[ViewVC] Diff of: cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.20 by root, Thu Jun 19 09:05:43 2008 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 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"
		45
		46	#ifdef EIO_STACKSIZE
		47	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		48	#endif
41	#include "xthread.h"	49	#include "xthread.h"
42		50
43	#include <errno.h>	51	#include <errno.h>
44	#include <stddef.h>	52	#include <stddef.h>
45	#include <stdlib.h>	53	#include <stdlib.h>
46	#include <string.h>	54	#include <string.h>
47	#include <errno.h>	55	#include <errno.h>
48	#include <sys/types.h>	56	#include <sys/types.h>
49	#include <sys/stat.h>	57	#include <sys/stat.h>
		58	#include <sys/statvfs.h>
50	#include <limits.h>	59	#include <limits.h>
51	#include <fcntl.h>	60	#include <fcntl.h>
52	#include <assert.h>	61	#include <assert.h>
53		62
54	#ifndef EIO_FINISH	63	#ifndef EIO_FINISH
…		…
64	#endif	73	#endif
65		74
66	#ifdef _WIN32	75	#ifdef _WIN32
67		76
68	/doh/	77	/doh/
69
70	#else	78	#else
71		79
72	# include "config.h"
73	# include <sys/time.h>	80	# include <sys/time.h>
74	# include <sys/select.h>	81	# include <sys/select.h>
75	# include <unistd.h>	82	# include <unistd.h>
76	# include <utime.h>	83	# include <utime.h>
77	# include <signal.h>	84	# include <signal.h>
78	# include <dirent.h>	85	# include <dirent.h>
79		86
		87	#if _POSIX_MEMLOCK \|\| _POSIX_MEMLOCK_RANGE \|\| _POSIX_MAPPED_FILES
		88	# include <sys/mman.h>
		89	#endif
		90
		91	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		92	# if __FreeBSD__ \|\| defined __NetBSD__ \|\| defined __OpenBSD__
		93	# define _DIRENT_HAVE_D_TYPE /* sigh */
		94	# define D_INO(de) (de)->d_fileno
		95	# define D_NAMLEN(de) (de)->d_namlen
		96	# elif __linux \|\| defined d_ino \|\| _XOPEN_SOURCE >= 600
		97	# define D_INO(de) (de)->d_ino
		98	# endif
		99
		100	#ifdef _D_EXACT_NAMLEN
		101	# undef D_NAMLEN
		102	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		103	#endif
		104
		105	# ifdef _DIRENT_HAVE_D_TYPE
		106	# define D_TYPE(de) (de)->d_type
		107	# endif
		108
80	# ifndef EIO_STRUCT_DIRENT	109	# ifndef EIO_STRUCT_DIRENT
81	# define EIO_STRUCT_DIRENT struct dirent	110	# define EIO_STRUCT_DIRENT struct dirent
82	# endif	111	# endif
83		112
84	#endif	113	#endif
85		114
86	#if HAVE_SENDFILE	115	#if HAVE_SENDFILE
87	# if __linux	116	# if __linux
88	# include <sys/sendfile.h>	117	# include <sys/sendfile.h>
89	# elif __freebsd	118	# elif __FreeBSD__ \|\| defined __APPLE__
90	# include <sys/socket.h>	119	# include <sys/socket.h>
91	# include <sys/uio.h>	120	# include <sys/uio.h>
92	# elif __hpux	121	# elif __hpux
93	# include <sys/socket.h>	122	# include <sys/socket.h>
94	# elif __solaris /* not yet */	123	# elif __solaris
95	# include <sys/sendfile.h>	124	# include <sys/sendfile.h>
96	# else	125	# else
97	# error sendfile support requested but not available	126	# error sendfile support requested but not available
98	# endif	127	# endif
99	#endif	128	#endif
100		129
101	/* number of seconds after which an idle threads exit */	130	#ifndef D_TYPE
102	#define IDLE_TIMEOUT 10	131	# define D_TYPE(de) 0
		132	#endif
		133	#ifndef D_INO
		134	# define D_INO(de) 0
		135	#endif
		136	#ifndef D_NAMLEN
		137	# define D_NAMLEN(de) strlen ((de)->d_name)
		138	#endif
103		139
104	/* used for struct dirent, AIX doesn't provide it */	140	/* used for struct dirent, AIX doesn't provide it */
105	#ifndef NAME_MAX	141	#ifndef NAME_MAX
106	# define NAME_MAX 4096	142	# define NAME_MAX 4096
		143	#endif
		144
		145	/* used for readlink etc. */
		146	#ifndef PATH_MAX
		147	# define PATH_MAX 4096
107	#endif	148	#endif
108		149
109	/* buffer size for various temporary buffers */	150	/* buffer size for various temporary buffers */
110	#define EIO_BUFSIZE 65536	151	#define EIO_BUFSIZE 65536
111		152
…		…
117	errno = ENOMEM; \	158	errno = ENOMEM; \
118	if (!eio_buf) \	159	if (!eio_buf) \
119	return -1;	160	return -1;
120		161
121	#define EIO_TICKS ((1000000 + 1023) >> 10)	162	#define EIO_TICKS ((1000000 + 1023) >> 10)
		163
		164	/*****************************************************************************/
		165
		166	#if __GNUC__ >= 3
		167	# define expect(expr,value) __builtin_expect ((expr),(value))
		168	#else
		169	# define expect(expr,value) (expr)
		170	#endif
		171
		172	#define expect_false(expr) expect ((expr) != 0, 0)
		173	#define expect_true(expr) expect ((expr) != 0, 1)
122		174
123	/*****************************************************************************/	175	/*****************************************************************************/
124		176
125	#define ETP_PRI_MIN EIO_PRI_MIN	177	#define ETP_PRI_MIN EIO_PRI_MIN
126	#define ETP_PRI_MAX EIO_PRI_MAX	178	#define ETP_PRI_MAX EIO_PRI_MAX
…		…
144	if (wrk->dirp) \	196	if (wrk->dirp) \
145	{ \	197	{ \
146	closedir (wrk->dirp); \	198	closedir (wrk->dirp); \
147	wrk->dirp = 0; \	199	wrk->dirp = 0; \
148	}	200	}
		201
149	#define ETP_WORKER_COMMON \	202	#define ETP_WORKER_COMMON \
150	void *dbuf; \	203	void *dbuf; \
151	DIR *dirp;	204	DIR *dirp;
152		205
153	/*****************************************************************************/	206	/*****************************************************************************/
154		207
155	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	208	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156		209
157	/* calculcate time difference in ~1/EIO_TICKS of a second */	210	/* calculate time difference in ~1/EIO_TICKS of a second */
158	static int tvdiff (struct timeval tv1, struct timeval tv2)	211	static int tvdiff (struct timeval tv1, struct timeval tv2)
159	{	212	{
160	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	213	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	214	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
162	}	215	}
…		…
170	static unsigned int max_poll_reqs; /* reslock */	223	static unsigned int max_poll_reqs; /* reslock */
171		224
172	static volatile unsigned int nreqs; /* reqlock */	225	static volatile unsigned int nreqs; /* reqlock */
173	static volatile unsigned int nready; /* reqlock */	226	static volatile unsigned int nready; /* reqlock */
174	static volatile unsigned int npending; /* reqlock */	227	static volatile unsigned int npending; /* reqlock */
175	static volatile unsigned int max_idle = 4;	228	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		229	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
176		230
		231	static xmutex_t wrklock;
		232	static xmutex_t reslock;
		233	static xmutex_t reqlock;
		234	static xcond_t reqwait;
		235
		236	#if !HAVE_PREADWRITE
		237	/*
		238	* make our pread/pwrite emulation safe against themselves, but not against
		239	* normal read/write by using a mutex. slows down execution a lot,
		240	* but that's your problem, not mine.
