ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.16 by root, Fri May 30 04:50:21 2008 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 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);
373		452
374	want_poll_cb = want_poll;	453	want_poll_cb = want_poll;
375	done_poll_cb = done_poll;	454	done_poll_cb = done_poll;
		455
		456	return 0;
376	}	457	}
377		458
378	X_THREAD_PROC (etp_proc);	459	X_THREAD_PROC (etp_proc);
379		460
		461	static void ecb_cold
380	static void etp_start_thread (void)	462	etp_start_thread (void)
381	{	463	{
382	etp_worker *wrk = calloc (1, sizeof (etp_worker));	464	etp_worker *wrk = calloc (1, sizeof (etp_worker));
383		465
384	/*TODO*/	466	/*TODO*/
385	assert (("unable to allocate worker thread data", wrk));	467	assert (("unable to allocate worker thread data", wrk));
…		…
398	free (wrk);	480	free (wrk);
399		481
400	X_UNLOCK (wrklock);	482	X_UNLOCK (wrklock);
401	}	483	}
402		484
		485	static void
403	static void etp_maybe_start_thread (void)	486	etp_maybe_start_thread (void)
404	{	487	{
405	if (etp_nthreads () >= wanted)	488	if (ecb_expect_true (etp_nthreads () >= wanted))
406	return;	489	return;
407		490
408	/* todo: maybe use idle here, but might be less exact */	491	/* todo: maybe use idle here, but might be less exact */
409	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 ()))
410	return;	493	return;
411		494
412	etp_start_thread ();	495	etp_start_thread ();
413	}	496	}
414		497
		498	static void ecb_cold
415	static void etp_end_thread (void)	499	etp_end_thread (void)
416	{	500	{
417	eio_req *req = calloc (1, sizeof (eio_req));	501	eio_req *req = calloc (1, sizeof (eio_req));
418		502
419	req->type = -1;	503	req->type = -1;
420	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	504	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
427	X_LOCK (wrklock);	511	X_LOCK (wrklock);
428	--started;	512	--started;
429	X_UNLOCK (wrklock);	513	X_UNLOCK (wrklock);
430	}	514	}
431		515
432	static int etp_poll (void)	516	static int
		517	etp_poll (void)
433	{	518	{
434	unsigned int maxreqs;	519	unsigned int maxreqs;
435	unsigned int maxtime;	520	unsigned int maxtime;
436	struct timeval tv_start, tv_now;	521	struct timeval tv_start, tv_now;
437		522
…		…
467		552
468	X_LOCK (reqlock);	553	X_LOCK (reqlock);
469	--nreqs;	554	--nreqs;
470	X_UNLOCK (reqlock);	555	X_UNLOCK (reqlock);
471		556
472	if (req->type == EIO_GROUP && req->size)	557	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
473	{	558	{
474	req->int1 = 1; /* mark request as delayed */	559	req->int1 = 1; /* mark request as delayed */
475	continue;	560	continue;
476	}	561	}
477	else	562	else
478	{	563	{
479	int res = ETP_FINISH (req);	564	int res = ETP_FINISH (req);
480	if (res)	565	if (ecb_expect_false (res))
481	return res;	566	return res;
482	}	567	}
483		568
484	if (maxreqs && !--maxreqs)	569	if (ecb_expect_false (maxreqs && !--maxreqs))
485	break;	570	break;
486		571
487	if (maxtime)	572	if (maxtime)
488	{	573	{
489	gettimeofday (&tv_now, 0);	574	gettimeofday (&tv_now, 0);
…		…
495		580
496	errno = EAGAIN;	581	errno = EAGAIN;
497	return -1;	582	return -1;
498	}	583	}
499		584
		585	static void
500	static void etp_cancel (ETP_REQ *req)	586	etp_cancel (ETP_REQ *req)
501	{	587	{
502	X_LOCK (wrklock);	588	X_LOCK (wrklock);
503	req->flags |= EIO_FLAG_CANCELLED;	589	req->flags |= EIO_FLAG_CANCELLED;
504	X_UNLOCK (wrklock);	590	X_UNLOCK (wrklock);
505		591
506	eio_grp_cancel (req);	592	eio_grp_cancel (req);
507	}	593	}
508		594
		595	static void
509	static void etp_submit (ETP_REQ *req)	596	etp_submit (ETP_REQ *req)
510	{	597	{
511	req->pri -= ETP_PRI_MIN;	598	req->pri -= ETP_PRI_MIN;
512		599
513	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;
514	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;
515		602
