ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.19 by root, Tue Jun 17 23:34:20 2008 UTC vs.
Revision 1.76 by root, Tue Jul 5 09:24:12 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5	* All rights reserved.	5	* All rights reserved.
6	*	6	*
7	* Redistribution and use in source and binary forms, with or without modifica-	7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:	8	* tion, are permitted provided that the following conditions are met:
9	*	9	*
…		…
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"
		46
		47	#ifdef EIO_STACKSIZE
		48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		49	#endif
41	#include "xthread.h"	50	#include "xthread.h"
42		51
43	#include <errno.h>	52	#include <errno.h>
44	#include <stddef.h>	53	#include <stddef.h>
45	#include <stdlib.h>	54	#include <stdlib.h>
46	#include <string.h>	55	#include <string.h>
47	#include <errno.h>	56	#include <errno.h>
48	#include <sys/types.h>	57	#include <sys/types.h>
49	#include <sys/stat.h>	58	#include <sys/stat.h>
		59	#include <sys/statvfs.h>
50	#include <limits.h>	60	#include <limits.h>
51	#include <fcntl.h>	61	#include <fcntl.h>
52	#include <assert.h>	62	#include <assert.h>
53		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
54	#ifndef EIO_FINISH	70	#ifndef EIO_FINISH
55	# 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
56	#endif	72	#endif
57		73
58	#ifndef EIO_DESTROY	74	#ifndef EIO_DESTROY
…		…
64	#endif	80	#endif
65		81
66	#ifdef _WIN32	82	#ifdef _WIN32
67		83
68	/*doh*/	84	/*doh*/
69
70	#else	85	#else
71		86
72	# include "config.h"
73	# include <sys/time.h>	87	# include <sys/time.h>
74	# include <sys/select.h>	88	# include <sys/select.h>
75	# include <unistd.h>	89	# include <unistd.h>
76	# include <utime.h>	90	# include <utime.h>
77	# include <signal.h>	91	# include <signal.h>
78	# include <dirent.h>	92	# include <dirent.h>
79		93
		94	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		95	# include <sys/mman.h>
		96	#endif
		97
		98	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		99	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		100	# define _DIRENT_HAVE_D_TYPE /* sigh */
		101	# define D_INO(de) (de)->d_fileno
		102	# define D_NAMLEN(de) (de)->d_namlen
		103	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		104	# define D_INO(de) (de)->d_ino
		105	# endif
		106
		107	#ifdef _D_EXACT_NAMLEN
		108	# undef D_NAMLEN
		109	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		110	#endif
		111
		112	# ifdef _DIRENT_HAVE_D_TYPE
		113	# define D_TYPE(de) (de)->d_type
		114	# endif
		115
80	# ifndef EIO_STRUCT_DIRENT	116	# ifndef EIO_STRUCT_DIRENT
81	# define EIO_STRUCT_DIRENT struct dirent	117	# define EIO_STRUCT_DIRENT struct dirent
82	# endif	118	# endif
83		119
84	#endif	120	#endif
85		121
86	#if HAVE_SENDFILE	122	#if HAVE_SENDFILE
87	# if __linux	123	# if __linux
88	# include <sys/sendfile.h>	124	# include <sys/sendfile.h>
89	# elif __freebsd	125	# elif __FreeBSD__ || defined __APPLE__
90	# include <sys/socket.h>	126	# include <sys/socket.h>
91	# include <sys/uio.h>	127	# include <sys/uio.h>
92	# elif __hpux	128	# elif __hpux
93	# include <sys/socket.h>	129	# include <sys/socket.h>
94	# elif __solaris /* not yet */	130	# elif __solaris
95	# include <sys/sendfile.h>	131	# include <sys/sendfile.h>
96	# else	132	# else
97	# error sendfile support requested but not available	133	# error sendfile support requested but not available
98	# endif	134	# endif
99	#endif	135	#endif
100		136
101	/* number of seconds after which an idle threads exit */	137	#ifndef D_TYPE
102	#define IDLE_TIMEOUT 10	138	# define D_TYPE(de) 0
		139	#endif
		140	#ifndef D_INO
		141	# define D_INO(de) 0
		142	#endif
		143	#ifndef D_NAMLEN
		144	# define D_NAMLEN(de) strlen ((de)->d_name)
		145	#endif
103		146
104	/* used for struct dirent, AIX doesn't provide it */	147	/* used for struct dirent, AIX doesn't provide it */
105	#ifndef NAME_MAX	148	#ifndef NAME_MAX
106	# 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
107	#endif	155	#endif
108		156
109	/* buffer size for various temporary buffers */	157	/* buffer size for various temporary buffers */
110	#define EIO_BUFSIZE 65536	158	#define EIO_BUFSIZE 65536
111		159
…		…
117	errno = ENOMEM; \	165	errno = ENOMEM; \
118	if (!eio_buf) \	166	if (!eio_buf) \
119	return -1;	167	return -1;
120		168
121	#define EIO_TICKS ((1000000 + 1023) >> 10)	169	#define EIO_TICKS ((1000000 + 1023) >> 10)
122
123	/*****************************************************************************/
124		170
125	#define ETP_PRI_MIN EIO_PRI_MIN	171	#define ETP_PRI_MIN EIO_PRI_MIN
126	#define ETP_PRI_MAX EIO_PRI_MAX	172	#define ETP_PRI_MAX EIO_PRI_MAX
127		173
128	struct etp_worker;	174	struct etp_worker;
…		…
144	if (wrk->dirp) \	190	if (wrk->dirp) \
145	{ \	191	{ \
146	closedir (wrk->dirp); \	192	closedir (wrk->dirp); \
147	wrk->dirp = 0; \	193	wrk->dirp = 0; \
148	}	194	}
		195
149	#define ETP_WORKER_COMMON \	196	#define ETP_WORKER_COMMON \
150	void *dbuf; \	197	void *dbuf; \
151	DIR *dirp;	198	DIR *dirp;
152		199
153	/*****************************************************************************/	200	/*****************************************************************************/
154		201
155	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	202	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156		203
157	/* 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
158	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	206	tvdiff (struct timeval *tv1, struct timeval *tv2)
159	{	207	{
160	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	208	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	209	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
162	}	210	}
163		211
…		…
170	static unsigned int max_poll_reqs; /* reslock */	218	static unsigned int max_poll_reqs; /* reslock */
171		219
172	static volatile unsigned int nreqs; /* reqlock */	220	static volatile unsigned int nreqs; /* reqlock */
173	static volatile unsigned int nready; /* reqlock */	221	static volatile unsigned int nready; /* reqlock */
174	static volatile unsigned int npending; /* reqlock */	222	static volatile unsigned int npending; /* reqlock */
175	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 */
176		225
		226	static xmutex_t wrklock;
		227	static xmutex_t reslock;
		228	static xmutex_t reqlock;
		229	static xcond_t reqwait;
		230
		231	#if !HAVE_PREADWRITE
		232	/*
		233	* make our pread/pwrite emulation safe against themselves, but not against
		234	* normal read/write by using a mutex. slows down execution a lot,
		235	* but that's your problem, not mine.
