[ViewVC] Diff of: 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 1.53 by root, Sat Jan 9 10:03:09 2010 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 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>
		79	# include <sys/mman.h>
75	# include <unistd.h>	80	# include <unistd.h>
76	# include <utime.h>	81	# include <utime.h>
77	# include <signal.h>	82	# include <signal.h>
78	# include <dirent.h>	83	# include <dirent.h>
79		84
		85	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		86	# if __FreeBSD__ \|\| defined __NetBSD__ \|\| defined __OpenBSD__
		87	# define _DIRENT_HAVE_D_TYPE /* sigh */
		88	# define D_INO(de) (de)->d_fileno
		89	# define D_NAMLEN(de) (de)->d_namlen
		90	# elif __linux \|\| defined d_ino \|\| _XOPEN_SOURCE >= 600
		91	# define D_INO(de) (de)->d_ino
		92	# endif
		93
		94	#ifdef _D_EXACT_NAMLEN
		95	# undef D_NAMLEN
		96	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		97	#endif
		98
		99	# ifdef _DIRENT_HAVE_D_TYPE
		100	# define D_TYPE(de) (de)->d_type
		101	# endif
		102
80	# ifndef EIO_STRUCT_DIRENT	103	# ifndef EIO_STRUCT_DIRENT
81	# define EIO_STRUCT_DIRENT struct dirent	104	# define EIO_STRUCT_DIRENT struct dirent
82	# endif	105	# endif
83		106
84	#endif	107	#endif
85		108
86	#if HAVE_SENDFILE	109	#if HAVE_SENDFILE
87	# if __linux	110	# if __linux
88	# include <sys/sendfile.h>	111	# include <sys/sendfile.h>
89	# elif __freebsd	112	# elif __FreeBSD__ \|\| defined __APPLE__
90	# include <sys/socket.h>	113	# include <sys/socket.h>
91	# include <sys/uio.h>	114	# include <sys/uio.h>
92	# elif __hpux	115	# elif __hpux
93	# include <sys/socket.h>	116	# include <sys/socket.h>
94	# elif __solaris /* not yet */	117	# elif __solaris
95	# include <sys/sendfile.h>	118	# include <sys/sendfile.h>
96	# else	119	# else
97	# error sendfile support requested but not available	120	# error sendfile support requested but not available
98	# endif	121	# endif
		122	#endif
		123
		124	#ifndef D_TYPE
		125	# define D_TYPE(de) 0
		126	#endif
		127	#ifndef D_INO
		128	# define D_INO(de) 0
		129	#endif
		130	#ifndef D_NAMLEN
		131	# define D_NAMLEN(de) strlen ((de)->d_name)
99	#endif	132	#endif
100		133
101	/* number of seconds after which an idle threads exit */	134	/* number of seconds after which an idle threads exit */
102	#define IDLE_TIMEOUT 10	135	#define IDLE_TIMEOUT 10
103		136
…		…
120		153
121	#define EIO_TICKS ((1000000 + 1023) >> 10)	154	#define EIO_TICKS ((1000000 + 1023) >> 10)
122		155
123	/*****************************************************************************/	156	/*****************************************************************************/
124		157
		158	#if __GNUC__ >= 3
		159	# define expect(expr,value) __builtin_expect ((expr),(value))
		160	#else
		161	# define expect(expr,value) (expr)
		162	#endif
		163
		164	#define expect_false(expr) expect ((expr) != 0, 0)
		165	#define expect_true(expr) expect ((expr) != 0, 1)
		166
		167	/*****************************************************************************/
		168
125	#define ETP_PRI_MIN EIO_PRI_MIN	169	#define ETP_PRI_MIN EIO_PRI_MIN
126	#define ETP_PRI_MAX EIO_PRI_MAX	170	#define ETP_PRI_MAX EIO_PRI_MAX
		171
		172	struct etp_worker;
127		173
128	#define ETP_REQ eio_req	174	#define ETP_REQ eio_req
129	#define ETP_DESTROY(req) eio_destroy (req)	175	#define ETP_DESTROY(req) eio_destroy (req)
130	static int eio_finish (eio_req *req);	176	static int eio_finish (eio_req *req);
131	#define ETP_FINISH(req) eio_finish (req)	177	#define ETP_FINISH(req) eio_finish (req)
		178	static void eio_execute (struct etp_worker self, eio_req req);
		179	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
132		180
133	#define ETP_WORKER_CLEAR(req) \	181	#define ETP_WORKER_CLEAR(req) \
134	if (wrk->dbuf) \	182	if (wrk->dbuf) \
135	{ \	183	{ \
136	free (wrk->dbuf); \	184	free (wrk->dbuf); \
…		…
140	if (wrk->dirp) \	188	if (wrk->dirp) \
141	{ \	189	{ \
142	closedir (wrk->dirp); \	190	closedir (wrk->dirp); \
143	wrk->dirp = 0; \	191	wrk->dirp = 0; \
144	}	192	}
		193
145	#define ETP_WORKER_COMMON \	194	#define ETP_WORKER_COMMON \
146	void *dbuf; \	195	void *dbuf; \
147	DIR *dirp;	196	DIR *dirp;
148		197
149	/*****************************************************************************/	198	/*****************************************************************************/
150		199
151	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	200	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
152		201
153	/* calculcate time difference in ~1/EIO_TICKS of a second */	202	/* calculate time difference in ~1/EIO_TICKS of a second */
154	static int tvdiff (struct timeval tv1, struct timeval tv2)	203	static int tvdiff (struct timeval tv1, struct timeval tv2)
155	{	204	{
156	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	205	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
157	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	206	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
158	}	207	}
…		…
172		221
173	static mutex_t wrklock = X_MUTEX_INIT;	222	static mutex_t wrklock = X_MUTEX_INIT;
174	static mutex_t reslock = X_MUTEX_INIT;	223	static mutex_t reslock = X_MUTEX_INIT;
175	static mutex_t reqlock = X_MUTEX_INIT;	224	static mutex_t reqlock = X_MUTEX_INIT;
176	static cond_t reqwait = X_COND_INIT;	225	static cond_t reqwait = X_COND_INIT;
		226
		227	#if !HAVE_PREADWRITE
		228	/*
		229	* make our pread/pwrite emulation safe against themselves, but not against
		230	* normal read/write by using a mutex. slows down execution a lot,
		231	* but that's your problem, not mine.