		603	if (ecb_expect_false (req->type == EIO_GROUP))
		604	{
		605	/* I hope this is worth it :/ */
516	X_LOCK (reqlock);	606	X_LOCK (reqlock);
517	++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;
518	++nready;	623	++nready;
519	reqq_push (&req_queue, req);	624	reqq_push (&req_queue, req);
520	X_COND_SIGNAL (reqwait);	625	X_COND_SIGNAL (reqwait);
521	X_UNLOCK (reqlock);	626	X_UNLOCK (reqlock);
522		627
523	etp_maybe_start_thread ();	628	etp_maybe_start_thread ();
		629	}
524	}	630	}
525		631
		632	static void ecb_cold
526	static void etp_set_max_poll_time (double nseconds)	633	etp_set_max_poll_time (double nseconds)
527	{	634	{
528	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	635	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
529	max_poll_time = nseconds;	636	max_poll_time = nseconds * EIO_TICKS;
530	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	637	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
531	}	638	}
532		639
		640	static void ecb_cold
533	static void etp_set_max_poll_reqs (unsigned int maxreqs)	641	etp_set_max_poll_reqs (unsigned int maxreqs)
534	{	642	{
535	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	643	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
536	max_poll_reqs = maxreqs;	644	max_poll_reqs = maxreqs;
537	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	645	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
538	}	646	}
539		647
		648	static void ecb_cold
540	static void etp_set_max_idle (unsigned int nthreads)	649	etp_set_max_idle (unsigned int nthreads)
541	{	650	{
542	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	651	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
543	max_idle = nthreads <= 0 ? 1 : nthreads;	652	max_idle = nthreads;
544	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	653	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
545	}	654	}
546		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
547	static void etp_set_min_parallel (unsigned int nthreads)	665	etp_set_min_parallel (unsigned int nthreads)
548	{	666	{
549	if (wanted < nthreads)	667	if (wanted < nthreads)
550	wanted = nthreads;	668	wanted = nthreads;
551	}	669	}
552		670
		671	static void ecb_cold
553	static void etp_set_max_parallel (unsigned int nthreads)	672	etp_set_max_parallel (unsigned int nthreads)
554	{	673	{
555	if (wanted > nthreads)	674	if (wanted > nthreads)
556	wanted = nthreads;	675	wanted = nthreads;
557		676
558	while (started > wanted)	677	while (started > wanted)
559	etp_end_thread ();	678	etp_end_thread ();
560	}	679	}
561		680
562	/*****************************************************************************/	681	/*****************************************************************************/
563		682
		683	static void
564	static void grp_try_feed (eio_req *grp)	684	grp_try_feed (eio_req *grp)
565	{	685	{
566	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	686	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
567	{	687	{
568	int old_len = grp->size;	688	grp->flags &= ~EIO_FLAG_GROUPADD;
569		689
570	EIO_FEED (grp);	690	EIO_FEED (grp);
571		691
572	/* stop if no progress has been made */	692	/* stop if no progress has been made */
573	if (old_len == grp->size)	693	if (!(grp->flags & EIO_FLAG_GROUPADD))
574	{	694	{
575	grp->feed = 0;	695	grp->feed = 0;
576	break;	696	break;
577	}	697	}
578	}	698	}
579	}	699	}
580		700
		701	static int
581	static int grp_dec (eio_req *grp)	702	grp_dec (eio_req *grp)
582	{	703	{
583	--grp->size;	704	--grp->size;
584		705
585	/* call feeder, if applicable */	706	/* call feeder, if applicable */
586	grp_try_feed (grp);	707	grp_try_feed (grp);
…		…
590	return eio_finish (grp);	711	return eio_finish (grp);
591	else	712	else
592	return 0;	713	return 0;
593	}	714	}
594		715
		716	void
595	void eio_destroy (eio_req *req)	717	eio_destroy (eio_req *req)
