ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.32 by root, Sat Jun 6 18:06:55 2009 UTC vs.
Revision 1.89 by root, Thu Jul 14 22:36:17 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5	* All rights reserved.	5	* All rights reserved.
6	*	6	*
7	* Redistribution and use in source and binary forms, with or without modifica-	7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:	8	* tion, are permitted provided that the following conditions are met:
9	*	9	*
…		…
35	* and other provisions required by the GPL. If you do not delete the	35	* and other provisions required by the GPL. If you do not delete the
36	* provisions above, a recipient may use your version of this file under	36	* provisions above, a recipient may use your version of this file under
37	* either the BSD or the GPL.	37	* either the BSD or the GPL.
38	*/	38	*/
39		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
40	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
41		46
42	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
43	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
44	#endif	49	#endif
45	#include "xthread.h"	50	#include "xthread.h"
…		…
53	#include <sys/stat.h>	58	#include <sys/stat.h>
54	#include <limits.h>	59	#include <limits.h>
55	#include <fcntl.h>	60	#include <fcntl.h>
56	#include <assert.h>	61	#include <assert.h>
57		62
		63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		65	#if HAVE_STDINT_H
		66	# include <stdint.h>
		67	#endif
		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72
		73	static void eio_destroy (eio_req *req);
		74
58	#ifndef EIO_FINISH	75	#ifndef EIO_FINISH
59	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	76	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60	#endif	77	#endif
61		78
62	#ifndef EIO_DESTROY	79	#ifndef EIO_DESTROY
…		…
65		82
66	#ifndef EIO_FEED	83	#ifndef EIO_FEED
67	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	84	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68	#endif	85	#endif
69		86
		87	#ifndef EIO_FD_TO_WIN32_HANDLE
		88	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		89	#endif
		90	#ifndef EIO_WIN32_HANDLE_TO_FD
		91	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		92	#endif
		93
		94	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		95
		96	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		97
70	#ifdef _WIN32	98	#ifdef _WIN32
71		99
72	/*doh*/	100	#define PAGESIZE 4096 /* GetSystemInfo? */
		101
		102	#ifdef EIO_STRUCT_STATI64
		103	#define stat(path,buf) _stati64 (path,buf)
		104	#define fstat(fd,buf) _fstati64 (path,buf)
		105	#endif
		106	#define lstat(path,buf) stat (path,buf)
		107	#define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		108	#define mkdir(path,mode) _mkdir (path)
		109	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		110
		111	#define chown(path,uid,gid) EIO_ENOSYS ()
		112	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		113	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		114	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		115	#define mknod(path,mode,dev) EIO_ENOSYS ()
		116	#define sync() EIO_ENOSYS ()
		117
		118	/* we could even stat and see if it exists */
		119	static int
		120	symlink (const char *old, const char *neu)
		121	{
		122	if (CreateSymbolicLink (neu, old, 1))
		123	return 0;
		124
		125	if (CreateSymbolicLink (neu, old, 0))
		126	return 0;
		127
		128	return EIO_ERRNO (ENOENT, -1);
		129	}
		130
73	#else	131	#else
74		132
75	# include "config.h"
76	# include <sys/time.h>	133	#include <sys/time.h>
77	# include <sys/select.h>	134	#include <sys/select.h>
		135	#include <sys/statvfs.h>
78	# include <unistd.h>	136	#include <unistd.h>
79	# include <utime.h>	137	#include <utime.h>
80	# include <signal.h>	138	#include <signal.h>
81	# include <dirent.h>	139	#include <dirent.h>
82		140
		141	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		142	#include <sys/mman.h>
		143	#endif
		144
83	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	145	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)	146	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
85	# define D_INO(de) (de)->d_fileno
86	# define _DIRENT_HAVE_D_TYPE /* sigh */	147	#define _DIRENT_HAVE_D_TYPE /* sigh */
87	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600	148	#define D_INO(de) (de)->d_fileno
		149	#define D_NAMLEN(de) (de)->d_namlen
		150	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
88	# define D_INO(de) (de)->d_ino	151	#define D_INO(de) (de)->d_ino
89	# endif	152	#endif
90		153
		154	#ifdef _D_EXACT_NAMLEN
		155	#undef D_NAMLEN
		156	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		157	#endif
		158
91	# ifdef _DIRENT_HAVE_D_TYPE	159	#ifdef _DIRENT_HAVE_D_TYPE
92	# define D_TYPE(de) (de)->d_type	160	#define D_TYPE(de) (de)->d_type
93	# endif	161	#endif
94		162
95	# ifndef EIO_STRUCT_DIRENT	163	#ifndef EIO_STRUCT_DIRENT
96	# define EIO_STRUCT_DIRENT struct dirent	164	#define EIO_STRUCT_DIRENT struct dirent
97	# endif	165	#endif
98		166
99	#endif	167	#endif
100		168
101	#if HAVE_SENDFILE	169	#if HAVE_SENDFILE
102	# if __linux	170	# if __linux
103	# include <sys/sendfile.h>	171	# include <sys/sendfile.h>
104	# elif __freebsd	172	# elif __FreeBSD__ || defined __APPLE__
105	# include <sys/socket.h>	173	# include <sys/socket.h>
106	# include <sys/uio.h>	174	# include <sys/uio.h>
107	# elif __hpux	175	# elif __hpux
108	# include <sys/socket.h>	176	# include <sys/socket.h>
109	# elif __solaris /* not yet */	177	# elif __solaris
110	# include <sys/sendfile.h>	178	# include <sys/sendfile.h>
111	# else	179	# else
112	# error sendfile support requested but not available	180	# error sendfile support requested but not available
113	# endif	181	# endif
114	#endif	182	#endif
115		183
116	#ifndef D_TYPE	184	#ifndef D_TYPE
117	# define D_TYPE(de) 0	185	# define D_TYPE(de) 0
118	#endif	186	#endif
119
120	#ifndef D_INO	187	#ifndef D_INO
121	# define D_INO(de) 0	188	# define D_INO(de) 0
122	#endif	189	#endif
123		190	#ifndef D_NAMLEN
124	/* number of seconds after which an idle threads exit */	191	# define D_NAMLEN(de) strlen ((de)->d_name)
125	#define IDLE_TIMEOUT 10	192	#endif
126		193
127	/* used for struct dirent, AIX doesn't provide it */	194	/* used for struct dirent, AIX doesn't provide it */
128	#ifndef NAME_MAX	195	#ifndef NAME_MAX
129	# define NAME_MAX 4096	196	# define NAME_MAX 4096
		197	#endif
		198
		199	/* used for readlink etc. */
		200	#ifndef PATH_MAX
		201	# define PATH_MAX 4096
130	#endif	202	#endif
131		203
132	/* buffer size for various temporary buffers */	204	/* buffer size for various temporary buffers */
133	#define EIO_BUFSIZE 65536	205	#define EIO_BUFSIZE 65536
134		206
…		…
140	errno = ENOMEM; \	212	errno = ENOMEM; \
141	if (!eio_buf) \	213	if (!eio_buf) \
142	return -1;	214	return -1;
143		215
144	#define EIO_TICKS ((1000000 + 1023) >> 10)	216	#define EIO_TICKS ((1000000 + 1023) >> 10)
145
146	/*****************************************************************************/
147
148	#if __GNUC__ >= 3
149	# define expect(expr,value) __builtin_expect ((expr),(value))
150	#else
151	# define expect(expr,value) (expr)
152	#endif
153
154	#define expect_false(expr) expect ((expr) != 0, 0)
155	#define expect_true(expr) expect ((expr) != 0, 1)
156
157	/*****************************************************************************/
158		217
159	#define ETP_PRI_MIN EIO_PRI_MIN	218	#define ETP_PRI_MIN EIO_PRI_MIN
160	#define ETP_PRI_MAX EIO_PRI_MAX	219	#define ETP_PRI_MAX EIO_PRI_MAX
161		220
162	struct etp_worker;	221	struct etp_worker;
…		…
187		246
188	/*****************************************************************************/	247	/*****************************************************************************/
189		248
190	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	249	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
191		250
192	/* calculcate time difference in ~1/EIO_TICKS of a second */	251	/* calculate time difference in ~1/EIO_TICKS of a second */
		252	ecb_inline int
193	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	253	tvdiff (struct timeval *tv1, struct timeval *tv2)
194	{	254	{
195	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	255	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
196	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	256	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
197	}	257	}
198		258
…		…
205	static unsigned int max_poll_reqs; /* reslock */	265	static unsigned int max_poll_reqs; /* reslock */
206		266
207	static volatile unsigned int nreqs; /* reqlock */	267	static volatile unsigned int nreqs; /* reqlock */
208	static volatile unsigned int nready; /* reqlock */	268	static volatile unsigned int nready; /* reqlock */
209	static volatile unsigned int npending; /* reqlock */	269	static volatile unsigned int npending; /* reqlock */
210	static volatile unsigned int max_idle = 4;	270	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		271	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
211		272
212	static mutex_t wrklock = X_MUTEX_INIT;	273	static xmutex_t wrklock;
213	static mutex_t reslock = X_MUTEX_INIT;	274	static xmutex_t reslock;
214	static mutex_t reqlock = X_MUTEX_INIT;	275	static xmutex_t reqlock;
215	static cond_t reqwait = X_COND_INIT;	276	static xcond_t reqwait;
216		277
217	#if !HAVE_PREADWRITE	278	#if !HAVE_PREADWRITE
218	/*	279	/*
219	* make our pread/pwrite emulation safe against themselves, but not against	280	* make our pread/pwrite emulation safe against themselves, but not against
220	* normal read/write by using a mutex. slows down execution a lot,	281	* normal read/write by using a mutex. slows down execution a lot,
221	* but that's your problem, not mine.	282	* but that's your problem, not mine.