		236	*/
177	static mutex_t wrklock = X_MUTEX_INIT;	237	static xmutex_t preadwritelock = X_MUTEX_INIT;
178	static mutex_t reslock = X_MUTEX_INIT;	238	#endif
179	static mutex_t reqlock = X_MUTEX_INIT;
180	static cond_t reqwait = X_COND_INIT;
181		239
182	typedef struct etp_worker	240	typedef struct etp_worker
183	{	241	{
184	/* locked by wrklock */	242	/* locked by wrklock */
185	struct etp_worker *prev, *next;	243	struct etp_worker *prev, *next;
186		244
187	thread_t tid;	245	xthread_t tid;
188		246
189	/* locked by reslock, reqlock or wrklock */	247	/* locked by reslock, reqlock or wrklock */
190	ETP_REQ *req; /* currently processed request */	248	ETP_REQ *req; /* currently processed request */
191		249
192	ETP_WORKER_COMMON	250	ETP_WORKER_COMMON
…		…
197	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	255	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
198	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	256	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
199		257
200	/* worker threads management */	258	/* worker threads management */
201		259
		260	static void ecb_cold
202	static void etp_worker_clear (etp_worker *wrk)	261	etp_worker_clear (etp_worker *wrk)
203	{	262	{
204	ETP_WORKER_CLEAR (wrk);	263	ETP_WORKER_CLEAR (wrk);
205	}	264	}
206		265
		266	static void ecb_cold
207	static void etp_worker_free (etp_worker *wrk)	267	etp_worker_free (etp_worker *wrk)
208	{	268	{
209	wrk->next->prev = wrk->prev;	269	wrk->next->prev = wrk->prev;
210	wrk->prev->next = wrk->next;	270	wrk->prev->next = wrk->next;
211		271
212	free (wrk);	272	free (wrk);
213	}	273	}
214		274
215	static unsigned int etp_nreqs (void)	275	static unsigned int
		276	etp_nreqs (void)
216	{	277	{
217	int retval;	278	int retval;
218	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	279	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
219	retval = nreqs;	280	retval = nreqs;
220	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	281	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
221	return retval;	282	return retval;
222	}	283	}
223		284
224	static unsigned int etp_nready (void)	285	static unsigned int
		286	etp_nready (void)
225	{	287	{
226	unsigned int retval;	288	unsigned int retval;
227		289
228	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	290	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
229	retval = nready;	291	retval = nready;
230	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	292	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
231		293
232	return retval;	294	return retval;
233	}	295	}
234		296
235	static unsigned int etp_npending (void)	297	static unsigned int
		298	etp_npending (void)
236	{	299	{
237	unsigned int retval;	300	unsigned int retval;
238		301
239	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
240	retval = npending;	303	retval = npending;
241	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	304	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
242		305
243	return retval;	306	return retval;
244	}	307	}
245		308
246	static unsigned int etp_nthreads (void)	309	static unsigned int
		310	etp_nthreads (void)
247	{	311	{
248	unsigned int retval;	312	unsigned int retval;
249		313
250	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	314	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
251	retval = started;	315	retval = started;
…		…
265	} etp_reqq;	329	} etp_reqq;
266		330
267	static etp_reqq req_queue;	331	static etp_reqq req_queue;
268	static etp_reqq res_queue;	332	static etp_reqq res_queue;
269		333
		334	static int ecb_noinline
270	static int reqq_push (etp_reqq *q, ETP_REQ *req)	335	reqq_push (etp_reqq *q, ETP_REQ *req)
271	{	336	{
272	int pri = req->pri;	337	int pri = req->pri;
273	req->next = 0;	338	req->next = 0;
274		339
275	if (q->qe[pri])	340	if (q->qe[pri])
…		…
281	q->qe[pri] = q->qs[pri] = req;	346	q->qe[pri] = q->qs[pri] = req;
282		347
283	return q->size++;	348	return q->size++;
284	}	349	}
285		350
		351	static ETP_REQ * ecb_noinline
286	static ETP_REQ *reqq_shift (etp_reqq *q)	352	reqq_shift (etp_reqq *q)
287	{	353	{
288	int pri;	354	int pri;
289		355
290	if (!q->size)	356	if (!q->size)
291	return 0;	357	return 0;
…		…
306	}	372	}
307		373
308	abort ();	374	abort ();
309	}	375	}
310		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
311	static void etp_atfork_prepare (void)	387	etp_atfork_prepare (void)
312	{	388	{
313	X_LOCK (wrklock);	389	X_LOCK (wrklock);
314	X_LOCK (reqlock);	390	X_LOCK (reqlock);
315	X_LOCK (reslock);	391	X_LOCK (reslock);
316	#if !HAVE_PREADWRITE	392	#if !HAVE_PREADWRITE
317	X_LOCK (preadwritelock);	393	X_LOCK (preadwritelock);
318	#endif	394	#endif
319	}	395	}
320		396
		397	static void ecb_cold
321	static void etp_atfork_parent (void)	398	etp_atfork_parent (void)
322	{	399	{
323	#if !HAVE_PREADWRITE	400	#if !HAVE_PREADWRITE
324	X_UNLOCK (preadwritelock);	401	X_UNLOCK (preadwritelock);
325	#endif	402	#endif
326	X_UNLOCK (reslock);	403	X_UNLOCK (reslock);
327	X_UNLOCK (reqlock);	404	X_UNLOCK (reqlock);
328	X_UNLOCK (wrklock);	405	X_UNLOCK (wrklock);
329	}	406	}
330		407
		408	static void ecb_cold
331	static void etp_atfork_child (void)	409	etp_atfork_child (void)
332	{	410	{
333	ETP_REQ *prv;	411	ETP_REQ *prv;
334		412
335	while (prv = reqq_shift (&req_queue))	413	while ((prv = reqq_shift (&req_queue)))
336	ETP_DESTROY (prv);	414	ETP_DESTROY (prv);
337		415
338	while (prv = reqq_shift (&res_queue))	416	while ((prv = reqq_shift (&res_queue)))
339	ETP_DESTROY (prv);	417	ETP_DESTROY (prv);
340		418
341	while (wrk_first.next != &wrk_first)	419	while (wrk_first.next != &wrk_first)
342	{	420	{
343	etp_worker *wrk = wrk_first.next;	421	etp_worker *wrk = wrk_first.next;
…		…
353	idle = 0;	431	idle = 0;
354	nreqs = 0;	432	nreqs = 0;
355	nready = 0;	433	nready = 0;
356	npending = 0;	434	npending = 0;
357		435
358	etp_atfork_parent ();	436	etp_thread_init ();
359	}	437	}
360		438
361	static void	439	static void ecb_cold
362	etp_once_init (void)	440	etp_once_init (void)
363	{	441	{
		442	etp_thread_init ();
364	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);
365	}	444	}
366		445
367	static int	446	static int ecb_cold
368	etp_init (void (*want_poll)(void), void (*done_poll)(void))	447	etp_init (void (*want_poll)(void), void (*done_poll)(void))
369	{	448	{
370	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	449	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
371		450
372	pthread_once (&doinit, etp_once_init);	451	pthread_once (&doinit, etp_once_init);
…		…
377	return 0;	456	return 0;
378	}	457	}
379		458
380	X_THREAD_PROC (etp_proc);	459	X_THREAD_PROC (etp_proc);
381		460
		461	static void ecb_cold
382	static void etp_start_thread (void)	462	etp_start_thread (void)
383	{	463	{
384	etp_worker *wrk = calloc (1, sizeof (etp_worker));	464	etp_worker *wrk = calloc (1, sizeof (etp_worker));
385		465
386	/*TODO*/	466	/*TODO*/
387	assert (("unable to allocate worker thread data", wrk));	467	assert (("unable to allocate worker thread data", wrk));
…		…
400	free (wrk);	480	free (wrk);
401		481
402	X_UNLOCK (wrklock);	482	X_UNLOCK (wrklock);
403	}	483	}
404		484
		485	static void
405	static void etp_maybe_start_thread (void)	486	etp_maybe_start_thread (void)
406	{	487	{
407	if (etp_nthreads () >= wanted)	488	if (ecb_expect_true (etp_nthreads () >= wanted))
408	return;	489	return;
409		490