596	{	718	{
597	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	719	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
598	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	720	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
599		721
600	EIO_DESTROY (req);	722	EIO_DESTROY (req);
601	}	723	}
602		724
		725	static int
603	static int eio_finish (eio_req *req)	726	eio_finish (eio_req *req)
604	{	727	{
605	int res = EIO_FINISH (req);	728	int res = EIO_FINISH (req);
606		729
607	if (req->grp)	730	if (req->grp)
608	{	731	{
…		…
625	eio_destroy (req);	748	eio_destroy (req);
626		749
627	return res;	750	return res;
628	}	751	}
629		752
		753	void
630	void eio_grp_cancel (eio_req *grp)	754	eio_grp_cancel (eio_req *grp)
631	{	755	{
632	for (grp = grp->grp_first; grp; grp = grp->grp_next)	756	for (grp = grp->grp_first; grp; grp = grp->grp_next)
633	eio_cancel (grp);	757	eio_cancel (grp);
634	}	758	}
635		759
		760	void
636	void eio_cancel (eio_req *req)	761	eio_cancel (eio_req *req)
637	{	762	{
638	etp_cancel (req);	763	etp_cancel (req);
639	}	764	}
640		765
		766	void
641	void eio_submit (eio_req *req)	767	eio_submit (eio_req *req)
642	{	768	{
643	etp_submit (req);	769	etp_submit (req);
644	}	770	}
645		771
646	unsigned int eio_nreqs (void)	772	unsigned int
		773	eio_nreqs (void)
647	{	774	{
648	return etp_nreqs ();	775	return etp_nreqs ();
649	}	776	}
650		777
651	unsigned int eio_nready (void)	778	unsigned int
		779	eio_nready (void)
652	{	780	{
653	return etp_nready ();	781	return etp_nready ();
654	}	782	}
655		783
656	unsigned int eio_npending (void)	784	unsigned int
		785	eio_npending (void)
657	{	786	{
658	return etp_npending ();	787	return etp_npending ();
659	}	788	}
660		789
661	unsigned int eio_nthreads (void)	790	unsigned int ecb_cold
		791	eio_nthreads (void)
662	{	792	{
663	return etp_nthreads ();	793	return etp_nthreads ();
664	}	794	}
665		795
		796	void ecb_cold
666	void eio_set_max_poll_time (double nseconds)	797	eio_set_max_poll_time (double nseconds)
667	{	798	{
668	etp_set_max_poll_time (nseconds);	799	etp_set_max_poll_time (nseconds);
669	}	800	}
670		801
		802	void ecb_cold
671	void eio_set_max_poll_reqs (unsigned int maxreqs)	803	eio_set_max_poll_reqs (unsigned int maxreqs)
672	{	804	{
673	etp_set_max_poll_reqs (maxreqs);	805	etp_set_max_poll_reqs (maxreqs);
674	}	806	}
675		807
		808	void ecb_cold
676	void eio_set_max_idle (unsigned int nthreads)	809	eio_set_max_idle (unsigned int nthreads)
677	{	810	{
678	etp_set_max_idle (nthreads);	811	etp_set_max_idle (nthreads);
679	}	812	}
680		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
681	void eio_set_min_parallel (unsigned int nthreads)	821	eio_set_min_parallel (unsigned int nthreads)
682	{	822	{
683	etp_set_min_parallel (nthreads);	823	etp_set_min_parallel (nthreads);
684	}	824	}
685		825
		826	void ecb_cold
686	void eio_set_max_parallel (unsigned int nthreads)	827	eio_set_max_parallel (unsigned int nthreads)
687	{	828	{
688	etp_set_max_parallel (nthreads);	829	etp_set_max_parallel (nthreads);
689	}	830	}
690		831
691	int eio_poll (void)	832	int eio_poll (void)
…		…
695		836
696	/*****************************************************************************/	837	/*****************************************************************************/
697	/* work around various missing functions */	838	/* work around various missing functions */
698		839
699	#if !HAVE_PREADWRITE	840	#if !HAVE_PREADWRITE
		841	# undef pread
		842	# undef pwrite
700	# define pread eio__pread	843	# define pread eio__pread
701	# define pwrite eio__pwrite	844	# define pwrite eio__pwrite
702
703	/*
704	* make our pread/pwrite safe against themselves, but not against
705	* normal read/write by using a mutex. slows down execution a lot,
706	* but that's your problem, not mine.