		241	*/
177	static mutex_t wrklock = X_MUTEX_INIT;	242	static xmutex_t preadwritelock = X_MUTEX_INIT;
178	static mutex_t reslock = X_MUTEX_INIT;	243	#endif
179	static mutex_t reqlock = X_MUTEX_INIT;
180	static cond_t reqwait = X_COND_INIT;
181		244
182	typedef struct etp_worker	245	typedef struct etp_worker
183	{	246	{
184	/* locked by wrklock */	247	/* locked by wrklock */
185	struct etp_worker prev, next;	248	struct etp_worker prev, next;
186		249
187	thread_t tid;	250	xthread_t tid;
188		251
189	/* locked by reslock, reqlock or wrklock */	252	/* locked by reslock, reqlock or wrklock */
190	ETP_REQ req; / currently processed request */	253	ETP_REQ req; / currently processed request */
191		254
192	ETP_WORKER_COMMON	255	ETP_WORKER_COMMON
…		…
306	}	369	}
307		370
308	abort ();	371	abort ();
309	}	372	}
310		373
		374	static 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
311	static void etp_atfork_prepare (void)	382	static void etp_atfork_prepare (void)
312	{	383	{
313	X_LOCK (wrklock);	384	X_LOCK (wrklock);
314	X_LOCK (reqlock);	385	X_LOCK (reqlock);
315	X_LOCK (reslock);	386	X_LOCK (reslock);
…		…
330		401
331	static void etp_atfork_child (void)	402	static void etp_atfork_child (void)
332	{	403	{
333	ETP_REQ *prv;	404	ETP_REQ *prv;
334		405
335	while (prv = reqq_shift (&req_queue))	406	while ((prv = reqq_shift (&req_queue)))
336	ETP_DESTROY (prv);	407	ETP_DESTROY (prv);
337		408
338	while (prv = reqq_shift (&res_queue))	409	while ((prv = reqq_shift (&res_queue)))
339	ETP_DESTROY (prv);	410	ETP_DESTROY (prv);
340		411
341	while (wrk_first.next != &wrk_first)	412	while (wrk_first.next != &wrk_first)
342	{	413	{
343	etp_worker *wrk = wrk_first.next;	414	etp_worker *wrk = wrk_first.next;
…		…
353	idle = 0;	424	idle = 0;
354	nreqs = 0;	425	nreqs = 0;
355	nready = 0;	426	nready = 0;
356	npending = 0;	427	npending = 0;
357		428
358	etp_atfork_parent ();	429	etp_thread_init ();
359	}	430	}
360		431
361	static void	432	static void
362	etp_once_init (void)	433	etp_once_init (void)
363	{	434	{
		435	etp_thread_init ();
364	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);
365	}	437	}
366		438
367	static int	439	static int
368	etp_init (void (want_poll)(void), void (done_poll)(void))	440	etp_init (void (want_poll)(void), void (done_poll)(void))
…		…
402	X_UNLOCK (wrklock);	474	X_UNLOCK (wrklock);
403	}	475	}
404		476
405	static void etp_maybe_start_thread (void)	477	static void etp_maybe_start_thread (void)
406	{	478	{
407	if (etp_nthreads () >= wanted)	479	if (expect_true (etp_nthreads () >= wanted))
408	return;	480	return;
409		481
410	/* todo: maybe use idle here, but might be less exact */	482	/* todo: maybe use idle here, but might be less exact */
411	if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())	483	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
412	return;	484	return;
413		485
414	etp_start_thread ();	486	etp_start_thread ();
415	}	487	}
416		488
…		…
469		541
470	X_LOCK (reqlock);	542	X_LOCK (reqlock);
471	--nreqs;	543	--nreqs;
472	X_UNLOCK (reqlock);	544	X_UNLOCK (reqlock);
473		545
474	if (req->type == EIO_GROUP && req->size)	546	if (expect_false (req->type == EIO_GROUP && req->size))
475	{	547	{
476	req->int1 = 1; /* mark request as delayed */	548	req->int1 = 1; /* mark request as delayed */
477	continue;	549	continue;
478	}	550	}
479	else	551	else
480	{	552	{
481	int res = ETP_FINISH (req);	553	int res = ETP_FINISH (req);
482	if (res)	554	if (expect_false (res))
483	return res;	555	return res;
484	}	556	}
485		557
486	if (maxreqs && !--maxreqs)	558	if (expect_false (maxreqs && !--maxreqs))
487	break;	559	break;
488		560
489	if (maxtime)	561	if (maxtime)
490	{	562	{
491	gettimeofday (&tv_now, 0);	563	gettimeofday (&tv_now, 0);
…		…
510		582
511	static void etp_submit (ETP_REQ *req)	583	static void etp_submit (ETP_REQ *req)
512	{	584	{
513	req->pri -= ETP_PRI_MIN;	585	req->pri -= ETP_PRI_MIN;
514		586
515	if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	587	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
