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 1.62 by root, Tue Feb 15 03:15:16 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 *
… …
74#else 74#else
75 75
76# include "config.h" 76# include "config.h"
77# include <sys/time.h> 77# include <sys/time.h>
78# include <sys/select.h> 78# include <sys/select.h>
79# include <sys/mman.h>
80# include <unistd.h> 79# include <unistd.h>
81# include <utime.h> 80# include <utime.h>
82# include <signal.h> 81# include <signal.h>
83# include <dirent.h> 82# include <dirent.h>
83
84#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85# include <sys/mman.h>
86#endif
84 87
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */ 90# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno 91# define D_INO(de) (de)->d_fileno
… …
129#endif 132#endif
130#ifndef D_NAMLEN 133#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 134# define D_NAMLEN(de) strlen ((de)->d_name)
132#endif 135#endif
133 136
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 */ 137/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 138#ifndef NAME_MAX
139# define NAME_MAX 4096 139# define NAME_MAX 4096
140#endif
141
142/* used for readlink etc. */
143#ifndef PATH_MAX
144# define PATH_MAX 4096
140#endif 145#endif
141 146
142/* buffer size for various temporary buffers */ 147/* buffer size for various temporary buffers */
143#define EIO_BUFSIZE 65536 148#define EIO_BUFSIZE 65536
144 149
… …
215static unsigned int max_poll_reqs; /* reslock */ 220static unsigned int max_poll_reqs; /* reslock */
216 221
217static volatile unsigned int nreqs; /* reqlock */ 222static volatile unsigned int nreqs; /* reqlock */
218static volatile unsigned int nready; /* reqlock */ 223static volatile unsigned int nready; /* reqlock */
219static volatile unsigned int npending; /* reqlock */ 224static volatile unsigned int npending; /* reqlock */
220static volatile unsigned int max_idle = 4; 225static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
226static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
221 227
222static mutex_t wrklock = X_MUTEX_INIT; 228static xmutex_t wrklock;
223static mutex_t reslock = X_MUTEX_INIT; 229static xmutex_t reslock;
224static mutex_t reqlock = X_MUTEX_INIT; 230static xmutex_t reqlock;
225static cond_t reqwait = X_COND_INIT; 231static xcond_t reqwait;
226 232
227#if !HAVE_PREADWRITE 233#if !HAVE_PREADWRITE
228/* 234/*
229 * make our pread/pwrite emulation safe against themselves, but not against 235 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot, 236 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine. 237 * but that's your problem, not mine.
232 */ 238 */
233static mutex_t preadwritelock = X_MUTEX_INIT; 239static xmutex_t preadwritelock = X_MUTEX_INIT;
234#endif 240#endif
235 241
236typedef struct etp_worker 242typedef struct etp_worker
237{ 243{
238 /* locked by wrklock */ 244 /* locked by wrklock */
239 struct etp_worker *prev, *next; 245 struct etp_worker *prev, *next;
240 246
241 thread_t tid; 247 xthread_t tid;
242 248
243 /* locked by reslock, reqlock or wrklock */ 249 /* locked by reslock, reqlock or wrklock */
244 ETP_REQ *req; /* currently processed request */ 250 ETP_REQ *req; /* currently processed request */
245 251
246 ETP_WORKER_COMMON 252 ETP_WORKER_COMMON
… …
360 } 366 }
361 367
362 abort (); 368 abort ();
363} 369}
364 370
371static void etp_thread_init (void)
372{
373 X_MUTEX_CREATE (wrklock);
374 X_MUTEX_CREATE (reslock);
375 X_MUTEX_CREATE (reqlock);
376 X_COND_CREATE (reqwait);
377}
378
365static void etp_atfork_prepare (void) 379static void etp_atfork_prepare (void)
366{ 380{
367 X_LOCK (wrklock); 381 X_LOCK (wrklock);
368 X_LOCK (reqlock); 382 X_LOCK (reqlock);
369 X_LOCK (reslock); 383 X_LOCK (reslock);
… …
407 idle = 0; 421 idle = 0;
408 nreqs = 0; 422 nreqs = 0;
409 nready = 0; 423 nready = 0;
410 npending = 0; 424 npending = 0;
411 425
412 etp_atfork_parent (); 426 etp_thread_init ();
413} 427}
414 428
415static void 429static void
416etp_once_init (void) 430etp_once_init (void)
417{ 431{
432 etp_thread_init ();
418 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 433 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419} 434}
420 435
421static int 436static int
422etp_init (void (*want_poll)(void), void (*done_poll)(void)) 437etp_init (void (*want_poll)(void), void (*done_poll)(void))
… …
613} 628}
614 629
615static void etp_set_max_idle (unsigned int nthreads) 630static void etp_set_max_idle (unsigned int nthreads)
616{ 631{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 632 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 max_idle = nthreads <= 0 ? 1 : nthreads; 633 max_idle = nthreads;
634 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
635}
636
637static void etp_set_idle_timeout (unsigned int seconds)
638{
639 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
640 idle_timeout = seconds;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 641 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620} 642}
621 643
622static void etp_set_min_parallel (unsigned int nthreads) 644static void etp_set_min_parallel (unsigned int nthreads)
623{ 645{
