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

Comparing libeio/eio.c (file contents):
Revision 1.56 by root, Sun Sep 12 03:36:28 2010 UTC vs.
Revision 1.65 by root, Mon May 30 12:56:50 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 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
… …
71#ifdef _WIN32 75#ifdef _WIN32
72 76
73 /*doh*/ 77 /*doh*/
74#else 78#else
75 79
76# include "config.h"
77# include <sys/time.h> 80# include <sys/time.h>
78# include <sys/select.h> 81# include <sys/select.h>
79# include <unistd.h> 82# include <unistd.h>
80# include <utime.h> 83# include <utime.h>
81# include <signal.h> 84# include <signal.h>
82# include <dirent.h> 85# include <dirent.h>
83 86
84#if _POSIX_MEMLOCK || _POSIX_MAPPED_FILES 87#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85# include <sys/mman.h> 88# include <sys/mman.h>
86#endif 89#endif
87 90
88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 91/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 92# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
… …
132#endif 135#endif
133#ifndef D_NAMLEN 136#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name) 137# define D_NAMLEN(de) strlen ((de)->d_name)
135#endif 138#endif
136 139
137/* number of seconds after which an idle threads exit */
138#define IDLE_TIMEOUT 10
139
140/* used for struct dirent, AIX doesn't provide it */ 140/* used for struct dirent, AIX doesn't provide it */
141#ifndef NAME_MAX 141#ifndef NAME_MAX
142# define NAME_MAX 4096 142# define NAME_MAX 4096
143#endif 143#endif
144 144
… …
223static unsigned int max_poll_reqs; /* reslock */ 223static unsigned int max_poll_reqs; /* reslock */
224 224
225static volatile unsigned int nreqs; /* reqlock */ 225static volatile unsigned int nreqs; /* reqlock */
226static volatile unsigned int nready; /* reqlock */ 226static volatile unsigned int nready; /* reqlock */
227static volatile unsigned int npending; /* reqlock */ 227static volatile unsigned int npending; /* reqlock */
228static volatile unsigned int max_idle = 4; 228static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
229static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
229 230
230static xmutex_t wrklock = X_MUTEX_INIT; 231static xmutex_t wrklock;
231static xmutex_t reslock = X_MUTEX_INIT; 232static xmutex_t reslock;
232static xmutex_t reqlock = X_MUTEX_INIT; 233static xmutex_t reqlock;
233static xcond_t reqwait = X_COND_INIT; 234static xcond_t reqwait;
234 235
235#if !HAVE_PREADWRITE 236#if !HAVE_PREADWRITE
236/* 237/*
237 * make our pread/pwrite emulation safe against themselves, but not against 238 * make our pread/pwrite emulation safe against themselves, but not against
238 * normal read/write by using a mutex. slows down execution a lot, 239 * normal read/write by using a mutex. slows down execution a lot,
… …
368 } 369 }
369 370
370 abort (); 371 abort ();
371} 372}
372 373
374static void etp_thread_init (void)
375{
376 X_MUTEX_CREATE (wrklock);
377 X_MUTEX_CREATE (reslock);
378 X_MUTEX_CREATE (reqlock);
379 X_COND_CREATE (reqwait);
380}
381
373static void etp_atfork_prepare (void) 382static void etp_atfork_prepare (void)
374{ 383{
375 X_LOCK (wrklock); 384 X_LOCK (wrklock);
376 X_LOCK (reqlock); 385 X_LOCK (reqlock);
377 X_LOCK (reslock); 386 X_LOCK (reslock);
… …
415 idle = 0; 424 idle = 0;
416 nreqs = 0; 425 nreqs = 0;
417 nready = 0; 426 nready = 0;
418 npending = 0; 427 npending = 0;
419 428
420 etp_atfork_parent (); 429 etp_thread_init ();
421} 430}
422 431
423static void 432static void
424etp_once_init (void) 433etp_once_init (void)
425{ 434{
435 etp_thread_init ();
426 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 436 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427} 437}
428 438
429static int 439static int
430etp_init (void (*want_poll)(void), void (*done_poll)(void)) 440etp_init (void (*want_poll)(void), void (*done_poll)(void))
… …
621} 631}
622 632
623static void etp_set_max_idle (unsigned int nthreads) 633static void etp_set_max_idle (unsigned int nthreads)
624{ 634{
625 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 635 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626 max_idle = nthreads <= 0 ? 1 : nthreads; 636 max_idle = nthreads;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
638}
639
640static void etp_set_idle_timeout (unsigned int seconds)
641{
642 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
643 idle_timeout = seconds;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 644 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628} 645}
629 646
630static void etp_set_min_parallel (unsigned int nthreads) 647static void etp_set_min_parallel (unsigned int nthreads)
631{ 648{
… …
759void eio_set_max_idle (unsigned int nthreads) 776void eio_set_max_idle (unsigned int nthreads)
760{ 777{
761 etp_set_max_idle (nthreads); 778 etp_set_max_idle (nthreads);
762} 779}
763 780
781void eio_set_idle_timeout (unsigned int seconds)
782{
783 etp_set_idle_timeout (seconds);
784}
785
764void eio_set_min_parallel (unsigned int nthreads) 786void eio_set_min_parallel (unsigned int nthreads)
765{ 787{
766 etp_set_min_parallel (nthreads); 788 etp_set_min_parallel (nthreads);
767} 789}
768 790
… …
816 838
817 return res; 839 return res;
818} 840}
819#endif 841#endif
820 842
821#ifndef HAVE_FUTIMES 843#ifndef HAVE_UTIMES
822 844
823# undef utimes 845# undef utimes
824# undef futimes
825# define utimes(path,times) eio__utimes (path, times) 846# define utimes(path,times) eio__utimes (path, times)
826# define futimes(fd,times) eio__futimes (fd, times)
827 847
828static int 848static int
829eio__utimes (const char *filename, const struct timeval times[2]) 849eio__utimes (const char *filename, const struct timeval times[2])
830{ 850{
831 if (times) 851 if (times)
… …
838 return utime (filename, &buf); 858 return utime (filename, &buf);
839 } 859 }
840 else 860 else
841 return utime (filename, 0); 861 return utime (filename, 0);
842} 862}
863
864#endif
865
866#ifndef HAVE_FUTIMES
867
868# undef futimes
869# define futimes(fd,times) eio__futimes (fd, times)
843 870
844static int eio__futimes (int fd, const struct timeval tv[2]) 871static int eio__futimes (int fd, const struct timeval tv[2])
845{ 872{
846 errno = ENOSYS; 873 errno = ENOSYS;
847 return -1; 874 return -1;
… …
1036} 1063}
1037 1064
1038static signed char 1065static signed char
1039eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1066eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1040{ 1067{
1041 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1068 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1042 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1069 : a->inode < b->inode ? -1
1070 : a->inode > b->inode ? 1
1071 : 0;
1043} 1072}
1044 1073
1045#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1074#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1046 1075
1047#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1076#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
… …
1053 unsigned char bits [9 + sizeof (ino_t) * 8]; 1082 unsigned char bits [9 + sizeof (ino_t) * 8];
1054 unsigned char *bit = bits; 1083 unsigned char *bit = bits;
1055 1084
1056 assert (CHAR_BIT == 8); 1085 assert (CHAR_BIT == 8);
1057 assert (sizeof (eio_dirent) * 8 < 256); 1086 assert (sizeof (eio_dirent) * 8 < 256);
1058 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1087 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1059 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1088 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1060 1089
1061 if (size <= EIO_SORT_FAST) 1090 if (size <= EIO_SORT_FAST)
1062 return; 1091 return;
1063 1092
1064 /* first prepare an array of bits to test in our radix sort */ 1093 /* first prepare an array of bits to test in our radix sort */
… …
1219 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1248 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1220 1249
1221 X_LOCK (wrklock); 1250 X_LOCK (wrklock);
1222 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1251 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1223 self->dirp = dirp = opendir (req->ptr1); 1252 self->dirp = dirp = opendir (req->ptr1);
1253
1224 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1254 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1225 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1255 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1226 req->ptr2 = names = malloc (namesalloc); 1256 req->ptr2 = names = malloc (namesalloc);
1227 X_UNLOCK (wrklock); 1257 X_UNLOCK (wrklock);
1228 1258
… …
1240 /* sort etc. */ 1270 /* sort etc. */
1241 req->int1 = flags; 1271 req->int1 = flags;
1242 req->result = dentoffs; 1272 req->result = dentoffs;
1243 1273
1244 if (flags & EIO_READDIR_STAT_ORDER) 1274 if (flags & EIO_READDIR_STAT_ORDER)
1245 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1275 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1246 else if (flags & EIO_READDIR_DIRS_FIRST) 1276 else if (flags & EIO_READDIR_DIRS_FIRST)
1247 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1277 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1248 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1278 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1249 else 1279 else
1250 { 1280 {
… …
1252 eio_dirent *oth = dents + dentoffs; 1282 eio_dirent *oth = dents + dentoffs;
1253 eio_dirent *dir = dents; 1283 eio_dirent *dir = dents;
1254 1284
1255 /* now partition dirs to the front, and non-dirs to the back */ 1285 /* now partition dirs to the front, and non-dirs to the back */
1256 /* by walking from both sides and swapping if necessary */ 1286 /* by walking from both sides and swapping if necessary */
1257 /* also clear score, so it doesn't influence sorting */
1258 while (oth > dir) 1287 while (oth > dir)
1259 { 1288 {
1260 if (dir->type == EIO_DT_DIR) 1289 if (dir->type == EIO_DT_DIR)
1261 ++dir; 1290 ++dir;
1262 else if ((--oth)->type == EIO_DT_DIR) 1291 else if ((--oth)->type == EIO_DT_DIR)
… …
1265 1294
1266 ++dir; 1295 ++dir;
1267 } 1296 }
1268 } 1297 }
1269 1298
1270 /* now sort the dirs only */ 1299 /* now sort the dirs only (dirs all have the same score) */
1271 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1300 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1272 } 1301 }
1273 1302
1274 break; 1303 break;
1275 } 1304 }
… …
1426 /* round up length */ 1455 /* round up length */
1427 *length = (*length + mask) & ~mask; 1456 *length = (*length + mask) & ~mask;
1428} 1457}
1429 1458
1430#if !_POSIX_MEMLOCK 1459#if !_POSIX_MEMLOCK
1431# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1432# define eio__mlockall(a) ((errno = ENOSYS), -1) 1460# define eio__mlockall(a) ((errno = ENOSYS), -1)
