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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.19 by root, Tue Jun 17 23:34:20 2008 UTC vs.
Revision 1.64 by root, Thu May 26 04:05:18 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	*
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	#include "eio.h"	40	#include "eio.h"
		41
		42	#ifdef EIO_STACKSIZE
		43	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		44	#endif
41	#include "xthread.h"	45	#include "xthread.h"
42		46
43	#include <errno.h>	47	#include <errno.h>
44	#include <stddef.h>	48	#include <stddef.h>
45	#include <stdlib.h>	49	#include <stdlib.h>
46	#include <string.h>	50	#include <string.h>
47	#include <errno.h>	51	#include <errno.h>
48	#include <sys/types.h>	52	#include <sys/types.h>
49	#include <sys/stat.h>	53	#include <sys/stat.h>
		54	#include <sys/statvfs.h>
50	#include <limits.h>	55	#include <limits.h>
51	#include <fcntl.h>	56	#include <fcntl.h>
52	#include <assert.h>	57	#include <assert.h>
53		58
54	#ifndef EIO_FINISH	59	#ifndef EIO_FINISH
…		…
64	#endif	69	#endif
65		70
66	#ifdef _WIN32	71	#ifdef _WIN32
67		72
68	/doh/	73	/doh/
69
70	#else	74	#else
71		75
72	# include "config.h"	76	# include "config.h"
73	# include <sys/time.h>	77	# include <sys/time.h>
74	# include <sys/select.h>	78	# include <sys/select.h>
75	# include <unistd.h>	79	# include <unistd.h>
76	# include <utime.h>	80	# include <utime.h>
77	# include <signal.h>	81	# include <signal.h>
78	# include <dirent.h>	82	# include <dirent.h>
79		83
		84	#if _POSIX_MEMLOCK \|\| _POSIX_MEMLOCK_RANGE \|\| _POSIX_MAPPED_FILES
		85	# include <sys/mman.h>
		86	#endif
		87
		88	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		89	# if __FreeBSD__ \|\| defined __NetBSD__ \|\| defined __OpenBSD__
		90	# define _DIRENT_HAVE_D_TYPE /* sigh */
		91	# define D_INO(de) (de)->d_fileno
		92	# define D_NAMLEN(de) (de)->d_namlen
		93	# elif __linux \|\| defined d_ino \|\| _XOPEN_SOURCE >= 600
		94	# define D_INO(de) (de)->d_ino
		95	# endif
		96
		97	#ifdef _D_EXACT_NAMLEN
		98	# undef D_NAMLEN
		99	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		100	#endif
		101
		102	# ifdef _DIRENT_HAVE_D_TYPE
		103	# define D_TYPE(de) (de)->d_type
		104	# endif
		105
80	# ifndef EIO_STRUCT_DIRENT	106	# ifndef EIO_STRUCT_DIRENT
81	# define EIO_STRUCT_DIRENT struct dirent	107	# define EIO_STRUCT_DIRENT struct dirent
82	# endif	108	# endif
83		109
84	#endif	110	#endif
85		111
86	#if HAVE_SENDFILE	112	#if HAVE_SENDFILE
87	# if __linux	113	# if __linux
88	# include <sys/sendfile.h>	114	# include <sys/sendfile.h>
89	# elif __freebsd	115	# elif __FreeBSD__ \|\| defined __APPLE__
90	# include <sys/socket.h>	116	# include <sys/socket.h>
91	# include <sys/uio.h>	117	# include <sys/uio.h>
92	# elif __hpux	118	# elif __hpux
93	# include <sys/socket.h>	119	# include <sys/socket.h>
94	# elif __solaris /* not yet */	120	# elif __solaris
95	# include <sys/sendfile.h>	121	# include <sys/sendfile.h>
96	# else	122	# else
97	# error sendfile support requested but not available	123	# error sendfile support requested but not available
98	# endif	124	# endif
99	#endif	125	#endif
100		126
101	/* number of seconds after which an idle threads exit */	127	#ifndef D_TYPE
102	#define IDLE_TIMEOUT 10	128	# define D_TYPE(de) 0
		129	#endif
		130	#ifndef D_INO
		131	# define D_INO(de) 0
		132	#endif
		133	#ifndef D_NAMLEN
		134	# define D_NAMLEN(de) strlen ((de)->d_name)
		135	#endif
103		136
104	/* used for struct dirent, AIX doesn't provide it */	137	/* used for struct dirent, AIX doesn't provide it */
105	#ifndef NAME_MAX	138	#ifndef NAME_MAX
106	# 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
107	#endif	145	#endif
108		146
109	/* buffer size for various temporary buffers */	147	/* buffer size for various temporary buffers */
110	#define EIO_BUFSIZE 65536	148	#define EIO_BUFSIZE 65536
111		149
…		…
117	errno = ENOMEM; \	155	errno = ENOMEM; \
118	if (!eio_buf) \	156	if (!eio_buf) \
119	return -1;	157	return -1;
120		158
121	#define EIO_TICKS ((1000000 + 1023) >> 10)	159	#define EIO_TICKS ((1000000 + 1023) >> 10)
		160
		161	/*****************************************************************************/
		162
		163	#if __GNUC__ >= 3
		164	# define expect(expr,value) __builtin_expect ((expr),(value))
		165	#else
		166	# define expect(expr,value) (expr)
		167	#endif
		168
		169	#define expect_false(expr) expect ((expr) != 0, 0)
		170	#define expect_true(expr) expect ((expr) != 0, 1)
122		171
123	/*****************************************************************************/	172	/*****************************************************************************/
124		173
125	#define ETP_PRI_MIN EIO_PRI_MIN	174	#define ETP_PRI_MIN EIO_PRI_MIN
126	#define ETP_PRI_MAX EIO_PRI_MAX	175	#define ETP_PRI_MAX EIO_PRI_MAX
…		…
144	if (wrk->dirp) \	193	if (wrk->dirp) \
145	{ \	194	{ \
146	closedir (wrk->dirp); \	195	closedir (wrk->dirp); \
147	wrk->dirp = 0; \	196	wrk->dirp = 0; \
148	}	197	}
		198
149	#define ETP_WORKER_COMMON \	199	#define ETP_WORKER_COMMON \
150	void *dbuf; \	200	void *dbuf; \
151	DIR *dirp;	201	DIR *dirp;
152		202
153	/*****************************************************************************/	203	/*****************************************************************************/
154		204
155	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	205	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156		206
157	/* calculcate time difference in ~1/EIO_TICKS of a second */	207	/* calculate time difference in ~1/EIO_TICKS of a second */
158	static int tvdiff (struct timeval tv1, struct timeval tv2)	208	static int tvdiff (struct timeval tv1, struct timeval tv2)
159	{	209	{
160	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	210	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	211	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
162	}	212	}
…		…
170	static unsigned int max_poll_reqs; /* reslock */	220	static unsigned int max_poll_reqs; /* reslock */
171		221
172	static volatile unsigned int nreqs; /* reqlock */	222	static volatile unsigned int nreqs; /* reqlock */
173	static volatile unsigned int nready; /* reqlock */	223	static volatile unsigned int nready; /* reqlock */
174	static volatile unsigned int npending; /* reqlock */	224	static volatile unsigned int npending; /* reqlock */
175	static volatile unsigned int max_idle = 4;	225	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		226	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
176		227
		228	static xmutex_t wrklock;
		229	static xmutex_t reslock;
		230	static xmutex_t reqlock;
		231	static xcond_t reqwait;
		232
		233	#if !HAVE_PREADWRITE
		234	/*
		235	* make our pread/pwrite emulation safe against themselves, but not against
		236	* normal read/write by using a mutex. slows down execution a lot,
		237	* but that's your problem, not mine.
