ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.14 by root, Sat May 17 12:17:25 2008 UTC vs.
Revision 1.62 by root, Tue Feb 15 03:15:16 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008 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
…		…
120		158
121	#define EIO_TICKS ((1000000 + 1023) >> 10)	159	#define EIO_TICKS ((1000000 + 1023) >> 10)
122		160
123	/*****************************************************************************/	161	/*****************************************************************************/
124		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)
		171
		172	/*****************************************************************************/
		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
		176
		177	struct etp_worker;
127		178
128	#define ETP_REQ eio_req	179	#define ETP_REQ eio_req
129	#define ETP_DESTROY(req) eio_destroy (req)	180	#define ETP_DESTROY(req) eio_destroy (req)
130	static int eio_finish (eio_req *req);	181	static int eio_finish (eio_req *req);
131	#define ETP_FINISH(req) eio_finish (req)	182	#define ETP_FINISH(req) eio_finish (req)
		183	static void eio_execute (struct etp_worker *self, eio_req *req);
		184	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
132		185
133	#define ETP_WORKER_CLEAR(req) \	186	#define ETP_WORKER_CLEAR(req) \
134	if (wrk->dbuf) \	187	if (wrk->dbuf) \
135	{ \	188	{ \
136	free (wrk->dbuf); \	189	free (wrk->dbuf); \
…		…
140	if (wrk->dirp) \	193	if (wrk->dirp) \
141	{ \	194	{ \
142	closedir (wrk->dirp); \	195	closedir (wrk->dirp); \
143	wrk->dirp = 0; \	196	wrk->dirp = 0; \
144	}	197	}
		198
145	#define ETP_WORKER_COMMON \	199	#define ETP_WORKER_COMMON \
146	void *dbuf; \	200	void *dbuf; \
147	DIR *dirp;	201	DIR *dirp;
148		202
149	/*****************************************************************************/	203	/*****************************************************************************/
150		204
151	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	205	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
152		206
153	/* calculcate time difference in ~1/EIO_TICKS of a second */	207	/* calculate time difference in ~1/EIO_TICKS of a second */
154	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	208	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
155	{	209	{
156	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	210	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
157	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	211	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
158	}	212	}
…		…
166	static unsigned int max_poll_reqs; /* reslock */	220	static unsigned int max_poll_reqs; /* reslock */
167		221
168	static volatile unsigned int nreqs; /* reqlock */	222	static volatile unsigned int nreqs; /* reqlock */
169	static volatile unsigned int nready; /* reqlock */	223	static volatile unsigned int nready; /* reqlock */
170	static volatile unsigned int npending; /* reqlock */	224	static volatile unsigned int npending; /* reqlock */
171	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 */
172		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	*/
173	static mutex_t wrklock = X_MUTEX_INIT;	239	static xmutex_t preadwritelock = X_MUTEX_INIT;
174	static mutex_t reslock = X_MUTEX_INIT;	240	#endif
175	static mutex_t reqlock = X_MUTEX_INIT;
176	static cond_t reqwait = X_COND_INIT;
177		241
178	typedef struct etp_worker	242	typedef struct etp_worker
179	{	243	{
180	/* locked by wrklock */	244	/* locked by wrklock */
181	struct etp_worker *prev, *next;	245	struct etp_worker *prev, *next;
182		246
183	thread_t tid;	247	xthread_t tid;
184		248
185	/* locked by reslock, reqlock or wrklock */	249	/* locked by reslock, reqlock or wrklock */
186	ETP_REQ *req; /* currently processed request */	250	ETP_REQ *req; /* currently processed request */
187		251
188	ETP_WORKER_COMMON	252	ETP_WORKER_COMMON
…		…
302	}	366	}
303		367
304	abort ();	368	abort ();
305	}	369	}
306		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