		232	*/
		233	static mutex_t preadwritelock = X_MUTEX_INIT;
		234	#endif
177		235
178	typedef struct etp_worker	236	typedef struct etp_worker
179	{	237	{
180	/* locked by wrklock */	238	/* locked by wrklock */
181	struct etp_worker prev, next;	239	struct etp_worker prev, next;
…		…
310	X_LOCK (reqlock);	368	X_LOCK (reqlock);
311	X_LOCK (reslock);	369	X_LOCK (reslock);
312	#if !HAVE_PREADWRITE	370	#if !HAVE_PREADWRITE
313	X_LOCK (preadwritelock);	371	X_LOCK (preadwritelock);
314	#endif	372	#endif
315	#if !HAVE_READDIR_R
316	X_LOCK (readdirlock);
317	#endif
318	}	373	}
319		374
320	static void etp_atfork_parent (void)	375	static void etp_atfork_parent (void)
321	{	376	{
322	#if !HAVE_READDIR_R
323	X_UNLOCK (readdirlock);
324	#endif
325	#if !HAVE_PREADWRITE	377	#if !HAVE_PREADWRITE
326	X_UNLOCK (preadwritelock);	378	X_UNLOCK (preadwritelock);
327	#endif	379	#endif
328	X_UNLOCK (reslock);	380	X_UNLOCK (reslock);
329	X_UNLOCK (reqlock);	381	X_UNLOCK (reqlock);
…		…
332		384
333	static void etp_atfork_child (void)	385	static void etp_atfork_child (void)
334	{	386	{
335	ETP_REQ *prv;	387	ETP_REQ *prv;
336		388
337	while (prv = reqq_shift (&req_queue))	389	while ((prv = reqq_shift (&req_queue)))
338	ETP_DESTROY (prv);	390	ETP_DESTROY (prv);
339		391
340	while (prv = reqq_shift (&res_queue))	392	while ((prv = reqq_shift (&res_queue)))
341	ETP_DESTROY (prv);	393	ETP_DESTROY (prv);
342		394
343	while (wrk_first.next != &wrk_first)	395	while (wrk_first.next != &wrk_first)
344	{	396	{
345	etp_worker *wrk = wrk_first.next;	397	etp_worker *wrk = wrk_first.next;
…		…
373		425
374	pthread_once (&doinit, etp_once_init);	426	pthread_once (&doinit, etp_once_init);
375		427
376	want_poll_cb = want_poll;	428	want_poll_cb = want_poll;
377	done_poll_cb = done_poll;	429	done_poll_cb = done_poll;
		430
		431	return 0;
378	}	432	}
379		433
380	X_THREAD_PROC (etp_proc);	434	X_THREAD_PROC (etp_proc);
381		435
382	static void etp_start_thread (void)	436	static void etp_start_thread (void)
…		…
402	X_UNLOCK (wrklock);	456	X_UNLOCK (wrklock);
403	}	457	}
404		458
405	static void etp_maybe_start_thread (void)	459	static void etp_maybe_start_thread (void)
406	{	460	{
407	if (etp_nthreads () >= wanted)	461	if (expect_true (etp_nthreads () >= wanted))
408	return;	462	return;
409		463
410	/* todo: maybe use idle here, but might be less exact */	464	/* todo: maybe use idle here, but might be less exact */
411	if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())	465	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
412	return;	466	return;
413		467
414	etp_start_thread ();	468	etp_start_thread ();
415	}	469	}
416		470
…		…
469		523
470	X_LOCK (reqlock);	524	X_LOCK (reqlock);
471	--nreqs;	525	--nreqs;
472	X_UNLOCK (reqlock);	526	X_UNLOCK (reqlock);
473		527
474	if (req->type == EIO_GROUP && req->size)	528	if (expect_false (req->type == EIO_GROUP && req->size))
475	{	529	{
476	req->int1 = 1; /* mark request as delayed */	530	req->int1 = 1; /* mark request as delayed */
477	continue;	531	continue;
478	}	532	}
479	else	533	else
480	{	534	{
481	int res = ETP_FINISH (req);	535	int res = ETP_FINISH (req);
482	if (res)	536	if (expect_false (res))
483	return res;	537	return res;
484	}	538	}
485		539
486	if (maxreqs && !--maxreqs)	540	if (expect_false (maxreqs && !--maxreqs))
487	break;	541	break;
488		542
489	if (maxtime)	543	if (maxtime)
490	{	544	{
491	gettimeofday (&tv_now, 0);	545	gettimeofday (&tv_now, 0);
…		…
510		564
511	static void etp_submit (ETP_REQ *req)	565	static void etp_submit (ETP_REQ *req)
512	{	566	{
513	req->pri -= ETP_PRI_MIN;	567	req->pri -= ETP_PRI_MIN;
514		568
515	if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	569	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;	570	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
517		571
		572	if (expect_false (req->type == EIO_GROUP))
		573	{
		574	/* I hope this is worth it :/ */
518	X_LOCK (reqlock);	575	X_LOCK (reqlock);
519	++nreqs;	576	++nreqs;
		577	X_UNLOCK (reqlock);
		578
		579	X_LOCK (reslock);
		580
		581	++npending;
		582
		583	if (!reqq_push (&res_queue, req) && want_poll_cb)
		584	want_poll_cb ();
		585
		586	X_UNLOCK (reslock);
		587	}
		588	else
		589	{
		590	X_LOCK (reqlock);
		591	++nreqs;
520	++nready;	592	++nready;
521	reqq_push (&req_queue, req);	593	reqq_push (&req_queue, req);
522	X_COND_SIGNAL (reqwait);	594	X_COND_SIGNAL (reqwait);
523	X_UNLOCK (reqlock);	595	X_UNLOCK (reqlock);
524		596
525	etp_maybe_start_thread ();	597	etp_maybe_start_thread ();
		598	}
526	}	599	}
527		600
528	static void etp_set_max_poll_time (double nseconds)	601	static void etp_set_max_poll_time (double nseconds)
529	{	602	{
530	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	603	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
531	max_poll_time = nseconds;	604	max_poll_time = nseconds * EIO_TICKS;
532	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	605	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
533	}	606	}
534		607