		238	*/
177	static mutex_t wrklock = X_MUTEX_INIT;	239	static xmutex_t preadwritelock = X_MUTEX_INIT;
178	static mutex_t reslock = X_MUTEX_INIT;	240	#endif
179	static mutex_t reqlock = X_MUTEX_INIT;
180	static cond_t reqwait = X_COND_INIT;
181		241
182	typedef struct etp_worker	242	typedef struct etp_worker
183	{	243	{
184	/* locked by wrklock */	244	/* locked by wrklock */
185	struct etp_worker prev, next;	245	struct etp_worker prev, next;
186		246
187	thread_t tid;	247	xthread_t tid;
188		248
189	/* locked by reslock, reqlock or wrklock */	249	/* locked by reslock, reqlock or wrklock */
190	ETP_REQ req; / currently processed request */	250	ETP_REQ req; / currently processed request */
191		251
192	ETP_WORKER_COMMON	252	ETP_WORKER_COMMON
…		…
306	}	366	}
307		367
308	abort ();	368	abort ();
309	}	369	}
310		370
		371	static 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
311	static void etp_atfork_prepare (void)	379	static void etp_atfork_prepare (void)
312	{	380	{
313	X_LOCK (wrklock);	381	X_LOCK (wrklock);
314	X_LOCK (reqlock);	382	X_LOCK (reqlock);
315	X_LOCK (reslock);	383	X_LOCK (reslock);
…		…
330		398
331	static void etp_atfork_child (void)	399	static void etp_atfork_child (void)
332	{	400	{
333	ETP_REQ *prv;	401	ETP_REQ *prv;
334		402
335	while (prv = reqq_shift (&req_queue))	403	while ((prv = reqq_shift (&req_queue)))
336	ETP_DESTROY (prv);	404	ETP_DESTROY (prv);
337		405
338	while (prv = reqq_shift (&res_queue))	406	while ((prv = reqq_shift (&res_queue)))
339	ETP_DESTROY (prv);	407	ETP_DESTROY (prv);
340		408
341	while (wrk_first.next != &wrk_first)	409	while (wrk_first.next != &wrk_first)
342	{	410	{
343	etp_worker *wrk = wrk_first.next;	411	etp_worker *wrk = wrk_first.next;
…		…
353	idle = 0;	421	idle = 0;
354	nreqs = 0;	422	nreqs = 0;
355	nready = 0;	423	nready = 0;
356	npending = 0;	424	npending = 0;
357		425
358	etp_atfork_parent ();	426	etp_thread_init ();
359	}	427	}
360		428
361	static void	429	static void
362	etp_once_init (void)	430	etp_once_init (void)
363	{	431	{
		432	etp_thread_init ();
364	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);
365	}	434	}
366		435
367	static int	436	static int
368	etp_init (void (want_poll)(void), void (done_poll)(void))	437	etp_init (void (want_poll)(void), void (done_poll)(void))
…		…
402	X_UNLOCK (wrklock);	471	X_UNLOCK (wrklock);
403	}	472	}
404		473
405	static void etp_maybe_start_thread (void)	474	static void etp_maybe_start_thread (void)
406	{	475	{
407	if (etp_nthreads () >= wanted)	476	if (expect_true (etp_nthreads () >= wanted))
408	return;	477	return;
409		478
410	/* todo: maybe use idle here, but might be less exact */	479	/* todo: maybe use idle here, but might be less exact */
411	if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())	480	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
412	return;	481	return;
413		482
414	etp_start_thread ();	483	etp_start_thread ();
415	}	484	}
416		485
…		…
469		538
470	X_LOCK (reqlock);	539	X_LOCK (reqlock);
471	--nreqs;	540	--nreqs;
472	X_UNLOCK (reqlock);	541	X_UNLOCK (reqlock);
473		542
474	if (req->type == EIO_GROUP && req->size)	543	if (expect_false (req->type == EIO_GROUP && req->size))
475	{	544	{
476	req->int1 = 1; /* mark request as delayed */	545	req->int1 = 1; /* mark request as delayed */
477	continue;	546	continue;
478	}	547	}
479	else	548	else
480	{	549	{
481	int res = ETP_FINISH (req);	550	int res = ETP_FINISH (req);
482	if (res)	551	if (expect_false (res))
483	return res;	552	return res;
484	}	553	}
485		554
486	if (maxreqs && !--maxreqs)	555	if (expect_false (maxreqs && !--maxreqs))
487	break;	556	break;
488		557
489	if (maxtime)	558	if (maxtime)
490	{	559	{
491	gettimeofday (&tv_now, 0);	560	gettimeofday (&tv_now, 0);
…		…
510		579
511	static void etp_submit (ETP_REQ *req)	580	static void etp_submit (ETP_REQ *req)
512	{	581	{
513	req->pri -= ETP_PRI_MIN;	582	req->pri -= ETP_PRI_MIN;
514		583
515	if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	584	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;	585	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
517		586
		587	if (expect_false (req->type == EIO_GROUP))
		588	{
		589	/* I hope this is worth it :/ */
518	X_LOCK (reqlock);	590	X_LOCK (reqlock);
519	++nreqs;	591	++nreqs;
		592	X_UNLOCK (reqlock);
		593
		594	X_LOCK (reslock);
		595
		596	++npending;
		597
		598	if (!reqq_push (&res_queue, req) && want_poll_cb)
		599	want_poll_cb ();
		600
		601	X_UNLOCK (reslock);
		602	}
		603	else
		604	{
		605	X_LOCK (reqlock);
		606	++nreqs;
520	++nready;	607	++nready;
521	reqq_push (&req_queue, req);	608	reqq_push (&req_queue, req);
522	X_COND_SIGNAL (reqwait);	609	X_COND_SIGNAL (reqwait);
523	X_UNLOCK (reqlock);	610	X_UNLOCK (reqlock);
524		611
525	etp_maybe_start_thread ();	612	etp_maybe_start_thread ();
		613	}
526	}	614	}
527		615
528	static void etp_set_max_poll_time (double nseconds)	616	static void etp_set_max_poll_time (double nseconds)
529	{	617	{
530	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	618	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
531	max_poll_time = nseconds;	619	max_poll_time = nseconds * EIO_TICKS;
532	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	620	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
533	}	621	}
534		622
535	static void etp_set_max_poll_reqs (unsigned int maxreqs)	623	static void etp_set_max_poll_reqs (unsigned int maxreqs)
536	{	624	{
…		…
540	}	628	}
541		629
542	static void etp_set_max_idle (unsigned int nthreads)	630	static void etp_set_max_idle (unsigned int nthreads)
543	{	631	{
544	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	632	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
