ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.32 by root, Sat Jun 6 18:06:55 2009 UTC vs.
Revision 1.72 by root, Fri Jun 10 12:45:20 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5	* All rights reserved.	5	* All rights reserved.
6	*	6	*
7	* Redistribution and use in source and binary forms, with or without modifica-	7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:	8	* tion, are permitted provided that the following conditions are met:
9	*	9	*
…		…
35	* and other provisions required by the GPL. If you do not delete the	35	* and other provisions required by the GPL. If you do not delete the
36	* provisions above, a recipient may use your version of this file under	36	* provisions above, a recipient may use your version of this file under
37	* either the BSD or the GPL.	37	* either the BSD or the GPL.
38	*/	38	*/
39		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
40	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
41		46
42	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
43	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
44	#endif	49	#endif
45	#include "xthread.h"	50	#include "xthread.h"
…		…
49	#include <stdlib.h>	54	#include <stdlib.h>
50	#include <string.h>	55	#include <string.h>
51	#include <errno.h>	56	#include <errno.h>
52	#include <sys/types.h>	57	#include <sys/types.h>
53	#include <sys/stat.h>	58	#include <sys/stat.h>
		59	#include <sys/statvfs.h>
54	#include <limits.h>	60	#include <limits.h>
55	#include <fcntl.h>	61	#include <fcntl.h>
56	#include <assert.h>	62	#include <assert.h>
57		63
		64	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		65	/* intptr_t only comes form stdint.h, says idiot openbsd coder */
		66	#if HAVE_STDINT_H
		67	# include <stdint.h>
		68	#endif
		69
58	#ifndef EIO_FINISH	70	#ifndef EIO_FINISH
59	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	71	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60	#endif	72	#endif
61		73
62	#ifndef EIO_DESTROY	74	#ifndef EIO_DESTROY
…		…
70	#ifdef _WIN32	82	#ifdef _WIN32
71		83
72	/*doh*/	84	/*doh*/
73	#else	85	#else
74		86
75	# include "config.h"
76	# include <sys/time.h>	87	# include <sys/time.h>
77	# include <sys/select.h>	88	# include <sys/select.h>
78	# include <unistd.h>	89	# include <unistd.h>
79	# include <utime.h>	90	# include <utime.h>
80	# include <signal.h>	91	# include <signal.h>
81	# include <dirent.h>	92	# include <dirent.h>
82		93
		94	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		95	# include <sys/mman.h>
		96	#endif
		97
83	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	98	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)	99	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		100	# define _DIRENT_HAVE_D_TYPE /* sigh */
85	# define D_INO(de) (de)->d_fileno	101	# define D_INO(de) (de)->d_fileno
86	# define _DIRENT_HAVE_D_TYPE /* sigh */	102	# define D_NAMLEN(de) (de)->d_namlen
87	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600	103	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
88	# define D_INO(de) (de)->d_ino	104	# define D_INO(de) (de)->d_ino
89	# endif	105	# endif
		106
		107	#ifdef _D_EXACT_NAMLEN
		108	# undef D_NAMLEN
		109	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		110	#endif
90		111
91	# ifdef _DIRENT_HAVE_D_TYPE	112	# ifdef _DIRENT_HAVE_D_TYPE
92	# define D_TYPE(de) (de)->d_type	113	# define D_TYPE(de) (de)->d_type
93	# endif	114	# endif
94		115
…		…
99	#endif	120	#endif
100		121
101	#if HAVE_SENDFILE	122	#if HAVE_SENDFILE
102	# if __linux	123	# if __linux
103	# include <sys/sendfile.h>	124	# include <sys/sendfile.h>
104	# elif __freebsd	125	# elif __FreeBSD__ || defined __APPLE__
105	# include <sys/socket.h>	126	# include <sys/socket.h>
106	# include <sys/uio.h>	127	# include <sys/uio.h>
107	# elif __hpux	128	# elif __hpux
108	# include <sys/socket.h>	129	# include <sys/socket.h>
109	# elif __solaris /* not yet */	130	# elif __solaris
110	# include <sys/sendfile.h>	131	# include <sys/sendfile.h>
111	# else	132	# else
112	# error sendfile support requested but not available	133	# error sendfile support requested but not available
113	# endif	134	# endif
114	#endif	135	#endif
115		136
116	#ifndef D_TYPE	137	#ifndef D_TYPE
117	# define D_TYPE(de) 0	138	# define D_TYPE(de) 0
118	#endif	139	#endif
119
120	#ifndef D_INO	140	#ifndef D_INO
121	# define D_INO(de) 0	141	# define D_INO(de) 0
122	#endif	142	#endif
123		143	#ifndef D_NAMLEN
124	/* number of seconds after which an idle threads exit */	144	# define D_NAMLEN(de) strlen ((de)->d_name)
125	#define IDLE_TIMEOUT 10	145	#endif
126		146
127	/* used for struct dirent, AIX doesn't provide it */	147	/* used for struct dirent, AIX doesn't provide it */
128	#ifndef NAME_MAX	148	#ifndef NAME_MAX
129	# define NAME_MAX 4096	149	# define NAME_MAX 4096
		150	#endif
		151
		152	/* used for readlink etc. */
		153	#ifndef PATH_MAX
		154	# define PATH_MAX 4096
130	#endif	155	#endif
131		156
132	/* buffer size for various temporary buffers */	157	/* buffer size for various temporary buffers */
133	#define EIO_BUFSIZE 65536	158	#define EIO_BUFSIZE 65536
134		159
…		…
140	errno = ENOMEM; \	165	errno = ENOMEM; \
141	if (!eio_buf) \	166	if (!eio_buf) \
142	return -1;	167	return -1;
143		168
144	#define EIO_TICKS ((1000000 + 1023) >> 10)	169	#define EIO_TICKS ((1000000 + 1023) >> 10)
145
146	/*****************************************************************************/
147
148	#if __GNUC__ >= 3
149	# define expect(expr,value) __builtin_expect ((expr),(value))
150	#else
151	# define expect(expr,value) (expr)
152	#endif
153
154	#define expect_false(expr) expect ((expr) != 0, 0)
155	#define expect_true(expr) expect ((expr) != 0, 1)
156
157	/*****************************************************************************/
158		170
159	#define ETP_PRI_MIN EIO_PRI_MIN	171	#define ETP_PRI_MIN EIO_PRI_MIN
160	#define ETP_PRI_MAX EIO_PRI_MAX	172	#define ETP_PRI_MAX EIO_PRI_MAX
161		173
162	struct etp_worker;	174	struct etp_worker;
…		…
187		199
188	/*****************************************************************************/	200	/*****************************************************************************/
189		201
190	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	202	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
191		203
192	/* calculcate time difference in ~1/EIO_TICKS of a second */	204	/* calculate time difference in ~1/EIO_TICKS of a second */
		205	ECB_INLINE int
193	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	206	tvdiff (struct timeval *tv1, struct timeval *tv2)
194	{	207	{
195	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	208	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
196	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	209	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
197	}	210	}
198		211
…		…
205	static unsigned int max_poll_reqs; /* reslock */	218	static unsigned int max_poll_reqs; /* reslock */
206		219
207	static volatile unsigned int nreqs; /* reqlock */	220	static volatile unsigned int nreqs; /* reqlock */
208	static volatile unsigned int nready; /* reqlock */	221	static volatile unsigned int nready; /* reqlock */
209	static volatile unsigned int npending; /* reqlock */	222	static volatile unsigned int npending; /* reqlock */
210	static volatile unsigned int max_idle = 4;	223	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		224	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
211		225
212	static mutex_t wrklock = X_MUTEX_INIT;	226	static xmutex_t wrklock;
213	static mutex_t reslock = X_MUTEX_INIT;	227	static xmutex_t reslock;
214	static mutex_t reqlock = X_MUTEX_INIT;	228	static xmutex_t reqlock;
215	static cond_t reqwait = X_COND_INIT;	229	static xcond_t reqwait;
216		230
217	#if !HAVE_PREADWRITE	231	#if !HAVE_PREADWRITE
218	/*	232	/*
219	* make our pread/pwrite emulation safe against themselves, but not against	233	* make our pread/pwrite emulation safe against themselves, but not against
220	* normal read/write by using a mutex. slows down execution a lot,	234	* normal read/write by using a mutex. slows down execution a lot,
221	* but that's your problem, not mine.	235	* but that's your problem, not mine.