222	*/	283	*/
223	static mutex_t preadwritelock = X_MUTEX_INIT;	284	static xmutex_t preadwritelock = X_MUTEX_INIT;
224	#endif	285	#endif
225		286
226	typedef struct etp_worker	287	typedef struct etp_worker
227	{	288	{
228	/* locked by wrklock */	289	/* locked by wrklock */
229	struct etp_worker *prev, *next;	290	struct etp_worker *prev, *next;
230		291
231	thread_t tid;	292	xthread_t tid;
232		293
233	/* locked by reslock, reqlock or wrklock */	294	/* locked by reslock, reqlock or wrklock */
234	ETP_REQ *req; /* currently processed request */	295	ETP_REQ *req; /* currently processed request */
235		296
236	ETP_WORKER_COMMON	297	ETP_WORKER_COMMON
…		…
241	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	302	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
242	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	303	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
243		304
244	/* worker threads management */	305	/* worker threads management */
245		306
		307	static void ecb_cold
246	static void etp_worker_clear (etp_worker *wrk)	308	etp_worker_clear (etp_worker *wrk)
247	{	309	{
248	ETP_WORKER_CLEAR (wrk);	310	ETP_WORKER_CLEAR (wrk);
249	}	311	}
250		312
		313	static void ecb_cold
251	static void etp_worker_free (etp_worker *wrk)	314	etp_worker_free (etp_worker *wrk)
252	{	315	{
253	wrk->next->prev = wrk->prev;	316	wrk->next->prev = wrk->prev;
254	wrk->prev->next = wrk->next;	317	wrk->prev->next = wrk->next;
255		318
256	free (wrk);	319	free (wrk);
257	}	320	}
258		321
259	static unsigned int etp_nreqs (void)	322	static unsigned int
		323	etp_nreqs (void)
260	{	324	{
261	int retval;	325	int retval;
262	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	326	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
263	retval = nreqs;	327	retval = nreqs;
264	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	328	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
265	return retval;	329	return retval;
266	}	330	}
267		331
268	static unsigned int etp_nready (void)	332	static unsigned int
		333	etp_nready (void)
269	{	334	{
270	unsigned int retval;	335	unsigned int retval;
271		336
272	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	337	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273	retval = nready;	338	retval = nready;
274	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	339	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275		340
276	return retval;	341	return retval;
277	}	342	}
278		343
279	static unsigned int etp_npending (void)	344	static unsigned int
		345	etp_npending (void)
280	{	346	{
281	unsigned int retval;	347	unsigned int retval;
282		348
283	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	349	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
284	retval = npending;	350	retval = npending;
285	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	351	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
286		352
287	return retval;	353	return retval;
288	}	354	}
289		355
290	static unsigned int etp_nthreads (void)	356	static unsigned int
		357	etp_nthreads (void)
291	{	358	{
292	unsigned int retval;	359	unsigned int retval;
293		360
294	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	361	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
295	retval = started;	362	retval = started;
…		…
309	} etp_reqq;	376	} etp_reqq;
310		377
311	static etp_reqq req_queue;	378	static etp_reqq req_queue;
312	static etp_reqq res_queue;	379	static etp_reqq res_queue;
313		380
		381	static int ecb_noinline
314	static int reqq_push (etp_reqq *q, ETP_REQ *req)	382	reqq_push (etp_reqq *q, ETP_REQ *req)
315	{	383	{
316	int pri = req->pri;	384	int pri = req->pri;
317	req->next = 0;	385	req->next = 0;
318		386
319	if (q->qe[pri])	387	if (q->qe[pri])
…		…
325	q->qe[pri] = q->qs[pri] = req;	393	q->qe[pri] = q->qs[pri] = req;
326		394
327	return q->size++;	395	return q->size++;
328	}	396	}
329		397
		398	static ETP_REQ * ecb_noinline
330	static ETP_REQ *reqq_shift (etp_reqq *q)	399	reqq_shift (etp_reqq *q)
331	{	400	{
332	int pri;	401	int pri;
333		402
334	if (!q->size)	403	if (!q->size)
335	return 0;	404	return 0;
…		…
350	}	419	}
351		420
352	abort ();	421	abort ();
353	}	422	}
354		423
355	static void etp_atfork_prepare (void)	424	static void ecb_cold
		425	etp_thread_init (void)
356	{	426	{
357	X_LOCK (wrklock);
358	X_LOCK (reqlock);
359	X_LOCK (reslock);
360	#if !HAVE_PREADWRITE	427	#if !HAVE_PREADWRITE
361	X_LOCK (preadwritelock);	428	X_MUTEX_CREATE (preadwritelock);
362	#endif	429	#endif
		430	X_MUTEX_CREATE (wrklock);
		431	X_MUTEX_CREATE (reslock);
		432	X_MUTEX_CREATE (reqlock);
		433	X_COND_CREATE (reqwait);
363	}	434	}
364		435
		436	static void ecb_cold
		437	etp_atfork_prepare (void)
		438	{
		439	}
		440
		441	static void ecb_cold
365	static void etp_atfork_parent (void)	442	etp_atfork_parent (void)
366	{	443	{
367	#if !HAVE_PREADWRITE
368	X_UNLOCK (preadwritelock);
369	#endif
370	X_UNLOCK (reslock);
371	X_UNLOCK (reqlock);
372	X_UNLOCK (wrklock);
373	}	444	}
374		445
		446	static void ecb_cold
375	static void etp_atfork_child (void)	447	etp_atfork_child (void)
376	{	448	{
377	ETP_REQ *prv;	449	ETP_REQ *prv;
378		450
379	while ((prv = reqq_shift (&req_queue)))	451	while ((prv = reqq_shift (&req_queue)))
380	ETP_DESTROY (prv);	452	ETP_DESTROY (prv);
…		…
397	idle = 0;	469	idle = 0;
398	nreqs = 0;	470	nreqs = 0;
399	nready = 0;	471	nready = 0;
400	npending = 0;	472	npending = 0;
401		473
402	etp_atfork_parent ();	474	etp_thread_init ();
403	}	475	}
404		476
405	static void	477	static void ecb_cold
406	etp_once_init (void)	478	etp_once_init (void)
407	{	479	{
		480	etp_thread_init ();
408	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	481	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
409	}	482	}
410		483
411	static int	484	static int ecb_cold
412	etp_init (void (*want_poll)(void), void (*done_poll)(void))	485	etp_init (void (*want_poll)(void), void (*done_poll)(void))
413	{	486	{
414	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	487	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
415		488
416	pthread_once (&doinit, etp_once_init);	489	pthread_once (&doinit, etp_once_init);
…		…
421	return 0;	494	return 0;
422	}	495	}
423		496
424	X_THREAD_PROC (etp_proc);	497	X_THREAD_PROC (etp_proc);
425		498
		499	static void ecb_cold
426	static void etp_start_thread (void)	500	etp_start_thread (void)
427	{	501	{
428	etp_worker *wrk = calloc (1, sizeof (etp_worker));	502	etp_worker *wrk = calloc (1, sizeof (etp_worker));
429		503
430	/*TODO*/	504	/*TODO*/
431	assert (("unable to allocate worker thread data", wrk));	505	assert (("unable to allocate worker thread data", wrk));
…		…
444	free (wrk);	518	free (wrk);
445		519
446	X_UNLOCK (wrklock);	520	X_UNLOCK (wrklock);
447	}	521	}
448		522
		523	static void
449	static void etp_maybe_start_thread (void)	524	etp_maybe_start_thread (void)
450	{	525	{
451	if (expect_true (etp_nthreads () >= wanted))	526	if (ecb_expect_true (etp_nthreads () >= wanted))
452	return;	527	return;
453		528
454	/* todo: maybe use idle here, but might be less exact */	529	/* todo: maybe use idle here, but might be less exact */
455	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	530	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
456	return;	531	return;
457		532
458	etp_start_thread ();	533	etp_start_thread ();
459	}	534	}
460		535
		536	static void ecb_cold
461	static void etp_end_thread (void)	537	etp_end_thread (void)
462	{	538	{
463	eio_req *req = calloc (1, sizeof (eio_req));	539	eio_req *req = calloc (1, sizeof (eio_req));
464		540
465	req->type = -1;	541	req->type = -1;
466	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	542	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