545	max_idle = nthreads <= 0 ? 1 : nthreads;	633	max_idle = nthreads;
		634	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		635	}
		636
		637	static void etp_set_idle_timeout (unsigned int seconds)
		638	{
		639	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		640	idle_timeout = seconds;
546	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	641	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
547	}	642	}
548		643
549	static void etp_set_min_parallel (unsigned int nthreads)	644	static void etp_set_min_parallel (unsigned int nthreads)
550	{	645	{
…		…
565		660
566	static void grp_try_feed (eio_req *grp)	661	static void grp_try_feed (eio_req *grp)
567	{	662	{
568	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	663	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569	{	664	{
570	int old_len = grp->size;	665	grp->flags &= ~EIO_FLAG_GROUPADD;
571		666
572	EIO_FEED (grp);	667	EIO_FEED (grp);
573		668
574	/* stop if no progress has been made */	669	/* stop if no progress has been made */
575	if (old_len == grp->size)	670	if (!(grp->flags & EIO_FLAG_GROUPADD))
576	{	671	{
577	grp->feed = 0;	672	grp->feed = 0;
578	break;	673	break;
579	}	674	}
580	}	675	}
…		…
678	void eio_set_max_idle (unsigned int nthreads)	773	void eio_set_max_idle (unsigned int nthreads)
679	{	774	{
680	etp_set_max_idle (nthreads);	775	etp_set_max_idle (nthreads);
681	}	776	}
682		777
		778	void eio_set_idle_timeout (unsigned int seconds)
		779	{
		780	etp_set_idle_timeout (seconds);
		781	}
		782
683	void eio_set_min_parallel (unsigned int nthreads)	783	void eio_set_min_parallel (unsigned int nthreads)
684	{	784	{
685	etp_set_min_parallel (nthreads);	785	etp_set_min_parallel (nthreads);
686	}	786	}
687		787
…		…
697		797
698	/*****************************************************************************/	798	/*****************************************************************************/
699	/* work around various missing functions */	799	/* work around various missing functions */
700		800
701	#if !HAVE_PREADWRITE	801	#if !HAVE_PREADWRITE
		802	# undef pread
		803	# undef pwrite
702	# define pread eio__pread	804	# define pread eio__pread
703	# define pwrite eio__pwrite	805	# 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		806
712	static ssize_t	807	static ssize_t
713	eio__pread (int fd, void *buf, size_t count, off_t offset)	808	eio__pread (int fd, void *buf, size_t count, off_t offset)
714	{	809	{
715	ssize_t res;	810	ssize_t res;
…		…
733		828
734	X_LOCK (preadwritelock);	829	X_LOCK (preadwritelock);
735	ooffset = lseek (fd, 0, SEEK_CUR);	830	ooffset = lseek (fd, 0, SEEK_CUR);
736	lseek (fd, offset, SEEK_SET);	831	lseek (fd, offset, SEEK_SET);
737	res = write (fd, buf, count);	832	res = write (fd, buf, count);
738	lseek (fd, offset, SEEK_SET);	833	lseek (fd, ooffset, SEEK_SET);
739	X_UNLOCK (preadwritelock);	834	X_UNLOCK (preadwritelock);
740		835
741	return res;	836	return res;
742	}	837	}
743	#endif	838	#endif
744		839
745	#ifndef HAVE_FUTIMES	840	#ifndef HAVE_UTIMES
746		841
		842	# undef utimes
747	# define utimes(path,times) eio__utimes (path, times)	843	# define utimes(path,times) eio__utimes (path, times)
748	# define futimes(fd,times) eio__futimes (fd, times)
749		844
750	static int	845	static int
751	eio__utimes (const char *filename, const struct timeval times[2])	846	eio__utimes (const char *filename, const struct timeval times[2])
752	{	847	{
753	if (times)	848	if (times)
…		…
761	}	856	}
762	else	857	else
763	return utime (filename, 0);	858	return utime (filename, 0);
764	}	859	}
765		860
		861	#endif
		862
		863	#ifndef HAVE_FUTIMES
		864
		865	# undef futimes
		866	# define futimes(fd,times) eio__futimes (fd, times)
		867
766	static int eio__futimes (int fd, const struct timeval tv[2])	868	static int eio__futimes (int fd, const struct timeval tv[2])
767	{	869	{
768	errno = ENOSYS;	870	errno = ENOSYS;
769	return -1;	871	return -1;
770	}	872	}
771		873
772	#endif	874	#endif
773		875
774	#if !HAVE_FDATASYNC	876	#if !HAVE_FDATASYNC
		877	# undef fdatasync
775	# define fdatasync fsync	878	# define fdatasync(fd) fsync (fd)
776	#endif	879	#endif
		880
		881	/* sync_file_range always needs emulation */
		882	int
		883	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		884	{
		885	#if HAVE_SYNC_FILE_RANGE
		886	int res;
		887
		888	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		889	\|\| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		890	\|\| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		891	{
		892	flags = 0
		893	\| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		894	\| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		895	\| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		896	}
		897
		898	res = sync_file_range (fd, offset, nbytes, flags);
		899
		900	if (!res \|\| errno != ENOSYS)
		901	return res;
		902	#endif
		903
		904	/* even though we could play tricks with the flags, it's better to always
		905	* call fdatasync, as that matches the expectation of its users best */
		906	return fdatasync (fd);
		907	}
777		908
778	#if !HAVE_READAHEAD	909	#if !HAVE_READAHEAD
		910	# undef readahead
779	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	911	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780		912
781	static ssize_t	913	static ssize_t
782	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	914	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783	{	915	{
…		…
810		942
811	#if HAVE_SENDFILE	943	#if HAVE_SENDFILE
812	# if __linux	944	# if __linux
813	res = sendfile (ofd, ifd, &offset, count);	945	res = sendfile (ofd, ifd, &offset, count);
814		946
815	# elif __freebsd	947	# elif __FreeBSD__