222	*/	236	*/
223	static mutex_t preadwritelock = X_MUTEX_INIT;	237	static xmutex_t preadwritelock = X_MUTEX_INIT;
224	#endif	238	#endif
225		239
226	typedef struct etp_worker	240	typedef struct etp_worker
227	{	241	{
228	/* locked by wrklock */	242	/* locked by wrklock */
229	struct etp_worker *prev, *next;	243	struct etp_worker *prev, *next;
230		244
231	thread_t tid;	245	xthread_t tid;
232		246
233	/* locked by reslock, reqlock or wrklock */	247	/* locked by reslock, reqlock or wrklock */
234	ETP_REQ *req; /* currently processed request */	248	ETP_REQ *req; /* currently processed request */
235		249
236	ETP_WORKER_COMMON	250	ETP_WORKER_COMMON
…		…
241	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	255	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
242	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	256	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
243		257
244	/* worker threads management */	258	/* worker threads management */
245		259
		260	static void ecb_cold
246	static void etp_worker_clear (etp_worker *wrk)	261	etp_worker_clear (etp_worker *wrk)
247	{	262	{
248	ETP_WORKER_CLEAR (wrk);	263	ETP_WORKER_CLEAR (wrk);
249	}	264	}
250		265
		266	static void ecb_cold
251	static void etp_worker_free (etp_worker *wrk)	267	etp_worker_free (etp_worker *wrk)
252	{	268	{
253	wrk->next->prev = wrk->prev;	269	wrk->next->prev = wrk->prev;
254	wrk->prev->next = wrk->next;	270	wrk->prev->next = wrk->next;
255		271
256	free (wrk);	272	free (wrk);
257	}	273	}
258		274
259	static unsigned int etp_nreqs (void)	275	static unsigned int
		276	etp_nreqs (void)
260	{	277	{
261	int retval;	278	int retval;
262	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	279	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
263	retval = nreqs;	280	retval = nreqs;
264	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	281	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
265	return retval;	282	return retval;
266	}	283	}
267		284
268	static unsigned int etp_nready (void)	285	static unsigned int
		286	etp_nready (void)
269	{	287	{
270	unsigned int retval;	288	unsigned int retval;
271		289
272	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	290	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273	retval = nready;	291	retval = nready;
274	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	292	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275		293
276	return retval;	294	return retval;
277	}	295	}
278		296
279	static unsigned int etp_npending (void)	297	static unsigned int
		298	etp_npending (void)
280	{	299	{
281	unsigned int retval;	300	unsigned int retval;
282		301
283	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
284	retval = npending;	303	retval = npending;
285	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	304	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
286		305
287	return retval;	306	return retval;
288	}	307	}
289		308
290	static unsigned int etp_nthreads (void)	309	static unsigned int
		310	etp_nthreads (void)
291	{	311	{
292	unsigned int retval;	312	unsigned int retval;
293		313
294	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	314	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
295	retval = started;	315	retval = started;
…		…
309	} etp_reqq;	329	} etp_reqq;
310		330
311	static etp_reqq req_queue;	331	static etp_reqq req_queue;
312	static etp_reqq res_queue;	332	static etp_reqq res_queue;
313		333
		334	static int ecb_noinline
314	static int reqq_push (etp_reqq *q, ETP_REQ *req)	335	reqq_push (etp_reqq *q, ETP_REQ *req)
315	{	336	{
316	int pri = req->pri;	337	int pri = req->pri;
317	req->next = 0;	338	req->next = 0;
318		339
319	if (q->qe[pri])	340	if (q->qe[pri])
…		…
325	q->qe[pri] = q->qs[pri] = req;	346	q->qe[pri] = q->qs[pri] = req;
326		347
327	return q->size++;	348	return q->size++;
328	}	349	}
329		350
		351	static ETP_REQ * ecb_noinline
330	static ETP_REQ *reqq_shift (etp_reqq *q)	352	reqq_shift (etp_reqq *q)
331	{	353	{
332	int pri;	354	int pri;
333		355
334	if (!q->size)	356	if (!q->size)
335	return 0;	357	return 0;
…		…
350	}	372	}
351		373
352	abort ();	374	abort ();
353	}	375	}
354		376
		377	static void ecb_cold
		378	etp_thread_init (void)
		379	{
		380	X_MUTEX_CREATE (wrklock);
		381	X_MUTEX_CREATE (reslock);
		382	X_MUTEX_CREATE (reqlock);
		383	X_COND_CREATE (reqwait);
		384	}
		385
		386	static void ecb_cold
355	static void etp_atfork_prepare (void)	387	etp_atfork_prepare (void)
356	{	388	{
357	X_LOCK (wrklock);	389	X_LOCK (wrklock);
358	X_LOCK (reqlock);	390	X_LOCK (reqlock);
359	X_LOCK (reslock);	391	X_LOCK (reslock);
360	#if !HAVE_PREADWRITE	392	#if !HAVE_PREADWRITE
361	X_LOCK (preadwritelock);	393	X_LOCK (preadwritelock);
362	#endif	394	#endif
363	}	395	}
364		396
		397	static void ecb_cold
365	static void etp_atfork_parent (void)	398	etp_atfork_parent (void)
366	{	399	{
367	#if !HAVE_PREADWRITE	400	#if !HAVE_PREADWRITE
368	X_UNLOCK (preadwritelock);	401	X_UNLOCK (preadwritelock);
369	#endif	402	#endif
370	X_UNLOCK (reslock);	403	X_UNLOCK (reslock);
371	X_UNLOCK (reqlock);	404	X_UNLOCK (reqlock);
372	X_UNLOCK (wrklock);	405	X_UNLOCK (wrklock);
373	}	406	}
374		407
		408	static void ecb_cold
375	static void etp_atfork_child (void)	409	etp_atfork_child (void)
376	{	410	{
377	ETP_REQ *prv;	411	ETP_REQ *prv;
378		412
379	while ((prv = reqq_shift (&req_queue)))	413	while ((prv = reqq_shift (&req_queue)))
380	ETP_DESTROY (prv);	414	ETP_DESTROY (prv);
…		…
397	idle = 0;	431	idle = 0;
398	nreqs = 0;	432	nreqs = 0;
399	nready = 0;	433	nready = 0;
400	npending = 0;	434	npending = 0;
401		435
402	etp_atfork_parent ();	436	etp_thread_init ();
403	}	437	}
404		438
405	static void	439	static void ecb_cold
406	etp_once_init (void)	440	etp_once_init (void)
407	{	441	{
		442	etp_thread_init ();
408	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	443	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
409	}	444	}
410		445
411	static int	446	static int ecb_cold
412	etp_init (void (*want_poll)(void), void (*done_poll)(void))	447	etp_init (void (*want_poll)(void), void (*done_poll)(void))
413	{	448	{
414	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	449	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
415		450
416	pthread_once (&doinit, etp_once_init);	451	pthread_once (&doinit, etp_once_init);
…		…
421	return 0;	456	return 0;
422	}	457	}
423		458
424	X_THREAD_PROC (etp_proc);	459	X_THREAD_PROC (etp_proc);
425		460
		461	static void ecb_cold
426	static void etp_start_thread (void)	462	etp_start_thread (void)
427	{	463	{
428	etp_worker *wrk = calloc (1, sizeof (etp_worker));	464	etp_worker *wrk = calloc (1, sizeof (etp_worker));
429		465
430	/*TODO*/	466	/*TODO*/