535	static void etp_set_max_poll_reqs (unsigned int maxreqs)	608	static void etp_set_max_poll_reqs (unsigned int maxreqs)
536	{	609	{
…		…
565		638
566	static void grp_try_feed (eio_req *grp)	639	static void grp_try_feed (eio_req *grp)
567	{	640	{
568	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	641	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569	{	642	{
570	int old_len = grp->size;	643	grp->flags &= ~EIO_FLAG_GROUPADD;
571		644
572	EIO_FEED (grp);	645	EIO_FEED (grp);
573		646
574	/* stop if no progress has been made */	647	/* stop if no progress has been made */
575	if (old_len == grp->size)	648	if (!(grp->flags & EIO_FLAG_GROUPADD))
576	{	649	{
577	grp->feed = 0;	650	grp->feed = 0;
578	break;	651	break;
579	}	652	}
580	}	653	}
…		…
697		770
698	/*****************************************************************************/	771	/*****************************************************************************/
699	/* work around various missing functions */	772	/* work around various missing functions */
700		773
701	#if !HAVE_PREADWRITE	774	#if !HAVE_PREADWRITE
		775	# undef pread
		776	# undef pwrite
702	# define pread eio__pread	777	# define pread eio__pread
703	# define pwrite eio__pwrite	778	# 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		779
712	static ssize_t	780	static ssize_t
713	eio__pread (int fd, void *buf, size_t count, off_t offset)	781	eio__pread (int fd, void *buf, size_t count, off_t offset)
714	{	782	{
715	ssize_t res;	783	ssize_t res;
…		…
733		801
734	X_LOCK (preadwritelock);	802	X_LOCK (preadwritelock);
735	ooffset = lseek (fd, 0, SEEK_CUR);	803	ooffset = lseek (fd, 0, SEEK_CUR);
736	lseek (fd, offset, SEEK_SET);	804	lseek (fd, offset, SEEK_SET);
737	res = write (fd, buf, count);	805	res = write (fd, buf, count);
738	lseek (fd, offset, SEEK_SET);	806	lseek (fd, ooffset, SEEK_SET);
739	X_UNLOCK (preadwritelock);	807	X_UNLOCK (preadwritelock);
740		808
741	return res;	809	return res;
742	}	810	}
743	#endif	811	#endif
744		812
745	#ifndef HAVE_FUTIMES	813	#ifndef HAVE_FUTIMES
746		814
		815	# undef utimes
		816	# undef futimes
747	# define utimes(path,times) eio__utimes (path, times)	817	# define utimes(path,times) eio__utimes (path, times)
748	# define futimes(fd,times) eio__futimes (fd, times)	818	# define futimes(fd,times) eio__futimes (fd, times)
749		819
750	static int	820	static int
751	eio__utimes (const char *filename, const struct timeval times[2])	821	eio__utimes (const char *filename, const struct timeval times[2])
…		…
770	}	840	}
771		841
772	#endif	842	#endif
773		843
774	#if !HAVE_FDATASYNC	844	#if !HAVE_FDATASYNC
		845	# undef fdatasync
775	# define fdatasync fsync	846	# define fdatasync(fd) fsync (fd)
776	#endif	847	#endif
		848
		849	/* sync_file_range always needs emulation */
		850	int
		851	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		852	{
		853	#if HAVE_SYNC_FILE_RANGE
		854	int res;
		855
		856	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		857	\|\| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		858	\|\| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		859	{
		860	flags = 0
		861	\| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		862	\| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		863	\| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		864	}
		865
		866	res = sync_file_range (fd, offset, nbytes, flags);
		867
		868	if (!res \|\| errno != ENOSYS)
		869	return res;
		870	#endif
		871
		872	/* even though we could play tricks with the flags, it's better to always
		873	* call fdatasync, as that matches the expectation of its users best */
		874	return fdatasync (fd);
		875	}
777		876
778	#if !HAVE_READAHEAD	877	#if !HAVE_READAHEAD
		878	# undef readahead
779	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	879	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780		880
781	static ssize_t	881	static ssize_t
782	eio__readahead (int fd, off_t offset, size_t count, worker *self)	882	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783	{	883	{
784	size_t todo = count;	884	size_t todo = count;
785	dBUF;	885	dBUF;
786		886
787	while (todo > 0)	887	while (todo > 0)
…		…
797	return count;	897	return count;
798	}	898	}
799		899
800	#endif	900	#endif
801		901
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 */	902	/* sendfile always needs emulation */
834	static ssize_t	903	static ssize_t
835	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	904	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
836	{	905	{
837	ssize_t res;	906	ssize_t res;
…		…
841		910
842	#if HAVE_SENDFILE	911	#if HAVE_SENDFILE
843	# if __linux	912	# if __linux
844	res = sendfile (ofd, ifd, &offset, count);	913	res = sendfile (ofd, ifd, &offset, count);
845		914
846	# elif __freebsd	915	# elif __FreeBSD__
847	/*	916	/*
848	* Of course, the freebsd sendfile is a dire hack with no thoughts	917	* Of course, the freebsd sendfile is a dire hack with no thoughts
849	* wasted on making it similar to other I/O functions.	918	* wasted on making it similar to other I/O functions.