473	X_LOCK (wrklock);	549	X_LOCK (wrklock);
474	--started;	550	--started;
475	X_UNLOCK (wrklock);	551	X_UNLOCK (wrklock);
476	}	552	}
477		553
478	static int etp_poll (void)	554	static int
		555	etp_poll (void)
479	{	556	{
480	unsigned int maxreqs;	557	unsigned int maxreqs;
481	unsigned int maxtime;	558	unsigned int maxtime;
482	struct timeval tv_start, tv_now;	559	struct timeval tv_start, tv_now;
483		560
…		…
513		590
514	X_LOCK (reqlock);	591	X_LOCK (reqlock);
515	--nreqs;	592	--nreqs;
516	X_UNLOCK (reqlock);	593	X_UNLOCK (reqlock);
517		594
518	if (expect_false (req->type == EIO_GROUP && req->size))	595	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
519	{	596	{
520	req->int1 = 1; /* mark request as delayed */	597	req->int1 = 1; /* mark request as delayed */
521	continue;	598	continue;
522	}	599	}
523	else	600	else
524	{	601	{
525	int res = ETP_FINISH (req);	602	int res = ETP_FINISH (req);
526	if (expect_false (res))	603	if (ecb_expect_false (res))
527	return res;	604	return res;
528	}	605	}
529		606
530	if (expect_false (maxreqs && !--maxreqs))	607	if (ecb_expect_false (maxreqs && !--maxreqs))
531	break;	608	break;
532		609
533	if (maxtime)	610	if (maxtime)
534	{	611	{
535	gettimeofday (&tv_now, 0);	612	gettimeofday (&tv_now, 0);
…		…
541		618
542	errno = EAGAIN;	619	errno = EAGAIN;
543	return -1;	620	return -1;
544	}	621	}
545		622
		623	static void
546	static void etp_cancel (ETP_REQ *req)	624	etp_cancel (ETP_REQ *req)
547	{	625	{
548	X_LOCK (wrklock);	626	req->cancelled = 1;
549	req->flags |= EIO_FLAG_CANCELLED;
550	X_UNLOCK (wrklock);
551		627
552	eio_grp_cancel (req);	628	eio_grp_cancel (req);
553	}	629	}
554		630
		631	static void
555	static void etp_submit (ETP_REQ *req)	632	etp_submit (ETP_REQ *req)
556	{	633	{
557	req->pri -= ETP_PRI_MIN;	634	req->pri -= ETP_PRI_MIN;
558		635
559	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	636	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
560	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	637	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
561		638
562	if (expect_false (req->type == EIO_GROUP))	639	if (ecb_expect_false (req->type == EIO_GROUP))
563	{	640	{
564	/* I hope this is worth it :/ */	641	/* I hope this is worth it :/ */
565	X_LOCK (reqlock);	642	X_LOCK (reqlock);
566	++nreqs;	643	++nreqs;
567	X_UNLOCK (reqlock);	644	X_UNLOCK (reqlock);
…		…
586		663
587	etp_maybe_start_thread ();	664	etp_maybe_start_thread ();
588	}	665	}
589	}	666	}
590		667
		668	static void ecb_cold
591	static void etp_set_max_poll_time (double nseconds)	669	etp_set_max_poll_time (double nseconds)
592	{	670	{
593	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	671	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
594	max_poll_time = nseconds;	672	max_poll_time = nseconds * EIO_TICKS;
595	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	673	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
596	}	674	}
597		675
		676	static void ecb_cold
598	static void etp_set_max_poll_reqs (unsigned int maxreqs)	677	etp_set_max_poll_reqs (unsigned int maxreqs)
599	{	678	{
600	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	679	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
601	max_poll_reqs = maxreqs;	680	max_poll_reqs = maxreqs;
602	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	681	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
603	}	682	}
604		683
		684	static void ecb_cold
605	static void etp_set_max_idle (unsigned int nthreads)	685	etp_set_max_idle (unsigned int nthreads)
606	{	686	{
607	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	687	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
608	max_idle = nthreads <= 0 ? 1 : nthreads;	688	max_idle = nthreads;
609	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	689	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
610	}	690	}
611		691
		692	static void ecb_cold
		693	etp_set_idle_timeout (unsigned int seconds)
		694	{
		695	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		696	idle_timeout = seconds;
		697	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		698	}
		699
		700	static void ecb_cold
612	static void etp_set_min_parallel (unsigned int nthreads)	701	etp_set_min_parallel (unsigned int nthreads)
613	{	702	{
614	if (wanted < nthreads)	703	if (wanted < nthreads)
615	wanted = nthreads;	704	wanted = nthreads;
616	}	705	}
617		706
		707	static void ecb_cold
618	static void etp_set_max_parallel (unsigned int nthreads)	708	etp_set_max_parallel (unsigned int nthreads)
619	{	709	{
620	if (wanted > nthreads)	710	if (wanted > nthreads)
621	wanted = nthreads;	711	wanted = nthreads;
622		712
623	while (started > wanted)	713	while (started > wanted)
624	etp_end_thread ();	714	etp_end_thread ();
625	}	715	}
626		716
627	/*****************************************************************************/	717	/*****************************************************************************/
628		718
		719	static void
629	static void grp_try_feed (eio_req *grp)	720	grp_try_feed (eio_req *grp)
630	{	721	{
631	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	722	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
632	{	723	{
633	grp->flags &= ~EIO_FLAG_GROUPADD;	724	grp->flags &= ~EIO_FLAG_GROUPADD;
634		725
…		…
641	break;	732	break;
642	}	733	}
643	}	734	}
644	}	735	}
645		736
		737	static int
646	static int grp_dec (eio_req *grp)	738	grp_dec (eio_req *grp)
647	{	739	{
648	--grp->size;	740	--grp->size;
649		741
650	/* call feeder, if applicable */	742	/* call feeder, if applicable */
651	grp_try_feed (grp);	743	grp_try_feed (grp);
…		…
655	return eio_finish (grp);	747	return eio_finish (grp);
656	else	748	else
657	return 0;	749	return 0;
658	}	750	}
659		751
		752	static void
660	void eio_destroy (eio_req *req)	753	eio_destroy (eio_req *req)
661	{	754	{
662	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	755	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
663	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	756	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
664		757
665	EIO_DESTROY (req);	758	EIO_DESTROY (req);
666	}	759	}
667		760
		761	static int
668	static int eio_finish (eio_req *req)	762	eio_finish (eio_req *req)
669	{	763	{
670	int res = EIO_FINISH (req);	764	int res = EIO_FINISH (req);
671		765
672	if (req->grp)	766	if (req->grp)
673	{	767	{
…		…
681	if (grp->grp_first == req)	775	if (grp->grp_first == req)
682	grp->grp_first = req->grp_next;	776	grp->grp_first = req->grp_next;
683		777
684	res2 = grp_dec (grp);	778	res2 = grp_dec (grp);
685		779
686	if (!res && res2)	780	if (!res)
687	res = res2;	781	res = res2;
688	}	782	}
689		783
690	eio_destroy (req);	784	eio_destroy (req);
691		785
692	return res;	786	return res;
693	}	787	}
694		788
		789	void
695	void eio_grp_cancel (eio_req *grp)	790	eio_grp_cancel (eio_req *grp)
696	{	791	{
697	for (grp = grp->grp_first; grp; grp = grp->grp_next)	792	for (grp = grp->grp_first; grp; grp = grp->grp_next)
698	eio_cancel (grp);	793	eio_cancel (grp);
699	}	794	}
700		795
		796	void
701	void eio_cancel (eio_req *req)	797	eio_cancel (eio_req *req)
702	{	798	{
703	etp_cancel (req);	799	etp_cancel (req);
704	}	800	}
705		801
		802	void
706	void eio_submit (eio_req *req)	803	eio_submit (eio_req *req)
707	{	804	{
708	etp_submit (req);	805	etp_submit (req);
709	}	806	}
710		807
711	unsigned int eio_nreqs (void)	808	unsigned int
		809	eio_nreqs (void)
712	{	810	{
713	return etp_nreqs ();	811	return etp_nreqs ();
714	}	812	}
715		813
716	unsigned int eio_nready (void)	814	unsigned int
		815	eio_nready (void)
717	{	816	{
718	return etp_nready ();	817	return etp_nready ();
719	}	818	}
720		819
721	unsigned int eio_npending (void)	820	unsigned int
		821	eio_npending (void)
722	{	822	{