431	assert (("unable to allocate worker thread data", wrk));	467	assert (("unable to allocate worker thread data", wrk));
…		…
444	free (wrk);	480	free (wrk);
445		481
446	X_UNLOCK (wrklock);	482	X_UNLOCK (wrklock);
447	}	483	}
448		484
		485	static void
449	static void etp_maybe_start_thread (void)	486	etp_maybe_start_thread (void)
450	{	487	{
451	if (expect_true (etp_nthreads () >= wanted))	488	if (ecb_expect_true (etp_nthreads () >= wanted))
452	return;	489	return;
453		490
454	/* todo: maybe use idle here, but might be less exact */	491	/* todo: maybe use idle here, but might be less exact */
455	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	492	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
456	return;	493	return;
457		494
458	etp_start_thread ();	495	etp_start_thread ();
459	}	496	}
460		497
		498	static void ecb_cold
461	static void etp_end_thread (void)	499	etp_end_thread (void)
462	{	500	{
463	eio_req *req = calloc (1, sizeof (eio_req));	501	eio_req *req = calloc (1, sizeof (eio_req));
464		502
465	req->type = -1;	503	req->type = -1;
466	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	504	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
473	X_LOCK (wrklock);	511	X_LOCK (wrklock);
474	--started;	512	--started;
475	X_UNLOCK (wrklock);	513	X_UNLOCK (wrklock);
476	}	514	}
477		515
478	static int etp_poll (void)	516	static int
		517	etp_poll (void)
479	{	518	{
480	unsigned int maxreqs;	519	unsigned int maxreqs;
481	unsigned int maxtime;	520	unsigned int maxtime;
482	struct timeval tv_start, tv_now;	521	struct timeval tv_start, tv_now;
483		522
…		…
513		552
514	X_LOCK (reqlock);	553	X_LOCK (reqlock);
515	--nreqs;	554	--nreqs;
516	X_UNLOCK (reqlock);	555	X_UNLOCK (reqlock);
517		556
518	if (expect_false (req->type == EIO_GROUP && req->size))	557	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
519	{	558	{
520	req->int1 = 1; /* mark request as delayed */	559	req->int1 = 1; /* mark request as delayed */
521	continue;	560	continue;
522	}	561	}
523	else	562	else
524	{	563	{
525	int res = ETP_FINISH (req);	564	int res = ETP_FINISH (req);
526	if (expect_false (res))	565	if (ecb_expect_false (res))
527	return res;	566	return res;
528	}	567	}
529		568
530	if (expect_false (maxreqs && !--maxreqs))	569	if (ecb_expect_false (maxreqs && !--maxreqs))
531	break;	570	break;
532		571
533	if (maxtime)	572	if (maxtime)
534	{	573	{
535	gettimeofday (&tv_now, 0);	574	gettimeofday (&tv_now, 0);
…		…
541		580
542	errno = EAGAIN;	581	errno = EAGAIN;
543	return -1;	582	return -1;
544	}	583	}
545		584
		585	static void
546	static void etp_cancel (ETP_REQ *req)	586	etp_cancel (ETP_REQ *req)
547	{	587	{
548	X_LOCK (wrklock);	588	X_LOCK (wrklock);
549	req->flags |= EIO_FLAG_CANCELLED;	589	req->flags |= EIO_FLAG_CANCELLED;
550	X_UNLOCK (wrklock);	590	X_UNLOCK (wrklock);
551		591
552	eio_grp_cancel (req);	592	eio_grp_cancel (req);
553	}	593	}
554		594
		595	static void
555	static void etp_submit (ETP_REQ *req)	596	etp_submit (ETP_REQ *req)
556	{	597	{
557	req->pri -= ETP_PRI_MIN;	598	req->pri -= ETP_PRI_MIN;
558		599
559	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	600	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
560	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	601	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
561		602
562	if (expect_false (req->type == EIO_GROUP))	603	if (ecb_expect_false (req->type == EIO_GROUP))
563	{	604	{
564	/* I hope this is worth it :/ */	605	/* I hope this is worth it :/ */
565	X_LOCK (reqlock);	606	X_LOCK (reqlock);
566	++nreqs;	607	++nreqs;
567	X_UNLOCK (reqlock);	608	X_UNLOCK (reqlock);
…		…
586		627
587	etp_maybe_start_thread ();	628	etp_maybe_start_thread ();
588	}	629	}
589	}	630	}
590		631
		632	static void ecb_cold
591	static void etp_set_max_poll_time (double nseconds)	633	etp_set_max_poll_time (double nseconds)
592	{	634	{
593	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	635	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
594	max_poll_time = nseconds;	636	max_poll_time = nseconds * EIO_TICKS;
595	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	637	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
596	}	638	}
597		639
		640	static void ecb_cold
598	static void etp_set_max_poll_reqs (unsigned int maxreqs)	641	etp_set_max_poll_reqs (unsigned int maxreqs)
599	{	642	{
600	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	643	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
601	max_poll_reqs = maxreqs;	644	max_poll_reqs = maxreqs;
602	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	645	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
603	}	646	}
604		647
		648	static void ecb_cold
605	static void etp_set_max_idle (unsigned int nthreads)	649	etp_set_max_idle (unsigned int nthreads)
606	{	650	{
607	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	651	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
608	max_idle = nthreads <= 0 ? 1 : nthreads;	652	max_idle = nthreads;
609	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	653	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
610	}	654	}
611		655
		656	static void ecb_cold
		657	etp_set_idle_timeout (unsigned int seconds)
		658	{
		659	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		660	idle_timeout = seconds;
		661	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		662	}
		663
		664	static void ecb_cold
612	static void etp_set_min_parallel (unsigned int nthreads)	665	etp_set_min_parallel (unsigned int nthreads)
613	{	666	{
614	if (wanted < nthreads)	667	if (wanted < nthreads)
615	wanted = nthreads;	668	wanted = nthreads;
616	}	669	}
617		670
		671	static void ecb_cold
618	static void etp_set_max_parallel (unsigned int nthreads)	672	etp_set_max_parallel (unsigned int nthreads)
619	{	673	{
620	if (wanted > nthreads)	674	if (wanted > nthreads)
621	wanted = nthreads;	675	wanted = nthreads;
622		676
623	while (started > wanted)	677	while (started > wanted)
624	etp_end_thread ();	678	etp_end_thread ();
625	}	679	}
626		680
627	/*****************************************************************************/	681	/*****************************************************************************/
628		682
		683	static void
629	static void grp_try_feed (eio_req *grp)	684	grp_try_feed (eio_req *grp)
630	{	685	{
631	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	686	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
632	{	687	{
633	grp->flags &= ~EIO_FLAG_GROUPADD;	688	grp->flags &= ~EIO_FLAG_GROUPADD;
634		689
…		…
641	break;	696	break;
642	}	697	}
643	}	698	}
644	}	699	}
645		700
		701	static int
646	static int grp_dec (eio_req *grp)	702	grp_dec (eio_req *grp)
647	{	703	{
648	--grp->size;	704	--grp->size;