410	/* todo: maybe use idle here, but might be less exact */	491	/* todo: maybe use idle here, but might be less exact */
411	if (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 ()))
412	return;	493	return;
413		494
414	etp_start_thread ();	495	etp_start_thread ();
415	}	496	}
416		497
		498	static void ecb_cold
417	static void etp_end_thread (void)	499	etp_end_thread (void)
418	{	500	{
419	eio_req *req = calloc (1, sizeof (eio_req));	501	eio_req *req = calloc (1, sizeof (eio_req));
420		502
421	req->type = -1;	503	req->type = -1;
422	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	504	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
429	X_LOCK (wrklock);	511	X_LOCK (wrklock);
430	--started;	512	--started;
431	X_UNLOCK (wrklock);	513	X_UNLOCK (wrklock);
432	}	514	}
433		515
434	static int etp_poll (void)	516	static int
		517	etp_poll (void)
435	{	518	{
436	unsigned int maxreqs;	519	unsigned int maxreqs;
437	unsigned int maxtime;	520	unsigned int maxtime;
438	struct timeval tv_start, tv_now;	521	struct timeval tv_start, tv_now;
439		522
…		…
469		552
470	X_LOCK (reqlock);	553	X_LOCK (reqlock);
471	--nreqs;	554	--nreqs;
472	X_UNLOCK (reqlock);	555	X_UNLOCK (reqlock);
473		556
474	if (req->type == EIO_GROUP && req->size)	557	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
475	{	558	{
476	req->int1 = 1; /* mark request as delayed */	559	req->int1 = 1; /* mark request as delayed */
477	continue;	560	continue;
478	}	561	}
479	else	562	else
480	{	563	{
481	int res = ETP_FINISH (req);	564	int res = ETP_FINISH (req);
482	if (res)	565	if (ecb_expect_false (res))
483	return res;	566	return res;
484	}	567	}
485		568
486	if (maxreqs && !--maxreqs)	569	if (ecb_expect_false (maxreqs && !--maxreqs))
487	break;	570	break;
488		571
489	if (maxtime)	572	if (maxtime)
490	{	573	{
491	gettimeofday (&tv_now, 0);	574	gettimeofday (&tv_now, 0);
…		…
497		580
498	errno = EAGAIN;	581	errno = EAGAIN;
499	return -1;	582	return -1;
500	}	583	}
501		584
		585	static void
502	static void etp_cancel (ETP_REQ *req)	586	etp_cancel (ETP_REQ *req)
503	{	587	{
504	X_LOCK (wrklock);	588	X_LOCK (wrklock);
505	req->flags |= EIO_FLAG_CANCELLED;	589	req->flags |= EIO_FLAG_CANCELLED;
506	X_UNLOCK (wrklock);	590	X_UNLOCK (wrklock);
507		591
508	eio_grp_cancel (req);	592	eio_grp_cancel (req);
509	}	593	}
510		594
		595	static void
511	static void etp_submit (ETP_REQ *req)	596	etp_submit (ETP_REQ *req)
512	{	597	{
513	req->pri -= ETP_PRI_MIN;	598	req->pri -= ETP_PRI_MIN;
514		599
515	if (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;
516	if (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;
517		602
		603	if (ecb_expect_false (req->type == EIO_GROUP))
		604	{
		605	/* I hope this is worth it :/ */
518	X_LOCK (reqlock);	606	X_LOCK (reqlock);
519	++nreqs;	607	++nreqs;
		608	X_UNLOCK (reqlock);
		609
		610	X_LOCK (reslock);
		611
		612	++npending;
		613
		614	if (!reqq_push (&res_queue, req) && want_poll_cb)
		615	want_poll_cb ();
		616
		617	X_UNLOCK (reslock);
		618	}
		619	else
		620	{
		621	X_LOCK (reqlock);
		622	++nreqs;
520	++nready;	623	++nready;
521	reqq_push (&req_queue, req);	624	reqq_push (&req_queue, req);
522	X_COND_SIGNAL (reqwait);	625	X_COND_SIGNAL (reqwait);
523	X_UNLOCK (reqlock);	626	X_UNLOCK (reqlock);
524		627
525	etp_maybe_start_thread ();	628	etp_maybe_start_thread ();
		629	}
526	}	630	}
527		631
		632	static void ecb_cold
528	static void etp_set_max_poll_time (double nseconds)	633	etp_set_max_poll_time (double nseconds)
529	{	634	{
530	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	635	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
531	max_poll_time = nseconds;	636	max_poll_time = nseconds * EIO_TICKS;
532	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	637	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
533	}	638	}
534		639
		640	static void ecb_cold
535	static void etp_set_max_poll_reqs (unsigned int maxreqs)	641	etp_set_max_poll_reqs (unsigned int maxreqs)
536	{	642	{
537	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	643	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
538	max_poll_reqs = maxreqs;	644	max_poll_reqs = maxreqs;
539	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	645	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
540	}	646	}
541		647
		648	static void ecb_cold
542	static void etp_set_max_idle (unsigned int nthreads)	649	etp_set_max_idle (unsigned int nthreads)
543	{	650	{
544	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	651	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
545	max_idle = nthreads <= 0 ? 1 : nthreads;	652	max_idle = nthreads;
546	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	653	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
547	}	654	}
548		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
549	static void etp_set_min_parallel (unsigned int nthreads)	665	etp_set_min_parallel (unsigned int nthreads)
550	{	666	{
551	if (wanted < nthreads)	667	if (wanted < nthreads)
552	wanted = nthreads;	668	wanted = nthreads;
553	}	669	}
554		670
		671	static void ecb_cold
555	static void etp_set_max_parallel (unsigned int nthreads)	672	etp_set_max_parallel (unsigned int nthreads)
556	{	673	{
557	if (wanted > nthreads)	674	if (wanted > nthreads)
558	wanted = nthreads;	675	wanted = nthreads;
559		676
560	while (started > wanted)	677	while (started > wanted)
561	etp_end_thread ();	678	etp_end_thread ();
562	}	679	}
563		680
564	/*****************************************************************************/	681	/*****************************************************************************/
565		682
		683	static void
566	static void grp_try_feed (eio_req *grp)	684	grp_try_feed (eio_req *grp)
567	{	685	{
568	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	686	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569	{	687	{
570	int old_len = grp->size;	688	grp->flags &= ~EIO_FLAG_GROUPADD;
571		689
572	EIO_FEED (grp);	690	EIO_FEED (grp);
573		691
574	/* stop if no progress has been made */	692	/* stop if no progress has been made */
575	if (old_len == grp->size)	693	if (!(grp->flags & EIO_FLAG_GROUPADD))
576	{	694	{
577	grp->feed = 0;	695	grp->feed = 0;
578	break;	696	break;
579	}	697	}
580	}	698	}
581	}	699	}
582		700
		701	static int
583	static int grp_dec (eio_req *grp)	702	grp_dec (eio_req *grp)
584	{	703	{
585	--grp->size;	704	--grp->size;
586		705
587	/* call feeder, if applicable */	706	/* call feeder, if applicable */
588	grp_try_feed (grp);	707	grp_try_feed (grp);
…		…
592	return eio_finish (grp);	711	return eio_finish (grp);
593	else	712	else
594	return 0;	713	return 0;
595	}	714	}
596		715
		716	void
597	void eio_destroy (eio_req *req)	717	eio_destroy (eio_req *req)
598	{	718	{
599	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	719	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
600	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	720	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
601		721
602	EIO_DESTROY (req);	722	EIO_DESTROY (req);
603	}	723	}
604		724
		725	static int
605	static int eio_finish (eio_req *req)	726	eio_finish (eio_req *req)
606	{	727	{
607	int res = EIO_FINISH (req);	728	int res = EIO_FINISH (req);
608		729
609	if (req->grp)	730	if (req->grp)
610	{	731	{
…		…
627	eio_destroy (req);	748	eio_destroy (req);
628		749
629	return res;	750	return res;
630	}	751	}
631		752
		753	void
632	void eio_grp_cancel (eio_req *grp)	754	eio_grp_cancel (eio_req *grp)