816	/*	948	/*
817	* Of course, the freebsd sendfile is a dire hack with no thoughts	949	* Of course, the freebsd sendfile is a dire hack with no thoughts
818	* wasted on making it similar to other I/O functions.	950	* wasted on making it similar to other I/O functions.
819	*/	951	*/
820	{	952	{
821	off_t sbytes;	953	off_t sbytes;
822	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	954	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
823		955
824	if (res < 0 && sbytes)	956	#if 0 /* according to the manpage, this is correct, but broken behaviour */
825	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	957	/* freebsd' sendfile will return 0 on success */
		958	/* freebsd 8 documents it as only setting sbytes on EINTR and EAGAIN, but /
		959	/* not on e.g. EIO or EPIPE - sounds broken */
		960	if ((res < 0 && (errno == EAGAIN \|\| errno == EINTR) && sbytes) \|\| res == 0)
		961	res = sbytes;
		962	#endif
		963
		964	/* according to source inspection, this is correct, and useful behaviour */
		965	if (sbytes)
		966	res = sbytes;
		967	}
		968
		969	# elif defined (__APPLE__)
		970
		971	{
		972	off_t sbytes = count;
		973	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		974
		975	/* according to the manpage, sbytes is always valid */
		976	if (sbytes)
826	res = sbytes;	977	res = sbytes;
827	}	978	}
828		979
829	# elif __hpux	980	# elif __hpux
830	res = sendfile (ofd, ifd, offset, count, 0, 0);	981	res = sendfile (ofd, ifd, offset, count, 0, 0);
…		…
844	if (res < 0 && sbytes)	995	if (res < 0 && sbytes)
845	res = sbytes;	996	res = sbytes;
846	}	997	}
847		998
848	# endif	999	# endif
		1000
		1001	#elif defined (_WIN32)
		1002
		1003	/* does not work, just for documentation of what would need to be done */
		1004	{
		1005	HANDLE h = TO_SOCKET (ifd);
		1006	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1007	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1008	}
		1009
849	#else	1010	#else
850	res = -1;	1011	res = -1;
851	errno = ENOSYS;	1012	errno = ENOSYS;
852	#endif	1013	#endif
853		1014
854	if (res < 0	1015	if (res < 0
855	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK	1016	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK
		1017	/* BSDs */
		1018	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1019	\|\| errno == ENOTSUP
		1020	#endif
		1021	\|\| errno == EOPNOTSUPP /* BSDs */
856	#if __solaris	1022	#if __solaris
857	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE	1023	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE
858	#endif	1024	#endif
859	)	1025	)
860	)	1026	)
…		…
891	}	1057	}
892		1058
893	return res;	1059	return res;
894	}	1060	}
895		1061
		1062	static signed char
		1063	eio_dent_cmp (const eio_dirent a, const eio_dirent b)
		1064	{
		1065	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1066	: a->inode < b->inode ? -1
		1067	: a->inode > b->inode ? 1
		1068	: 0;
		1069	}
		1070
		1071	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1072
		1073	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1074	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1075
		1076	static void
		1077	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1078	{
		1079	unsigned char bits [9 + sizeof (ino_t) * 8];
		1080	unsigned char *bit = bits;
		1081
		1082	assert (CHAR_BIT == 8);
		1083	assert (sizeof (eio_dirent) * 8 < 256);
		1084	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1085	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1086
		1087	if (size <= EIO_SORT_FAST)
		1088	return;
		1089
		1090	/* first prepare an array of bits to test in our radix sort */
		1091	/* try to take endianness into account, as well as differences in ino_t sizes */
		1092	/* inode_bits must contain all inodes ORed together */
		1093	/* which is used to skip bits that are 0 everywhere, which is very common */
		1094	{
		1095	ino_t endianness;
		1096	int i, j;
		1097
		1098	/* we store the byte offset of byte n into byte n of "endianness" */
		1099	for (i = 0; i < sizeof (ino_t); ++i)
		1100	((unsigned char *)&endianness)[i] = i;
		1101
		1102	*bit++ = 0;
		1103
		1104	for (i = 0; i < sizeof (ino_t); ++i)
		1105	{
		1106	/* shifting off the byte offsets out of "endianness" */
		1107	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1108	endianness >>= 8;
		1109
		1110	for (j = 0; j < 8; ++j)
		1111	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1112	*bit++ = offs + j;
		1113	}
		1114
		1115	for (j = 0; j < 8; ++j)
		1116	if (score_bits & (1 << j))
		1117	bit++ = offsetof (eio_dirent, score) 8 + j;
		1118	}
		1119
		1120	/* now actually do the sorting (a variant of MSD radix sort) */
		1121	{
		1122	eio_dirent base_stk [9 + sizeof (ino_t) 8], *base;
		1123	eio_dirent end_stk [9 + sizeof (ino_t) 8], *end;
		1124	unsigned char bit_stk [9 + sizeof (ino_t) 8];
		1125	int stk_idx = 0;
		1126
		1127	base_stk [stk_idx] = dents;
		1128	end_stk [stk_idx] = dents + size;
		1129	bit_stk [stk_idx] = bit - 1;
		1130
		1131	do
		1132	{
		1133	base = base_stk [stk_idx];
		1134	end = end_stk [stk_idx];
		1135	bit = bit_stk [stk_idx];
		1136
		1137	for (;;)
		1138	{
		1139	unsigned char O = *bit >> 3;
		1140	unsigned char M = 1 << (*bit & 7);
		1141
		1142	eio_dirent *a = base;
		1143	eio_dirent *b = end;
		1144
		1145	if (b - a < EIO_SORT_CUTOFF)
		1146	break;
		1147
		1148	/* now bit-partition the array on the bit */
		1149	/* this ugly asymmetric loop seems to perform much better than typical */
		1150	/* partition algos found in the literature */
		1151	do
		1152	if (!(((unsigned char *)a)[O] & M))
		1153	++a;
		1154	else if (!(((unsigned char *)--b)[O] & M))
		1155	{
		1156	eio_dirent tmp = a; a = b; b = tmp;