649		705
650	/* call feeder, if applicable */	706	/* call feeder, if applicable */
651	grp_try_feed (grp);	707	grp_try_feed (grp);
…		…
655	return eio_finish (grp);	711	return eio_finish (grp);
656	else	712	else
657	return 0;	713	return 0;
658	}	714	}
659		715
		716	void
660	void eio_destroy (eio_req *req)	717	eio_destroy (eio_req *req)
661	{	718	{
662	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	719	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
663	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	720	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
664		721
665	EIO_DESTROY (req);	722	EIO_DESTROY (req);
666	}	723	}
667		724
		725	static int
668	static int eio_finish (eio_req *req)	726	eio_finish (eio_req *req)
669	{	727	{
670	int res = EIO_FINISH (req);	728	int res = EIO_FINISH (req);
671		729
672	if (req->grp)	730	if (req->grp)
673	{	731	{
…		…
690	eio_destroy (req);	748	eio_destroy (req);
691		749
692	return res;	750	return res;
693	}	751	}
694		752
		753	void
695	void eio_grp_cancel (eio_req *grp)	754	eio_grp_cancel (eio_req *grp)
696	{	755	{
697	for (grp = grp->grp_first; grp; grp = grp->grp_next)	756	for (grp = grp->grp_first; grp; grp = grp->grp_next)
698	eio_cancel (grp);	757	eio_cancel (grp);
699	}	758	}
700		759
		760	void
701	void eio_cancel (eio_req *req)	761	eio_cancel (eio_req *req)
702	{	762	{
703	etp_cancel (req);	763	etp_cancel (req);
704	}	764	}
705		765
		766	void
706	void eio_submit (eio_req *req)	767	eio_submit (eio_req *req)
707	{	768	{
708	etp_submit (req);	769	etp_submit (req);
709	}	770	}
710		771
711	unsigned int eio_nreqs (void)	772	unsigned int
		773	eio_nreqs (void)
712	{	774	{
713	return etp_nreqs ();	775	return etp_nreqs ();
714	}	776	}
715		777
716	unsigned int eio_nready (void)	778	unsigned int
		779	eio_nready (void)
717	{	780	{
718	return etp_nready ();	781	return etp_nready ();
719	}	782	}
720		783
721	unsigned int eio_npending (void)	784	unsigned int
		785	eio_npending (void)
722	{	786	{
723	return etp_npending ();	787	return etp_npending ();
724	}	788	}
725		789
726	unsigned int eio_nthreads (void)	790	unsigned int ecb_cold
		791	eio_nthreads (void)
727	{	792	{
728	return etp_nthreads ();	793	return etp_nthreads ();
729	}	794	}
730		795
		796	void ecb_cold
731	void eio_set_max_poll_time (double nseconds)	797	eio_set_max_poll_time (double nseconds)
732	{	798	{
733	etp_set_max_poll_time (nseconds);	799	etp_set_max_poll_time (nseconds);
734	}	800	}
735		801
		802	void ecb_cold
736	void eio_set_max_poll_reqs (unsigned int maxreqs)	803	eio_set_max_poll_reqs (unsigned int maxreqs)
737	{	804	{
738	etp_set_max_poll_reqs (maxreqs);	805	etp_set_max_poll_reqs (maxreqs);
739	}	806	}
740		807
		808	void ecb_cold
741	void eio_set_max_idle (unsigned int nthreads)	809	eio_set_max_idle (unsigned int nthreads)
742	{	810	{
743	etp_set_max_idle (nthreads);	811	etp_set_max_idle (nthreads);
744	}	812	}
745		813
		814	void ecb_cold
		815	eio_set_idle_timeout (unsigned int seconds)
		816	{
		817	etp_set_idle_timeout (seconds);
		818	}
		819
		820	void ecb_cold
746	void eio_set_min_parallel (unsigned int nthreads)	821	eio_set_min_parallel (unsigned int nthreads)
747	{	822	{
748	etp_set_min_parallel (nthreads);	823	etp_set_min_parallel (nthreads);
749	}	824	}
750		825
		826	void ecb_cold
751	void eio_set_max_parallel (unsigned int nthreads)	827	eio_set_max_parallel (unsigned int nthreads)
752	{	828	{
753	etp_set_max_parallel (nthreads);	829	etp_set_max_parallel (nthreads);
754	}	830	}
755		831
756	int eio_poll (void)	832	int eio_poll (void)
…		…
798		874
799	return res;	875	return res;
800	}	876	}
801	#endif	877	#endif
802		878
803	#ifndef HAVE_FUTIMES	879	#ifndef HAVE_UTIMES
804		880
805	# undef utimes	881	# undef utimes
806	# undef futimes
807	# define utimes(path,times) eio__utimes (path, times)	882	# define utimes(path,times) eio__utimes (path, times)
808	# define futimes(fd,times) eio__futimes (fd, times)
809		883
810	static int	884	static int
811	eio__utimes (const char *filename, const struct timeval times[2])	885	eio__utimes (const char *filename, const struct timeval times[2])
812	{	886	{
813	if (times)	887	if (times)
…		…
821	}	895	}
822	else	896	else
823	return utime (filename, 0);	897	return utime (filename, 0);
824	}	898	}
825		899
		900	#endif
		901
		902	#ifndef HAVE_FUTIMES
		903
		904	# undef futimes
		905	# define futimes(fd,times) eio__futimes (fd, times)
		906
		907	static int
826	static int eio__futimes (int fd, const struct timeval tv[2])	908	eio__futimes (int fd, const struct timeval tv[2])
827	{	909	{
828	errno = ENOSYS;	910	errno = ENOSYS;
829	return -1;	911	return -1;
830	}	912	}
831		913
…		…
835	# undef fdatasync	917	# undef fdatasync
836	# define fdatasync(fd) fsync (fd)	918	# define fdatasync(fd) fsync (fd)
837	#endif	919	#endif
838		920
839	/* sync_file_range always needs emulation */	921	/* sync_file_range always needs emulation */
840	int	922	static int
841	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	923	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
842	{	924	{
843	#if HAVE_SYNC_FILE_RANGE	925	#if HAVE_SYNC_FILE_RANGE
844	int res;	926	int res;
845		927
…		…
858	if (!res || errno != ENOSYS)	940	if (!res || errno != ENOSYS)
859	return res;	941	return res;
860	#endif	942	#endif
861		943
862	/* even though we could play tricks with the flags, it's better to always	944	/* even though we could play tricks with the flags, it's better to always
863	* call fdatasync, as thta matches the expectation of it's users best */	945	* call fdatasync, as that matches the expectation of its users best */
864	return fdatasync (fd);	946	return fdatasync (fd);
865	}	947	}
866		948
867	#if !HAVE_READAHEAD	949	#if !HAVE_READAHEAD
868	# undef readahead	950	# undef readahead
…		…
891		973
892	/* sendfile always needs emulation */	974	/* sendfile always needs emulation */
893	static ssize_t	975	static ssize_t
894	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	976	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
895	{	977	{
		978	ssize_t written = 0;
896	ssize_t res;	979	ssize_t res;
897		980
898	if (!count)	981	if (!count)
899	return 0;	982	return 0;
900		983
		984	for (;;)
		985	{
901	#if HAVE_SENDFILE	986	#if HAVE_SENDFILE
902	# if __linux	987	# if __linux
		988	off_t soffset = offset;
903	res = sendfile (ofd, ifd, &offset, count);	989	res = sendfile (ofd, ifd, &soffset, count);
904		990
905	# elif __freebsd	991	# elif __FreeBSD__
906	/*	992	/*
907	* Of course, the freebsd sendfile is a dire hack with no thoughts	993	* Of course, the freebsd sendfile is a dire hack with no thoughts
908	* wasted on making it similar to other I/O functions.	994	* wasted on making it similar to other I/O functions.