1433#else 1461#else
1434
1435static int
1436eio__mlock (void *addr, size_t length)
1437{
1438 eio_page_align (&addr, &length);
1439
1440 mlock (addr, length);
1441}
1442 1462
1443static int 1463static int
1444eio__mlockall (int flags) 1464eio__mlockall (int flags)
1445{ 1465{
1446 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 1466 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
… …
1454 flags = 0 1474 flags = 0
1455 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) 1475 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1456 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); 1476 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1457 } 1477 }
1458 1478
1459 mlockall (flags); 1479 return mlockall (flags);
1460} 1480}
1481#endif
1482
1483#if !_POSIX_MEMLOCK_RANGE
1484# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1485#else
1486
1487static int
1488eio__mlock (void *addr, size_t length)
1489{
1490 eio_page_align (&addr, &length);
1491
1492 return mlock (addr, length);
1493}
1494
1461#endif 1495#endif
1462 1496
1463#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1497#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1464# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1498# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1465#else 1499#else
… …
1541 if (req) 1575 if (req)
1542 break; 1576 break;
1543 1577
1544 ++idle; 1578 ++idle;
1545 1579
1546 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1580 ts.tv_sec = time (0) + idle_timeout;
1547 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1581 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1548 { 1582 {
1549 if (idle > max_idle) 1583 if (idle > max_idle)
1550 { 1584 {
1551 --idle; 1585 --idle;
… …
1630 return 0; \ 1664 return 0; \
1631 } 1665 }
1632 1666
1633static void eio_execute (etp_worker *self, eio_req *req) 1667static void eio_execute (etp_worker *self, eio_req *req)
1634{ 1668{
1635 errno = 0;
1636
1637 switch (req->type) 1669 switch (req->type)
1638 { 1670 {
1639 case EIO_READ: ALLOC (req->size); 1671 case EIO_READ: ALLOC (req->size);
1640 req->result = req->offs >= 0 1672 req->result = req->offs >= 0
1641 ? pread (req->int1, req->ptr2, req->size, req->offs) 1673 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1673 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1705 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1674 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1706 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1675 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1707 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1676 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1708 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1677 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1709 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1678 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1710 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1679 1711
1680 case EIO_READLINK: ALLOC (PATH_MAX); 1712 case EIO_READLINK: ALLOC (PATH_MAX);
1681 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 1713 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1682 1714
1683 case EIO_SYNC: req->result = 0; sync (); break; 1715 case EIO_SYNC: req->result = 0; sync (); break;
… …
1691 1723
1692 case EIO_READDIR: eio__scandir (req, self); break; 1724 case EIO_READDIR: eio__scandir (req, self); break;
1693 1725
1694 case EIO_BUSY: 1726 case EIO_BUSY:
1695#ifdef _WIN32 1727#ifdef _WIN32
1696 Sleep (req->nv1 * 1000.); 1728 Sleep (req->nv1 * 1e3);
1697#else 1729#else
1698 { 1730 {
1699 struct timeval tv; 1731 struct timeval tv;
1700 1732
1701 tv.tv_sec = req->nv1; 1733 tv.tv_sec = req->nv1;
1702 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1734 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1703 1735
1704 req->result = select (0, 0, 0, 0, &tv); 1736 req->result = select (0, 0, 0, 0, &tv);
1705 } 1737 }
1706#endif 1738#endif
1707 break; 1739 break;
… …
1722 times = tv; 1754 times = tv;
1723 } 1755 }
1724 else 1756 else
1725 times = 0; 1757 times = 0;
1726 1758
1727
1728 req->result = req->type == EIO_FUTIME 1759 req->result = req->type == EIO_FUTIME
1729 ? futimes (req->int1, times) 1760 ? futimes (req->int1, times)
1730 : utimes (req->ptr1, times); 1761 : utimes (req->ptr1, times);
1731 } 1762 }
1732 break; 1763 break;
… …
1741 case EIO_CUSTOM: 1772 case EIO_CUSTOM:
1742 ((void (*)(eio_req *))req->feed) (req); 1773 ((void (*)(eio_req *))req->feed) (req);
1743 break; 1774 break;
1744 1775
1745 default: 1776 default:
1777 errno = ENOSYS;
1746 req->result = -1; 1778 req->result = -1;
1747 break; 1779 break;
1748 } 1780 }
1749 1781
1750 req->errorno = errno; 1782 req->errorno = errno;
… …
1933 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1965 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1934} 1966}
1935 1967
1936eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1968eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1937{ 1969{
1938 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1970 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1939} 1971}
1940 1972
1941static eio_req * 1973static eio_req *
1942eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1974eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1943{ 1975{

Diff Legend

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