850	*/	919	*/
851	{	920	{
852	off_t sbytes;	921	off_t sbytes;
853	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	922	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
854		923
855	if (res < 0 && sbytes)	924	#if 0 /* according to the manpage, this is correct, but broken behaviour */
856	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	925	/* freebsd' sendfile will return 0 on success */
		926	/* freebsd 8 documents it as only setting sbytes on EINTR and EAGAIN, but /
		927	/* not on e.g. EIO or EPIPE - sounds broken */
		928	if ((res < 0 && (errno == EAGAIN \|\| errno == EINTR) && sbytes) \|\| res == 0)
		929	res = sbytes;
		930	#endif
		931
		932	/* according to source inspection, this is correct, and useful behaviour */
		933	if (sbytes)
		934	res = sbytes;
		935	}
		936
		937	# elif defined (__APPLE__)
		938
		939	{
		940	off_t sbytes = count;
		941	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		942
		943	/* according to the manpage, sbytes is always valid */
		944	if (sbytes)
857	res = sbytes;	945	res = sbytes;
858	}	946	}
859		947
860	# elif __hpux	948	# elif __hpux
861	res = sendfile (ofd, ifd, offset, count, 0, 0);	949	res = sendfile (ofd, ifd, offset, count, 0, 0);
…		…
875	if (res < 0 && sbytes)	963	if (res < 0 && sbytes)
876	res = sbytes;	964	res = sbytes;
877	}	965	}
878		966
879	# endif	967	# endif
		968
		969	#elif defined (_WIN32)
		970
		971	/* does not work, just for documentation of what would need to be done */
		972	{
		973	HANDLE h = TO_SOCKET (ifd);
		974	SetFilePointer (h, offset, 0, FILE_BEGIN);
		975	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		976	}
		977
880	#else	978	#else
881	res = -1;	979	res = -1;
882	errno = ENOSYS;	980	errno = ENOSYS;
883	#endif	981	#endif
884		982
885	if (res < 0	983	if (res < 0
886	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK	984	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK
		985	/* BSDs */
		986	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		987	\|\| errno == ENOTSUP
		988	#endif
		989	\|\| errno == EOPNOTSUPP /* BSDs */
887	#if __solaris	990	#if __solaris
888	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE	991	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE
889	#endif	992	#endif
890	)	993	)
891	)	994	)
…		…
922	}	1025	}
923		1026
924	return res;	1027	return res;
925	}	1028	}
926		1029
		1030	static signed char
		1031	eio_dent_cmp (const eio_dirent a, const eio_dirent b)
		1032	{
		1033	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1034	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
		1035	}
		1036
		1037	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1038
		1039	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1040	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1041
		1042	static void
		1043	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1044	{
		1045	unsigned char bits [9 + sizeof (ino_t) * 8];
		1046	unsigned char *bit = bits;
		1047
		1048	assert (CHAR_BIT == 8);
		1049	assert (sizeof (eio_dirent) * 8 < 256);
		1050	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
		1051	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
		1052
		1053	if (size <= EIO_SORT_FAST)
		1054	return;
		1055
		1056	/* first prepare an array of bits to test in our radix sort */
		1057	/* try to take endianness into account, as well as differences in ino_t sizes */
		1058	/* inode_bits must contain all inodes ORed together */
		1059	/* which is used to skip bits that are 0 everywhere, which is very common */
		1060	{
		1061	ino_t endianness;
		1062	int i, j;
		1063
		1064	/* we store the byte offset of byte n into byte n of "endianness" */
		1065	for (i = 0; i < sizeof (ino_t); ++i)
		1066	((unsigned char *)&endianness)[i] = i;
		1067
		1068	*bit++ = 0;
		1069
		1070	for (i = 0; i < sizeof (ino_t); ++i)
		1071	{
		1072	/* shifting off the byte offsets out of "endianness" */
		1073	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1074	endianness >>= 8;
		1075
		1076	for (j = 0; j < 8; ++j)
		1077	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1078	*bit++ = offs + j;
		1079	}
		1080
		1081	for (j = 0; j < 8; ++j)
		1082	if (score_bits & (1 << j))
		1083	bit++ = offsetof (eio_dirent, score) 8 + j;
		1084	}
		1085
		1086	/* now actually do the sorting (a variant of MSD radix sort) */
		1087	{
		1088	eio_dirent base_stk [9 + sizeof (ino_t) 8], *base;
		1089	eio_dirent end_stk [9 + sizeof (ino_t) 8], *end;
		1090	unsigned char bit_stk [9 + sizeof (ino_t) 8];
		1091	int stk_idx = 0;
		1092
		1093	base_stk [stk_idx] = dents;
		1094	end_stk [stk_idx] = dents + size;
		1095	bit_stk [stk_idx] = bit - 1;
		1096
		1097	do
		1098	{
		1099	base = base_stk [stk_idx];
		1100	end = end_stk [stk_idx];
		1101	bit = bit_stk [stk_idx];
		1102
		1103	for (;;)
		1104	{
		1105	unsigned char O = *bit >> 3;
		1106	unsigned char M = 1 << (*bit & 7);
		1107