516	if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	588	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
517		589
		590	if (expect_false (req->type == EIO_GROUP))
		591	{
		592	/* I hope this is worth it :/ */
518	X_LOCK (reqlock);	593	X_LOCK (reqlock);
519	++nreqs;	594	++nreqs;
		595	X_UNLOCK (reqlock);
		596
		597	X_LOCK (reslock);
		598
		599	++npending;
		600
		601	if (!reqq_push (&res_queue, req) && want_poll_cb)
		602	want_poll_cb ();
		603
		604	X_UNLOCK (reslock);
		605	}
		606	else
		607	{
		608	X_LOCK (reqlock);
		609	++nreqs;
520	++nready;	610	++nready;
521	reqq_push (&req_queue, req);	611	reqq_push (&req_queue, req);
522	X_COND_SIGNAL (reqwait);	612	X_COND_SIGNAL (reqwait);
523	X_UNLOCK (reqlock);	613	X_UNLOCK (reqlock);
524		614
525	etp_maybe_start_thread ();	615	etp_maybe_start_thread ();
		616	}
526	}	617	}
527		618
528	static void etp_set_max_poll_time (double nseconds)	619	static void etp_set_max_poll_time (double nseconds)
529	{	620	{
530	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	621	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
531	max_poll_time = nseconds;	622	max_poll_time = nseconds * EIO_TICKS;
532	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	623	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
533	}	624	}
534		625
535	static void etp_set_max_poll_reqs (unsigned int maxreqs)	626	static void etp_set_max_poll_reqs (unsigned int maxreqs)
536	{	627	{
…		…
540	}	631	}
541		632
542	static void etp_set_max_idle (unsigned int nthreads)	633	static void etp_set_max_idle (unsigned int nthreads)
543	{	634	{
544	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	635	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
545	max_idle = nthreads <= 0 ? 1 : nthreads;	636	max_idle = nthreads;
		637	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		638	}
		639
		640	static void etp_set_idle_timeout (unsigned int seconds)
		641	{
		642	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		643	idle_timeout = seconds;
546	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	644	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
547	}	645	}
548		646
549	static void etp_set_min_parallel (unsigned int nthreads)	647	static void etp_set_min_parallel (unsigned int nthreads)
550	{	648	{
…		…
565		663
566	static void grp_try_feed (eio_req *grp)	664	static void grp_try_feed (eio_req *grp)
567	{	665	{
568	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	666	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569	{	667	{
570	int old_len = grp->size;	668	grp->flags &= ~EIO_FLAG_GROUPADD;
571		669
572	EIO_FEED (grp);	670	EIO_FEED (grp);
573		671
574	/* stop if no progress has been made */	672	/* stop if no progress has been made */
575	if (old_len == grp->size)	673	if (!(grp->flags & EIO_FLAG_GROUPADD))
576	{	674	{
577	grp->feed = 0;	675	grp->feed = 0;
578	break;	676	break;
579	}	677	}
580	}	678	}
…		…
678	void eio_set_max_idle (unsigned int nthreads)	776	void eio_set_max_idle (unsigned int nthreads)
679	{	777	{
680	etp_set_max_idle (nthreads);	778	etp_set_max_idle (nthreads);
681	}	779	}
682		780
		781	void eio_set_idle_timeout (unsigned int seconds)
		782	{
		783	etp_set_idle_timeout (seconds);
		784	}
		785
683	void eio_set_min_parallel (unsigned int nthreads)	786	void eio_set_min_parallel (unsigned int nthreads)
684	{	787	{
685	etp_set_min_parallel (nthreads);	788	etp_set_min_parallel (nthreads);
686	}	789	}
687		790
…		…
697		800
698	/*****************************************************************************/	801	/*****************************************************************************/
699	/* work around various missing functions */	802	/* work around various missing functions */
700		803
701	#if !HAVE_PREADWRITE	804	#if !HAVE_PREADWRITE
		805	# undef pread
		806	# undef pwrite
702	# define pread eio__pread	807	# define pread eio__pread
703	# define pwrite eio__pwrite	808	# define pwrite eio__pwrite
704
705	/*
706	* make our pread/pwrite safe against themselves, but not against
707	* normal read/write by using a mutex. slows down execution a lot,
708	* but that's your problem, not mine.