909	*/	995	*/
910	{
911	off_t sbytes;	996	off_t sbytes;
912	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	997	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
913		998
914	if (res < 0 && sbytes)	999	#if 0 /* according to the manpage, this is correct, but broken behaviour */
915	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1000	/* freebsd' sendfile will return 0 on success */
		1001	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1002	/* not on e.g. EIO or EPIPE - sounds broken */
		1003	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
916	res = sbytes;	1004	res = sbytes;
917	}	1005	#endif
		1006
		1007	/* according to source inspection, this is correct, and useful behaviour */
		1008	if (sbytes)
		1009	res = sbytes;
		1010
		1011	# elif defined (__APPLE__)
		1012	off_t sbytes = count;
		1013	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1014
		1015	/* according to the manpage, sbytes is always valid */
		1016	if (sbytes)
		1017	res = sbytes;
918		1018
919	# elif __hpux	1019	# elif __hpux
920	res = sendfile (ofd, ifd, offset, count, 0, 0);	1020	res = sendfile (ofd, ifd, offset, count, 0, 0);
921		1021
922	# elif __solaris	1022	# elif __solaris
923	{
924	struct sendfilevec vec;	1023	struct sendfilevec vec;
925	size_t sbytes;	1024	size_t sbytes;
926		1025
927	vec.sfv_fd = ifd;	1026	vec.sfv_fd = ifd;
928	vec.sfv_flag = 0;	1027	vec.sfv_flag = 0;
929	vec.sfv_off = offset;	1028	vec.sfv_off = offset;
930	vec.sfv_len = count;	1029	vec.sfv_len = count;
931		1030
932	res = sendfilev (ofd, &vec, 1, &sbytes);	1031	res = sendfilev (ofd, &vec, 1, &sbytes);
933		1032
934	if (res < 0 && sbytes)	1033	if (res < 0 && sbytes)
935	res = sbytes;	1034	res = sbytes;
936	}
937		1035
938	# endif	1036	# endif
		1037
		1038	#elif defined (_WIN32)
		1039	/* does not work, just for documentation of what would need to be done */
		1040	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1041	/* libeio guarantees that the file offset does not change, and windows */
		1042	/* has no way to get an independent handle to the same file description */
		1043	HANDLE h = TO_SOCKET (ifd);
		1044	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1045	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1046
939	#else	1047	#else
940	res = -1;	1048	res = -1;
941	errno = ENOSYS;	1049	errno = ENOSYS;
942	#endif	1050	#endif
943		1051
		1052	/* we assume sendfile can copy at least 128mb in one go */
		1053	if (res <= 128 * 1024 * 1024)
		1054	{
		1055	if (res > 0)
		1056	written += res;
		1057
		1058	if (written)
		1059	return written;
		1060
		1061	break;
		1062	}
		1063	else
		1064	{
		1065	/* if we requested more, then probably the kernel was lazy */
		1066	written += res;
		1067	offset += res;
		1068	count -= res;
		1069
		1070	if (!count)
		1071	return written;
		1072	}
		1073	}
		1074
944	if (res < 0	1075	if (res < 0
945	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1076	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1077	/* BSDs */
		1078	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1079	|| errno == ENOTSUP
		1080	#endif
		1081	|| errno == EOPNOTSUPP /* BSDs */
946	#if __solaris	1082	#if __solaris
947	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1083	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
948	#endif	1084	#endif
949	)	1085	)
950	)	1086	)
…		…
981	}	1117	}
982		1118
983	return res;	1119	return res;
984	}	1120	}
985		1121
986	static int	1122	static signed char
987	eio_dent_cmp (const void *a_, const void *b_)	1123	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
988	{	1124	{
989	const eio_dirent *a = (const eio_dirent *)a_;	1125	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
990	const eio_dirent *b = (const eio_dirent *)b_;	1126	: a->inode < b->inode ? -1
		1127	: a->inode > b->inode ? 1
		1128	: 0;
		1129	}
991		1130
992	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score	1131	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
993	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */	1132
		1133	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1134	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1135
		1136	static void
		1137	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1138	{
		1139	unsigned char bits [9 + sizeof (ino_t) * 8];
		1140	unsigned char *bit = bits;
		1141
		1142	assert (CHAR_BIT == 8);
		1143	assert (sizeof (eio_dirent) * 8 < 256);
		1144	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1145	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1146
		1147	if (size <= EIO_SORT_FAST)
		1148	return;
		1149
		1150	/* first prepare an array of bits to test in our radix sort */
		1151	/* try to take endianness into account, as well as differences in ino_t sizes */
		1152	/* inode_bits must contain all inodes ORed together */
		1153	/* which is used to skip bits that are 0 everywhere, which is very common */
		1154	{
		1155	ino_t endianness;
		1156	int i, j;
		1157
		1158	/* we store the byte offset of byte n into byte n of "endianness" */
		1159	for (i = 0; i < sizeof (ino_t); ++i)
		1160	((unsigned char *)&endianness)[i] = i;
		1161
		1162	*bit++ = 0;
		1163
		1164	for (i = 0; i < sizeof (ino_t); ++i)
		1165	{
		1166	/* shifting off the byte offsets out of "endianness" */
		1167	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1168	endianness >>= 8;
		1169
		1170	for (j = 0; j < 8; ++j)
		1171	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1172	*bit++ = offs + j;
		1173	}
		1174
		1175	for (j = 0; j < 8; ++j)
		1176	if (score_bits & (1 << j))
		1177	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1178	}
		1179
		1180	/* now actually do the sorting (a variant of MSD radix sort) */
		1181	{
		1182	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1183	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1184	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1185	int stk_idx = 0;
		1186
		1187	base_stk [stk_idx] = dents;
		1188	end_stk [stk_idx] = dents + size;
		1189	bit_stk [stk_idx] = bit - 1;
		1190
		1191	do
		1192	{
		1193	base = base_stk [stk_idx];
		1194	end = end_stk [stk_idx];
		1195	bit = bit_stk [stk_idx];
		1196
		1197	for (;;)
		1198	{
		1199	unsigned char O = *bit >> 3;
		1200	unsigned char M = 1 << (*bit & 7);
		1201
		1202	eio_dirent *a = base;
		1203	eio_dirent *b = end;
		1204
		1205	if (b - a < EIO_SORT_CUTOFF)
		1206	break;
		1207
		1208	/* now bit-partition the array on the bit */
		1209	/* this ugly asymmetric loop seems to perform much better than typical */
		1210	/* partition algos found in the literature */
		1211	do
		1212	if (!(((unsigned char *)a)[O] & M))
		1213	++a;
		1214	else if (!(((unsigned char *)--b)[O] & M))
		1215	{
		1216	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1217	++a;
		1218	}
		1219	while (b > a);
		1220
		1221	/* next bit, or stop, if no bits left in this path */
		1222	if (!*--bit)
		1223	break;
		1224
		1225	base_stk [stk_idx] = a;
		1226	end_stk [stk_idx] = end;
		1227	bit_stk [stk_idx] = bit;
		1228	++stk_idx;
		1229
		1230	end = a;
		1231	}
		1232	}
		1233	while (stk_idx--);
		1234	}
		1235	}
		1236
		1237	static void
		1238	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1239	{
		1240	/* first move the smallest element to the front, to act as a sentinel */
		1241	{
		1242	int i;
		1243	eio_dirent *min = dents;
		1244
		1245	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1246	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1247	if (EIO_DENT_CMP (dents [i], <, *min))
		1248	min = &dents [i];
		1249
		1250	/* swap elements 0 and j (minimum) */
		1251	{
		1252	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1253	}
		1254	}
		1255
		1256	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1257	{
		1258	eio_dirent *i, *j;
		1259
		1260	for (i = dents + 1; i < dents + size; ++i)
		1261	{
		1262	eio_dirent value = *i;
		1263
		1264	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1265	j [1] = j [0];
		1266
		1267	j [1] = value;
		1268	}
		1269	}
		1270	}
		1271
		1272	static void
		1273	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1274	{
		1275	if (size <= 1)
		1276	return; /* our insertion sort relies on size > 0 */
		1277
		1278	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1279	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1280	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1281