		1108	eio_dirent *a = base;
		1109	eio_dirent *b = end;
		1110
		1111	if (b - a < EIO_SORT_CUTOFF)
		1112	break;
		1113
		1114	/* now bit-partition the array on the bit */
		1115	/* this ugly asymmetric loop seems to perform much better than typical */
		1116	/* partition algos found in the literature */
		1117	do
		1118	if (!(((unsigned char *)a)[O] & M))
		1119	++a;
		1120	else if (!(((unsigned char *)--b)[O] & M))
		1121	{
		1122	eio_dirent tmp = a; a = b; b = tmp;
		1123	++a;
		1124	}
		1125	while (b > a);
		1126
		1127	/* next bit, or stop, if no bits left in this path */
		1128	if (!*--bit)
		1129	break;
		1130
		1131	base_stk [stk_idx] = a;
		1132	end_stk [stk_idx] = end;
		1133	bit_stk [stk_idx] = bit;
		1134	++stk_idx;
		1135
		1136	end = a;
		1137	}
		1138	}
		1139	while (stk_idx--);
		1140	}
		1141	}
		1142
		1143	static void
		1144	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1145	{
		1146	/* first move the smallest element to the front, to act as a sentinel */
		1147	{
		1148	int i;
		1149	eio_dirent *min = dents;
		1150
		1151	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1152	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1153	if (EIO_DENT_CMP (dents [i], <, *min))
		1154	min = &dents [i];
		1155
		1156	/* swap elements 0 and j (minimum) */
		1157	{
		1158	eio_dirent tmp = dents; dents = min; min = tmp;
		1159	}
		1160	}
		1161
		1162	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1163	{
		1164	eio_dirent i, j;
		1165
		1166	for (i = dents + 1; i < dents + size; ++i)
		1167	{
		1168	eio_dirent value = *i;
		1169
		1170	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1171	j [1] = j [0];
		1172
		1173	j [1] = value;
		1174	}
		1175	}
		1176	}
		1177
		1178	static void
		1179	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1180	{
		1181	if (size <= 1)
		1182	return; /* our insertion sort relies on size > 0 */
		1183
		1184	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1185	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1186	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1187
		1188	/* use an insertion sort at the end, or for small arrays, */
		1189	/* as insertion sort is more efficient for small partitions */
		1190	eio_dent_insertion_sort (dents, size);
		1191	}
		1192
927	/* read a full directory */	1193	/* read a full directory */
928	static void	1194	static void
929	eio__scandir (eio_req req, etp_worker self)	1195	eio__scandir (eio_req req, etp_worker self)
930	{	1196	{
931	DIR *dirp;	1197	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;	1198	EIO_STRUCT_DIRENT *entp;
938	char name, names;	1199	char name, names;
939	int memlen = 4096;	1200	int namesalloc = 4096;
940	int memofs = 0;	1201	int namesoffs = 0;
		1202	int flags = req->int1;
		1203	eio_dirent *dents = 0;
		1204	int dentalloc = 128;
941	int res = 0;	1205	int dentoffs = 0;
		1206	ino_t inode_bits = 0;
		1207
		1208	req->result = -1;
		1209
		1210	if (!(flags & EIO_READDIR_DENTS))
		1211	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);
942		1212
943	X_LOCK (wrklock);	1213	X_LOCK (wrklock);
		1214	/* the corresponding closedir is in ETP_WORKER_CLEAR */
944	self->dirp = dirp = opendir (req->ptr1);	1215	self->dirp = dirp = opendir (req->ptr1);
945	self->dbuf = u = malloc (sizeof (*u));
946	req->flags \|= EIO_FLAG_PTR2_FREE;	1216	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;
		1217	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
947	req->ptr2 = names = malloc (memlen);	1218	req->ptr2 = names = malloc (namesalloc);
948	X_UNLOCK (wrklock);	1219	X_UNLOCK (wrklock);
949		1220
950	if (dirp && u && names)	1221	if (dirp && names && (!flags \|\| dents))
951	for (;;)	1222	for (;;)
952	{	1223	{
953	errno = 0;	1224	errno = 0;
954	readdir_r (dirp, &u->d, &entp);	1225	entp = readdir (dirp);
955		1226
956	if (!entp)	1227	if (!entp)
		1228	{
		1229	if (errno)
		1230	break;
		1231
		1232	/* sort etc. */
		1233	req->int1 = flags;
		1234	req->result = dentoffs;
		1235
		1236	if (flags & EIO_READDIR_STAT_ORDER)
		1237	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
		1238	else if (flags & EIO_READDIR_DIRS_FIRST)
		1239	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1240	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1241	else
		1242	{
		1243	/* in this case, all is known, and we just put dirs first and sort them */
		1244	eio_dirent *oth = dents + dentoffs;
		1245	eio_dirent *dir = dents;
		1246
		1247	/* now partition dirs to the front, and non-dirs to the back */
		1248	/* by walking from both sides and swapping if necessary */
		1249	/* also clear score, so it doesn't influence sorting */
		1250	while (oth > dir)
		1251	{
		1252	if (dir->type == EIO_DT_DIR)
		1253	++dir;
		1254	else if ((--oth)->type == EIO_DT_DIR)
		1255	{
		1256	eio_dirent tmp = dir; dir = oth; oth = tmp;
		1257
		1258	++dir;
		1259	}
		1260	}
		1261