723	return etp_npending ();	823	return etp_npending ();
724	}	824	}
725		825
726	unsigned int eio_nthreads (void)	826	unsigned int ecb_cold
		827	eio_nthreads (void)
727	{	828	{
728	return etp_nthreads ();	829	return etp_nthreads ();
729	}	830	}
730		831
		832	void ecb_cold
731	void eio_set_max_poll_time (double nseconds)	833	eio_set_max_poll_time (double nseconds)
732	{	834	{
733	etp_set_max_poll_time (nseconds);	835	etp_set_max_poll_time (nseconds);
734	}	836	}
735		837
		838	void ecb_cold
736	void eio_set_max_poll_reqs (unsigned int maxreqs)	839	eio_set_max_poll_reqs (unsigned int maxreqs)
737	{	840	{
738	etp_set_max_poll_reqs (maxreqs);	841	etp_set_max_poll_reqs (maxreqs);
739	}	842	}
740		843
		844	void ecb_cold
741	void eio_set_max_idle (unsigned int nthreads)	845	eio_set_max_idle (unsigned int nthreads)
742	{	846	{
743	etp_set_max_idle (nthreads);	847	etp_set_max_idle (nthreads);
744	}	848	}
745		849
		850	void ecb_cold
		851	eio_set_idle_timeout (unsigned int seconds)
		852	{
		853	etp_set_idle_timeout (seconds);
		854	}
		855
		856	void ecb_cold
746	void eio_set_min_parallel (unsigned int nthreads)	857	eio_set_min_parallel (unsigned int nthreads)
747	{	858	{
748	etp_set_min_parallel (nthreads);	859	etp_set_min_parallel (nthreads);
749	}	860	}
750		861
		862	void ecb_cold
751	void eio_set_max_parallel (unsigned int nthreads)	863	eio_set_max_parallel (unsigned int nthreads)
752	{	864	{
753	etp_set_max_parallel (nthreads);	865	etp_set_max_parallel (nthreads);
754	}	866	}
755		867
756	int eio_poll (void)	868	int eio_poll (void)
…		…
765	# undef pread	877	# undef pread
766	# undef pwrite	878	# undef pwrite
767	# define pread eio__pread	879	# define pread eio__pread
768	# define pwrite eio__pwrite	880	# define pwrite eio__pwrite
769		881
770	static ssize_t	882	static eio_ssize_t
771	eio__pread (int fd, void *buf, size_t count, off_t offset)	883	eio__pread (int fd, void *buf, size_t count, off_t offset)
772	{	884	{
773	ssize_t res;	885	eio_ssize_t res;
774	off_t ooffset;	886	off_t ooffset;
775		887
776	X_LOCK (preadwritelock);	888	X_LOCK (preadwritelock);
777	ooffset = lseek (fd, 0, SEEK_CUR);	889	ooffset = lseek (fd, 0, SEEK_CUR);
778	lseek (fd, offset, SEEK_SET);	890	lseek (fd, offset, SEEK_SET);
…		…
781	X_UNLOCK (preadwritelock);	893	X_UNLOCK (preadwritelock);
782		894
783	return res;	895	return res;
784	}	896	}
785		897
786	static ssize_t	898	static eio_ssize_t
787	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	899	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
788	{	900	{
789	ssize_t res;	901	eio_ssize_t res;
790	off_t ooffset;	902	off_t ooffset;
791		903
792	X_LOCK (preadwritelock);	904	X_LOCK (preadwritelock);
793	ooffset = lseek (fd, 0, SEEK_CUR);	905	ooffset = lseek (fd, 0, SEEK_CUR);
794	lseek (fd, offset, SEEK_SET);	906	lseek (fd, offset, SEEK_SET);
…		…
798		910
799	return res;	911	return res;
800	}	912	}
801	#endif	913	#endif
802		914
803	#ifndef HAVE_FUTIMES	915	#ifndef HAVE_UTIMES
804		916
805	# undef utimes	917	# undef utimes
806	# undef futimes
807	# define utimes(path,times) eio__utimes (path, times)	918	# define utimes(path,times) eio__utimes (path, times)
808	# define futimes(fd,times) eio__futimes (fd, times)
809		919
810	static int	920	static int
811	eio__utimes (const char *filename, const struct timeval times[2])	921	eio__utimes (const char *filename, const struct timeval times[2])
812	{	922	{
813	if (times)	923	if (times)
…		…
821	}	931	}
822	else	932	else
823	return utime (filename, 0);	933	return utime (filename, 0);
824	}	934	}
825		935
		936	#endif
		937
		938	#ifndef HAVE_FUTIMES
		939
		940	# undef futimes
		941	# define futimes(fd,times) eio__futimes (fd, times)
		942
		943	static int
826	static int eio__futimes (int fd, const struct timeval tv[2])	944	eio__futimes (int fd, const struct timeval tv[2])
827	{	945	{
828	errno = ENOSYS;	946	errno = ENOSYS;
829	return -1;	947	return -1;
830	}	948	}
831		949
…		…
835	# undef fdatasync	953	# undef fdatasync
836	# define fdatasync(fd) fsync (fd)	954	# define fdatasync(fd) fsync (fd)
837	#endif	955	#endif
838		956
839	/* sync_file_range always needs emulation */	957	/* sync_file_range always needs emulation */
840	int	958	static int
841	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	959	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
842	{	960	{
843	#if HAVE_SYNC_FILE_RANGE	961	#if HAVE_SYNC_FILE_RANGE
844	int res;	962	int res;
845		963
…		…
858	if (!res || errno != ENOSYS)	976	if (!res || errno != ENOSYS)
859	return res;	977	return res;
860	#endif	978	#endif
861		979
862	/* even though we could play tricks with the flags, it's better to always	980	/* even though we could play tricks with the flags, it's better to always
863	* call fdatasync, as thta matches the expectation of it's users best */	981	* call fdatasync, as that matches the expectation of its users best */
864	return fdatasync (fd);	982	return fdatasync (fd);
		983	}
		984
		985	static int
		986	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		987	{
		988	#if HAVE_FALLOCATE
		989	return fallocate (fd, mode, offset, len);
		990	#else
		991	errno = ENOSYS;
		992	return -1;
		993	#endif
865	}	994	}
866		995
867	#if !HAVE_READAHEAD	996	#if !HAVE_READAHEAD
868	# undef readahead	997	# undef readahead
869	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	998	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
870		999
871	static ssize_t	1000	static eio_ssize_t
872	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1001	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
873	{	1002	{
874	size_t todo = count;	1003	size_t todo = count;
875	dBUF;	1004	dBUF;
876		1005
…		…
888	}	1017	}
889		1018
890	#endif	1019	#endif
891		1020
892	/* sendfile always needs emulation */	1021	/* sendfile always needs emulation */
893	static ssize_t	1022	static eio_ssize_t
894	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1023	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
895	{	1024	{
		1025	eio_ssize_t written = 0;
896	ssize_t res;	1026	eio_ssize_t res;
897		1027
898	if (!count)	1028	if (!count)
899	return 0;	1029	return 0;
900		1030
		1031	for (;;)
		1032	{
		1033	#ifdef __APPLE__
		1034	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1035	#endif
901	#if HAVE_SENDFILE	1036	#if HAVE_SENDFILE
902	# if __linux	1037	# if __linux
		1038	off_t soffset = offset;
903	res = sendfile (ofd, ifd, &offset, count);	1039	res = sendfile (ofd, ifd, &soffset, count);
904		1040
905	# elif __freebsd	1041	# elif __FreeBSD__
906	/*	1042	/*
907	* Of course, the freebsd sendfile is a dire hack with no thoughts	1043	* Of course, the freebsd sendfile is a dire hack with no thoughts
908	* wasted on making it similar to other I/O functions.	1044	* wasted on making it similar to other I/O functions.
909	*/	1045	*/
910	{
911	off_t sbytes;	1046	off_t sbytes;
912	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1047	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
913		1048
914	if (res < 0 && sbytes)	1049	#if 0 /* according to the manpage, this is correct, but broken behaviour */
915	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1050	/* freebsd' sendfile will return 0 on success */
		1051	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1052	/* not on e.g. EIO or EPIPE - sounds broken */
		1053	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
916	res = sbytes;	1054	res = sbytes;
917	}	1055	#endif
		1056
		1057	/* according to source inspection, this is correct, and useful behaviour */
		1058	if (sbytes)
		1059	res = sbytes;
		1060
		1061	# elif defined (__APPLE__)
		1062	off_t sbytes = count;
		1063	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1064
		1065	/* according to the manpage, sbytes is always valid */
		1066	if (sbytes)