		1282	/* use an insertion sort at the end, or for small arrays, */
		1283	/* as insertion sort is more efficient for small partitions */
		1284	eio_dent_insertion_sort (dents, size);
994	}	1285	}
995		1286
996	/* read a full directory */	1287	/* read a full directory */
997	static void	1288	static void
998	eio__scandir (eio_req *req, etp_worker *self)	1289	eio__scandir (eio_req *req, etp_worker *self)
999	{	1290	{
1000	DIR *dirp;	1291	DIR *dirp;
1001	EIO_STRUCT_DIRENT *entp;	1292	EIO_STRUCT_DIRENT *entp;
1002	unsigned char *name, *names;	1293	char *name, *names;
1003	int namesalloc = 4096;	1294	int namesalloc = 4096;
1004	int namesoffs = 0;	1295	int namesoffs = 0;
1005	int flags = req->int1;	1296	int flags = req->int1;
1006	eio_dirent *dents = 0;	1297	eio_dirent *dents = 0;
1007	int dentalloc = 128;	1298	int dentalloc = 128;
1008	int dentoffs = 0;	1299	int dentoffs = 0;
		1300	ino_t inode_bits = 0;
1009		1301
1010	req->result = -1;	1302	req->result = -1;
1011		1303
1012	if (!(flags & EIO_READDIR_DENTS))	1304	if (!(flags & EIO_READDIR_DENTS))
1013	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1305	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1014		1306
1015	X_LOCK (wrklock);	1307	X_LOCK (wrklock);
1016	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1308	/* the corresponding closedir is in ETP_WORKER_CLEAR */
1017	self->dirp = dirp = opendir (req->ptr1);	1309	self->dirp = dirp = opendir (req->ptr1);
		1310
1018	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1311	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1019	req->ptr1 = names = malloc (namesalloc);
1020	req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1312	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
		1313	req->ptr2 = names = malloc (namesalloc);
1021	X_UNLOCK (wrklock);	1314	X_UNLOCK (wrklock);
1022		1315
1023	if (dirp && names && (!flags || dents))	1316	if (dirp && names && (!flags || dents))
1024	for (;;)	1317	for (;;)
1025	{	1318	{
…		…
1033		1326
1034	/* sort etc. */	1327	/* sort etc. */
1035	req->int1 = flags;	1328	req->int1 = flags;
1036	req->result = dentoffs;	1329	req->result = dentoffs;
1037		1330
1038	if (dents)
1039	{
1040	eio_dirent *ent = dents + dentoffs;
1041
1042	while (ent > dents)
1043	(--ent)->name = names + (size_t)ent->name;
1044	}
1045
1046	if (flags & EIO_READDIR_STAT_ORDER	1331	if (flags & EIO_READDIR_STAT_ORDER)
1047	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1332	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1048	{
1049	/* pray your qsort doesn't use quicksort */
1050	qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1051	}
1052	else if (flags & EIO_READDIR_DIRS_FIRST)	1333	else if (flags & EIO_READDIR_DIRS_FIRST)
		1334	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1335	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1336	else
1053	{	1337	{
1054	/* in this case, all is known, and we just put dirs first and sort them */	1338	/* in this case, all is known, and we just put dirs first and sort them */
1055	eio_dirent *ent = dents + dentoffs;	1339	eio_dirent *oth = dents + dentoffs;
1056	eio_dirent *dir = dents;	1340	eio_dirent *dir = dents;
1057		1341
		1342	/* now partition dirs to the front, and non-dirs to the back */
		1343	/* by walking from both sides and swapping if necessary */
1058	while (ent > dir)	1344	while (oth > dir)
1059	{	1345	{
1060	if (dir->type == DT_DIR)	1346	if (dir->type == EIO_DT_DIR)
1061	++dir;	1347	++dir;
1062	else
1063	{
1064	--ent;
1065
1066	if (ent->type == DT_DIR)	1348	else if ((--oth)->type == EIO_DT_DIR)
1067	{	1349	{
1068	eio_dirent tmp = *dir;	1350	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1069	*dir = *ent;
1070	*ent = tmp;
1071		1351
1072	++dir;	1352	++dir;
1073	}	1353	}
1074	}	1354	}
		1355
		1356	/* now sort the dirs only (dirs all have the same score) */
		1357	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1075	}	1358	}
1076
1077	/* now sort the dirs only */
1078	qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1079	}
1080
1081	/* only provide the names array unless DENTS is specified */
1082	if (!(flags & EIO_READDIR_DENTS))
1083	{
1084	X_LOCK (wrklock);
1085	assert (!dents);
1086	req->ptr1 = 0;
1087	req->ptr2 = names;
1088	X_UNLOCK (wrklock);
1089	}
1090		1359
1091	break;	1360	break;
1092	}	1361	}
1093		1362
1094	/* now add the entry to our list(s) */	1363	/* now add the entry to our list(s) */
1095	name = entp->d_name;	1364	name = entp->d_name;
1096		1365
1097	/* skip . and .. entries */	1366	/* skip . and .. entries */
1098	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1367	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1099	{	1368	{
1100	int len = strlen (name) + 1;	1369	int len = D_NAMLEN (entp) + 1;
1101		1370
1102	while (expect_false (namesoffs + len > namesalloc))	1371	while (ecb_expect_false (namesoffs + len > namesalloc))
1103	{	1372	{
1104	namesalloc *= 2;	1373	namesalloc *= 2;
1105	X_LOCK (wrklock);	1374	X_LOCK (wrklock);
1106	req->ptr1 = names = realloc (names, namesalloc);	1375	req->ptr2 = names = realloc (names, namesalloc);
1107	X_UNLOCK (wrklock);	1376	X_UNLOCK (wrklock);
1108		1377
1109	if (!names)	1378	if (!names)
1110	break;	1379	break;
1111	}	1380	}
…		…
1114		1383
1115	if (dents)	1384	if (dents)
1116	{	1385	{
1117	struct eio_dirent *ent;	1386	struct eio_dirent *ent;
1118		1387
1119	if (expect_false (dentoffs == dentalloc))	1388	if (ecb_expect_false (dentoffs == dentalloc))
1120	{	1389	{
1121	dentalloc *= 2;	1390	dentalloc *= 2;
1122	X_LOCK (wrklock);	1391	X_LOCK (wrklock);
1123	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1392	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124	X_UNLOCK (wrklock);	1393	X_UNLOCK (wrklock);
1125		1394
1126	if (!dents)	1395	if (!dents)
1127	break;	1396	break;
1128	}	1397	}
1129		1398
1130	ent = dents + dentoffs;	1399	ent = dents + dentoffs;
1131		1400
1132	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1401	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1133	ent->namelen = len - 1;	1402	ent->namelen = len - 1;
1134	ent->inode = D_INO (entp);	1403	ent->inode = D_INO (entp);
		1404
		1405	inode_bits |= ent->inode;
1135		1406
1136	switch (D_TYPE (entp))	1407	switch (D_TYPE (entp))
1137	{	1408	{
1138	default:	1409	default:
1139	ent->type = EIO_DT_UNKNOWN;	1410	ent->type = EIO_DT_UNKNOWN;
…		…
1144	case DT_FIFO: ent->type = EIO_DT_FIFO; break;	1415	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145	#endif	1416	#endif
1146	#ifdef DT_CHR	1417	#ifdef DT_CHR
1147	case DT_CHR: ent->type = EIO_DT_CHR; break;	1418	case DT_CHR: ent->type = EIO_DT_CHR; break;
1148	#endif	1419	#endif
		1420	#ifdef DT_MPC
		1421	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1422	#endif
1149	#ifdef DT_DIR	1423	#ifdef DT_DIR
1150	case DT_DIR: ent->type = EIO_DT_DIR; break;	1424	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1425	#endif
		1426	#ifdef DT_NAM
		1427	case DT_NAM: ent->type = EIO_DT_NAM; break;
1151	#endif	1428	#endif
1152	#ifdef DT_BLK	1429	#ifdef DT_BLK
1153	case DT_BLK: ent->type = EIO_DT_BLK; break;	1430	case DT_BLK: ent->type = EIO_DT_BLK; break;
1154	#endif	1431	#endif
		1432	#ifdef DT_MPB
		1433	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1434	#endif
1155	#ifdef DT_REG	1435	#ifdef DT_REG
1156	case DT_REG: ent->type = EIO_DT_REG; break;	1436	case DT_REG: ent->type = EIO_DT_REG; break;
		1437	#endif
		1438	#ifdef DT_NWK
		1439	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1440	#endif
		1441	#ifdef DT_CMP
		1442	case DT_CMP: ent->type = EIO_DT_CMP; break;
1157	#endif	1443	#endif
1158	#ifdef DT_LNK	1444	#ifdef DT_LNK
1159	case DT_LNK: ent->type = EIO_DT_LNK; break;	1445	case DT_LNK: ent->type = EIO_DT_LNK; break;
1160	#endif	1446	#endif
1161	#ifdef DT_SOCK	1447	#ifdef DT_SOCK
1162	case DT_SOCK: ent->type = EIO_DT_SOCK; break;	1448	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163	#endif	1449	#endif
		1450	#ifdef DT_DOOR
		1451	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1452	#endif
1164	#ifdef DT_WHT	1453	#ifdef DT_WHT
1165	case DT_WHT: ent->type = EIO_DT_WHT; break;	1454	case DT_WHT: ent->type = EIO_DT_WHT; break;
1166	#endif	1455	#endif
1167	}	1456	}
1168		1457
1169	ent->score = 0;	1458	ent->score = 7;
1170		1459
1171	if (flags & EIO_READDIR_DIRS_FIRST)	1460	if (flags & EIO_READDIR_DIRS_FIRST)