707	*/
708	static mutex_t preadwritelock = X_MUTEX_INIT;
709		845
710	static ssize_t	846	static ssize_t
711	eio__pread (int fd, void *buf, size_t count, off_t offset)	847	eio__pread (int fd, void *buf, size_t count, off_t offset)
712	{	848	{
713	ssize_t res;	849	ssize_t res;
…		…
731		867
732	X_LOCK (preadwritelock);	868	X_LOCK (preadwritelock);
733	ooffset = lseek (fd, 0, SEEK_CUR);	869	ooffset = lseek (fd, 0, SEEK_CUR);
734	lseek (fd, offset, SEEK_SET);	870	lseek (fd, offset, SEEK_SET);
735	res = write (fd, buf, count);	871	res = write (fd, buf, count);
736	lseek (fd, offset, SEEK_SET);	872	lseek (fd, ooffset, SEEK_SET);
737	X_UNLOCK (preadwritelock);	873	X_UNLOCK (preadwritelock);
738		874
739	return res;	875	return res;
740	}	876	}
741	#endif	877	#endif
742		878
743	#ifndef HAVE_FUTIMES	879	#ifndef HAVE_UTIMES
744		880
		881	# undef utimes
745	# define utimes(path,times) eio__utimes (path, times)	882	# define utimes(path,times) eio__utimes (path, times)
746	# define futimes(fd,times) eio__futimes (fd, times)
747		883
748	static int	884	static int
749	eio__utimes (const char *filename, const struct timeval times[2])	885	eio__utimes (const char *filename, const struct timeval times[2])
750	{	886	{
751	if (times)	887	if (times)
…		…
759	}	895	}
760	else	896	else
761	return utime (filename, 0);	897	return utime (filename, 0);
762	}	898	}
763		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
764	static int eio__futimes (int fd, const struct timeval tv[2])	908	eio__futimes (int fd, const struct timeval tv[2])
765	{	909	{
766	errno = ENOSYS;	910	errno = ENOSYS;
767	return -1;	911	return -1;
768	}	912	}
769		913
770	#endif	914	#endif
771		915
772	#if !HAVE_FDATASYNC	916	#if !HAVE_FDATASYNC
		917	# undef fdatasync
773	# define fdatasync fsync	918	# define fdatasync(fd) fsync (fd)
774	#endif	919	#endif
		920
		921	/* sync_file_range always needs emulation */
		922	static int
		923	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		924	{
		925	#if HAVE_SYNC_FILE_RANGE
		926	int res;
		927
		928	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		929	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		930	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		931	{
		932	flags = 0
		933	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		934	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		935	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		936	}
		937
		938	res = sync_file_range (fd, offset, nbytes, flags);
		939
		940	if (!res || errno != ENOSYS)
		941	return res;
		942	#endif
		943
		944	/* even though we could play tricks with the flags, it's better to always
		945	* call fdatasync, as that matches the expectation of its users best */
		946	return fdatasync (fd);
		947	}
775		948
776	#if !HAVE_READAHEAD	949	#if !HAVE_READAHEAD
		950	# undef readahead
777	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	951	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
778		952
779	static ssize_t	953	static ssize_t
780	eio__readahead (int fd, off_t offset, size_t count, worker *self)	954	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
781	{	955	{
782	size_t todo = count;	956	size_t todo = count;
783	dBUF;	957	dBUF;
784		958
785	while (todo > 0)	959	while (todo > 0)
…		…
799		973
800	/* sendfile always needs emulation */	974	/* sendfile always needs emulation */
801	static ssize_t	975	static ssize_t
802	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)
803	{	977	{
		978	ssize_t written = 0;
804	ssize_t res;	979	ssize_t res;
805		980
806	if (!count)	981	if (!count)
807	return 0;	982	return 0;
808		983
		984	for (;;)
		985	{
809	#if HAVE_SENDFILE	986	#if HAVE_SENDFILE
810	# if __linux	987	# if __linux
		988	off_t soffset = offset;
811	res = sendfile (ofd, ifd, &offset, count);	989	res = sendfile (ofd, ifd, &soffset, count);
812		990
813	# elif __freebsd	991	# elif __FreeBSD__
814	/*	992	/*
815	* 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
816	* wasted on making it similar to other I/O functions.	994	* wasted on making it similar to other I/O functions.