		1157	++a;
		1158	}
		1159	while (b > a);
		1160
		1161	/* next bit, or stop, if no bits left in this path */
		1162	if (!*--bit)
		1163	break;
		1164
		1165	base_stk [stk_idx] = a;
		1166	end_stk [stk_idx] = end;
		1167	bit_stk [stk_idx] = bit;
		1168	++stk_idx;
		1169
		1170	end = a;
		1171	}
		1172	}
		1173	while (stk_idx--);
		1174	}
		1175	}
		1176
		1177	static void
		1178	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1179	{
		1180	/* first move the smallest element to the front, to act as a sentinel */
		1181	{
		1182	int i;
		1183	eio_dirent *min = dents;
		1184
		1185	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1186	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1187	if (EIO_DENT_CMP (dents [i], <, *min))
		1188	min = &dents [i];
		1189
		1190	/* swap elements 0 and j (minimum) */
		1191	{
		1192	eio_dirent tmp = dents; dents = min; min = tmp;
		1193	}
		1194	}
		1195
		1196	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1197	{
		1198	eio_dirent i, j;
		1199
		1200	for (i = dents + 1; i < dents + size; ++i)
		1201	{
		1202	eio_dirent value = *i;
		1203
		1204	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1205	j [1] = j [0];
		1206
		1207	j [1] = value;
		1208	}
		1209	}
		1210	}
		1211
		1212	static void
		1213	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1214	{
		1215	if (size <= 1)
		1216	return; /* our insertion sort relies on size > 0 */
		1217
		1218	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1219	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1220	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1221
		1222	/* use an insertion sort at the end, or for small arrays, */
		1223	/* as insertion sort is more efficient for small partitions */
		1224	eio_dent_insertion_sort (dents, size);
		1225	}
		1226
896	/* read a full directory */	1227	/* read a full directory */
897	static void	1228	static void
898	eio__scandir (eio_req req, etp_worker self)	1229	eio__scandir (eio_req req, etp_worker self)
899	{	1230	{
900	DIR *dirp;	1231	DIR *dirp;
901	EIO_STRUCT_DIRENT *entp;	1232	EIO_STRUCT_DIRENT *entp;
902	char name, names;	1233	char name, names;
903	int memlen = 4096;	1234	int namesalloc = 4096;
904	int memofs = 0;	1235	int namesoffs = 0;
		1236	int flags = req->int1;
		1237	eio_dirent *dents = 0;
		1238	int dentalloc = 128;
905	int res = 0;	1239	int dentoffs = 0;
		1240	ino_t inode_bits = 0;
		1241
		1242	req->result = -1;
		1243
		1244	if (!(flags & EIO_READDIR_DENTS))
		1245	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);
906		1246
907	X_LOCK (wrklock);	1247	X_LOCK (wrklock);
		1248	/* the corresponding closedir is in ETP_WORKER_CLEAR */
908	self->dirp = dirp = opendir (req->ptr1);	1249	self->dirp = dirp = opendir (req->ptr1);
		1250
909	req->flags \|= EIO_FLAG_PTR2_FREE;	1251	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;
		1252	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
910	req->ptr2 = names = malloc (memlen);	1253	req->ptr2 = names = malloc (namesalloc);
911	X_UNLOCK (wrklock);	1254	X_UNLOCK (wrklock);
912		1255
913	if (dirp && names)	1256	if (dirp && names && (!flags \|\| dents))
914	for (;;)	1257	for (;;)
915	{	1258	{
916	errno = 0;	1259	errno = 0;
917	entp = readdir (dirp);	1260	entp = readdir (dirp);
918		1261
919	if (!entp)	1262	if (!entp)
		1263	{
		1264	if (errno)
		1265	break;
		1266
		1267	/* sort etc. */
		1268	req->int1 = flags;
		1269	req->result = dentoffs;
		1270
		1271	if (flags & EIO_READDIR_STAT_ORDER)
		1272	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1273	else if (flags & EIO_READDIR_DIRS_FIRST)
		1274	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1275	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1276	else
		1277	{
		1278	/* in this case, all is known, and we just put dirs first and sort them */
		1279	eio_dirent *oth = dents + dentoffs;
		1280	eio_dirent *dir = dents;
		1281
		1282	/* now partition dirs to the front, and non-dirs to the back */
		1283	/* by walking from both sides and swapping if necessary */
		1284	while (oth > dir)
		1285	{
		1286	if (dir->type == EIO_DT_DIR)
		1287	++dir;
		1288	else if ((--oth)->type == EIO_DT_DIR)
		1289	{
		1290	eio_dirent tmp = dir; dir = oth; oth = tmp;
		1291
		1292	++dir;
		1293	}
		1294	}
		1295
		1296	/* now sort the dirs only (dirs all have the same score) */
		1297	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1298	}
		1299
920	break;	1300	break;
		1301	}
921		1302
		1303	/* now add the entry to our list(s) */
922	name = entp->d_name;	1304	name = entp->d_name;
923		1305
		1306	/* skip . and .. entries */
924	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))	1307	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))
925	{	1308	{
926	int len = strlen (name) + 1;	1309	int len = D_NAMLEN (entp) + 1;
927		1310
928	res++;	1311	while (expect_false (namesoffs + len > namesalloc))
929
930	while (memofs + len > memlen)
931	{	1312	{
932	memlen *= 2;	1313	namesalloc *= 2;
933	X_LOCK (wrklock);	1314	X_LOCK (wrklock);
934	req->ptr2 = names = realloc (names, memlen);	1315	req->ptr2 = names = realloc (names, namesalloc);
935	X_UNLOCK (wrklock);	1316	X_UNLOCK (wrklock);
936		1317
937	if (!names)	1318	if (!names)
938	break;	1319	break;
939	}	1320	}
940		1321
941	memcpy (names + memofs, name, len);	1322	memcpy (names + namesoffs, name, len);
		1323
		1324	if (dents)
		1325	{
		1326	struct eio_dirent *ent;
		1327
		1328	if (expect_false (dentoffs == dentalloc))
		1329	{
		1330	dentalloc *= 2;
		1331	X_LOCK (wrklock);
		1332	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1333	X_UNLOCK (wrklock);
		1334
		1335	if (!dents)
		1336	break;
		1337	}
		1338
