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

Comparing libeio/eio.c (file contents):
Revision 1.52 by root, Fri Jan 8 09:29:56 2010 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) 2007,2008,2009,2010 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
968 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* freebsd' sendfile will return 0 on success */ 969 /* freebsd' sendfile will return 0 on success */
925 /* 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 */
926 /* not on e.g. EIO or EPIPE - sounds broken */ 971 /* not on e.g. EIO or EPIPE - sounds broken */
927 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 972 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
928 res = sbytes; 973 res = sbytes;
929 } 974 #endif
975
976 /* according to source inspection, this is correct, and useful behaviour */
977 if (sbytes)
978 res = sbytes;
930 979
931# elif defined (__APPLE__) 980# elif defined (__APPLE__)
932
933 {
934 off_t sbytes = count; 981 off_t sbytes = count;
935 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
936 983
937 /* according to the manpage, sbytes is always valid */ 984 /* according to the manpage, sbytes is always valid */
938 if (sbytes) 985 if (sbytes)
939 res = sbytes; 986 res = sbytes;
940 }
941 987
942# elif __hpux 988# elif __hpux
943 res = sendfile (ofd, ifd, offset, count, 0, 0); 989 res = sendfile (ofd, ifd, offset, count, 0, 0);
944 990
945# elif __solaris 991# elif __solaris
946 {
947 struct sendfilevec vec; 992 struct sendfilevec vec;
948 size_t sbytes; 993 size_t sbytes;
949 994
950 vec.sfv_fd = ifd; 995 vec.sfv_fd = ifd;
951 vec.sfv_flag = 0; 996 vec.sfv_flag = 0;
952 vec.sfv_off = offset; 997 vec.sfv_off = offset;
953 vec.sfv_len = count; 998 vec.sfv_len = count;
954 999
955 res = sendfilev (ofd, &vec, 1, &sbytes); 1000 res = sendfilev (ofd, &vec, 1, &sbytes);
956 1001
957 if (res < 0 && sbytes) 1002 if (res < 0 && sbytes)
958 res = sbytes; 1003 res = sbytes;
959 }
960 1004
961# endif 1005# endif
962 1006
963#elif defined (_WIN32) 1007#elif defined (_WIN32)
964
965 /* 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 */
966 {
967 HANDLE h = TO_SOCKET (ifd); 1009 HANDLE h = TO_SOCKET (ifd);
968 SetFilePointer (h, offset, 0, FILE_BEGIN); 1010 SetFilePointer (h, offset, 0, FILE_BEGIN);
969 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1011 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
970 }
971 1012
972#else 1013#else
973 res = -1; 1014 res = -1;
974 errno = ENOSYS; 1015 errno = ENOSYS;
975#endif 1016#endif
976 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
977 if (res < 0 1041 if (res < 0
978 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1042 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
979 /* BSDs */ 1043 /* BSDs */
980#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1044#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
981 || errno == ENOTSUP 1045 || errno == ENOTSUP
982#endif 1046#endif
… …
1022} 1086}
1023 1087
1024static signed char 1088static signed char
1025eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1089eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1026{ 1090{
1027 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 */
1028 : 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;
1029} 1095}
1030 1096
1031#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
1032 1098
1033#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 */
… …
1039 unsigned char bits [9 + sizeof (ino_t) * 8]; 1105 unsigned char bits [9 + sizeof (ino_t) * 8];
1040 unsigned char *bit = bits; 1106 unsigned char *bit = bits;
1041 1107
1042 assert (CHAR_BIT == 8); 1108 assert (CHAR_BIT == 8);
1043 assert (sizeof (eio_dirent) * 8 < 256); 1109 assert (sizeof (eio_dirent) * 8 < 256);
1044 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1110 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1045 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1111 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1046 1112
1047 if (size <= EIO_SORT_FAST) 1113 if (size <= EIO_SORT_FAST)
1048 return; 1114 return;
1049 1115
1050 /* 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 */
… …
1205 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1271 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1206 1272
1207 X_LOCK (wrklock); 1273 X_LOCK (wrklock);
1208 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1274 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1209 self->dirp = dirp = opendir (req->ptr1); 1275 self->dirp = dirp = opendir (req->ptr1);
1276