		1067	res = sbytes;
918		1068
919	# elif __hpux	1069	# elif __hpux
920	res = sendfile (ofd, ifd, offset, count, 0, 0);	1070	res = sendfile (ofd, ifd, offset, count, 0, 0);
921		1071
922	# elif __solaris	1072	# elif __solaris
923	{
924	struct sendfilevec vec;	1073	struct sendfilevec vec;
925	size_t sbytes;	1074	size_t sbytes;
926		1075
927	vec.sfv_fd = ifd;	1076	vec.sfv_fd = ifd;
928	vec.sfv_flag = 0;	1077	vec.sfv_flag = 0;
929	vec.sfv_off = offset;	1078	vec.sfv_off = offset;
930	vec.sfv_len = count;	1079	vec.sfv_len = count;
931		1080
932	res = sendfilev (ofd, &vec, 1, &sbytes);	1081	res = sendfilev (ofd, &vec, 1, &sbytes);
933		1082
934	if (res < 0 && sbytes)	1083	if (res < 0 && sbytes)
935	res = sbytes;	1084	res = sbytes;
936	}
937		1085
938	# endif	1086	# endif
		1087
		1088	#elif defined (_WIN32) && 0
		1089	/* does not work, just for documentation of what would need to be done */
		1090	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1091	/* libeio guarantees that the file offset does not change, and windows */
		1092	/* has no way to get an independent handle to the same file description */
		1093	HANDLE h = TO_SOCKET (ifd);
		1094	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1095	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1096
939	#else	1097	#else
940	res = -1;	1098	res = -1;
941	errno = ENOSYS;	1099	errno = ENOSYS;
942	#endif	1100	#endif
943		1101
		1102	/* we assume sendfile can copy at least 128mb in one go */
		1103	if (res <= 128 * 1024 * 1024)
		1104	{
		1105	if (res > 0)
		1106	written += res;
		1107
		1108	if (written)
		1109	return written;
		1110
		1111	break;
		1112	}
		1113	else
		1114	{
		1115	/* if we requested more, then probably the kernel was lazy */
		1116	written += res;
		1117	offset += res;
		1118	count -= res;
		1119
		1120	if (!count)
		1121	return written;
		1122	}
		1123	}
		1124
944	if (res < 0	1125	if (res < 0
945	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1126	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1127	/* BSDs */
		1128	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1129	|| errno == ENOTSUP
		1130	#endif
		1131	|| errno == EOPNOTSUPP /* BSDs */
946	#if __solaris	1132	#if __solaris
947	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1133	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
948	#endif	1134	#endif
949	)	1135	)
950	)	1136	)
…		…
954		1140
955	res = 0;	1141	res = 0;
956		1142
957	while (count)	1143	while (count)
958	{	1144	{
959	ssize_t cnt;	1145	eio_ssize_t cnt;
960		1146
961	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1147	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
962		1148
963	if (cnt <= 0)	1149	if (cnt <= 0)
964	{	1150	{
…		…
981	}	1167	}
982		1168
983	return res;	1169	return res;
984	}	1170	}
985		1171
		1172	#ifdef PAGESIZE
		1173	# define eio_pagesize() PAGESIZE
		1174	#else
		1175	static intptr_t
		1176	eio_pagesize (void)
		1177	{
		1178	static intptr_t page;
		1179
		1180	if (!page)
		1181	page = sysconf (_SC_PAGESIZE);
		1182
		1183	return page;
		1184	}
		1185	#endif
		1186
		1187	static void
		1188	eio_page_align (void **addr, size_t *length)
		1189	{
		1190	intptr_t mask = eio_pagesize () - 1;
		1191
		1192	/* round down addr */
		1193	intptr_t adj = mask & (intptr_t)*addr;
		1194
		1195	*addr = (void *)((intptr_t)*addr - adj);
		1196	*length += adj;
		1197
		1198	/* round up length */
		1199	*length = (*length + mask) & ~mask;
		1200	}
		1201
		1202	#if !_POSIX_MEMLOCK
		1203	# define eio__mlockall(a) EIO_ENOSYS ()
		1204	#else
		1205
986	static int	1206	static int
		1207	eio__mlockall (int flags)
		1208	{
		1209	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1210	extern int mallopt (int, int);
		1211	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1212	#endif
		1213
		1214	if (EIO_MCL_CURRENT != MCL_CURRENT
		1215	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1216	{
		1217	flags = 0
		1218	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1219	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1220	}
		1221
		1222	return mlockall (flags);
		1223	}
		1224	#endif
		1225
		1226	#if !_POSIX_MEMLOCK_RANGE
		1227	# define eio__mlock(a,b) EIO_ENOSYS ()
		1228	#else
		1229
		1230	static int
		1231	eio__mlock (void *addr, size_t length)
		1232	{
		1233	eio_page_align (&addr, &length);
		1234
		1235	return mlock (addr, length);
		1236	}
		1237
		1238	#endif
		1239
		1240	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1241	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1242	#else
		1243
		1244	static int
		1245	eio__msync (void *mem, size_t len, int flags)
		1246	{
		1247	eio_page_align (&mem, &len);
		1248
		1249	if (EIO_MS_ASYNC != MS_SYNC
		1250	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1251	|| EIO_MS_SYNC != MS_SYNC)
		1252	{
		1253	flags = 0
		1254	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1255	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1256	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1257	}
		1258
		1259	return msync (mem, len, flags);
		1260	}
		1261
		1262	#endif
		1263
		1264	static int
		1265	eio__mtouch (eio_req *req)
		1266	{
		1267	void *mem = req->ptr2;
		1268	size_t len = req->size;
		1269	int flags = req->int1;
		1270
		1271	eio_page_align (&mem, &len);
		1272
		1273	{
		1274	intptr_t addr = (intptr_t)mem;
		1275	intptr_t end = addr + len;
		1276	intptr_t page = eio_pagesize ();
		1277
		1278	if (addr < end)
		1279	if (flags & EIO_MT_MODIFY) /* modify */
		1280	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1281	else
		1282	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1283	}
		1284
		1285	return 0;
		1286	}
		1287
		1288	/*****************************************************************************/
		1289	/* requests implemented outside eio_execute, because they are so large */
		1290
		1291	static void
		1292	eio__realpath (eio_req *req, etp_worker *self)
		1293	{
		1294	char *rel = req->ptr1;
		1295	char *res;
		1296	char *tmp1, *tmp2;
		1297	#if SYMLOOP_MAX > 32
		1298	int symlinks = SYMLOOP_MAX;
		1299	#else
		1300	int symlinks = 32;
		1301	#endif
		1302
		1303	req->result = -1;
		1304
		1305	errno = EINVAL;
		1306	if (!rel)
		1307	return;
		1308
		1309	errno = ENOENT;
		1310	if (!*rel)
		1311	return;
		1312
		1313	if (!req->ptr2)
		1314	{
		1315	X_LOCK (wrklock);
		1316	req->flags |= EIO_FLAG_PTR2_FREE;
		1317	X_UNLOCK (wrklock);
		1318	req->ptr2 = malloc (PATH_MAX * 3);
		1319
		1320	errno = ENOMEM;
		1321	if (!req->ptr2)
		1322	return;
		1323	}
		1324
		1325	res = req->ptr2;
		1326	tmp1 = res + PATH_MAX;
		1327	tmp2 = tmp1 + PATH_MAX;
		1328
		1329	#if 0 /* disabled, the musl way to do things is just too racy */
		1330	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1331	/* on linux we may be able to ask the kernel */
		1332	{
		1333	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1334
		1335	if (fd >= 0)
		1336	{
		1337	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1338	req->result = readlink (tmp1, res, PATH_MAX);
		1339	close (fd);
		1340
		1341	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1342
		1343	if (req->result > 0)
		1344	goto done;
		1345	}
		1346	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1347	return;
		1348	}
		1349	#endif
		1350	#endif
		1351
		1352	if (*rel != '/')
		1353	{
		1354	if (!getcwd (res, PATH_MAX))
		1355	return;
		1356
		1357	if (res [1]) /* only use if not / */
		1358	res += strlen (res);
		1359	}
		1360
		1361	while (*rel)
		1362	{
		1363	eio_ssize_t len, linklen;
		1364	char *beg = rel;
		1365
		1366	while (*rel && *rel != '/')
		1367	++rel;
		1368
		1369	len = rel - beg;
		1370
		1371	if (!len) /* skip slashes */
		1372	{
		1373	++rel;
		1374	continue;
		1375	}
		1376
		1377	if (beg [0] == '.')