709	*/
710	static mutex_t preadwritelock = X_MUTEX_INIT;
711		809
712	static ssize_t	810	static ssize_t
713	eio__pread (int fd, void *buf, size_t count, off_t offset)	811	eio__pread (int fd, void *buf, size_t count, off_t offset)
714	{	812	{
715	ssize_t res;	813	ssize_t res;
…		…
733		831
734	X_LOCK (preadwritelock);	832	X_LOCK (preadwritelock);
735	ooffset = lseek (fd, 0, SEEK_CUR);	833	ooffset = lseek (fd, 0, SEEK_CUR);
736	lseek (fd, offset, SEEK_SET);	834	lseek (fd, offset, SEEK_SET);
737	res = write (fd, buf, count);	835	res = write (fd, buf, count);
738	lseek (fd, offset, SEEK_SET);	836	lseek (fd, ooffset, SEEK_SET);
739	X_UNLOCK (preadwritelock);	837	X_UNLOCK (preadwritelock);
740		838
741	return res;	839	return res;
742	}	840	}
743	#endif	841	#endif
744		842
745	#ifndef HAVE_FUTIMES	843	#ifndef HAVE_UTIMES
746		844
		845	# undef utimes
747	# define utimes(path,times) eio__utimes (path, times)	846	# define utimes(path,times) eio__utimes (path, times)
748	# define futimes(fd,times) eio__futimes (fd, times)
749		847
750	static int	848	static int
751	eio__utimes (const char *filename, const struct timeval times[2])	849	eio__utimes (const char *filename, const struct timeval times[2])
752	{	850	{
753	if (times)	851	if (times)
…		…
761	}	859	}
762	else	860	else
763	return utime (filename, 0);	861	return utime (filename, 0);
764	}	862	}
765		863
		864	#endif
		865
		866	#ifndef HAVE_FUTIMES
		867
		868	# undef futimes
		869	# define futimes(fd,times) eio__futimes (fd, times)
		870
766	static int eio__futimes (int fd, const struct timeval tv[2])	871	static int eio__futimes (int fd, const struct timeval tv[2])
767	{	872	{
768	errno = ENOSYS;	873	errno = ENOSYS;
769	return -1;	874	return -1;
770	}	875	}
771		876
772	#endif	877	#endif
773		878
774	#if !HAVE_FDATASYNC	879	#if !HAVE_FDATASYNC
		880	# undef fdatasync
775	# define fdatasync fsync	881	# define fdatasync(fd) fsync (fd)
776	#endif	882	#endif
		883
		884	/* sync_file_range always needs emulation */
		885	int
		886	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		887	{
		888	#if HAVE_SYNC_FILE_RANGE
		889	int res;
		890
		891	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		892	\|\| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		893	\|\| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		894	{
		895	flags = 0
		896	\| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		897	\| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		898	\| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		899	}
		900
		901	res = sync_file_range (fd, offset, nbytes, flags);
		902
		903	if (!res \|\| errno != ENOSYS)
		904	return res;
		905	#endif
		906
		907	/* even though we could play tricks with the flags, it's better to always
		908	* call fdatasync, as that matches the expectation of its users best */
		909	return fdatasync (fd);
		910	}
777		911
778	#if !HAVE_READAHEAD	912	#if !HAVE_READAHEAD
		913	# undef readahead
779	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	914	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780		915
781	static ssize_t	916	static ssize_t
782	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	917	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783	{	918	{
…		…
810		945
811	#if HAVE_SENDFILE	946	#if HAVE_SENDFILE
812	# if __linux	947	# if __linux
813	res = sendfile (ofd, ifd, &offset, count);	948	res = sendfile (ofd, ifd, &offset, count);
814		949
815	# elif __freebsd	950	# elif __FreeBSD__
816	/*	951	/*
817	* Of course, the freebsd sendfile is a dire hack with no thoughts	952	* Of course, the freebsd sendfile is a dire hack with no thoughts
818	* wasted on making it similar to other I/O functions.	953	* wasted on making it similar to other I/O functions.
819	*/	954	*/
820	{	955	{
821	off_t sbytes;	956	off_t sbytes;
822	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	957	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
823		958
824	if (res < 0 && sbytes)	959	#if 0 /* according to the manpage, this is correct, but broken behaviour */
825	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	960	/* freebsd' sendfile will return 0 on success */
		961	/* freebsd 8 documents it as only setting sbytes on EINTR and EAGAIN, but /
		962	/* not on e.g. EIO or EPIPE - sounds broken */
		963	if ((res < 0 && (errno == EAGAIN \|\| errno == EINTR) && sbytes) \|\| res == 0)
		964	res = sbytes;
		965	#endif
		966
		967	/* according to source inspection, this is correct, and useful behaviour */
		968	if (sbytes)
		969	res = sbytes;
		970	}
		971
		972	# elif defined (__APPLE__)
		973
		974	{
		975	off_t sbytes = count;
		976	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		977
		978	/* according to the manpage, sbytes is always valid */
		979	if (sbytes)
826	res = sbytes;	980	res = sbytes;
827	}	981	}
828		982
829	# elif __hpux	983	# elif __hpux
830	res = sendfile (ofd, ifd, offset, count, 0, 0);	984	res = sendfile (ofd, ifd, offset, count, 0, 0);
…		…
844	if (res < 0 && sbytes)	998	if (res < 0 && sbytes)
845	res = sbytes;	999	res = sbytes;
846	}	1000	}
847		1001
848	# endif	1002	# endif
		1003
		1004	#elif defined (_WIN32)
		1005
		1006	/* does not work, just for documentation of what would need to be done */
		1007	{
		1008	HANDLE h = TO_SOCKET (ifd);
		1009	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1010	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1011	}
		1012
849	#else	1013	#else
850	res = -1;	1014	res = -1;
851	errno = ENOSYS;	1015	errno = ENOSYS;
852	#endif	1016	#endif
853		1017
854	if (res < 0	1018	if (res < 0
855	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK	1019	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK
		1020	/* BSDs */
		1021	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1022	\|\| errno == ENOTSUP
		1023	#endif
		1024	\|\| errno == EOPNOTSUPP /* BSDs */
856	#if __solaris	1025	#if __solaris
857	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE	1026	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE
858	#endif	1027	#endif
859	)	1028	)
860	)	1029	)
…		…
891	}	1060	}
892		1061
893	return res;	1062	return res;
894	}	1063	}
895		1064
		1065	static signed char
		1066	eio_dent_cmp (const eio_dirent a, const eio_dirent b)
		1067	{
		1068	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1069	: a->inode < b->inode ? -1
		1070	: a->inode > b->inode ? 1