… …
751void eio_set_max_idle (unsigned int nthreads) 773void eio_set_max_idle (unsigned int nthreads)
752{ 774{
753 etp_set_max_idle (nthreads); 775 etp_set_max_idle (nthreads);
754} 776}
755 777
778void eio_set_idle_timeout (unsigned int seconds)
779{
780 etp_set_idle_timeout (seconds);
781}
782
756void eio_set_min_parallel (unsigned int nthreads) 783void eio_set_min_parallel (unsigned int nthreads)
757{ 784{
758 etp_set_min_parallel (nthreads); 785 etp_set_min_parallel (nthreads);
759} 786}
760 787
… …
808 835
809 return res; 836 return res;
810} 837}
811#endif 838#endif
812 839
813#ifndef HAVE_FUTIMES 840#ifndef HAVE_UTIMES
814 841
815# undef utimes 842# undef utimes
816# undef futimes
817# define utimes(path,times) eio__utimes (path, times) 843# define utimes(path,times) eio__utimes (path, times)
818# define futimes(fd,times) eio__futimes (fd, times)
819 844
820static int 845static int
821eio__utimes (const char *filename, const struct timeval times[2]) 846eio__utimes (const char *filename, const struct timeval times[2])
822{ 847{
823 if (times) 848 if (times)
… …
830 return utime (filename, &buf); 855 return utime (filename, &buf);
831 } 856 }
832 else 857 else
833 return utime (filename, 0); 858 return utime (filename, 0);
834} 859}
860
861#endif
862
863#ifndef HAVE_FUTIMES
864
865# undef futimes
866# define futimes(fd,times) eio__futimes (fd, times)
835 867
836static int eio__futimes (int fd, const struct timeval tv[2]) 868static int eio__futimes (int fd, const struct timeval tv[2])
837{ 869{
838 errno = ENOSYS; 870 errno = ENOSYS;
839 return -1; 871 return -1;
… …
1387 break; 1419 break;
1388 } 1420 }
1389 } 1421 }
1390} 1422}
1391 1423
1424#ifdef PAGESIZE
1425# define eio_pagesize() PAGESIZE
1426#else
1427static intptr_t
1428eio_pagesize (void)
1429{
1430 static intptr_t page;
1431
1432 if (!page)
1433 page = sysconf (_SC_PAGESIZE);
1434
1435 return page;
1436}
1437#endif
1438
1439static void
1440eio_page_align (void **addr, size_t *length)
1441{
1442 intptr_t mask = eio_pagesize () - 1;
1443
1444 /* round down addr */
1445 intptr_t adj = mask & (intptr_t)*addr;
1446
1447 *addr = (void *)((intptr_t)*addr - adj);
1448 *length += adj;
1449
1450 /* round up length */
1451 *length = (*length + mask) & ~mask;
1452}
1453
1454#if !_POSIX_MEMLOCK
1455# define eio__mlockall(a) ((errno = ENOSYS), -1)
1456#else
1457
1458static int
1459eio__mlockall (int flags)
1460{
1461 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1462 extern int mallopt (int, int);
1463 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1464 #endif
1465
1466 if (EIO_MCL_CURRENT != MCL_CURRENT
1467 || EIO_MCL_FUTURE != MCL_FUTURE)
1468 {
1469 flags = 0
1470 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1471 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1472 }
1473
1474 return mlockall (flags);
1475}
1476#endif
1477
1478#if !_POSIX_MEMLOCK_RANGE
1479# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1480#else
1481
1482static int
1483eio__mlock (void *addr, size_t length)
1484{
1485 eio_page_align (&addr, &length);
1486
1487 return mlock (addr, length);
1488}
1489
1490#endif
1491
1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1492#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1493# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1394#else 1494#else
1395 1495
1396int 1496int
1397eio__msync (void *mem, size_t len, int flags) 1497eio__msync (void *mem, size_t len, int flags)
1398{ 1498{
1499 eio_page_align (&mem, &len);
1500
1399 if (EIO_MS_ASYNC != MS_SYNC 1501 if (EIO_MS_ASYNC != MS_SYNC
1400 || EIO_MS_INVALIDATE != MS_INVALIDATE 1502 || EIO_MS_INVALIDATE != MS_INVALIDATE
1401 || EIO_MS_SYNC != MS_SYNC) 1503 || EIO_MS_SYNC != MS_SYNC)
1402 { 1504 {
1403 flags = 0 1505 flags = 0
… …
1412#endif 1514#endif
1413 1515
1414int 1516int
1415eio__mtouch (void *mem, size_t len, int flags) 1517eio__mtouch (void *mem, size_t len, int flags)
1416{ 1518{
1519 eio_page_align (&mem, &len);
1520
1521 {
1417 intptr_t addr = (intptr_t)mem; 1522 intptr_t addr = (intptr_t)mem;
1418 intptr_t end = addr + len; 1523 intptr_t end = addr + len;
1419#ifdef PAGESIZE 1524 intptr_t page = eio_pagesize ();
1420 const intptr_t page = PAGESIZE;
1421#else
1422 static intptr_t page;
1423 1525
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) 1526 if (addr < end)
1432 if (flags & EIO_MT_MODIFY) /* modify */ 1527 if (flags & EIO_MT_MODIFY) /* modify */
1433 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1528 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1434 else 1529 else
1435 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1530 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1531 }
1436 1532
1437 return 0; 1533 return 0;
1438} 1534}
1439 1535
1440/*****************************************************************************/ 1536/*****************************************************************************/
… …
1474 if (req) 1570 if (req)
1475 break; 1571 break;
1476 1572
1477 ++idle; 1573 ++idle;
1478 1574
1479 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1575 ts.tv_sec = time (0) + idle_timeout;