817	*/	995	*/
818	{
819	off_t sbytes;	996	off_t sbytes;
820	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	997	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
821		998
822	if (res < 0 && sbytes)	999	#if 0 /* according to the manpage, this is correct, but broken behaviour */
823	/* 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)
824	res = sbytes;	1004	res = sbytes;
825	}	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;
826		1018
827	# elif __hpux	1019	# elif __hpux
828	res = sendfile (ofd, ifd, offset, count, 0, 0);	1020	res = sendfile (ofd, ifd, offset, count, 0, 0);
829		1021
830	# elif __solaris	1022	# elif __solaris
831	{
832	struct sendfilevec vec;	1023	struct sendfilevec vec;
833	size_t sbytes;	1024	size_t sbytes;
834		1025
835	vec.sfv_fd = ifd;	1026	vec.sfv_fd = ifd;
836	vec.sfv_flag = 0;	1027	vec.sfv_flag = 0;
837	vec.sfv_off = offset;	1028	vec.sfv_off = offset;
838	vec.sfv_len = count;	1029	vec.sfv_len = count;
839		1030
840	res = sendfilev (ofd, &vec, 1, &sbytes);	1031	res = sendfilev (ofd, &vec, 1, &sbytes);
841		1032
842	if (res < 0 && sbytes)	1033	if (res < 0 && sbytes)
843	res = sbytes;	1034	res = sbytes;
844	}
845		1035
846	# 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
847	#else	1047	#else
848	res = -1;	1048	res = -1;
849	errno = ENOSYS;	1049	errno = ENOSYS;
850	#endif	1050	#endif
851		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
852	if (res < 0	1075	if (res < 0
853	&& (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 */
854	#if __solaris	1082	#if __solaris
855	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1083	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
856	#endif	1084	#endif
857	)	1085	)
858	)	1086	)
…		…
889	}	1117	}
890		1118
891	return res;	1119	return res;
892	}	1120	}
893		1121
		1122	static signed char
		1123	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1124	{
		1125	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1126	: a->inode < b->inode ? -1
		1127	: a->inode > b->inode ? 1
		1128	: 0;
		1129	}
		1130
		1131	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		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);
		1285	}
		1286
894	/* read a full directory */	1287	/* read a full directory */
895	static void	1288	static void
896	eio__scandir (eio_req *req, etp_worker *self)	1289	eio__scandir (eio_req *req, etp_worker *self)
897	{	1290	{
898	DIR *dirp;	1291	DIR *dirp;
899	EIO_STRUCT_DIRENT *entp;	1292	EIO_STRUCT_DIRENT *entp;
900	char *name, *names;	1293	char *name, *names;
901	int memlen = 4096;	1294	int namesalloc = 4096;
902	int memofs = 0;	1295	int namesoffs = 0;
		1296	int flags = req->int1;
		1297	eio_dirent *dents = 0;
		1298	int dentalloc = 128;
903	int res = 0;	1299	int dentoffs = 0;
		1300	ino_t inode_bits = 0;
		1301
		1302	req->result = -1;
		1303
		1304	if (!(flags & EIO_READDIR_DENTS))
		1305	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
904		1306
905	X_LOCK (wrklock);	1307	X_LOCK (wrklock);
		1308	/* the corresponding closedir is in ETP_WORKER_CLEAR */
906	self->dirp = dirp = opendir (req->ptr1);	1309	self->dirp = dirp = opendir (req->ptr1);
		1310
907	req->flags |= EIO_FLAG_PTR2_FREE;	1311	if (req->flags & EIO_FLAG_PTR1_FREE)
		1312	free (req->ptr1);
		1313
		1314	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1315	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
908	req->ptr2 = names = malloc (memlen);	1316	req->ptr2 = names = malloc (namesalloc);
909	X_UNLOCK (wrklock);	1317	X_UNLOCK (wrklock);
910		1318
911	if (dirp && names)	1319	if (dirp && names && (!flags || dents))
912	for (;;)	1320	for (;;)
913	{	1321	{
914	errno = 0;	1322	errno = 0;
915	entp = readdir (dirp);	1323	entp = readdir (dirp);
916		1324
917	if (!entp)	1325	if (!entp)
		1326	{
		1327	if (errno)
		1328	break;
		1329
		1330	/* sort etc. */
		1331	req->int1 = flags;
		1332	req->result = dentoffs;
		1333
		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
		1340	{
		1341	/* in this case, all is known, and we just put dirs first and sort them */
		1342	eio_dirent *oth = dents + dentoffs;
		1343	eio_dirent *dir = dents;
		1344
		1345	/* now partition dirs to the front, and non-dirs to the back */
		1346	/* by walking from both sides and swapping if necessary */
		1347	while (oth > dir)
		1348	{
		1349	if (dir->type == EIO_DT_DIR)
		1350	++dir;
		1351	else if ((--oth)->type == EIO_DT_DIR)
		1352	{
		1353	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1354
		1355	++dir;
		1356	}
		1357	}
		1358
		1359	/* now sort the dirs only (dirs all have the same score) */
		1360	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1361	}
		1362
918	break;	1363	break;
		1364	}
919		1365
		1366	/* now add the entry to our list(s) */
920	name = entp->d_name;	1367	name = entp->d_name;
921		1368
		1369	/* skip . and .. entries */
922	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1370	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