		1071	: 0;
		1072	}
		1073
		1074	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1075
		1076	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1077	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1078
		1079	static void
		1080	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1081	{
		1082	unsigned char bits [9 + sizeof (ino_t) * 8];
		1083	unsigned char *bit = bits;
		1084
		1085	assert (CHAR_BIT == 8);
		1086	assert (sizeof (eio_dirent) * 8 < 256);
		1087	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1088	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1089
		1090	if (size <= EIO_SORT_FAST)
		1091	return;
		1092
		1093	/* first prepare an array of bits to test in our radix sort */
		1094	/* try to take endianness into account, as well as differences in ino_t sizes */
		1095	/* inode_bits must contain all inodes ORed together */
		1096	/* which is used to skip bits that are 0 everywhere, which is very common */
		1097	{
		1098	ino_t endianness;
		1099	int i, j;
		1100
		1101	/* we store the byte offset of byte n into byte n of "endianness" */
		1102	for (i = 0; i < sizeof (ino_t); ++i)
		1103	((unsigned char *)&endianness)[i] = i;
		1104
		1105	*bit++ = 0;
		1106
		1107	for (i = 0; i < sizeof (ino_t); ++i)
		1108	{
		1109	/* shifting off the byte offsets out of "endianness" */
		1110	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1111	endianness >>= 8;
		1112
		1113	for (j = 0; j < 8; ++j)
		1114	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1115	*bit++ = offs + j;
		1116	}
		1117
		1118	for (j = 0; j < 8; ++j)
		1119	if (score_bits & (1 << j))
		1120	bit++ = offsetof (eio_dirent, score) 8 + j;
		1121	}
		1122
		1123	/* now actually do the sorting (a variant of MSD radix sort) */
		1124	{
		1125	eio_dirent base_stk [9 + sizeof (ino_t) 8], *base;
		1126	eio_dirent end_stk [9 + sizeof (ino_t) 8], *end;
		1127	unsigned char bit_stk [9 + sizeof (ino_t) 8];
		1128	int stk_idx = 0;
		1129
		1130	base_stk [stk_idx] = dents;
		1131	end_stk [stk_idx] = dents + size;
		1132	bit_stk [stk_idx] = bit - 1;
		1133
		1134	do
		1135	{
		1136	base = base_stk [stk_idx];
		1137	end = end_stk [stk_idx];
		1138	bit = bit_stk [stk_idx];
		1139
		1140	for (;;)
		1141	{
		1142	unsigned char O = *bit >> 3;
		1143	unsigned char M = 1 << (*bit & 7);
		1144
		1145	eio_dirent *a = base;
		1146	eio_dirent *b = end;
		1147
		1148	if (b - a < EIO_SORT_CUTOFF)
		1149	break;
		1150
		1151	/* now bit-partition the array on the bit */
		1152	/* this ugly asymmetric loop seems to perform much better than typical */
		1153	/* partition algos found in the literature */
		1154	do
		1155	if (!(((unsigned char *)a)[O] & M))
		1156	++a;
		1157	else if (!(((unsigned char *)--b)[O] & M))
		1158	{
		1159	eio_dirent tmp = a; a = b; b = tmp;
		1160	++a;
		1161	}
		1162	while (b > a);
		1163
		1164	/* next bit, or stop, if no bits left in this path */
		1165	if (!*--bit)
		1166	break;
		1167
		1168	base_stk [stk_idx] = a;
		1169	end_stk [stk_idx] = end;
		1170	bit_stk [stk_idx] = bit;
		1171	++stk_idx;
		1172
		1173	end = a;
		1174	}
		1175	}
		1176	while (stk_idx--);
		1177	}
		1178	}
		1179
		1180	static void
		1181	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1182	{
		1183	/* first move the smallest element to the front, to act as a sentinel */
		1184	{
		1185	int i;
		1186	eio_dirent *min = dents;
		1187
		1188	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1189	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1190	if (EIO_DENT_CMP (dents [i], <, *min))
		1191	min = &dents [i];
		1192
		1193	/* swap elements 0 and j (minimum) */
		1194	{
		1195	eio_dirent tmp = dents; dents = min; min = tmp;
		1196	}
		1197	}
		1198
		1199	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1200	{
		1201	eio_dirent i, j;
		1202
		1203	for (i = dents + 1; i < dents + size; ++i)
		1204	{
		1205	eio_dirent value = *i;
		1206
		1207	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1208	j [1] = j [0];
		1209
		1210	j [1] = value;
		1211	}
		1212	}
		1213	}
		1214
		1215	static void
		1216	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1217	{
		1218	if (size <= 1)
		1219	return; /* our insertion sort relies on size > 0 */
		1220
		1221	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1222	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1223	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1224
		1225	/* use an insertion sort at the end, or for small arrays, */
		1226	/* as insertion sort is more efficient for small partitions */
		1227	eio_dent_insertion_sort (dents, size);
		1228	}
		1229
896	/* read a full directory */	1230	/* read a full directory */
897	static void	1231	static void
898	eio__scandir (eio_req req, etp_worker self)	1232	eio__scandir (eio_req req, etp_worker self)
899	{	1233	{
900	DIR *dirp;	1234	DIR *dirp;
901	EIO_STRUCT_DIRENT *entp;	1235	EIO_STRUCT_DIRENT *entp;
902	char name, names;	1236	char name, names;
903	int memlen = 4096;	1237	int namesalloc = 4096;
904	int memofs = 0;	1238	int namesoffs = 0;
		1239	int flags = req->int1;
		1240	eio_dirent *dents = 0;
		1241	int dentalloc = 128;
905	int res = 0;	1242	int dentoffs = 0;
		1243	ino_t inode_bits = 0;
		1244
		1245	req->result = -1;
		1246
		1247	if (!(flags & EIO_READDIR_DENTS))
		1248	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);
906		1249
907	X_LOCK (wrklock);	1250	X_LOCK (wrklock);
908	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1251	/* the corresponding closedir is in ETP_WORKER_CLEAR */
909	self->dirp = dirp = opendir (req->ptr1);	1252	self->dirp = dirp = opendir (req->ptr1);
		1253
910	req->flags \|= EIO_FLAG_PTR2_FREE;	1254	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;
		1255	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
911	req->ptr2 = names = malloc (memlen);	1256	req->ptr2 = names = malloc (namesalloc);
912	X_UNLOCK (wrklock);	1257	X_UNLOCK (wrklock);
913		1258
914	if (dirp && names)	1259	if (dirp && names && (!flags \|\| dents))
915	for (;;)	1260	for (;;)
916	{	1261	{
917	errno = 0;	1262	errno = 0;
918	entp = readdir (dirp);	1263	entp = readdir (dirp);
919		1264
920	if (!entp)	1265	if (!entp)
		1266	{
		1267	if (errno)
		1268	break;
		1269
		1270	/* sort etc. */
		1271	req->int1 = flags;
		1272	req->result = dentoffs;
		1273
		1274	if (flags & EIO_READDIR_STAT_ORDER)