1480 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1576 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1481 { 1577 {
1482 if (idle > max_idle) 1578 if (idle > max_idle)
1483 { 1579 {
1484 --idle; 1580 --idle;
… …
1563 return 0; \ 1659 return 0; \
1564 } 1660 }
1565 1661
1566static void eio_execute (etp_worker *self, eio_req *req) 1662static void eio_execute (etp_worker *self, eio_req *req)
1567{ 1663{
1568 errno = 0;
1569
1570 switch (req->type) 1664 switch (req->type)
1571 { 1665 {
1572 case EIO_READ: ALLOC (req->size); 1666 case EIO_READ: ALLOC (req->size);
1573 req->result = req->offs >= 0 1667 req->result = req->offs >= 0
1574 ? pread (req->int1, req->ptr2, req->size, req->offs) 1668 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1606 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1700 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1607 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1701 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; 1702 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1609 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1703 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1610 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1704 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; 1705 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1612 1706
1613 case EIO_READLINK: ALLOC (NAME_MAX); 1707 case EIO_READLINK: ALLOC (PATH_MAX);
1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1708 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1615 1709
1616 case EIO_SYNC: req->result = 0; sync (); break; 1710 case EIO_SYNC: req->result = 0; sync (); break;
1617 case EIO_FSYNC: req->result = fsync (req->int1); break; 1711 case EIO_FSYNC: req->result = fsync (req->int1); break;
1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1712 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1619 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1713 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1620 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1714 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1715 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1716 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; 1717 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1622 1718
1623 case EIO_READDIR: eio__scandir (req, self); break; 1719 case EIO_READDIR: eio__scandir (req, self); break;
1624 1720
1625 case EIO_BUSY: 1721 case EIO_BUSY:
1626#ifdef _WIN32 1722#ifdef _WIN32
1627 Sleep (req->nv1 * 1000.); 1723 Sleep (req->nv1 * 1e3);
1628#else 1724#else
1629 { 1725 {
1630 struct timeval tv; 1726 struct timeval tv;
1631 1727
1632 tv.tv_sec = req->nv1; 1728 tv.tv_sec = req->nv1;
1633 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1729 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1634 1730
1635 req->result = select (0, 0, 0, 0, &tv); 1731 req->result = select (0, 0, 0, 0, &tv);
1636 } 1732 }
1637#endif 1733#endif
1638 break; 1734 break;
… …
1653 times = tv; 1749 times = tv;
1654 } 1750 }
1655 else 1751 else
1656 times = 0; 1752 times = 0;
1657 1753
1658
1659 req->result = req->type == EIO_FUTIME 1754 req->result = req->type == EIO_FUTIME
1660 ? futimes (req->int1, times) 1755 ? futimes (req->int1, times)
1661 : utimes (req->ptr1, times); 1756 : utimes (req->ptr1, times);
1662 } 1757 }
1663 break; 1758 break;
… …
1672 case EIO_CUSTOM: 1767 case EIO_CUSTOM:
1673 ((void (*)(eio_req *))req->feed) (req); 1768 ((void (*)(eio_req *))req->feed) (req);
1674 break; 1769 break;
1675 1770
1676 default: 1771 default:
1772 errno = ENOSYS;
1677 req->result = -1; 1773 req->result = -1;
1678 break; 1774 break;
1679 } 1775 }
1680 1776
1681 req->errorno = errno; 1777 req->errorno = errno;
… …
1711eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1807eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1712{ 1808{
1713 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1809 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1714} 1810}
1715 1811
1812eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1813{
1814 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1815}
1816
1817eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1818{
1819 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1820}
1821
1716eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1822eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1717{ 1823{
1718 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1824 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1719} 1825}
1720 1826
… …
1854 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1960 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1855} 1961}
1856 1962
1857eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1963eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1858{ 1964{
1859 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1965 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1860} 1966}
1861 1967
1862static eio_req * 1968static eio_req *
1863eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1969eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 1970{

Diff Legend

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