633	{	755	{
634	for (grp = grp->grp_first; grp; grp = grp->grp_next)	756	for (grp = grp->grp_first; grp; grp = grp->grp_next)
635	eio_cancel (grp);	757	eio_cancel (grp);
636	}	758	}
637		759
		760	void
638	void eio_cancel (eio_req *req)	761	eio_cancel (eio_req *req)
639	{	762	{
640	etp_cancel (req);	763	etp_cancel (req);
641	}	764	}
642		765
		766	void
643	void eio_submit (eio_req *req)	767	eio_submit (eio_req *req)
644	{	768	{
645	etp_submit (req);	769	etp_submit (req);
646	}	770	}
647		771
648	unsigned int eio_nreqs (void)	772	unsigned int
		773	eio_nreqs (void)
649	{	774	{
650	return etp_nreqs ();	775	return etp_nreqs ();
651	}	776	}
652		777
653	unsigned int eio_nready (void)	778	unsigned int
		779	eio_nready (void)
654	{	780	{
655	return etp_nready ();	781	return etp_nready ();
656	}	782	}
657		783
658	unsigned int eio_npending (void)	784	unsigned int
		785	eio_npending (void)
659	{	786	{
660	return etp_npending ();	787	return etp_npending ();
661	}	788	}
662		789
663	unsigned int eio_nthreads (void)	790	unsigned int ecb_cold
		791	eio_nthreads (void)
664	{	792	{
665	return etp_nthreads ();	793	return etp_nthreads ();
666	}	794	}
667		795
		796	void ecb_cold
668	void eio_set_max_poll_time (double nseconds)	797	eio_set_max_poll_time (double nseconds)
669	{	798	{
670	etp_set_max_poll_time (nseconds);	799	etp_set_max_poll_time (nseconds);
671	}	800	}
672		801
		802	void ecb_cold
673	void eio_set_max_poll_reqs (unsigned int maxreqs)	803	eio_set_max_poll_reqs (unsigned int maxreqs)
674	{	804	{
675	etp_set_max_poll_reqs (maxreqs);	805	etp_set_max_poll_reqs (maxreqs);
676	}	806	}
677		807
		808	void ecb_cold
678	void eio_set_max_idle (unsigned int nthreads)	809	eio_set_max_idle (unsigned int nthreads)
679	{	810	{
680	etp_set_max_idle (nthreads);	811	etp_set_max_idle (nthreads);
681	}	812	}
682		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
683	void eio_set_min_parallel (unsigned int nthreads)	821	eio_set_min_parallel (unsigned int nthreads)
684	{	822	{
685	etp_set_min_parallel (nthreads);	823	etp_set_min_parallel (nthreads);
686	}	824	}
687		825
		826	void ecb_cold
688	void eio_set_max_parallel (unsigned int nthreads)	827	eio_set_max_parallel (unsigned int nthreads)
689	{	828	{
690	etp_set_max_parallel (nthreads);	829	etp_set_max_parallel (nthreads);
691	}	830	}
692		831
693	int eio_poll (void)	832	int eio_poll (void)
…		…
696	}	835	}
697		836
698	/*****************************************************************************/	837	/*****************************************************************************/
699	/* work around various missing functions */	838	/* work around various missing functions */
700		839
		840	#if _POSIX_VERSION < 200809L
		841	# define realpath(path,resolved_path) (errno = ENOSYS, 0)
		842	#endif
		843
701	#if !HAVE_PREADWRITE	844	#if !HAVE_PREADWRITE
		845	# undef pread
		846	# undef pwrite
702	# define pread eio__pread	847	# define pread eio__pread
703	# define pwrite eio__pwrite	848	# define pwrite eio__pwrite
704
705	/*
706	* make our pread/pwrite safe against themselves, but not against
707	* normal read/write by using a mutex. slows down execution a lot,
708	* but that's your problem, not mine.
709	*/
710	static mutex_t preadwritelock = X_MUTEX_INIT;
711		849
712	static ssize_t	850	static ssize_t
713	eio__pread (int fd, void *buf, size_t count, off_t offset)	851	eio__pread (int fd, void *buf, size_t count, off_t offset)
714	{	852	{
715	ssize_t res;	853	ssize_t res;
…		…
733		871
734	X_LOCK (preadwritelock);	872	X_LOCK (preadwritelock);
735	ooffset = lseek (fd, 0, SEEK_CUR);	873	ooffset = lseek (fd, 0, SEEK_CUR);
736	lseek (fd, offset, SEEK_SET);	874	lseek (fd, offset, SEEK_SET);
737	res = write (fd, buf, count);	875	res = write (fd, buf, count);
738	lseek (fd, offset, SEEK_SET);	876	lseek (fd, ooffset, SEEK_SET);
739	X_UNLOCK (preadwritelock);	877	X_UNLOCK (preadwritelock);
740		878
741	return res;	879	return res;
742	}	880	}
743	#endif	881	#endif
744		882
745	#ifndef HAVE_FUTIMES	883	#ifndef HAVE_UTIMES
746		884
		885	# undef utimes
747	# define utimes(path,times) eio__utimes (path, times)	886	# define utimes(path,times) eio__utimes (path, times)
748	# define futimes(fd,times) eio__futimes (fd, times)
749		887
750	static int	888	static int
751	eio__utimes (const char *filename, const struct timeval times[2])	889	eio__utimes (const char *filename, const struct timeval times[2])
752	{	890	{
753	if (times)	891	if (times)
…		…
761	}	899	}
762	else	900	else
763	return utime (filename, 0);	901	return utime (filename, 0);
764	}	902	}
765		903
		904	#endif
		905
		906	#ifndef HAVE_FUTIMES
		907
		908	# undef futimes
		909	# define futimes(fd,times) eio__futimes (fd, times)
		910
		911	static int
766	static int eio__futimes (int fd, const struct timeval tv[2])	912	eio__futimes (int fd, const struct timeval tv[2])
767	{	913	{
768	errno = ENOSYS;	914	errno = ENOSYS;
769	return -1;	915	return -1;
770	}	916	}
771		917
772	#endif	918	#endif
773		919
774	#if !HAVE_FDATASYNC	920	#if !HAVE_FDATASYNC
		921	# undef fdatasync
775	# define fdatasync fsync	922	# define fdatasync(fd) fsync (fd)
776	#endif	923	#endif
		924
		925	/* sync_file_range always needs emulation */
		926	static int
		927	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		928	{
		929	#if HAVE_SYNC_FILE_RANGE
		930	int res;
		931
		932	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		933	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		934	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		935	{
		936	flags = 0
		937	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		938	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		939	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		940	}
		941
		942	res = sync_file_range (fd, offset, nbytes, flags);
		943
		944	if (!res || errno != ENOSYS)
		945	return res;
		946	#endif
		947
		948	/* even though we could play tricks with the flags, it's better to always
		949	* call fdatasync, as that matches the expectation of its users best */
		950	return fdatasync (fd);
		951	}
777		952
778	#if !HAVE_READAHEAD	953	#if !HAVE_READAHEAD
		954	# undef readahead
779	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	955	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780		956
781	static ssize_t	957	static ssize_t
782	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	958	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783	{	959	{
…		…
801		977
802	/* sendfile always needs emulation */	978	/* sendfile always needs emulation */
803	static ssize_t	979	static ssize_t
804	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	980	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
805	{	981	{
		982	ssize_t written = 0;
806	ssize_t res;	983	ssize_t res;
807		984
808	if (!count)	985	if (!count)
809	return 0;	986	return 0;
810		987
		988	for (;;)
		989	{
811	#if HAVE_SENDFILE	990	#if HAVE_SENDFILE
812	# if __linux	991	# if __linux
		992	off_t soffset = offset;
813	res = sendfile (ofd, ifd, &offset, count);	993	res = sendfile (ofd, ifd, &soffset, count);
814		994
815	# elif __freebsd	995	# elif __FreeBSD__
816	/*	996	/*
817	* Of course, the freebsd sendfile is a dire hack with no thoughts	997	* Of course, the freebsd sendfile is a dire hack with no thoughts
818	* wasted on making it similar to other I/O functions.	998	* wasted on making it similar to other I/O functions.