		1339	ent = dents + dentoffs;
		1340
		1341	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1342	ent->namelen = len - 1;
		1343	ent->inode = D_INO (entp);
		1344
		1345	inode_bits \|= ent->inode;
		1346
		1347	switch (D_TYPE (entp))
		1348	{
		1349	default:
		1350	ent->type = EIO_DT_UNKNOWN;
		1351	flags \|= EIO_READDIR_FOUND_UNKNOWN;
		1352	break;
		1353
		1354	#ifdef DT_FIFO
		1355	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1356	#endif
		1357	#ifdef DT_CHR
		1358	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1359	#endif
		1360	#ifdef DT_MPC
		1361	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1362	#endif
		1363	#ifdef DT_DIR
		1364	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1365	#endif
		1366	#ifdef DT_NAM
		1367	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1368	#endif
		1369	#ifdef DT_BLK
		1370	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1371	#endif
		1372	#ifdef DT_MPB
		1373	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1374	#endif
		1375	#ifdef DT_REG
		1376	case DT_REG: ent->type = EIO_DT_REG; break;
		1377	#endif
		1378	#ifdef DT_NWK
		1379	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1380	#endif
		1381	#ifdef DT_CMP
		1382	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1383	#endif
		1384	#ifdef DT_LNK
		1385	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1386	#endif
		1387	#ifdef DT_SOCK
		1388	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1389	#endif
		1390	#ifdef DT_DOOR
		1391	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1392	#endif
		1393	#ifdef DT_WHT
		1394	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1395	#endif
		1396	}
		1397
		1398	ent->score = 7;
		1399
		1400	if (flags & EIO_READDIR_DIRS_FIRST)
		1401	{
		1402	if (ent->type == EIO_DT_UNKNOWN)
		1403	{
		1404	if (name == '.') / leading dots are likely directories, and, in any case, rare */
		1405	ent->score = 1;
		1406	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1407	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1408	}
		1409	else if (ent->type == EIO_DT_DIR)
		1410	ent->score = 0;
		1411	}
		1412	}
		1413
942	memofs += len;	1414	namesoffs += len;
		1415	++dentoffs;
		1416	}
		1417
		1418	if (EIO_CANCELLED (req))
		1419	{
		1420	errno = ECANCELED;
		1421	break;
943	}	1422	}
944	}	1423	}
		1424	}
945		1425
946	if (errno)	1426	#ifdef PAGESIZE
947	res = -1;	1427	# define eio_pagesize() PAGESIZE
		1428	#else
		1429	static intptr_t
		1430	eio_pagesize (void)
		1431	{
		1432	static intptr_t page;
		1433
		1434	if (!page)
		1435	page = sysconf (_SC_PAGESIZE);
		1436
		1437	return page;
		1438	}
		1439	#endif
		1440
		1441	static void
		1442	eio_page_align (void *addr, size_t length)
		1443	{
		1444	intptr_t mask = eio_pagesize () - 1;
		1445
		1446	/* round down addr */
		1447	intptr_t adj = mask & (intptr_t)*addr;
		1448
		1449	addr = (void )((intptr_t)*addr - adj);
		1450	*length += adj;
		1451
		1452	/* round up length */
		1453	length = (length + mask) & ~mask;
		1454	}
		1455
		1456	#if !_POSIX_MEMLOCK
		1457	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1458	#else
		1459
		1460	static int
		1461	eio__mlockall (int flags)
		1462	{
		1463	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1464	extern int mallopt (int, int);
		1465	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1466	#endif
		1467
		1468	if (EIO_MCL_CURRENT != MCL_CURRENT
		1469	\|\| EIO_MCL_FUTURE != MCL_FUTURE)
		1470	{
		1471	flags = 0
		1472	\| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1473	\| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1474	}
		1475
		1476	return mlockall (flags);
		1477	}
		1478	#endif
		1479
		1480	#if !_POSIX_MEMLOCK_RANGE
		1481	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1482	#else
		1483
		1484	static int
		1485	eio__mlock (void *addr, size_t length)
		1486	{
		1487	eio_page_align (&addr, &length);
		1488
		1489	return mlock (addr, length);
		1490	}
		1491
		1492	#endif
		1493
		1494	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1495	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1496	#else
		1497
		1498	int
		1499	eio__msync (void *mem, size_t len, int flags)
		1500	{
		1501	eio_page_align (&mem, &len);
		1502
		1503	if (EIO_MS_ASYNC != MS_SYNC
		1504	\|\| EIO_MS_INVALIDATE != MS_INVALIDATE
		1505	\|\| EIO_MS_SYNC != MS_SYNC)
		1506	{
		1507	flags = 0
		1508	\| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1509	\| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1510	\| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1511	}
		1512
		1513	return msync (mem, len, flags);
		1514	}
		1515
		1516	#endif
		1517
		1518	int
		1519	eio__mtouch (void *mem, size_t len, int flags)
		1520	{
		1521	eio_page_align (&mem, &len);
		1522
948		1523	{
949	req->result = res;	1524	intptr_t addr = (intptr_t)mem;
		1525	intptr_t end = addr + len;
		1526	intptr_t page = eio_pagesize ();
		1527
		1528	if (addr < end)
		1529	if (flags & EIO_MT_MODIFY) /* modify */
		1530	do { ((volatile sig_atomic_t )addr) \|= 0; } while ((addr += page) < len);
		1531	else
		1532	do { ((volatile sig_atomic_t )addr) ; } while ((addr += page) < len);
		1533	}
		1534
		1535	return 0;
950	}	1536	}
951		1537
952	/*****************************************************************************/	1538	/*****************************************************************************/
953		1539
954	#define ALLOC(len) \	1540	#define ALLOC(len) \
…		…
986	if (req)	1572	if (req)
987	break;	1573	break;
988		1574
989	++idle;	1575	++idle;
990		1576
991	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1577	ts.tv_sec = time (0) + idle_timeout;