1172	{	1461	{
1173	if (ent->type == EIO_DT_UNKNOWN)	1462	if (ent->type == EIO_DT_UNKNOWN)
1174	{	1463	{
1175	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1464	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176	ent->score = 98;	1465	ent->score = 1;
1177	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1466	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1178	ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */	1467	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1179	}	1468	}
1180	else if (ent->type == DT_DIR)	1469	else if (ent->type == EIO_DT_DIR)
1181	ent->score = 100;	1470	ent->score = 0;
1182	}	1471	}
1183	}	1472	}
1184		1473
1185	namesoffs += len;	1474	namesoffs += len;
1186	++dentoffs;	1475	++dentoffs;
1187	}	1476	}
		1477
		1478	if (EIO_CANCELLED (req))
		1479	{
		1480	errno = ECANCELED;
		1481	break;
		1482	}
1188	}	1483	}
1189	else
1190	req->result = -1;
1191	}	1484	}
1192		1485
1193	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1194	# undef msync
1195	# define msync(a,b,c) ((errno = ENOSYS), -1)
1196	#endif
1197
1198	int
1199	eio__mtouch (void *mem, size_t len, int flags)
1200	{
1201	intptr_t addr = (intptr_t)mem;
1202	intptr_t end = addr + len;
1203	#ifdef PAGESIZE	1486	#ifdef PAGESIZE
1204	const intptr_t page = PAGESIZE;	1487	# define eio_pagesize() PAGESIZE
1205	#else	1488	#else
		1489	static intptr_t
		1490	eio_pagesize (void)
		1491	{
1206	static intptr_t page;	1492	static intptr_t page;
1207		1493
1208	if (!page)	1494	if (!page)
1209	page = sysconf (_SC_PAGESIZE);	1495	page = sysconf (_SC_PAGESIZE);
		1496
		1497	return page;
		1498	}
		1499	#endif
		1500
		1501	static void
		1502	eio_page_align (void **addr, size_t *length)
		1503	{
		1504	intptr_t mask = eio_pagesize () - 1;
		1505
		1506	/* round down addr */
		1507	intptr_t adj = mask & (intptr_t)*addr;
		1508
		1509	*addr = (void *)((intptr_t)*addr - adj);
		1510	*length += adj;
		1511
		1512	/* round up length */
		1513	*length = (*length + mask) & ~mask;
		1514	}
		1515
		1516	#if !_POSIX_MEMLOCK
		1517	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1518	#else
		1519
		1520	static int
		1521	eio__mlockall (int flags)
		1522	{
		1523	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1524	extern int mallopt (int, int);
		1525	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1210	#endif	1526	#endif
1211		1527
1212	addr &= ~(page - 1); /* assume page size is always a power of two */	1528	if (EIO_MCL_CURRENT != MCL_CURRENT
		1529	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1530	{
		1531	flags = 0
		1532	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1533	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1534	}
1213		1535
		1536	return mlockall (flags);
		1537	}
		1538	#endif
		1539
		1540	#if !_POSIX_MEMLOCK_RANGE
		1541	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1542	#else
		1543
		1544	static int
		1545	eio__mlock (void *addr, size_t length)
		1546	{
		1547	eio_page_align (&addr, &length);
		1548
		1549	return mlock (addr, length);
		1550	}
		1551
		1552	#endif
		1553
		1554	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1555	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1556	#else
		1557
		1558	static int
		1559	eio__msync (void *mem, size_t len, int flags)
		1560	{
		1561	eio_page_align (&mem, &len);
		1562
		1563	if (EIO_MS_ASYNC != MS_SYNC
		1564	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1565	|| EIO_MS_SYNC != MS_SYNC)
		1566	{
		1567	flags = 0
		1568	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1569	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1570	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1571	}
		1572
		1573	return msync (mem, len, flags);
		1574	}
		1575
		1576	#endif
		1577
		1578	static int
		1579	eio__mtouch (eio_req *req)
		1580	{
		1581	void *mem = req->ptr2;
		1582	size_t len = req->size;
		1583	int flags = req->int1;
		1584
		1585	eio_page_align (&mem, &len);
		1586
		1587	{
		1588	intptr_t addr = (intptr_t)mem;
		1589	intptr_t end = addr + len;
		1590	intptr_t page = eio_pagesize ();
		1591
1214	if (addr < end)	1592	if (addr < end)
1215	if (flags) /* modify */	1593	if (flags & EIO_MT_MODIFY) /* modify */
1216	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);	1594	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1217	else	1595	else
1218	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);	1596	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1597	}
1219		1598
1220	return 0;	1599	return 0;
1221	}	1600	}
1222		1601
1223	/*****************************************************************************/	1602	/*****************************************************************************/
…		…
1257	if (req)	1636	if (req)
1258	break;	1637	break;
1259		1638
1260	++idle;	1639	++idle;
1261		1640
1262	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1641	ts.tv_sec = time (0) + idle_timeout;
1263	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1642	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1264	{	1643	{
1265	if (idle > max_idle)	1644	if (idle > max_idle)
1266	{	1645	{
1267	--idle;	1646	--idle;
…		…
1310	return 0;	1689	return 0;
1311	}	1690	}
1312		1691
1313	/*****************************************************************************/	1692	/*****************************************************************************/
1314		1693
		1694	int ecb_cold
1315	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1695	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1316	{	1696	{
1317	return etp_init (want_poll, done_poll);	1697	return etp_init (want_poll, done_poll);
1318	}	1698	}
1319		1699
		1700	ECB_INLINE void
1320	static void eio_api_destroy (eio_req *req)	1701	eio_api_destroy (eio_req *req)
1321	{	1702	{
1322	free (req);	1703	free (req);
1323	}	1704	}
1324		1705
1325	#define REQ(rtype) \	1706	#define REQ(rtype) \
…		…
1344	{ \	1725	{ \
1345	eio_api_destroy (req); \	1726	eio_api_destroy (req); \
1346	return 0; \	1727	return 0; \
1347	}	1728	}
1348		1729
		1730	static void
1349	static void eio_execute (etp_worker *self, eio_req *req)	1731	eio_execute (etp_worker *self, eio_req *req)
1350	{	1732	{
1351	errno = 0;
1352
1353	switch (req->type)	1733	switch (req->type)
1354	{	1734	{
1355	case EIO_READ: ALLOC (req->size);	1735	case EIO_READ: ALLOC (req->size);
1356	req->result = req->offs >= 0	1736	req->result = req->offs >= 0
1357	? pread (req->int1, req->ptr2, req->size, req->offs)	1737	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1368	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1748	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1369	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1749	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1370	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1750	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1371	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1751	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1372		1752
		1753	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1754	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1755	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1756	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1757
1373	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1758	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1374	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1759	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1375	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	1760	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1376	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1761	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1377	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	1762	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1384	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1769	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1385	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	1770	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1386	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	1771	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1387	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	1772	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1388	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	1773	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1389	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	1774	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1390		1775