307	static void etp_atfork_prepare (void)	379	static void etp_atfork_prepare (void)
308	{	380	{
309	X_LOCK (wrklock);	381	X_LOCK (wrklock);
310	X_LOCK (reqlock);	382	X_LOCK (reqlock);
311	X_LOCK (reslock);	383	X_LOCK (reslock);
312	#if !HAVE_PREADWRITE	384	#if !HAVE_PREADWRITE
313	X_LOCK (preadwritelock);	385	X_LOCK (preadwritelock);
314	#endif	386	#endif
315	#if !HAVE_READDIR_R
316	X_LOCK (readdirlock);
317	#endif
318	}	387	}
319		388
320	static void etp_atfork_parent (void)	389	static void etp_atfork_parent (void)
321	{	390	{
322	#if !HAVE_READDIR_R
323	X_UNLOCK (readdirlock);
324	#endif
325	#if !HAVE_PREADWRITE	391	#if !HAVE_PREADWRITE
326	X_UNLOCK (preadwritelock);	392	X_UNLOCK (preadwritelock);
327	#endif	393	#endif
328	X_UNLOCK (reslock);	394	X_UNLOCK (reslock);
329	X_UNLOCK (reqlock);	395	X_UNLOCK (reqlock);
…		…
332		398
333	static void etp_atfork_child (void)	399	static void etp_atfork_child (void)
334	{	400	{
335	ETP_REQ *prv;	401	ETP_REQ *prv;
336		402
337	while (prv = reqq_shift (&req_queue))	403	while ((prv = reqq_shift (&req_queue)))
338	ETP_DESTROY (prv);	404	ETP_DESTROY (prv);
339		405
340	while (prv = reqq_shift (&res_queue))	406	while ((prv = reqq_shift (&res_queue)))
341	ETP_DESTROY (prv);	407	ETP_DESTROY (prv);
342		408
343	while (wrk_first.next != &wrk_first)	409	while (wrk_first.next != &wrk_first)
344	{	410	{
345	etp_worker *wrk = wrk_first.next;	411	etp_worker *wrk = wrk_first.next;
…		…
355	idle = 0;	421	idle = 0;
356	nreqs = 0;	422	nreqs = 0;
357	nready = 0;	423	nready = 0;
358	npending = 0;	424	npending = 0;
359		425
360	etp_atfork_parent ();	426	etp_thread_init ();
361	}	427	}
362		428
363	static void	429	static void
364	etp_once_init (void)	430	etp_once_init (void)
365	{	431	{
		432	etp_thread_init ();
366	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);
367	}	434	}
368		435
369	static int	436	static int
370	etp_init (void (*want_poll)(void), void (*done_poll)(void))	437	etp_init (void (*want_poll)(void), void (*done_poll)(void))
…		…
373		440
374	pthread_once (&doinit, etp_once_init);	441	pthread_once (&doinit, etp_once_init);
375		442
376	want_poll_cb = want_poll;	443	want_poll_cb = want_poll;
377	done_poll_cb = done_poll;	444	done_poll_cb = done_poll;
		445
		446	return 0;
378	}	447	}
379		448
380	X_THREAD_PROC (etp_proc);	449	X_THREAD_PROC (etp_proc);
381		450
382	static void etp_start_thread (void)	451	static void etp_start_thread (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, worker *self)	914	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783	{	915	{
784	size_t todo = count;	916	size_t todo = count;
785	dBUF;	917	dBUF;
786		918
787	while (todo > 0)	919	while (todo > 0)
…		…
797	return count;	929	return count;
798	}	930	}
799		931
800	#endif	932	#endif
801		933
802	#if !HAVE_READDIR_R
803	# define readdir_r eio__readdir_r
804
805	static mutex_t readdirlock = X_MUTEX_INIT;
806
807	static int
808	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
809	{
810	EIO_STRUCT_DIRENT *e;
811	int errorno;
812
813	X_LOCK (readdirlock);
814
815	e = readdir (dirp);
816	errorno = errno;
817
818	if (e)
819	{
820	*res = ent;
821	strcpy (ent->d_name, e->d_name);
822	}
823	else
824	*res = 0;
825
826	X_UNLOCK (readdirlock);
827
828	errno = errorno;
829	return e ? 0 : -1;
830	}
831	#endif
832
833	/* sendfile always needs emulation */	934	/* sendfile always needs emulation */
834	static ssize_t	935	static ssize_t
835	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	936	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
836	{	937	{
837	ssize_t res;	938	ssize_t res;
…		…
841		942
842	#if HAVE_SENDFILE	943	#if HAVE_SENDFILE
843	# if __linux	944	# if __linux
844	res = sendfile (ofd, ifd, &offset, count);	945	res = sendfile (ofd, ifd, &offset, count);
845		946
846	# elif __freebsd	947	# elif __FreeBSD__
847	/*	948	/*
848	* 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
849	* wasted on making it similar to other I/O functions.	950	* wasted on making it similar to other I/O functions.
