
1677 req->result = -1; 1808 req->result = -1;
1678 break; 1809 break;
1679 } 1810 }
1680 1811
1681 req->errorno = errno; 1812 req->errorno = errno;
… …
1711eio_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)
1712{ 1843{
1713 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;
1714} 1845}
1715 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
1716eio_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)
1717{ 1858{
1718 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;
1719} 1860}
1720 1861
… …
1854 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1995 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1855} 1996}
1856 1997
1857eio_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)
1858{ 1999{
1859 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;
1860} 2001}
1861 2002
1862static eio_req * 2003static eio_req *
1863eio__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)
1864{ 2005{
… …
1888eio_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)
1889{ 2030{
1890 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2031 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1891} 2032}
1892 2033
1893eio_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);
1894{ 2035{
1895 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2036 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1896} 2037}
1897 2038
1898#endif 2039#endif
1899 2040
1900eio_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