		1378	{
		1379	if (len == 1)
		1380	continue; /* . - nop */
		1381
		1382	if (beg [1] == '.' && len == 2)
		1383	{
		1384	/* .. - back up one component, if possible */
		1385
		1386	while (res != req->ptr2)
		1387	if (*--res == '/')
		1388	break;
		1389
		1390	continue;
		1391	}
		1392	}
		1393
		1394	errno = ENAMETOOLONG;
		1395	if (res + 1 + len + 1 >= tmp1)
		1396	return;
		1397
		1398	/* copy one component */
		1399	*res = '/';
		1400	memcpy (res + 1, beg, len);
		1401
		1402	/* zero-terminate, for readlink */
		1403	res [len + 1] = 0;
		1404
		1405	/* now check if it's a symlink */
		1406	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1407
		1408	if (linklen < 0)
		1409	{
		1410	if (errno != EINVAL)
		1411	return;
		1412
		1413	/* it's a normal directory. hopefully */
		1414	res += len + 1;
		1415	}
		1416	else
		1417	{
		1418	/* yay, it was a symlink - build new path in tmp2 */
		1419	int rellen = strlen (rel);
		1420
		1421	errno = ENAMETOOLONG;
		1422	if (linklen + 1 + rellen >= PATH_MAX)
		1423	return;
		1424
		1425	errno = ELOOP;
		1426	if (!--symlinks)
		1427	return;
		1428
		1429	if (*tmp1 == '/')
		1430	res = req->ptr2; /* symlink resolves to an absolute path */
		1431
		1432	/* we need to be careful, as rel might point into tmp2 already */
		1433	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1434	tmp2 [linklen] = '/';
		1435	memcpy (tmp2, tmp1, linklen);
		1436
		1437	rel = tmp2;
		1438	}
		1439	}
		1440
		1441	/* special case for the lone root path */
		1442	if (res == req->ptr2)
		1443	*res++ = '/';
		1444
		1445	req->result = res - (char *)req->ptr2;
		1446
		1447	done:
		1448	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1449	}
		1450
		1451	static signed char
987	eio_dent_cmp (const void *a_, const void *b_)	1452	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
988	{	1453	{
989	const eio_dirent *a = (const eio_dirent *)a_;	1454	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
990	const eio_dirent *b = (const eio_dirent *)b_;	1455	: a->inode < b->inode ? -1
		1456	: a->inode > b->inode ? 1
		1457	: 0;
		1458	}
991		1459
992	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score	1460	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
993	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */	1461
		1462	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1463	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1464
		1465	static void
		1466	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1467	{
		1468	unsigned char bits [9 + sizeof (ino_t) * 8];
		1469	unsigned char *bit = bits;
		1470
		1471	assert (CHAR_BIT == 8);
		1472	assert (sizeof (eio_dirent) * 8 < 256);
		1473	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1474	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1475
		1476	if (size <= EIO_SORT_FAST)
		1477	return;
		1478
		1479	/* first prepare an array of bits to test in our radix sort */
		1480	/* try to take endianness into account, as well as differences in ino_t sizes */
		1481	/* inode_bits must contain all inodes ORed together */
		1482	/* which is used to skip bits that are 0 everywhere, which is very common */
		1483	{
		1484	ino_t endianness;
		1485	int i, j;
		1486
		1487	/* we store the byte offset of byte n into byte n of "endianness" */
		1488	for (i = 0; i < sizeof (ino_t); ++i)
		1489	((unsigned char *)&endianness)[i] = i;
		1490
		1491	*bit++ = 0;
		1492
		1493	for (i = 0; i < sizeof (ino_t); ++i)
		1494	{
		1495	/* shifting off the byte offsets out of "endianness" */
		1496	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1497	endianness >>= 8;
		1498
		1499	for (j = 0; j < 8; ++j)
		1500	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1501	*bit++ = offs + j;
		1502	}
		1503
		1504	for (j = 0; j < 8; ++j)
		1505	if (score_bits & (1 << j))
		1506	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1507	}
		1508
		1509	/* now actually do the sorting (a variant of MSD radix sort) */
		1510	{
		1511	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1512	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1513	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1514	int stk_idx = 0;
		1515
		1516	base_stk [stk_idx] = dents;
		1517	end_stk [stk_idx] = dents + size;
		1518	bit_stk [stk_idx] = bit - 1;
		1519
		1520	do
		1521	{
		1522	base = base_stk [stk_idx];
		1523	end = end_stk [stk_idx];
		1524	bit = bit_stk [stk_idx];
		1525
		1526	for (;;)
		1527	{
		1528	unsigned char O = *bit >> 3;
		1529	unsigned char M = 1 << (*bit & 7);
		1530
		1531	eio_dirent *a = base;
		1532	eio_dirent *b = end;
		1533
		1534	if (b - a < EIO_SORT_CUTOFF)
		1535	break;
		1536
		1537	/* now bit-partition the array on the bit */
		1538	/* this ugly asymmetric loop seems to perform much better than typical */
		1539	/* partition algos found in the literature */
		1540	do
		1541	if (!(((unsigned char *)a)[O] & M))
		1542	++a;
		1543	else if (!(((unsigned char *)--b)[O] & M))
		1544	{
		1545	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1546	++a;
		1547	}
		1548	while (b > a);
		1549
		1550	/* next bit, or stop, if no bits left in this path */
		1551	if (!*--bit)
		1552	break;
		1553
		1554	base_stk [stk_idx] = a;
		1555	end_stk [stk_idx] = end;
		1556	bit_stk [stk_idx] = bit;
		1557	++stk_idx;
		1558
		1559	end = a;
		1560	}
		1561	}
		1562	while (stk_idx--);
		1563	}
		1564	}
		1565
		1566	static void
		1567	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1568	{
		1569	/* first move the smallest element to the front, to act as a sentinel */
		1570	{
		1571	int i;
		1572	eio_dirent *min = dents;
		1573
		1574	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1575	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1576	if (EIO_DENT_CMP (dents [i], <, *min))
		1577	min = &dents [i];
		1578
		1579	/* swap elements 0 and j (minimum) */
		1580	{
		1581	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1582	}
		1583	}
		1584
		1585	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1586	{
		1587	eio_dirent *i, *j;
		1588
		1589	for (i = dents + 1; i < dents + size; ++i)
		1590	{
		1591	eio_dirent value = *i;
		1592
		1593	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1594	j [1] = j [0];
		1595
		1596	j [1] = value;
		1597	}
		1598	}
		1599	}
		1600
		1601	static void
		1602	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1603	{
		1604	if (size <= 1)
		1605	return; /* our insertion sort relies on size > 0 */
		1606
		1607	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1608	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1609	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1610
		1611	/* use an insertion sort at the end, or for small arrays, */
		1612	/* as insertion sort is more efficient for small partitions */
		1613	eio_dent_insertion_sort (dents, size);
994	}	1614	}
995		1615
996	/* read a full directory */	1616	/* read a full directory */
997	static void	1617	static void
998	eio__scandir (eio_req *req, etp_worker *self)	1618	eio__scandir (eio_req *req, etp_worker *self)
999	{	1619	{
1000	DIR *dirp;	1620	DIR *dirp;
1001	EIO_STRUCT_DIRENT *entp;	1621	EIO_STRUCT_DIRENT *entp;
1002	unsigned char *name, *names;	1622	char *name, *names;
1003	int namesalloc = 4096;	1623	int namesalloc = 4096;
1004	int namesoffs = 0;	1624	int namesoffs = 0;
1005	int flags = req->int1;	1625	int flags = req->int1;
1006	eio_dirent *dents = 0;	1626	eio_dirent *dents = 0;
1007	int dentalloc = 128;	1627	int dentalloc = 128;
1008	int dentoffs = 0;	1628	int dentoffs = 0;
		1629	ino_t inode_bits = 0;
1009		1630
1010	req->result = -1;	1631	req->result = -1;
1011		1632
1012	if (!(flags & EIO_READDIR_DENTS))	1633	if (!(flags & EIO_READDIR_DENTS))
1013	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1634	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1014		1635
1015	X_LOCK (wrklock);	1636	X_LOCK (wrklock);
1016	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1637	/* the corresponding closedir is in ETP_WORKER_CLEAR */
1017	self->dirp = dirp = opendir (req->ptr1);	1638	self->dirp = dirp = opendir (req->ptr1);
		1639
		1640	if (req->flags & EIO_FLAG_PTR1_FREE)
		1641	free (req->ptr1);
		1642
1018	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1643	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1019	req->ptr1 = names = malloc (namesalloc);
1020	req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1644	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
		1645	req->ptr2 = names = malloc (namesalloc);
1021	X_UNLOCK (wrklock);	1646	X_UNLOCK (wrklock);
1022		1647
1023	if (dirp && names && (!flags || dents))	1648	if (dirp && names && (!flags || dents))
1024	for (;;)	1649	for (;;)
1025	{	1650	{
…		…
1033		1658
1034	/* sort etc. */	1659	/* sort etc. */
1035	req->int1 = flags;	1660	req->int1 = flags;
1036	req->result = dentoffs;	1661	req->result = dentoffs;
1037		1662
1038	if (dents)
1039	{
1040	eio_dirent *ent = dents + dentoffs;
1041
1042	while (ent > dents)
1043	(--ent)->name = names + (size_t)ent->name;
1044	}
1045
1046	if (flags & EIO_READDIR_STAT_ORDER	1663	if (flags & EIO_READDIR_STAT_ORDER)
1047	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1664	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1048	{
1049	/* pray your qsort doesn't use quicksort */
1050	qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1051	}
1052	else if (flags & EIO_READDIR_DIRS_FIRST)	1665	else if (flags & EIO_READDIR_DIRS_FIRST)
		1666	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1667	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1668	else
1053	{	1669	{
1054	/* in this case, all is known, and we just put dirs first and sort them */	1670	/* in this case, all is known, and we just put dirs first and sort them */
1055	eio_dirent *ent = dents + dentoffs;	1671	eio_dirent *oth = dents + dentoffs;
1056	eio_dirent *dir = dents;	1672	eio_dirent *dir = dents;
1057		1673
		1674	/* now partition dirs to the front, and non-dirs to the back */
		1675	/* by walking from both sides and swapping if necessary */
1058	while (ent > dir)	1676	while (oth > dir)
1059	{	1677	{
1060	if (dir->type == DT_DIR)	1678	if (dir->type == EIO_DT_DIR)
1061	++dir;	1679	++dir;
1062	else
1063	{
1064	--ent;
1065
1066	if (ent->type == DT_DIR)	1680	else if ((--oth)->type == EIO_DT_DIR)
1067	{	1681	{
1068	eio_dirent tmp = *dir;	1682	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1069	*dir = *ent;
1070	*ent = tmp;
1071		1683
1072	++dir;	1684	++dir;
1073	}	1685	}
1074	}	1686	}
		1687
		1688	/* now sort the dirs only (dirs all have the same score) */
		1689	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1075	}	1690	}
1076
1077	/* now sort the dirs only */
1078	qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1079	}
1080
1081	/* only provide the names array unless DENTS is specified */
1082	if (!(flags & EIO_READDIR_DENTS))
1083	{
1084	X_LOCK (wrklock);
1085	assert (!dents);
1086	req->ptr1 = 0;
1087	req->ptr2 = names;
1088	X_UNLOCK (wrklock);
1089	}
1090		1691
1091	break;	1692	break;
1092	}	1693	}
1093		1694
1094	/* now add the entry to our list(s) */	1695	/* now add the entry to our list(s) */