850	*/	951	*/
851	{	952	{
852	off_t sbytes;	953	off_t sbytes;
853	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	954	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
854		955
855	if (res < 0 && sbytes)	956	#if 0 /* according to the manpage, this is correct, but broken behaviour */
856	/* 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)
857	res = sbytes;	977	res = sbytes;
858	}	978	}
859		979
860	# elif __hpux	980	# elif __hpux
861	res = sendfile (ofd, ifd, offset, count, 0, 0);	981	res = sendfile (ofd, ifd, offset, count, 0, 0);
…		…
875	if (res < 0 && sbytes)	995	if (res < 0 && sbytes)
876	res = sbytes;	996	res = sbytes;
877	}	997	}
878		998
879	# 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
880	#else	1010	#else
881	res = -1;	1011	res = -1;
882	errno = ENOSYS;	1012	errno = ENOSYS;
883	#endif	1013	#endif
884		1014
885	if (res < 0	1015	if (res < 0
886	&& (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 */
887	#if __solaris	1022	#if __solaris
888	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1023	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
889	#endif	1024	#endif
890	)	1025	)
891	)	1026	)
…		…
922	}	1057	}
923		1058
924	return res;	1059	return res;
925	}	1060	}
926		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 : a->inode > b->inode ? 1 : 0;
		1067	}
		1068
		1069	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1070
		1071	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1072	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1073
		1074	static void
		1075	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1076	{
		1077	unsigned char bits [9 + sizeof (ino_t) * 8];
		1078	unsigned char *bit = bits;
		1079
		1080	assert (CHAR_BIT == 8);
		1081	assert (sizeof (eio_dirent) * 8 < 256);
		1082	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
		1083	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
		1084
		1085	if (size <= EIO_SORT_FAST)
		1086	return;
		1087
		1088	/* first prepare an array of bits to test in our radix sort */
		1089	/* try to take endianness into account, as well as differences in ino_t sizes */
		1090	/* inode_bits must contain all inodes ORed together */
		1091	/* which is used to skip bits that are 0 everywhere, which is very common */
		1092	{
		1093	ino_t endianness;
		1094	int i, j;
		1095
		1096	/* we store the byte offset of byte n into byte n of "endianness" */
		1097	for (i = 0; i < sizeof (ino_t); ++i)
		1098	((unsigned char *)&endianness)[i] = i;
		1099
		1100	*bit++ = 0;
		1101
		1102	for (i = 0; i < sizeof (ino_t); ++i)
		1103	{
		1104	/* shifting off the byte offsets out of "endianness" */
		1105	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1106	endianness >>= 8;
		1107
		1108	for (j = 0; j < 8; ++j)
		1109	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1110	*bit++ = offs + j;
		1111	}
		1112
		1113	for (j = 0; j < 8; ++j)
		1114	if (score_bits & (1 << j))
		1115	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1116	}
		1117
		1118	/* now actually do the sorting (a variant of MSD radix sort) */
		1119	{
		1120	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1121	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1122	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1123	int stk_idx = 0;
		1124
		1125	base_stk [stk_idx] = dents;
		1126	end_stk [stk_idx] = dents + size;
		1127	bit_stk [stk_idx] = bit - 1;
		1128
		1129	do
		1130	{
		1131	base = base_stk [stk_idx];
		1132	end = end_stk [stk_idx];
		1133	bit = bit_stk [stk_idx];
		1134
		1135	for (;;)
		1136	{
		1137	unsigned char O = *bit >> 3;
		1138	unsigned char M = 1 << (*bit & 7);
		1139
		1140	eio_dirent *a = base;
		1141	eio_dirent *b = end;
		1142
		1143	if (b - a < EIO_SORT_CUTOFF)
		1144	break;
		1145
		1146	/* now bit-partition the array on the bit */
		1147	/* this ugly asymmetric loop seems to perform much better than typical */
		1148	/* partition algos found in the literature */
		1149	do
		1150	if (!(((unsigned char *)a)[O] & M))
		1151	++a;
		1152	else if (!(((unsigned char *)--b)[O] & M))
		1153	{
		1154	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1155	++a;
		1156	}
		1157	while (b > a);
		1158
		1159	/* next bit, or stop, if no bits left in this path */
		1160	if (!*--bit)
		1161	break;
		1162
		1163	base_stk [stk_idx] = a;
		1164	end_stk [stk_idx] = end;
		1165	bit_stk [stk_idx] = bit;
		1166	++stk_idx;
		1167
		1168	end = a;
		1169	}
		1170	}
		1171	while (stk_idx--);
		1172	}
		1173	}
		1174
		1175	static void
		1176	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1177	{
		1178	/* first move the smallest element to the front, to act as a sentinel */
		1179	{
		1180	int i;