923	{	1371	{
924	int len = strlen (name) + 1;	1372	int len = D_NAMLEN (entp) + 1;
925		1373
926	res++;	1374	while (ecb_expect_false (namesoffs + len > namesalloc))
927
928	while (memofs + len > memlen)
929	{	1375	{
930	memlen *= 2;	1376	namesalloc *= 2;
931	X_LOCK (wrklock);	1377	X_LOCK (wrklock);
932	req->ptr2 = names = realloc (names, memlen);	1378	req->ptr2 = names = realloc (names, namesalloc);
933	X_UNLOCK (wrklock);	1379	X_UNLOCK (wrklock);
934		1380
935	if (!names)	1381	if (!names)
936	break;	1382	break;
937	}	1383	}
938		1384
939	memcpy (names + memofs, name, len);	1385	memcpy (names + namesoffs, name, len);
		1386
		1387	if (dents)
		1388	{
		1389	struct eio_dirent *ent;
		1390
		1391	if (ecb_expect_false (dentoffs == dentalloc))
		1392	{
		1393	dentalloc *= 2;
		1394	X_LOCK (wrklock);
		1395	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1396	X_UNLOCK (wrklock);
		1397
		1398	if (!dents)
		1399	break;
		1400	}
		1401
		1402	ent = dents + dentoffs;
		1403
		1404	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1405	ent->namelen = len - 1;
		1406	ent->inode = D_INO (entp);
		1407
		1408	inode_bits |= ent->inode;
		1409
		1410	switch (D_TYPE (entp))
		1411	{
		1412	default:
		1413	ent->type = EIO_DT_UNKNOWN;
		1414	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1415	break;
		1416
		1417	#ifdef DT_FIFO
		1418	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1419	#endif
		1420	#ifdef DT_CHR
		1421	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1422	#endif
		1423	#ifdef DT_MPC
		1424	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1425	#endif
		1426	#ifdef DT_DIR
		1427	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1428	#endif
		1429	#ifdef DT_NAM
		1430	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1431	#endif
		1432	#ifdef DT_BLK
		1433	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1434	#endif
		1435	#ifdef DT_MPB
		1436	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1437	#endif
		1438	#ifdef DT_REG
		1439	case DT_REG: ent->type = EIO_DT_REG; break;
		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
		1447	#ifdef DT_LNK
		1448	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1449	#endif
		1450	#ifdef DT_SOCK
		1451	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1452	#endif
		1453	#ifdef DT_DOOR
		1454	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1455	#endif
		1456	#ifdef DT_WHT
		1457	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1458	#endif
		1459	}
		1460
		1461	ent->score = 7;
		1462
		1463	if (flags & EIO_READDIR_DIRS_FIRST)
		1464	{
		1465	if (ent->type == EIO_DT_UNKNOWN)
		1466	{
		1467	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1468	ent->score = 1;
		1469	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1470	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1471	}
		1472	else if (ent->type == EIO_DT_DIR)
		1473	ent->score = 0;
		1474	}
		1475	}
		1476
940	memofs += len;	1477	namesoffs += len;
		1478	++dentoffs;
		1479	}
		1480
		1481	if (EIO_CANCELLED (req))
		1482	{
		1483	errno = ECANCELED;
		1484	break;
941	}	1485	}
942	}	1486	}
		1487	}
943		1488
944	if (errno)	1489	#ifdef PAGESIZE
945	res = -1;	1490	# define eio_pagesize() PAGESIZE
		1491	#else
		1492	static intptr_t
		1493	eio_pagesize (void)
		1494	{
		1495	static intptr_t page;
		1496
		1497	if (!page)
		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 */
		1529	#endif
		1530
		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	}
		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
946		1590	{
947	req->result = res;	1591	intptr_t addr = (intptr_t)mem;
		1592	intptr_t end = addr + len;
		1593	intptr_t page = eio_pagesize ();
		1594
		1595	if (addr < end)
		1596	if (flags & EIO_MT_MODIFY) /* modify */
		1597	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1598	else
		1599	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1600	}
		1601
		1602	return 0;
948	}	1603	}
949		1604
950	/*****************************************************************************/	1605	/*****************************************************************************/
951		1606
952	#define ALLOC(len) \	1607	#define ALLOC(len) \
…		…
984	if (req)	1639	if (req)
985	break;	1640	break;
986		1641
987	++idle;	1642	++idle;
988		1643
989	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1644	ts.tv_sec = time (0) + idle_timeout;
990	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1645	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
991	{	1646	{
992	if (idle > max_idle)	1647	if (idle > max_idle)
993	{	1648	{
994	--idle;	1649	--idle;
…		…
1037	return 0;	1692	return 0;
1038	}	1693	}
1039		1694
1040	/*****************************************************************************/	1695	/*****************************************************************************/
1041		1696
		1697	int ecb_cold
1042	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1698	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1043	{	1699	{