		1262	/* now sort the dirs only */
		1263	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1264	}
		1265
957	break;	1266	break;
		1267	}
958		1268
		1269	/* now add the entry to our list(s) */
959	name = entp->d_name;	1270	name = entp->d_name;
960		1271
		1272	/* skip . and .. entries */
961	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))	1273	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))
962	{	1274	{
963	int len = strlen (name) + 1;	1275	int len = D_NAMLEN (entp) + 1;
964		1276
965	res++;	1277	while (expect_false (namesoffs + len > namesalloc))
966
967	while (memofs + len > memlen)
968	{	1278	{
969	memlen *= 2;	1279	namesalloc *= 2;
970	X_LOCK (wrklock);	1280	X_LOCK (wrklock);
971	req->ptr2 = names = realloc (names, memlen);	1281	req->ptr2 = names = realloc (names, namesalloc);
972	X_UNLOCK (wrklock);	1282	X_UNLOCK (wrklock);
973		1283
974	if (!names)	1284	if (!names)
975	break;	1285	break;
976	}	1286	}
977		1287
978	memcpy (names + memofs, name, len);	1288	memcpy (names + namesoffs, name, len);
		1289
		1290	if (dents)
		1291	{
		1292	struct eio_dirent *ent;
		1293
		1294	if (expect_false (dentoffs == dentalloc))
		1295	{
		1296	dentalloc *= 2;
		1297	X_LOCK (wrklock);
		1298	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1299	X_UNLOCK (wrklock);
		1300
		1301	if (!dents)
		1302	break;
		1303	}
		1304
		1305	ent = dents + dentoffs;
		1306
		1307	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1308	ent->namelen = len - 1;
		1309	ent->inode = D_INO (entp);
		1310
		1311	inode_bits \|= ent->inode;
		1312
		1313	switch (D_TYPE (entp))
		1314	{
		1315	default:
		1316	ent->type = EIO_DT_UNKNOWN;
		1317	flags \|= EIO_READDIR_FOUND_UNKNOWN;
		1318	break;
		1319
		1320	#ifdef DT_FIFO
		1321	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1322	#endif
		1323	#ifdef DT_CHR
		1324	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1325	#endif
		1326	#ifdef DT_MPC
		1327	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1328	#endif
		1329	#ifdef DT_DIR
		1330	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1331	#endif
		1332	#ifdef DT_NAM
		1333	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1334	#endif
		1335	#ifdef DT_BLK
		1336	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1337	#endif
		1338	#ifdef DT_MPB
		1339	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1340	#endif
		1341	#ifdef DT_REG
		1342	case DT_REG: ent->type = EIO_DT_REG; break;
		1343	#endif
		1344	#ifdef DT_NWK
		1345	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1346	#endif
		1347	#ifdef DT_CMP
		1348	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1349	#endif
		1350	#ifdef DT_LNK
		1351	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1352	#endif
		1353	#ifdef DT_SOCK
		1354	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1355	#endif
		1356	#ifdef DT_DOOR
		1357	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1358	#endif
		1359	#ifdef DT_WHT
		1360	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1361	#endif
		1362	}
		1363
		1364	ent->score = 7;
		1365
		1366	if (flags & EIO_READDIR_DIRS_FIRST)
		1367	{
		1368	if (ent->type == EIO_DT_UNKNOWN)
		1369	{
		1370	if (name == '.') / leading dots are likely directories, and, in any case, rare */
		1371	ent->score = 1;
		1372	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1373	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1374	}
		1375	else if (ent->type == EIO_DT_DIR)
		1376	ent->score = 0;
		1377	}
		1378	}
		1379
979	memofs += len;	1380	namesoffs += len;
		1381	++dentoffs;
		1382	}
		1383
		1384	if (EIO_CANCELLED (req))
		1385	{
		1386	errno = ECANCELED;
		1387	break;
980	}	1388	}
981	}	1389	}
		1390	}
982		1391
983	if (errno)	1392	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
984	res = -1;	1393	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
985		1394	#else
986	req->result = res;	1395
		1396	int
		1397	eio__msync (void *mem, size_t len, int flags)
		1398	{
		1399	if (EIO_MS_ASYNC != MS_SYNC
		1400	\|\| EIO_MS_INVALIDATE != MS_INVALIDATE
		1401	\|\| EIO_MS_SYNC != MS_SYNC)
		1402	{
		1403	flags = 0
		1404	\| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1405	\| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1406	\| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1407	}
		1408
		1409	return msync (mem, len, flags);
		1410	}
		1411
		1412	#endif
		1413
		1414	int
		1415	eio__mtouch (void *mem, size_t len, int flags)
		1416	{
		1417	intptr_t addr = (intptr_t)mem;
		1418	intptr_t end = addr + len;
		1419	#ifdef PAGESIZE
		1420	const intptr_t page = PAGESIZE;
		1421	#else
		1422	static intptr_t page;
		1423
		1424	if (!page)
		1425	page = sysconf (_SC_PAGESIZE);
		1426	#endif
		1427
		1428	/* round down to start of page, although this is probably useless */
		1429	addr &= ~(page - 1); /* assume page size is always a power of two */
		1430
		1431	if (addr < end)