		1275	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1276	else if (flags & EIO_READDIR_DIRS_FIRST)
		1277	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1278	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1279	else
		1280	{
		1281	/* in this case, all is known, and we just put dirs first and sort them */
		1282	eio_dirent *oth = dents + dentoffs;
		1283	eio_dirent *dir = dents;
		1284
		1285	/* now partition dirs to the front, and non-dirs to the back */
		1286	/* by walking from both sides and swapping if necessary */
		1287	while (oth > dir)
		1288	{
		1289	if (dir->type == EIO_DT_DIR)
		1290	++dir;
		1291	else if ((--oth)->type == EIO_DT_DIR)
		1292	{
		1293	eio_dirent tmp = dir; dir = oth; oth = tmp;
		1294
		1295	++dir;
		1296	}
		1297	}
		1298
		1299	/* now sort the dirs only (dirs all have the same score) */
		1300	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1301	}
		1302
921	break;	1303	break;
		1304	}
922		1305
		1306	/* now add the entry to our list(s) */
923	name = entp->d_name;	1307	name = entp->d_name;
924		1308
		1309	/* skip . and .. entries */
925	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))	1310	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))
926	{	1311	{
927	int len = strlen (name) + 1;	1312	int len = D_NAMLEN (entp) + 1;
928		1313
929	res++;	1314	while (expect_false (namesoffs + len > namesalloc))
930
931	while (memofs + len > memlen)
932	{	1315	{
933	memlen *= 2;	1316	namesalloc *= 2;
934	X_LOCK (wrklock);	1317	X_LOCK (wrklock);
935	req->ptr2 = names = realloc (names, memlen);	1318	req->ptr2 = names = realloc (names, namesalloc);
936	X_UNLOCK (wrklock);	1319	X_UNLOCK (wrklock);
937		1320
938	if (!names)	1321	if (!names)
939	break;	1322	break;
940	}	1323	}
941		1324
942	memcpy (names + memofs, name, len);	1325	memcpy (names + namesoffs, name, len);
		1326
		1327	if (dents)
		1328	{
		1329	struct eio_dirent *ent;
		1330
		1331	if (expect_false (dentoffs == dentalloc))
		1332	{
		1333	dentalloc *= 2;
		1334	X_LOCK (wrklock);
		1335	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1336	X_UNLOCK (wrklock);
		1337
		1338	if (!dents)
		1339	break;
		1340	}
		1341
		1342	ent = dents + dentoffs;
		1343
		1344	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1345	ent->namelen = len - 1;
		1346	ent->inode = D_INO (entp);
		1347
		1348	inode_bits \|= ent->inode;
		1349
		1350	switch (D_TYPE (entp))
		1351	{
		1352	default:
		1353	ent->type = EIO_DT_UNKNOWN;
		1354	flags \|= EIO_READDIR_FOUND_UNKNOWN;
		1355	break;
		1356
		1357	#ifdef DT_FIFO
		1358	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1359	#endif
		1360	#ifdef DT_CHR
		1361	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1362	#endif
		1363	#ifdef DT_MPC
		1364	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1365	#endif
		1366	#ifdef DT_DIR
		1367	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1368	#endif
		1369	#ifdef DT_NAM
		1370	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1371	#endif
		1372	#ifdef DT_BLK
		1373	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1374	#endif
		1375	#ifdef DT_MPB
		1376	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1377	#endif
		1378	#ifdef DT_REG
		1379	case DT_REG: ent->type = EIO_DT_REG; break;
		1380	#endif
		1381	#ifdef DT_NWK
		1382	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1383	#endif
		1384	#ifdef DT_CMP
		1385	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1386	#endif
		1387	#ifdef DT_LNK
		1388	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1389	#endif
		1390	#ifdef DT_SOCK
		1391	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1392	#endif
		1393	#ifdef DT_DOOR
		1394	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1395	#endif
		1396	#ifdef DT_WHT
		1397	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1398	#endif
		1399	}
		1400
		1401	ent->score = 7;
		1402
		1403	if (flags & EIO_READDIR_DIRS_FIRST)
		1404	{
		1405	if (ent->type == EIO_DT_UNKNOWN)
		1406	{
		1407	if (name == '.') / leading dots are likely directories, and, in any case, rare */
		1408	ent->score = 1;
		1409	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1410	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1411	}
		1412	else if (ent->type == EIO_DT_DIR)
		1413	ent->score = 0;
		1414	}
		1415	}
		1416
943	memofs += len;	1417	namesoffs += len;
		1418	++dentoffs;
		1419	}
		1420
		1421	if (EIO_CANCELLED (req))
		1422	{
		1423	errno = ECANCELED;
		1424	break;
944	}	1425	}
945	}	1426	}
		1427	}
946		1428
947	if (errno)	1429	#ifdef PAGESIZE
948	res = -1;	1430	# define eio_pagesize() PAGESIZE
		1431	#else
		1432	static intptr_t
		1433	eio_pagesize (void)
		1434	{
		1435	static intptr_t page;
		1436
		1437	if (!page)
		1438	page = sysconf (_SC_PAGESIZE);
		1439
		1440	return page;
		1441	}
		1442	#endif
		1443
		1444	static void
		1445	eio_page_align (void *addr, size_t length)
		1446	{
		1447	intptr_t mask = eio_pagesize () - 1;
		1448
		1449	/* round down addr */
		1450	intptr_t adj = mask & (intptr_t)*addr;
		1451
		1452	addr = (void )((intptr_t)*addr - adj);
		1453	*length += adj;
		1454
		1455	/* round up length */
		1456	length = (length + mask) & ~mask;
		1457	}
		1458
		1459	#if !_POSIX_MEMLOCK
		1460	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1461	#else
		1462
		1463	static int
		1464	eio__mlockall (int flags)
		1465	{
		1466	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1467	extern int mallopt (int, int);
		1468	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1469	#endif
		1470
		1471	if (EIO_MCL_CURRENT != MCL_CURRENT
		1472	\|\| EIO_MCL_FUTURE != MCL_FUTURE)
		1473	{
		1474	flags = 0
		1475	\| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1476	\| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1477	}
		1478
		1479	return mlockall (flags);
		1480	}
		1481	#endif
		1482
		1483	#if !_POSIX_MEMLOCK_RANGE
		1484	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1485	#else
		1486
		1487	static int
		1488	eio__mlock (void *addr, size_t length)
		1489	{
		1490	eio_page_align (&addr, &length);
		1491
		1492	return mlock (addr, length);
		1493	}
		1494
		1495	#endif
		1496
		1497	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1498	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1499	#else
		1500
		1501	int
		1502	eio__msync (void *mem, size_t len, int flags)
		1503	{