819	*/	999	*/
820	{
821	off_t sbytes;	1000	off_t sbytes;
822	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1001	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
823		1002
824	if (res < 0 && sbytes)	1003	#if 0 /* according to the manpage, this is correct, but broken behaviour */
825	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1004	/* freebsd' sendfile will return 0 on success */
		1005	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1006	/* not on e.g. EIO or EPIPE - sounds broken */
		1007	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
826	res = sbytes;	1008	res = sbytes;
827	}	1009	#endif
		1010
		1011	/* according to source inspection, this is correct, and useful behaviour */
		1012	if (sbytes)
		1013	res = sbytes;
		1014
		1015	# elif defined (__APPLE__) && 0 /* broken, as everything on os x */
		1016	off_t sbytes = count;
		1017	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1018
		1019	/* according to the manpage, sbytes is always valid */
		1020	if (sbytes)
		1021	res = sbytes;
828		1022
829	# elif __hpux	1023	# elif __hpux
830	res = sendfile (ofd, ifd, offset, count, 0, 0);	1024	res = sendfile (ofd, ifd, offset, count, 0, 0);
831		1025
832	# elif __solaris	1026	# elif __solaris
833	{
834	struct sendfilevec vec;	1027	struct sendfilevec vec;
835	size_t sbytes;	1028	size_t sbytes;
836		1029
837	vec.sfv_fd = ifd;	1030	vec.sfv_fd = ifd;
838	vec.sfv_flag = 0;	1031	vec.sfv_flag = 0;
839	vec.sfv_off = offset;	1032	vec.sfv_off = offset;
840	vec.sfv_len = count;	1033	vec.sfv_len = count;
841		1034
842	res = sendfilev (ofd, &vec, 1, &sbytes);	1035	res = sendfilev (ofd, &vec, 1, &sbytes);
843		1036
844	if (res < 0 && sbytes)	1037	if (res < 0 && sbytes)
845	res = sbytes;	1038	res = sbytes;
846	}
847		1039
848	# endif	1040	# endif
		1041
		1042	#elif defined (_WIN32)
		1043	/* does not work, just for documentation of what would need to be done */
		1044	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1045	/* libeio guarantees that the file offset does not change, and windows */
		1046	/* has no way to get an independent handle to the same file description */
		1047	HANDLE h = TO_SOCKET (ifd);
		1048	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1049	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1050
849	#else	1051	#else
850	res = -1;	1052	res = -1;
851	errno = ENOSYS;	1053	errno = ENOSYS;
852	#endif	1054	#endif
853		1055
		1056	/* we assume sendfile can copy at least 128mb in one go */
		1057	if (res <= 128 * 1024 * 1024)
		1058	{
		1059	if (res > 0)
		1060	written += res;
		1061
		1062	if (written)
		1063	return written;
		1064
		1065	break;
		1066	}
		1067	else
		1068	{
		1069	/* if we requested more, then probably the kernel was lazy */
		1070	written += res;
		1071	offset += res;
		1072	count -= res;
		1073
		1074	if (!count)
		1075	return written;
		1076	}
		1077	}
		1078
854	if (res < 0	1079	if (res < 0
855	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1080	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1081	/* BSDs */
		1082	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1083	|| errno == ENOTSUP
		1084	#endif
		1085	|| errno == EOPNOTSUPP /* BSDs */
856	#if __solaris	1086	#if __solaris
857	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1087	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
858	#endif	1088	#endif
859	)	1089	)
860	)	1090	)
…		…
891	}	1121	}
892		1122
893	return res;	1123	return res;
894	}	1124	}
895		1125
		1126	static signed char
		1127	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1128	{
		1129	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1130	: a->inode < b->inode ? -1
		1131	: a->inode > b->inode ? 1
		1132	: 0;
		1133	}
		1134
		1135	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1136
		1137	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1138	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1139
		1140	static void
		1141	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1142	{
		1143	unsigned char bits [9 + sizeof (ino_t) * 8];
		1144	unsigned char *bit = bits;
		1145
		1146	assert (CHAR_BIT == 8);
		1147	assert (sizeof (eio_dirent) * 8 < 256);
		1148	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1149	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1150
		1151	if (size <= EIO_SORT_FAST)
		1152	return;
		1153
		1154	/* first prepare an array of bits to test in our radix sort */
		1155	/* try to take endianness into account, as well as differences in ino_t sizes */
		1156	/* inode_bits must contain all inodes ORed together */
		1157	/* which is used to skip bits that are 0 everywhere, which is very common */
		1158	{
		1159	ino_t endianness;
		1160	int i, j;
		1161
		1162	/* we store the byte offset of byte n into byte n of "endianness" */
		1163	for (i = 0; i < sizeof (ino_t); ++i)
		1164	((unsigned char *)&endianness)[i] = i;
		1165
		1166	*bit++ = 0;
		1167
		1168	for (i = 0; i < sizeof (ino_t); ++i)
		1169	{
		1170	/* shifting off the byte offsets out of "endianness" */
		1171	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1172	endianness >>= 8;
		1173
		1174	for (j = 0; j < 8; ++j)
		1175	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1176	*bit++ = offs + j;
		1177	}
		1178
		1179	for (j = 0; j < 8; ++j)
		1180	if (score_bits & (1 << j))
		1181	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1182	}
		1183
		1184	/* now actually do the sorting (a variant of MSD radix sort) */
		1185	{
		1186	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1187	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1188	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1189	int stk_idx = 0;
		1190
		1191	base_stk [stk_idx] = dents;
		1192	end_stk [stk_idx] = dents + size;
		1193	bit_stk [stk_idx] = bit - 1;
		1194
		1195	do
		1196	{
		1197	base = base_stk [stk_idx];
		1198	end = end_stk [stk_idx];
		1199	bit = bit_stk [stk_idx];
		1200
		1201	for (;;)
		1202	{
		1203	unsigned char O = *bit >> 3;
		1204	unsigned char M = 1 << (*bit & 7);
		1205
		1206	eio_dirent *a = base;
		1207	eio_dirent *b = end;
		1208
		1209	if (b - a < EIO_SORT_CUTOFF)
		1210	break;
		1211
		1212	/* now bit-partition the array on the bit */
		1213	/* this ugly asymmetric loop seems to perform much better than typical */
		1214	/* partition algos found in the literature */
		1215	do
		1216	if (!(((unsigned char *)a)[O] & M))
		1217	++a;
		1218	else if (!(((unsigned char *)--b)[O] & M))
		1219	{
		1220	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1221	++a;
		1222	}
		1223	while (b > a);
		1224
		1225	/* next bit, or stop, if no bits left in this path */
		1226	if (!*--bit)
		1227	break;
		1228
		1229	base_stk [stk_idx] = a;
		1230	end_stk [stk_idx] = end;
		1231	bit_stk [stk_idx] = bit;
		1232	++stk_idx;
		1233
		1234	end = a;
		1235	}
		1236	}
		1237	while (stk_idx--);
		1238	}
		1239	}
		1240
		1241	static void
		1242	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1243	{
		1244	/* first move the smallest element to the front, to act as a sentinel */
		1245	{
		1246	int i;
		1247	eio_dirent *min = dents;
		1248
		1249	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1250	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1251	if (EIO_DENT_CMP (dents [i], <, *min))
		1252	min = &dents [i];
		1253
		1254	/* swap elements 0 and j (minimum) */
		1255	{
		1256	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1257	}
		1258	}
		1259
		1260	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1261	{
		1262	eio_dirent *i, *j;
		1263
		1264	for (i = dents + 1; i < dents + size; ++i)
		1265	{
		1266	eio_dirent value = *i;
		1267
		1268	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1269	j [1] = j [0];
		1270
		1271	j [1] = value;
		1272	}
		1273	}
		1274	}
		1275
		1276	static void