992	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1578	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
993	{	1579	{
994	if (idle > max_idle)	1580	if (idle > max_idle)
995	{	1581	{
996	--idle;	1582	--idle;
…		…
1075	return 0; \	1661	return 0; \
1076	}	1662	}
1077		1663
1078	static void eio_execute (etp_worker self, eio_req req)	1664	static void eio_execute (etp_worker self, eio_req req)
1079	{	1665	{
1080	errno = 0;
1081
1082	switch (req->type)	1666	switch (req->type)
1083	{	1667	{
1084	case EIO_READ: ALLOC (req->size);	1668	case EIO_READ: ALLOC (req->size);
1085	req->result = req->offs >= 0	1669	req->result = req->offs >= 0
1086	? pread (req->int1, req->ptr2, req->size, req->offs)	1670	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1097	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1681	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1682	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1099	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1683	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1100	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1684	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1101		1685
		1686	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1687	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1688	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1689	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1690
1102	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1691	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1103	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1692	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1104	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	1693	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1105	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1694	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1106	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	1695	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1113	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1702	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1114	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	1703	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1115	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	1704	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1116	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	1705	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1117	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	1706	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1118	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	1707	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1119		1708
1120	case EIO_READLINK: ALLOC (NAME_MAX);	1709	case EIO_READLINK: ALLOC (PATH_MAX);
1121	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	1710	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1122		1711
1123	case EIO_SYNC: req->result = 0; sync (); break;	1712	case EIO_SYNC: req->result = 0; sync (); break;
1124	case EIO_FSYNC: req->result = fsync (req->int1); break;	1713	case EIO_FSYNC: req->result = fsync (req->int1); break;
1125	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1714	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1715	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1716	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1717	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1718	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		1719	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1126		1720
1127	case EIO_READDIR: eio__scandir (req, self); break;	1721	case EIO_READDIR: eio__scandir (req, self); break;
1128		1722
1129	case EIO_BUSY:	1723	case EIO_BUSY:
1130	#ifdef _WIN32	1724	#ifdef _WIN32
1131	Sleep (req->nv1 * 1000.);	1725	Sleep (req->nv1 * 1e3);
1132	#else	1726	#else
1133	{	1727	{
1134	struct timeval tv;	1728	struct timeval tv;
1135		1729
1136	tv.tv_sec = req->nv1;	1730	tv.tv_sec = req->nv1;
1137	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	1731	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1138		1732
1139	req->result = select (0, 0, 0, 0, &tv);	1733	req->result = select (0, 0, 0, 0, &tv);
1140	}	1734	}
1141	#endif	1735	#endif
1142	break;	1736	break;
…		…
1157	times = tv;	1751	times = tv;
1158	}	1752	}
1159	else	1753	else
1160	times = 0;	1754	times = 0;
1161		1755
1162
1163	req->result = req->type == EIO_FUTIME	1756	req->result = req->type == EIO_FUTIME
1164	? futimes (req->int1, times)	1757	? futimes (req->int1, times)
1165	: utimes (req->ptr1, times);	1758	: utimes (req->ptr1, times);
1166	}	1759	}
		1760	break;
1167		1761
1168	case EIO_GROUP:	1762	case EIO_GROUP:
		1763	abort (); /* handled in eio_request */
		1764
1169	case EIO_NOP:	1765	case EIO_NOP:
1170	req->result = 0;	1766	req->result = 0;
1171	break;	1767	break;
1172		1768
1173	case EIO_CUSTOM:	1769	case EIO_CUSTOM:
1174	req->feed (req);	1770	((void ()(eio_req ))req->feed) (req);
1175	break;	1771	break;
1176		1772
1177	default:	1773	default:
		1774	errno = ENOSYS;
1178	req->result = -1;	1775	req->result = -1;
1179	break;	1776	break;
1180	}	1777	}
1181		1778
1182	req->errorno = errno;	1779	req->errorno = errno;
…		…
1202	eio_req eio_fsync (int fd, int pri, eio_cb cb, void data)	1799	eio_req eio_fsync (int fd, int pri, eio_cb cb, void data)
1203	{	1800	{
1204	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1801	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1205	}	1802	}
1206		1803
		1804	eio_req eio_msync (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1805	{
		1806	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1807	}
		1808