		1181	eio_dirent *min = dents;
		1182
		1183	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1184	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1185	if (EIO_DENT_CMP (dents [i], <, *min))
		1186	min = &dents [i];
		1187
		1188	/* swap elements 0 and j (minimum) */
		1189	{
		1190	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1191	}
		1192	}
		1193
		1194	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1195	{
		1196	eio_dirent *i, *j;
		1197
		1198	for (i = dents + 1; i < dents + size; ++i)
		1199	{
		1200	eio_dirent value = *i;
		1201
		1202	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1203	j [1] = j [0];
		1204
		1205	j [1] = value;
		1206	}
		1207	}
		1208	}
		1209
		1210	static void
		1211	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1212	{
		1213	if (size <= 1)
		1214	return; /* our insertion sort relies on size > 0 */
		1215
		1216	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1217	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1218	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1219
		1220	/* use an insertion sort at the end, or for small arrays, */
		1221	/* as insertion sort is more efficient for small partitions */
		1222	eio_dent_insertion_sort (dents, size);
		1223	}
		1224
927	/* read a full directory */	1225	/* read a full directory */
928	static void	1226	static void
929	eio__scandir (eio_req *req, etp_worker *self)	1227	eio__scandir (eio_req *req, etp_worker *self)
930	{	1228	{
931	DIR *dirp;	1229	DIR *dirp;
932	union
933	{
934	EIO_STRUCT_DIRENT d;
935	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
936	} *u;
937	EIO_STRUCT_DIRENT *entp;	1230	EIO_STRUCT_DIRENT *entp;
938	char *name, *names;	1231	char *name, *names;
939	int memlen = 4096;	1232	int namesalloc = 4096;
940	int memofs = 0;	1233	int namesoffs = 0;
		1234	int flags = req->int1;
		1235	eio_dirent *dents = 0;
		1236	int dentalloc = 128;
941	int res = 0;	1237	int dentoffs = 0;
		1238	ino_t inode_bits = 0;
		1239
		1240	req->result = -1;
		1241
		1242	if (!(flags & EIO_READDIR_DENTS))
		1243	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
942		1244
943	X_LOCK (wrklock);	1245	X_LOCK (wrklock);
		1246	/* the corresponding closedir is in ETP_WORKER_CLEAR */
944	self->dirp = dirp = opendir (req->ptr1);	1247	self->dirp = dirp = opendir (req->ptr1);
945	self->dbuf = u = malloc (sizeof (*u));
946	req->flags |= EIO_FLAG_PTR2_FREE;	1248	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1249	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
947	req->ptr2 = names = malloc (memlen);	1250	req->ptr2 = names = malloc (namesalloc);
948	X_UNLOCK (wrklock);	1251	X_UNLOCK (wrklock);
949		1252
950	if (dirp && u && names)	1253	if (dirp && names && (!flags || dents))
951	for (;;)	1254	for (;;)
952	{	1255	{
953	errno = 0;	1256	errno = 0;
954	readdir_r (dirp, &u->d, &entp);	1257	entp = readdir (dirp);
955		1258
956	if (!entp)	1259	if (!entp)
		1260	{
		1261	if (errno)
		1262	break;
		1263
		1264	/* sort etc. */
		1265	req->int1 = flags;
		1266	req->result = dentoffs;
		1267
		1268	if (flags & EIO_READDIR_STAT_ORDER)
		1269	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
		1270	else if (flags & EIO_READDIR_DIRS_FIRST)
		1271	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1272	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1273	else
		1274	{
		1275	/* in this case, all is known, and we just put dirs first and sort them */
		1276	eio_dirent *oth = dents + dentoffs;
		1277	eio_dirent *dir = dents;
		1278
		1279	/* now partition dirs to the front, and non-dirs to the back */
		1280	/* by walking from both sides and swapping if necessary */
		1281	/* also clear score, so it doesn't influence sorting */
		1282	while (oth > dir)
		1283	{
		1284	if (dir->type == EIO_DT_DIR)
		1285	++dir;
		1286	else if ((--oth)->type == EIO_DT_DIR)
		1287	{
		1288	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1289
		1290	++dir;
		1291	}
		1292	}
		1293
		1294	/* now sort the dirs only */
		1295	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1296	}
		1297
957	break;	1298	break;
		1299	}
958		1300
		1301	/* now add the entry to our list(s) */
959	name = entp->d_name;	1302	name = entp->d_name;
960		1303
		1304	/* skip . and .. entries */
961	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1305	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
962	{	1306	{
963	int len = strlen (name) + 1;	1307	int len = D_NAMLEN (entp) + 1;
964		1308
965	res++;	1309	while (expect_false (namesoffs + len > namesalloc))
966
967	while (memofs + len > memlen)
968	{	1310	{
969	memlen *= 2;	1311	namesalloc *= 2;
970	X_LOCK (wrklock);	1312	X_LOCK (wrklock);