		1277	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1278	{
		1279	if (size <= 1)
		1280	return; /* our insertion sort relies on size > 0 */
		1281
		1282	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1283	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1284	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1285
		1286	/* use an insertion sort at the end, or for small arrays, */
		1287	/* as insertion sort is more efficient for small partitions */
		1288	eio_dent_insertion_sort (dents, size);
		1289	}
		1290
896	/* read a full directory */	1291	/* read a full directory */
897	static void	1292	static void
898	eio__scandir (eio_req *req, etp_worker *self)	1293	eio__scandir (eio_req *req, etp_worker *self)
899	{	1294	{
900	DIR *dirp;	1295	DIR *dirp;
901	EIO_STRUCT_DIRENT *entp;	1296	EIO_STRUCT_DIRENT *entp;
902	char *name, *names;	1297	char *name, *names;
903	int memlen = 4096;	1298	int namesalloc = 4096;
904	int memofs = 0;	1299	int namesoffs = 0;
		1300	int flags = req->int1;
		1301	eio_dirent *dents = 0;
		1302	int dentalloc = 128;
905	int res = 0;	1303	int dentoffs = 0;
		1304	ino_t inode_bits = 0;
		1305
		1306	req->result = -1;
		1307
		1308	if (!(flags & EIO_READDIR_DENTS))
		1309	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
906		1310
907	X_LOCK (wrklock);	1311	X_LOCK (wrklock);
		1312	/* the corresponding closedir is in ETP_WORKER_CLEAR */
908	self->dirp = dirp = opendir (req->ptr1);	1313	self->dirp = dirp = opendir (req->ptr1);
		1314
909	req->flags |= EIO_FLAG_PTR2_FREE;	1315	if (req->flags & EIO_FLAG_PTR1_FREE)
		1316	free (req->ptr1);
		1317
		1318	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1319	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
910	req->ptr2 = names = malloc (memlen);	1320	req->ptr2 = names = malloc (namesalloc);
911	X_UNLOCK (wrklock);	1321	X_UNLOCK (wrklock);
912		1322
913	if (dirp && names)	1323	if (dirp && names && (!flags || dents))
914	for (;;)	1324	for (;;)
915	{	1325	{
916	errno = 0;	1326	errno = 0;
917	entp = readdir (dirp);	1327	entp = readdir (dirp);
918		1328
919	if (!entp)	1329	if (!entp)
		1330	{
		1331	if (errno)
		1332	break;
		1333
		1334	/* sort etc. */
		1335	req->int1 = flags;
		1336	req->result = dentoffs;
		1337
		1338	if (flags & EIO_READDIR_STAT_ORDER)
		1339	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1340	else if (flags & EIO_READDIR_DIRS_FIRST)
		1341	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1342	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1343	else
		1344	{
		1345	/* in this case, all is known, and we just put dirs first and sort them */
		1346	eio_dirent *oth = dents + dentoffs;
		1347	eio_dirent *dir = dents;
		1348
		1349	/* now partition dirs to the front, and non-dirs to the back */
		1350	/* by walking from both sides and swapping if necessary */
		1351	while (oth > dir)
		1352	{
		1353	if (dir->type == EIO_DT_DIR)
		1354	++dir;
		1355	else if ((--oth)->type == EIO_DT_DIR)
		1356	{
		1357	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1358
		1359	++dir;
		1360	}
		1361	}
		1362
		1363	/* now sort the dirs only (dirs all have the same score) */
		1364	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1365	}
		1366
920	break;	1367	break;
		1368	}
921		1369
		1370	/* now add the entry to our list(s) */
922	name = entp->d_name;	1371	name = entp->d_name;
923		1372
		1373	/* skip . and .. entries */
924	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1374	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
925	{	1375	{
926	int len = strlen (name) + 1;	1376	int len = D_NAMLEN (entp) + 1;
927		1377
928	res++;	1378	while (ecb_expect_false (namesoffs + len > namesalloc))
929
930	while (memofs + len > memlen)
931	{	1379	{
932	memlen *= 2;	1380	namesalloc *= 2;
933	X_LOCK (wrklock);	1381	X_LOCK (wrklock);
934	req->ptr2 = names = realloc (names, memlen);	1382	req->ptr2 = names = realloc (names, namesalloc);
935	X_UNLOCK (wrklock);	1383	X_UNLOCK (wrklock);
936		1384
937	if (!names)	1385	if (!names)
938	break;	1386	break;
939	}	1387	}
940		1388
941	memcpy (names + memofs, name, len);	1389	memcpy (names + namesoffs, name, len);
		1390
		1391	if (dents)
		1392	{
		1393	struct eio_dirent *ent;
		1394
		1395	if (ecb_expect_false (dentoffs == dentalloc))
		1396	{
		1397	dentalloc *= 2;
		1398	X_LOCK (wrklock);
		1399	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1400	X_UNLOCK (wrklock);
		1401
		1402	if (!dents)
		1403	break;
		1404	}
		1405
		1406	ent = dents + dentoffs;
		1407
		1408	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1409	ent->namelen = len - 1;
		1410	ent->inode = D_INO (entp);
		1411
		1412	inode_bits |= ent->inode;
		1413
		1414	switch (D_TYPE (entp))
		1415	{
		1416	default:
		1417	ent->type = EIO_DT_UNKNOWN;
		1418	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1419	break;
		1420
		1421	#ifdef DT_FIFO
		1422	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1423	#endif
		1424	#ifdef DT_CHR
		1425	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1426	#endif
		1427	#ifdef DT_MPC
		1428	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1429	#endif
		1430	#ifdef DT_DIR
		1431	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1432	#endif
		1433	#ifdef DT_NAM
		1434	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1435	#endif
		1436	#ifdef DT_BLK
		1437	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1438	#endif
		1439	#ifdef DT_MPB
		1440	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1441	#endif
		1442	#ifdef DT_REG
		1443	case DT_REG: ent->type = EIO_DT_REG; break;
		1444	#endif
		1445	#ifdef DT_NWK
		1446	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1447	#endif
		1448	#ifdef DT_CMP
		1449	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1450	#endif
		1451	#ifdef DT_LNK
		1452	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1453	#endif
		1454	#ifdef DT_SOCK
		1455	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1456	#endif
		1457	#ifdef DT_DOOR
		1458	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1459	#endif
		1460	#ifdef DT_WHT
		1461	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1462	#endif
		1463	}
		1464
		1465	ent->score = 7;
		1466
		1467	if (flags & EIO_READDIR_DIRS_FIRST)
		1468	{
		1469	if (ent->type == EIO_DT_UNKNOWN)
		1470	{
		1471	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1472	ent->score = 1;
		1473	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1474	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1475	}
		1476	else if (ent->type == EIO_DT_DIR)
		1477	ent->score = 0;
		1478	}
		1479	}
		1480
942	memofs += len;	1481	namesoffs += len;
		1482	++dentoffs;
		1483	}
		1484
		1485	if (EIO_CANCELLED (req))
		1486	{
		1487	errno = ECANCELED;
		1488	break;
943	}	1489	}
944	}	1490	}
		1491	}
945		1492
946	if (errno)	1493	#ifdef PAGESIZE
947	res = -1;	1494	# define eio_pagesize() PAGESIZE
		1495	#else
		1496	static intptr_t
		1497	eio_pagesize (void)
		1498	{
		1499	static intptr_t page;
		1500
		1501	if (!page)
		1502	page = sysconf (_SC_PAGESIZE);
		1503
		1504	return page;
		1505	}
		1506	#endif
		1507
		1508	static void
		1509	eio_page_align (void **addr, size_t *length)
		1510	{
		1511	intptr_t mask = eio_pagesize () - 1;
		1512
		1513	/* round down addr */
		1514	intptr_t adj = mask & (intptr_t)*addr;
		1515
		1516	*addr = (void *)((intptr_t)*addr - adj);
		1517	*length += adj;
		1518
		1519	/* round up length */
		1520	*length = (*length + mask) & ~mask;
		1521	}
		1522
		1523	#if !_POSIX_MEMLOCK
		1524	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1525	#else
		1526
		1527	static int
		1528	eio__mlockall (int flags)
		1529	{
		1530	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1531	extern int mallopt (int, int);
		1532	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1533	#endif