1210 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1277 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1211 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1278 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1212 req->ptr2 = names = malloc (namesalloc); 1279 req->ptr2 = names = malloc (namesalloc);
1213 X_UNLOCK (wrklock); 1280 X_UNLOCK (wrklock);
1214 1281
… …
1226 /* sort etc. */ 1293 /* sort etc. */
1227 req->int1 = flags; 1294 req->int1 = flags;
1228 req->result = dentoffs; 1295 req->result = dentoffs;
1229 1296
1230 if (flags & EIO_READDIR_STAT_ORDER) 1297 if (flags & EIO_READDIR_STAT_ORDER)
1231 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);
1232 else if (flags & EIO_READDIR_DIRS_FIRST) 1299 else if (flags & EIO_READDIR_DIRS_FIRST)
1233 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1300 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1234 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 */
1235 else 1302 else
1236 { 1303 {
… …
1238 eio_dirent *oth = dents + dentoffs; 1305 eio_dirent *oth = dents + dentoffs;
1239 eio_dirent *dir = dents; 1306 eio_dirent *dir = dents;
1240 1307
1241 /* 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 */
1242 /* by walking from both sides and swapping if necessary */ 1309 /* by walking from both sides and swapping if necessary */
1243 /* also clear score, so it doesn't influence sorting */
1244 while (oth > dir) 1310 while (oth > dir)
1245 { 1311 {
1246 if (dir->type == EIO_DT_DIR) 1312 if (dir->type == EIO_DT_DIR)
1247 ++dir; 1313 ++dir;
1248 else if ((--oth)->type == EIO_DT_DIR) 1314 else if ((--oth)->type == EIO_DT_DIR)
… …
1251 1317
1252 ++dir; 1318 ++dir;
1253 } 1319 }
1254 } 1320 }
1255 1321
1256 /* now sort the dirs only */ 1322 /* now sort the dirs only (dirs all have the same score) */
1257 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1323 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1258 } 1324 }
1259 1325
1260 break; 1326 break;
1261 } 1327 }
… …
1381 break; 1447 break;
1382 } 1448 }
1383 } 1449 }
1384} 1450}
1385 1451
1452#ifdef PAGESIZE
1453# define eio_pagesize() PAGESIZE
1454#else
1455static intptr_t
1456eio_pagesize (void)
1457{
1458 static intptr_t page;
1459
1460 if (!page)
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 */
1492 #endif
1493
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 }
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
1386#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1520#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1387# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1521# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1388#else 1522#else
1389 1523
1390int 1524int
1391eio__msync (void *mem, size_t len, int flags) 1525eio__msync (void *mem, size_t len, int flags)
1392{ 1526{
1527 eio_page_align (&mem, &len);
1528
1393 if (EIO_MS_ASYNC != MS_SYNC 1529 if (EIO_MS_ASYNC != MS_SYNC
1394 || EIO_MS_INVALIDATE != MS_INVALIDATE 1530 || EIO_MS_INVALIDATE != MS_INVALIDATE
1395 || EIO_MS_SYNC != MS_SYNC) 1531 || EIO_MS_SYNC != MS_SYNC)
1396 { 1532 {
1397 flags = 0 1533 flags = 0
… …
1404} 1540}
1405 1541
1406#endif 1542#endif
1407 1543
1408int 1544int
1409eio__mtouch (void *mem, size_t len, int flags) 1545eio__mtouch (eio_req *req)
1410{ 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 {
1411 intptr_t addr = (intptr_t)mem; 1554 intptr_t addr = (intptr_t)mem;
1412 intptr_t end = addr + len; 1555 intptr_t end = addr + len;
1413#ifdef PAGESIZE 1556 intptr_t page = eio_pagesize ();
1414 const intptr_t page = PAGESIZE;
1415#else
1416 static intptr_t page;
1417 1557
1418 if (!page)
1419 page = sysconf (_SC_PAGESIZE);
1420#endif
1421
1422 /* round down to start of page, although this is probably useless */
1423 addr &= ~(page - 1); /* assume page size is always a power of two */
1424
1425 if (addr < end) 1558 if (addr < end)
1426 if (flags & EIO_MT_MODIFY) /* modify */ 1559 if (flags & EIO_MT_MODIFY) /* modify */
1427 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));
1428 else 1561 else
1429 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 }
1430 1564
1431 return 0; 1565 return 0;
1432} 1566}
1433 1567
1434/*****************************************************************************/ 1568/*****************************************************************************/
… …
1468 if (req) 1602 if (req)
1469 break; 1603 break;
1470 1604
1471 ++idle; 1605 ++idle;
1472 1606
1473 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1607 ts.tv_sec = time (0) + idle_timeout;
1474 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1608 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1475 { 1609 {
1476 if (idle > max_idle) 1610 if (idle > max_idle)
1477 { 1611 {
1478 --idle; 1612 --idle;
… …
1557 return 0; \ 1691 return 0; \
1558 } 1692 }