971	req->ptr2 = names = realloc (names, memlen);	1313	req->ptr2 = names = realloc (names, namesalloc);
972	X_UNLOCK (wrklock);	1314	X_UNLOCK (wrklock);
973		1315
974	if (!names)	1316	if (!names)
975	break;	1317	break;
976	}	1318	}
977		1319
978	memcpy (names + memofs, name, len);	1320	memcpy (names + namesoffs, name, len);
		1321
		1322	if (dents)
		1323	{
		1324	struct eio_dirent *ent;
		1325
		1326	if (expect_false (dentoffs == dentalloc))
		1327	{
		1328	dentalloc *= 2;
		1329	X_LOCK (wrklock);
		1330	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1331	X_UNLOCK (wrklock);
		1332
		1333	if (!dents)
		1334	break;
		1335	}
		1336
		1337	ent = dents + dentoffs;
		1338
		1339	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1340	ent->namelen = len - 1;
		1341	ent->inode = D_INO (entp);
		1342
		1343	inode_bits |= ent->inode;
		1344
		1345	switch (D_TYPE (entp))
		1346	{
		1347	default:
		1348	ent->type = EIO_DT_UNKNOWN;
		1349	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1350	break;
		1351
		1352	#ifdef DT_FIFO
		1353	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1354	#endif
		1355	#ifdef DT_CHR
		1356	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1357	#endif
		1358	#ifdef DT_MPC
		1359	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1360	#endif
		1361	#ifdef DT_DIR
		1362	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1363	#endif
		1364	#ifdef DT_NAM
		1365	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1366	#endif
		1367	#ifdef DT_BLK
		1368	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1369	#endif
		1370	#ifdef DT_MPB
		1371	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1372	#endif
		1373	#ifdef DT_REG
		1374	case DT_REG: ent->type = EIO_DT_REG; break;
		1375	#endif
		1376	#ifdef DT_NWK
		1377	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1378	#endif
		1379	#ifdef DT_CMP
		1380	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1381	#endif
		1382	#ifdef DT_LNK
		1383	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1384	#endif
		1385	#ifdef DT_SOCK
		1386	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1387	#endif
		1388	#ifdef DT_DOOR
		1389	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1390	#endif
		1391	#ifdef DT_WHT
		1392	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1393	#endif
		1394	}
		1395
		1396	ent->score = 7;
		1397
		1398	if (flags & EIO_READDIR_DIRS_FIRST)
		1399	{
		1400	if (ent->type == EIO_DT_UNKNOWN)
		1401	{
		1402	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1403	ent->score = 1;
		1404	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1405	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1406	}
		1407	else if (ent->type == EIO_DT_DIR)
		1408	ent->score = 0;
		1409	}
		1410	}
		1411
979	memofs += len;	1412	namesoffs += len;
		1413	++dentoffs;
		1414	}
		1415
		1416	if (EIO_CANCELLED (req))
		1417	{
		1418	errno = ECANCELED;
		1419	break;
980	}	1420	}
981	}	1421	}
		1422	}
982		1423
983	if (errno)	1424	#ifdef PAGESIZE
984	res = -1;	1425	# define eio_pagesize() PAGESIZE
		1426	#else
		1427	static intptr_t
		1428	eio_pagesize (void)
		1429	{
		1430	static intptr_t page;
		1431
		1432	if (!page)
		1433	page = sysconf (_SC_PAGESIZE);
		1434
		1435	return page;
		1436	}
		1437	#endif
		1438
		1439	static void
		1440	eio_page_align (void **addr, size_t *length)
		1441	{
		1442	intptr_t mask = eio_pagesize () - 1;
		1443
		1444	/* round down addr */
		1445	intptr_t adj = mask & (intptr_t)*addr;
		1446
		1447	*addr = (void *)((intptr_t)*addr - adj);
		1448	*length += adj;
		1449
		1450	/* round up length */
		1451	*length = (*length + mask) & ~mask;
		1452	}
		1453
		1454	#if !_POSIX_MEMLOCK
		1455	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1456	#else
		1457
		1458	static int
		1459	eio__mlockall (int flags)
		1460	{
		1461	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1462	extern int mallopt (int, int);
		1463	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1464	#endif
		1465
		1466	if (EIO_MCL_CURRENT != MCL_CURRENT
		1467	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1468	{
		1469	flags = 0
		1470	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1471	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1472	}
		1473
		1474	return mlockall (flags);
		1475	}
		1476	#endif
		1477
		1478	#if !_POSIX_MEMLOCK_RANGE
		1479	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1480	#else
		1481
		1482	static int
		1483	eio__mlock (void *addr, size_t length)
		1484	{
		1485	eio_page_align (&addr, &length);
		1486
		1487	return mlock (addr, length);
		1488	}
		1489
		1490	#endif
		1491
		1492	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1493	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1494	#else
		1495
		1496	int
		1497	eio__msync (void *mem, size_t len, int flags)
		1498	{
		1499	eio_page_align (&mem, &len);
		1500
		1501	if (EIO_MS_ASYNC != MS_SYNC
		1502	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1503	|| EIO_MS_SYNC != MS_SYNC)
		1504	{
		1505	flags = 0
		1506	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1507	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1508	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1509	}
		1510
		1511	return msync (mem, len, flags);
		1512	}
		1513
		1514	#endif
		1515
		1516	int
		1517	eio__mtouch (void *mem, size_t len, int flags)
		1518	{
		1519	eio_page_align (&mem, &len);
		1520
985		1521	{
986	req->result = res;	1522	intptr_t addr = (intptr_t)mem;
		1523	intptr_t end = addr + len;
		1524	intptr_t page = eio_pagesize ();
		1525
		1526	if (addr < end)
		1527	if (flags & EIO_MT_MODIFY) /* modify */
		1528	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1529	else
		1530	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1531	}
		1532
		1533	return 0;
987	}	1534	}
988		1535
989	/*****************************************************************************/	1536	/*****************************************************************************/
990		1537
991	#define ALLOC(len) \	1538	#define ALLOC(len) \
…		…
1023	if (req)	1570	if (req)
1024	break;	1571	break;
1025		1572
1026	++idle;	1573	++idle;
1027		1574
1028	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1575	ts.tv_sec = time (0) + idle_timeout;
1029	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1576	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1030	{	1577	{
1031	if (idle > max_idle)	1578	if (idle > max_idle)
1032	{	1579	{
1033	--idle;	1580	--idle;
…		…
1047		1594
1048	--nready;	1595	--nready;
1049		1596
1050	X_UNLOCK (reqlock);	1597	X_UNLOCK (reqlock);
1051		1598
1052	errno = 0; /* strictly unnecessary */	1599	if (req->type < 0)
		1600	goto quit;
1053		1601
1054	if (!EIO_CANCELLED (req))	1602	if (!EIO_CANCELLED (req))
1055	switch (req->type)	1603	ETP_EXECUTE (self, req);
1056	{
1057	case EIO_READ: ALLOC (req->size);
1058	req->result = req->offs >= 0
1059	? pread (req->int1, req->ptr2, req->size, req->offs)
1060	: read (req->int1, req->ptr2, req->size); break;
1061	case EIO_WRITE: req->result = req->offs >= 0
1062	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1063	: write (req->int1, req->ptr2, req->size); break;
1064
1065	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1066	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1067
1068	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1069	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1070	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1071	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1072	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1073	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1074
1075	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1076	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1077	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1078	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1079	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1080	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1081
1082	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1083	case EIO_CLOSE: req->result = close (req->int1); break;
1084	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1085	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1086	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1087	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1088	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1089	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1090	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1091	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1092
1093	case EIO_READLINK: ALLOC (NAME_MAX);
1094	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1095
1096	case EIO_SYNC: req->result = 0; sync (); break;
1097	case EIO_FSYNC: req->result = fsync (req->int1); break;
1098	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1099
1100	case EIO_READDIR: eio__scandir (req, self); break;
1101
1102	case EIO_BUSY:
1103	#ifdef _WIN32
1104	Sleep (req->nv1 * 1000.);
1105	#else
1106	{
1107	struct timeval tv;
1108
1109	tv.tv_sec = req->nv1;
1110	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1111
1112	req->result = select (0, 0, 0, 0, &tv);
1113	}
1114	#endif
1115	break;
1116
1117	case EIO_UTIME:
1118	case EIO_FUTIME:
1119	{
1120	struct timeval tv[2];
1121	struct timeval *times;
1122
1123	if (req->nv1 != -1. || req->nv2 != -1.)