		1534
		1535	if (EIO_MCL_CURRENT != MCL_CURRENT
		1536	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1537	{
		1538	flags = 0
		1539	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1540	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1541	}
		1542
		1543	return mlockall (flags);
		1544	}
		1545	#endif
		1546
		1547	#if !_POSIX_MEMLOCK_RANGE
		1548	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1549	#else
		1550
		1551	static int
		1552	eio__mlock (void *addr, size_t length)
		1553	{
		1554	eio_page_align (&addr, &length);
		1555
		1556	return mlock (addr, length);
		1557	}
		1558
		1559	#endif
		1560
		1561	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1562	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1563	#else
		1564
		1565	static int
		1566	eio__msync (void *mem, size_t len, int flags)
		1567	{
		1568	eio_page_align (&mem, &len);
		1569
		1570	if (EIO_MS_ASYNC != MS_SYNC
		1571	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1572	|| EIO_MS_SYNC != MS_SYNC)
		1573	{
		1574	flags = 0
		1575	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1576	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1577	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1578	}
		1579
		1580	return msync (mem, len, flags);
		1581	}
		1582
		1583	#endif
		1584
		1585	static int
		1586	eio__mtouch (eio_req *req)
		1587	{
		1588	void *mem = req->ptr2;
		1589	size_t len = req->size;
		1590	int flags = req->int1;
		1591
		1592	eio_page_align (&mem, &len);
		1593
948		1594	{
949	req->result = res;	1595	intptr_t addr = (intptr_t)mem;
		1596	intptr_t end = addr + len;
		1597	intptr_t page = eio_pagesize ();
		1598
		1599	if (addr < end)
		1600	if (flags & EIO_MT_MODIFY) /* modify */
		1601	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1602	else
		1603	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1604	}
		1605
		1606	return 0;
950	}	1607	}
951		1608
952	/*****************************************************************************/	1609	/*****************************************************************************/
953		1610
954	#define ALLOC(len) \	1611	#define ALLOC(len) \
…		…
986	if (req)	1643	if (req)
987	break;	1644	break;
988		1645
989	++idle;	1646	++idle;
990		1647
991	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1648	ts.tv_sec = time (0) + idle_timeout;
992	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1649	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
993	{	1650	{
994	if (idle > max_idle)	1651	if (idle > max_idle)
995	{	1652	{
996	--idle;	1653	--idle;
…		…
1039	return 0;	1696	return 0;
1040	}	1697	}
1041		1698
1042	/*****************************************************************************/	1699	/*****************************************************************************/
1043		1700
		1701	int ecb_cold
1044	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1702	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1045	{	1703	{
1046	return etp_init (want_poll, done_poll);	1704	return etp_init (want_poll, done_poll);
1047	}	1705	}
1048		1706
		1707	ecb_inline void
1049	static void eio_api_destroy (eio_req *req)	1708	eio_api_destroy (eio_req *req)
1050	{	1709	{
1051	free (req);	1710	free (req);
1052	}	1711	}
1053		1712
1054	#define REQ(rtype) \	1713	#define REQ(rtype) \
…		…
1073	{ \	1732	{ \
1074	eio_api_destroy (req); \	1733	eio_api_destroy (req); \
1075	return 0; \	1734	return 0; \
1076	}	1735	}
1077		1736
		1737	static void
1078	static void eio_execute (etp_worker *self, eio_req *req)	1738	eio_execute (etp_worker *self, eio_req *req)
1079	{	1739	{
1080	errno = 0;
1081
1082	switch (req->type)	1740	switch (req->type)
1083	{	1741	{
1084	case EIO_READ: ALLOC (req->size);	1742	case EIO_READ: ALLOC (req->size);
1085	req->result = req->offs >= 0	1743	req->result = req->offs >= 0
1086	? pread (req->int1, req->ptr2, req->size, req->offs)	1744	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1097	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1755	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1756	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1099	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1757	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1100	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1758	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1101		1759
		1760	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1761	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1762	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1763	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1764
1102	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1765	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1103	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1766	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1104	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	1767	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1105	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1768	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1106	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	1769	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1113	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1776	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1114	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	1777	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1115	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	1778	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1116	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	1779	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1117	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	1780	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1118	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	1781	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1119		1782
		1783	case EIO_REALPATH: req->flags |= EIO_FLAG_PTR2_FREE;
		1784	req->ptr2 = realpath (req->ptr1, 0);
		1785	req->result = req->ptr2 ? strlen (req->ptr2) : -1;
		1786	break;
		1787
1120	case EIO_READLINK: ALLOC (NAME_MAX);	1788	case EIO_READLINK: ALLOC (PATH_MAX);
1121	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	1789	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1122		1790
1123	case EIO_SYNC: req->result = 0; sync (); break;	1791	case EIO_SYNC: req->result = 0; sync (); break;
1124	case EIO_FSYNC: req->result = fsync (req->int1); break;	1792	case EIO_FSYNC: req->result = fsync (req->int1); break;
1125	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1793	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1794	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1795	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		1796	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1797	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		1798	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1126		1799
1127	case EIO_READDIR: eio__scandir (req, self); break;	1800	case EIO_READDIR: eio__scandir (req, self); break;
1128		1801
1129	case EIO_BUSY:	1802	case EIO_BUSY:
1130	#ifdef _WIN32	1803	#ifdef _WIN32
1131	Sleep (req->nv1 * 1000.);	1804	Sleep (req->nv1 * 1e3);
1132	#else	1805	#else
1133	{	1806	{
1134	struct timeval tv;	1807	struct timeval tv;
1135		1808
1136	tv.tv_sec = req->nv1;	1809	tv.tv_sec = req->nv1;
1137	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	1810	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1138		1811
1139	req->result = select (0, 0, 0, 0, &tv);	1812	req->result = select (0, 0, 0, 0, &tv);
1140	}	1813	}
1141	#endif	1814	#endif
1142	break;	1815	break;
…		…
1157	times = tv;	1830	times = tv;
1158	}	1831	}
1159	else	1832	else
1160	times = 0;	1833	times = 0;
1161		1834
1162