		1432	if (flags & EIO_MT_MODIFY) /* modify */
		1433	do { ((volatile sig_atomic_t )addr) \|= 0; } while ((addr += page) < len);
		1434	else
		1435	do { ((volatile sig_atomic_t )addr) ; } while ((addr += page) < len);
		1436
		1437	return 0;
987	}	1438	}
988		1439
989	/*****************************************************************************/	1440	/*****************************************************************************/
990		1441
991	#define ALLOC(len) \	1442	#define ALLOC(len) \
…		…
1047		1498
1048	--nready;	1499	--nready;
1049		1500
1050	X_UNLOCK (reqlock);	1501	X_UNLOCK (reqlock);
1051		1502
1052	errno = 0; /* strictly unnecessary */	1503	if (req->type < 0)
		1504	goto quit;
1053		1505
1054	if (!EIO_CANCELLED (req))	1506	if (!EIO_CANCELLED (req))
1055	switch (req->type)	1507	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		1508
1156	X_LOCK (reslock);	1509	X_LOCK (reslock);
1157		1510
1158	++npending;	1511	++npending;
1159		1512
…		…
1176		1529
1177	/*****************************************************************************/	1530	/*****************************************************************************/
1178		1531
1179	int eio_init (void (want_poll)(void), void (done_poll)(void))	1532	int eio_init (void (want_poll)(void), void (done_poll)(void))
1180	{	1533	{
1181	etp_init (want_poll, done_poll);	1534	return etp_init (want_poll, done_poll);
1182	}	1535	}
1183		1536
1184	static void eio_api_destroy (eio_req *req)	1537	static void eio_api_destroy (eio_req *req)
1185	{	1538	{
1186	free (req);	1539	free (req);
…		…
1208	{ \	1561	{ \
1209	eio_api_destroy (req); \	1562	eio_api_destroy (req); \
1210	return 0; \	1563	return 0; \
1211	}	1564	}
1212		1565
		1566	static void eio_execute (etp_worker self, eio_req req)
		1567	{
		1568	errno = 0;
		1569
		1570	switch (req->type)
		1571	{
		1572	case EIO_READ: ALLOC (req->size);
		1573	req->result = req->offs >= 0
		1574	? pread (req->int1, req->ptr2, req->size, req->offs)
		1575	: read (req->int1, req->ptr2, req->size); break;
		1576	case EIO_WRITE: req->result = req->offs >= 0
		1577	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1578	: write (req->int1, req->ptr2, req->size); break;
		1579
		1580	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1581	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1582
		1583	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1584	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1585	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1586	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1587	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1588	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1589
		1590	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1591	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1592	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1593	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1594
		1595	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1596	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1597	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1598	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1599	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1600	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1601
		1602	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1603	case EIO_CLOSE: req->result = close (req->int1); break;
		1604	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1605	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1606	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1607	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1608	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1609	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1610	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1611	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1612
		1613	case EIO_READLINK: ALLOC (NAME_MAX);
		1614	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
		1615
		1616	case EIO_SYNC: req->result = 0; sync (); break;
		1617	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1618	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1619	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1620	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1621	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1622
		1623	case EIO_READDIR: eio__scandir (req, self); break;
		1624
		1625	case EIO_BUSY:
		1626	#ifdef _WIN32
		1627	Sleep (req->nv1 * 1000.);
		1628	#else
		1629	{
		1630	struct timeval tv;
		1631
		1632	tv.tv_sec = req->nv1;
		1633	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1634
		1635	req->result = select (0, 0, 0, 0, &tv);
		1636	}
		1637	#endif
		1638	break;
		1639
		1640	case EIO_UTIME:
		1641	case EIO_FUTIME:
		1642	{
		1643	struct timeval tv[2];
		1644	struct timeval *times;
		1645
		1646	if (req->nv1 != -1. \|\| req->nv2 != -1.)