		1504	eio_page_align (&mem, &len);
		1505
		1506	if (EIO_MS_ASYNC != MS_SYNC
		1507	\|\| EIO_MS_INVALIDATE != MS_INVALIDATE
		1508	\|\| EIO_MS_SYNC != MS_SYNC)
		1509	{
		1510	flags = 0
		1511	\| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1512	\| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1513	\| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1514	}
		1515
		1516	return msync (mem, len, flags);
		1517	}
		1518
		1519	#endif
		1520
		1521	int
		1522	eio__mtouch (void *mem, size_t len, int flags)
		1523	{
		1524	eio_page_align (&mem, &len);
		1525
949		1526	{
950	req->result = res;	1527	intptr_t addr = (intptr_t)mem;
		1528	intptr_t end = addr + len;
		1529	intptr_t page = eio_pagesize ();
		1530
		1531	if (addr < end)
		1532	if (flags & EIO_MT_MODIFY) /* modify */
		1533	do { ((volatile sig_atomic_t )addr) \|= 0; } while ((addr += page) < len);
		1534	else
		1535	do { ((volatile sig_atomic_t )addr) ; } while ((addr += page) < len);
		1536	}
		1537
		1538	return 0;
951	}	1539	}
952		1540
953	/*****************************************************************************/	1541	/*****************************************************************************/
954		1542
955	#define ALLOC(len) \	1543	#define ALLOC(len) \
…		…
987	if (req)	1575	if (req)
988	break;	1576	break;
989		1577
990	++idle;	1578	++idle;
991		1579
992	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1580	ts.tv_sec = time (0) + idle_timeout;
993	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1581	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
994	{	1582	{
995	if (idle > max_idle)	1583	if (idle > max_idle)
996	{	1584	{
997	--idle;	1585	--idle;
…		…
1076	return 0; \	1664	return 0; \
1077	}	1665	}
1078		1666
1079	static void eio_execute (etp_worker self, eio_req req)	1667	static void eio_execute (etp_worker self, eio_req req)
1080	{	1668	{
1081	errno = 0;
1082
1083	switch (req->type)	1669	switch (req->type)
1084	{	1670	{
1085	case EIO_READ: ALLOC (req->size);	1671	case EIO_READ: ALLOC (req->size);
1086	req->result = req->offs >= 0	1672	req->result = req->offs >= 0
1087	? pread (req->int1, req->ptr2, req->size, req->offs)	1673	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1098	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1684	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1099	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1685	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1100	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1686	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1101	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1687	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1102		1688
		1689	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1690	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1691	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1692	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1693
1103	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1694	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1104	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1695	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1105	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	1696	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1106	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1697	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1107	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	1698	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1114	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1705	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1115	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;
1116	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	1707	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1117	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	1708	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1118	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	1709	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1119	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;
1120		1711
1121	case EIO_READLINK: ALLOC (NAME_MAX);	1712	case EIO_READLINK: ALLOC (PATH_MAX);
1122	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	1713	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1123		1714
1124	case EIO_SYNC: req->result = 0; sync (); break;	1715	case EIO_SYNC: req->result = 0; sync (); break;
1125	case EIO_FSYNC: req->result = fsync (req->int1); break;	1716	case EIO_FSYNC: req->result = fsync (req->int1); break;
1126	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1717	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1718	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1719	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1720	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1721	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		1722	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1127		1723
1128	case EIO_READDIR: eio__scandir (req, self); break;	1724	case EIO_READDIR: eio__scandir (req, self); break;
1129		1725
1130	case EIO_BUSY:	1726	case EIO_BUSY:
1131	#ifdef _WIN32	1727	#ifdef _WIN32
1132	Sleep (req->nv1 * 1000.);	1728	Sleep (req->nv1 * 1e3);
1133	#else	1729	#else
1134	{	1730	{
1135	struct timeval tv;	1731	struct timeval tv;
1136		1732
1137	tv.tv_sec = req->nv1;	1733	tv.tv_sec = req->nv1;
1138	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	1734	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1139		1735
1140	req->result = select (0, 0, 0, 0, &tv);	1736	req->result = select (0, 0, 0, 0, &tv);
1141	}	1737	}
1142	#endif	1738	#endif
1143	break;	1739	break;
…		…
1158	times = tv;	1754	times = tv;
1159	}	1755	}
1160	else	1756	else
1161	times = 0;	1757	times = 0;
1162		1758
1163
1164	req->result = req->type == EIO_FUTIME	1759	req->result = req->type == EIO_FUTIME
1165	? futimes (req->int1, times)	1760	? futimes (req->int1, times)
1166	: utimes (req->ptr1, times);	1761	: utimes (req->ptr1, times);
1167	}	1762	}
		1763	break;
1168		1764
1169	case EIO_GROUP:	1765	case EIO_GROUP:
		1766	abort (); /* handled in eio_request */
		1767
1170	case EIO_NOP:	1768	case EIO_NOP:
1171	req->result = 0;	1769	req->result = 0;
1172	break;	1770	break;
1173		1771
1174	case EIO_CUSTOM:	1772	case EIO_CUSTOM:
1175	req->feed (req);	1773	((void ()(eio_req ))req->feed) (req);
1176	break;	1774	break;
1177		1775
1178	default:	1776	default:
		1777	errno = ENOSYS;
1179	req->result = -1;	1778	req->result = -1;
1180	break;	1779	break;
1181	}	1780	}
1182		1781
1183	req->errorno = errno;	1782	req->errorno = errno;
…		…
1203	eio_req eio_fsync (int fd, int pri, eio_cb cb, void data)	1802	eio_req eio_fsync (int fd, int pri, eio_cb cb, void data)
1204	{	1803	{
1205	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1804	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1206	}	1805	}
1207		1806
		1807	eio_req eio_msync (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1808	{
		1809	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1810	}
		1811
		1812	eio_req eio_mtouch (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1813	{
		1814	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1815	}
		1816
		1817	eio_req eio_mlock (void addr, size_t length, int pri, eio_cb cb, void *data)
		1818	{
		1819	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		1820	}
		1821
		1822	eio_req eio_mlockall (int flags, int pri, eio_cb cb, void data)
		1823	{
		1824	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		1825	}
		1826
		1827	eio_req eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void data)
		1828	{
		1829	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1830	}
		1831
1208	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)	1832	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)
1209	{	1833	{
1210	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1834	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1211	}	1835	}
1212		1836
…		…
1231	}	1855	}
1232		1856
1233	eio_req eio_fstat (int fd, int pri, eio_cb cb, void data)	1857	eio_req eio_fstat (int fd, int pri, eio_cb cb, void data)
1234	{	1858	{
1235	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1859	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1860	}
		1861
		1862	eio_req eio_fstatvfs (int fd, int pri, eio_cb cb, void data)
		1863	{
		1864	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1236	}	1865	}
1237		1866
1238	eio_req eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void data)	1867	eio_req eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void data)
1239	{	1868	{
1240	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1869	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1314	eio_req eio_lstat (const char path, int pri, eio_cb cb, void *data)	1943	eio_req eio_lstat (const char path, int pri, eio_cb cb, void *data)
1315	{	1944	{
1316	return eio__1path (EIO_LSTAT, path, pri, cb, data);	1945	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1317	}	1946	}
1318		1947
		1948	eio_req eio_statvfs (const char path, int pri, eio_cb cb, void *data)
		1949	{
		1950	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		1951	}
		1952
1319	eio_req eio_unlink (const char path, int pri, eio_cb cb, void *data)	1953	eio_req eio_unlink (const char path, int pri, eio_cb cb, void *data)
1320	{	1954	{
1321	return eio__1path (EIO_UNLINK, path, pri, cb, data);	1955	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1322	}	1956	}
1323		1957
1324	eio_req eio_rmdir (const char path, int pri, eio_cb cb, void *data)	1958	eio_req eio_rmdir (const char path, int pri, eio_cb cb, void *data)
1325	{	1959	{
1326	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1960	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1327	}	1961	}
1328		1962
1329	eio_req eio_readdir (const char path, int pri, eio_cb cb, void *data)	1963	eio_req eio_readdir (const char path, int flags, int pri, eio_cb cb, void *data)
1330	{	1964	{
1331	return eio__1path (EIO_READDIR, path, pri, cb, data);	1965	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1332	}	1966	}
1333		1967
1334	eio_req eio_mknod (const char path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1968	eio_req eio_mknod (const char path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1335	{	1969	{
1336	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;
1337	}	1971	}
1338		1972
1339	static eio_req *	1973	static eio_req *
1340	eio__2path (int type, const char path, const char new_path, int pri, eio_cb cb, void *data)	1974	eio__2path (int type, const char path, const char new_path, int pri, eio_cb cb, void *data)
1341	{	1975	{
…		…
1367	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2001	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1368	}	2002	}
1369		2003
1370	eio_req eio_custom (eio_cb execute, int pri, eio_cb cb, void data)	2004	eio_req eio_custom (eio_cb execute, int pri, eio_cb cb, void data)
1371	{	2005	{
1372	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2006	REQ (EIO_CUSTOM); req->feed = (void ()(eio_req ))execute; SEND;
1373	}	2007	}
1374		2008
1375	#endif	2009	#endif
1376		2010
1377	eio_req eio_grp (eio_cb cb, void data)	2011	eio_req eio_grp (eio_cb cb, void data)
…		…
1405		2039
1406	void eio_grp_add (eio_req grp, eio_req req)	2040	void eio_grp_add (eio_req grp, eio_req req)
1407	{	2041	{
1408	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2042	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1409		2043
		2044	grp->flags \|= EIO_FLAG_GROUPADD;
		2045
1410	++grp->size;	2046	++grp->size;
1411	req->grp = grp;	2047	req->grp = grp;
1412		2048
1413	req->grp_prev = 0;	2049	req->grp_prev = 0;
1414	req->grp_next = grp->grp_first;	2050	req->grp_next = grp->grp_first;
…		…
1423	/* misc garbage */	2059	/* misc garbage */
1424		2060
1425	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2061	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1426	{	2062	{
1427	etp_worker wrk;	2063	etp_worker wrk;
		2064	ssize_t ret;
1428		2065
1429	wrk.dbuf = 0;	2066	wrk.dbuf = 0;
1430		2067
1431	eio__sendfile (ofd, ifd, offset, count, &wrk);	2068	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1432		2069
1433	if (wrk.dbuf)	2070	if (wrk.dbuf)
1434	free (wrk.dbuf);	2071	free (wrk.dbuf);
1435	}
1436		2072
		2073	return ret;
		2074	}
		2075

Diff Legend

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

Comparing cvsroot/libeio/eio.c (file contents): Revision 1.20 by root, Thu Jun 19 09:05:43 2008 UTC vs. Revision 1.65 by root, Mon May 30 12:56:50 2011 UTC

Diff Legend

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.20 by root, Thu Jun 19 09:05:43 2008 UTC vs.
Revision 1.65 by root, Mon May 30 12:56:50 2011 UTC