1163	req->result = req->type == EIO_FUTIME	1835	req->result = req->type == EIO_FUTIME
1164	? futimes (req->int1, times)	1836	? futimes (req->int1, times)
1165	: utimes (req->ptr1, times);	1837	: utimes (req->ptr1, times);
1166	}	1838	}
		1839	break;
1167		1840
1168	case EIO_GROUP:	1841	case EIO_GROUP:
		1842	abort (); /* handled in eio_request */
		1843
1169	case EIO_NOP:	1844	case EIO_NOP:
1170	req->result = 0;	1845	req->result = 0;
1171	break;	1846	break;
1172		1847
1173	case EIO_CUSTOM:	1848	case EIO_CUSTOM:
1174	req->feed (req);	1849	req->feed (req);
1175	break;	1850	break;
1176		1851
1177	default:	1852	default:
		1853	errno = ENOSYS;
1178	req->result = -1;	1854	req->result = -1;
1179	break;	1855	break;
1180	}	1856	}
1181		1857
1182	req->errorno = errno;	1858	req->errorno = errno;
…		…
1202	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1878	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1203	{	1879	{
1204	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1880	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1205	}	1881	}
1206		1882
		1883	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1884	{
		1885	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1886	}
		1887
		1888	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1889	{
		1890	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1891	}
		1892
		1893	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		1894	{
		1895	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		1896	}
		1897
		1898	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		1899	{
		1900	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		1901	}
		1902
		1903	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1904	{
		1905	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1906	}
		1907
1207	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1908	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1208	{	1909	{
1209	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1910	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1210	}	1911	}
1211		1912
…		…
1230	}	1931	}
1231		1932
1232	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	1933	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1233	{	1934	{
1234	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1935	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1936	}
		1937
		1938	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		1939	{
		1940	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1235	}	1941	}
1236		1942
1237	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	1943	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1238	{	1944	{
1239	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1945	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1303	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2009	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1304	{	2010	{
1305	return eio__1path (EIO_READLINK, path, pri, cb, data);	2011	return eio__1path (EIO_READLINK, path, pri, cb, data);
1306	}	2012	}
1307		2013
		2014	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2015	{
		2016	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2017	}
		2018
1308	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2019	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1309	{	2020	{
1310	return eio__1path (EIO_STAT, path, pri, cb, data);	2021	return eio__1path (EIO_STAT, path, pri, cb, data);
1311	}	2022	}
1312		2023
1313	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2024	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1314	{	2025	{
1315	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2026	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1316	}	2027	}
1317		2028
		2029	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2030	{
		2031	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2032	}
		2033
1318	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2034	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1319	{	2035	{
1320	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2036	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1321	}	2037	}
1322		2038
1323	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	2039	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1324	{	2040	{
1325	return eio__1path (EIO_RMDIR, path, pri, cb, data);	2041	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1326	}	2042	}
1327		2043
1328	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	2044	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1329	{	2045	{
1330	return eio__1path (EIO_READDIR, path, pri, cb, data);	2046	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1331	}	2047	}
1332		2048
1333	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2049	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1334	{	2050	{
1335	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2051	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1336	}	2052	}
1337		2053
1338	static eio_req *	2054	static eio_req *
1339	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2055	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1340	{	2056	{
…		…
1364	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2080	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1365	{	2081	{
1366	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2082	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1367	}	2083	}
1368		2084
1369	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2085	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1370	{	2086	{
1371	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2087	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1372	}	2088	}
1373		2089
1374	#endif	2090	#endif
…		…
1385	#undef SEND	2101	#undef SEND
1386		2102
1387	/*****************************************************************************/	2103	/*****************************************************************************/
1388	/* grp functions */	2104	/* grp functions */
1389		2105
		2106	void
1390	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2107	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1391	{	2108	{
1392	grp->int2 = limit;	2109	grp->int2 = limit;
1393	grp->feed = feed;	2110	grp->feed = feed;
1394		2111
1395	grp_try_feed (grp);	2112	grp_try_feed (grp);
1396	}	2113	}
1397		2114
		2115	void
1398	void eio_grp_limit (eio_req *grp, int limit)	2116	eio_grp_limit (eio_req *grp, int limit)
1399	{	2117	{
1400	grp->int2 = limit;	2118	grp->int2 = limit;
1401		2119
1402	grp_try_feed (grp);	2120	grp_try_feed (grp);
1403	}	2121	}
1404		2122
		2123	void
1405	void eio_grp_add (eio_req *grp, eio_req *req)	2124	eio_grp_add (eio_req *grp, eio_req *req)
1406	{	2125	{
1407	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2126	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2127
		2128	grp->flags |= EIO_FLAG_GROUPADD;
1408		2129
1409	++grp->size;	2130	++grp->size;
1410	req->grp = grp;	2131	req->grp = grp;
1411		2132
1412	req->grp_prev = 0;	2133	req->grp_prev = 0;
…		…
1419	}	2140	}
1420		2141
1421	/*****************************************************************************/	2142	/*****************************************************************************/
1422	/* misc garbage */	2143	/* misc garbage */
1423		2144
		2145	ssize_t
1424	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2146	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1425	{	2147	{
1426	etp_worker wrk;	2148	etp_worker wrk;
		2149	ssize_t ret;
1427		2150
1428	wrk.dbuf = 0;	2151	wrk.dbuf = 0;
1429		2152
1430	eio__sendfile (ofd, ifd, offset, count, &wrk);	2153	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1431		2154
1432	if (wrk.dbuf)	2155	if (wrk.dbuf)
1433	free (wrk.dbuf);	2156	free (wrk.dbuf);
1434	}
1435		2157
		2158	return ret;
		2159	}
		2160

Diff Legend

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