1559 1693
1560static void eio_execute (etp_worker *self, eio_req *req) 1694static void eio_execute (etp_worker *self, eio_req *req)
1561{ 1695{
1562 errno = 0;
1563
1564 switch (req->type) 1696 switch (req->type)
1565 { 1697 {
1566 case EIO_READ: ALLOC (req->size); 1698 case EIO_READ: ALLOC (req->size);
1567 req->result = req->offs >= 0 1699 req->result = req->offs >= 0
1568 ? pread (req->int1, req->ptr2, req->size, req->offs) 1700 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1600 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1732 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1601 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;
1602 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1734 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1603 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1735 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1604 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1736 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1605 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;
1606 1738
1607 case EIO_READLINK: ALLOC (NAME_MAX); 1739 case EIO_READLINK: ALLOC (PATH_MAX);
1608 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1740 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1609 1741
1610 case EIO_SYNC: req->result = 0; sync (); break; 1742 case EIO_SYNC: req->result = 0; sync (); break;
1611 case EIO_FSYNC: req->result = fsync (req->int1); break; 1743 case EIO_FSYNC: req->result = fsync (req->int1); break;
1612 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1744 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1613 case EIO_MSYNC: req->result = eio__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;
1614 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;
1615 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;
1616 1750
1617 case EIO_READDIR: eio__scandir (req, self); break; 1751 case EIO_READDIR: eio__scandir (req, self); break;
1618 1752
1619 case EIO_BUSY: 1753 case EIO_BUSY:
1620#ifdef _WIN32 1754#ifdef _WIN32
1621 Sleep (req->nv1 * 1000.); 1755 Sleep (req->nv1 * 1e3);
1622#else 1756#else
1623 { 1757 {
1624 struct timeval tv; 1758 struct timeval tv;
1625 1759
1626 tv.tv_sec = req->nv1; 1760 tv.tv_sec = req->nv1;
1627 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1761 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1628 1762
1629 req->result = select (0, 0, 0, 0, &tv); 1763 req->result = select (0, 0, 0, 0, &tv);
1630 } 1764 }
1631#endif 1765#endif
1632 break; 1766 break;
… …
1647 times = tv; 1781 times = tv;
1648 } 1782 }
1649 else 1783 else
1650 times = 0; 1784 times = 0;
1651 1785
1652
1653 req->result = req->type == EIO_FUTIME 1786 req->result = req->type == EIO_FUTIME
1654 ? futimes (req->int1, times) 1787 ? futimes (req->int1, times)
1655 : utimes (req->ptr1, times); 1788 : utimes (req->ptr1, times);
1656 } 1789 }
1657 break; 1790 break;
… …
1662 case EIO_NOP: 1795 case EIO_NOP:
1663 req->result = 0; 1796 req->result = 0;
1664 break; 1797 break;
1665 1798
1666 case EIO_CUSTOM: 1799 case EIO_CUSTOM:
1667 ((void (*)(eio_req *))req->feed) (req); 1800 req->feed (req);
1668 break; 1801 break;
1669 1802
1670 default: 1803 default:
1804 errno = ENOSYS;
1671 req->result = -1; 1805 req->result = -1;
1672 break; 1806 break;
1673 } 1807 }
1674 1808
1675 req->errorno = errno; 1809 req->errorno = errno;
… …
1705eio_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)
1706{ 1840{
1707 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;
1708} 1842}
1709 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
1710eio_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)
1711{ 1855{
1712 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;
1713} 1857}
1714 1858
… …
1848 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1992 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1849} 1993}
1850 1994
1851eio_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)
1852{ 1996{
1853 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;
1854} 1998}
1855 1999
1856static eio_req * 2000static eio_req *
1857eio__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)
1858{ 2002{
… …
1882eio_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)
1883{ 2027{
1884 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2028 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1885} 2029}
1886 2030
1887eio_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);
1888{ 2032{
1889 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2033 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1890} 2034}
1891 2035
1892#endif 2036#endif
1893 2037
1894eio_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