1124	{
1125	tv[0].tv_sec = req->nv1;
1126	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1127	tv[1].tv_sec = req->nv2;
1128	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1129
1130	times = tv;
1131	}
1132	else
1133	times = 0;
1134
1135
1136	req->result = req->type == EIO_FUTIME
1137	? futimes (req->int1, times)
1138	: utimes (req->ptr1, times);
1139	}
1140
1141	case EIO_GROUP:
1142	case EIO_NOP:
1143	req->result = 0;
1144	break;
1145
1146	case -1:
1147	goto quit;
1148
1149	default:
1150	req->result = -1;
1151	break;
1152	}
1153
1154	req->errorno = errno;
1155		1604
1156	X_LOCK (reslock);	1605	X_LOCK (reslock);
1157		1606
1158	++npending;	1607	++npending;
1159		1608
…		…
1176		1625
1177	/*****************************************************************************/	1626	/*****************************************************************************/
1178		1627
1179	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1628	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1180	{	1629	{
1181	etp_init (want_poll, done_poll);	1630	return etp_init (want_poll, done_poll);
1182	}	1631	}
1183		1632
1184	static void eio_api_destroy (eio_req *req)	1633	static void eio_api_destroy (eio_req *req)
1185	{	1634	{
1186	free (req);	1635	free (req);
…		…
1208	{ \	1657	{ \
1209	eio_api_destroy (req); \	1658	eio_api_destroy (req); \
1210	return 0; \	1659	return 0; \
1211	}	1660	}
1212		1661
		1662	static void eio_execute (etp_worker *self, eio_req *req)
		1663	{
		1664	switch (req->type)
		1665	{
		1666	case EIO_READ: ALLOC (req->size);
		1667	req->result = req->offs >= 0
		1668	? pread (req->int1, req->ptr2, req->size, req->offs)
		1669	: read (req->int1, req->ptr2, req->size); break;
		1670	case EIO_WRITE: req->result = req->offs >= 0
		1671	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1672	: write (req->int1, req->ptr2, req->size); break;
		1673
		1674	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1675	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1676
		1677	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1678	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1679	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1680	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1681	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1682	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1683
		1684	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1685	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1686	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1687	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1688
		1689	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1690	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1691	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1692	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1693	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1694	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1695
		1696	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1697	case EIO_CLOSE: req->result = close (req->int1); break;
		1698	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1699	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1700	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1701	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1702	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1703	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1704	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1705	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		1706
		1707	case EIO_READLINK: ALLOC (PATH_MAX);
		1708	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
		1709
		1710	case EIO_SYNC: req->result = 0; sync (); break;
		1711	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1712	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1713	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1714	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1715	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1716	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		1717	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1718
		1719	case EIO_READDIR: eio__scandir (req, self); break;
		1720
		1721	case EIO_BUSY:
		1722	#ifdef _WIN32
		1723	Sleep (req->nv1 * 1e3);
		1724	#else
		1725	{
		1726	struct timeval tv;
		1727
		1728	tv.tv_sec = req->nv1;
		1729	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
		1730
		1731	req->result = select (0, 0, 0, 0, &tv);
		1732	}
		1733	#endif
		1734	break;
		1735
		1736	case EIO_UTIME:
		1737	case EIO_FUTIME:
		1738	{
		1739	struct timeval tv[2];
		1740	struct timeval *times;
		1741
		1742	if (req->nv1 != -1. || req->nv2 != -1.)
		1743	{
		1744	tv[0].tv_sec = req->nv1;
		1745	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1746	tv[1].tv_sec = req->nv2;
		1747	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1748
		1749	times = tv;
		1750	}
		1751	else
		1752	times = 0;
		1753
		1754	req->result = req->type == EIO_FUTIME
		1755	? futimes (req->int1, times)
		1756	: utimes (req->ptr1, times);
		1757	}
		1758	break;
		1759
		1760	case EIO_GROUP:
		1761	abort (); /* handled in eio_request */
		1762
		1763	case EIO_NOP:
		1764	req->result = 0;
		1765	break;
		1766
		1767	case EIO_CUSTOM:
		1768	((void (*)(eio_req *))req->feed) (req);
		1769	break;
		1770
		1771	default:
		1772	errno = ENOSYS;
		1773	req->result = -1;
		1774	break;
		1775	}
		1776
		1777	req->errorno = errno;
		1778	}
		1779
1213	#ifndef EIO_NO_WRAPPERS	1780	#ifndef EIO_NO_WRAPPERS
1214		1781
1215	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1782	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1216	{	1783	{
1217	REQ (EIO_NOP); SEND;	1784	REQ (EIO_NOP); SEND;
…		…
1230	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1797	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1231	{	1798	{
1232	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1799	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1233	}	1800	}
1234		1801
		1802	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1803	{
		1804	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1805	}
		1806
		1807	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1808	{
		1809	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1810	}
		1811
		1812	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		1813	{
		1814	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		1815	}
		1816
		1817	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		1818	{
		1819	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		1820	}
		1821
		1822	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1823	{
		1824	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1825	}
		1826
1235	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1827	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1236	{	1828	{
1237	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1829	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1238	}	1830	}
1239		1831
…		…
1258	}	1850	}
1259		1851
1260	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	1852	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1261	{	1853	{
1262	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1854	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1855	}
		1856
		1857	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		1858	{
		1859	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1263	}	1860	}
1264		1861
1265	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	1862	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1266	{	1863	{
1267	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1864	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1341	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	1938	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1342	{	1939	{
1343	return eio__1path (EIO_LSTAT, path, pri, cb, data);	1940	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1344	}	1941	}
1345		1942
		1943	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		1944	{
		1945	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		1946	}
		1947
1346	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	1948	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1347	{	1949	{
1348	return eio__1path (EIO_UNLINK, path, pri, cb, data);	1950	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1349	}	1951	}
1350		1952
1351	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	1953	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1352	{	1954	{
1353	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1955	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1354	}	1956	}
1355		1957
1356	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	1958	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1357	{	1959	{
1358	return eio__1path (EIO_READDIR, path, pri, cb, data);	1960	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1359	}	1961	}
1360		1962
1361	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1963	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1362	{	1964	{
1363	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1965	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1364	}	1966	}
1365		1967
1366	static eio_req *	1968	static eio_req *
1367	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1969	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1368	{	1970	{
…		…
1392	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1994	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1393	{	1995	{
1394	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1996	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1395	}	1997	}
1396		1998
		1999	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
		2000	{
		2001	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
		2002	}
		2003
1397	#endif	2004	#endif
1398		2005
1399	eio_req *eio_grp (eio_cb cb, void *data)	2006	eio_req *eio_grp (eio_cb cb, void *data)
1400	{	2007	{
1401	const int pri = EIO_PRI_MAX;	2008	const int pri = EIO_PRI_MAX;
…		…
1427		2034
1428	void eio_grp_add (eio_req *grp, eio_req *req)	2035	void eio_grp_add (eio_req *grp, eio_req *req)
1429	{	2036	{
1430	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2037	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1431		2038
		2039	grp->flags |= EIO_FLAG_GROUPADD;
		2040
1432	++grp->size;	2041	++grp->size;
1433	req->grp = grp;	2042	req->grp = grp;
1434		2043
1435	req->grp_prev = 0;	2044	req->grp_prev = 0;
1436	req->grp_next = grp->grp_first;	2045	req->grp_next = grp->grp_first;
…		…
1445	/* misc garbage */	2054	/* misc garbage */
1446		2055
1447	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2056	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1448	{	2057	{
1449	etp_worker wrk;	2058	etp_worker wrk;
		2059	ssize_t ret;
1450		2060
1451	wrk.dbuf = 0;	2061	wrk.dbuf = 0;
1452		2062
1453	eio__sendfile (ofd, ifd, offset, count, &wrk);	2063	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1454		2064
1455	if (wrk.dbuf)	2065	if (wrk.dbuf)
1456	free (wrk.dbuf);	2066	free (wrk.dbuf);
1457	}
1458		2067
		2068	return ret;
		2069	}
		2070

Diff Legend

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