1391	case EIO_READLINK: ALLOC (NAME_MAX);	1776	case EIO_READLINK: ALLOC (PATH_MAX);
1392	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	1777	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1393		1778
1394	case EIO_SYNC: req->result = 0; sync (); break;	1779	case EIO_SYNC: req->result = 0; sync (); break;
1395	case EIO_FSYNC: req->result = fsync (req->int1); break;	1780	case EIO_FSYNC: req->result = fsync (req->int1); break;
1396	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1781	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1397	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	1782	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1783	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1398	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	1784	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1785	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1399	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	1786	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1400		1787
1401	case EIO_READDIR: eio__scandir (req, self); break;	1788	case EIO_READDIR: eio__scandir (req, self); break;
1402		1789
1403	case EIO_BUSY:	1790	case EIO_BUSY:
1404	#ifdef _WIN32	1791	#ifdef _WIN32
1405	Sleep (req->nv1 * 1000.);	1792	Sleep (req->nv1 * 1e3);
1406	#else	1793	#else
1407	{	1794	{
1408	struct timeval tv;	1795	struct timeval tv;
1409		1796
1410	tv.tv_sec = req->nv1;	1797	tv.tv_sec = req->nv1;
1411	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	1798	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1412		1799
1413	req->result = select (0, 0, 0, 0, &tv);	1800	req->result = select (0, 0, 0, 0, &tv);
1414	}	1801	}
1415	#endif	1802	#endif
1416	break;	1803	break;
…		…
1431	times = tv;	1818	times = tv;
1432	}	1819	}
1433	else	1820	else
1434	times = 0;	1821	times = 0;
1435		1822
1436
1437	req->result = req->type == EIO_FUTIME	1823	req->result = req->type == EIO_FUTIME
1438	? futimes (req->int1, times)	1824	? futimes (req->int1, times)
1439	: utimes (req->ptr1, times);	1825	: utimes (req->ptr1, times);
1440	}	1826	}
1441	break;	1827	break;
…		…
1446	case EIO_NOP:	1832	case EIO_NOP:
1447	req->result = 0;	1833	req->result = 0;
1448	break;	1834	break;
1449		1835
1450	case EIO_CUSTOM:	1836	case EIO_CUSTOM:
1451	((void (*)(eio_req *))req->feed) (req);	1837	req->feed (req);
1452	break;	1838	break;
1453		1839
1454	default:	1840	default:
		1841	errno = ENOSYS;
1455	req->result = -1;	1842	req->result = -1;
1456	break;	1843	break;
1457	}	1844	}
1458		1845
1459	req->errorno = errno;	1846	req->errorno = errno;
…		…
1489	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	1876	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1490	{	1877	{
1491	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	1878	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1492	}	1879	}
1493		1880
		1881	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		1882	{
		1883	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		1884	}
		1885
		1886	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		1887	{
		1888	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		1889	}
		1890
1494	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	1891	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1495	{	1892	{
1496	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	1893	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1497	}	1894	}
1498		1895
…		…
1522	}	1919	}
1523		1920
1524	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	1921	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1525	{	1922	{
1526	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1923	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1924	}
		1925
		1926	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		1927	{
		1928	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1527	}	1929	}
1528		1930
1529	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	1931	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1530	{	1932	{
1531	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1933	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1605	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2007	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1606	{	2008	{
1607	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2009	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1608	}	2010	}
1609		2011
		2012	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2013	{
		2014	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2015	}
		2016
1610	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2017	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1611	{	2018	{
1612	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2019	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1613	}	2020	}
1614		2021
…		…
1622	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2029	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1623	}	2030	}
1624		2031
1625	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2032	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1626	{	2033	{
1627	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2034	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1628	}	2035	}
1629		2036
1630	static eio_req *	2037	static eio_req *
1631	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2038	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1632	{	2039	{
…		…
1656	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2063	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1657	{	2064	{
1658	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2065	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1659	}	2066	}
1660		2067
1661	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2068	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1662	{	2069	{
1663	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2070	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1664	}	2071	}
1665		2072
1666	#endif	2073	#endif
1667		2074
1668	eio_req *eio_grp (eio_cb cb, void *data)	2075	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1677	#undef SEND	2084	#undef SEND
1678		2085
1679	/*****************************************************************************/	2086	/*****************************************************************************/
1680	/* grp functions */	2087	/* grp functions */
1681		2088
		2089	void
1682	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2090	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1683	{	2091	{
1684	grp->int2 = limit;	2092	grp->int2 = limit;
1685	grp->feed = feed;	2093	grp->feed = feed;
1686		2094
1687	grp_try_feed (grp);	2095	grp_try_feed (grp);
1688	}	2096	}
1689		2097
		2098	void
1690	void eio_grp_limit (eio_req *grp, int limit)	2099	eio_grp_limit (eio_req *grp, int limit)
1691	{	2100	{
1692	grp->int2 = limit;	2101	grp->int2 = limit;
1693		2102
1694	grp_try_feed (grp);	2103	grp_try_feed (grp);
1695	}	2104	}
1696		2105
		2106	void
1697	void eio_grp_add (eio_req *grp, eio_req *req)	2107	eio_grp_add (eio_req *grp, eio_req *req)
1698	{	2108	{
1699	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2109	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1700		2110
1701	grp->flags |= EIO_FLAG_GROUPADD;	2111	grp->flags |= EIO_FLAG_GROUPADD;
1702		2112
…		…
1713	}	2123	}
1714		2124
1715	/*****************************************************************************/	2125	/*****************************************************************************/
1716	/* misc garbage */	2126	/* misc garbage */
1717		2127
		2128	ssize_t
1718	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2129	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1719	{	2130	{
1720	etp_worker wrk;	2131	etp_worker wrk;
		2132	ssize_t ret;
1721		2133
1722	wrk.dbuf = 0;	2134	wrk.dbuf = 0;
1723		2135
1724	eio__sendfile (ofd, ifd, offset, count, &wrk);	2136	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1725		2137
1726	if (wrk.dbuf)	2138	if (wrk.dbuf)
1727	free (wrk.dbuf);	2139	free (wrk.dbuf);
1728	}
1729		2140
		2141	return ret;
		2142	}
		2143

Diff Legend

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