		1647	{
		1648	tv[0].tv_sec = req->nv1;
		1649	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1650	tv[1].tv_sec = req->nv2;
		1651	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1652
		1653	times = tv;
		1654	}
		1655	else
		1656	times = 0;
		1657
		1658
		1659	req->result = req->type == EIO_FUTIME
		1660	? futimes (req->int1, times)
		1661	: utimes (req->ptr1, times);
		1662	}
		1663	break;
		1664
		1665	case EIO_GROUP:
		1666	abort (); /* handled in eio_request */
		1667
		1668	case EIO_NOP:
		1669	req->result = 0;
		1670	break;
		1671
		1672	case EIO_CUSTOM:
		1673	((void ()(eio_req ))req->feed) (req);
		1674	break;
		1675
		1676	default:
		1677	req->result = -1;
		1678	break;
		1679	}
		1680
		1681	req->errorno = errno;
		1682	}
		1683
1213	#ifndef EIO_NO_WRAPPERS	1684	#ifndef EIO_NO_WRAPPERS
1214		1685
1215	eio_req eio_nop (int pri, eio_cb cb, void data)	1686	eio_req eio_nop (int pri, eio_cb cb, void data)
1216	{	1687	{
1217	REQ (EIO_NOP); SEND;	1688	REQ (EIO_NOP); SEND;
…		…
1230	eio_req eio_fsync (int fd, int pri, eio_cb cb, void data)	1701	eio_req eio_fsync (int fd, int pri, eio_cb cb, void data)
1231	{	1702	{
1232	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1703	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1233	}	1704	}
1234		1705
		1706	eio_req eio_msync (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1707	{
		1708	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1709	}
		1710
		1711	eio_req eio_mtouch (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1712	{
		1713	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1714	}
		1715
		1716	eio_req eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void data)
		1717	{
		1718	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1719	}
		1720
1235	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)	1721	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)
1236	{	1722	{
1237	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1723	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1238	}	1724	}
1239		1725
…		…
1258	}	1744	}
1259		1745
1260	eio_req eio_fstat (int fd, int pri, eio_cb cb, void data)	1746	eio_req eio_fstat (int fd, int pri, eio_cb cb, void data)
1261	{	1747	{
1262	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1748	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1749	}
		1750
		1751	eio_req eio_fstatvfs (int fd, int pri, eio_cb cb, void data)
		1752	{
		1753	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1263	}	1754	}
1264		1755
1265	eio_req eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void data)	1756	eio_req eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void data)
1266	{	1757	{
1267	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1758	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)	1832	eio_req eio_lstat (const char path, int pri, eio_cb cb, void *data)
1342	{	1833	{
1343	return eio__1path (EIO_LSTAT, path, pri, cb, data);	1834	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1344	}	1835	}
1345		1836
		1837	eio_req eio_statvfs (const char path, int pri, eio_cb cb, void *data)
		1838	{
		1839	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		1840	}
		1841
1346	eio_req eio_unlink (const char path, int pri, eio_cb cb, void *data)	1842	eio_req eio_unlink (const char path, int pri, eio_cb cb, void *data)
1347	{	1843	{
1348	return eio__1path (EIO_UNLINK, path, pri, cb, data);	1844	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1349	}	1845	}
1350		1846
1351	eio_req eio_rmdir (const char path, int pri, eio_cb cb, void *data)	1847	eio_req eio_rmdir (const char path, int pri, eio_cb cb, void *data)
1352	{	1848	{
1353	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1849	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1354	}	1850	}
1355		1851
1356	eio_req eio_readdir (const char path, int pri, eio_cb cb, void *data)	1852	eio_req eio_readdir (const char path, int flags, int pri, eio_cb cb, void *data)
1357	{	1853	{
1358	return eio__1path (EIO_READDIR, path, pri, cb, data);	1854	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1359	}	1855	}
1360		1856
1361	eio_req eio_mknod (const char path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1857	eio_req eio_mknod (const char path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1362	{	1858	{
1363	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1859	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1364	}	1860	}
1365		1861
1366	static eio_req *	1862	static eio_req *
1367	eio__2path (int type, const char path, const char new_path, int pri, eio_cb cb, void *data)	1863	eio__2path (int type, const char path, const char new_path, int pri, eio_cb cb, void *data)
1368	{	1864	{
…		…
1392	eio_req eio_rename (const char path, const char new_path, int pri, eio_cb cb, void data)	1888	eio_req eio_rename (const char path, const char new_path, int pri, eio_cb cb, void data)
1393	{	1889	{
1394	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1890	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1395	}	1891	}
1396		1892
		1893	eio_req eio_custom (eio_cb execute, int pri, eio_cb cb, void data)
		1894	{
		1895	REQ (EIO_CUSTOM); req->feed = (void ()(eio_req ))execute; SEND;
		1896	}
		1897
1397	#endif	1898	#endif
1398		1899
1399	eio_req eio_grp (eio_cb cb, void data)	1900	eio_req eio_grp (eio_cb cb, void data)
1400	{	1901	{
1401	const int pri = EIO_PRI_MAX;	1902	const int pri = EIO_PRI_MAX;
…		…
1427		1928
1428	void eio_grp_add (eio_req grp, eio_req req)	1929	void eio_grp_add (eio_req grp, eio_req req)
1429	{	1930	{
1430	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1931	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1431		1932
		1933	grp->flags \|= EIO_FLAG_GROUPADD;
		1934
1432	++grp->size;	1935	++grp->size;
1433	req->grp = grp;	1936	req->grp = grp;
1434		1937
1435	req->grp_prev = 0;	1938	req->grp_prev = 0;
1436	req->grp_next = grp->grp_first;	1939	req->grp_next = grp->grp_first;
…		…
1445	/* misc garbage */	1948	/* misc garbage */
1446		1949
1447	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1950	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1448	{	1951	{
1449	etp_worker wrk;	1952	etp_worker wrk;
		1953	ssize_t ret;
1450		1954
1451	wrk.dbuf = 0;	1955	wrk.dbuf = 0;
1452		1956
1453	eio__sendfile (ofd, ifd, offset, count, &wrk);	1957	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1454		1958
1455	if (wrk.dbuf)	1959	if (wrk.dbuf)
1456	free (wrk.dbuf);	1960	free (wrk.dbuf);
1457	}
1458		1961
		1962	return ret;
		1963	}
		1964

Diff Legend

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

Comparing libeio/eio.c (file contents): Revision 1.14 by root, Sat May 17 12:17:25 2008 UTC vs. Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC

Diff Legend

Comparing libeio/eio.c (file contents):
Revision 1.14 by root, Sat May 17 12:17:25 2008 UTC vs.
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC