ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.31 by root, Sat Jun 6 17:25:13 2009 UTC vs.
Revision 1.74 by root, Mon Jun 20 07:28:15 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;
…		…
178	if (wrk->dirp) \	190	if (wrk->dirp) \
179	{ \	191	{ \
180	closedir (wrk->dirp); \	192	closedir (wrk->dirp); \
181	wrk->dirp = 0; \	193	wrk->dirp = 0; \
182	}	194	}
		195
183	#define ETP_WORKER_COMMON \	196	#define ETP_WORKER_COMMON \
184	void *dbuf; \	197	void *dbuf; \
185	DIR *dirp;	198	DIR *dirp;
186		199
187	/*****************************************************************************/	200	/*****************************************************************************/
188		201
189	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	202	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
190		203
191	/* 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
192	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	206	tvdiff (struct timeval *tv1, struct timeval *tv2)
193	{	207	{
194	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	208	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
195	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	209	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
196	}	210	}
197		211
…		…
204	static unsigned int max_poll_reqs; /* reslock */	218	static unsigned int max_poll_reqs; /* reslock */
205		219
206	static volatile unsigned int nreqs; /* reqlock */	220	static volatile unsigned int nreqs; /* reqlock */
207	static volatile unsigned int nready; /* reqlock */	221	static volatile unsigned int nready; /* reqlock */
208	static volatile unsigned int npending; /* reqlock */	222	static volatile unsigned int npending; /* reqlock */
209	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 */
210		225
211	static mutex_t wrklock = X_MUTEX_INIT;	226	static xmutex_t wrklock;
212	static mutex_t reslock = X_MUTEX_INIT;	227	static xmutex_t reslock;
213	static mutex_t reqlock = X_MUTEX_INIT;	228	static xmutex_t reqlock;
214	static cond_t reqwait = X_COND_INIT;	229	static xcond_t reqwait;
215		230
216	#if !HAVE_PREADWRITE	231	#if !HAVE_PREADWRITE
217	/*	232	/*
218	* make our pread/pwrite emulation safe against themselves, but not against	233	* make our pread/pwrite emulation safe against themselves, but not against
219	* 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,
220	* but that's your problem, not mine.	235	* but that's your problem, not mine.
221	*/	236	*/
222	static mutex_t preadwritelock = X_MUTEX_INIT;	237	static xmutex_t preadwritelock = X_MUTEX_INIT;
223	#endif	238	#endif
224		239
225	typedef struct etp_worker	240	typedef struct etp_worker
226	{	241	{
227	/* locked by wrklock */	242	/* locked by wrklock */
228	struct etp_worker *prev, *next;	243	struct etp_worker *prev, *next;
229		244
230	thread_t tid;	245	xthread_t tid;
231		246
232	/* locked by reslock, reqlock or wrklock */	247	/* locked by reslock, reqlock or wrklock */
233	ETP_REQ *req; /* currently processed request */	248	ETP_REQ *req; /* currently processed request */
234		249
235	ETP_WORKER_COMMON	250	ETP_WORKER_COMMON
…		…
240	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	255	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
241	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	256	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
242		257
243	/* worker threads management */	258	/* worker threads management */
244		259
		260	static void ecb_cold
245	static void etp_worker_clear (etp_worker *wrk)	261	etp_worker_clear (etp_worker *wrk)
246	{	262	{
247	ETP_WORKER_CLEAR (wrk);	263	ETP_WORKER_CLEAR (wrk);
248	}	264	}
249		265
		266	static void ecb_cold
250	static void etp_worker_free (etp_worker *wrk)	267	etp_worker_free (etp_worker *wrk)
251	{	268	{
252	wrk->next->prev = wrk->prev;	269	wrk->next->prev = wrk->prev;
253	wrk->prev->next = wrk->next;	270	wrk->prev->next = wrk->next;
254		271
255	free (wrk);	272	free (wrk);
256	}	273	}
257		274
258	static unsigned int etp_nreqs (void)	275	static unsigned int
		276	etp_nreqs (void)
259	{	277	{
260	int retval;	278	int retval;
261	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	279	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
262	retval = nreqs;	280	retval = nreqs;
263	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	281	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
264	return retval;	282	return retval;
265	}	283	}
266		284
267	static unsigned int etp_nready (void)	285	static unsigned int
		286	etp_nready (void)
268	{	287	{
269	unsigned int retval;	288	unsigned int retval;
270		289
271	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	290	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272	retval = nready;	291	retval = nready;
273	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	292	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274		293
275	return retval;	294	return retval;
276	}	295	}
277		296
278	static unsigned int etp_npending (void)	297	static unsigned int
		298	etp_npending (void)
279	{	299	{
280	unsigned int retval;	300	unsigned int retval;
281		301
282	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
283	retval = npending;	303	retval = npending;
284	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	304	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
285		305
286	return retval;	306	return retval;
287	}	307	}
288		308
289	static unsigned int etp_nthreads (void)	309	static unsigned int
		310	etp_nthreads (void)
290	{	311	{
291	unsigned int retval;	312	unsigned int retval;
292		313
293	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	314	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
294	retval = started;	315	retval = started;
…		…
308	} etp_reqq;	329	} etp_reqq;
309		330
310	static etp_reqq req_queue;	331	static etp_reqq req_queue;
311	static etp_reqq res_queue;	332	static etp_reqq res_queue;
312		333
		334	static int ecb_noinline
313	static int reqq_push (etp_reqq *q, ETP_REQ *req)	335	reqq_push (etp_reqq *q, ETP_REQ *req)
314	{	336	{
315	int pri = req->pri;	337	int pri = req->pri;
316	req->next = 0;	338	req->next = 0;
317		339
318	if (q->qe[pri])	340	if (q->qe[pri])
…		…
324	q->qe[pri] = q->qs[pri] = req;	346	q->qe[pri] = q->qs[pri] = req;
325		347
326	return q->size++;	348	return q->size++;
327	}	349	}
328		350
		351	static ETP_REQ * ecb_noinline
329	static ETP_REQ *reqq_shift (etp_reqq *q)	352	reqq_shift (etp_reqq *q)
330	{	353	{
331	int pri;	354	int pri;
332		355
333	if (!q->size)	356	if (!q->size)
334	return 0;	357	return 0;
…		…
349	}	372	}
350		373
351	abort ();	374	abort ();
352	}	375	}
353		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
354	static void etp_atfork_prepare (void)	387	etp_atfork_prepare (void)
355	{	388	{
356	X_LOCK (wrklock);	389	X_LOCK (wrklock);
357	X_LOCK (reqlock);	390	X_LOCK (reqlock);
358	X_LOCK (reslock);	391	X_LOCK (reslock);
359	#if !HAVE_PREADWRITE	392	#if !HAVE_PREADWRITE
360	X_LOCK (preadwritelock);	393	X_LOCK (preadwritelock);
361	#endif	394	#endif
362	}	395	}
363		396
		397	static void ecb_cold
364	static void etp_atfork_parent (void)	398	etp_atfork_parent (void)
365	{	399	{
366	#if !HAVE_PREADWRITE	400	#if !HAVE_PREADWRITE
367	X_UNLOCK (preadwritelock);	401	X_UNLOCK (preadwritelock);
368	#endif	402	#endif
369	X_UNLOCK (reslock);	403	X_UNLOCK (reslock);
370	X_UNLOCK (reqlock);	404	X_UNLOCK (reqlock);
371	X_UNLOCK (wrklock);	405	X_UNLOCK (wrklock);
372	}	406	}
373		407
		408	static void ecb_cold
374	static void etp_atfork_child (void)	409	etp_atfork_child (void)
375	{	410	{
376	ETP_REQ *prv;	411	ETP_REQ *prv;
377		412
378	while ((prv = reqq_shift (&req_queue)))	413	while ((prv = reqq_shift (&req_queue)))
379	ETP_DESTROY (prv);	414	ETP_DESTROY (prv);
…		…
396	idle = 0;	431	idle = 0;
397	nreqs = 0;	432	nreqs = 0;
398	nready = 0;	433	nready = 0;
399	npending = 0;	434	npending = 0;
400		435
401	etp_atfork_parent ();	436	etp_thread_init ();
402	}	437	}
403		438
404	static void	439	static void ecb_cold
405	etp_once_init (void)	440	etp_once_init (void)
406	{	441	{
		442	etp_thread_init ();
407	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);
408	}	444	}
409		445
410	static int	446	static int ecb_cold
411	etp_init (void (*want_poll)(void), void (*done_poll)(void))	447	etp_init (void (*want_poll)(void), void (*done_poll)(void))
412	{	448	{
413	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	449	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
414		450
415	pthread_once (&doinit, etp_once_init);	451	pthread_once (&doinit, etp_once_init);
…		…
420	return 0;	456	return 0;
421	}	457	}
422		458
423	X_THREAD_PROC (etp_proc);	459	X_THREAD_PROC (etp_proc);
424		460
		461	static void ecb_cold
425	static void etp_start_thread (void)	462	etp_start_thread (void)
426	{	463	{
427	etp_worker *wrk = calloc (1, sizeof (etp_worker));	464	etp_worker *wrk = calloc (1, sizeof (etp_worker));
428		465
429	/*TODO*/	466	/*TODO*/
430	assert (("unable to allocate worker thread data", wrk));	467	assert (("unable to allocate worker thread data", wrk));
…		…
443	free (wrk);	480	free (wrk);
444		481
445	X_UNLOCK (wrklock);	482	X_UNLOCK (wrklock);
446	}	483	}
447		484
		485	static void
448	static void etp_maybe_start_thread (void)	486	etp_maybe_start_thread (void)
449	{	487	{
450	if (expect_true (etp_nthreads () >= wanted))	488	if (ecb_expect_true (etp_nthreads () >= wanted))
451	return;	489	return;
452		490
453	/* todo: maybe use idle here, but might be less exact */	491	/* todo: maybe use idle here, but might be less exact */
454	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 ()))
455	return;	493	return;
456		494
457	etp_start_thread ();	495	etp_start_thread ();
458	}	496	}
459		497
		498	static void ecb_cold
460	static void etp_end_thread (void)	499	etp_end_thread (void)
461	{	500	{
462	eio_req *req = calloc (1, sizeof (eio_req));	501	eio_req *req = calloc (1, sizeof (eio_req));
463		502
464	req->type = -1;	503	req->type = -1;
465	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	504	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
472	X_LOCK (wrklock);	511	X_LOCK (wrklock);
473	--started;	512	--started;
474	X_UNLOCK (wrklock);	513	X_UNLOCK (wrklock);
475	}	514	}
476		515
477	static int etp_poll (void)	516	static int
		517	etp_poll (void)
478	{	518	{
479	unsigned int maxreqs;	519	unsigned int maxreqs;
480	unsigned int maxtime;	520	unsigned int maxtime;
481	struct timeval tv_start, tv_now;	521	struct timeval tv_start, tv_now;
482		522
…		…
512		552
513	X_LOCK (reqlock);	553	X_LOCK (reqlock);
514	--nreqs;	554	--nreqs;
515	X_UNLOCK (reqlock);	555	X_UNLOCK (reqlock);
516		556
517	if (expect_false (req->type == EIO_GROUP && req->size))	557	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
518	{	558	{
519	req->int1 = 1; /* mark request as delayed */	559	req->int1 = 1; /* mark request as delayed */
520	continue;	560	continue;
521	}	561	}
522	else	562	else
523	{	563	{
524	int res = ETP_FINISH (req);	564	int res = ETP_FINISH (req);
525	if (expect_false (res))	565	if (ecb_expect_false (res))
526	return res;	566	return res;
527	}	567	}
528		568
529	if (expect_false (maxreqs && !--maxreqs))	569	if (ecb_expect_false (maxreqs && !--maxreqs))
530	break;	570	break;
531		571
532	if (maxtime)	572	if (maxtime)
533	{	573	{
534	gettimeofday (&tv_now, 0);	574	gettimeofday (&tv_now, 0);
…		…
540		580
541	errno = EAGAIN;	581	errno = EAGAIN;
542	return -1;	582	return -1;
543	}	583	}
544		584
		585	static void
545	static void etp_cancel (ETP_REQ *req)	586	etp_cancel (ETP_REQ *req)
546	{	587	{
547	X_LOCK (wrklock);	588	X_LOCK (wrklock);
548	req->flags |= EIO_FLAG_CANCELLED;	589	req->flags |= EIO_FLAG_CANCELLED;
549	X_UNLOCK (wrklock);	590	X_UNLOCK (wrklock);
550		591
551	eio_grp_cancel (req);	592	eio_grp_cancel (req);
552	}	593	}
553		594
		595	static void
554	static void etp_submit (ETP_REQ *req)	596	etp_submit (ETP_REQ *req)
555	{	597	{
556	req->pri -= ETP_PRI_MIN;	598	req->pri -= ETP_PRI_MIN;
557		599
558	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;
559	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;
560		602
561	if (expect_false (req->type == EIO_GROUP))	603	if (ecb_expect_false (req->type == EIO_GROUP))
562	{	604	{
563	/* I hope this is worth it :/ */	605	/* I hope this is worth it :/ */
564	X_LOCK (reqlock);	606	X_LOCK (reqlock);
565	++nreqs;	607	++nreqs;
566	X_UNLOCK (reqlock);	608	X_UNLOCK (reqlock);
…		…
585		627
586	etp_maybe_start_thread ();	628	etp_maybe_start_thread ();
587	}	629	}
588	}	630	}
589		631
		632	static void ecb_cold
590	static void etp_set_max_poll_time (double nseconds)	633	etp_set_max_poll_time (double nseconds)
591	{	634	{
592	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	635	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
593	max_poll_time = nseconds;	636	max_poll_time = nseconds * EIO_TICKS;
594	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	637	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
595	}	638	}
596		639
		640	static void ecb_cold
597	static void etp_set_max_poll_reqs (unsigned int maxreqs)	641	etp_set_max_poll_reqs (unsigned int maxreqs)
598	{	642	{
599	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	643	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
600	max_poll_reqs = maxreqs;	644	max_poll_reqs = maxreqs;
601	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	645	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
602	}	646	}
603		647
		648	static void ecb_cold
604	static void etp_set_max_idle (unsigned int nthreads)	649	etp_set_max_idle (unsigned int nthreads)
605	{	650	{
606	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	651	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
607	max_idle = nthreads <= 0 ? 1 : nthreads;	652	max_idle = nthreads;
608	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	653	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
609	}	654	}
610		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
611	static void etp_set_min_parallel (unsigned int nthreads)	665	etp_set_min_parallel (unsigned int nthreads)
612	{	666	{
613	if (wanted < nthreads)	667	if (wanted < nthreads)
614	wanted = nthreads;	668	wanted = nthreads;
615	}	669	}
616		670
		671	static void ecb_cold
617	static void etp_set_max_parallel (unsigned int nthreads)	672	etp_set_max_parallel (unsigned int nthreads)
618	{	673	{
619	if (wanted > nthreads)	674	if (wanted > nthreads)
620	wanted = nthreads;	675	wanted = nthreads;
621		676
622	while (started > wanted)	677	while (started > wanted)
623	etp_end_thread ();	678	etp_end_thread ();
624	}	679	}
625		680
626	/*****************************************************************************/	681	/*****************************************************************************/
627		682
		683	static void
628	static void grp_try_feed (eio_req *grp)	684	grp_try_feed (eio_req *grp)
629	{	685	{
630	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	686	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
631	{	687	{
632	grp->flags &= ~EIO_FLAG_GROUPADD;	688	grp->flags &= ~EIO_FLAG_GROUPADD;
633		689
…		…
640	break;	696	break;
641	}	697	}
642	}	698	}
643	}	699	}
644		700
		701	static int
645	static int grp_dec (eio_req *grp)	702	grp_dec (eio_req *grp)
646	{	703	{
647	--grp->size;	704	--grp->size;
648		705
649	/* call feeder, if applicable */	706	/* call feeder, if applicable */
650	grp_try_feed (grp);	707	grp_try_feed (grp);
…		…
654	return eio_finish (grp);	711	return eio_finish (grp);
655	else	712	else
656	return 0;	713	return 0;
657	}	714	}
658		715
		716	void
659	void eio_destroy (eio_req *req)	717	eio_destroy (eio_req *req)
660	{	718	{
661	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	719	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
662	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	720	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
663		721
664	EIO_DESTROY (req);	722	EIO_DESTROY (req);
665	}	723	}
666		724
		725	static int
667	static int eio_finish (eio_req *req)	726	eio_finish (eio_req *req)
668	{	727	{
669	int res = EIO_FINISH (req);	728	int res = EIO_FINISH (req);
670		729
671	if (req->grp)	730	if (req->grp)
672	{	731	{
…		…
689	eio_destroy (req);	748	eio_destroy (req);
690		749
691	return res;	750	return res;
692	}	751	}
693		752
		753	void
694	void eio_grp_cancel (eio_req *grp)	754	eio_grp_cancel (eio_req *grp)
695	{	755	{
696	for (grp = grp->grp_first; grp; grp = grp->grp_next)	756	for (grp = grp->grp_first; grp; grp = grp->grp_next)
697	eio_cancel (grp);	757	eio_cancel (grp);
698	}	758	}
699		759
		760	void
700	void eio_cancel (eio_req *req)	761	eio_cancel (eio_req *req)
701	{	762	{
702	etp_cancel (req);	763	etp_cancel (req);
703	}	764	}
704		765
		766	void
705	void eio_submit (eio_req *req)	767	eio_submit (eio_req *req)
706	{	768	{
707	etp_submit (req);	769	etp_submit (req);
708	}	770	}
709		771
710	unsigned int eio_nreqs (void)	772	unsigned int
		773	eio_nreqs (void)
711	{	774	{
712	return etp_nreqs ();	775	return etp_nreqs ();
713	}	776	}
714		777
715	unsigned int eio_nready (void)	778	unsigned int
		779	eio_nready (void)
716	{	780	{
717	return etp_nready ();	781	return etp_nready ();
718	}	782	}
719		783
720	unsigned int eio_npending (void)	784	unsigned int
		785	eio_npending (void)
721	{	786	{
722	return etp_npending ();	787	return etp_npending ();
723	}	788	}
724		789
725	unsigned int eio_nthreads (void)	790	unsigned int ecb_cold
		791	eio_nthreads (void)
726	{	792	{
727	return etp_nthreads ();	793	return etp_nthreads ();
728	}	794	}
729		795
		796	void ecb_cold
730	void eio_set_max_poll_time (double nseconds)	797	eio_set_max_poll_time (double nseconds)
731	{	798	{
732	etp_set_max_poll_time (nseconds);	799	etp_set_max_poll_time (nseconds);
733	}	800	}
734		801
		802	void ecb_cold
735	void eio_set_max_poll_reqs (unsigned int maxreqs)	803	eio_set_max_poll_reqs (unsigned int maxreqs)
736	{	804	{
737	etp_set_max_poll_reqs (maxreqs);	805	etp_set_max_poll_reqs (maxreqs);
738	}	806	}
739		807
		808	void ecb_cold
740	void eio_set_max_idle (unsigned int nthreads)	809	eio_set_max_idle (unsigned int nthreads)
741	{	810	{
742	etp_set_max_idle (nthreads);	811	etp_set_max_idle (nthreads);
743	}	812	}
744		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
745	void eio_set_min_parallel (unsigned int nthreads)	821	eio_set_min_parallel (unsigned int nthreads)
746	{	822	{
747	etp_set_min_parallel (nthreads);	823	etp_set_min_parallel (nthreads);
748	}	824	}
749		825
		826	void ecb_cold
750	void eio_set_max_parallel (unsigned int nthreads)	827	eio_set_max_parallel (unsigned int nthreads)
751	{	828	{
752	etp_set_max_parallel (nthreads);	829	etp_set_max_parallel (nthreads);
753	}	830	}
754		831
755	int eio_poll (void)	832	int eio_poll (void)
…		…
797		874
798	return res;	875	return res;
799	}	876	}
800	#endif	877	#endif
801		878
802	#ifndef HAVE_FUTIMES	879	#ifndef HAVE_UTIMES
803		880
804	# undef utimes	881	# undef utimes
805	# undef futimes
806	# define utimes(path,times) eio__utimes (path, times)	882	# define utimes(path,times) eio__utimes (path, times)
807	# define futimes(fd,times) eio__futimes (fd, times)
808		883
809	static int	884	static int
810	eio__utimes (const char *filename, const struct timeval times[2])	885	eio__utimes (const char *filename, const struct timeval times[2])
811	{	886	{
812	if (times)	887	if (times)
…		…
820	}	895	}
821	else	896	else
822	return utime (filename, 0);	897	return utime (filename, 0);
823	}	898	}
824		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
825	static int eio__futimes (int fd, const struct timeval tv[2])	908	eio__futimes (int fd, const struct timeval tv[2])
826	{	909	{
827	errno = ENOSYS;	910	errno = ENOSYS;
828	return -1;	911	return -1;
829	}	912	}
830		913
…		…
834	# undef fdatasync	917	# undef fdatasync
835	# define fdatasync(fd) fsync (fd)	918	# define fdatasync(fd) fsync (fd)
836	#endif	919	#endif
837		920
838	/* sync_file_range always needs emulation */	921	/* sync_file_range always needs emulation */
839	int	922	static int
840	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)
841	{	924	{
842	#if HAVE_SYNC_FILE_RANGE	925	#if HAVE_SYNC_FILE_RANGE
843	int res;	926	int res;
844		927
…		…
857	if (!res || errno != ENOSYS)	940	if (!res || errno != ENOSYS)
858	return res;	941	return res;
859	#endif	942	#endif
860		943
861	/* 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
862	* call fdatasync, as thta matches the expectation of it's users best */	945	* call fdatasync, as that matches the expectation of its users best */
863	return fdatasync (fd);	946	return fdatasync (fd);
864	}	947	}
865		948
866	#if !HAVE_READAHEAD	949	#if !HAVE_READAHEAD
867	# undef readahead	950	# undef readahead
…		…
890		973
891	/* sendfile always needs emulation */	974	/* sendfile always needs emulation */
892	static ssize_t	975	static ssize_t
893	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)
894	{	977	{
		978	ssize_t written = 0;
895	ssize_t res;	979	ssize_t res;
896		980
897	if (!count)	981	if (!count)
898	return 0;	982	return 0;
899		983
		984	for (;;)
		985	{
900	#if HAVE_SENDFILE	986	#if HAVE_SENDFILE
901	# if __linux	987	# if __linux
		988	off_t soffset = offset;
902	res = sendfile (ofd, ifd, &offset, count);	989	res = sendfile (ofd, ifd, &soffset, count);
903		990
904	# elif __freebsd	991	# elif __FreeBSD__
905	/*	992	/*
906	* 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
907	* wasted on making it similar to other I/O functions.	994	* wasted on making it similar to other I/O functions.
908	*/	995	*/
909	{
910	off_t sbytes;	996	off_t sbytes;
911	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	997	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
912		998
913	if (res < 0 && sbytes)	999	#if 0 /* according to the manpage, this is correct, but broken behaviour */
914	/* 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)
915	res = sbytes;	1004	res = sbytes;
916	}	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;
917		1018
918	# elif __hpux	1019	# elif __hpux
919	res = sendfile (ofd, ifd, offset, count, 0, 0);	1020	res = sendfile (ofd, ifd, offset, count, 0, 0);
920		1021
921	# elif __solaris	1022	# elif __solaris
922	{
923	struct sendfilevec vec;	1023	struct sendfilevec vec;
924	size_t sbytes;	1024	size_t sbytes;
925		1025
926	vec.sfv_fd = ifd;	1026	vec.sfv_fd = ifd;
927	vec.sfv_flag = 0;	1027	vec.sfv_flag = 0;
928	vec.sfv_off = offset;	1028	vec.sfv_off = offset;
929	vec.sfv_len = count;	1029	vec.sfv_len = count;
930		1030
931	res = sendfilev (ofd, &vec, 1, &sbytes);	1031	res = sendfilev (ofd, &vec, 1, &sbytes);
932		1032
933	if (res < 0 && sbytes)	1033	if (res < 0 && sbytes)
934	res = sbytes;	1034	res = sbytes;
935	}
936		1035
937	# 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
938	#else	1047	#else
939	res = -1;	1048	res = -1;
940	errno = ENOSYS;	1049	errno = ENOSYS;
941	#endif	1050	#endif
942		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
943	if (res < 0	1075	if (res < 0
944	&& (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 */
945	#if __solaris	1082	#if __solaris
946	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1083	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
947	#endif	1084	#endif
948	)	1085	)
949	)	1086	)
…		…
980	}	1117	}
981		1118
982	return res;	1119	return res;
983	}	1120	}
984		1121
985	static int	1122	static signed char
986	eio_dent_cmp (const void *a_, const void *b_)	1123	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
987	{	1124	{
988	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 */
989	const eio_dirent *b = (const eio_dirent *)b_;	1126	: a->inode < b->inode ? -1
		1127	: a->inode > b->inode ? 1
		1128	: 0;
		1129	}
990		1130
991	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
992	: 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);
993	}	1285	}
994		1286
995	/* read a full directory */	1287	/* read a full directory */
996	static void	1288	static void
997	eio__scandir (eio_req *req, etp_worker *self)	1289	eio__scandir (eio_req *req, etp_worker *self)
998	{	1290	{
999	DIR *dirp;	1291	DIR *dirp;
1000	EIO_STRUCT_DIRENT *entp;	1292	EIO_STRUCT_DIRENT *entp;
1001	unsigned char *name, *names;	1293	char *name, *names;
1002	int namesalloc = 4096;	1294	int namesalloc = 4096;
1003	int namesoffs = 0;	1295	int namesoffs = 0;
1004	int flags = req->int1;	1296	int flags = req->int1;
1005	eio_dirent *dents = 0;	1297	eio_dirent *dents = 0;
1006	int dentalloc = 128;	1298	int dentalloc = 128;
1007	int dentoffs = 0;	1299	int dentoffs = 0;
		1300	ino_t inode_bits = 0;
1008		1301
1009	req->result = -1;	1302	req->result = -1;
1010		1303
1011	if (!(flags & EIO_READDIR_DENTS))	1304	if (!(flags & EIO_READDIR_DENTS))
1012	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1305	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1013		1306
1014	X_LOCK (wrklock);	1307	X_LOCK (wrklock);
1015
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
		1311	if (req->flags & EIO_FLAG_PTR1_FREE)
		1312	free (req->ptr1);
		1313
1018	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1314	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;	1315	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1021		1316	req->ptr2 = names = malloc (namesalloc);
1022	X_UNLOCK (wrklock);	1317	X_UNLOCK (wrklock);
1023		1318
1024	if (dirp && names && (!flags || dents))	1319	if (dirp && names && (!flags || dents))
1025	for (;;)	1320	for (;;)
1026	{	1321	{
…		…
1034		1329
1035	/* sort etc. */	1330	/* sort etc. */
1036	req->int1 = flags;	1331	req->int1 = flags;
1037	req->result = dentoffs;	1332	req->result = dentoffs;
1038		1333
1039	if (flags & EIO_READDIR_STAT_ORDER || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))	1334	if (flags & EIO_READDIR_STAT_ORDER)
		1335	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1336	else if (flags & EIO_READDIR_DIRS_FIRST)
		1337	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1338	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1339	else
1040	{	1340	{
1041	/* pray your qsort doesn't use quicksort */
1042	qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1043	}
1044	else if (flags & EIO_READDIR_DIRS_FIRST && !(flags & EIO_READDIR_FOUND_UNKNOWN))
1045	{
1046	/* in this case, all is known, and we just put dirs first and sort them */	1341	/* in this case, all is known, and we just put dirs first and sort them */
1047	eio_dirent *ent = dents + dentoffs;	1342	eio_dirent *oth = dents + dentoffs;
1048	eio_dirent *dir = dents;	1343	eio_dirent *dir = dents;
1049		1344
		1345	/* now partition dirs to the front, and non-dirs to the back */
		1346	/* by walking from both sides and swapping if necessary */
1050	while (ent > dir)	1347	while (oth > dir)
1051	{	1348	{
1052	if (dir->type == DT_DIR)	1349	if (dir->type == EIO_DT_DIR)
1053	++dir;	1350	++dir;
1054	else
1055	{
1056	--ent;
1057
1058	if (ent->type == DT_DIR)	1351	else if ((--oth)->type == EIO_DT_DIR)
1059	{	1352	{
1060	eio_dirent tmp = *dir;	1353	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1061	*dir = *ent;
1062	*ent = tmp;
1063		1354
1064	++dir;	1355	++dir;
1065	}	1356	}
1066	}	1357	}
		1358
		1359	/* now sort the dirs only (dirs all have the same score) */
		1360	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1067	}	1361	}
1068
1069	/* now sort the dirs only */
1070	qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1071	}
1072
1073	{int i; for(i=0;i<dentoffs;++i){eio_dirent *e=dents+i; printf ("%9ld %3d %s\n", e->inode,e->score,e->name);}}//D
1074		1362
1075	break;	1363	break;
1076	}	1364	}
1077		1365
		1366	/* now add the entry to our list(s) */
1078	name = entp->d_name;	1367	name = entp->d_name;
1079		1368
		1369	/* skip . and .. entries */
1080	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1370	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1081	{	1371	{
1082	int len = strlen (name) + 1;	1372	int len = D_NAMLEN (entp) + 1;
1083		1373
1084	while (expect_false (namesoffs + len > namesalloc))	1374	while (ecb_expect_false (namesoffs + len > namesalloc))
1085	{	1375	{
1086	namesalloc *= 2;	1376	namesalloc *= 2;
1087	X_LOCK (wrklock);	1377	X_LOCK (wrklock);
1088	req->ptr1 = names = realloc (names, namesalloc);	1378	req->ptr2 = names = realloc (names, namesalloc);
1089	X_UNLOCK (wrklock);	1379	X_UNLOCK (wrklock);
1090		1380
1091	if (!names)	1381	if (!names)
1092	break;	1382	break;
1093	}	1383	}
…		…
1096		1386
1097	if (dents)	1387	if (dents)
1098	{	1388	{
1099	struct eio_dirent *ent;	1389	struct eio_dirent *ent;
1100		1390
1101	if (expect_false (dentoffs == dentalloc))	1391	if (ecb_expect_false (dentoffs == dentalloc))
1102	{	1392	{
1103	dentalloc *= 2;	1393	dentalloc *= 2;
1104	X_LOCK (wrklock);	1394	X_LOCK (wrklock);
1105	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1395	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1106	X_UNLOCK (wrklock);	1396	X_UNLOCK (wrklock);
1107		1397
1108	if (!dents)	1398	if (!dents)
1109	break;	1399	break;
1110	}	1400	}
1111		1401
1112	ent = dents + dentoffs;	1402	ent = dents + dentoffs;
1113		1403
1114	ent->name = names + namesoffs;	1404	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1115	ent->namelen = len - 1;	1405	ent->namelen = len - 1;
1116	ent->inode = D_INO (entp);	1406	ent->inode = D_INO (entp);
		1407
		1408	inode_bits |= ent->inode;
1117		1409
1118	switch (D_TYPE (entp))	1410	switch (D_TYPE (entp))
1119	{	1411	{
1120	default:	1412	default:
1121	ent->type = EIO_DT_UNKNOWN;	1413	ent->type = EIO_DT_UNKNOWN;
…		…
1126	case DT_FIFO: ent->type = EIO_DT_FIFO; break;	1418	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1127	#endif	1419	#endif
1128	#ifdef DT_CHR	1420	#ifdef DT_CHR
1129	case DT_CHR: ent->type = EIO_DT_CHR; break;	1421	case DT_CHR: ent->type = EIO_DT_CHR; break;
1130	#endif	1422	#endif
		1423	#ifdef DT_MPC
		1424	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1425	#endif
1131	#ifdef DT_DIR	1426	#ifdef DT_DIR
1132	case DT_DIR: ent->type = EIO_DT_DIR; break;	1427	case DT_DIR: ent->type = EIO_DT_DIR; break;
1133	#endif	1428	#endif
		1429	#ifdef DT_NAM
		1430	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1431	#endif
1134	#ifdef DT_BLK	1432	#ifdef DT_BLK
1135	case DT_BLK: ent->type = EIO_DT_BLK; break;	1433	case DT_BLK: ent->type = EIO_DT_BLK; break;
1136	#endif	1434	#endif
		1435	#ifdef DT_MPB
		1436	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1437	#endif
1137	#ifdef DT_REG	1438	#ifdef DT_REG
1138	case DT_REG: ent->type = EIO_DT_REG; break;	1439	case DT_REG: ent->type = EIO_DT_REG; break;
1139	#endif	1440	#endif
		1441	#ifdef DT_NWK
		1442	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1443	#endif
		1444	#ifdef DT_CMP
		1445	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1446	#endif
1140	#ifdef DT_LNK	1447	#ifdef DT_LNK
1141	case DT_LNK: ent->type = EIO_DT_LNK; break;	1448	case DT_LNK: ent->type = EIO_DT_LNK; break;
1142	#endif	1449	#endif
1143	#ifdef DT_SOCK	1450	#ifdef DT_SOCK
1144	case DT_SOCK: ent->type = EIO_DT_SOCK; break;	1451	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1145	#endif	1452	#endif
		1453	#ifdef DT_DOOR
		1454	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1455	#endif
1146	#ifdef DT_WHT	1456	#ifdef DT_WHT
1147	case DT_WHT: ent->type = EIO_DT_WHT; break;	1457	case DT_WHT: ent->type = EIO_DT_WHT; break;
1148	#endif	1458	#endif
1149	}	1459	}
1150		1460
1151	ent->score = 0;	1461	ent->score = 7;
1152		1462
1153	if (flags & EIO_READDIR_DIRS_FIRST)	1463	if (flags & EIO_READDIR_DIRS_FIRST)
1154	{	1464	{
1155	if (ent->type == EIO_DT_UNKNOWN)	1465	if (ent->type == EIO_DT_UNKNOWN)
1156	{	1466	{
1157	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1467	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1158	ent->score = 98;	1468	ent->score = 1;
1159	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1469	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1160	ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */	1470	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1161	}	1471	}
1162	else if (ent->type == DT_DIR)	1472	else if (ent->type == EIO_DT_DIR)
1163	ent->score = 100;	1473	ent->score = 0;
1164	}	1474	}
1165	}	1475	}
1166		1476
1167	namesoffs += len;	1477	namesoffs += len;
1168	++dentoffs;	1478	++dentoffs;
1169	}	1479	}
		1480
		1481	if (EIO_CANCELLED (req))
		1482	{
		1483	errno = ECANCELED;
		1484	break;
		1485	}
1170	}	1486	}
1171	else
1172	req->result = -1;
1173
1174	/* if user doesn't want the dents, do not provide it */
1175	if (!(flags & EIO_READDIR_DENTS))
1176	{
1177	X_LOCK (wrklock);
1178	free (dents);
1179	req->ptr2 = req->ptr1;
1180	req->ptr1 = 0;
1181	X_UNLOCK (wrklock);
1182	}
1183	}	1487	}
1184		1488
1185	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1186	# undef msync
1187	# define msync(a,b,c) ((errno = ENOSYS), -1)
1188	#endif
1189
1190	int
1191	eio__mtouch (void *mem, size_t len, int flags)
1192	{
1193	intptr_t addr = (intptr_t)mem;
1194	intptr_t end = addr + len;
1195	#ifdef PAGESIZE	1489	#ifdef PAGESIZE
1196	const intptr_t page = PAGESIZE;	1490	# define eio_pagesize() PAGESIZE
1197	#else	1491	#else
		1492	static intptr_t
		1493	eio_pagesize (void)
		1494	{
1198	static intptr_t page;	1495	static intptr_t page;
1199		1496
1200	if (!page)	1497	if (!page)
1201	page = sysconf (_SC_PAGESIZE);	1498	page = sysconf (_SC_PAGESIZE);
		1499
		1500	return page;
		1501	}
		1502	#endif
		1503
		1504	static void
		1505	eio_page_align (void **addr, size_t *length)
		1506	{
		1507	intptr_t mask = eio_pagesize () - 1;
		1508
		1509	/* round down addr */
		1510	intptr_t adj = mask & (intptr_t)*addr;
		1511
		1512	*addr = (void *)((intptr_t)*addr - adj);
		1513	*length += adj;
		1514
		1515	/* round up length */
		1516	*length = (*length + mask) & ~mask;
		1517	}
		1518
		1519	#if !_POSIX_MEMLOCK
		1520	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1521	#else
		1522
		1523	static int
		1524	eio__mlockall (int flags)
		1525	{
		1526	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1527	extern int mallopt (int, int);
		1528	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1202	#endif	1529	#endif
1203		1530
1204	addr &= ~(page - 1); /* assume page size is always a power of two */	1531	if (EIO_MCL_CURRENT != MCL_CURRENT
		1532	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1533	{
		1534	flags = 0
		1535	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1536	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1537	}
1205		1538
		1539	return mlockall (flags);
		1540	}
		1541	#endif
		1542
		1543	#if !_POSIX_MEMLOCK_RANGE
		1544	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1545	#else
		1546
		1547	static int
		1548	eio__mlock (void *addr, size_t length)
		1549	{
		1550	eio_page_align (&addr, &length);
		1551
		1552	return mlock (addr, length);
		1553	}
		1554
		1555	#endif
		1556
		1557	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1558	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1559	#else
		1560
		1561	static int
		1562	eio__msync (void *mem, size_t len, int flags)
		1563	{
		1564	eio_page_align (&mem, &len);
		1565
		1566	if (EIO_MS_ASYNC != MS_SYNC
		1567	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1568	|| EIO_MS_SYNC != MS_SYNC)
		1569	{
		1570	flags = 0
		1571	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1572	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1573	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1574	}
		1575
		1576	return msync (mem, len, flags);
		1577	}
		1578
		1579	#endif
		1580
		1581	static int
		1582	eio__mtouch (eio_req *req)
		1583	{
		1584	void *mem = req->ptr2;
		1585	size_t len = req->size;
		1586	int flags = req->int1;
		1587
		1588	eio_page_align (&mem, &len);
		1589
		1590	{
		1591	intptr_t addr = (intptr_t)mem;
		1592	intptr_t end = addr + len;
		1593	intptr_t page = eio_pagesize ();
		1594
1206	if (addr < end)	1595	if (addr < end)
1207	if (flags) /* modify */	1596	if (flags & EIO_MT_MODIFY) /* modify */
1208	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);	1597	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1209	else	1598	else
1210	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);	1599	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1600	}
1211		1601
1212	return 0;	1602	return 0;
1213	}	1603	}
1214		1604
1215	/*****************************************************************************/	1605	/*****************************************************************************/
…		…
1249	if (req)	1639	if (req)
1250	break;	1640	break;
1251		1641
1252	++idle;	1642	++idle;
1253		1643
1254	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1644	ts.tv_sec = time (0) + idle_timeout;
1255	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1645	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1256	{	1646	{
1257	if (idle > max_idle)	1647	if (idle > max_idle)
1258	{	1648	{
1259	--idle;	1649	--idle;
…		…
1302	return 0;	1692	return 0;
1303	}	1693	}
1304		1694
1305	/*****************************************************************************/	1695	/*****************************************************************************/
1306		1696
		1697	int ecb_cold
1307	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1698	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1308	{	1699	{
1309	return etp_init (want_poll, done_poll);	1700	return etp_init (want_poll, done_poll);
1310	}	1701	}
1311		1702
		1703	ecb_inline void
1312	static void eio_api_destroy (eio_req *req)	1704	eio_api_destroy (eio_req *req)
1313	{	1705	{
1314	free (req);	1706	free (req);
1315	}	1707	}
1316		1708
1317	#define REQ(rtype) \	1709	#define REQ(rtype) \
…		…
1336	{ \	1728	{ \
1337	eio_api_destroy (req); \	1729	eio_api_destroy (req); \
1338	return 0; \	1730	return 0; \
1339	}	1731	}
1340		1732
		1733	static void
1341	static void eio_execute (etp_worker *self, eio_req *req)	1734	eio_execute (etp_worker *self, eio_req *req)
1342	{	1735	{
1343	errno = 0;
1344
1345	switch (req->type)	1736	switch (req->type)
1346	{	1737	{
1347	case EIO_READ: ALLOC (req->size);	1738	case EIO_READ: ALLOC (req->size);
1348	req->result = req->offs >= 0	1739	req->result = req->offs >= 0
1349	? pread (req->int1, req->ptr2, req->size, req->offs)	1740	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1360	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1751	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1361	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1752	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1362	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1753	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1363	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1754	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1364		1755
		1756	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1757	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1758	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1759	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1760
1365	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1761	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1366	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1762	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1367	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	1763	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1368	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1764	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1369	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	1765	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1376	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1772	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1377	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	1773	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1378	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	1774	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1379	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	1775	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1380	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	1776	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1381	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	1777	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1382		1778
1383	case EIO_READLINK: ALLOC (NAME_MAX);	1779	case EIO_READLINK: ALLOC (PATH_MAX);
1384	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	1780	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1385		1781
1386	case EIO_SYNC: req->result = 0; sync (); break;	1782	case EIO_SYNC: req->result = 0; sync (); break;
1387	case EIO_FSYNC: req->result = fsync (req->int1); break;	1783	case EIO_FSYNC: req->result = fsync (req->int1); break;
1388	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1784	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1389	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	1785	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1786	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1390	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	1787	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1788	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1391	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	1789	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1392		1790
1393	case EIO_READDIR: eio__scandir (req, self); break;	1791	case EIO_READDIR: eio__scandir (req, self); break;
1394		1792
1395	case EIO_BUSY:	1793	case EIO_BUSY:
1396	#ifdef _WIN32	1794	#ifdef _WIN32
1397	Sleep (req->nv1 * 1000.);	1795	Sleep (req->nv1 * 1e3);
1398	#else	1796	#else
1399	{	1797	{
1400	struct timeval tv;	1798	struct timeval tv;
1401		1799
1402	tv.tv_sec = req->nv1;	1800	tv.tv_sec = req->nv1;
1403	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	1801	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1404		1802
1405	req->result = select (0, 0, 0, 0, &tv);	1803	req->result = select (0, 0, 0, 0, &tv);
1406	}	1804	}
1407	#endif	1805	#endif
1408	break;	1806	break;
…		…
1423	times = tv;	1821	times = tv;
1424	}	1822	}
1425	else	1823	else
1426	times = 0;	1824	times = 0;
1427		1825
1428
1429	req->result = req->type == EIO_FUTIME	1826	req->result = req->type == EIO_FUTIME
1430	? futimes (req->int1, times)	1827	? futimes (req->int1, times)
1431	: utimes (req->ptr1, times);	1828	: utimes (req->ptr1, times);
1432	}	1829	}
1433	break;	1830	break;
…		…
1438	case EIO_NOP:	1835	case EIO_NOP:
1439	req->result = 0;	1836	req->result = 0;
1440	break;	1837	break;
1441		1838
1442	case EIO_CUSTOM:	1839	case EIO_CUSTOM:
1443	((void (*)(eio_req *))req->feed) (req);	1840	req->feed (req);
1444	break;	1841	break;
1445		1842
1446	default:	1843	default:
		1844	errno = ENOSYS;
1447	req->result = -1;	1845	req->result = -1;
1448	break;	1846	break;
1449	}	1847	}
1450		1848
1451	req->errorno = errno;	1849	req->errorno = errno;
…		…
1481	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	1879	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1482	{	1880	{
1483	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	1881	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1484	}	1882	}
1485		1883
		1884	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		1885	{
		1886	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		1887	}
		1888
		1889	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		1890	{
		1891	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		1892	}
		1893
1486	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	1894	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1487	{	1895	{
1488	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	1896	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1489	}	1897	}
1490		1898
…		…
1514	}	1922	}
1515		1923
1516	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	1924	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1517	{	1925	{
1518	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1926	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1927	}
		1928
		1929	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		1930	{
		1931	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1519	}	1932	}
1520		1933
1521	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	1934	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1522	{	1935	{
1523	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1936	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1597	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2010	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1598	{	2011	{
1599	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2012	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1600	}	2013	}
1601		2014
		2015	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2016	{
		2017	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2018	}
		2019
1602	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2020	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1603	{	2021	{
1604	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2022	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1605	}	2023	}
1606		2024
…		…
1614	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2032	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1615	}	2033	}
1616		2034
1617	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2035	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1618	{	2036	{
1619	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2037	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1620	}	2038	}
1621		2039
1622	static eio_req *	2040	static eio_req *
1623	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2041	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1624	{	2042	{
…		…
1648	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2066	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1649	{	2067	{
1650	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2068	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1651	}	2069	}
1652		2070
1653	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2071	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1654	{	2072	{
1655	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2073	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1656	}	2074	}
1657		2075
1658	#endif	2076	#endif
1659		2077
1660	eio_req *eio_grp (eio_cb cb, void *data)	2078	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1669	#undef SEND	2087	#undef SEND
1670		2088
1671	/*****************************************************************************/	2089	/*****************************************************************************/
1672	/* grp functions */	2090	/* grp functions */
1673		2091
		2092	void
1674	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2093	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1675	{	2094	{
1676	grp->int2 = limit;	2095	grp->int2 = limit;
1677	grp->feed = feed;	2096	grp->feed = feed;
1678		2097
1679	grp_try_feed (grp);	2098	grp_try_feed (grp);
1680	}	2099	}
1681		2100
		2101	void
1682	void eio_grp_limit (eio_req *grp, int limit)	2102	eio_grp_limit (eio_req *grp, int limit)
1683	{	2103	{
1684	grp->int2 = limit;	2104	grp->int2 = limit;
1685		2105
1686	grp_try_feed (grp);	2106	grp_try_feed (grp);
1687	}	2107	}
1688		2108
		2109	void
1689	void eio_grp_add (eio_req *grp, eio_req *req)	2110	eio_grp_add (eio_req *grp, eio_req *req)
1690	{	2111	{
1691	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2112	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1692		2113
1693	grp->flags |= EIO_FLAG_GROUPADD;	2114	grp->flags |= EIO_FLAG_GROUPADD;
1694		2115
…		…
1705	}	2126	}
1706		2127
1707	/*****************************************************************************/	2128	/*****************************************************************************/
1708	/* misc garbage */	2129	/* misc garbage */
1709		2130
		2131	ssize_t
1710	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2132	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1711	{	2133	{
1712	etp_worker wrk;	2134	etp_worker wrk;
		2135	ssize_t ret;
1713		2136
1714	wrk.dbuf = 0;	2137	wrk.dbuf = 0;
1715		2138
1716	eio__sendfile (ofd, ifd, offset, count, &wrk);	2139	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1717		2140
1718	if (wrk.dbuf)	2141	if (wrk.dbuf)
1719	free (wrk.dbuf);	2142	free (wrk.dbuf);
1720	}
1721		2143
		2144	return ret;
		2145	}
		2146

Diff Legend

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