		1809	eio_req eio_mtouch (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1810	{
		1811	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1812	}
		1813
		1814	eio_req eio_mlock (void addr, size_t length, int pri, eio_cb cb, void *data)
		1815	{
		1816	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		1817	}
		1818
		1819	eio_req eio_mlockall (int flags, int pri, eio_cb cb, void data)
		1820	{
		1821	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		1822	}
		1823
		1824	eio_req eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void data)
		1825	{
		1826	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1827	}
		1828
1207	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)	1829	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)
1208	{	1830	{
1209	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1831	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1210	}	1832	}
1211		1833
…		…
1230	}	1852	}
1231		1853
1232	eio_req eio_fstat (int fd, int pri, eio_cb cb, void data)	1854	eio_req eio_fstat (int fd, int pri, eio_cb cb, void data)
1233	{	1855	{
1234	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1856	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1857	}
		1858
		1859	eio_req eio_fstatvfs (int fd, int pri, eio_cb cb, void data)
		1860	{
		1861	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1235	}	1862	}
1236		1863
1237	eio_req eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void data)	1864	eio_req eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void data)
1238	{	1865	{
1239	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1866	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1313	eio_req eio_lstat (const char path, int pri, eio_cb cb, void *data)	1940	eio_req eio_lstat (const char path, int pri, eio_cb cb, void *data)
1314	{	1941	{
1315	return eio__1path (EIO_LSTAT, path, pri, cb, data);	1942	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1316	}	1943	}
1317		1944
		1945	eio_req eio_statvfs (const char path, int pri, eio_cb cb, void *data)
		1946	{
		1947	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		1948	}
		1949
1318	eio_req eio_unlink (const char path, int pri, eio_cb cb, void *data)	1950	eio_req eio_unlink (const char path, int pri, eio_cb cb, void *data)
1319	{	1951	{
1320	return eio__1path (EIO_UNLINK, path, pri, cb, data);	1952	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1321	}	1953	}
1322		1954
1323	eio_req eio_rmdir (const char path, int pri, eio_cb cb, void *data)	1955	eio_req eio_rmdir (const char path, int pri, eio_cb cb, void *data)
1324	{	1956	{
1325	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1957	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1326	}	1958	}
1327		1959
1328	eio_req eio_readdir (const char path, int pri, eio_cb cb, void *data)	1960	eio_req eio_readdir (const char path, int flags, int pri, eio_cb cb, void *data)
1329	{	1961	{
1330	return eio__1path (EIO_READDIR, path, pri, cb, data);	1962	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1331	}	1963	}
1332		1964
1333	eio_req eio_mknod (const char path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1965	eio_req eio_mknod (const char path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1334	{	1966	{
1335	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	1967	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1336	}	1968	}
1337		1969
1338	static eio_req *	1970	static eio_req *
1339	eio__2path (int type, const char path, const char new_path, int pri, eio_cb cb, void *data)	1971	eio__2path (int type, const char path, const char new_path, int pri, eio_cb cb, void *data)
1340	{	1972	{
…		…
1366	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1998	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1367	}	1999	}
1368		2000
1369	eio_req eio_custom (eio_cb execute, int pri, eio_cb cb, void data)	2001	eio_req eio_custom (eio_cb execute, int pri, eio_cb cb, void data)
1370	{	2002	{
1371	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2003	REQ (EIO_CUSTOM); req->feed = (void ()(eio_req ))execute; SEND;
1372	}	2004	}
1373		2005
1374	#endif	2006	#endif
1375		2007
1376	eio_req eio_grp (eio_cb cb, void data)	2008	eio_req eio_grp (eio_cb cb, void data)
…		…
1404		2036
1405	void eio_grp_add (eio_req grp, eio_req req)	2037	void eio_grp_add (eio_req grp, eio_req req)
1406	{	2038	{
1407	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2039	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1408		2040
		2041	grp->flags \|= EIO_FLAG_GROUPADD;
		2042
1409	++grp->size;	2043	++grp->size;
1410	req->grp = grp;	2044	req->grp = grp;
1411		2045
1412	req->grp_prev = 0;	2046	req->grp_prev = 0;
1413	req->grp_next = grp->grp_first;	2047	req->grp_next = grp->grp_first;
…		…
1422	/* misc garbage */	2056	/* misc garbage */
1423		2057
1424	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2058	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1425	{	2059	{
1426	etp_worker wrk;	2060	etp_worker wrk;
		2061	ssize_t ret;
1427		2062
1428	wrk.dbuf = 0;	2063	wrk.dbuf = 0;
1429		2064
1430	eio__sendfile (ofd, ifd, offset, count, &wrk);	2065	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1431		2066
1432	if (wrk.dbuf)	2067	if (wrk.dbuf)
1433	free (wrk.dbuf);	2068	free (wrk.dbuf);
1434	}
1435		2069
		2070	return ret;
		2071	}
		2072

Diff Legend

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

Comparing cvsroot/libeio/eio.c (file contents): Revision 1.19 by root, Tue Jun 17 23:34:20 2008 UTC vs. Revision 1.64 by root, Thu May 26 04:05:18 2011 UTC

Diff Legend

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.19 by root, Tue Jun 17 23:34:20 2008 UTC vs.
Revision 1.64 by root, Thu May 26 04:05:18 2011 UTC