1044	etp_init (want_poll, done_poll);	1700	return etp_init (want_poll, done_poll);
1045	}	1701	}
1046		1702
		1703	ecb_inline void
1047	static void eio_api_destroy (eio_req *req)	1704	eio_api_destroy (eio_req *req)
1048	{	1705	{
1049	free (req);	1706	free (req);
1050	}	1707	}
1051		1708
1052	#define REQ(rtype) \	1709	#define REQ(rtype) \
…		…
1071	{ \	1728	{ \
1072	eio_api_destroy (req); \	1729	eio_api_destroy (req); \
1073	return 0; \	1730	return 0; \
1074	}	1731	}
1075		1732
		1733	static void
1076	static void eio_execute (etp_worker *self, eio_req *req)	1734	eio_execute (etp_worker *self, eio_req *req)
1077	{	1735	{
1078	errno = 0;
1079
1080	switch (req->type)	1736	switch (req->type)
1081	{	1737	{
1082	case EIO_READ: ALLOC (req->size);	1738	case EIO_READ: ALLOC (req->size);
1083	req->result = req->offs >= 0	1739	req->result = req->offs >= 0
1084	? pread (req->int1, req->ptr2, req->size, req->offs)	1740	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1094	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1750	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1095	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1751	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1096	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1752	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1097	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1753	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1754	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		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;
1099		1760
1100	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;
1101	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;
1102	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;
1103	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;
…		…
1113	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	1774	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1114	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	1775	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1115	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	1776	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1116	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;	1777	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1117		1778
1118	case EIO_READLINK: ALLOC (NAME_MAX);	1779	case EIO_READLINK: ALLOC (PATH_MAX);
1119	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	1780	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1120		1781
1121	case EIO_SYNC: req->result = 0; sync (); break;	1782	case EIO_SYNC: req->result = 0; sync (); break;
1122	case EIO_FSYNC: req->result = fsync (req->int1); break;	1783	case EIO_FSYNC: req->result = fsync (req->int1); break;
1123	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1784	case EIO_FDATASYNC: req->result = fdatasync (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;
		1787	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1788	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		1789	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1124		1790
1125	case EIO_READDIR: eio__scandir (req, self); break;	1791	case EIO_READDIR: eio__scandir (req, self); break;
1126		1792
1127	case EIO_BUSY:	1793	case EIO_BUSY:
1128	#ifdef _WIN32	1794	#ifdef _WIN32
1129	Sleep (req->nv1 * 1000.);	1795	Sleep (req->nv1 * 1e3);
1130	#else	1796	#else
1131	{	1797	{
1132	struct timeval tv;	1798	struct timeval tv;
1133		1799
1134	tv.tv_sec = req->nv1;	1800	tv.tv_sec = req->nv1;
1135	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	1801	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1136		1802
1137	req->result = select (0, 0, 0, 0, &tv);	1803	req->result = select (0, 0, 0, 0, &tv);
1138	}	1804	}
1139	#endif	1805	#endif
1140	break;	1806	break;
…		…
1155	times = tv;	1821	times = tv;
1156	}	1822	}
1157	else	1823	else
1158	times = 0;	1824	times = 0;
1159		1825
1160
1161	req->result = req->type == EIO_FUTIME	1826	req->result = req->type == EIO_FUTIME
1162	? futimes (req->int1, times)	1827	? futimes (req->int1, times)
1163	: utimes (req->ptr1, times);	1828	: utimes (req->ptr1, times);
1164	}	1829	}
		1830	break;
1165		1831
1166	case EIO_GROUP:	1832	case EIO_GROUP:
		1833	abort (); /* handled in eio_request */
		1834
1167	case EIO_NOP:	1835	case EIO_NOP:
1168	req->result = 0;	1836	req->result = 0;
1169	break;	1837	break;
1170		1838
1171	case EIO_CUSTOM:	1839	case EIO_CUSTOM:
1172	req->feed (req);	1840	req->feed (req);
1173	break;	1841	break;
1174		1842
1175	default:	1843	default:
		1844	errno = ENOSYS;
1176	req->result = -1;	1845	req->result = -1;
1177	break;	1846	break;
1178	}	1847	}
1179		1848
1180	req->errorno = errno;	1849	req->errorno = errno;
…		…
1200	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1869	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1201	{	1870	{
1202	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1871	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1203	}	1872	}