1095	name = entp->d_name;	1696	name = entp->d_name;
1096		1697
1097	/* skip . and .. entries */	1698	/* skip . and .. entries */
1098	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1699	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1099	{	1700	{
1100	int len = strlen (name) + 1;	1701	int len = D_NAMLEN (entp) + 1;
1101		1702
1102	while (expect_false (namesoffs + len > namesalloc))	1703	while (ecb_expect_false (namesoffs + len > namesalloc))
1103	{	1704	{
1104	namesalloc *= 2;	1705	namesalloc *= 2;
1105	X_LOCK (wrklock);	1706	X_LOCK (wrklock);
1106	req->ptr1 = names = realloc (names, namesalloc);	1707	req->ptr2 = names = realloc (names, namesalloc);
1107	X_UNLOCK (wrklock);	1708	X_UNLOCK (wrklock);
1108		1709
1109	if (!names)	1710	if (!names)
1110	break;	1711	break;
1111	}	1712	}
…		…
1114		1715
1115	if (dents)	1716	if (dents)
1116	{	1717	{
1117	struct eio_dirent *ent;	1718	struct eio_dirent *ent;
1118		1719
1119	if (expect_false (dentoffs == dentalloc))	1720	if (ecb_expect_false (dentoffs == dentalloc))
1120	{	1721	{
1121	dentalloc *= 2;	1722	dentalloc *= 2;
1122	X_LOCK (wrklock);	1723	X_LOCK (wrklock);
1123	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1724	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124	X_UNLOCK (wrklock);	1725	X_UNLOCK (wrklock);
1125		1726
1126	if (!dents)	1727	if (!dents)
1127	break;	1728	break;
1128	}	1729	}
1129		1730
1130	ent = dents + dentoffs;	1731	ent = dents + dentoffs;
1131		1732
1132	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1733	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1133	ent->namelen = len - 1;	1734	ent->namelen = len - 1;
1134	ent->inode = D_INO (entp);	1735	ent->inode = D_INO (entp);
		1736
		1737	inode_bits |= ent->inode;
1135		1738
1136	switch (D_TYPE (entp))	1739	switch (D_TYPE (entp))
1137	{	1740	{
1138	default:	1741	default:
1139	ent->type = EIO_DT_UNKNOWN;	1742	ent->type = EIO_DT_UNKNOWN;
…		…
1144	case DT_FIFO: ent->type = EIO_DT_FIFO; break;	1747	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145	#endif	1748	#endif
1146	#ifdef DT_CHR	1749	#ifdef DT_CHR
1147	case DT_CHR: ent->type = EIO_DT_CHR; break;	1750	case DT_CHR: ent->type = EIO_DT_CHR; break;
1148	#endif	1751	#endif
		1752	#ifdef DT_MPC
		1753	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1754	#endif
1149	#ifdef DT_DIR	1755	#ifdef DT_DIR
1150	case DT_DIR: ent->type = EIO_DT_DIR; break;	1756	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1757	#endif
		1758	#ifdef DT_NAM
		1759	case DT_NAM: ent->type = EIO_DT_NAM; break;
1151	#endif	1760	#endif
1152	#ifdef DT_BLK	1761	#ifdef DT_BLK
1153	case DT_BLK: ent->type = EIO_DT_BLK; break;	1762	case DT_BLK: ent->type = EIO_DT_BLK; break;
1154	#endif	1763	#endif
		1764	#ifdef DT_MPB
		1765	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1766	#endif
1155	#ifdef DT_REG	1767	#ifdef DT_REG
1156	case DT_REG: ent->type = EIO_DT_REG; break;	1768	case DT_REG: ent->type = EIO_DT_REG; break;
		1769	#endif
		1770	#ifdef DT_NWK
		1771	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1772	#endif
		1773	#ifdef DT_CMP
		1774	case DT_CMP: ent->type = EIO_DT_CMP; break;
1157	#endif	1775	#endif
1158	#ifdef DT_LNK	1776	#ifdef DT_LNK
1159	case DT_LNK: ent->type = EIO_DT_LNK; break;	1777	case DT_LNK: ent->type = EIO_DT_LNK; break;
1160	#endif	1778	#endif
1161	#ifdef DT_SOCK	1779	#ifdef DT_SOCK
1162	case DT_SOCK: ent->type = EIO_DT_SOCK; break;	1780	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163	#endif	1781	#endif
		1782	#ifdef DT_DOOR
		1783	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1784	#endif
1164	#ifdef DT_WHT	1785	#ifdef DT_WHT
1165	case DT_WHT: ent->type = EIO_DT_WHT; break;	1786	case DT_WHT: ent->type = EIO_DT_WHT; break;
1166	#endif	1787	#endif
1167	}	1788	}
1168		1789
1169	ent->score = 0;	1790	ent->score = 7;
1170		1791
1171	if (flags & EIO_READDIR_DIRS_FIRST)	1792	if (flags & EIO_READDIR_DIRS_FIRST)
1172	{	1793	{
1173	if (ent->type == EIO_DT_UNKNOWN)	1794	if (ent->type == EIO_DT_UNKNOWN)
1174	{	1795	{
1175	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1796	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176	ent->score = 98;	1797	ent->score = 1;
1177	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1798	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1178	ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */	1799	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1179	}	1800	}
1180	else if (ent->type == DT_DIR)	1801	else if (ent->type == EIO_DT_DIR)
1181	ent->score = 100;	1802	ent->score = 0;
1182	}	1803	}
1183	}	1804	}
1184		1805
1185	namesoffs += len;	1806	namesoffs += len;
1186	++dentoffs;	1807	++dentoffs;
1187	}	1808	}
		1809
		1810	if (EIO_CANCELLED (req))
		1811	{
		1812	errno = ECANCELED;
		1813	break;
		1814	}
1188	}	1815	}
1189	else
1190	req->result = -1;
1191	}
1192
1193	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1194	# undef msync
1195	# define msync(a,b,c) ((errno = ENOSYS), -1)
1196	#endif
1197
1198	int
1199	eio__mtouch (void *mem, size_t len, int flags)
1200	{
1201	intptr_t addr = (intptr_t)mem;
1202	intptr_t end = addr + len;
1203	#ifdef PAGESIZE
1204	const intptr_t page = PAGESIZE;
1205	#else
1206	static intptr_t page;
1207
1208	if (!page)
1209	page = sysconf (_SC_PAGESIZE);
1210	#endif
1211
1212	addr &= ~(page - 1); /* assume page size is always a power of two */
1213
1214	if (addr < end)
1215	if (flags) /* modify */
1216	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1217	else
1218	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1219
1220	return 0;
1221	}	1816	}
1222		1817
1223	/*****************************************************************************/	1818	/*****************************************************************************/
1224		1819
1225	#define ALLOC(len) \	1820	#define ALLOC(len) \
…		…
1240	X_THREAD_PROC (etp_proc)	1835	X_THREAD_PROC (etp_proc)
1241	{	1836	{
1242	ETP_REQ *req;	1837	ETP_REQ *req;
1243	struct timespec ts;	1838	struct timespec ts;
1244	etp_worker *self = (etp_worker *)thr_arg;	1839	etp_worker *self = (etp_worker *)thr_arg;
		1840	int timeout;
1245		1841
1246	/* try to distribute timeouts somewhat randomly */	1842	/* try to distribute timeouts somewhat evenly */
1247	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1843	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1248		1844
1249	for (;;)	1845	for (;;)
1250	{	1846	{
		1847	ts.tv_sec = 0;
		1848
1251	X_LOCK (reqlock);	1849	X_LOCK (reqlock);
1252		1850
1253	for (;;)	1851	for (;;)
1254	{	1852	{
1255	self->req = req = reqq_shift (&req_queue);	1853	self->req = req = reqq_shift (&req_queue);
1256		1854
1257	if (req)	1855	if (req)
1258	break;	1856	break;
1259		1857
		1858	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1859	{
		1860	X_UNLOCK (reqlock);
		1861	X_LOCK (wrklock);
		1862	--started;
		1863	X_UNLOCK (wrklock);
		1864	goto quit;
		1865	}
		1866
1260	++idle;	1867	++idle;
1261		1868
1262	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1869	if (idle <= max_idle)
1263	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1870	/* we are allowed to idle, so do so without any timeout */
		1871	X_COND_WAIT (reqwait, reqlock);
		1872	else
1264	{	1873	{
1265	if (idle > max_idle)	1874	/* initialise timeout once */
1266	{	1875	if (!ts.tv_sec)
1267	--idle;	1876	ts.tv_sec = time (0) + idle_timeout;
1268	X_UNLOCK (reqlock);
1269	X_LOCK (wrklock);
1270	--started;
1271	X_UNLOCK (wrklock);
1272	goto quit;
1273	}
1274		1877
1275	/* we are allowed to idle, so do so without any timeout */
1276	X_COND_WAIT (reqwait, reqlock);	1878	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1879	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1277	}	1880	}
1278		1881
1279	--idle;	1882	--idle;
1280	}	1883	}
1281		1884
…		…
1284	X_UNLOCK (reqlock);	1887	X_UNLOCK (reqlock);
1285		1888
1286	if (req->type < 0)	1889	if (req->type < 0)
1287	goto quit;	1890	goto quit;
1288		1891
1289	if (!EIO_CANCELLED (req))
1290	ETP_EXECUTE (self, req);	1892	ETP_EXECUTE (self, req);
1291		1893
1292	X_LOCK (reslock);	1894	X_LOCK (reslock);
1293		1895
1294	++npending;	1896	++npending;
1295		1897
…		…
1310	return 0;	1912	return 0;
1311	}	1913	}
1312		1914
1313	/*****************************************************************************/	1915	/*****************************************************************************/
1314		1916
		1917	int ecb_cold
1315	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1918	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1316	{	1919	{
1317	return etp_init (want_poll, done_poll);	1920	return etp_init (want_poll, done_poll);
1318	}	1921	}
1319		1922
		1923	ecb_inline void
1320	static void eio_api_destroy (eio_req *req)	1924	eio_api_destroy (eio_req *req)
1321	{	1925	{
1322	free (req);	1926	free (req);
1323	}	1927	}
1324		1928
1325	#define REQ(rtype) \	1929	#define REQ(rtype) \
…		…
1344	{ \	1948	{ \
1345	eio_api_destroy (req); \	1949	eio_api_destroy (req); \
1346	return 0; \	1950	return 0; \
1347	}	1951	}
1348		1952
		1953	static void
1349	static void eio_execute (etp_worker *self, eio_req *req)	1954	eio_execute (etp_worker *self, eio_req *req)
1350	{	1955	{
1351	errno = 0;	1956	if (ecb_expect_false (EIO_CANCELLED (req)))
		1957	{
		1958	req->result = -1;
		1959	req->errorno = ECANCELED;
		1960	return;
		1961	}
1352		1962
1353	switch (req->type)	1963	switch (req->type)
1354	{	1964	{
1355	case EIO_READ: ALLOC (req->size);	1965	case EIO_READ: ALLOC (req->size);
1356	req->result = req->offs >= 0	1966	req->result = req->offs >= 0
…		…
1368	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1978	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1369	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1979	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1370	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1980	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1371	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1981	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1372		1982
		1983	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1984	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1985	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1986	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1987
1373	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1988	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1374	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1989	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1375	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	1990	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1376	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1991	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1377	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	1992	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1384	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1999	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1385	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	2000	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1386	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2001	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1387	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2002	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1388	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2003	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1389	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	2004	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1390		2005