1204		1873
		1874	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1875	{
		1876	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1877	}
		1878
		1879	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1880	{
		1881	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1882	}
		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
		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)
		1895	{
		1896	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1897	}
		1898
1205	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1899	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1206	{	1900	{
1207	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1901	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1208	}	1902	}
1209		1903
…		…
1228	}	1922	}
1229		1923
1230	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)
1231	{	1925	{
1232	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;
1233	}	1932	}
1234		1933
1235	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)
1236	{	1935	{
1237	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;
…		…
1311	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)
1312	{	2011	{
1313	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2012	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1314	}	2013	}
1315		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
1316	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)
1317	{	2021	{
1318	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2022	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1319	}	2023	}
1320		2024
1321	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	2025	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1322	{	2026	{
1323	return eio__1path (EIO_RMDIR, path, pri, cb, data);	2027	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1324	}	2028	}
1325		2029
1326	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	2030	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1327	{	2031	{
1328	return eio__1path (EIO_READDIR, path, pri, cb, data);	2032	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1329	}	2033	}
1330		2034
1331	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)
1332	{	2036	{
1333	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	2037	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1334	}	2038	}
1335		2039
1336	static eio_req *	2040	static eio_req *
1337	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)
1338	{	2042	{
…		…
1362	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)
1363	{	2067	{
1364	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2068	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1365	}	2069	}
1366		2070
1367	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)
1368	{	2072	{
1369	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2073	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1370	}	2074	}
1371		2075
1372	#endif	2076	#endif
…		…
1383	#undef SEND	2087	#undef SEND
1384		2088
1385	/*****************************************************************************/	2089	/*****************************************************************************/
1386	/* grp functions */	2090	/* grp functions */
1387		2091
		2092	void
1388	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)
1389	{	2094	{
1390	grp->int2 = limit;	2095	grp->int2 = limit;
1391	grp->feed = feed;	2096	grp->feed = feed;
1392		2097
1393	grp_try_feed (grp);	2098	grp_try_feed (grp);
1394	}	2099	}
1395		2100
		2101	void
1396	void eio_grp_limit (eio_req *grp, int limit)	2102	eio_grp_limit (eio_req *grp, int limit)
1397	{	2103	{
1398	grp->int2 = limit;	2104	grp->int2 = limit;
1399		2105
1400	grp_try_feed (grp);	2106	grp_try_feed (grp);
1401	}	2107	}
1402		2108
		2109	void
1403	void eio_grp_add (eio_req *grp, eio_req *req)	2110	eio_grp_add (eio_req *grp, eio_req *req)
1404	{	2111	{
1405	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));
		2113
		2114	grp->flags |= EIO_FLAG_GROUPADD;
1406		2115
1407	++grp->size;	2116	++grp->size;
1408	req->grp = grp;	2117	req->grp = grp;
1409		2118
1410	req->grp_prev = 0;	2119	req->grp_prev = 0;
…		…
1417	}	2126	}
1418		2127
1419	/*****************************************************************************/	2128	/*****************************************************************************/
1420	/* misc garbage */	2129	/* misc garbage */
1421		2130
		2131	ssize_t
1422	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)
1423	{	2133	{
1424	etp_worker wrk;	2134	etp_worker wrk;
		2135	ssize_t ret;
1425		2136
1426	wrk.dbuf = 0;	2137	wrk.dbuf = 0;
1427		2138
1428	eio__sendfile (ofd, ifd, offset, count, &wrk);	2139	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1429		2140
1430	if (wrk.dbuf)	2141	if (wrk.dbuf)
1431	free (wrk.dbuf);	2142	free (wrk.dbuf);
1432	}
1433		2143
		2144	return ret;
		2145	}
		2146

Diff Legend

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