		2006	case EIO_REALPATH: eio__realpath (req, self); break;
		2007
1391	case EIO_READLINK: ALLOC (NAME_MAX);	2008	case EIO_READLINK: ALLOC (PATH_MAX);
1392	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2009	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1393		2010
1394	case EIO_SYNC: req->result = 0; sync (); break;	2011	case EIO_SYNC: req->result = 0; sync (); break;
1395	case EIO_FSYNC: req->result = fsync (req->int1); break;	2012	case EIO_FSYNC: req->result = fsync (req->int1); break;
1396	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2013	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1397	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	2014	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2015	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1398	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	2016	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2017	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1399	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2018	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2019	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1400		2020
1401	case EIO_READDIR: eio__scandir (req, self); break;	2021	case EIO_READDIR: eio__scandir (req, self); break;
1402		2022
1403	case EIO_BUSY:	2023	case EIO_BUSY:
1404	#ifdef _WIN32	2024	#ifdef _WIN32
1405	Sleep (req->nv1 * 1000.);	2025	Sleep (req->nv1 * 1e3);
1406	#else	2026	#else
1407	{	2027	{
1408	struct timeval tv;	2028	struct timeval tv;
1409		2029
1410	tv.tv_sec = req->nv1;	2030	tv.tv_sec = req->nv1;
1411	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2031	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1412		2032
1413	req->result = select (0, 0, 0, 0, &tv);	2033	req->result = select (0, 0, 0, 0, &tv);
1414	}	2034	}
1415	#endif	2035	#endif
1416	break;	2036	break;
…		…
1431	times = tv;	2051	times = tv;
1432	}	2052	}
1433	else	2053	else
1434	times = 0;	2054	times = 0;
1435		2055
1436
1437	req->result = req->type == EIO_FUTIME	2056	req->result = req->type == EIO_FUTIME
1438	? futimes (req->int1, times)	2057	? futimes (req->int1, times)
1439	: utimes (req->ptr1, times);	2058	: utimes (req->ptr1, times);
1440	}	2059	}
1441	break;	2060	break;
…		…
1446	case EIO_NOP:	2065	case EIO_NOP:
1447	req->result = 0;	2066	req->result = 0;
1448	break;	2067	break;
1449		2068
1450	case EIO_CUSTOM:	2069	case EIO_CUSTOM:
1451	((void (*)(eio_req *))req->feed) (req);	2070	req->feed (req);
1452	break;	2071	break;
1453		2072
1454	default:	2073	default:
		2074	errno = ENOSYS;
1455	req->result = -1;	2075	req->result = -1;
1456	break;	2076	break;
1457	}	2077	}
1458		2078
1459	req->errorno = errno;	2079	req->errorno = errno;
…		…
1489	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2109	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1490	{	2110	{
1491	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2111	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1492	}	2112	}
1493		2113
		2114	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2115	{
		2116	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2117	}
		2118
		2119	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2120	{
		2121	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2122	}
		2123
1494	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2124	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1495	{	2125	{
1496	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2126	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1497	}	2127	}
1498		2128
		2129	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2130	{
		2131	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2132	}
		2133
1499	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2134	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1500	{	2135	{
1501	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2136	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1502	}	2137	}
1503		2138
…		…
1524	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2159	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1525	{	2160	{
1526	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2161	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1527	}	2162	}
1528		2163
		2164	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2165	{
		2166	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2167	}
		2168
1529	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2169	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1530	{	2170	{
1531	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2171	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1532	}	2172	}
1533		2173
…		…
1539	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2179	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1540	{	2180	{
1541	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2181	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1542	}	2182	}
1543		2183
1544	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2184	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1545	{	2185	{
1546	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2186	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1547	}	2187	}
1548		2188
1549	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2189	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1569	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2209	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1570	{	2210	{
1571	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2211	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1572	}	2212	}
1573		2213
1574	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2214	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1575	{	2215	{
1576	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2216	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1577	}	2217	}
1578		2218
1579	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2219	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1595	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2235	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1596	{	2236	{
1597	return eio__1path (EIO_READLINK, path, pri, cb, data);	2237	return eio__1path (EIO_READLINK, path, pri, cb, data);
1598	}	2238	}
1599		2239
		2240	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2241	{
		2242	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2243	}
		2244
1600	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2245	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1601	{	2246	{
1602	return eio__1path (EIO_STAT, path, pri, cb, data);	2247	return eio__1path (EIO_STAT, path, pri, cb, data);
1603	}	2248	}
1604		2249
1605	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2250	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1606	{	2251	{
1607	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2252	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1608	}	2253	}
1609		2254
		2255	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2256	{
		2257	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2258	}
		2259
1610	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2260	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1611	{	2261	{
1612	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2262	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1613	}	2263	}
1614		2264
…		…
1622	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2272	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1623	}	2273	}
1624		2274
1625	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2275	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1626	{	2276	{
1627	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2277	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1628	}	2278	}
1629		2279
1630	static eio_req *	2280	static eio_req *
1631	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2281	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1632	{	2282	{
…		…
1656	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2306	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1657	{	2307	{
1658	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2308	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1659	}	2309	}
1660		2310
1661	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2311	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1662	{	2312	{
1663	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2313	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1664	}	2314	}
1665		2315
1666	#endif	2316	#endif
1667		2317
1668	eio_req *eio_grp (eio_cb cb, void *data)	2318	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1677	#undef SEND	2327	#undef SEND
1678		2328
1679	/*****************************************************************************/	2329	/*****************************************************************************/
1680	/* grp functions */	2330	/* grp functions */
1681		2331
		2332	void
1682	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2333	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1683	{	2334	{
1684	grp->int2 = limit;	2335	grp->int2 = limit;
1685	grp->feed = feed;	2336	grp->feed = feed;
1686		2337
1687	grp_try_feed (grp);	2338	grp_try_feed (grp);
1688	}	2339	}
1689		2340
		2341	void
1690	void eio_grp_limit (eio_req *grp, int limit)	2342	eio_grp_limit (eio_req *grp, int limit)
1691	{	2343	{
1692	grp->int2 = limit;	2344	grp->int2 = limit;
1693		2345
1694	grp_try_feed (grp);	2346	grp_try_feed (grp);
1695	}	2347	}
1696		2348
		2349	void
1697	void eio_grp_add (eio_req *grp, eio_req *req)	2350	eio_grp_add (eio_req *grp, eio_req *req)
1698	{	2351	{
1699	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2352	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1700		2353
1701	grp->flags |= EIO_FLAG_GROUPADD;	2354	grp->flags |= EIO_FLAG_GROUPADD;
1702		2355
…		…
1713	}	2366	}
1714		2367
1715	/*****************************************************************************/	2368	/*****************************************************************************/
1716	/* misc garbage */	2369	/* misc garbage */
1717		2370
		2371	eio_ssize_t
1718	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2372	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1719	{	2373	{
1720	etp_worker wrk;	2374	etp_worker wrk;
		2375	eio_ssize_t ret;
1721		2376
1722	wrk.dbuf = 0;	2377	wrk.dbuf = 0;
1723		2378
1724	eio__sendfile (ofd, ifd, offset, count, &wrk);	2379	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1725		2380
1726	if (wrk.dbuf)	2381	if (wrk.dbuf)
1727	free (wrk.dbuf);	2382	free (wrk.dbuf);
1728	}
1729		2383
		2384	return ret;
		2385	}
		2386

Diff Legend

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